the international, interdisciplinary society devoted to ... · program administrator editorial...

General Issue

Volume 50 Number 2 March/April 2016

The International, Interdisciplinary Society Devoted to Ocean and Marine Engineering, Science, and Policy

Journal

Featuring Expanded Papers from OCEANS ‘15 MTS/IEEE Conference

BOA RD OF D IR EC T ORSPresidentRay TollOld Dominion UniversityPresident-electDonna KocakHARRIS CorporationImmediate Past PresidentDrew MichelROV Technologies, Inc. VP—Section AffairsDoug WilsonCaribbean Wind LLCVP—Education and ResearchLiesl HotalingCOSEEVP—Industry and TechnologyDr. Andrew M. Clark, P.E.CypSeisNet, Inc.VP—PublicationsDr. Erika MontagueOceanGate, Inc.Treasurer and VP—Budget and FinanceMike PintoMBARI (Ret.)VP—Government and Public AffairsJustin ManleyTeledyne Benthos

SEC T IONSCanadian MaritimeVacantFlorida Erica MoultonSt. Pete MakersGreat Lakes Hans Van SumerenNorthwestern Michigan College Gulf CoastJohn CousinsGeneral Dynamics Information TechnologyHampton RoadsJim HaluskaOld Dominion UniversityHawaiiAlan HiltonCenter of Excellence for Research in Ocean Sciences (CEROS)HoustonTerry DahlkeGeospace OffshoreIndiaDr.R.VenkatesanNational Institute of Ocean Technology JapanHideyuki SuzukiUniversity of TokyoMonterey BayJill ZandeMATE CenterNew EnglandRhonda MonizSeaBotixNorwayTBDNewfoundland and LabradorDarrell O’NeillDept. of Innovation, Trade, and Rural DevelopmentOregonJeremy ChildressOregon State UniversityCollege of Oceanic & Atmospheric SciencesPuget SoundFritz StahrUniversity of WashingtonSan DiegoAlan KennyTeledyne RD Instruments

South KoreaProf. B.G. LeeMaritime & Ocean Engineering Research Inst.(MOERI/KORDI)Washington, D.C.Jake SobinKongsberg

P ROF E S SION A L C OMMI T T E E SIndustry and TechnologyBuoy TechnologyRick ColeRDSEA International, Inc. Cables and ConnectorsJennifer SnyderSAICDeepwater Field Development TechnologyDr. Benton BaughRadoil, Inc.DivingMichael LombardiOcean OpportunityMichael MaxHydrate Energy International LLCDynamic PositioningPete FougerePortsmouth, NHManned Underwater VehiclesWilliam KohnenHYDROSPACE GroupMooringsJack RowleySAICOceanographic InstrumentationDr. Carol JanzenSea-Bird Electronics, Inc.Offshore StructuresDr. Peter W. MarshallMHP Systems EngineeringRemotely Operated VehiclesChuck RichardsC.A. Richards and Associates, Inc.Ropes and Tension MembersEvan ZimmermanDelmar Systems, Inc.Seafloor EngineeringCraig EtkaScorpio Ventures, LLCUnderwater ImagingDr. Fraser DalgleishHarbor Branch Oceanographic InstituteUnmanned Maritime VehiclesRafael MandujanoVehicle Control Technologies, Inc.

Education and ResearchMarine ArchaeologyDr. Stephen WoodFlorida Institute of TechnologyMarine EducationErica MoultonMATE CenterMarine Geodetic Information SystemsDave ZilkoskiNOAAMarine MaterialsDr. R. VenkatesanNational Institute of Ocean Technology Ocean ExplorationErika MontagueOceanGate, Inc.Physical Oceanography/MeteorologyDr. Richard L. CroutNaval Research LaboratoryRemote SensingRyan VandermeulenUniversity of Southern Mississippi

Government and Public AffairsArctic TechnologyBasel AbdallaWood Group Kenny Marine Law and PolicyBill GlennRoyston Rayzor Vickery & Williams LLP Marine Mineral ResourcesAlex BarnardMarine SecurityDallas MeggittSound & Sea TechnologyOcean Economic PotentialVacantOcean Observing SystemsVembu SubramanianSECOORAIan WalshWET LabsOcean PollutionRyan MortonAnadarko PetroleumRenewable EnergyMaggie L. MerrillMarine Marketing Services and New England MREC at UMass DartmouthKenneth BaldwinUniversity of New Hampshire

S T UDEN T SEC T IONSAlpena Community CollegeArizona State University College of William & MaryDalhousie UniversityDuke UniversityFlorida Atlantic UniversityFlorida Institute of TechnologyLong Beach City CollegeMarine Institute of Newfoundland and LabradorMassachusetts Institute of TechnologyMemorial University (MUN)Monterey Peninsula College/Hartnell CollegeNorth Carolina State UniversityOregon State UniversityRutgers UniversityStockton UniversityTexas A&M University—College StationTexas A&M—Corpus ChristiTexas A&M University—GalvestonUnited States Naval AcademyUniversity of Florida University of HawaiiUniversity of HoustonUniversity of North Carolina—CharlotteUniversity of Puerto Rico, MayagüezUniversity of Southern MississippiUniversity of WashingtonWebb Institute

HONOR A RY MEMBE RS (†deceased)†Robert B. Abel†Charles H. Bussmann†John C. Calhoun, Jr.John P. Craven†Paul M. FyeDavid S. Potter†Athelstan Spilhaus†E. C. Stephan†Allyn C. Vine†James H. Wakelin, Jr.

Marine Technology Society Officers

Cover Images: Front: Randeni et al., Figure 2a (in this issue). Back (top to bottom): A Spray Glider under the surface about to embark on a 2-month mission off the West Florida Shelf (Photo credit: Brian Cousin/Harbor Branch Oceanographic Institute); Randeni et al., Figures 3c and 1b (in this issue); Cai et al., Figure 2a (this issue).

3Message From the MTS Journal EditorAnni Vuorenkoski Dalgleish

5Rendezvous Point Technique for Multivehicle Mine Countermeasure Operations in Communication-Constrained EnvironmentsVeronika Yordanova, Hugh Griffiths

17Autonomous Underwater Vehicle Motion Response: A Nonacoustic Tool for Blue Water NavigationSupun A. T. Randeni P., Alexander L. Forrest, Remo Cossu, Zhi Quan Leong, Peter D. King, Dev Ranmuthugala

27Coupled Modeling of Hydrodynamics and Sound in Coastal Ocean for Renewable Ocean Energy DevelopmentWen Long, Zhaoqing Yang, Andrea Copping, Z. Daniel Deng

37Development and Application of a New Mobile pH Calibrator for Real-Time Monitoring of pH in Diffuse Flow Hydrothermal Vent FluidsChunyang Tan, Kang Ding, William E Seyfried, Jr

Volume 50, Number 2, March/April 2016

General Issue

48Micro-Modem for Short-Range Underwater Mobile Communication SystemsJun-Ho Jeon, Sung-Joon Park

54Characteristics of Microstructure Turbulence Measurements With a Moored Instrument in the South China SeaXin Luan, Xiuyan Liu, Hua Yang, Shuxin Wang, Guojia Hou, Dalei Song

63Study and Design of a Heat Dissipation System in a Junction Box for Chinese Experimental Ocean Observatory NetworkDejun Li, Jun Wang, Jianshe Feng, Canjun Yang, Yanhu Chen

75Study of a Novel Underwater Cable With Whole Cable MonitoringJiawang Chen, Chunying Xu, Ying Chen, Jian Ji, Dongxu Yan

83Effect of Low Temperature and High Pressure on Deep-Sea Oil-Filled Brushless DC MotorsMinjian Cai, Shijun Wu, Canjun Yang

Text: SPi Cover and Graphics:Michele A. Danoff, Graphics By Design

The Marine Technology Society Journal (ISSN 0025-3324) is published by the Marine Technology Society, Inc., 1100 H Street NW, Suite LL-100 Washington, DC 20005

Annual Subscription Prices (all prices include shipping): Online access to the MTS Journal (including 14 years of archives) is FREE for MTS members. Members can purchase the printed Journal for $38 domestic, $55 Canada, and $125 international. Non-members and library subscriptions are: Online only (including 14 years of archives): $465; Online and print (including 14 years of archives): $499; Print only: $155 domestic, $175 Canada, and $250 international. Single-issue (hardcopy) price is $18 for members and $20 for nonmembers, plus $7 domestic shipping or $14 international shipping for all single-issue orders. Pay-per-view (worldwide): $28/article. Postage for periodicals is paid at Washington, DC and additional mailing offices.

P O S T M A ST E R :Please send address changes to:

Marine Technology Society Journal1100 H Street NW, Suite LL-100Washington, DC 20005Tel: (202) 717-8705; Fax: (202) 347-4302

Copyright © 2016 Marine Technology Society, Inc.

In This Issue

2 Marine Technology Society Journal

Anni Vuorenkoski Dalgleish, Ph.D.EditorHarbor Branch Oceanographic Institute at Florida Atlantic University

Ann E. Jochens, Ph.D.Texas A&M University

Donna KocakHARRIS Corporation

Scott Kraus, Ph.D.New England Aquarium

Dhugal Lindsay, Ph.D.Japan Agency for Marine-Earth Science & Technology

Eugene MassionMonterey Bay Aquarium Research Institute

Ralph Rayner, Ph.D.BMT Limited

Brock RosenthalOcean Innovations

EditorialErika Montague, Ph.D.VP of Publications

Anni Vuorenkoski Dalgleish, Ph.D.Editor

Amy MorganteManaging Editor

AdministrationRay TollPresident

Ms. Chris BarrettInterim Executive Director

Josh CohenMember Groups and Outreach Manager

Jeanne GloverMembership Manager

Suzanne VoelkerProgram Administrator

Editorial BoardThe Marine Technology Society is a not-for-profit, international professional society. Established in 1963, the Society’s mission is to promote the exchange of information in ocean and marine engineer-ing, technology, science, and policy.

Please send all correspondence to:The Marine Technology Society1100 H Street NW, Suite LL-100Washington, DC 20005(202) 717-8705 Tel. (202) 347-4302 FaxMTS Journal: [email protected]: [email protected]: [email protected]: [email protected]: [email protected]: [email protected]: www.mtsociety.org

MEMBERSHIP INFORMATIONmay be obtained by contacting the Marine Technology Society. Benefits include:■ Free subscription to the online Marine

Technology Society Journal, with highly reduced rates for the paper version

■ Free subscription to the bimonthly news -let ter, Currents, covering events, business news, science and technology, and people in marine technology

■ Member discounts on all MTS publications■ Reduced registration rates to all MTS and

MTS-sponsored conferences and workshops■ Member-only access to an expansive Job

Bank and Member Directory■ Reduced advertising rates in MTS

publications■ National recognition through our Awards

ProgramIndividual dues are $75 per year. Life membership is available for a one-time fee of $1,000. Patron, Student, Emeritus, Institutional, Business, and Corporate memberships are also available.

ADVERTISINGAdvertising is accepted by the Marine Tech nology Society Journal. For more informa-tion on MTS advertising and policy, please contact [email protected]

COPYRIGHTCopyright © 2016 by the Marine Technology Society, Inc. Authorization to photocopy items for internal or personal use, or the internal or personal use of specific clients, is granted by the Marine Technology Society, provided that the base fee of $1.00 per copy, plus .20 per page is paid directly to Copyright Clearance Center, 222 Rosewood Dr., Danvers, MA 01923.

For those organizations that have been granted a photocopy license by CCC, a sepa-rate sys tem of payment has been arranged. The fee code for users of the Transactional Reporting Service is 0025-3324/89 $1.00 + .20. Papers by U.S Government employees are declared works of the U.S. Government and are therefore in the public domain.

The Marine Technology Society cannot be held responsible for the opinions given and the statements made in any of the articles published.

ABSTRACTSMTS Journal article abstracts, if available, can be accessed for free at http://www.ingentaconnect.com/content/mts/mtsj.

Print and electronic abstracts of MTS Journal articles are also available through GeoRef <http://www.agiweb.org/georef/>, Aquatic Sciences and Fisheries Abstracts, published by Cambridge Scientific Abstracts <http://www.csa.com/factsheets/aquclust-set-c.php>, and Geobase’s Oceanbase published by Elsevier Science.

CONTRIBUTORSContributors can obtain an information and style sheet by contacting the managing editor. Sub missions that are relevant to the concerns of the Society are welcome. All papers are sub-jected to a stringent review procedure directed by the editor and the editorial board. The Journal focuses on technical material that may not otherwise be available, and thus technical papers and notes that have not been published previously are given priority. General commen-taries are also accepted and are subject to review and approval by the editorial board.

I N T R O D U C T I O N

Message From the MTS Journal Editor

Anni Vuorenkoski Dalgleish

This issue of Marine Technology Society Journal includes three papers featuring researchpresented at the OCEANS ’15 MTS/IEEE conference held on October 19–22, 2015, inWashington, DC. In addition, six papers in this issue describe recent advances covering arange of core general topics in the field of marine technology, including papers describing con-tributions to advances in ocean sensing technology, underwater communications systems, andunderwater platforms, as well as technology related to fixed and floating offshore structures.

In their paper, Yordanova andGriffiths address a number of challenges related to strategiesof using autonomous underwater vehicles (AUVs) for mine counter measure (MCM) appli-cations. The authors provide an overview of methods currently used for searching, detecting,classifying, and identifying marine mines, and propose a novel, more efficient approach uti-lizing an adaptive network of AUVs. The paper proposes a novel multivehicle path planningarchitecture resulting in reduction in mission time and cost by optimizing vehicle task allo-cation and by adaptive scheduling of rendezvous points. The underwater environment posessignificant challenges for implementing a dynamic strategy for a network of deterministicallymaneuvering vehicles; the authors include discussion on the limited range and bandwidth offree-space underwater communication systems and also on the limited ability to coordinateand configure positions. The simulation results presented in this paper pave a way for furtherexplorations into beneficial usage of networked AUVs, and although the method is particu-larly discussed in the context of MCM missions, it can be modified and applied for othermarine sectors.

AUVs are also the topic of Randeni et al. in their paper, which also is an expanded studybased on the authors’OCEANS ’15 presentation. The paper proposes a potential solution to awell-known and documented problem related toAUVnavigation systems. The authors provideresults from recent field experiments demonstrating and assessing the performance of a novelnonacoustic method to resolve the near-field flow velocity components around an AUV.In another expanded OCEANS ’15 paper, Long et al. present a theoretical analysis of thesound propagation from a marine hydrokinetic (MHK) device. The method is coupled withthe Finite Volume Community Ocean Model with the objective of studying MHK-inducedacoustic pressure fields, which may potentially interfere withmarine mammal communication.

In situ measurement of pH is essential in geochemical analysis of seawater and is the topicof the paper by Tan et al. The authors present a new mobile pH calibrator (MpHC), whichwas deployed on theDSRVAlvin to make in situ pHmeasurements at hydrothermal vents on

March/April 2016 Volume 50 Number 2 3

4 Marine Tec

the Juan de Fuca Ridge. The paper by Jeon and Park describes the design and implemen-tation of a micro modem for mobile underwater acoustic communications systems, wherethe modem could be integrated to a small mobile node such as a biomimetic fish robot.Luan et al. obtained long-term ocean turbulence measurements in the South China Seawith a moored instrument. Authors present the experimental data and processing methodsto analyze the effects of a turbulence field on the platform stability, tracking flexibility, andmicroscale shear spectra. In another contribution, Li et al. report the results from a theoreticalheat dissipation simulation and experimental study to assess the performance of a novel coolingsystemdesign for a junction box in a seafloor observatory network. Also in this issue, Chen et al.propose a low-cost method for underwater cable detection based on flexible sensors, whereasCai et al. describe an application and verificationmethod of a deep-seaDCmotor to verify themotor’s power loss and performance under low temperature and extreme pressure.

The Editorial Board would like to thank Jake Sobin for contributing to this issue by solicit-ing expanded papers from OCEANS ’15 presenting authors. The Editorial Board would alsolike to thank allMTSmembers who have contributed to theMTS Journal by submittingmanu-scripts and serving as Guest Editors or reviewers. As always, the Board encourages the MTScommunity to continue submitting manuscripts and proposals for future Special Editions.

hnology Society Journal

P A P E R

Rendezvous Point Technique for MultivehicleMine Countermeasure Operations inCommunication-Constrained Environments
A U T H O R SVeronika YordanovaHugh GriffithsUniversity College London
A B S T R A C T
Advantages of using a multivehicle network over a single autonomous under-
water vehicle platform include extended coverage area, potential cost and timeefficiency, and more robust performance. A common issue that slows advancementin the field is the limited available communication between the platforms. The ap-proach we propose is based on assigning a sequence of rendezvous points (RPs)where the vehicles can meet and exchange information. The work we present in thispaper applies principally to mine countermeasure and suggests that, despite thedisadvantage of time to allow for the vehicles to reach the RPs, there are techniquesthat can minimize the losses and provide advantages such as easier coordinationand access points for operator monitoring and system modifications. The resultswe present in this paper give an estimate of the reduction in loss if such an approachis employed. We make a comparison between the RP and a benchmark case byanalyzing numerical simulations.Keywords: rendezvous, multivehicle, mine countermeasures

do not propagate well underwater, andtherefore, acoustic transmissions are the

Introduction

There has been increased interest inthe last decade in using autonomous un-derwater vehicles (AUVs) in a networkconfiguration. However, the inabil-ity to maintain robust communicationcontinuously is a major constraint forthe development of such technology.Radio frequency and optical signals

most common choice. However, thereare severe limitations in range and band-width of the acoustic communication,and the network operation must takethis into account (Akyildiz et al., 2005).

The approach we propose is basedon assigning dynamically a sequenceof rendezvous points (RPs) throughoutthe mission, a location and time whereall agents in the network agree to meetand the vehicles can exchange informa-tion. This way a complete lack of con-nection can be assumed outside of theRP perimeter, thus providing a meansfor the system to operate under severechannel conditions. The work we pres-ent appliesmainly tomine countermea-sures (MCMs). This paper presents theidea of applying RP to MCM by pro-posing adaptive RP scheduling. Theoverall goal of this new approach is toenable an adaptive reallocation of sys-

tem resources to maximize search areawhile making explicit the rule of revi-siting all contacts on the way.

The remainder of the paper is orga-nized as follows: Section 2 gives detailson different types of MCMs; Section 3focuses on relevant work published re-cently on underwater networks and au-tonomy; Section 4 gives an overview oftheRP idea; Section 5 explains themeth-odologies adopted for evaluating the RPmethod and analyzing the suitable con-ditions for application; Section 6 showssimulation results; Section 7 presents theanalysis and limitations of the approach.The last section concludes the work andsuggests future directions.

Evolution of MCMSystemsTheKorean andGulf wars are exam-

ples where effective mine warfare was

March

applied. Warships were damaged andamphibious assaults aborted due to theinability of the navies to counter thisasymmetric threat. Currently, the sheernumber of existing naval mines is an-other reason to treat the problem as achallenging and diverse task: it is esti-mated that a million mines, of morethan 300 types, are stored by 60 naviesworldwide (this excludesU.S. weapons);mine production exists in more than30 countries, and export is done bymore than 20. These figures do not ac-count for improvised explosive devices,which are considered affordable andrelatively easy to make (Truver, 2012).TheMCMproblem arises from the dif-ficulty of distinguishing between the realmines and the false alarms (FAs) due tomine-like seafloor objects (Sariel, Balch,& Erdogan, 2008), as well as from in-efficient means of clearing them.

/April 2016 Volume 50 Number 2 5

Due to the large diversity of minetypes, their actuation mechanisms andmeans of deployment, it is hard to iden-tify a single best method to deal with theproblem of minefields. Currently, theconventional MCM approaches aresweeping and hunting. Minesweepingis used for removing mines by causingtheir detonation or capturing them.The design of the minesweeping vesselshould be stealthy such that it does nottrigger themine itself, but instead the ex-plosion occurs at a safe distance wherethe towed body with the triggeringmechanism is. Minesweepers can alsocapture the chain or cable of mooredmines, which was the predominantmine type untilWWI, but not that com-mon presently. The disadvantage ofusing such a sweeping technique is thatthere can be no assurance that the areais clear of mines. The other commonlyemployed technique, mine hunting, in-volves prior detection and classificationbefore any neutralization action istaken. This brings the advantage of pro-viding a probabilistic evaluation of thethreat level of the area. The sensor usedto detect mines is sonar, and the ac-quired imagery is processed by humanoperators to classify any contacts thatcould be actual mines. Once a decisionis made, the object can be neutralized.

However, some countermeasuresare becoming obsolete with the ad-vancement of mine technology, andnew solutions are being sought.

In minesweeping, the acoustic andmagnetic vessel signature that wouldactuate a mine is mimicked by theminesweeper vessel in an attempt totrigger it prematurely. This is becomingless effective as modern mines rely onmultiple signatures, which are not al-ways possible to simulate all together(Truver, 2012). On the other hand,even if we disregard for a moment thecomplicatedmultiple triggeringmecha-


nism, the sweepingmethod does not re-sult in any certainty that an area is minefree (Cornish, 2003). There is the pos-sibility that a mine did not activate eventhough it detected a suitable target. Theactuation mechanism sometimes in-volves randomized control that selectsa target from a sequence of detectionsin order to avoid multiple mines beingtriggered by the same contact or hitonly the first vessel from a convoy.

Mine hunting is considered morereliable than sweeping as at the end ofthe mission a level of confidence can bereached that can be input to a decision-making process on whether to drive aship or convoy through an area (Cao& Bell, 1999). The process includes adetection, classification, and identifica-tion stage performed using imageryfrom sonar towed by a ship. Once acertain target is located, a neutraliza-tion unit, usually a remotely operatedvehicle (ROV), is sent to dispose of it.However, there are some issues withtraditional mine hunting techniques.Mine hunting ships require designwith a minimal vessel signature so thatit does not trigger the mines in its vicin-ity (Schwarz, 2014). The bigger issuethat remains is that no matter howstealthy the ship is, there still needs tobe people on board to control the mis-sion. An alternative is to use a remotelyoperated vessel that is controlled froma safe base on the shore. While this istechnology that is advancing (Benjamin& Curcio, 2004), there is still the issueof the sonar not being able to explorethe contacts from close proximity.The method also relies on a single sen-sor that makes repeated scans over thearea, which might require long missiontimes. The interest in using AUVs formine hunting has been increasingin the last decade, due to cumulativework in multiple relevant fields toallow collaboration between vehicles

and lowering the price of commercialhardware. Recent advances that havecontributed to the area include workin communication, navigation, locali-zation, mapping, vehicle design, under-water swarms, autonomy (surface andunderwater), networking, internationalexperiments with defense and scientificapplications, etc. (Kalwa et al., 2015;Dugelay et al., 2015).

Another reason to adopt autono-mous vehicles for MCM applicationsis that there are still some unconven-tional methods in use, such as sendingdivers or mammals to perform the neu-tralization and search phases. Althoughnot that common, these techniquesdo exist. Mine neutralization methodsconducted by human divers remain themost reliable. Mammals have more en-durance and could be trained forMCMpurposes, and there are several existingprograms that have trained dolphinsand sea lions for such a mission. How-ever, the issue of misunderstanding be-tween the animal and the handler existsduring a mission. Overall, for bothhumans and mammals, the major dis-advantage is risking their lives by send-ing them into a minefield (Brown et al.,2012).

Using AUVs for MCM gives theadvantage of keeping all personnelat a safe distance by allowing for auton-omous operation. When a group ofnetworked vehicles is available, thishas the potential to reduce time, cost,and efforts compared to single platformsand current conventional methods.Such a configuration could also intro-duce distributed and more efficientarea coverage.

Related WorkEfforts to improve underwater sensor

network (UWSN) performance are com-monly aimed at modem development

and network protocol design (Partanet al., 2007; Kong et al., 2005; Cuiet al., 2006). When moving plat-forms are the focus of the networkconfiguration, solutions are drivenby ideas adopted by the robotics com-munity. Examples include adapt-ing coordination techniques, such asauction mechanisms (Sariel et al.,2008; DeMarco et al., 2011). Oftensuch approaches do not take into ac-count the limitations imposed by thecommunication in the underwaterchannel. Some methods for reducingreliance on communication usingprediction models are also available(Sotzing & Lane, 2010); however,they also rely on anticipated environ-mental conditions.

The work in this paper was devel-opedwith the focus of adapting optimi-zation techniques to the appropriateapplication constraints. Relevant ideashave been used for a group of net-worked surface and underwater vehiclesto adapt their formation to the waterbasin borders (Kemna et al., 2015).The possibility of a group of vehiclesreconfiguring their positions has alsobeen recognized when the nodes needto adapt to unexpected conditions orto seek optimal placement (Braca et al.,2014; Yilmaz et al., 2008). However,such solutions aimed at improving au-tonomy often discount or neglect theissue of communication.

The idea of synchronous rendez-vous has been recognized and adoptedfor ad hoc networks of mobile autono-mous agents (Cortes et al., 2006);however, it is not a typical approachfor UWSN. Although it provides ameans to avoid the communication re-striction by allowing all nodes to meetand plan further actions, one majordrawback is that part of the resourcesin the system are sacrificed to allowthe nodes to travel to the appointed

place. To reduce the lost time, thesepoints can be preplanned in a static se-quence to guarantee optimality. This,however, introduces rigidness andlack of adaptability to external events.Therefore, we propose a rendezvousapproach that allows dynamic onlinepoint allocation based on the informa-tion gathered by the vehicles and theirfuture goals.

RP ApproachTo improve the resource usage in

the system, we consider a specific ap-plication and scenario to measure theloss and evaluate the significance of pa-rameters for optimizing the RP sched-uling. This paper looks into applyingthe RP approach toMCMand the typ-ical operation specifics are presented.Furthermore, the loss mechanism isexplained with regards to the selectedscenario.

MCM Phases and Operationfor AUVs

A typical mine hunting operationhas five phases:■ Search: An area is scanned formine-

like objects (MLO). To secure com-plete coverage, often the platform ismoved in a lawnmower pattern.

■ Detection: Contact data are receivedfrom sensors, location is recorded,and a message of the contact andits location is created.

■ Classification: A decision is made asto whether an object is a mine-like ornon-mine-like. This is done by usingautonomous target recognition soft-ware or a human operator. However,currently this decision is not trustedto be made autonomously.

■ Identification: This determines thetype of the mine so further neutral-ization strategies can be employed.Often, there is a long delay between

March

Identification and Neutralizationphases, due to the system lackingthe ability of autonomous Classifi-cation and Identification, whichmeans the datafirst have to be recov-ered and processed off-board at theend of the searchmission and beforethe neutralization phase (Brownet al., 2012).

■ Neutralization: A mine is consid-ered neutralized once its locationis defined so it can be avoided. Incase the platform cannot evade themine, other measures are adopted.Those can include destroyingthe mine, disabling its detonationability, or disabling its ability todetect.Thework in this paper is concerned

with the Search and Identify phases,without taking into account the sensorspecifics or the autonomous target rec-ognition restrictions. Instead, it focuseson optimizing the collaboration be-tween multiple platforms. The Neutral-ize phase is excluded as often a ROV,rather than AUV, is used to properlyguide a disabling mechanism.

A typical multivehicle MCM mis-sion configuration includes two typesof sensor packages: Search-Classify-Map (SCM) and Reacquire-Identify(RI) (Freitag et al., 2005). The SCMrelies on a coarse side-scan sonar allow-ing faster speed during the searchingphase. The RI phase makes use of ahigh-resolution sensor, such as multi-beam sonar, that collects images for finalidentification and decision making.Often these two tasks are performedby separate vehicles. The SCMor searchvehicle follows a lawnmower pattern,while the RI vehicle relocates the targetand further examines it.

RP Loss in MCM ScenarioThe limitation of using sepa-

rate vehicles for distinct tasks is not


necessarily due to hardware restric-tions. Modern vehicles can havemultiple sensors mounted on thesame platform. However, coordinatedposition and task reconfigurability ofmultiple networked vehicles is stilla big issue underwater. Using RPsthroughout the mission enables com-munication between the AUVs andthus allows for dynamic task realloca-tion. This can aid higher resource uti-lization in the network. On the otherhand, there is a trade-off with thetime spent for the platforms to travelto RP. Evaluating and minimizingthis time loss is vital for the approachto be applicable.

In order to schedule an RP, allvehicles in the system have to agreeon the most convenient time andlocation to meet. This is done mul-tiple times throughout the mission,and the intervals between RPs areadapted based on the number of con-tacts found. We divide the RP sched-uling into two stages: (1) select thetime and (2) select the coordinates ofthe RP. To minimize the vehicle trav-eling time toward the RP, we wantto minimize the total number of RPsand push the next point as far in timeas possible. On the other hand, wewant to give the system nodes reg-ular chances to adjust future strategyand provide updates to the operator.It is useful to define a suitable timeinterval that satisfies the opposingdemands by showing when the re-source loss becomes prohibitive forthe mission.

The resource loss per vehicle is de-fined as the time spent by each plat-form to travel to the RP instead ofdoing mission-related task. That is,travel time from the point when thevehicle stops its search function andgoes to the RP. Equation 1 gives thisrelation by calculating what fraction


of the time window between RPs isspent for reaching the RP:

loss ¼x

v � tRP� 12

1� 1n

� �; n > 1

xv � tRP

� 12n

; n ¼ 1

8>><>>:

ð1Þ

where x is the width of the search area,n is the number of vehicles (the casesand transformations relevant to n areexplained further), v is speed, and tRPis the time until the next RP. An im-portant note is the reasoning behindthe choice to evaluate the loss per sin-gle vehicle and per single RP interval.First, a single platform loss gives flexi-bility for reconfigurability throughoutthe mission. Second, the single RP in-terval evaluation, as opposed to totalmission time, comes from the factthat all RP intervals vary, as they area function of the number and locationof contacts found during the searchphase. Therefore, the resulting lossfor adopting the RP approach will beadditive, but nevertheless each timeinterval will be unique.

In order to apply Equation 1, someassumptions are made: the area issearched sequentially; the vehicles arehomogeneous and have the samespeed. Since the resource loss calcula-tion is very dependent on the geom-etry of the search area, Figure 1 givesa graphical representation of the sce-nario considered in this paper forevaluating the RP approach. After de-fining the favorable parameters forminimizing the loss, some conclusionscan be drawn on when this approachmight not be applicable due to prohib-itive losses.

The mission scenario consideredfor the remainder of the simulationsin this paper is demining a strip nearthe shore. Typically, the width would

be much smaller compared to thelength of the area. Another assumptionis that the vehicles will not be able tocover the whole area before their batte-ries are exhausted. The overall purposeof the mission is to explore as much ofthe area as possible, while having theconstraint of revisiting the detectedMLOs.

In Figure 1, x and y are the dimen-sions of the searched area in the simu-lated scenario. The red circle on thetop left side pinpoints the starting po-sition of three available vehicles. Theyare all equipped with search (side-scansonar) and RI (multibeam) sensors andcan perform both SCM and RI tasks.The first RP is predefined and thespeed of the vehicles is known, sothis can give a good idea of the locationof the next meeting point as well as thearea that will be covered by each plat-form. Since the vehicles are homoge-neous, all three search areas will beequal (areaveh1 = areaveh2 = areaveh3).

The right-hand side of Figure 1gives an example of the continuationof the mission after the first RP. Atthe RP, all vehicles have shared the lo-cation of the contacts they have en-countered. The path and the requiredtime to revisit them have been calcu-lated. A decision has been made thatone vehicle will be reallocated withan RI task and follow all known con-tacts (the route is drawn in red in Fig-ure 1) and two platforms will continuein search mode, but now with changedarea patterns.

Given the scenario from Figure 1,the loss calculation from Equation 1can be further clarified. Normally,the AUVs would search in a lawn-mower pattern, as shown in the firstvehicle’s search box (‘veh 1’, left-handside of Figure 1). At the time of theRP, this will bring the AUV either atthe far or close corner of its search

box, which will define the distance andthe subsequent resource loss to the RP.However, the simulation is not opti-mized to always position the nodes atthe near corner. To account for this,the function in Equation 1 is adjustedto be proportional to the number ofvehicles performing the search phase.The special case when n = 1 is requiredas otherwise it would result in loss = 0.The current penalty for this case re-sembles the loss of n = 2.With this cor-rection, the calculation always assumesthe distance between the middle pointof the platform’s search area (on they axis, same as where the RP is) andthe RP point (this distance is notedon both sides of Figure 1). The distancepenalty is proportionate and increaseswhen increasing the number of searchplatforms. The loss calculation also de-pends on the speed of the platforms—the faster they are, the less time iswasted to travel to the RP. And lastly,the time to the next RP defines whatfraction of the total time will be used

for task-related purposes and whatpart will be traveling to the meetingpoint.

It is obvious that the longer thetime between the rendezvous, thesmaller fraction of the time will belost in traveling there. On the otherhand, there is a limitation on thistime depending on how often the op-erator would need an update from thenetwork. To evaluate when the lossbecomes prohibitive or to define theminimum time for RP, we have pa-rameterized Equation 1, and the resultis shown on Figure 2. The graph pre-sents the loss for a group of two andthree vehicles, respectively, for differ-ent speed values, v, and width of thesearched area, x, while the RP time in-terval and the length of the search areaare kept constant (RP = 1 h, y = 5000m).Then, in Figure 3, we have selectedfavorable, but realistic, conditions—3 vehicles, speed, 2 m/s and width of2,500 m. The plot shows what per-centage of the total time is lost if

March

we vary the time of the RP. This canbe used as a rule of thumb guidanceof the minimum time limit of thenext RP.

In the scenario we have selected inFigure 3, RP between 1 and 2 h gives aloss between 5% and 12% of the totaltime. If this is considered unacceptableby the operator, the time window canbe moved further in time. The calcu-lations in the next sections adopt theparameters used in Figure 3 (n = 3,x = 2,500 m, v = 2 m/s) and definean RP interval of 1 h assuming thereis no other parameter to base the deci-sion on. The scenario from Figure 1 isused throughout all simulations for theremainder of the paper.

RP Schedulingand Analysis

Defining the lower boundary forscheduling RP is useful when there isno other information available. How-ever, the main advantage of the RPapproach comes from the ability toreallocate vehicle tasks based on theinformation they have gathered duringthe search.

For example, if we assume the sce-nario from Figure 1, on the right, butwithout utilizing the RP technique, atthe start of the mission two vehicleswill be performing search tasks andthey will send the locations of the de-tected MLOs to the third vehicle thatis tasked with RI. However, if there isno prior information about what num-ber of targets to expect, themission canbe completed with few detectedMLOsor with a very large number. In the for-mer case, the RI vehicle will be idlemost of the time; in the latter, the mis-sion will be incomplete with targetsobserved only with low-resolution sen-sor. This scenario was assumed as abenchmark case and adopted in all

FIGURE 1

Rendezvous point approach: Left—starting point to first RP. Right—example scenario between twoRPs. (Color version of figures are available online at: http://www.ingentaconnect.com/content/mts/mtsj/2016/00000050/00000002.)


simulations throughout the remainderof this paper for comparison with theRP approach.

A higher amount of the system re-source could be utilized if the vehiclesare retasked adaptively according tothe number of detected contacts. Dur-ing an RP, a decision is made whether

10 Marine Technology Society Journa

there is a need to retask a search vehicleinto an RI vehicle and vice versa basedon how many detections were madeduring the previous RP interval. There-fore, the number of detected targets orthe time it takes for the RI vehicle(s)to revisit them is the main parameterthat decides how to utilize the avail-

l

able vehicles as well as how to schedulethe RP:

tRP ¼ f tRI þ treaquire� �

; min;max½ � ð2Þ

where tRI is a calculation of the timerequired for a single vehicle to travelin an optimal path and visit all detectedMLOs to identify them. Additionaltime for reacquiring the target locationsonce the AUV is at the reported coor-dinates and collecting high-resolutionsensory information was added bythe treaquire term. For more realistic cal-culation, this parameter needs to beadjusted depending on the missionconditions and the platform sensors.The [min, max] interval is derived byusing Equation 1 and Figure 3.

The second stage of scheduling thenext RP is the position estimationof the vehicles at the time when theywill be advancing toward it. This cal-culation is based on predicting thevehicles’ positions at the next tRP,, orthe area coverage of the search AUVs.The Areasearch parameter in Equation 3gives the area that will be covered bythe search vehicles in the time inter-val between RPs. This calculation isperformed during the meeting at RP,after the following RP time is definedby Equation 2. The position of thesearch vehicles coincides with they-axis from Figure 1, where the RP ispositioned.

Areasearch ¼ v � sw� tRP � loss� n

ð3Þ

where v is speed of a vehicle, sw isswath width of the vehicle’s sensor,tRP is time until next RP, loss is the re-source waste for traveling to RP usingEquation 1, and n is number of AUVs.However, since all platforms are as-sumed to be of the same type, at time

FIGURE 3

Loss of time resource (y axis) vs. total time between RP (x axis)—calculation based on Equation 1(parameters used: n = 3; x = 2,500 m; v = 2m/s).

FIGURE 2

Loss calculation for two vehicles (lines) and three vehicles (dots): different speed values andwidth ofsearch area (x), fixed RP time (1 h), and fixed length of the y dimension of searched space (5,000m).

tRP, all nodes will be aligned at thesame y coordinate in the search space.Therefore, the location of the RP is themiddle of the width of the search areaat this y coordinate, as seen in Figure 1.Equation 3 accounts only for the posi-tions of the search vehicles. The RIvehicles follow the shortest path be-tween the contacts they will be revisit-ing and reach back to the current RP.The advancement over the y-axis madeby the search vehicles has not beenconsidered for the RI vehicles.

These relations were used toexplore the gains of network re-configurability by applying the RPapproach. The steps used in our sim-ulation are summarized in the pseudocode in Algorithm 1, as well as therule-based decision making for howmany platforms to employ search andRI tasks.

At line 3, the first RP is predefined.It is assumed that no prior informationis available at the start of the mission;hence, the RP time is solely driven bythe loss calculation or given the mini-mum value from Equation 2. The sim-ulation runs until the mission_timeor the sum of the RP time windowsexceeds a predefined threshold. Thisthreshold was selected as a percentageof typical battery capacity of an AUV—70% of 10 h (line 7). Dependingon the selected parameters, the aver-age resource loss and search area arecalculated (lines 8 and 9). Once allplatforms reach the first RP, they willshare information about the detectedtargets. Communication at RP is as-sumed available. The MLOs are simu-lated by generating a random numberof targets, limited in number, andwith locations constrained within thearea that have been searched in thetime window (line 10). The shortestpath to revisit them is then calculated(line 11).

ALGORITHM 1

Dynamic allocation of vehicle tasks and RP.

The decision making on when to schedule the next RP and howmany vehiclesto send is given between lines 12 and 30. Simple rule-based logic is used. If theRI time falls in the interval between min_int and 2*min_int (line15), the time forthe next RP is selected as the time it will take an RI vehicle to check all targets, thesearch vehicles are reduced by one and the targets in the searched area are consid-ered identified. In case the time for the RI task exceeds the 2*min_int limit − line20 (due to too many targets), then two vehicles are tasked to perform the RI taskand the time for the next RP is half of the RI time. The number of search vehicles isreduced by two, and also all targets are considered identified. In the case when theRI time is below themin_int threshold, it is better to leave the targets unidentifieduntil the next RP cycle. The advantage is that all vehicles will perform search ratherthan one vehicle tasked with RI and then being idle for portion of the time. Thedisadvantage is that, on the next cycle, when there are enough targets to identify,the path to travel will be longer. In the next section, this trade-off is explored fur-ther. Lines 12 and 13 ensure that the simulation will be terminated in the case thatmore than two RI vehicles are required to identify the contacts from the previousRP interval.

ResultsThe function fromAlgorithm 1was used in aMATLAB simulation to evaluate

the performance of the RP approach when variable numbers of MLOs are de-tected. This assumes the case where no prior intelligence is available about theexpected number of contacts, and thus, the mission operators would be unableto manage the resources in the system offline.

The aim of this work is to show how applying the RP method gives the op-portunity for the system to adapt to an unexpected and varying need for retaskingthe vehicles. On the other hand, a benchmark scenario was designed where the


vehicles’ functionality is predefined, aswould be the case if no autonomy wasimplemented. Our expectations werethat the results would show such archi-tecture makes the system rigid and in-efficient. To represent this base case,at the start of our simulated missions,from the total of three vehicles, tomatch the RP case, two AUVs weredesignated to perform search and onewas tasked with identifying a detectedcontact. The reason to put more vehi-cles in search was that the objective ofthe mission is to maximize the overallsearch area. The advantage here is thatthere is no loss introduced from regu-larly traveling to a meeting point aswhen the RP method is applied. Thebenchmark case assumes that the vehi-cles have the ability to broadcast targetlocations continuously throughout theduration of the mission with no colli-sion or message loss.

Figure 4 shows three graphs, eachrepresenting simulation results withdifferent numbers of detected targets.At the top is a scenario with low num-ber of targets (0–10 generated targetsper RP window), the middle graphshows average number of targets (0–20 per RP), and at the bottom is alarge number of simulated contacts(0–30 per RP). The choice of targetnumber intervals is somewhat arbi-trary, as this parameter is not basedon literature or experiments. Sincethis variable is hard to determine, wehave adopted these intervals basedon the load that would be generatedwithin the assumed system resources.Thus, each graph shows results for100 repetitions of Algorithm 1 wherethe randomization of number and loca-tion of MLOs accounts for variabilityin losses and gains in the system.

The graphs in Figure 4 show thetotal area searched by the vehicles(depicted on the y axis) throughout


the available mission time (shown inminutes on the x axis). The scatteredblack dots are full mission simula-tions with the RP approach applied,while the red dots are simulationswith designated vehicles tasks or thebenchmark case. It is essential to clarifythat the mission time per simulationis defined by the RP approach, as de-

l

scribed in Algorithm 1 on line 7, andthen this limit is applied to the bench-mark case, where the same input oftarget distribution and number is usedto calculate the overall area searchedby this architecture. This creates a pairof RP and benchmark case simulationsthat give a comparable output as all in-puts are the same. In Figure 4, there are100 pairs in each graph, which allow anevaluation of the loss or gain of apply-ing the differentmethods, even thoughmultiple variables are present in thesimulations, such as number and loca-tion of the targets, as well as missionduration.

It can be seen on the graphs thatin all cases the red dots follow a linearrelationship—the more mission timeavailable, the larger area is searched.On the other hand, the black dots arescattered below and above this red line,showing lower resource utilization ifthey are below the red line and higherif they are above.

As expected, in the example offewer targets (top graph in Figure 4),most of the simulations show betterperformance for the RP approach,which allows for function reconfigur-ability, compared to the case of prede-fined tasks for the AUVs. The reasonis that the RI vehicle from the bench-mark mission is being idle for the ma-jority of the time, as there are notenough targets to fill its schedule. Onthe other hand, the third vehicle in theRP simulations is performing a searchfunction during most of the RP win-dows as there were not enough targetsfound to justify the RI task. It can beobserved that with increasing the num-ber of targets, the advantage of the RPis lost due to multiple vehicles taskedwith RI throughout themission.How-ever, in the case of a cluttered envi-ronment with many contacts present,the vehicles from the benchmark case

FIGURE 4

Comparison between the RP and benchmarkapproaches applied to MCM mission by plot-ting the search area gained by each method.The three graphs show how the results changewhen the number of targets increases: topgraph gives a simulation with low number oftargets (0 to 10), middle one doubles thetargets (0 to 20), and bottom graph shows acluttered environment (0 to 30 contacts perRP window).

are not able to revisit all detectionsas only one vehicle is tasked with RI,and its resources are not enough. Thecomparison between the two ap-proaches in their ability to revisit andidentify the contacts has been analyzedfurther in this paper, and simulationresults are available in Figure 7.

Figure 5 compares the distribu-tion of the search resource in the RPand benchmark case. The y axis gives

the average number of search vehiclesutilized per simulation, and the x axisprovides the simulation number.Since the base case is constantly usingtwo search vehicles, regardless of themission circumstances, there are 100equally spaced black dots at y = 2for every simulation number on allthree graphs. When RP is applied,the search platforms change theirnumber at every RP window depend-ing on the available MLOs, as some-times the RI vehicles can account for0, 1, or 2 of the total number of ve-hicles (Algorithm 1, lines 15–28).Each data point from the blue linesin Figure 5 corresponds to the aver-age number of search vehicles utilizedthroughout each simulation.

This result can be correlated withthe graphs in Figure 4, explaining thehigher search area covered in the casewhen less vehicles are tasked with RI(top graphs in both figures where lesstargets are found) and thus the aver-age search platform number is muchhigher. The bottom graph in Fig-ure 5, where the RP network of ve-hicles had to adapt to a large numberof targets, shows that in a proportionof the simulations less than two vehi-cles on average were available for thewhole mission. This can explain themajority of black scattered dots mov-ing toward the bottom right cornerin the third graph of Figure 4. Thisis the undesirable situation when moretime spent at the MCMmission yieldsless area searched.

On the other hand, even in the un-favorable event of encountering manytargets, the RP guarantees that all de-tections are revisited. The approachwas built with this core objective inmind and it defined a condition driv-ing the whole decision process in Algo-rithm 1. For the benchmark case, somecontacts were not revisited as the RI

March/A

vehicle resource would not be enough.This is another advantage for the basescenario, together with the unlimitedcommunication mentioned earlier,which makes its results to look morefavorable. However, the objective toprovide high-resolution images of allMLOs is violated in the benchmarkcase and thus making the RP approachshow more pessimistic results than itwould in a fair comparison. The RI suc-cess rate for both methods is furtheranalyzed in the next section.

AnalysisIn order to give a more suitable rep-

resentation of the outcomes showingthe efficiency gains and losses of theRP method, Figure 6 displays thesame data as used to present results inFigure 4, but in a different format. Toachieve a better visualization to distin-guish between the different approaches,the benchmark case data points wererotated and translated to coincide withthe x = 0 axis and plotted as a red line.Then, the difference between the areasearched by using RP and benchmarkwas calculated. This emphasized thediscrepancy in overall area betweenthe RP and base case pairs. To capturethe dynamics of this variation, thesepairs were further sorted in a descend-ing order, which resulted in the blueplotted line on all graphs in Figure 6.Such result representation is easier toevaluate by clearly differentiating howlikely it is for the RP approach to bebeneficial. Above the red line or thezero of the y axis, the RP approachgains additional search area even thoughthe vehicles waste time resource formultiple meetings. If the blue line isbelow the red one, then the simulationinstance results in a loss of search area.

In the top graph of Figure 6, show-ing results from a search with low

FIGURE 5

Comparison between the RP and benchmarkapproaches by giving the average number ofsearch vehicles used per mission. The graphsshow the decreasing availability of searchresource (top to bottom) in the RP case byincreasing the number of detections in thesimulations.

pril 2016 Volume 50 Number 2 13

number of contacts, the RP gives a sig-nificant gain over the benchmark casein themajority of the simulations, eventhough the loss for vehicles meeting isaccounted for. It is obvious that thisis the most advantageous situation toapply RP.

The middle graph gives the result-ing area search difference when dou-bling the number of targets from theprevious case. Now it is visible thatthe majority of the blue line is under


the red line; i.e., most simulationsindicate a loss to the system whenapplying the RP. However, a largepart of the simulation points fall verynear the red line. In about 40% ofthe simulations, the results are within0.2 km2 of total search area. This ac-counts for about 3 to 4 min of a single

l

vehicles operation time, accordingto Equation 1. This shows that, halfof the time, the RP approach is almostindistinguishably as good as thebenchmark case. However, it adds theoperational advantages discussed inSection 4 for system monitoring andperiodic data gathering.

The final graph presents a statewhen there is an extensive number ofcontacts simulated during the missionand shows the expected undesirableresults from the RP approach—a lotof resources would be spent on relocat-ing these contacts, and this will add tothe time loss brought from the regularmeetings. In reality, most of the con-tacts would be FAs. Environmentsthat are characterized by many FAsin the vehicle sensor, such as rockysea bottom, are unfavorable for theRP approach. It can be seen on thegraph that only very few data pointshave the blue line above the red, andin the majority of the cases this ap-proach brings loss. In addition, theloss propagates to significant differencein search area, in the order of 1.5 km2

and above, for the 30% of the cases.So far, only the search resource in the

system has been analyzed. Figure 7 givesan insight how the benchmark and RPapproach spend their RI resources. Thex axis on all graphs is the simulationnumber and the y axis measures thetime it would require to spend in RIphase per mission. In order to makethe comparison clearer, the results ofthe 100 repeated simulations weresorted in a descending order. The redline is the mission time (different foreach iteration of the algorithm). Theblue line follows the benchmark results,and the green line follows the corre-sponding RP results. It is clear thatthe benchmark case always uses lessresources compared to when RP is ap-plied (the blue line is always below the

FIGURE 6

Normalized and sorted simulation resultscomparing the gain of search area achievedby applying RP and base case. The top graphgives a simulation with low number of targets(0 to 10), middle one doubles the targets (0 to20), and the bottom graph shows a clutteredenvironment (0 to 30 contacts per RPwindow).

FIGURE 7

Normalized and sorted simulation resultscomparing the time required to reacquire andidentify (RI) all detected contacts for applyingRP versus the benchmark case. The top graphgives a simulation with a low number of tar-gets (0 to 10), the middle one doubles the tar-gets (0 to 20), and the bottom graph showsa cluttered environment (0 to 30 contacts perRP window).

green line). This comes from the factthat the RI vehicles in the RP simula-tions have to go back to each contactlocation only after its route is definedat the meeting point, which adds atime overhead. On the other hand,the benchmark case assumed perfectcommunication, so the two search ve-hicles can instantly send their contactlocations and the RI vehicle can selectthe shortest path without delay.

In all the cases when the bench-mark is above the red line or thelimit of the mission time, some con-tacts remain unidentified, as the RIresource is not sufficient. This is evi-dent in the bottom graph where themajority of the simulations result incontacts requiring extra RI time. Onthe other hand, the RP approach adap-tively reallocates part of its resourcefrom search into RI, resulting in morethan one RI vehicle per mission onaverage, as can be seen in Figure 5.Therefore, all contacts are identifiedby the end of the mission. This flexi-bility gives an advantage to the RP ap-proach that can surpass the overall lossof time to meet and the reduced totalsearch area.

A disadvantage of the RP in relationto the RI resource is that the contactsare always revisited after a meetingpoint. This results in a delayed reac-tion, and the current simulation doesnot allow for the contacts from thelast RP interval to be revisited (as thealgorithm terminates once the missiontime reaches the selected threshold—line 7 in Algorithm 1). This is an over-sight of the approach, but it does notviolate the conclusions made in thispaper, as both the benchmark and theapplied RP simulations disregard thesecontacts. In a real MCMor experimen-tal setting, where revisiting the contactsis crucial, this can be easily amendedby setting the last RP interval to force

each search vehicle to perform the RItask for its contacts before it goesback to the mission end point. This,however, would not contribute to thecurrent evaluation and was not in-cluded in the simulations.

Conclusion andFuture Direction

This paper has presented the idea ofadaptive scheduling of RPs for MCMapplication with AUVs. The benefitof using RP is that the vehicles canbe utilized at a constant rate indepen-dent of the number of targets detectedthroughout the mission. In contrast,when adopting a typical configurationwhere the functions of search and IDvehicles are separated from the startof the mission, the overall time toachieve the same result could be signif-icantly increased if there are large num-bers of MLOs or the ID vehicle couldbe underutilized if their number is low.The conventional approach thus be-comes less efficient compared to thepresented RP method.

This work can be improved byrelaxing or refining some of the as-sumptions which forced pessimisticcalculations, such as adjusting theposition of the search vehicle whencalculating the loss and the time re-quired for reacquiring a target. An-other important change would beallowing different kinds of vehicleswith varying speed and sensors thatwill add diversity to the simulationand make it more realistic. Finally, re-laxing the assumption of sequentialsearch allows for the use of a probabi-listic model of selecting which area tobe given priority for exploration andwhich one needs a repeated coverage.This also calls for more sophisticateddecision making, such as using Markovprocesses.

March/A

AcknowledgmentThis project is funded by Atlas

Elektronik UK and EPSRC DoctoralTraining Centre no: EP/G037264/1.The authors would like to thankA. Gibbert, A. Charlish, and R. Brindfor being available and providing earlyfeedback to this work.

AuthorsVeronika Yordanova andHughGriffithsDepartment of Electronic andElectrical EngineeringUniversity College LondonTorrington Place,London WC1E 7JE, UKEmail: [email protected]; [email protected]

ReferencesAkyildiz, I.F., Pompili, D., & Melodia, T.

2005. Underwater acoustic sensor networks:

research challenges. Ad Hoc Netw. 3(3):

257-79. http://dx.doi.org/10.1016/j.adhoc.

2005.01.004.

Benjamin, M.R., & Curcio, J.A. 2004.

COLREGS-based navigation of autonomous

marine vehicles. In: Proceedings of Autono-

mous Underwater Vehicles. pp. 32-9. Sebasco,

ME: IEEE/OES.

Braca, P., Goldhahn, R., LePage, K.D.,

Marano, S., Matta, V., & Willett, P. 2014.

Cognitive multistatic AUV networks. In:

17th International Conference on Information

Fusion (FUSION), 1-7. Salamanca: IEEE.

Brown, T., Damiano, J., Jhala, S., Moore, R.,

Morgan, B., Nguyen, V., & Turk, J. 2012.

Next generation mine countermeasures for

the very shallow water zone in support of

amphibious operations (No. NPS-SE-001).

Naval Postgraduate School Monterey CA

Department of Systems Engineering.

Cao, P., & Bell, M.J. 1999. New methods

to evaluate the detection performance of a

minehunter (No. DSTO-TN-0200). Defence

Science and Technology Organisation

Canberra (Australia).


Cornish, G.J. 2003. US Naval Mine Warfare

Strategy: Analysis of the Way Ahead. Army

War Coll Carlisle Barracks PA.

Cortes, J., Martínez, S., & Bullo, F. 2006.

Robust rendezvous for mobile autonomous

agents via proximity graphs in arbitrary

dimensions. IEEE T Automat Contr.

51(8):1289-98. http://dx.doi.org/10.1109/

TAC.2006.878713.

Cui, J.H., Kong, J., Gerla, M., & Zhou, S.

2006. The challenges of building mobile un-

derwater wireless networks for aquatic appli-

cations. IEEE Netw. 20(3):12-8. http://

dx.doi.org/10.1109/MNET.2006.1637927.

DeMarco, K., West, M.E., & Collins, T.R.

2011. An implementation of ROS on the

Yellowfin autonomous underwater vehicle (AUV).

In: OCEANS 2011, 1-7. Kona: MTS/IEEE.

Dugelay, S., Williams, D.P., Okopal, G.,

Connors, W.A., Midtgaard, Ø., Sæbø, T., …

Hesby, A. 2015. MANEX’14: Experimental

description and preliminary results. Tech.

Rep.: NATO STOCenter forMarine Research

and Experimentation, La Spezia, Italy: CMRE-

FR-2014-013.

Freitag, L., Grund,M., Von Alt, C., Stokey, R.,

& Austin, T. 2005. A shallow water acoustic

network for mine countermeasures opera-

tions with autonomous underwater vehicles.

In: Undersea Defence Technology (UDT),

Amsterdam: UDT. pp. 1-6.

Kalwa, J., Pascoal, A., Ridao, P., Birk, A.,

Glotzbach, T., Brignone, L., & Bibuli, M.

2015. EU project MORPH: Current status

after 3 years of cooperation under and

above water. IFAC-PapersOnLine. 48(2):

119-24. http://dx.doi.org/10.1016/j.ifacol.

2015.06.019.

Kemna, S., Caron, D.A., & Sukhatme, G.S.

2015. Constraint-induced formation switch-

ing for adaptive environmental sampling. In:

OCEANS 2015, 1-7. Geneva: MTS/IEEE.

http://dx.doi.org/10.1109/oceans-genova.

2015.7271361.

Kong, J., Cui, J.H., Wu, D., & Gerla, M.

2005. Building underwater ad-hoc networks

and sensor networks for large scale real-time


aquatic applications. IEEE Military Com-

munications Conference, MILCOM 2005.

3:1535-41. http://dx.doi.org/10.1109/

MILCOM.2005.1605894.

Partan, J., Kurose, J., & Levine, B.N. 2007.

A survey of practical issues in underwater

networks. ACM SIGMOBILE Mobile Com-

puting and Communications Review. 11(4):

23-33. http://dx.doi.org/10.1145/1347364.

1347372.

Sariel, S., Balch, T., & Erdogan, N. 2008.

Naval mine countermeasure missions. IEEE

Robot Autom Mag. 15(1):45-52. http://

dx.doi.org/10.1109/M-RA.2007.914920.

Schwarz, M. 2014. Future mine counter-

measures: No easy solutions. Naval War Coll

Rev. 67(3):123.

Sotzing, C.C., & Lane, D.M. 2010. Im-

proving the coordination efficiency of limited

communication multi-autonomus under-

water vehicle operations using a multiagent

architecture. J Field Robot. 27(4):412-29.

http://dx.doi.org/10.1002/rob.20340.

Truver, S.C. 2012. Taking mines seriously.

Naval War Coll Rev. 65(2):1-37.

Yilmaz, N.K., Evangelinos, C., Lermusiaux,

P.F., & Patrikalakis, N.M. 2008. Path

planning of autonomous underwater vehicles

for adaptive sampling using mixed integer

linear programming. IEEE J Oceanic Eng.

33(4):522-37. http://dx.doi.org/10.1109/

JOE.2008.2002105.

l

P A P E R

Autonomous Underwater Vehicle MotionResponse: A Nonacoustic Tool forBlue Water Navigation
A U T H O R SSupun A. T. Randeni P.Australian Maritime College,University of Tasmania
Alexander L. ForrestDepartment of Civil andEnvironmental Engineering,University of California-Davis,and Australian Maritime College,University of Tasmania

Remo CossuSchool of Civil Engineering,University of Queensland, andAustralian Maritime College,University of Tasmania

Zhi Quan LeongPeter D. KingDev RanmuthugalaAustralian Maritime College,University of Tasmania

A B S T R A C T
Autonomous underwater vehicles (AUVs) use secondary velocity over ground
measurements to aid the Inertial Navigation System (INS) to avoid unbounded driftin the point-to-point navigation solution. When operating in deep open ocean (i.e.,in blue water—beyond the frequency-specific instrument range), the velocity mea-surements are either based on water column velocities or completely unavailable. Insuch scenarios, the velocity-relative-to-water measurements from an acousticDoppler current profiler (ADCP) are often used for INS aiding. ADCPs have a blank-ing distance (typically ranging between 0.5 and 5 m) in proximity to the device inwhich the flow velocity data are undetectable. Hence, water velocities used to aid theINS solution can be significantly different from that near the vehicle and are subjectedto significant noise. Previously, the authors introduced a nonacoustic method to cal-culate the water velocity components of a turbulent water column within the ADCPdead zone using the AUV motion response (referred to as the WVAM method). Thecurrent study analyzes the feasibility of incorporating the WVAMmethod within theINS by investigating the accuracy of it at different turbulence levels of the watercolumn. Findings of this work demonstrate that the threshold limits of the methodcan be improved in the nonlinear ranges (i.e., at low and high levels of energy);however, by estimating a more accurate representation of vehicle hydrodynamiccoefficients, this method has proven robust in a range of tidally induced flow con-ditions. TheWVAMmethod, in its current state, offers significant potential to make akey contribution to blue water navigation when integrated within the vehicle’s INS.Keywords: water column velocity, INS, autonomous underwater vehicles (AUVs),acoustic Doppler current profilers (ADCPs), WVAM

utonomous underwater vehicles(AUVs) are submarine robots that are

Introduction

Aable to carry out ocean sampling cam-paigns (Curtin et al., 1993), bathymet-ric data collections (Grasmueck et al.,2006), and military and security exer-cises in unstructured environments(Paull et al., 2014). Accurate point-to-point guidance of an AUV (i.e., naviga-tion) as well as precise knowledge of itsposition within a 3-D domain (i.e., lo-calization) are mandatory (Paull et al.,2014). Despite the AUVs are beingdeveloped since the 1970s, the cur-rent navigation and localization tech-

niques need improvement, particularlyfor blue water operations (Hegrenæs&Berglund, 2009).

Inertial navigation is one of the keynavigation techniques used by AUVswhere the rotational and translationaccelerations of the AUV are deter-mined using the Inertial NavigationSystem (INS) sensors. Inertial naviga-tion is most effective when the acceler-ation solution from the INS is aidedwith the velocity over ground mea-surements from a bottom-tracking

March/A

Doppler Velocity Logger (DVL) or aGPS. An INS determines the position,velocity, and orientation of the vehicleusing the data from inertial measure-ment units (IMUs) relative to iner-tial space. Due to inherent errors, theINS navigation solution will have anunbounded drift unless counteractedwith the speed over ground velocityestimates (Hegrenæs & Hallingstad,2011). The DVL bottom track, wherethe velocity relative to the ground is de-termined from the Doppler frequency


shift of soundwaves reflected off theseabed, is commonly used for this pur-pose. This technology is limited byfrequency-specific penetration of tensto hundreds of meters through thewater column (Hegrenaes et al., 2008).

In blue water, specifically duringdeep descents/ascents, operating inthe mid-water zone and over roughbathymetry, DVL bottom track datamay be intermittently or completelyunavailable (Hegrenaes et al., 2008).In such cases, as illustrated in Figures 1aand 1b, an alternative approach is touse the DVL water-tracking mode,that is, acoustic Doppler current pro-filer (ADCP) mode, in conjunctionwith a real-time current estimationmethods to aid the INS navigation so-


lution (Hegrenæs & Berglund, 2009).This method has proven to producebetter navigation solutions and lessfree inertial drift compared to systemsutilizing INS alone; however, they areinherently susceptible to instrumentnoise (Hegrenæs & Berglund, 2009).

In addition to instrument noise,ADCPs or water-tracking DVLs havea blanking distance (i.e., a dead zone)in proximity to the device in which theflow velocity data remain unresolved(Simpson, 2001). The blanking dis-tance can typically span from 0.5 to5 m from the vehicle depending on thesampling frequency and selected binsize of the instrument (see Figure 1a).Therefore, the water velocity usedto aid the INS navigation is at least

l

0.5 m and sometimes even up to 5 mfrom the vehicle resulting in an uncer-tainty of the water velocity near the ve-hicle. This can induce error and causeadverse effects on the navigation andlocalization solutions.

The authors previously introduceda nonacoustic method to calculate thewater velocity components of a tur-bulent water column using the AUVmotion response without the aid ofan ADCP referred to as the WVAMmethod (Randeni et al., 2015b). Inthis method, the water velocities aredetermined by comparing the motionresponse of the vehicle when operatingwithin turbulent and calm water en-vironments. A key advantage of theWVAM method is that it is able to es-timate the flow velocities at the vehi-cle’s center of buoyancy; however,the authors indicated that the accuracyof the WVAM method might varywith the intensity of the measuring ve-locity components (Randeni et al.,2015b). That is, the precision of thewater velocity measurements from ahighly turbulent environment with rel-atively large velocities may be differentthan those of low turbulence con-ditions with lower velocities. The re-maining unanswered question waswhether the WVAM method is abletomeasure low and high flow velocitiesthat are likely to be encountered inblue water conditions.

In this context, the current work wasconducted as a feasibility study to findthe threshold limits by investigatingthe variation of the WVAM method’saccuracy in varying turbulence levelsin the water column through field de-ployments in a river estuary that exhib-its strong tidal currents up to 2 m s−1.The uncertainty of the method was cal-culated based on a direct comparisonwith velocity measurements obtainedfrom the AUV’s ADCP at different

FIGURE 1

(a) An AUV descending to a test site in blue water where the bottom track velocities of the vehicleare unavailable to the INS since the DVL beam span is unable to reach the seabed. (b) DVL isincapable of providing continuous bottom-track velocity measurements when traveling overrough bathymetry.

stages within the tidal cycle. Additionalsteps to improve the applicability of theWVAM method for blue water navi-gation are also discussed.

MethodologyInstruments

A Gavia -c lass modular AUVwas used to test the WVAM method(Randeni et al., 2015a), with the vehi-cle configured to an overall length of2.7 m, diameter of 0.2 m, and a dryweight in air of approximately 70 kg(see Figure 2a). The modularized vehi-cle in the tested configuration consistedof a nose cone, battery, GeoSwath inter-ferometric sonar, 1,200-kHz TeledyneRDI ADCP/DVL, Kearfott T24 INS,and control and propulsion modules,as shown in Figure 2b. The ADCPmodule of the AUV included two4-beam ADCPs arranged in a verticalplane to make both upward- and

downward-oriented velocity measure-ments (see Figure 2c). The ADCPswere set to profile the 9.94 m of watercolumn in 0.5-m range bins so that thethree directional water velocity compo-nents in the vehicle’s body-fixed coor-dinate system are measured in eachbin. Adjacent to the transducers (bothabove and below), there was a blank-ing distance of 0.44 m, as shown inFigure 2c. The maximum uncertaintymargin of the DVL in measuring thespeeds over ground is ±0.03 m s−1

(Hildebrandt & Hilljegerdes, 2010).The Kearfott T24 INS together

with the ADCP/DVL module mea-sured the orientation, velocities in 6 de-grees of freedom (6 DOFs), and theposition of the AUV. The depth ofthe vehicle was obtained from the pres-sure sensor on-board the AUV. Thesesensor measurements were recordedin the vehicle log at a frequency of0.87 Hz. The percentage uncertainty

March/A

of the pressure sensor is 0.1% givinga depth rate uncertainty margin of±1.0 × 10−4 m s−1 (Hildebrandt &Hilljegerdes, 2010). The respectiveuncertainties of the INS in providingthe pitch and yaw rates of the AUVare ±7.96 × 10−5 rad s−1 and ±1.60 ×10−4 rad s−1.

Site DescriptionThe objective of this study was

to determine the accuracy of theWVAMmethod in different flow con-ditions in order to assess the feasibil-ity of using the method for blue waterAUV navigation. To achieve this, theWVAM method was tested in theTamar estuary near the Batman Bridge(see Figures 3a and 3b), located inTasmania, Australia. Due to the prox-imity to the open sea and the flow con-striction of the river bed, the Tamarestuary exhibits strong tidal currentswith maximum flow velocities of upto 2.5 m s−1 (see Figure 3c).

Nine AUV runs were conductedalong the track line shown inFigure 3b.The first five were conducted on April14, 2015, and the last four were con-ducted on April 15, 2015. Runs 1–3and 6–9 were conducted during slackwater as shown in the tidal curvegiven in Figure 4. The velocity of thetidal currents during the first runs oneach day (i.e., Runs 1 and 6) was ap-proximately 0.25 m s−1. Due to the de-velopment of the strong flood tide, theflow speeds increased rapidly duringthe subsequent runs on each day.Runs 4 and 5 were carried out at par-tially and fully developed flood tideconditions with strong tidal currents(around 2 m s−1).

WVAM MethodFor the WVAM method (Randeni

et al., 2015a, 2015b), the AUV needsto undergo a straight line, constant

FIGURE 2

(a) Omarama Primary School (Lake Ohau, New Zealand) students inspecting the Gavia, the mod-ular AUV that was utilized to test theWVAMmethod; (b) the tested configuration of the vehicle; and(c) the body-fixed coordinate system (origin at the center of buoyancy—marked by the circle)showing the ADCP beam geometry.


depth trajectory through the regionwhere the water column velocities areto be measured. Typically, when anAUV is operating in an environmentwith fluctuating water velocities, theforces and moments induced by thesevelocities can interrupt the control sta-


bility and change the vehicle speed,depth, pitch, and yaw angles fromthe desired values. In order to compen-sate for such changes in performance,the vehicle’s control system adjuststhe revolution speed of the propellerand the angles of the four control sur-

l

faces. In response to these adjustments,the motion of the AUV will change inorder to return the AUV to the pre-scribed mission track unless the pro-peller and the control surfaces areunable to compensate for the externalforces (Kim & Ura, 2003).

TheWVAMmethod uses the com-pensation commands given by the vehi-cle’s control system to the propulsionmotor and control surfaces as recordedin the vehicle log and executes thesecommands within a simulation modelrepresenting a calmwater environment.Since there are no disturbing forces dueto flow variations in calm water condi-tions, the simulated vehicle motion willbe different than the actual motion.The difference between the twomotionresponses provides an estimation of theabsolute water column velocities in thebody-fixed coordinate system.

Equation 1 gives a generalized formof the water velocity calculation usedwithin the WVAM method:

v⇀water tð Þ ¼ v⇀AUV turbulentð Þ tð Þ � v⇀AUV calmð Þ tð Þð1Þ

where v⇀water is the velocity component ofthe surroundingwater column relative tothe earth in the body-fixed coordinatesystem (see Figure 2c), v⇀AUV turbulentð Þis the velocity component of theAUV observed in the turbulent envi-ronment, and v⇀AUV calmð Þ is the velocitycomponent obtained from the calmwater simulation when the controlcommands recorded during the fieldtests were simulated. Subscript t in-dicates the time step. To estimate thewater velocity component along thex, y, and z axes, v⇀ is replaced withthe surge, sway, and heave velocitycomponents (i.e., u, v, and w) of thevehicle in the body-fixed frame of ref-erence, respectively.

FIGURE 3

(a) The experimental field site in Tasmania, Australia (inset). (b) The Tamar estuary with a bidirec-tional arrow representing the AUV track. (c) Due to the proximity to the open sea and the flowconstriction of the river bed, the site exhibited strong tidal currents.

FIGURE 4

The water level relative to the mean sea level (MSL) observed on April 14 and 15, 2015, with theperiods that the AUV runs were conducted indicated by filled diamond markers.

Simulation Model andHydrodynamic Coefficients

The simulation model of the GaviaAUV was developed to reproducethe vehicle’s trajectory within a calmwater environment in response to thetime series of the control commands.It requires an accurate approximationof the associated hydrodynamic coef-ficients (i.e., a representation of theforces and moments acting on the ve-hicle at different orientations, veloci-ties) to adequately predict the motionof the AUV. Generally, the forces andmoments acting on submerged bodiesin 6 DOFs are highly nonlinear (Lewis,1988). For example, the vertical hydro-dynamic force acting on an AUV var-ies linearly with its pitch angle up toa value of around ±8°, beyond whichit becomes nonlinear (Randeni et al.,2015a). Similar threshold values existfor other hydrodynamic forces andmoments transiting between their re-spective linear and nonlinear rangers.Therefore, the hydrodynamic coeffi-cients estimated for the linear rangesare only valid up to a certain thresholdvalue (Lewis, 1988).

During the initial development ofthe WVAM method, a basic curve fit-ting method was utilized to determinethe hydrodynamic coefficients due toits relative simplicity (Randeni et al.,2015a). The coefficients obtained fromthis method were limited to small anglesof incidence (i.e., generally below 8°) re-stricting them to the linear range.Whenan AUV operates in turbulent environ-ments, its pitch and yaw angles typicallyfluctuate around the baseline values.The magnitude of these fluctuations in-creases with increasing levels of turbu-lence due to the inability of the AUV ’scontrol system to compensate for thesevere disturbance forces (Kim & Ura,2003). Therefore, in extremely turbu-lent water columns, these fluctuation

angles will be greater than ±8°. Thus,a simulation model that is limited tolinear hydrodynamics data is unableto adequately replicate the motion ofthe vehicle in extreme environments.

Results and DiscussionValidation of the WVAM Method

Figures 5a, 5c, and 5e illustrate thevariations of the ADCP-measured

March/A

water velocity components (i.e., veloc-ities in x, y, and z directions, respec-tively) with the vertical distance fromthe AUV. Figures 5b, 5d, and 5f pres-ent the respective flow velocity com-ponents estimated with the WVAMmethod. The velocity data shown inFigure 5 were recorded during Run 1(i.e., when the AUV was movingwith the predominant tidal currents).As seen, theWVAM velocity estimates

FIGURE 5

Panels a, c, and e illustrate the ADCP-measured variations of the water column velocity components inx, y, and z directions (respectively)with the vertical distance from theAUV. The respectiveflowvelocitycomponents estimated by the WVAM method are presented in Panels b, d, and f. The illustrated ve-locity data were obtained from Run 1 when the AUVwas moving with the predominant tidal currents.


well correlate with the ADCP measurements, especially around the bins closer tothe vehicle.

The uncertainty of the water velocity measurements from theWVAMmethodcompared to the ADCP results was quantified using Equation 2 that approximatesthe standard error (SE) with a percentage confidence of 99.7% (Devore, 2011).

SE ¼ 3ffiffiffin

p

ffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiXnt¼1

vwater ADCPð Þ � vwater WVAMð Þ� �2

n

vuuutð2Þ

where vwater ADCPð Þ is the water velocity measured using the ADCP, vwater WVAMð Þ isthe water velocity calculated using the WVAM method, and n is the number oftime steps. The SEs for the velocity components in the x, y, and z directions for thefirst run were ±0.068, ±0.017, and ±0.045 m s−1, respectively. These numbersrepresent the difference between WVAM and ADCP velocity predictions ineach of the three directions, with the greatest error seen in the x direction.

It is evident from these plots that the WVAM method provides a good repli-cation of the flow velocities measured using the onboard ADCP. Since the vehiclewas moving with the predominant tidal flow direction, a positive water velocityalong the x direction is seen in Figures 6a and 6b. The negative water velocitycomponent in the y direction (Figures 6c and 6d) indicates that the transverseflow direction is from northeast side to southwest. Although the transversewater velocity does not follow the ADCP velocity trend for some periods, theaveraged velocities are in favorable agreement. However, further studies will beconducted to investigate the reasons for this. The best replica between theWVAM and ADCP velocities is seen in the vertical velocity component. The larg-est mismatches in the vertical velocities are seen at the peaks. The hydrodynamiccoefficients of the simulation model estimated using the basic system identifica-tion method were only valid for small angles of incidence of the vehicle, where thecoefficients are in their linear ranges. Therefore, as the yaw and pitch angle fluc-tuations become larger, the accuracy of the simulation model decreases, adverselyaffecting the WVAM velocity prediction (Randeni et al., 2015a). The disparity atpeaks of the vertical velocity component is due to the hydrodynamic coefficientsexceeding their linear ranges causing a reduction in the accuracy of the simulationmodel.

A recent study (Green, 2015) compared the water velocity measurements ob-tained from theGavia AUV ’s onboard ADCP with a stationary ADCPmoored tothe seabed. This investigation was carried out in the same test location (i.e., Tamarestuary) with the sameGavia-class AUV as used in this study. Green (2015) foundvery good agreements between the AUV-mounted and stationary ADCPmeasure-ments. In addition, the stationary ADCP data set was used to validate estimateswith theWVAM for a period when the AUVwas in close proximity to themooredADCP. Similar to the findings from the AUV-ADCP and stationary ADCP com-parison, the velocities between WVAM and stationary ADCP showed a goodagreement with differences of 0.05, 0.08, and 0.01m s−1 for the respective velocitycomponents in the x, y, and z directions (see Figure 6).


Accuracy of the WVAM MethodWith the Level of Turbulence

Runs 1–3 and 6–9 were conductedin lower turbulent environments, witha vertical water velocity range of around−1 to 1 m s−1 compared to the Runs 4and 5, which were at vertical water ve-locity range of −2 to 0.5 m s−1. Dueto the developing flood tide, the levelof turbulence increased graduallywith each run. The averaged fluctua-tions of the vehicle’s yaw angle, pitchangle, and surge speed from the tar-get values are given in Table 1. If theAUV ’s control system is capable ofguiding the vehicle accurately alongthe prescribed path in turbulent envi-ronments, these values would be closeto zero. As seen from Table 1, the fluc-tuations have raised with the increas-ing level of turbulence.

The SEs of the WVAM watervelocity predictions compared to theADCP measurements for each runare presented in Table 1 and Figure 7.

FIGURE 6

Comparison of the horizontal water velocitiesobtained from the WVAM method, stationaryADCP, and the AUV-fixed ADCP; modifiedfrom Green (2015).

The uncertainties of the vertical andtransverse water velocity estimatesincrease with the averaged fluctua-tions of the pitch and yaw angles, re-spectively. For example, in Run 2, theaveraged deviation of the yaw angle isaround ±17°, and the SE of the trans-verse velocity prediction is much largerthan for Run 1 (i.e., 0.153 m s−1).Run 1 has a smaller deviation of theyaw angle of around ±1.5°, which re-sults in a much smaller uncertainty inthe vicinity of 0.017 m s−1. A similaroutcome is seen in the vertical water

velocity component. In Run 9, theaveraged deviation of the pitch angleis ±7.5° giving an SE in the verticalwater velocity prediction of around0.1 m s−1. However, in Run 6, thedeviation of the pitch angle is compar-atively lower at ±3.1° resulting in asmaller SE for the velocity of around0.05 m s−1. Although the tides werenot fully developed during the Runs 2and 8, the yaw angle deviations are rel-atively higher. This could be as a resultof a strong crosscurrent acting on thevehicle at the particular water height.

March/A

Generally, an SE between theADCPandWVAM that results up to 0.1 m s−1

is acceptable as the uncertainty of themeasurements from an AUV-mountedADCP is of similar magnitude (Fong& Jones, 2006). The threshold aver-aged deviation of the pitch and yawangles that provide the water columnvelocities with an SE below 0.1 m s−1

is around 7°–8°. Above this thresholdangle, the hydrodynamic coefficientsusually become nonlinear. The hydro-dynamic coefficients estimated for thesimulation model using the curve fit-ting method are only valid for smallangles of incidence of the vehicle,that is, when the coefficients are inthe linear range (Randeni et al., 2015b).Therefore, when the pitch and yawangles are above the linear range, theaccuracy of the simulation modeldecreases, and the uncertainty of thevertical and transverse water veloc-ity components obtained from theWVAM method increases. Hence,the WVAM method, in its currentstate, is less accurate in determiningtransverse and vertical water velocitiesin high turbulent environments.

TABLE 1

SEs of the water velocity components determined from the WVAM method compared to the onboard ADCP measurements and the associatedaveraged deviations for the prescribed parameters.

Run Number

SE
Averaged Deviation From the Prescribed Value
u
v w Yaw Angle(Degrees)
Pitch Angle(Degrees)

pril 2016 Volume 50

Surge Speed(ms−1)

1
0.068 0.017 0.045 ±1.5 ±3.6 ±0.2
2
0.029 0.153 0.075 ±17.4 ±7.1 ±0.6
3
0.041 0.095 0.108 ±3.5 ±7.5 ±0.7
4
0.063 0.061 0.241 ±2.8 ±9.2 ±1.6
5
0.035 0.079 0.191 ±5.6 ±9.7 ±1.6
6
0.042 0.063 0.052 ±3.1 ±3.1 ±0.1
7
0.048 0.058 0.067 ±2.6 ±6.4 ±0.3
8
0.025 0.104 0.092 ±14.5 ±7.4 ±0.4
9
0.054 0.082 0.097 ±5.8 ±7.5 ±0.6
FIGURE 7

The variations of the WVAM method’s SEs with the averaged fluctuations of the vehicle surgespeed and yaw and pitch angles from the prescribed values. The water velocity components inx, y, and z directions correspond with the surge speed, yaw angle, and pitch angle, respectively.

Number 2 23

The accuracy of the water velocitycomponent in the x direction remainsgenerally the same for all the runs re-gardless of the averaged fluctuationsof the surge speed (see Figure 7). Duringthe development of the simulationmodel, the hydrodynamic coefficientsdominating the motion in the x direc-tion were estimated for a propellerspeed range of 525–825 RPM (i.e.,an approximate vehicle speed rangeof 1.43 to 2.46 m s−1 in a calm waterenvironment). During this study, theAUV field runs were conducted at700 RPM providing a mean forwardspeed of around 2.04 m s−1 in calmwater conditions with a standard devi-ation of 0.01 m s−1. The observedspeed during the runs varied as muchas ±1.6 m s−1 from the calm waterspeed of 2.04 m s−1, especially duringRuns 5 and 6 due to the strong drag onthe vehicle imparted by the strong tidalcurrents. Although the actual vehiclespeed deviated from the simulatedspeed, the prescribed propeller speedof 700 RPM provides a calm watersimulated speed of around 2.04 m s−1,which is within the identified forwardspeed range of 1.43–2.46m s−1. There-fore, the simulationmodel provided anaccurate representation of the forwardspeed of the AUV in calm water. It canthus be concluded that the SE resultsbetween the WVAM and ADCP inthe x direction generally remained un-related to the turbulence level of thewater column.

Applicability of theWVAM Method for BlueWater Navigation

The outcome of the above analysisshows that the WVAM method cap-tures the water velocity up to around1–1.5 m s−1 with an acceptable accura-cy. Therefore, it can be used to aid theINS navigation solution for the situa-


tions discussed in Figure 1 without fur-ther improvements. However, thethreshold of the WVAM method inmeasuring velocity components inthe y and z directions can be improvedfor higher turbulent environments by amore accurate estimation of the vehiclehydrodynamic coefficients withintheir linear as well as nonlinear ranges.Captive model experiments and com-putational fluid dynamic (CFD) simu-lations are capable of determining thenonlinear hydrodynamic coefficientsof AUVs with a greater accuracy; albeitthe associated experimental costs andthe computational times are higher(Randeni et al., 2015c). The coeffi-cients obtained from such methodsare only valid for the vehicle configura-tion that was tested during the experi-ments. However, the arrangements ofmodular AUVs change with the mis-sion and addition/removal of payloadsoccurs frequently. As it is not feasibleto conduct cost-intensive experimentsand simulations for each alteration, es-timating the nonlinear hydrodynamiccoefficients using a system identificationmethod such as least squres optimiza-tion is more suitable (Ljung, 1998).

Ascend and Descend Parallelto Sea Currents

Gavia-class AUVs are under-actuated vehicles (i.e., they have alower number of actuators than thevehicle’s DOF). Its actuator control islimited to the surge speed, roll angle,pitch angle, and yaw angle of the vehi-cle. Depth and transverse displacementis obtained by changing the pitch andyaw angles, respectively. Generally, un-deractuated underwater vehicles are dif-ficult to control when external forcessuch as currents are acting parallel tothe underactuated directions (Aguiar& Pascoal, 2007). For example, thecontrol system of theGaviaAUV strug-

l

gles to maintain the trajectory whencrosscurrents are acting perpendicularto the AUV ’s motion.

In such situations, the vehicle tendsto oscillate, resulting in higher unbounddrifts in the INS navigation solutioncompared to normal operations, unlessproperly counteracted with the velocityover ground measurements from theDVL bottom track. These drift fluc-tuations are considerably less when thevehicle is traveling in line with the cur-rents (i.e., currents are acting along thex axis of the AUV). A positive outcomeof this study is that water column veloc-ities in x direction obtained from theWVAM method were accurate for allturbulence levels present during thefield campaign in the Tamar estuary.The results suggest that deep water as-cents and descents should be carriedout with the vehicle in line with the cur-rents to improve the accuracy of theWVAM-INS navigation solution (i.e.,as seen in Figure 5, the accuracy is high-er in the flow measurements along thex direction compared to y direction),although this may be challengingin unknown environments. Occasion-ally, a reconnaissance mission may beconducted over the test site prior tothe deep dive to ensure the safety ofthe vehicle and to adjust instrumentsettings, for instance, sonar parametersof the multibeam unit. In such cases,the main flow direction could be sim-ply and accurately determined usingthe WVAM method during this mis-sion, and an algorithm can be createdto autonomously decide the directionof ascent/descent.

Velocity OvergroundMeasurements for theWVAM Method

During the vehicle operations out-lined in Figure 1, continuous velocitymeasurements of the AUV relative to

the ground are not obtainable fromDVL bottom tracking to assist theINS navigation solution. Therefore,the INS could be aided with velocityestimates from secondary instrumentsor methods, although this does notrepresent typical operations. An accu-rate simulation model is able to predictthe vehicle velocities in real time whenthe control commands of the AUV areprovided (Hegrenæs & Hallingstad,2011). The hydrodynamic coefficientsof simulation models generally repre-sent the calm water operational condi-tion of the AUV. Therefore, vehicledrift resulting from currents in thewater column will not be included inthe velocity predictions, causing in-accuracies in the AUV position esti-mated by the INS (Augenstein &Rock, 2008; Hegrenæs & Berglund,2009). The results discussed in theabove sections show that the WVAMmethod could be successfully utilizedto determine the water column veloci-ties that can be used in conjunctionwithmodel-predicted vehicle velocitiesto aid the INS navigation solution.

A disadvantage of the WVAMmethod is that it requires an estimationof the vehicle velocity relative to theground in order to measure the watercolumn velocities. The position andvelocity estimates of the vehicle can bedetermined using acoustic transponders(i.e., ultra short baseline or long base-line), albeit with an infrequent updaterate. The intermittent velocity updatesfrom the transponders will be used fortheWVAM to determine the water col-umn velocities, and the estimated flowmeasurements will then be assumed toremain constant until the next reliablevelocity update is received.

As an alternative to the velocity es-timates from acoustic transponders,the velocity of the AUV relative tothe ground can be computed in a prop-

agative manner starting from the initialmeasurements taken at the beginningof the process when DVL bottomtrack or GPS data are available. Thatis, the initial vehicle velocity measure-ments will be used to determine thewater column velocities using theWVAM method and will be used inconjunction with the velocities pre-dicted from the simulation model torecalculate the vehicle velocities rela-tive to the ground. The initially mea-sured and calculated vehicle velocitieswill be compared, and, if the correla-tion is satisfied, the computation willbe iteratively updated by obtainingthe velocity solution of the currenttime stamp using the previous timestamp’s velocity information. How-ever, further measures should be takento reduce the error accumulation. TheINS navigation solution is more robustand less prone to errors when velocityoverground readings from various sen-sors and estimates are considered, forexample, by using a Kalman filter,since the signal outputs from indi-vidual sensors can be discontinuous(Hegrenæs & Hallingstad, 2011).Thus, the INS aidedwith the vehicle ve-locity estimates determined from theproposed method is expected to pro-vide a better localization performance.

ConclusionsThe WVAM method is a non-

acoustic technique to determine thewater velocity components of a tur-bulent water column using the motionresponse of an AUV. The accuracy ofthe WVAM method was examinedat different turbulence levels of thewater column. Nine AUV runs wereconducted along the same track lineat different times in the tidal cyclein the Tamar estuary in Tasmania,Australia. Typically, when an AUV

March/A

undertakes missions in rough waterenvironments, the pitch and yaw an-gles of the vehicle fluctuate aroundthe target values due to the inabilityof the AUV’s dynamic controller toadequately compensate for the externaldisturbing forces. The greater the tur-bulence level of the water, the largerthe fluctuations. The estimated watervelocity components in the y and z di-rections using the WVAM methodagreed well with the experimental mea-surements obtained from the AUV ’sonboard ADCP for low turbulent con-ditions (with a vertical water veloc-ity range of around −1 to +1 m s−1),where the averaged deviations of thevehicle’s pitch and yaw angles arebelow 7°–8°. The correlation reducedwhen the averaged deviations wereabove 7°–8°. The accuracy of the watervelocity component in the x directionremained generally the same for all theruns regardless of the turbulence levelof the water column.

The hydrodynamic coefficients forthe simulation model utilized in theWVAM method were determinedusing a curve fitting technique. Theseestimated coefficients were only validfor small angles of incidence of the ve-hicle, where the coefficients are withintheir linear range. Therefore, as thepitch and yaw angle fluctuations be-come larger, the accuracy of the sim-ulation model decreases adverselyaffecting the prediction of the verticaland transverse water velocity com-ponents obtained from the WVAMmethod. During the AUV missions,the vehicle speed remained within theidentified forward speed range of1.43–2.46 m s−1. Therefore, the simu-lation model was able to provide an ac-curate prediction of the forward speedof the vehicle enabling the WVAMmethod to accurately determine thewater velocities in the x direction.


The WVAM method is capable ofcapturing the velocity up to around 1–1.5 m s−1 with an acceptable accuracy.Therefore, it could be incorporated toaid the INS navigation solutionwithoutfurther improvements for situationswhere the DVL bottom track is inter-mittently or completely unavailableor ineffective. However, the accuracyof the method will be increased by up-grading the simulation model to repli-cate the motion response of the vehiclein both the linear and nonlinear ranges.

AcknowledgmentsThis study was partially funded

through the Research EnhancementGrant Scheme of the University ofTasmania (UTAS). The authors wouldlike to thank Jeff Watts, NathanKemp, and Isak Bowden-Floyd at theAustralian Maritime College, UTAS,for their support during the AUV fieldexperiments. The authors also thankHelgi Þorgilsson (Senior Systems Engi-neer at Teledyne Gavia) and HordurJohannsson (Senior Software Engineerat Teledyne Gavia) for the continuedtechnical support and assistance.

Corresponding AuthorSupun A. T. Randeni P.Locked Bag 1395,Launceston, TAS 7250, AustraliaEmail: [email protected]

ReferencesAguiar, A.P., & Pascoal, A.M. 2007. Dy-

namic positioning and way-point tracking of

underactuated AUVs in the presence of ocean

currents. Int J Control. 80(7):1092-108.

Available at: http://dx.doi.org/10.1080/

00207170701268882.

Augenstein, S., & Rock, S. 2008. Estimating

inertial position and current in the midwater.

Paper presented at the OCEANS 2008.


Quebec City, Canada: IEEE. Available at: http://

dx.doi.org/10.1109/OCEANS.2008.5152057.

Curtin, T.B., Bellingham, J.G., Catipovic, J.,

& Webb, D. 1993. Autonomous oceanographic

sampling networks. Oceanography. 6(3):

86-94. Available at: http://dx.doi.org/10.5670/

oceanog.1993.03.

Devore, J. 2011. Probability and Statistics for

Engineering and the Sciences. Cengage Learning.

Fong, D.A., & Jones, N.L. 2006. Evaluation

of AUV‐based ADCP measurements. Limnol

Oceanogr-Meth. 4(3):58-67. Available at:

http://dx.doi.org/10.4319/lom.2006.4.58.

Grasmueck, M., Eberli, G.P., Viggiano,

D.A., Correa, T., Rathwell, G., & Luo, J.

(2006). Autonomous underwater vehicle

(AUV) mapping reveals coral mound distri-

bution, morphology, and oceanography in

deep water of the Straits of Florida. Geophys

Res Lett. 33:L23616. Available at: http://

dx.doi.org/10.1029/2006GL027734.

Green, S.M. 2015. Tidal Site Characterisation

Using Stationary and AUV-Mounted ADCPs

in a Highly Dynamic Environment. Launceston,

Tasmania, Australia: (Bachelor of Engineering),

Australian Maritime College, University of

Tasmania, Launceston.

Hegrenæs, Ø., & Berglund, E. 2009. Doppler

water-track aided inertial navigation for

autonomous underwater vehicle. Paper presented

at the OCEANS 2009-EUROPE. Bremen,

Germany: IEEE. Available at: http://dx.doi.org/

10.1109/OCEANSE.2009.5278307.

Hegrenaes, O., Berglund, E., & Hallingstad, O.

2008. Model-aided inertial navigation for

underwater vehicles. Paper presented at the

IEEE International Conference on Robotics

and Automation, 2008. ICRA 2008. Pasadena,

California, USA: IEEE. Available at: http://

dx.doi.org/10.1109/ROBOT.2008.4543346.

Hegrenæs, Ø., & Hallingstad, O. 2011.

Model-aided INS with sea current estimation

for robust underwater navigation. IEEE J

Oceanic Eng. 36(2):316-37. Available at:

http://dx.doi.org/10.1109/JOE.2010.2100470.

Hildebrandt, M., & Hilljegerdes, J. 2010.

Design of a versatile AUV for high precision

l

visual mapping and algorithm evaluation. Paper

presented at the Autonomous Underwater

Vehicles (AUV), 2010 IEEE/OES. Available

at: http://dx.doi.org/10.1109/AUV.2010.

5779663. Monterey, California, USA: IEEE

Kim, K., & Ura, T. 2003. Fuel-optimal

guidance and tracking control of AUV under

current interaction. Paper presented at the

The Thirteenth International Offshore and

Polar Engineering Conference. San Diego,

California, USA: IEEE.

Lewis, E.V. 1988. Principles of Naval Archi-

tecture: Motions in Waves and Controllability

(Vol. 3). Jersey City, New Jersey, USA: Society

of Naval Architects & Marine Engineers.

Ljung, L. 1998. System identification. New

Jersey, USA: Springer. Available at: http://

dx.doi.org/10.1007/978-1-4612-1768-8_11.

Paull, L., Saeedi, S., Seto, M., & Li, H. 2014.

AUV navigation and localization: A review.

IEEE J Oceanic Eng. 39(1):131-49. Available

at: http://dx.doi.org/10.1109/JOE.2013.

2278891.

Randeni, S.A.T., Forrest, A.L., Cossu, R.,

Leong, Z.Q., & Ranmuthugala, D. 2015a. Esti-

mating flow velocities of the water column using

the motion response of an autonomous under-

water vehicle (AUV). Paper presented at the

OCEANS ’15 MTS/IEEE, Washington, DC.

Randeni, S.A.T., Forrest, A.L., Cossu, R.,

Leong, Z.Q., Zarruk, G.A., & Ranmuthugala,

D. 2015b. Determining the horizontal and verti-

cal water velocity components of a turbulent water

column using the motion response of an autono-

mous underwater vehicle. J Atmos Ocean Tech.

Randeni, S.A.T., Leong, Z.Q., Ranmuthugala,

D., Forrest, A.L., &Duffy, J. 2015c. Numerical

investigation of the hydrodynamic interaction

between two underwater bodies in relative mo-

tion. Appl OceanRes. 51(0):14-24. Available at:

http://dx.doi.org/10.1016/j.apor.2015.02.006.

Simpson, M.R. 2001. Discharge measure-

ments using a broad-band acoustic Doppler

current profiler. Sacramento, California, USA:

U.S. Department of the Interior, U.S. Geo-

logical Survey.

P A P E R

Coupled Modeling of Hydrodynamics andSound in Coastal Ocean for RenewableOcean Energy Development
A U T H O R SWen LongKi Won JungZhaoqing YangAndrea CoppingZ. Daniel DengPacific NorthwestNational Laboratory
A B S T R A C T
The rapid growth of renewable offshore energy development has raised concerns
that underwater noise from construction and operation of offshore devices may in-terfere with communication of marine animals. An underwater soundmodel was de-veloped to simulate sound propagation frommarine and hydrokinetic energy (MHK)devices or offshore wind (OSW) energy platforms. Finite difference methods weredeveloped to solve the 3-D Helmholtz equation for sound propagation in the coastalenvironment. A 3-D sparse matrix solver with complex coefficients was formed forsolving the resulting acoustic pressure field. The complex shifted Laplacian precondi-tioner (CSLP) method was applied to solve the matrix system iteratively with MessagePassing Interface (MPI) parallelization using a high-performance cluster. The soundmodel was then coupled with the Finite Volume Community OceanModel (FVCOM) forsimulating sound propagation generated by human activities, such as construction ofOSW turbines or tidal stream turbine operations, in a range-dependent setting. As aproof of concept, the validation of the solver is tested with an ideal case against twoother methods, and its application is then presented for two coastal wedge problems.This sound model can be useful for evaluating impacts on marine mammals due todeployment of MHK devices and OSW energy platforms.Keywords: sound, coastal ocean, renewable ocean energy, coupling, Helmholtzequation

32 km of the coastline.However, stake-holders and regulators have concerns

Introduction

Renewable energy generated fromrenewable offshore resources—fromwaves, tidal streams, and ocean cur-rents and from offshore wind energy—has the potential to contribute tothe U.S. low carbon “all of the above”portfolio of renewable energy sources.These devices are commonly deployedin estuaries and coastal waters within

that the extraction of marine renewableenergymay pose a risk to the well-beingof marine mammals and fish (Bailey,2012; NRC, 2005). One of the poten-tial risks is from the addition of unnat-ural sound generated by energy devicesthat could injure, disorient, or alter themovements of marine animals. Theeffects of sound from wave and tidalstream devices and sound from piledriving for offshore wind turbines aresome of the most significant concerns.

There have been efforts to under-stand the sound propagation from theanthropogenic noise sources in range-dependent (e.g., varying bathymetry)settings. Ray tracing was used to study

the impact of the noise from the seis-mic air guns on fish behavior (Hovemet al., 2012) and to estimate the re-ceived sonar signals by the northernbottlenose whales (Miller et al., 2015).Parabolic equation (PE) was also uti-lized to estimate the sound exposurelevel to seals around the pile drivinglocation near a wind farm constructionsite (Hastie et al., 2015). DeRuiteret al. (2006) used both ray tracingand PE to model acoustic propagationof air-gun array pulses and comparedthe results with the recorded signalsby tag-equipped sperm whales. Al-though a range dependency was con-sidered in terms of bathymetry and a

March/A

speed of sound profile, it was mostly2-D variation based on axis symmetry,and none of the previous works consid-ered the 3-D heterogeneity in salinity,temperature, sound speed, and bathym-etry. This becomes important whenthere are river plumes in the coastalarea of interest, because the freshwatercan alter both salinity and temperature,both of which affect sound speed inwater.

Sound transmission is closely relatedto ocean water density, which is in turndependent on water temperature, sa-linity, and depth (pressure) (Lurton,2002). Typical vertical density stratifi-cation due to temperature and salinity


in the ocean was described by Talley et al. (2011). A simple regression of the data provides sound speed C (m/s) as (Medwin,1975)

C ¼ 1449:2þ 4:6T -0:055T 2 þ 0:00029T 3 þ 1:34-0:01Tð Þ S-35ð Þ þ 0:016D ð1Þ

where S is salinity in practical salinity unit (PSU), T is temperature in degrees Celsius, and D is depth in meters. For typicaldeep ocean areas, the main contribution to variation of sound speed is due to temperature stratification and pressure (depth).This relationship sets up a minimum speed at around 700-m depth. As an example, sound generated at about 1,100-m waterdepth propagates at small angles around the horizontal direction and is refracted up and down around the same depth level,resulting in horizontal propagation over very long distances. However, in coastal zones only with a consideration of depth, thisvariability in C becomes less than 5 m/s because of its shallow depth less than 200 m. Instead, in coastal zones, where salinitycould vary from 0 to 35 PSU, the variation ofC due to salinity could be as much as ±50m/s.Moreover, sound transmission incoastal waters is temporally and spatially more variable than that in the open ocean, for example, due to seasonal influences ofriver plumes, effects of meteorological wind stress and heat fluxes, acoustic scattering from internal waves, and reflections ofsound from complex coastal geometries. An advanced model can be used to account for this variability to accurately predictsound fields. However, coupled hydrodynamic and acoustic models have not been developed because of the intensive com-putational resources needed.

We improved the ability to predict sound propagation in coastal regions by coupling of the 3-D ocean hydrodynamiccirculationmodel FVCOM (Finite VolumeCommunityOceanModel; Chen et al., 2003, 2006, 2007; Lai et al., 2010; Yanget al., 2015), which resolves the variations of water density and sea surface elevation, with our 3-D acoustic model that solvesunderwater sound transmission loss (TL), as well as reflection from lateral, bottom, and surface boundaries and refraction dueto varying density fields.

Hydrodynamic ModelFVCOM solves the continuity and momentum equations using the finite-volume method and sigma-stretched coordi-

nate transformation in the vertical direction.

Formulation of Hydrodynamics in FVCOMThemomentum equations of FVCOM in the horizontal directions have the general form (Chen et al., 2003, 2006, 2007;

Lai et al., 2010) of

∂uD∂t

þ ∂u2D∂x

þ ∂uvD∂y

þ ∂uω∂σ

� fvD ¼ � Dρo

∂p∂x

þ 1D

∂∂σ

Km∂u∂σ

� �þ DFx ð2Þ

∂vD∂t

þ ∂uvD∂x

þ ∂v2D∂y

þ ∂vω∂σ

þ fuD ¼ � Dρo

∂p∂y

þ 1D

∂∂σ

Km∂v∂σ

� �þ DFy ð3Þ

where x, y, and σ are the east, north, and vertical axes in the sigma coordinates; u, v, and ω are the three velocity com-ponents in the x, y, and σ directions; D is the total water depth; Fx and Fy are the horizontal momentum diffusivity termsin the x and y directions; Km is the vertical eddy viscosity coefficient; ρ is water density; p is pressure; and f is the Coriolisparameter. σ is the stretched vertical coordinate system based on an algebraic transformation from vertical Cartesiancoordinate z.

σ ¼ z � ξD

; ð4Þ


with ξ(t, x, y) being the surface elevation andD¼ hþ ξ, where h(x, y) is still water depth. The surface and bottom boundariesare corresponding toσ¼ 0 andσ¼�1, respectively. Similarly, the temperatureT(x, y,σ, t) and salinity S(x, y,σ, t) are solvedin 3-D using the transport equations below.

∂TD∂t

þ ∂uTD∂x

þ ∂vTD∂y

þ ∂ωT∂σ

¼ 1D

∂∂σ

KT∂T∂σ

� �þ DbH þ DFT ð5Þ

∂SD∂t

þ ∂uSD∂x

þ ∂vSD∂y

þ ∂ωS∂σ

¼ 1D

∂∂σ

KS∂S∂σ

� �þ DFS ð6Þ

where bH(x, y) is the heating source/sink for a given vertical location; FT and FS are the horizontal turbulent eddy diffusionterms for temperature and salinity, respectively; KT and KS are for vertical eddy diffusivities of temperature and salinity.The above equations are solved using finite volume integration based on spatial discretization with triangular meshes inhorizontal plane (x, y) and fixed σ layers in vertical direction. For derivations of the discretized system, readers are referredto the publications on FVCOM (Chen et al., 2003, 2006, 2007; Lai et al., 2010) for details.

Application of FVCOM for a Coastal Wedge Case with a River PlumeA typical coastal wedge (idealized coastal domain with a wedge shape for the water part) application is depicted in Figure 1.

The water depth at the shoreline (right-hand side) is 2 m, and the shoreline is about 2 km long. The cross-shore bathymetryis described as a piece-wise linear sloping bottom with maximum depth at the open boundary being 44 m. A river dischargeof 150 m/s is placed at the center of the coastline, delivering a freshwater plume to the coastal zone. For such a relativelysimple geometry, an unstructured grid mesh is chosen such that all the triangular nodes collapse into a structured mesh. TheFVCOMmodel was run for 39 days of simulation with time step Δt¼ 1 s from initially evenly distributed salinity of 30 PSUand quiescent flow field (i.e., (u, v, ω) ¼ (0, 0, 0)). Salinity distribution at the last hour of the 39-day simulation is plottedusing colored contours (Figure 1). For simplicity, Coriolis forcing is set to zero, and temperature simulation is disabled.

Coastal Acoustic ModelMathematical Description of Sound in Coastal Oceans

In Cartesian coordinates (x, y, z), the pressure disturbance caused by sound in ocean p(x, y, z, t) is governed by the 3-Dwave equation below.

∂2p∂t2

� C2∇2p ¼ F x; y; z; tð Þ ð7Þ

FIGURE 1

Coastal wedge simulation of river plume. (Color version of figures are available online at: http://www.ingentaconnect.com/content/mts/mtsj/2016/00000050/00000002.)


whereC(x, y, z) is the sound speed (unit: m/s) and F is the sound source strength. For a given frequency f, the pressure p can bewritten in harmonic form (using complex numbers, and the physical pressure will always be the real part of the complexnumber)

p x; y; z; tð Þ ¼ bp x; y; zð Þe iωt ð8Þ

Substituting this harmonic equation into equation (7), the Helmholtz equation is obtained as

∇2bp þ k2bp ¼ � 1C2

bF ð9Þ

where F ¼ bFeiωt and ω ¼ 2πf is the angular frequency and k x; y; zð Þ ¼ωC is the angular wave number. In coastal oceans, C

and k are spatially variable due to the variation of temperature and salinity. For applications with attenuation, k2 is replaced by(1 � αi)k2, where α is the attenuation coefficient and i is the notation for the imaginary part of a complex number.

Finite Difference FormulationFor implementation with a structured grid system of constant grid spacing Δx, Δy, Δz in a 3-D regular box-shaped do-

main, denoting bp as indexed by (i, j, l ) in x, y, and z directions, equation (9) can be discretized using the finite differencemethod as

bpi�1; j;l � 2bpi; j;l þ bpiþ1; j;l

� �Δx2

þbpi; j�1;l � 2bpi; j;l þ bpi; jþ1;l

� �Δy2

þbpi; j;l�1 � 2bpi; j;l þ bpi; j;lþ1

� �Δz2

þ k2bpi; j;l ¼ � 1C2

bFi; j;l ð10Þ

where i = 2…,M� 1, j¼ 2…,N� 1, l¼ 2…, L� 1 andM,N, and L are the number of grid points in x, y, and z directions,respectively. In addition, the surface boundary condition is bpi; j;l¼1 ¼ 0. Equation (10) can also be arranged in the followingform for all bp values.

Abp ¼ b ð11Þ

where A is an M × N × L by M × N × L sparse matrix.

Iterative Helmholtz Solver Based on Complex Shifted Laplacian Preconditioner MethodThe matrix A in equation (11) is typically a very large sparse matrix of complex numbers. For high-frequency sound and

large domain, the matrix equation (11) becomes difficult to solve directly. The Complex Shifted Laplacian Preconditioner(CSLP) method by Vuik et al. (2003) with either Bi-CGSTAB (Van de Vorst, 1992) or GMRES (Saad & Schultz, 1986) isused for this research. The CSLP method is based on construction of a preconditioning matrix MP

MP ¼ B þ β1 � iβ2ð Þk2I ð12Þ

where I is the unit matrix and B ¼ A � k2I. Instead of solving equation (11), the following left preconditioned equation issolved for bp.

M�1P Abp ¼ M�1

P b ð13Þ

β1 and β2 have been optimized for the best convergence rate. In our simulations, we used β1 ¼ 0 and β2 ¼ 0.5. MP�1 was

simply calculated using an SOR (Young, 1971) iteration method from MP. To solve for bp from equation (13), the Krylov


subspace methods called Bi-CGSTAB(Van de Vorst, 1992) or GMRES(Saad & Schultz, 1986) can be used.These two methods are implementedin the PETSc parallel computationlibrary (version 3.6.0) (Balay et al.,2015).

The 3-D wedge case with heteroge-neous wave number (Erlangga, 2005)is tested for the speed of computationwith different wave numbers and num-ber of grids on a high performancecluster with 16 CPUs (Intel XeonCPU E5-2670 v3, 2.3 GHz). The re-sults are given in Table 1. We foundthat (β1, β2)¼ (0, 0.5) gave significantlybetter results than (β1, β2)¼ (1, 0.5) thatwas reported in Erlangga et al. (2006).

Radiation Boundary TreatmentThe radiation boundary is needed

for open-domain simulations. How-ever, because an imperfect radiationcondition can reflect in the simulationdomain of interest, a treatment thatdoes not return the sound wave shouldbe applied. This is implemented usingthe perfectly matched layer (PML)method (Chew & Weedon, 1994).Typically, this is done by placing50 grids at the open boundary to ab-sorb outgoing sound waves with alarge damping term. Within the PMLzone, k2 in equation (9) is replaced by

(1� iα)k2, where α is the attenuationfactor in the PML zone, which isramped up from zero at the physicaldomain boundary to 0.475 at the inte-rior end of the PML layer. The externalboundaries of the PML layers are ter-minated by the first-order Sommerfeldradiation boundary condition, defined

asd bpdn

¼ ikbp, where n is the direction

normal to the boundary.Figure 2 depicts the bp solution (real

part) of a 2-D simulation using the

March/A

PML layer in a homogeneous (uni-form sound speed) domain with openboundaries. This demonstrates thatour implementation of the PML layershas effectively removed artificial reflec-tions from boundaries.

Validation of AcousticPropagation Model

The finite difference scheme (equa-tion (10)) along with the CSLPmethod(equation (13)) is tested on a wedgediscussed by Lin et al. (2012) for vali-dation. As the geometry is illustrated inFigure 3, the slope angle of the wedgeis 2.86°, and the cross-shore distanceis 4 km. The resulting water depth is200m at 4 km away from the shoreline.Here, because the validation is onlyfocused on the TL across the wedge,the width of the wedge is reduced to600m. The water is assumed to be uni-form in density with the sound speedC ¼ 1,500 m/s. While the density ofthe seabed is assumed to be identical

TABLE 1

The test result is shown in terms of the iteration number (the running time is in the parentheses) asa function of wave number. The test was done using 16 CPUs and the restart value of 20 in GMRES.

Wave Number [1/m](# of Grid Points)

(β1, β2)
(β1, β2)
(1, 0.5)
(0, 0.5)
20 (35,937)
1,157 (0.432 s) 710 (0.257 s)
40 (274,625)
2,196 (7.62 s) 1,253 (4.51 s)
60 (912,673)
3,629 (49.1 s) 1,842 (26.1 s)
80 (2,146,689)
5,187 (175 s) 2,470 (84 s)
200 (33,076,161)
25,677 (17640 s) 6,318 (4160 s)
FIGURE 2

2-D benchmark testing of PML for absorbing sound waves at the boundary. The real part of thepressure is shown in a 2-D homogeneous medium with the wave number of 0.4189 (1/m) withPML (not shown). The number of grid points per wavelength is set at 10. The point source isplaced on the top line of the figure.


to that of the water, the seabed is mod-eled with a uniform sound speed C ¼1,700 m/s. The attenuation in theseabed is set as 0.5 dB/wavelength. Apoint sound source with the frequencyf ¼ 25 Hz is placed at x ¼ �4,000 m,y ¼ 0 m, and z ¼ �100 m (shown asthe red sphere in Figure 3). In orderto minimize the reflections from theboundaries surrounding the propaga-tion medium, PML zones are addedto the bottom and lateral boundaries.The grid sizes are defined as Δx andΔy ¼ 5 m and Δz ¼ 2.5 m.

Figure 4 shows the TL defined as

TL ¼ �20log10j bpj ð14Þ

at z¼�30 m (70 m above the source)on the plane perpendicular to theshoreline as a function of distancefrom the source (i.e., along the lineAB in Figure 3). Here, the pressureamplitude jbpj is calibrated to become1 Pa at the distance of 1 m from thesource. For comparison, solutionsusing a PE-based RAMGEO (Collins,1993) and using the image method(Deane&Buckingham, 1993) are pre-


sented as well. Figure 4 suggests thatthe acoustic propagation model basedon equation (10) predicts the TL acrossthe wedge slope correctly.

Application in IdealizedCoastal Wedge Examples

The application of the acousticpropagation model based on 3-DHelmholtz equation with the CSLP

l

method is presented for two idealizedcoastal wedge examples.

3-D ASA WedgeThe first example is for an idealized

coastal water domain, which consistsof shoreline coastal water confined ina wedge shape. Lin et al. (2012) dis-cussed (called ASA wedge here) simu-lating the sound in such a domainusing several PE-based methods. Here,a similar wedge domain is constructedand simulated. Figure 5 shows thestraight wedge-shaped coastal zonewith a shoreline depth of zero and anoffshore depth of 200 m, a cross-shore distance of 4 km and an along-shore distance of 20 km. Differentfrom the wedge case used for the valida-tion, a unit sound source of frequencyf ¼ 25 Hz is placed at x ¼ �2,000 m,y¼ 0m, and z¼�50m (shown as thered sphere in Figure 5). The water isassumed to be uniform in density witha sound speed C ¼ 1,500 m/s; theseabed is also depicted and extends toz ¼ �200 m with the sound speedbeing C ¼ 1,700 m/s. PML zones are

FIGURE 3

Geometry and sound speed profile of ASA wedge for the validation of the acoustic propagationmodel. The sound source location is indicated by a sphere below point A.

FIGURE 4

The TL curve along the line AB at z = −30 m. For comparison, the TL curves based on the imagemethod (Deane & Buckingham, 1993) and a PE-based RAMGEO (Collins, 1993) are also presented.

added for all lateral sides and bottom(not shown) for the simulation.

Figure 6 shows the computed TLat z¼ �15 m, that is, 15 m below sur-face. In order to obtain the TL valuedirectly, bp is scaled by pressure at thedistance of 1m from the acoustic source.This results in a significant reflectionfrom the shoreline. Figure 7 shows thevertical transect view of the xz plane aty ¼ 0; the sound penetrates into theseabed below the sound source. Thereis also considerable penetration on theshoreline, whereas the area below theseabed offshore is sheltered.

Figure 8 shows the TL along lineCD of Figure 5, which is located 15 mbelow the surface.

Coastal Wedge witha River Plume

The second example includes acoastal wedge with a river plume. TheFVCOM model simulation of thecoastal wedge (Figure 1) providedthe salinity field at a constant tem-perature of 20°C. The grid size of thehydrodynamic model in Figure 1 is62.5 m in horizontal plane with 10vertical layers used for the water col-umn discretization. We interpolated

March/A

the temperature and salinity fieldsonto finite difference grids to calculatethe sound speed based on equation (1).A unit sound source of frequencyf ¼ 100 Hz is placed 5 m below thesurface. The finite difference schemeformulation (equation (10)) is appliedwith a uniform grid size of Δx, Δy, andΔz ¼ 1.25 m for the domain with theseabed extending below the maximumwater depth (44 m deep at the openboundary) to z ¼ �144 m with asound speed C ¼ 2,500 m/s belowseabed. There are also 60 PML gridsadded in all lateral boundaries andthat extended below the seabed forthis test case. The surface is treated asa pressure-release boundary condition(zero acoustic pressure). The simula-tion is carried out using 25 parallelcomputer nodes, each with 24 cores.The model run-time is about 4 h forthe simulation. Figure 9 shows theplane (xy) view of TL at z ¼ �5 m,that is, the same depth as the source

FIGURE 6

TL of xy plane at z ¼ �15 m.

FIGURE 5

Geometry and sound speed profile of ASA wedge; sound source location is indicated by a spherebelow point C.

FIGURE 7

TL of xz plane results at y ¼ 0 m; the dark line indicates the seabed.

FIGURE 8

TL along line CD at z ¼ �15 m (35 m above source).


for the sound speed field correspond-ing to the last hour of hydrodynamicsimulation in day 39 (Figure 1). Fig-ure 10 shows the xz plane view aty¼1,000m at themiddle of the shore-line. Similar to the wedge shown inFigure 7, the sound penetrates intothe seabed. There are multiple reflec-tions along the shoreline, ocean surface,and seabed, which form standing wavepatterns. Figure 11 shows the differ-ence of TL between the fully developedriver plume condition at the end ofday 39 and the initial uniform condi-tions (T ¼ 20°C and S ¼ 30 PSU) atz ¼ �5 m. The presence of the riverplume changes the sound field signifi-cantly in the far-field.

Conclusion andFuture Work

An underwater sound model usinga finite difference method was devel-oped to simulate sound transmissionin the coastal ocean. The hydrody-namic model FVCOM was used tosimulate the varying distributions ofdensity in coastal waters and to calcu-late the speed of sound in the coastalenvironment. The CSLP method wasused for fast iteration to solve the re-sulting large and sparse matrix systemon a parallel cluster. The validation offinite difference has been presented


using an ASA benchmark wedge interms of the TL as a function of thedistance from the source along theupslope direction of the wedge, andthe model was applied to two wedgeexamples.

l

This underwater sound model,forced by sound speed predictionsfrom a hydrodynamic model, can beuseful for evaluating impact on marinemammals due to construction of off-shore wind (OSW) turbines and de-ployment and operation of marineand hydrokinetic energy devices incoastal regions. Pile driving for instal-lation of OSW turbines may causesignificant underwater noise that willdisturb marine mammals includinghighly endangered whales. Similarly,operational sounds from tidal andocean current turbines may disturbmarine mammals and fish that are af-forded special regulatory protection.The availability of a linked acoustic-hydrodynamic model that easily

FIGURE 9

TL at z ¼ �5 m based on hydrodynamic prediction of sound speed field at the end of 39-daysimulation (Figure 1).

FIGURE 10

TL for the xz plane at y ¼ 1,000 m based on hydrodynamic prediction of sound speed field at theend of 39-day simulation.

FIGURE 11

Difference of TL between using hydrodynamic predictions of sound field at the end of 39-day sim-ulation (with the river plume) and the initial state (T¼ 20°C and S¼ 30 PSU) without a river plumeat z ¼ �5 m.

provides a solution for underwateracoustic outputs from these activitiescould allay regulatory and stakeholderconcerns as well as allow rapid re-sponse for installers to mitigate noisethat may be harmful or enable engi-neers to design turbines that produceless sound.

We are further developing themodel using finite volume formula-tions of the Helmholtz equationswith sigma coordinates that are com-patible with the FVCOM model,followed by validation of the modelusing field observations in nearshorecoastal waters in Puget Sound. Inaddition, in order for this model tobe of practical use, the characteristicsof acoustic sources and its effect onthemodel need to be fully investigated.For instance, the frequency range froma pile driving, which is between 100 and1,000 Hz (Erbe, 2009; Bailey et al.,2010), can cause a large burden to thecomputational cost because the numberof grids is proportional to the frequencycubed. Therefore, an effort needs to bemade to optimize a trade-off betweenthe fast running time and the amountof required computational resource.

AcknowledgmentsThis study was funded by the Pacif-

ic Northwest National Laboratory’sLaboratory Directed Research and De-velopment program under projectnumber #66978. Drs. Taiping Wangand Genevra E. Harker from PNNLprovided internal review of this manu-script. We appreciate Dr. ZhongxiangZhao from the Applied Physics Labo-ratory and one anonymous scientistwho served as external reviewers andhelped with their insights and proof-reading. We also want to express grat-itude to Dr. C. Vuik’s group at Delft

University of Technology for provid-ing sample code with 2-D examplesand email communications.Mr. SatishBalay from Argonne National Labora-tory was very helpful with the usage ofPETSc library for high-performancecomputing.

Corresponding Author:Ki Won JungPacific Northwest National LaboratoryP.O. Box 999, Richland, WA 99352Email: [email protected]

ReferencesBailey, H., Senior, B., Simmons, D., Rusin,

J., Picken, G., & Thompson, P.M. 2010.

Assessing underwater noise levels during pile-

driving at an offshore windfarm and its po-

tential effects on marine mammals. Mar Pollut

Bull. 60(6):888-97. http://dx.doi.org/10.1016/

j.marpolbul.2010.01.003.

Bailey, H. 2012. Impacts of wind turbine

construction on marine mammals: what can

be done? Int Sustain Energ Rev. 6(1):34-7.

Balay, S., Abhyankar, S., Adams, M.F.,

Brown, J., Brune, P., Buschelman, K., …

Zhang, H. 2015. PETSc Users Manual,

ANL-95/11—Revision 3.6.

Chen, C., Liu, H., & Beardsley, R.C. 2003.

An unstructured, finite-volume, three-

dimensional, primitive equation ocean model:

application to coastal ocean and estuaries.

J Atmos Ocean Tech. 20:159-86. http://

dx.doi.org/10.1175/1520-0426(2003)020

<0159:AUGFVT>2.0.CO;2.

Chen, C., Beardsley, R.C., & Cowles, G.

2006. An unstructured grid, finite-volume

coastal ocean model: FVCOM user manual.

School for Marine Science and Technology,

University of Massachusetts Dartmouth.

315 pp.

Chen, C., Huang, H., Beardsley, R., Liu, H.,

Xu, Q., & Cowles, G. 2007. A finite volume

numerical approach for coastal ocean circula-

March/A

tion studies: Comparisons with finite differ-

ence models. J Geophys Res. 112:C03018.

doi:10.0129/2006JC003485.

Chew, W.C., & Weedon, W.H. 1994. A 3d

perfectly matched medium from modified

Maxwell’s equations with stretched coordinates.

Microw Opt Techn Let. 7(13):599-604. http://

dx.doi.org/10.1002/mop.4650071304.

Collins, M.D. 1993. A split‐step Padé

solution for the parabolic equation method.

J Acoust Soc Am. 93(4):1736-42. http://

dx.doi.org/10.1121/1.406739.

Deane, G.B., & Buckingham, M.J. 1993.

An analysis of the three‐dimensional sound

field in a penetrable wedge with a stratified

fluid or elastic basement. J Acoust Soc Am.

93(3):1319-28. http://dx.doi.org/10.1121/

1.405417.

DeRuiter, S.L., Tyack, P.L., Lin, Y.T.,

Newhall, A.E., Lynch, J.F., & Miller, P.J.

2006. Modeling acoustic propagation of air-

gun array pulses recorded on tagged sperm

whales (Physeter macrocephalus). J Acoust

Soc Am. 120(6):4100-14. http://dx.doi.org/

10.1121/1.2359705.

Erbe, C. 2009. Underwater noise from pile

driving in Moreton Bay, Qld. Acoust Aust.

37(3):87-92.

Erlangga, Y.A. 2005. A robust and efficient

iterative method for the numerical solution

of the Helmholtz equation. Ph.D. thesis, Delft

University of Technology. 126 pp.

Erlangga, Y.A., Oosterlee, C.W., & Vuik, C.

2006. A novel multigrid based preconditioner

for heterogeneous Helmholtz problems.

SIAM J Sci Comput. 27(4):1471-92. http://

dx.doi.org/10.1137/040615195.

Hastie, G.D., Russell, D.J., McConnell, B.,

Moss, S., Thompson, D., & Janik, V.M.

2015. Sound exposure in harbour seals during

the installation of an offshore wind farm:

predictions of auditory damage. J Appl Ecol.

52(3):631-40. http://dx.doi.org/10.1111/

1365-2664.12403.

Hovem, J.M., Tronstad, T.V., Karlsen, H.E.,

& Lokkeborg, S. 2012. Modeling propagation


of seismic airgun sounds and the effects on

fish behavior. IEEE J Oceanic Eng. 37(4):

576-88. http://dx.doi.org/10.1109/JOE.2012.

2206189.

Lai, Z., Chen, C., Colwes, G.W., & Beardsley,

R.C. 2010. A nonhydrostatic version of

FVCOM, 1. Validation experiments. J Geophys

Res-Oceans. 115(C11). http://dx.doi.org/

10.1029/2009JC005525.

Lin, Y., Colis, J.M., & Duda, T.F. 2012. A

three-dimensional parabolic equation model

of sound propagation using higher-order

operator splitting and Pade approximants.

J Acoust Soc Am. 132(5):EL364-70. http://

dx.doi.org/10.1121/1.4754421.

Lurton, X. 2002. An Introduction to Under-

water Acoustics. London: Springer. 347 pp.

Medwin, H. 1975. Speed of sound in water:

A simple equation for realistic parameters.

J Acoust Soc Am. 58(6):1318-9. http://dx.doi.

org/10.1121/1.380790.

Miller, P.J.O., Kvadsheim, P.H., Lam, F.P.A.,

Tyack, P.L., Curé, C., DeRuiter, S.L., …

Hooker, S.K. 2015. First indications that

northern bottlenose whales are sensitive to be-

havioural disturbance from anthropogenic

noise. Roy Soc Open Sci. 2(6):140484. http://

dx.doi.org/10.1098/rsos.140484.

National Research Council. 2005. Marine

Mammal Populations and Ocean Noise: De-

termining When Noise Causes Biologically

Significant Effects, Committee on Character-

izing Biologically Significant, Marine Mammal

Behavior. Available at: http://www.nap.edu/

catalog/11147.html.

Saad, Y., & Schultz, M.H. 1986. GMRES:

A generalized minimal residual algorithm for

solving nonsymmetric linear systems. SIAM J

Sci Stat Comp. 7:856-69. http://dx.doi.org/

10.1137/0907058.

Talley, L.D., Pickard, G.L., Emery, W.J.,

& Swift, J.H. 2011. Descriptive Physical

Oceanography, Sixth Edition. Boston, MA:

Elsevier. 560 pp.

Van de Vorst, H.A. 1992. Bi-CGSTAB: a fast

and smoothly converging variant of Bi-CG

for solution of non-symmetric linear systems.


SIAM J Sci Stat Comp. 13:631-44. http://

dx.doi.org/10.1137/0913035.

Vuik, C., Erlangga, Y.A., & Oosterlee, C.W.

2003. Shifted Laplace preconditioners for the

Helmholtz equations. Report# 03-18, ISSN

1389-6520, Delft University of Technology.

Yang, Z., Wang, T., Voisin, N., & Copping,

A. 2015. Estuarine response to river flow and

sea-level rise under future climate change

and human development. Estuar Coast Shelf S.

156:19-30. http://dx.doi.org/10.1016/j.ecss.

2014.08.015.

Young, D.M. 1971. Iterative Solution of

Large Linear Systems. New York: Academic

Press. 570 pp.

l

P A P E R

Development and Application of a NewMobile pH Calibrator for Real-TimeMonitoring of pH in Diffuse FlowHydrothermal Vent Fluids
A U T H O R SChunyang TanKang DingWilliam E Seyfried, JrUniversity of Minnesota
A B S T R A C T
In situmeasurement of pH in diffuse flow hydrothermal vent fluids is necessary to
investigate the feedback between geochemical and biochemical processes. AccuratepH determination has been unusually challenging owing to temperature and pressureeffects that place severe constraints on the performance of a wide variety of pH sensorsystems. In this paper, we describe a newly developed mobile pH calibrator (MpHC),whichmakes use of in situ calibration protocols that enhance the accuracy of pHmea-surement and monitoring on the ocean floor at deep-sea hydrothermal vents. TheMpHC combines the physically robust and highly sensitive iridium solid-state pHelectrode with a flow control system to perform 2-point calibration with on-boardpH buffer solutions. The small size and novel design of the sensor probe allowmore effective access to seafloor hydrothermal vent fluids and their associated sulfidestructures and biological communities. TheMpHC is capable of in situ deployment bysubmersible via ICL (inductively couple link) communication around hydrothermalvents at pressures and temperatures up to 45 MPa and 100°C, respectively. In thispaper, we also present results of in situ calibration methods used to correct the stan-dard potential and slope (mV/pH) of the solid-state electrode for temperature effects.The MpHC has been deployed most recently using the submersible Alvin duringcruise AT26-17 to Axial Seamount and Main Endeavour Field, Juan De Fuca Ridgein the NE Pacific. With in situ calibration functionality, the MpHC offers the prospectofmore successful longer-termmeasurements in keeping with power availability pro-vided by cabled seafloor observatories coming online in the NE Pacific.Keywords: in situ, pH, calibration, hydrothermal, electrode

and their environments, which aredominated by mixing between cold

Introduction

I t is widely recognized that theinvestigation of the geochemical andbiochemical processes of hydrothermalvent systems relies on constrainingkey chemical and physical componentsof vent fluids (Childress et al., 1991;Von Damm, 1995). In this regard,pH is considered especially importantbecause of the role it plays in the solu-bility of minerals, with direct impli-cations for the temporal evolution ofseafloor sulfide deposits. Moreover,tracing pH variation and distributioncan help scientists understand the in-teractions between the vent organisms

oxygen bearing seawater and highlyreduced hydrothermal fluid (Sarrazinet al., 1999; Von Damm & Lilley,2004). Thus, the need is great for thedevelopment and application of in situinstruments that can measure and mon-itor pH and related components inunusually challenging environmentswith high resolution and accuracy inspace and time (Shank et al., 1998).

In situ electrochemicalmeasurementshave been widely used in chemical

oceanography for decades (Reimers,2007). With advances in electroana-lytical chemistry and microelectronicstechnology, the pH electrode is nowwell suited to operate at high pressureand variable temperatures, potentiallyproviding fundamental insight on com-plex geochemical and biogeochemicalprocesses inherent to hydrothermal ventsystems (Ding & Seyfried, 2007; Prien,2007). To achieve the full potential thatthe new generation of electrochemical

March/A

sensors may provide, however, limita-tions imposed by long-term deploy-ment in challenging and changingchemical and physical environmentsmust be overcome; so too, the robust-ness of the electrodes must also be im-proved, if the full range of deep-seavent applications is to be realized (LeBris et al., 2001).

In the past few years, we devel-oped an in situ pH calibration device(Deployable Calibrator), which could


perform 2-point calibration of pHin situ with pH buffers optimized forspecific applications (Tan et al., 2010,2012, 2014). The in situ calibrationmethod minimizes potential sourcesof drift, enhancing accuracy of pHmeasurements. This instrument wasused more recently to investigate thepH of diffuse flow hydrothermal ventfluid and also modified for use withcabled ocean observatories for real timepH monitoring (Tan et al., 2014). Al-though successful, the size of pH probeand functionality of the sensor systemcould be improved upon to enhancegeochemical applications at deep-seahydrothermal vents. To address theseissues, a so-called mobile pH calibrator(MpHC) was developed. This highlyintegrated pHmeasurement and mon-itoring system was smaller than its pre-decessor, inherently more precise, andmore tolerant of temperature variabil-ity (0–100°C). The calibrator and sup-porting electronics can be easily andeffectively mounted on the basket ofthe submersible, Alvin, facilitatingdeployment at different and diversevent sites. The calibrator was testedduring a series of dives to cold seepsin connection with the Alvin ScienceVerification Cruise (AT26-12) andagain during Alvin/Atlantis operations(AT26-17) at hydrothermal vents atAxial Seamount and theMain Endeav-our Hydrothermal Field (Juan de FucaRidge). Here we emphasize techno-logical modifications and operationalcapabilities inherent to the new electro-chemical system, while also brieflyreviewing results of preliminary datafrom the recent deployments of thecalibrator at NE Pacific vent sites.

Materials and MethodsThe MpHC is designed around

three critical components: (1) a slim


profile (20 mm OD) terminating inthe sensor probe that allows precisepositioning in unusually restrictiveareas, inherent to seafloor hydrother-mal vent applications; (2) simulta-neous pH, redox, and temperaturemeasurement at temperatures up to100°C; and (3) in situ calibration ofthe electrochemical sensors with nopH buffer leakage into the ambientenvironment. In its simplest configu-ration, the MpHC device consists oftwo parts: the sensor probe itself, fordetection, and the main body, whichis comprised of solenoid valves, preci-sion pumps for fluids delivery, andsupporting electronics, all containedwithin an oil-filled pressure housing.

The Flow Control Systemof MpHC

Importantly, the MpHC was spe-cifically designed to make use of aflow control system that can delivermultiple pH buffer solutions neededto calibrate the electrodes in situ (Fig-ure 1). The electrodes (Figure 2) aresealed in the flow cell at one end ofthe probe to facilitate interaction withthe hydrothermal fluids. Subsequentto acquisition of vent fluid pH and

l

temperature data, in situ calibration isactivated. This in situ calibration capa-bility represents one of the most im-portant characteristics of the sensorsystem. In practice, this entails flowof two different pH buffer fluids intothe electrode (sensor probe) bearingflow cell. All the time, the probe ismaintained at a constant position, ful-filling the promise of in situ calibra-tion. Outflow of the spent bufferenters an on-board reservoir for stor-age, precluding contamination of thelocal environment. The sample fluidinlet port is sealed during the calibra-tion interval in response to automatedactivation of an in-line check valve. Inaddition to sealing the flow cell, thecheck valve also acts to reduce the vol-ume of the cell effectively precludingthermal convection of the hydrothermalfluid (Figure 1).

During in situ measurement of pHand temperature, the flow control sys-tem activates the micrometering pump(FMI, Inc.) to continuously draw thesample fluid through the check valveinto the flow cell. The fluid passesthe normally open port of valves V4and V5 (LVM10 series valves, SMC,Inc.) and is eventually released into

FIGURE 1

Schematic illustration of process control system used for in situ and autonomous operations in-herent to the functionality and performance of the MpHC. Arrows indicate the direction of fluidflow. Hydrothermal fluid sampling and analysis require only pump operation, while calibrationrequires autonomous operation of both pump and valves.

the surrounding seawater. In contrast,during the calibration mode of opera-tion, the pump action is reversed, andpower to the valves permits access andmeasurement of buffer fluids. Thus,the pH buffers are pumped into theflow cell, while the check valve isforced into the closed position bythe confining pressure, as describedabove. This pattern of fluid flow min-imizes the number of valves used andpower consumption during pH mea-surements and calibration cycles. Theflow control system of valves andpumps is contained in an oil-filledchamber. The flow rate is preset to20 ml/min and is adjustable up to60 ml/min depending on the specificapplication. All wetted parts are con-structed by highly corrosive resistmaterial, such as Teflon (PTFE) andPEEK.

The Sensor ProbeThe electrochemical sensors inher-

ent to the design and functionality ofthe MpHC device include solid-stateelectrodes for measurement of pH(Ir-IrOx), redox (platinum), dissolvedsulfide (Ag-Ag2S), and referenceelectrode (Ag-|AgCl) (Figure 2c). Thesmall size (~1 mmOD) and solid-statedesign of the pH cell enhance dura-bility and deployment in and aroundrigid sulfide and rock structures. Theelectrochemical array also includesa titanium-sheathed thermistor with±0.1°C resolution. Owing to thechemical and thermal heterogeneityof diffuse flow vent fluids and theimportance of pH on microbial andmineralization processes, the in situmeasurement and calibration proto-cols we demonstrate represent an ini-tial, but essential step, if long-term

March/A

measurement and accurate interpreta-tion of biogeochemical processes asso-ciated with these systems is to becomea reality.

The sensor probe for pH measure-ment is constructed of three essentialcomponents: (1) oil-filled chamberfor hydrostatic pressure compensation,(2) titanium housing for structuralsupport, and, (3) flow cell assembly,which automates sampling and cali-bration procedures (Figure 2a). Theoil-filled chamber contains dielectricsilicone oil and is equipped with aT handle to facilitate deployment bysubmersible manipulator action. Theflow cell assembly is situated at the endof the titanium housing and connectsto the oil filled electronics chamber,as summarized above (Figure 2b). Amodified Conax (Conax Technologies,Inc.) fitting with PEEK and Viton seal-ants holds the electrodes and two PEEKfluid circulation tubes (OD 1.58 mm)in place. Applying torque on the Vitonsealant forms a watertight seal betweenthe flow cell and the silicone oil-fluidtitanium tubing. The small internalvolume (~1 ml) of the flow cell mini-mizes dead volume and enhances signalresponse to samples and standards. APEEK check valve cartridge (OptimizeTechnologies) with ruby ball and sap-phire seat was selected tomeet the strin-gent fluid flow requirements discussedabove. The fluid circulation tubing al-lows for distribution of fluid samplesand pH buffer fluids in and out of theelectrochemical sensor cell. A PTFEmesh is used at the sample inlet to re-move large particle entrained in thefluid samples.

The upper and lower portions ofthe check valve assembly unit are con-structed entirely of PEEK. The dis-tance between the sensing portion ofthe pH electrode and the sample inletport is 20mm.TheODof the flow cell

FIGURE 2

Schematic illustration of the pH probe of mobile calibrator showing integrated configuration ofelectrodes and check valve (a), expanded view of the flow cell assembly (b), and sensor assemblywith Viton sealant and Conax fitting (c).


assembly is ~16 mm, and the totallength of the titanium tubing includ-ing the flow cell assembly is 280 mm(Figure 2a). As noted, the toleranceto relatively high fluid temperaturestogether with narrow width of theprobe facilitates deployment in envi-ronments in and around diffuse flowhydrothermal vent sites.

Lab experiments demonstrated thatapproximately 5 min is required tocomplete one calibration cycle. Thus,the MpHC can complete more than16 calibration cycles with the currentdesign constraints and on-board storagecapacity of the waste buffer solutions.

General Description of the MpHCThe MpHC, including the flow

control system, battery compartment,electronics housing, and buffer fluidreservoirs, is placed in a plastic con-tainer that occupies approximately0.5 m2 of “basket” space on the Alvinsubmersible (Figure 3). The sensorprobe (external to the control systemcontainer box) is stored in a verticalPVC tube designed to facilitate its


removal and return. The control sys-tem and probe, however, were linkedby communication cable (3 m). TheMpHC weighs approximately 27 kgin air and 16 kg in water.

Sensor signals are processed at arate of one measurement every 5 sand stored on a 14-bit data logger.The flow system is linked to a controlcircuit, which, like the data logger,is of low power design and consistsof ICL (inductively couple link) com-munication (Fornari et al, 1997) andon-board flash memory.

Before deployment, the flow cell isfilled with artificial saltwater to main-tain hydrostatic pressure balance atthe seafloor. In practice, in situ pHmeasurement is coordinated betweenthe Alvin pilot and the scientist operat-ing the MpHC via ICL communica-tion. A control GUI was developedusing LabVIEW software and loadedon a Samsung XE700T tablet to facil-itate operation.When the manipulatorpositions the sensor probe at a targetlocation, the operator turns on theflow control system to draw in hydro-

l

thermal fluid, while the sensor signalis displayed on a computer screen.Subsequently, the operator will acti-vate the calibration subroutine, whilethe manipulator maintains the sensorprobe in a stationary position. TheGUI stores the MpHC status and elec-trode readings on the hard drive of thetablet. Accordingly, all the data arecopied to onboard memory for sub-sequent processing.

In situ pH CalculationTemperature and pressure play an

important role in sample and bufferpH and electrochemical characteristicsof the electrodes. Theoretically, sampleacquisition and calibration should beperformed at the same temperature;however, this is not always possible be-cause of abrupt temperature changescaused by mixing between hydrother-mal fluids and seawater. Accordingly,sharp temperature gradients are com-mon during a measurement or cali-bration cycle, for which laboratorycorrection needs to be applied.

Laboratory CalibrationTo correct pH buffer solutions for

temperature variability experienced onthe seafloor at deep-sea vents, three pHbuffer solutions were subjected to a se-ries of laboratory experiments. ThesepH buffers are commercially availablefrom Fisher Scientific, with productidentification numbers B79, B82,and B77, respectively. As with our sea-floor buffers, the dissolved chlorideconcentration of each of these bufferswas adjusted to minimize liquid junc-tion effects when used in seawater-typefluids. This, of course, requires re-assessment of the buffer pH value, es-pecially when exposed to temperaturevariability. The temperature calibra-tion test of each buffer was performed

FIGURE 3

MpHC mounted on basket of DSRV Alvin prior to dives to the Ashes vent area, Axial Seamount,NE Pacific Ocean in July 2014. The entire system occupies approximately 0.2 m2 of basket spaceand weighs approximately 16 kg in water. The probe is stowed in a PVC chamber on basket.

using a Ross pH combination elec-trode (Thermo Scientific), which wasconnected to an ATC probe (Orion)for temperature compensation. Inpractice, the tests were performed ina water bath where temperaturescould be adjusted from 2°C to 50°C.The pH buffers were determined tohave pH values of 3.74, 7.05, and10.39 at 25°C and 1 atm. Althoughtests such as these allow buffer fluidpH response to be calibrated at tem-peratures broadly analogous to therange of vent fluid temperatures thatmay be encountered on the seafloor,uncertainties associated with the cor-responding response of the Ir-IrOx

solid-state pH electrode also needto be determined. Thus, we performeda separate series of measurements usingthe same Ir-based solid-state pH sensoractually used for our vent studies. Theelectrochemical response of this pHsensor was previously established at el-evated temperatures (Pan & Seyfried,2008), but until now this has notbeen verified at temperatures lessthan 25°C, in terms of both Nernstianresponse and response time. Accord-ingly, the Ir-IrOx|Ag-AgCl electrodes,combined with Orion ATC probe,were sealed in a glass reaction cellinto which the previously calibratedpH buffer fluids were incrementallyadded. In this case, however, onlypH buffers 3.74 and 7.05 (25°C)were utilized, since these buffers areless sensitive to changing temperature,allowing us to more closely and effec-tively examine the effect of tempera-ture and pH on the electrochemicalresponse inherent to the Ir-IrOx sen-sor. Temperature and electrode po-tential were simultaneously measuredand recorded at 1°C increments from2°C to 50°C using a modified ThermoA211 pH meter. With benefit ofthe previously established values of

the pH buffers with temperature, theslope of the electrodes could be un-ambiguously determined (Figure 4).The measured slope, Smeasured, is inexcellent agreement with the theoret-ical Nernstian slope, Stheoretical, overthe full temperature range. The po-tential of the electrode with buffer3.74 was also shown to vary linearlywith temperature with a coefficient of−0.8625 mV/°C.

Thus, laboratory calibration helpedto establish the effect of temperaturevariability on pH measurement andcalibration. In effect, with two in situcalibration cycles, the Nernst slopecan be determined, providing accuratepH data for diffuse flow hydrothermalvent fluids. It is important to em-phasize, however, that this approachassumes that the effect of tempera-ture on the buffer solutions used forthe in situ calibration is known, eitherfrom lab tests, as performed here,or based on theoretical calculationsthat explicitly consider the effect of

March/A

temperature on the distribution ofaqueous species.

In situ pH Calculation MethodIn situ calculation of pH at seafloor

vents using the MpHC instrumentis accomplished by a three-step pro-cedure. First, at an initial temperature(T1), the potential of the pH and ref-erence electrodes are determined in re-sponse to reaction with two bufferfluids (A and B). This procedure allowsdetermination of the Nernst slope(ST1). It is important to emphasizethat the choice of buffer A is basedon our previously determined testsshowing that the pH value of the solu-tion is largely unaffected by temper-ature. Thus, buffer A with the pHvalue pHA canmore effectively providea reference or anchor potential againstwhich temperature effects can be com-pared. Subsequently, the procedureis repeated for the same two buffersat a different temperature (T2), whichthen allows determination of the slope

FIGURE 4

Measured mV/pH response of Ir-IrOx (pH) and Ag-AgCl (reference) electrodes with temperature.The measured data show a linear correlation with temperature equivalent to 0.198 mV·pH−1·°C−1.This is in excellent agreement with that predicted from the theoretical Nernstian slope over thetemperature range tested. The y-intercept is a unique function of the standard potential (E°) ofthe electrode.


(ST2). Finally, actual pH measure-ments of the hydrothermal fluid arecarried out at some other temperature(Tx), with the slope (STx) and referencepotential obtained from buffer A (EATx),as defined from the following set oflinear equations:

STx ¼ ST1 þ Tx � T1ð Þ ST2 � ST1T2 � T1

ð1Þ

EATx ¼ EAT1 þ Tx � T1ð ÞEAT2 � EAT1

T2 � T1

ð2Þ

With equations (1) and (2), thein situ pH of a sample (pHSTx) can bedetermined from the measured poten-tial ESTx as follows:

pHSTx ¼ pHA þ ESTx � EATxSTx

ð3Þ

In effect, by performing two in situcalibration cycles at any temperature,the temperature dependence of thestandard potential of the Ir-IrOx sensoris accounted for and corrected. Usingthis procedure, the accuracy of thein situ pH derived from the raw poten-tial is significantly improved. The hy-drothermal diffuse flow environmentat deep-sea vents is characterized byrelatively large temporal temperatureand pH variability and is well suitedto test the procedures proposed forpH calibration and measurement.The first calibration cycle is typicallyconducted in ambient seawater atdepth as the vent system is approached,whereas the second calibration isconducted at the point of the targethydrothermal fluids, and immediatelyfollowed by the hydrothermal mea-surement. Thus, two calibrations atdifferent temperatures are obtained inkeeping with the calibration scheme


outlined above. In some cases, onlyone calibration can be performed dueto the schedule of a certain dive. Insuch cases, the coefficients of the slopeand buffer potential obtained in the labare used to calculate the in situ pHvalue, but this is not optimal, sincethe effect of pressure is not beingaccounted for, rendering small, butuncertain error (Le Bris et al., 2001).

Field Deployment andInitial ResultField Test

TheMpHC has undergone a num-ber of field trials that have providedconfirmatory evidence of the effective-ness of the in situ auto calibration pro-cedures as a means of monitoring pHin variably high temperature subsea-floor hydrothermal environments.For example, during cruise AT26-17( July 2014), the MpHC was usedto measure the pH of diffuse hydro-thermal fluids at Axial Seamount, inthe NE Pacific (45°55′N, 130°0.0′W)and also at similar low-temperaturevents within Main Endeavour Field,Juan De Fuca Ridge (47°57.0′N,129°6.6′W). These vent systems havebeen long-studied, since both are tar-get locations for the U.S. Ocean Ob-servatory Initiative and the Neptunecabled observatory (Oceans NetworkCanada), respectively (Kelly et al.,2012, 2014). The MpHC completedthree deployments facilitated by assetsintrinsic to DSRV Alvin. Importantly,during dive 4743 (Sully diffuse flowvent site, MEF), the MpHC was usedfor pH and redox measurementsin vent fluids at temperatures as highas 70°C. The water depth at this siteis 2,189 m.

Longer-term deployment opportu-nities were available at the Axial Sea-mount. In particular, vent fluids were

l

sampled at the base of the Phoenixsulfide structure within the ASHESfield (45°55.99′N, 130°0.82′W) at adepth of 1,544 m. The temperatureof the diffuse flow fluid ranged from3°C to 25°C, whereas the nearby high-temperature (black smoker) vent fluidissuing from the central portion of thesame structure was 312°C. The tem-perature distribution underscores thatthe mixing with cold ambient seawatercontributes to diffuse flow hydrother-mal processes, while still providingcomponents sufficient to fuel micro-bial communities and associatedmacro-faunal assemblages (Figure 5), whichare intimately related to the low tem-perature vent fluid.

For this dive, buffer fluids with pHvalues (25°C) of 3.74 and 10.39 wereutilized to calibrate the chemical sen-sor system on the seafloor. Accordinglyand in keeping with previously de-fined procedures, the MpHC calibra-tion cycle was first performed at theambient seawater temperature 2.6°C.

FIGURE 5

MpHC in operation in and around a tubewormcolony at Phoenix hydrothermal site at Ashesvent field (Axial Seamount) during Alvin dive4741, July 21, 2014 at a depth of 1,544 m.The temperature of the diffuse flow fluidranged from approximately 5°C to 25°C. Notethe close proximity of a tube worm colonyand the diffuse flow to high-temperature ventswhere fluid temperatures exceed 300°C.

Upon initial deployment at the ventsite, the MpHC unit performed a10-min measurement of the tempera-ture and pH of the hydrothermalfluid and then conducted a 2-point cal-ibration for the next 20 min. Duringthe entire process, the Alvinmanipula-tor maintained the sensor probe at thesame location (Figure 6). At the onsetof the calibration cycle potential (mV)readings changed rapidly and consis-tently owing to constraints imposedby temperature variability and thevalues of pH buffer fluids relative toambient seawater and the conditionsof the diffuse flow hydrothermal envi-ronment. Indeed, data indicate thatthe temperature increased from ap-proximately 2.6 to ~23°C within thediffuse flow fluid and then decreased.

The potential/pH relationship ofthe two calibration cycles and corre-sponding hydrothermal fluid measure-ment cycles are shown in Figure 7(data of the calibration cycles are

March/A

shown in Table A1). As emphasizedearlier, the potential response of the3.74-pH buffer was used to calculatethe EATx in Equation 2. Figure 7 alsoreveals the steepening of the mV/pHslope with increasing temperature atseafloor pressure. This is caused bythe effect of temperature and pressureon the thermodynamic properties ofthe Ir-IrOx sensor, which influencethe sensors standard potential (E °).Thus, using equations (1)–(3), a pHvalue of 6.14 is obtained for the hydro-thermal fluid at 23°C (Figure 8). Theaccuracy of this number is difficult toconstrain unambiguously owing tothe inherent temperature, pressure,and compositional variability thatcharacterizes diffuse flow hydrother-mal fluids and the limited bottomtime available, precluding repeatmeasurements at similar conditions.However, we did conduct multiple

FIGURE 6

In situ potential of Ir-IrOx pH electrode with coregistered temperature data fromMpHC deploymentduring Alvin dive 4741, 1,544 m (see text) to Phoenix hydrothermal site in ASHES vent field (AxialSeamount). Initial pH calibration was performed in seawater at 2.6°C immediately before reposi-tioning the sensor system in the diffuse flow fluid where fluid temperature increased sharply to~23.5°C, while temperature and electrochemical data were continuously recorded. The change intemperate of the vent fluid to approximately 12°C elicited a second calibration cycle. The two buffercalibration cycles, in situ, at two different temperatures allowed.

FIGURE 7

Potential/pH relationships over two calibration cycles at the Axial Seamount study area. These dataprovide temperature-dependent mV/pH response to correct for temperature variability during cal-ibration and measurement at 2.6°C and 12°C. Based on the two calibration cycles (Table A1), thein situ pH at any other temperature within the range of the calibration scheme (e.g., Tx ) can bedetermined. Predicted mV/pH for Tx = 23°C is shown for comparison (see text). Buffer 3.74 (25°C)was used to obtain the standard potential EATx due to its pH value (pHA) shows relative thermalstability. The variability of the slopes of the temperature dependent E/pH trends is largely a prop-erty of the effect of temperature on the standard potential of the Ir-IrOx pH sensor, underscoring theneed for in situ calibration.


measurements of seawater at a depthof 1,450 m at Axial Seamount witha similar instrument with the sameelectrochemical and buffer control sys-tems (Tan et al., 2012). These dataindicate seawater pH of 7.65 ± 0.1.The extent to which this test appliesto the uncertainty attending the diffusefluid pHmeasurements reported aboveis largely dependent on the effectivenessof the temperature dependent correc-tion procedures developed and appliedduring the present investigation.

Comparing Data With OtherSites of Diffuse FlowHydrothermal Activity

It has long been recognized basedon mass balance and limited geochem-ical modeling results that diffuse flowfluids are best explained as nearly con-servative mixes of high temperaturevent fluids with seawater, with someadditional conductive heating of theseawater ventmix (VonDamm&Lilley,2004; Luther et al., 2012; Pester et al.,2008; Butterfield et al., 2004). Data


from these and other studies confirmthis for diffuse flow fluids at the AxialSeamount study area (Edmond et al.,1979) and at EPR 9°N (Von Dammet al., 2004), where the extent of sea-water mixing is, as usual, best indicatedby the near seawater concentrations ofdissolved Mg. In spite of this, however,there still exist noteworthy concentra-tions of transition metals and dissolvedgases needed to fuel microbial metabo-lism, while the relatively high dissolvedsilica provides evidence of the magni-tude of high-temperature source fluidsthat is added to the mix (Von Dammet al., 2004; Butterfield et al., 2004;Le Bris et al., 2003). Almost always,pH values are less than typical of sea-water, as reported during the presentstudy, although in a more generalsense, pH decrease often correlatespositively with temperature and dis-solved sulfide (Luther et al., 2012).In situ pH values reported here (6.14at 23°C) are generally within therange of analogous values reported forvent sites at EPR 9°N for fluids with

l

similar dissolved sulfide concentra-tions, although substantial variabilityin pH (in situ) is the norm not theexception (Luther et al., 2012). Partof the variability likely relates to con-straints imposed by the compositionof the high-temperature hydrothermalcomponent, while other factors in-cluding biological activity and uncer-tain chemical reaction processes inthe mixing zone, also likely play a role.Brief, but repeated spikes in tempera-ture and fluid chemistry, includingpH, are a well-recognized if not a char-acteristic feature of diffuse flow vent-ing, further complicating cause andeffect phenomena. Only by extend-ing the time series and deploying insitu samplers with internal calibratingcapabilities at different vent sites formuch longer times can we expect tounderstand unambiguously the inter-play among chemical, physical, andbiological processes at deep-sea vents.Efforts reported here represent an im-portant first step in realizing this goal.

ConclusionsThe MpHC system was developed

to perform reliable and precise pHmeasurements in and around hydro-thermal diffuse flow vents on the sea-floor at midocean ridges. A criticalelement of the new device involvesthe capability for autonomous calibra-tion in situ on the seafloor. Accord-ingly, multiple on-board buffers aresequentially delivered to a small vol-ume flow cell in which solid-statepH, reference, and redox electrodes,as well as temperature sensors, are con-tained. Computer-controlled valvesand pumps activated in response totemperature or time facilitate the insitu calibration procedures, includingrinse cycles with ambient seawater.The small size and low dead volume

FIGURE 8

pH and temperature data obtained for diffuse flow vent fluid at Axial Seamount, NE Pacific Ocean.The ~4-min record reveals steady state temperature and pH of 23°C and 6.14, respectively. Theaccuracy of the pH data benefits from temperature and pressure calibration procedures outlined inthe text and summarized in Figures 6 and 7.

of the electrode-bearing flow cell en-hance signal response time, while min-imizing consumption of buffer fluid.

The in situ calibration capabilityof the MpHC is essential to addressthe effect of temperature and pressureon electrochemical performance, espe-cially in the diffuse flow vent systems,where the mixture between high-temperature vent fluid and seawateroften produces temperature variabilityin time and space. Lab measurementshave confirmed that temperature vari-ability influences not only the pHbuffer speciation but more impor-tantly the standard potential (E °) ofthe solid-state IrOx sensor. In situ cali-bration on the seafloor at two differenttemperatures with two different pHbuffers has been found to be sufficientto correct for temperature effects per-mitting accurate pH measurement ofvent fluids at temperatures differentfrom the temperature of calibrationcycle, broadening the efficiency andeffectiveness of collecting vent fluidpH data.

The MpHC has been deployed bythe submersible Alvin to measure andmonitor pH at sites in the Gulf ofMexico and also at diffuse flow ventsassociated with hydrothermal activityat the Ashes vent field, Axial Caldera,northeast Pacific Ocean. In bothcases, data indicate that the auto cali-bration procedures provide an effec-tive means to assure the accuracy ofthe pH data and offer promise forlonger-term monitoring of pH andother components with power avail-able from cabled observatories.

AcknowledgmentsWe would like to thank the Alvin

operations team, the captain and crewof R/V Atlantis, and the Deep Submer-gence Science Committee (DeSSC)

for their help during the deploymentof MpHC. This work is funded byU.S. NSF grant 0927615, ChinaCAS grant XDB06040300, andChina NSF grant 41406110.

Corresponding Author:Chunyang TanDepartment of Earth Sciences,University of Minnesota310 Pillsbury DR SE,Minneapolis, MN 55455Email: [email protected]

ReferencesButterfield, D.A., Roe, K.K., Lilley, M.D.,

Huber, J.A., Baross, J.A., Embley, R.W., &

Massoth, G.J. 2004. Mixing, reaction and

microbial activity in the sub-seafloor revealed

by temporal and spatial variation in diffuse flow

vents at Axial Volcano. In: Subseafloor Bio-

sphere at Mid-Ocean Ranges, eds. Wilcock,

W.S.D., DeLong, E.F., Kelley, D.S., Baross,

J.A., & Cary, S.C., pp. 269-89. Washington,

DC: American Geophysical Union.

Childress, J.J., Fisher, C.R., Favuzzi, J.A.,

Kochevar, R.E., Sanders, N.K., & Alayse,

A.M. 1991. Sulfide driven autotrophic balance

in the bacterial symbiont containing hydro-

thermal vent tubeworm,Riftia pachyptila Jones.

Biol Bull. 180(1):135-53. http://dx.doi.org/

10.2307/1542437.

Ding, K., & Seyfried, W.E., Jr. 2007. In situ

measurement of pH and dissolved H2 in

mid-ocean ridge hydrothermal fluids at elevated

temperatures and pressures. Chem Rev.

107(2):601-22. http://dx.doi.org/10.1021/

cr050367s.

Edmond, J.M., Measures, C., McDuff, R.E.,

Chan, L.H., Collier, R., Grant, B., … Corliss,

J.B. 1979. Ridge crest hydrothermal activity

and the balances of the major and minor ele-

ments in the ocean: the Galapagos data. Earth

Planet Sci Lett. 46(1):1-18. http://dx.doi.

org/10.1016/0012-821X(79)90061-X.

Fornari, D., Bradley, A., Humphris, S.,

Walden, B., Duester, B., Catanach, R., …

March/A

Hickey, J. 1997. Inductively coupled link

(ICL) temperature probes for hot hydrothermal

fluid sampling from ROV Jason and DSV

Alvin. Ridge Events. 8(1):26-31.

Kelley, D.S., Carbotte, S.M., Caress, D.W.,

Clague, D.A., Delaney, J.R., Gill, J.B., …

Wilcock, W.S.D. 2012. Endeavour Segment

of the Juan de Fuca Ridge: One of the most

remarkable places on Earth. Oceanography.

25(1):44-61. http://dx.doi.org/10.5670/oceanog.

2012.03.

Kelley, D.S., Delaney, J.R., & Kim Juniper,

S. 2014. Establishing a new era of submarine

volcanic observatories: Cabling Axial Seamount

and the Endeavour Segment of the Juan de

Fuca Ridge. Mar Geol. 352(1):426-50. http://

dx.doi.org/10.1016/j.margeo.2014.03.010.

Le Bris, N., Sarradin, P.M., & Pennec, S.

2001. A new deep-sea probe for in situ pH

measurement in the environment of hydro-

thermal vent biological communities. Deep-

Sea Res Pt I. 48(8):1941-51. http://dx.doi.org/

10.1016/S0967-0637(00)00112-6.

Le Bris, N., Sarradin, P., & Caprais, J. 2003.

Contrasted sulphide chemistries in the envi-

ronment of 13°N EPR vent fauna. Deep-Sea

Res Pt I. 50(6):737-47. http://dx.doi.org/

10.1016/S0967-0637(03)00051-7.

Luther, G.W., III, Gartman, A., Yücel, M.,

Madison, A.S., Moore, T.S., Nees, H.A., …

Fisher, C.R. 2012. Chemistry, temperature,

and faunal distributions at diffuse-flow hy-

drothermal vents: Comparison of two geo-

logically distinct ridge systems. Oceanography.

25(1):234-45. http://dx.doi.org/10.5670/

oceanog.2012.22.

Pan, Y., & Seyfried, W.E., Jr. 2008. Experi-

mental and theoretical constraints on pH

measurements with an iridium oxide electrode

in aqueous fluids from 25 to 175°C and

25 MPa. J Solution Chem. 37(8):1051-62.

http://dx.doi.org/10.1007/s10953-008-9293-z.

Pester, N.J., Butterfield, D.A., Foustoukos,

D.I., Roe, K., Ding, K., Shank, T.M., &

Seyfried, W.E., Jr. 2008. The chemistry of

diffuse-flow vent fluids on the Galapagos Rift

(86°W): Temporal variability and subseafloor

phase equilibria controls. In: Magma to


Microbe: Modeling Hydrothermal Processes

at Oceanic Spreading Centers, eds. Lowell,

R.P., Seewald, J., Metaxas, A., & Perfit, M.R.,

pp. 123-44. Washington, DC: American

Geophysical Union.

Prien, R.D. 2007. The future of chemical

in situ sensors. Mar Chem. 107(3):422-32.

http://dx.doi.org/doi:10.1016/j.marchem.

2007.01.014.

Reimers, C.E. 2007. Applications of micro-

electrodes to problems in chemical oceanog-

raphy. Chem Rev. 107(2):590-600. http://

dx.doi.org/10.1021/cr050363n.

Sarrazin, J., Juniper, S.K., Massoth, G., &

Legendre, P. 1999. Physical and chemical

factors influencing species distributions on

hydrothermal sulfide edifices of the Juan de

Fuca Ridge. Northeast Pacific. Mar Ecol-Prog

Ser. 190:89-112. http://dx.doi.org/10.3354/

meps190089.

Shank, T.M., Fornari, D.J., Von Damm,

K.L., Lilley, M.D., Haymon, R.M., & Lutz,

R.A. 1998. Temporal and spatial patterns of

biological community development at nascent

deep-sea hydrothermal vents (9°50′N, East

Pacific Rise). Deep-Sea Res Pt II. 45:465-515.

http://dx.doi.org/10.1016/S0967-0645(97)

00089-1.

Tan, C., Jin, Bo, Ding, K., Seyfried, W.E., Jr.,

& Chen, Y. 2010. A long-term in situ calibra-

tion system for chemistry analysis of seawater.

J Zhejiang Univ-Sc A. 11(9):701-8. http://

dx.doi.org/10.1631/jzus.A0900643

Tan, C., Ding, K., Jin, B., Seyfried, W.E., Jr.,

& Chen, Y. 2012. Development of an in situ

pH calibrator in deep sea environments.

IEEE-ASME T Mech. 17(1):8-15. http://

dx.doi.org/10.1109/TMECH.2011.2168419.

Tan, C., Ding, K., & Seyfried, W.E., Jr.

2014. In situ pH measurement and real-

time calibration: applications to ocean obser-

vatories at deep sea vents. Sea Technol. 55(6):

41-4.

Von Damm, K.L. 1995. Controls on the

chemistry and temporal variability of seafloor

hydrothermal fluids. In: Seafloor hydrother-

mal Systems: Physical, Chemical, Biological


and Geological Interactions, eds. Humphris,

S.E., Zierenberg, R.A., Mullineaux, L.S.,

& Thomson, R.E., pp. 222-47. Washington,

DC: American Geophysical Union. http://

dx.doi.org/10.1029/gm091p0222.

Von Damm, K.L., & Lilley, M.D. 2004.

Diffuse flow hydrothermal fluids from 9°50′N

East Pacific Rise: origin, evolution and bio-

geochemical controls. In: The Subseafloor Bio-

sphere at Mid-Oceanic Ridges, eds. Wilcock,

W.S.D., Cary, S.C., Delong, E.F., Kelly, D.S.,

& Baross, J.A., pp. 245-68. Washington, DC:

American Geophysical Union.

l

Appendix ASupporting Information

TABLE A1

Simultaneous pH and temperature data recorded during two calibration cycles performed in situ in diffuse fluid issuing from the base of the Phoenixchimney structure, Ashes vent field, Axial Seamount, NE Pacific Ocean in July 2014. The in situ potential reading of each pH buffer was derived fromthe averaged recording for the last full minute of operation during each calibration cycle. Lab tests were used to correct the data for the effect oftemperature, while much smaller pressure effects were corrected from available theoretical and empirical data. Thus, the influence of temperature andpressure on the slope was shown to be linear and result in a combined increase of 0.187mV·pH−1·°C−1. Although the pH 3.7 buffer solutionwas shownto be only slightly affected by temperature, the same was not the case for the pH 10.4 buffer. In any event, the measured slopes (E/pH) at 2.6 and 12°Cwere still within 90% of the theoretical value at each temperature.

Sample
pHin situ E (mV) T (°C) Measured slope(mV/pH)
March/April 2016 Volum

Theoretical slope(mV/pH)

Buffer A
3.70 277 ± 0.3 2.6 −50.4 −54.7
Buffer B
11.20 −100.6 ± 0.4
Buffer A
3.70 253.6 ± 0.5 12.0 −52.2 −56.6
Buffer B
10.87 −120.5 ± 2.0
e 50 Number 2 47

P A P E R

Micro-Modem for Short-Range UnderwaterMobile Communication Systems
A U T H O R SJun-Ho JeonSung-Joon ParkDepartment of ElectronicEngineering, Gangneung-WonjuNational University

A B S T R A C T

l

Recently, there has been great interest in short-range underwater communicationfor applications such as water pollution monitoring, fish farming, oceanographic datacollection, and underwater tactical surveillance based on underwater sensor networksystems. Because underwater wireless communication relies primarily on acoustics,the development of acoustic modems has been an important topic that needs to beaddressed. Furthermore, for years, underwater biomimetic fish robots have been stud-ied in the area of biomechanics for scientific use and reconnaissance missions. In thisarticle, we describe an underwater mobile communication systemwhere fish robots actas sensor nodes, which is different from the conventional concept of an underwatersensor network that is static. We describe the issues that need to be resolved to providemobility to a node, and we develop amicro-modem tomeet the requirements of a mov-ing node. Experiments conductedwith prototypes in both a lake and a river verify that theproposed system provides a new degree of freedom (mobility) and is a viable approach.Keywords: underwater communication, underwatermobile sensor network, underwaterrobot, micro-modem

underwater acoustic modems has beenconducted by both academia and in-

Introduction

For a considerable period of time,underwater communication systemshave focused mainly on applicationssuch as marine resource exploration,deep-sea probes, and large-sized under-water vehicles. As an enabling tech-nology, research on long-distance

dustry. Although acoustic waves areinherently resilient to attenuation in awater medium, a modem must emit ahigh signal power to achieve the desig-nated working range.

Quite recently, the field of under-water wire less sensor networks(UWSNs) has emerged as a rapidlygrowing area of research as a result ofits wide range of aquatic applicationsboth in inland waters and in the littoralseas (e.g., water pollution monitoring,fish farming, oceanographic data col-lection, and underwater tactical sur-veillance) (Akyildiz et al., 2005). Theincreasing demand for short-rangewireless connections between nodeshas introduced a new challenge tothe development of micro-modems,which have the properties of smallsize, low power consumption, andmoderate data rate and coverage. Theinitial work on small-scale acoustic

modems, named CORAL (Pandyaet al., 2005), spurred the researchand development of micro-modems(Freitag et al., 2005; Lee et al., 2014),including the well-known micro-modem developed by the Woods HoleOceanographic Institution.

The field of biomimetic fish robots,which are a type of underwater vehicle,is another interesting research area thatattempts to make the best use of thelaws of nature. Since a fish robot utiliz-ing biomechanics was first introduced(Streitlien et al, 1996), many bio-inspired underwater robots have beeninvestigated. The recently unveiledGhostSwimmer, developed by theU.S. Navy, is the newest water dronemimicking a tuna, which operates au-tonomously or with a tether in waterdepths ranging from 0.25 to 91 m.When operating independently, thefish robotmust be periodically brought

to the surface to deliver the data it hasgathered.

In this article, we propose the useof an underwater fish robot equippedwith an acoustic communicationmod-ule for efficient underwater mobilecommunication systems. If wirelesscommunication support is added to afish robot, it is possible to control therobot and to transfer the collected datain real time without the help of anumbilical cable or the use of periodicfloatation, which are troublesomeand time-consuming. Furthermore,unlike a static UWSN, the fish robotprovides insight into a new researcharea, underwater mobile sensor net-works using robotic fish as under-water sensor nodes. For this purpose,we first identify the immediate issuesthat need to be addressed in orderto develop a micro-modem for under-water mobile devices. In particular, we

focus on the characteristics of a spher-ical transducer acting as a commonendpoint of the transmitter and thereceiver, the underwater channelbased on a ray tracing model, and therequirements for the modem. Further-more, we investigate a frame formatand a symbol structure that are ade-quate for the data transfer requirementsof the target system, and we develop aprototype of a micro-modem. In or-der to verify the functionality of themodem and the feasibility of the aggre-gated system prototype, experimentsare conducted in a lake and river.

The remainder of this article isorganized as follows: First, we brieflyintroduce the objectives and the oper-ational scenarios of short-range under-water mobile communication systems.Next, we explain in detail the researchchallenges regarding the realization ofmicro-modems for use in underwatermobile communication systems anddescribe a design and implementationof a micro-modem conforming to re-quirements. Lastly, we present experi-mental results and give concludingremarks that include direction for fur-ther research.

Underwater MobileCommunication SystemsUnderwater Robot System

A prevalent assumption regardingsmall-sized underwater robots, suchas underwater vehicles and fish robots,is that they either move autonomouslyin water or are controlled with a tetherconnected to a mother ship at the sur-face of the water body. This is becauseacoustic modems developed to date aretypically too bulky to be mounted inunderwater robots. If it were possibleto transfer data wirelessly during oper-ation, the effectiveness of the systemwould increase significantly. This is

our motivation to propose an under-water robot system where a robotcontains an acoustic communicationmodule as illustrated in Figure 1a.In the presented operational scenario,a robot sends the collected under-water data or is controlled in real timevia an acoustic link. In the next section,the requirements of an acoustic modemfor small-sized underwater robots areinvestigated in detail.

Underwater Mobile SensorNetwork System

Figure 1b depicts a mobile systemmodel that monitors underwater en-vironments in real time. The systemconsists of a docking station andthree bio-inspired fish robots. Thedocking station contains an acousticmodem to communicate with thefish robots underwater and a gatewaythat controls the radio frequency (RF)communication to terrestrial networks.It is also equipped with docking, dry-ing, cleaning, and recharging modulesfor fish robots. In order to locate therobots, an ultra-short baseline (USBL)transducer is installed in the dock-ing station. Furthermore, each fishrobot is equipped with a thrust mod-ule, underwater sensors, a micro-modem, a USBL transponder, a global

March/A

positioning system module, and bat-tery packs.

The system operates as follows:Once a user located on shore gives acommand for underwater monitoring,the command is delivered via an RFcommunication link to the gatewaylocated in the docking station. Then,three robots navigate to a designatedwaypoint near the surface of the water.After arriving at the point, they sub-merge deeply in the water, sense en-vironmental data, and send the datathrough an acoustic communicationchannel. When the assigned missionis complete, the fish robots may returnto the docking station for cleaning andrecharging or they may navigate toanother place in order to gather dataaccording to a user’s command.

The specific procedure of theacoustic communication between thedocking station and the three fish ro-bots is as follows: Whenever the dock-ing station needs to transmit data tothe fish robots or to receive data fromthe fish robots, it sends a broadcastmessage to synchronize the entitiesparticipating in the communicationsystem. After the initial broadcastingperiod, four time slots are sequentiallyand exclusively dedicated to the gate-way, the first fish robot, the second

FIGURE 1

Underwater mobile communication systems.


fish robot, and the third fish robot. Inthis manner, each entity sends controlor monitoring data during its own timeslot through a single acoustic channel.

Research ChallengesTransducer

The most critical problem in thedevelopment of a micro-modem foran underwater robot arises from thetransducer that is used to convert elec-tric signals into acoustic waves and viceversa. The two main properties re-quired for a transducer to be adoptedin a robot are omnidirectivity and tinysize. In order to support acoustic com-munication irrespective of the positionand orientation of the robot, the use ofa spherical transducer, which has anomnidirectional radiation pattern butpresents a lower efficiency comparedwith directional transducers, is an in-dispensable option. Furthermore, thespherical transducer must be as smallas possible, because a large transduceraperture is liable to disturb the naturalmovement of a robot because of theshape and weight of the transducer.As a result, because the resonant fre-quency of a transducer is inversely pro-portional to its size, we have no choicebut to use a high-frequency carrier.

Figure 2 shows the ratio of the out-put voltage (VO) of the transducer atthe receiver to the input voltage (VI )of the transducer at the transmitter asa function of carrier frequency and dis-tance between the two transducers. Forboth subgraphs, cylindrical spreading(Kinsler et al., 1999), 10°C, pH 7,and −53 dB transducer loss for sound-ing and listening are assumed. Salinityis assumed to be 0 in Figure 2a and35‰ in Figure 2b to reflect freshwater and seawater, respectively. Fromthe results, it is obvious that the loss insignal strength increases significantly


with the increase of frequency as the dis-tance exceeds approximately 1,000 min fresh water and 100 m in seawater.In other words, the carrier frequencybecomes a crucial factor of concern be-yond 1,000m in fresh water and 100min seawater under the given conditions.Therefore, if we assume that the under-water mobile communication systemsdescribed above are operated at a dis-tance of several hundred meters in saltywater, the acoustic wave emitted froma small-sized transducer having a highresonant frequency will experience alarge loss as a result of absorption.

Underwater ChannelThe ray tracing model is one of the

widely used propagation models forunderwater acoustic signals in whichsound energy is considered as a set ofpropagating rays in a fluid with a con-stant sound speed. According to theray tracing model, a received signal isrepresented as the superposition of mul-tiple rays, as depicted in Figure 3a. Thesignal strength of each ray is affectedby spreading and absorption in water,transducer loss, and reflection coeffi-cients (Chitre, 2007). For example,the delay spreads under conditionsmodeling a salty lake and a river thatwill be further discussed in a later sec-

l

tion are provided in Figure 3b. Cylin-drical spreading, a carrier frequencyof 70 kHz, and the distance of 500 mare assumed for simulation. Reflec-tion losses at the water surface and atthe bottom are chosen to be 0.5 and6.0 dB, respectively. The water depthH and the depths of transmitter and re-ceiver, ht and hr , are set to 2 m, 0.5 m,and 0.5 m for the result in the lake,whereas they are set to 10 m, 1.5 m,and 1.5 m for the result in the river.The salinity in the lake is 35%, whereasthat in the river is 0. In addition, onlythe paths within six reflections at thesurface and the bottom are includedin the results for simplicity. A key ob-servation is that the delay of each pathdecreases as the ratio of the distance tothe water depth increases. Since thisratio is relatively high in the lake, thedelay spread becomes narrower whencompared with the result made in theriver. Thus, if the channel were suffi-ciently static, a data rate of a few kilo-bits per second could be achieved inthe lake. It is also seen that the signalstrength of the dominant path in thelake is much smaller than that in theriver, owing to the increase of absorp-tion by salinity as shown in Figure 2.

In addition to the static attenua-tion described above, an acoustic wave

FIGURE 2

Loss in signal strength with respect to distance and frequency in underwater channels. (Colorversion of figures are available online at: http://www.ingentaconnect.com/content/mts/mtsj/2016/00000050/00000002.)

traversing water suffers from ambientnoise generated by aquatic organismsand natural phenomena in the water.Surface agitation by wind is anotherfactor to consider because it makesthe reflection of acoustic waves atthe surface irregular. Thus, in a harshenvironment where a channel is time-varying as a result of the complexcombination of all natural and artifi-cial causes, it is impractical to analyzea channel theoretically. More informa-tion on underwater acoustic channelsfor general purposes can be found inStojanovic and Preisig (2009).

Modem CharacteristicsIn addition to the issues presented

by the transducer and the underwaterchannel, many other challenging issuesmust be considered in the devel-opment of a micro-modem. One ofthem is modem size. Because thelength of underwater robots can beon the order of tens of centimeters,the modem size should be as small aspossible so that the modem can becontained in a robot. Furthermore, afish robot having a streamlined shapeand a couple of articulated joints re-stricts the modem size significantly asillustrated in Figure 1a. A communi-cation protocol for data exchange inan underwater channel must also bedesigned. Because the data carried un-

derwater is normally short and bursty,a packet-based protocol is preferredbecause of its efficiency. On the otherhand, the data rate and working rangeare not critical matters when consider-ing the architecture and requirementsof small-scale underwater mobile com-munication systems.

Micro-ModemFor efficient data transfer in an

underwater channel, acoustic trans-mission technologies suitable for anapplication should be defined care-fully. In this section, we present asimple protocol targeted for systemsexecuting intermittent acoustic com-munication, such as an underwaterenvironment monitoring system withbiomimetic fish robots. Figure 4ashows a physical layer frame format de-signed for the support of packet-based

March/A

acoustic communication. The physicallayer header consists of a preamble, astart-of-frame delimiter (SFD), and aframe length field; it is followed bydata of variable length. Bearing inmind that underwater sensing data orrobot control data are typically shortin length, we allocate 255 bytes forthe maximum length of data that cor-responds to 1-byte frame length. Ad-ditional fields can be appended ifrequired.

The symbol structure used to gen-erate a modulated signal is illustratedin Figure 4b. Because the size of themicro-modem is highly restricted as aresult of the small size of the fishrobot, a binary modulation with non-coherent detection that requires a sim-ple structure and low computationalcomplexity is implemented. In orderto mitigate the effect of intersymbolinterference (ISI) caused by the delayspread in an underwater acousticchannel, the transmitting time (T ) isfollowed by the guard time (TG) foreach symbol duration (TS). Therefore,the sinusoidal carrier is generated dur-ing T and no signal is generated duringTG for the transmission of informationbit “1,” whereas no signal is generatedduring the whole TS for the transmis-sion of information bit “0.”

Figure 5 is an actual image of theimplemented modem hardware. Forease of loading into a fish robot, it is

FIGURE 4

Frame format and symbol structure.

FIGURE 3

Underwater channel modeled using ray tracing method.


designed as a cylindrical shape that isassembled by stacking up three circularboards—a digital board, an analogtransmission board, and an analog re-ception board. The digital board pro-vides frame generation and recoveryand controls the interface with analogboards and the other devices that actas upper layers. In the analog trans-mission board, a modulated signal isamplified up to 300 V peak-to-peakand then fed to a transducer. Analogsignal processing for detection in-cluding amplification and bandpassfiltering is performed in the analog re-ception board. For this application,a tiny spherical transducer havingan omnidirectional radiation patternwhose resonant frequency and diameterare 70 kHz and 34 mm, respectively,is utilized. The transducer is directlyconnected to the modem hardware tocomplete a micro-modem. The mainfeatures of the developed micro-modemare summarized in Table 1.

Performance EvaluationGyeongpo Lake and the Han River

were used as test sites for the outdoorexperiments, as shown in Figure 6.Gyeongpo Lake, located in Gangneung,Republic of Korea, is 1.6 km wide and0.7 km long and has amaximumdepthof 2m.Connected to the PacificOcean


through a small waterway, it containsmany aquatic plants and marine lifethat form a colony. Normally, theflow of water is calm except for an oc-casional surface agitation caused bywind. The other site is a confluencepoint of two streams of the Han Rivernear Seoul. The width and maximumdepth of the river are approximately500 and 10 m, respectively. As ex-pected, the water becomes shalloweras it approaches the riverbank. Theflow velocity in the quiet state is ap-proximately 1 knot. Small piers at thelakeside and the riverside are utilizedas the main bases for the experiments.

In accordance with the amount ofdata that is gathered by and transferredto one fish robot at a time, 16 bytes areallocated to the data field in the frame

l

format for the experiments. Accordingto the primary experiments, the outputlevel of the transducer at the transmit-ter increases gradually and reaches asteady state after five cycles of a 70-kHzsinusoid. Thus, not only to ensure thenormal operation of the transducer butalso tomaximize the avoidance of inter-symbol interference caused by multi-paths, the transmitting time T is set tofive cycles of a 70-kHz sinusoid. Therest of the symbol duration is entirelydedicated to the guard time. For ex-ample, when the data rates are 1 and5 kbps, the ratio of T to TS becomes5/70 and 5/14, respectively.

In order to examine the feasibil-ity of the developed micro-modem,point-to-point communication testswere conducted. In the lake and river

FIGURE 5

Micro-modem.

TABLE 1

Specifications of micro-modem.

Feature
Description
MCU
STM32F103 (Cortex-M3)
Modem size (Φ × H )
70 × 40 mm
Transducer size (Φ)
34 mm
Resonant frequency
70 kHz
Transducer loss
−53 dB
Interface
UART, SPI, USB
Supplied voltage
14.8 V/29.6 V
Power consumption
8 W
FIGURE 6

Experimental environments.

experiments, a modem was placed at apier and another modemwas located ata boat. For the sake of convenience,transducers were submerged only at0.5 and 1.5 m below the surface ofthe lake and the river, respectively. Ac-cording to the experiments, acousticcommunication was possible up to adata rate of 5 kbps in the lake and1 kbps in the river at a distance of500 m, as is expected to some extentfrom the ray tracing model shown inFigure 3b.

The system level prototype de-scribed in Figure 1b was deployedin the river for the integration tests.First, a functional test on the under-water acoustic communication systemwas performed using a surface gatewayand three underwater nodes withoutfish robots to establish the integrityof the underwater communicationnetwork. After that, a communicationtest with the gateway and one fishrobot equipped with the micro-modemwas conducted to confirm reliableunderwater acoustic communicationfor a moving object. Finally, a systemlevel experiment conforming to thewhole scenario explained in the previ-ous section was executed.

Conclusions andFuture Direction

In this article, we outlined the useof underwater mobile communicationsystems where biomimetic fish robotsact as moving nodes. We determinedthat providing mobility for a smallnode gives rise to serious restrictionson the design of the acoustic modem,and we conducted a review of the un-derwater channel characteristics thatrelate to short-range acoustic commu-nication. Based on these observations,we explored acoustic transmission tech-niques for the application of under-

water mobile communication systemsand developed a prototype of a micro-modem. According to the modem leveland system level experiments carriedout in a lake and a river, the proposedsystem equipped with the micro-modem is a viable option.

Owing to the mobility of under-water mobile communication systems,service coverage could be enlarged oreasily reconfiguredwith a small numberofmobile nodes; this translates into sys-tem and cost efficiencies. Beyond thediscussions presented in this article,it may be worthwhile to investigatemethods to improve the performanceof the micro-modem using high-speedintegrated circuits dealing with ad-vanced transmission and receptiontechniques. We believe that the discov-ery and deployment of new convergentsystems utilizing underwater robotswith embedded micro-modems arealso challenging issues to be addressed.

AcknowledgmentThe authors wish to thank Ocean

Sensor Network System TechnologyResearch Center at Gangneung-WonjuNational University and Korea Insti-tute of Industrial Technology for con-tributions to this article. This workwas supported in part by Korea Re-search Council for Industrial Scienceand Technology (B551179-10-02-00, Development of Bio MimeticRobot System) and in part by Mid-career Researcher Program throughNRF grant funded by the MSIP(2014R1A2A2A01006205).

Corresponding Author:Sung-Joon ParkDigital Communications Lab.Department of Electronic Engineering,Gangneung-Wonju National University

March/A

7 Jukheon-gil, Gangneung,Gangwon 25457, Republic of KoreaEmail: [email protected]

ReferencesAkyildiz, I.F., Pompili, D., & Melodia, T.

2005. Underwater acoustic sensor networks:

research challenges. Ad Hoc Netw. 3(2):

257-79. http://dx.doi.org/10.1016/j.adhoc.

2005.01.004.

Chitre, M. 2007. A high frequency warm

shallow water acoustic communication chan-

nel model and measurements. J Acoust

Soc Am. 122(5):2580-6. http://dx.doi.org/

10.1121/1.2782884.

Freitag, L., Grund, M., Singh, S., Partan, J.,

Koski, P., & Ball, K. 2005. The WHOI

micro-modem: an acoustic communications

and navigation system for multiple plat-

forms. In: Proceedings of IEEE OCEANS.

Washington, DC: IEEE. http://dx.doi.org/

10.1109/oceans.2005.1639901.

Kinsler, L.E., Frey, A.R., Coppens, A.B., &

Sanders, J.V. 1999. Fundamentals of Acoustics.

Hoboken, NJ: John Wiley & Sons. 548 pp.

Lee, W., Jeon, J.-H., & Park, S.-J. 2014.

Micro-modem for short-range underwater

communication systems. In: Proceedings of

MTS/IEEE OCEANS. St. John’s,Canada:

IEEE. http://dx.doi.org/10.1109/oceans.2014.

7003208.

Pandya, S., Engel, J., Chen, J., Fan, Z., &

Liu, C. 2005. CORAL: miniature acoustic com-

munication subsystem architecture for under-

water wireless sensor networks. In: Proceedings

of IEEE Sensors. Irvine, CA: IEEE. http://

dx.doi.org/10.1109/icsens.2005.1597661.

Stojanovic, M., & Preisig, J. 2009. Under-

water acoustic communication channels: propa-

gation models and statistical characterization.

IEEE Communications Magazine. 47(1):84-9.

http://dx.doi.org/10.1109/MCOM.2009.4752682.

Streitlien, K., Triantafyllou, G.S., &

Triantafyllou, M.S. 1996. Efficient foil propul-

sion through vortex control. AIAA J. 34(11):

2315-9. http://dx.doi.org/10.2514/3.13396.


P A P E R

Characteristics of Microstructure TurbulenceMeasurements With a Moored Instrumentin the South China Sea
A U T H O R SXin LuanXiuyan LiuHua YangShuxin WangCollege of InformationScience and Engineering,Ocean University of China
Guojia HouCollege of Information Engineering,Qingdao University

Dalei SongCollege of Engineering,Ocean University of China


A B S T R A C T

l

A newly designed autonomous instrument was evaluated for long-term oceanturbulence measurements in the South China Sea from October 19, 2013, toFebruary 10, 2014 at 21°09.90′N, 117°42.03′E. A shear velocity time series wascollected continuously for 115 days at a single depth of 250m using two orthogonalshear probes (PNS06). The mechanical design of wings is developed for an obser-vation platform to achieve characteristics of stability and flexibility of current direc-tion tracking. The depth and pitch in comparison with ambient current speed areanalyzed, and the results indicate that the platform has satisfied stability require-ments for turbulence measurement. The comparison between the instrument head-ing and current direction shows that the instrument flexibly responds to variationsof ambient currents. Throughout the observation period, 62% of the data sets is ofsignificant correlation with correlation coefficients above 0.5, and 4% is of strongcorrelation. The shear velocity spectra correspond well to the theoretical Nasmythspectrum, and the magnitude of dissipation rates estimated from shear is approx-imately Ο(10−8) W kg−1. The moored instrument is functional as designed and pro-vides a reliable observation platform for extended turbulent measurements.Keywords: turbulence measurements, moored instrument, stability, flexibility,shear spectra
Otant role in improving our understand-
ing of the effect of the turbulence

Introduction
cean turbulence plays an impor-
mixing process on ocean circulationdynamics and significantly improvingclimatemodels. It also significantly im-pacts the research of ocean dynamics,energy exchanges, the marine ecolog-ical environment, and the distributionof suspended materials (Nikurashin& Ferrari, 2010). Ocean turbulencehas received considerable attention inthe last few decades and has becomea popular subject in marine scienceresearch. Measurements of the turbu-lence mixing process are usually limitedto surveys of short duration and incom-plete and sporadic samples in timeand space (Thorpe, 2005). Measure-ments of dissipation rates of turbulentkinetic energy (TKE) over extended

periods yield a valuable data set thatprovides quantitative assessments ofdissipation rates and turbulent mixinglevel; the technologies for long-termmeasurements are therefore increas-ingly important.

Conventionally, ocean microstruc-ture turbulence is measured by fast re-sponse thermistors or two orthogonalairfoil shear probes measuring hori-zontal or vertical shear components(Lueck et al., 2002). Use of a micro-structure profiler with shear probes isa common method to measure shearvelocity data. However, the traditionalprofilers, whether horizontal or verti-cal, require fairly intensive labor andhighly trained personnel; it is there-

fore impossible to collect long-termtime series due to financial and logis-tical constraints (Lueck et al., 1997).By altering ships and scientific per-sonnel, the longest time series lasted30 days and cost more than $1 millionwith conventional profilers (Moumet al., 1995). Depending on the ad-vantages of not requiring professionaland technical personnel, the mooredsystem makes it possible to take un-attended turbulent observations. Theearlier research of moored microstruc-ture measurements was reported byLueck et al. (1997). They used fourshear probes and three pairs of Seabirdsensors to measure shear and fast tem-perature in the dissipation range of the

wavenumber spectrum. This mooredplatform obtained a long time seriesof turbulence measurements at a re-duced cost for personnel and ship time.Lueck and Huang (1999) deployed amoored instrument in a swift tidalchannel by using shear probes andFP07 thermistors to measure shear ve-locity and temperature fluctuations;the profilers collected time series ofocean turbulence at a fixed depth foronly 8 days. Song et al. (2013) designedthe mooring turbulence observationinstruments that implemented con-tinuous observation at a fixed level inthe Kiaochow Bay for several days. Bycomparing shear probes with micro-structure profiles MSS60, it was shownthat the moored system was reliableto detect the microscale turbulence,but not for long-term series measure-ments. An autonomous instrument(Fer & Paskyabi, 2014) was located ina wave-affected layer. Only the velocityspectra in the dissipation subrange canbe used for dissipation rate calcula-tions; the shear spectra in the inertialsubrange are severely contaminatedby platform vibration, and the tem-perature gradient spectrum in the dis-sipation subrange is not satisfactorilyresolved. While limited to a singledepth, the measurement systems col-lected time series for 3 weeks contin-uously. The latest work reported byLueck et al. (2015) was successfullydeployed over a 2-week period in IslaySound; the flow speed exceeded 3 m/s,and the depth of this channel was 53m.These environmental conditions rep-resented significant challenges for thedesign of the mooring measurements.In this article, we report on micro-structure turbulence measurementsat dissipation scales from a newlydesigned moored system in the SouthChina Sea (SCS). It was deployed at anapproximately 250-m depth in water

column and was less contaminated bywave orbital velocities and the interfer-ence of platform motion. The mooredinstrument has continuously collecteddata for 115 days with a giant storagecapacity, which represents significantchallenges for extended observationsof turbulent dissipation.

The outline of this article is as fol-lows. The sections of instruments andexperiments mainly describe experi-mental equipment and environmentalconditions, followed by the proposedmethod to normalize the measureddata between the heading and currentdirection. Subsequently, the resultsand discussion mainly show the char-acteristics of the moored instrumentincluding the stability, flexibility, andspectral analysis of the microstructureshear velocity data. The conclusionpresents some additional measures toimprove the mechanical design of thesystem.

InstrumentsMechanical Structure

The moored turbulence measur-ing instrument (MTMI; Figure 1) isan ocean turbulence measurement sys-tem designed to collect microstructureshear time series at a fixed level. Themechanical structure of the MTMImainly includes five divisions: the nose

March/A

section, the instrument cabin, the bat-tery cabin including rechargeable alka-line battery packs, the latter float bodymade of buoyancy materials, and thecrossed wings. The nose section is asemicircular head, in which two orthog-onal shear probes (PNS06; Wang et al.,2014) are mounted to measure fluctu-ations of cross-stream velocity (∂w/∂x,∂v/∂x) and hydrodynamic lift force in-duced by the ambient turbulent flow.Four guard rings are attached to pro-tect the shear probes from being dam-aged during the experiments. Themain circuit boards and attitude sen-sor (MTI) measured three-axis accel-eration signals (Ax, Ay, Az ), and themotion behaviors of the instrument(heading, ψ; pitch, θ; and roll, ϕ) areinstalled in the instrument cabin. TheMTMI is powered by battery packs,each rated for 40 Ah at 9-V DC, pro-viding an estimated operating time ofapproximately 6 months. Balanceweight is contained in the latter floatbody; the horizontal and vertical tailscontained in the wings are usedto maintain balance and control thedirection of the main floating bodyunder the sea. Particularly, the crossedwings make it possible for the MTMIto track variations of current directionand to maintain its position counter tothe coming flow. The instrument is alow-drag buoy that can be used as the

FIGURE 1

Mechanical features of the MTMI.


upper buoyancy element or integratedat the desired depth in a mooring line.To obtain accurate turbulence sig-nals, the platform should be stableunder severe interference (termed anti-interference) and responds rapidly tochanges of current direction. There-fore, the structure of the MTMI isdesigned to be streamlined to reduceresistance. The assembled MTMIweighs approximately 254.9 kg, witha net buoyancy of 12 kg in seawater.It has an overall length of 3.0 m anda maximum diameter of 50 cm. Themaximum working depth of the wholeMTMI is 4,000 m, which complieswith requirements for working in thedeep sea.

To permit the instrument head-ing to rotate in response to changes incurrent direction, the hydrodynamictorque of the wings’ force must belarger than the heading force, whichdepends on the mechanical structuresof the main body and the wings ofthe instrument. According to the re-quirements of fluid mechanics, thewings of the MTMI are designed tobe cruciform, and two vertical and hor-izontal fins of the wings are separatedin this experiment. The crossed finsat the rear provide the torque for rota-tion and stabilize the pitch motions.The area of countering flow for onefin is approximately 2.75 × 105 mm2

(Figure 2), and the torque of the wingshas been programmed to force largerthan the heading force. In short, thestreamlined body and crossed wingscontribute to improving the stabiliza-tion and flexibility of the platform.

Coordinate SystemTheMTMI system is a right-handed

Cartesian coordinate system that hasthe following axes: x, which is alongthe major axis of the instrument andis positive pointing forward. The y axis


points to the port side of the instru-ment and is positive athwartship, andz is directed positive upward (Figure 3).Accordingly, the instrument heading(ψ, rotation around the z axis) is posi-tive counterclockwise, ranging from−180° to 180°. Pitch (θ, rotation aroundthe y axis) is positive when the nose isdown, ranging from −90° to 90°, androll (ϕ, rotation around the x axis) ispositive when the instrument rolls portside up, ranging from −180° to 180°.For nonzero values of pitch and roll,the vertical axis is not aligned with theacceleration of gravity g. The MTImotion sensor mainly records the three-axis accelerations of the Ax, Ay, Az, andthese three acceleration signals alsofollow the above coordinate system.

ExperimentDeployment

The moored instrument MTMIwas deployed in the SCS on October

l

19, 2013, and recovered on February10, 2014, at 21°09.90′N, 117°42.03′E(Figure 4). It was anchored at a depthof approximately 250 m in the SCS;this method of deployment was sim-ilar to other moorings. First, the in-strument was lowered into the deepwater, followed by the flotation sphere.Subsequently, the remainder of moor-ing line was deposited under tensionuntil the anchor had boarded theship. During the experimental period,the instrument MTMI had alreadysampled 28.5 GB of turbulence shearvoltage data for a continuous 115 daysthat yielded a valuable database forfurther research of the turbulent dis-sipation and evolution processes inthe SCS.

The Mooring SystemThe whole mooring system used

for long-term ocean turbulence ob-servation can be divided into foursubdivisions (Figure 5) based on theirdifferent uses. The Sea-Bird Elec-tronics (SBE37SM-RS232) CTD sen-sor and the MTMI are attached inpart I. TheCTD systemmainly recordsthe water temperature, conductivity,and depth of the MTMI. The in-strument MTMI is assembled for ob-serving microstructure shear velocityfluctuations. To stabilize the obser-vation platform, a turbulence instru-ment needs to be embedded in thestreamlined floating body made ofbuoyant glass microsphere material.Part II contains the Doppler currentmeter (RCM 11) measuring the am-bient current speed (cm/s) and currentdirection (°) in the deep sea. Measure-ments are compensated for instrumenttilt and referred to magnetic North byusing an internal solid state compass.A microprocessor computes vector-averaged current speed and directionover the last sampling interval. An

FIGURE 2

Size of each fin of the wings.

FIGURE 3

The right-handed Cartesian coordinate systemof the instrument and rotation for pitch (θ), roll(ϕ), and heading (ψ).

acoustic release in part III is utilizedto recover the moored instrumentMTMI. The last part is the anchorblock, providing gravity for the wholesystem to make the instrument bein equilibrium. When the acousticrelease receives instructions to releasethe anchor block, the system willfloat up driven by the buoyancy ele-ments. The MTMI is located at adepth of approximately 250 m, andseveral groups of glass floats providebuoyancy, keeping the submerged

buoy straight together with the anchorat the bottom.

In a general diagram of the mooredsystem (Figure 5), the whole systemweighs 1636.13 kg including approxi-mately 1,000 kg of gravity anchor. Itsstatic buoyancy in water is 190.41 kg,and the mooring line is 300 m long.The type and layout of the instrumentscan be flexibly adjusted accordingto the actual situation. A variety ofocean instruments are hung in thesubmerged buoy, and the distribu-tions of the equipment are ordered ata certain depth.

SamplingThe battery packs are configured

in cyclic sampling, which can supplyenough power to allow the MTMI tooperate at predetermined intervals.The duty cycle of the moored instru-ment is 3 min on and 12 min off.The sampling rate of shear sensor isset to 1,024 Hz, and the MTI sensoris set to 120 Hz, which is mountedby a right-handed coordinate. Thesampling interval of RCM and theSea-Bird CTD system is set to 60 s.

March/A

Data are recorded in 16 CompactFlashmemory cards, with a capacity per cardof 16 GB.

MethodologyData Normalization

According to the requirements ofturbulence observations, platform mo-tion features are analyzed by usingthe experimental data for turbulencedetection; the instrument has beengreatly improved after several tests inthe coast. In the design of the mooredturbulence observation instrument,the system balance and the rotationin response to changes in the directionof flow orientations are two significantparameters; thus, an effective analysismethod is necessary to illustrate therational design of the turbulence plat-form. Accordingly, the instrumentheading is positive counterclockwise,ranging from −180° to 180°, but thesamples of current direction acquiredby RCM are positive clockwise, rang-ing between 0° and 360°. Due totheir different sampling rates andranges between the MTI motion sen-sor and the RCM, we need to normal-ize the two data arrays of the headingand current direction before process-ing and analyzing them. First, thevalues of the heading will be rotated180° to normalize with the current di-rection in the same range of 0°–360°.Second, the sample rates of the head-ing signals will be changed by re-sampling them as averaged in 3-minsegments in which there are morethan 20,000 points. During the re-sampling process, an anti-aliasing orlow-pass finite impulse response (FIR)filter is applied to compensate for thefilter’s delay; it provides an easy-to-use alternative method, relieving theuser’s need to supply a filter or com-pensate for the signal delay introduced

FIGURE 4

Map showing the deployment region and the location of the MTMI at 21°09.90′N, 117°42.03′E(red square).

FIGURE 5

General diagram of the whole mooring systemas deployed in the SCS.


by filtering. Finally, a statistical averagemethod is used to process the normal-ized data between heading and cur-rent direction. Actually, the averagemethod is an arithmetic average, whichhas the advantage of simplicity andpracticality in calculating average values.Although this method may result insome deviation between predicted val-ues and the actual values, it does notaffect long-term observations of track-ing the current direction due to the largeamount of heading data.

Correlation AnalysisThe correlation analysis is a sta-

tistical method to describe statisticalrelationships between two or morerandom variables or observed datavalues. It can be interpreted as a cor-relation coefficient that illustrates aquantitative measure of some type ofcorrelation and dependence. The vari-ations of correlation coefficients arealways between −1 and +1. If two var-iables are perfectly related in a positivelinear sense, the correlation coefficientis equal to +1. On the contrary, a cor-relation coefficient of −1 indicates thattwo variables are perfectly related in anegative linear sense. When the coeffi-cient is zero, it suggests that there is nolinear relationship between the twovariables. Different values of correla-tion coefficients indicate the differentrelated levels between two variables(Cheng et al., 2006; Sandomirski,2015). To develop the MTMI’s flexi-bility to track the changes of currentdirection, canonical correlation analysis(CCA) is introduced to analyze thecorrelation between the current direc-tion and heading. If the obtained datatend to co-occur in two arrays, they arehighly correlated and tend to be com-bined into a single dimension in thenew space. We can try to find projec-tions of each representation separately


such that the projections are maximallycorrelated. The correlation coefficient ρis given as in Equation (1).

ρX ;Y ¼ corr X ;Yð Þ ¼ cov X ; Yð ÞσXσY

ð1Þ

where X and Y are two variables inmatrices or vectors of the same size.Conventionally, the absolute value ofa correlation coefficient above 0.5 in-dicates that the related level betweentwo variables is significant (Lai & Fyfe,2012). Through the correlation anal-ysis between heading and current direc-tion, it is easy to deduce results aboutthe flexibility of theMTMI to track thevariation of current direction.

Results and DiscussionTo evaluate the effectiveness of a

moored turbulence observation plat-form, the stability of the instrumentat a fixed depth; the angle of attack(AOA), interpreted as the attitude ofpitch and roll (Moreau et al., 2005);and the flexibility of tracking currentdirection are crucial parameters thatneed to be estimated as accurately aspossible for high quality of dissipationrates. In short, the accuracy of mea-sured turbulence data is affected bythe stability of the moored system,the platform motion behavior underappropriate environmental conditions,and the flexibility.

Stability of the PlatformThe stability of the moored plat-

form mainly includes the variationof depth and pitch with respect tocurrent speed. Thus, the stability ofthe system is described and analyzedthrough measured data: depth andpitch, where depth data are acquiredby the CTD system and pitch is mea-sured by MTI sensor. Ideally, the

l

MTMI should be fixed at a pointabove the anchor. Actually, the instru-ment is forced by the buoyancy andtension from the mooring line. Theevolution of depth and pitch with cur-rent speed will stabilize the MTMI.On the one hand, the pitch of the in-strument is unaffected by the varia-tions of current speed. With furtherincreases in speed, the pitch rangesabout 0° ± 15° (Figure 6a, left). Statis-tics suggest that, for 85% of the dataset throughout the whole observationperiod, the pitch reaches zero, whichindicates that the orientation of theplatform is nearly always stable forlong-term turbulence measurements.On the other hand, the average depthof MTMI is about 247m for zero flow.With the current speed increasing, thedepth of the instrument ranges from247 to 270 m (Figure 6b, right). Theinstrument is located at a fixed depthof approximately 250 m because themooring line between the anchor andthe main float body is extremely tightand the depth limitation of MTMI isdetermined by the quality of syntacticfoam. In summary, the pitch is unaf-fected by current speed from large-scaletime series, and the platform is alwaysmoored at an approximate depth of250 m. The whole moored system hassatisfied the measured requirements forstability with an appropriate design forturbulence observations.

Platform Behaviorand Accelerations

The acceptable quality of turbulencemeasurements is limited to a smallsubset of environmental conditions—the flow speed, the AOA, and plat-form vibration accelerations. The timeseries of current speed measured bythe RCM instrument during thetime-extended deployment indicatesthat the environmental conditions

ranging from 0.1 to 0.42 m/s is rela-tively stable. In addition, the averagehorizontal current speed past sensorsis about 0.24 m/s (Figure 7a), whichis sufficiently greater than an estimateof the turbulent velocity. A good esti-mate of the AOA time series felt byshear probes will allow for reliable cal-culation of dissipation rates (ε) of turbu-lent energy. When the AOA is smallerthan ±20°, the output of shear probescan respond linearly to cross-stream ve-locity fluctuations times the horizontalvelocity U (Paskyabi & Fer, 2013).In this article, the AOA can be inter-preted as the instrument behavior thatis gauged by the variations of pitch androll (Lueck & Huang, 1999). Time se-ries of pitch and roll is roughly consid-ered as constant with the averagevalues of θ = 4.5° and ϕ = 5.44°, re-spectively, in 6 days (Figure 7b).Throughout the long-term observationperiod, the variations of pitch and rollof the instrument are relatively at zerofor large-scale turbulent measurements.The vibrations and accelerations mea-sured by MTI sensor are the other pa-rameters to describe the behavior of theplatform and environmental forcing.When the accelerations are scaled by

current speed, the signals from the lon-gitudinal (Ax) and athwartship (Ay) ac-celerations (Figure 7c) divided by thespeed are always small compared withthe shear probe signals.

The Flexibility of Tracking FlowThe alignment of the instrument

with the ambient current direction iscritical for successful turbulence mea-surements. We originally intended

March/A

to compare the instrument headingagainst the current direction reportedby the RCM. The RCM is rigidlyfixed to the instrument, and one axisis always parallel to the shaft of theshear probe, which allows for reliableestimates of turbulence. The abilityof the moored instrument to trackflow direction refers to the flexibilityof the heading rotating in response tovariation of flow orientations, which isused to describe the state of the mea-suring platform. The angle of approx-imately 180° between heading andcurrent direction indicates that the in-strument heading keeps counter to thestate of aligning with the coming cur-rent direction; that is to say, the axisheading should always be parallel tothe changes of current direction (Fer &Paskyabi, 2014; Lueck et al., 1997).Therefore, the flexibility of the MTMIresponding to the changes of currentorientations depends on the instru-ment heading and the mean ambientcurrent direction. From the perspectiveof Figure 8, the instrument heading isaligned with the ambient current di-

FIGURE 6

Large-scale time series of pitch and depth in comparison with current speed. (a) Time series ofpitch with current throughout the whole observation period of 115 days. (b) The location of themoored instrument in SCS for 115 days is assessed by depth data measured by the CTD systemand current speed measured by the RCM instrument.

FIGURE 7

Time series of the attitude sensor and accelerations in January 13–18, 2014, in the SCS experi-ment. (a) Horizontal current speed measured by RCM. (b) Pitch and roll measured by MTI motionsensor. (c) The longitudinal (Ax) and athwartship (Ay) accelerations.


rection, opposing the incoming flow(Figure 8a). For comparison with cur-rent direction, the heading has beensubtracted 180°, where the horizontalaxis of time series is expressed in daysand the vertical axis refers to the de-gree of direction. The black (solid)line is almost overlapped by the grayline (Figure 8b). The absolute differ-ence is lower than 10°, satisfying therequirement of AOA (Fer & Paskyabi,2014). In addition, the correlationcoefficient between the heading andcurrent direction is 0.66 suggestingthat their similarity is 66%. The dif-ference between the current directionand the axis of heading is small,which indicates that the mooredinstrument is applicable to measureocean turbulence.

The MTMI has been deployed fora total of 115 days in the SCS fromOctober 19, 2013, to February 10,2014. In addition to some incompleteand sporadic data for 18 days, there area total of 97 days with complete datasamples. A large proportion of correla-tion coefficient is focused on the range


of 0.50–0.9, and the proportion of allthe coefficients is shown in Table 1.For 62% of the data set, the correlationcoefficient is above 0.5, indicating asignificant correlation relationship be-tween the heading and current direc-tion, and 40% is of high correlation(Table 1). The larger correlation coef-ficients, the stronger the correlationbetween the measured heading andcurrent direction will be. The resultsillustrate that the instrument head-ing of MTMI has a good flexibilityin response to the changes of currentdirection. In short, under the given

l

mechanical design, the instrumentscan response flexibly to the variationsof current and consistently follow thecurrent direction. Throughout thewhole observation period, the correla-tion coefficients between heading andcurrent direction are above 0.5 for 62%of the data set, which means that 62%of the turbulence data are effectivelyavailable for the research.

Spectra AnalysisThe performance of the moored

instrument for measuring turbulentshear velocity is gauged by examiningsamples of unprocessed data.Generally,wavenumber spectrum is a commonlyused method to evaluate the character-istics of turbulence. The shear spectraΦ( f ) in frequency domain are con-verted into the wavenumber domainby using the Taylor’s frozen turbulencehypothesis, where Φ(k) = UΦ( f ) andk = f /U; U is the mean flow velocity.The dissipation rate of the TKE is cal-culated by integrating the wave num-ber spectrum for assuming isotropicturbulence (Voulgaris & Trowbridge,1998).

εi ¼ 7:5v∂ui∂z

� �2

¼ 7:5vðkck0

ψi kð Þdk m2s−3� � ð2Þ

FIGURE 8

The flexibility of the heading rotating in response to variations of the current direction. (a) Currentdirection measured by RCM (black), and the instrument heading measured by MTI attitude senor(gray). (b) Current direction (black) in comparison with the heading, which has been subtracted180° (gray). The correlation coefficient between the heading and current direction is 0.66.

TABLE 1

The proportion of correlation coefficients and the corresponding degree of correlation between theinstrument heading and current direction in 97 days.

Correlation Coefficients (−)
Proportion (%) Degree of Correlation (−)
0–0.3
19 Weak positive correlation
0.3–0.5
19 Low correlation
0.5–0.7
22 Significant correlation
0.7–0.9
36 High correlation
0.9–1.0
4 Strong correlation

where ψi(k) is the shear spectrum mea-sured by shear probe 1 (i =1) and shearprobe 2 (i = 2), v (v ≈ 1.64 × 10−6m2 s−1)is the kinematic molecular viscosity asa function of the local water tempera-ture, k denotes the wavenumber, andthe overbar denotes a spatial average.The k0 is the lower integration limit,and the kc is the Kolmogorov wavenumber, where kc = (ε/v

3)1/4. The em-pirical model for the turbulence spec-trum determined by Nasmyth (1970) isutilized to set the lower and upper inte-gration limits of the spectrum (Oakey,1982; Zhang & Moum, 2010). Theacceptable measurements of dissipationrate are obtained directly by using theshear probes.

In the wavenumber domain, thevariance of the shear resides mainly atwavenumbers between 1 and 100 cyclesper meter (cpm; wavelength of 0.01 to1 m). The noise signals are mainly in-duced by turbulence eddies, and thephenomenon of a Karman vortex streetexcludes the disturbance of gravitywaves. When the moored instrumentis located in the deep ocean to observeturbulence, viscous fluids flow throughthe bluff body, and the phenomenonof a Karman vortex street is generatedbehind a circular cylinder in the water

flow field. In short, this phenomenon isindirectly caused by interaction of themean flow with the cable, and resultsin vibration signals. The acceleration-coherent vibration can be removedfrom the shear signals with the tech-nique deduced by Goodman et al.(2006). The rate of dissipation has de-creased approximately two times forthe mean flow velocity U = 0.25 m s−1.Both the raw shear spectra (Figure 9,Sh1 and Sh2) and the cleaned spectra(Figure 9, Sh1-clean and Sh2-clean)conformwell to the theoreticalNasmythspectrum in shape and level. The peakof noisy signals in the shear spectra atapproximately 24 cpm induced by theKarman vortex street is partly removedusing a coherent noise removal algo-rithm, further proving that the platformis stable for measuring turbulence.

ConclusionsThe MTMI was successfully de-

ployed at a depth of approximately250 m in the SCS. It was demon-strated that effective turbulence datacan be measured and collected usingthe MTMI developed by OUC at afixed depth for a long-term observa-tion. The results show that 85% of

March/A

the pitch is unaffected by variationsof the current speed from the large-scale time series. For the whole obser-vation period, 62% of the correlationcoefficients is over 0.5, demonstratingthe significant correlation betweenheading and current direction. Withthe given mechanical design, the in-strument always keeps a counterflowstate and can respond flexibly to thechanges of current direction. Themag-nitude of dissipation rates of TKEestimated from the shear probes isapproximately Ο(10−8) W kg−1 at aknown mean flow speed U = 0.25 m/s,and the power spectrum curves matchclosely with the empirical Nasmythspectrum. The newly designed instru-ment MTMI has proven to be aneffective platform for long-term mea-surements of microstructure turbu-lence. In addition, to further improvethe stability and flexibility of mooredsystem, the design of the wings needsto be improved in some aspects, suchas altering the materials or increasingthe area of the wings.

AcknowledgmentsThis research was financially sup-

ported by the National High Tech-nology Research and DevelopmentProgram of China (“863” Program)under Grant No. 2012AA090901and Grant No. 2014AA093404, theNational Nature Science Foundationunder Grant No. 61203070, andthe Nature Science Foundation ofShandong Province under Grant No.ZR2012DQ011. We especially thankchief scientist Ji-Wei Tian and profes-sor Z. Daniel Deng for their guidance.We would also like to thank the OceanUniversity of China for the platform ofthe vessel Dong Fang Hong 2 and theentire research and development teamfor their help with this experiment.

FIGURE 9

Shear spectra of dissipation rates (ε) for 4-s-long segment, where x axis is expressed in wave-number domain and the y axis is the power density spectrum (PSD). The black and gray lines arethe raw shear data and cleaned data of shear probes 1 and 2, the dotted curves are the Nasmythspectrum, and the vertical dashed line is the Kolmogorov cutoff wavenumber of the integration.


Corresponding Author:Dalei SongCollege of Engineering,Ocean University of ChinaEmail: [email protected]

ReferencesCheng, Q.M., Jing, L.H., & Panahi, A. 2006.

Principal component analysis with optimum

order sample correlation coefficient for

image enhancement. Int J Remote Sens.

27(16):3387-401. http://dx.doi.org/10.1080/

01431160600606882.

Fer, I., & Paskyabi, M.B. 2014. Autonomous

ocean turbulence measurements using shear

probes on a moored instrument. J Atmos

Ocean Tech. 31(2):474-90. http://dx.doi.org/

10.1175/JTECH-D-13-00096.1.

Goodman, L., Levine, E.R., & Lueck, R.G.

2006. On measuring the terms of the turbu-

lent kinetic energy budget from an AUV.

J Atmos Ocean Tech. 23(7):977-90. http://

dx.doi.org/10.1175/JTECH1889.1.

Lai, P.L., & Fyfe, C. 2012. Kernel and

nonlinear canonical correlation analysis. Int J

Neural Syst. 10(5):365-77. http://dx.doi.org/

10.1142/S012906570000034X.

Lueck, R., & Huang, D. 1999. Dissipation

measurement with a moored instrument in

a swift tidal channel. J Atmos Ocean Tech.

16(11):1499-505. http://dx.doi.org/10.1175/

1520-0426(1999)016<1499:DMWAMI>2.0.

CO;2.

Lueck, R., Wolk, F., Hancyck, J., & Black, K.

2015. Hub-height time series measurements of

velocity and dissipation of turbulence kinetic en-

ergy in a tidal channel. In: Conference of Current,

Waves, Turbulence Measurement Workshop

(CWTM). Petersburg, FL: IEEE. http://

dx.doi.org/10.1109/CWTM.2015.7098143.

Lueck, R.G., Huang, D., Newman, D., &

Box, J. 1997. Turbulence measurement with a

moored instrument. J Atmos Ocean Tech.

14(1):143-61. http://dx.doi.org/10.1175/

1520-0426(1997)014<0143:TMWAMI>2.0.

CO;2.


Lueck, R.G., Wolk, F., & Yamazaki, H.

2002. Oceanic velocity microstructure mea-

surements in the 20th century. J Oceanogr.

58(1):153-74. http://dx.doi.org/10.1023/

A:1015837020019.

Moum, J.N., Gregg, M.C., Lien, R.C., &

Carr, M.E. 1995. Comparison of turbulence

kinetic energy dissipation rate estimates

from two ocean microstructure profiles.

J Atmos Ocean Tech. 12:346-66. http://

dx.doi.org/10.1175/1520-0426(1995)

012<0346:COTKED>2.0.CO;2.

Moreau, S., Roger, M., & Jurdic, V. 2005.

Effect of angle of attack and airfoil shape on

turbulence-interaction noise. In : 11th AIAA/

CEAS Aeroacoustics Conference. pp.1-20.

Monterey, CA: Airfoil Noise. http://dx.doi.

org/10.2514/6.2005-2973.

Nasmyth, P.W. 1970. Oceanic turbulence.

Ph.D. thesis. Department of Physics, Univer-

sity of British Columbia. 69 pp.

Nikurashin, M., & Ferrari, R. 2010. Radia-

tion and dissipation of internal waves generated

by geostrophic motions impinging on small-

scale topography: Application to the Southern

ocean. J Phys Oceanogr. 40(5):1055-74.

http://dx.doi.org/10.1175/2010JPO4315.1.

Oakey, N.S. 1982. Determination of the

rate of dissipation of turbulent energy from

simultaneous temperature and velocity

shear microstructure measurements. J Phys

Oceanogr. 12(3):256-71. http://dx.doi.org/

10.1175/1520-0485(1982)012<0256: DOT

ROD> 2.0.CO;2.

Paskyabi, M.B., & Fer, I. 2013. Turbulence

measurements in shallow water from a sub-

surface moored moving platform. Energy

Procedia. 35(9):307-16. http://dx.doi.org/

10.1016/j.egypro.2013.07.183.

Sandomirski, S.G. 2015. Effect of measure-

ment accuracy and range of variation of a

physical quantity on the correlation coeffi-

cient. Meas Tech+. 57(10):1113-20. http://

dx.doi.org/10.1007/s11018-015-0588-3.

Song, D.L., Sun, J.J., Xue, B., Jiang, Q.L., &

Wu, B.W. 2013. Mooring system of ocean

turbulence observation based on submerged

l

buoy. China Ocean Eng. 27(3):369-78. http://

dx.doi.org/10.1016/j.egypro.2013.07.183.

Thorpe, S.A. 2005. The Turbulent Ocean.

Cambridge, UK: Cambridge University Press.

http://dx.doi.org/10.1017/CBO9780511819933.

Voulgaris, G., & Trowbridge, J.H. 1998.

Evaluation of the Acoustic Doppler Velocim-

eter (ADV) for turbulence measurements.

J Atmos Ocean Tech. 15(15):272-89.

http://dx.doi.org/10.1175/1520-0426(1998)

015<0272:EOTADV>2.0.CO;2.

Wang, Y.F., Luan, X., Han, L., Yang, H.,

Wang, S.X., & Hou, G.J. 2014. A de-noising

algorithm of turbulence signal based on

maximum likelihood. Mar Technol Soc J.

48(6):42-51. http://dx.doi.org/10.4031/MTSJ.

48.6.1.

Zhang, Y.W., & Moum, J.N. 2010. Inertial-

convective subrange estimates of thermal

variance dissipation rate from moored tem-

perature measurements. J Atmos Ocean Tech.

27(11):1950-9. http://dx.doi.org/10.1175/

2010JTECHO746.1.

P A P E R

Study and Design of a Heat Dissipation Systemin a Junction Box for Chinese ExperimentalOcean Observatory Network
A U T H O R SDejun LiJun WangJianshe FengCanjun YangYanhu ChenState Key Laboratory of Fluid Powerand Mechatronic Systems,Zhejiang University
A B S T R A C T
Good heat dissipation is highly significant for long-term reliable functioning of a
junction box (JB) for a cable ocean observatory network. An innovative heat dissipa-tion system consisting of circumferentially equi-spaced chassis heat sinks and anadaptive elastic support structure is proposed based on the analysis of the coolingmechanism of the JB. A temperature field model of the JB is established. The param-eters include seawater pressure, precision of machine workpieces, seawater flowrate, and thermal contact resistance. A 3D dynamic simulation of heat dissipationis conducted using the commercial software ANSYS to describe the temperaturefield inside the JB. The proposed heat dissipation system is applied and testedon a laboratory setup of the Experimental Underwater Observatory Network Systemof China. Experimental results agree well with the theoretical model and simulatedresults. The increase in maximum temperature in the JB is between 7°C and 10°Cunder different sea conditions.Keywords: heat dissipation system, ocean observatory network, junction box,temperature field model, contact thermal resistance

oping a deep understanding of the roleof oceans and elucidating the complex

Introduction

The ocean covers 70%of the earth’ssurface and extensively influences theenvironment. However, the nature ofthe ocean is poorly understood. Devel-

physical, biological, chemical, and geo-logical processes operating therein rep-resented a major challenge in the early21st century (Favali et al., 2015). Ca-bled ocean observatory networks havebeen installed to helpmeet the challenge.These networks provide abundantpower and high bandwidth commu-nications to remote-controlled sensorsand scientific instruments which gatherreal-time data and imagery. Some rele-vant cabled ocean observatory networkshave been developed and deployedaround the world. North-East PacificTime-series Undersea NetworkedExperiments (NEPTUNE) Canada,Monterey Accelerated Research Systemin America, Dense Ocean Network forEarthquake and Tsunamis in Japan,and European Seafloor ObservatoryNetwork in Europe can provide long-

term, real-time, and continuous data(Chave et al., 2004; Priede et al.,2004; Howe et al., 2006; Barnes &Tunnicliffe, 2008; Kawaguchi et al.,2008; Aguzzi et al., 2011). A similarnetwork, the Experimental SeafloorObservatory Network System, hasbeen developed and will be deployedin the South China Sea in the near fu-ture. This network is the first attemptat building a regional cabled observa-tory network in China. The structureof the cabled ocean observatory net-work, which consists of a shore station(SS) layer, junction box (JB) layer, andterminal science instrument (SI) layer,is shown in Figure 1. Backbone andwater-tight cables are used to connectthe different layers.

The JB is the intermediate process-ing unit between the SS and SIs on

March/A

the seabed. This layer is responsiblefor power conversion, communicationtransmission, and time synchroniza-tion. Long-term working reliability ofthe JB is essential to the normal oper-ation of the entire observatory networkunder the extreme environment of thedeep sea (Chen et al., 2012). In thisstudy, the JB is divided into two sec-tions, namely, power conversion andcontrol cavities. The power conversionvessel is mainly used to realize powerconversion and distribution for thewhole underwater network. Powerconsumed by heat dissipation of eachpower conversion vessel, which is lim-ited by the power conversion efficiency,reaches hundreds of watts at full loadoperation. Electronic components andinstruments generally need a compara-tively large and ventilated environment


for effective heat dissipation.However,the submarine JB is usually designedas a sealed cylindrical cavity. Compactstructure and small space occupancy ispreferred because of deep-sea workingconditions and layout requirements.Therefore, traditional terrestrial heatdissipation technologies cannot be ap-plied in the JB because of the restrictivecircumstances. A proper and innova-tive heat dissipation scheme is designedfor the submarine JB through compre-hensive heat transfer analysis to meetthe demands for a highly reliable andlong-term operation.

The paper includes six sections.Heat dissipation mechanism and struc-tural design of the JB is presented inHeat DissipationMechanism and Struc-tural Design of the JB. Model for theThermalConduction of the JB addressesthe thermal conduction of the JBmodel,whichmainly includes themathematicalmodel of the temperature field, calcula-tionmodel for contact thermal resistance,and computation of the convective heattransfer coefficient. Software simula-tion analysis is presented in SoftwareSimulation Analysis. Experimental re-sults are shown and analyzed in Exper-imental Results. Conclusion concludesthe study.


Heat DissipationMechanism and StructuralDesign of the JB

The majority of the studies on theheat dissipation of deep-sea mechan-ical and electrical equipment havefocused on the fields of remotely oper-ated vehicles, autonomous underwatervehicles, and manned submersibles(McDonald & Naiman, 2002; Petittet al., 2004). Only a few studies haveinvestigated the heat dissipation of asubmarine JB, and hardly any tradi-tional terrestrial mature heat dispersaltechnique can be applied directly. Inthis study, a heat dissipation systemthat refers to the submersible Alvinof the Woods Hole OceanographicInstitution is designed by consideringthe heat dissipation mechanism, thepractical necessities of the JB, and thecharacteristics of the surrounding en-vironment. The JB is surrounded bycold water, which can be used to dis-perse the heat on the exterior wall ofthe vessel through heat convection.The solid-state electronic componentsand other heat source componentsshould be mounted onto the interiorwall of the cylindrical vessel to transferheat via direct contact, which is theo-

l

retically the fastest means to dissipateheat. However, this scheme is difficultto implement and cannot guaranteesystem reliability. A good heat transfermedium between heat sources and theinterior cylinder wall of a vessel withhigh thermal conductivity is designed.Thus, an effective heat dissipationscheme for the JB is realized.

The heat dissipation pathway fromthe internal distribution heat sourceof the JB to the external seawater is asfollows: heat source → interlayer heattransfer medium → cylinder wall ofthe JB → seawater. Figure 2 shows thethermal resistance network of the de-signed pathway. Rmi and Rju are thethermal resistances of the interlayerheat transfer medium and cylinder wallof the JB, respectively. Rso-mi representsthe contact thermal resistance betweenthe heat source and the interlayer heattransfer medium. Rmi-ju depicts the con-tact thermal resistance between the in-terlayer heat transfer medium and theinterior cylinder wall of the JB. Rju-amshows the contact thermal resistancebetween the exterior wall and seawater.Rso-mi , Rmi-ju , and Rmi are related tothe thermal conductivity coefficientof the interlayer heat transfer medium.In the “Observatorio Submarino Ex-pandible” (OBSEA) JB (del Río et al.,2013), air is used as the primary me-dium for heat transfer from the elec-tronic components to the interior wall.Rso-mi , Rmi-ju , and Rmi represent theheat resistance of the natural convectionbetween air and heat sources, betweenair and the interior wall of the vessel,and the heat resistance of air itself, re-spectively. The JB of NEPTUNE usesmetal, which has considerably higherthermal conductivity than air, as theinterlayer heat transfermedium and ob-tains good heat dissipation effect. Theelectronic components are mountedonto a chassis (i.e., an invertedT-shaped

FIGURE 1

The structure of the cabled ocean observatory network. (Color version of figures are availableonline at: http://www.ingentaconnect.com/content/mts/mtsj/2016/00000050/00000002.)

structure made of aluminum) for heatdissipation. However, the heat trans-fer pathway of the T-shaped structureis relatively long, and all the electroniccomponents mounted onto the samestructure centralize heat sources at ahigh local temperature (Woodroffeet al., 2008).

In the proposed design, four cir-cumferentially equi-spaced aluminumchassis heat sinks that coupled wellwith the interior cylinder wall of theJB are designed to disperse heat. Fig-ure 3 shows the structure of the JBwith part of the heat sinks and the in-ternal electronic components. Powerconverters and other heat sourcesare fixed onto the heat sinks. High-precision mechanical processing ofthe heat sink curved base can matchproperly with the interior cylinderwall of the vessel and allow verticalheat transfer between heat sources andseawater. Microdeformation occurswhen the JB is deployed on the sea-bed at a water depth of 2,000 m. Themaximum deformations on the radialand axial directions are approximately

0.225 and 0.732 mm, respectively, ac-cording to the finite element analysis.The deformation patterns of the vesselare shown in Figure 4. Therefore, anadaptive elastic support structure isdesigned to push the heat sinks thatare securely attached onto the interiorwall of the vessel. The supportingstructure can enable the heat sinks toadapt to wall deformation by flexingthe springs. This design increases thecooling area of the power converters,shortens the cooling path, and enlargesthe available space. Furthermore, com-pared with the oil immersion heat dis-sipation method, the proposed methodavoids the risks of oil immersion of theelectronic components and oil expan-sion when the oil is heated during theheat transfer process.

March/A

Model for the ThermalConduction of the JB

Figure 5 shows the heat dissipa-tion process of the JB, in which thepower converters initially dissipate heatthrough the chassis heat sinks. Thehigh machining accuracy of the con-tact surfaces and the high heat conduc-tivity of the aluminum chassis heat sinksmake the contact thermal resistanceRso-mi sufficiently small, and thus, itcan be disregarded. The thermal resis-tance of the middle-layer medium(chassis heat sinks), Rmi, significantlydecreases because of the direct verticalheat dissipation path from which theheat sources move outward. Contactthermal resistance Rmi-ju between thecurved base and the interior wall ofthe vessel is affected by machiningaccuracy and normal pressure. There-fore, increasing machining accuracyand spring length can reduce Rmi-ju.The values of the thermal resistanceof the titanium wall of the vessel Rjuand the heat resistance Rju-am betweenthe exterior wall and seawater are fixed.The proposed design can optimize thedissipation performance of the JB.

FIGURE 2

Heat dissipation pathway of the JB.

FIGURE 3

Structure of the JB.

FIGURE 4

The deformation patterns of the vessel.


Mathematical Model of theTemperature Field of the JB

Establishing an accurate model forthe temperature field is difficult be-cause of the complex structure of theJB. The geometry of the 1/4 cross sec-tion of the JB is depicted in Figure 6(a).The squeeze method is adopted to sim-plify the structure of the JB into two2D models that consist of multilayercylinder walls (Figures 6(b) and 6(c)).

The cooling condition of the actualstructure is obviously worse than that ofthe structure depicted in Figure 6(b)but better than that of the structuredepicted in Figure 6(c). As shown inFigure 6(b), the heat sinks are evenlydistributed on the interior wall of thecylinder and the thickness of the chassisheat sink is decreased. These charac-teristics reduce thermal resistance. Bycontrast, Figure 6(c) shows that heatsinks are unevenly distributed on theinterior wall of the cylinder and thatthe thickness of the chassis heat sinkis increased. These characteristics en-hance thermal resistance. The tempera-ture field models for the cross section inboth structures are established accord-ing to the heat conduction differentialequations and boundary conditions.Thus, the temperature range of the ac-tual structure is determined.

FIGURE 5

Heat dissipation of the JB.


The heat conduction differential equation of the cylindrical coordinate sys-tem is as follows (Kays et al., 2012):

ρc∂T∂τ

¼ 1r∂∂r

kr∂T∂τ

� �þ 1r2

∂∂φ

k∂T∂φ

� �þ ∂∂z

k∂T∂z

� �þΦ

: ð1Þ

The temperature field of the JB is in a steady state, so ρc∂T

. The heat source
∂τ
has a steady conduction, and the heat power density is Φ: ¼ 0. The problem is

simplified to 1D heat conduction along the radial direction when the symmetryof the structure is considered. The simplified heat conduction differential equa-tion can be written as follows:

ddr

rdTdr

� �¼ 0 ð2Þ

The analytical solution is as follows:

T ¼ C1 ln r þ C3 r0 < r < r1−ð ÞC2 ln r þ C4 r1þ < r < r2ð Þ

�ð3Þ

where C1, C2, C3, and C4 are all constants.The boundary conditions are as follows:

Φ ¼ �λ12πr1ldTdr

��r¼r1−

ð4Þ

Φ ¼ �λ22πr1ldTdr

��r¼r1þ

ð5Þ

FIGURE 6

(a) The geometry of 1/4 cross section of the JB. (b) 1/4 cross section of simplified model I. (c) 1/4cross section of simplified model II.

l

T r1�ð Þ � T r1þð Þ ¼ Rmi�juΦ ð6Þ

Φ ¼ hA T r2ð Þ � T∞½ � ð7Þ

The constants in Eq. (3) are written as follows:

C1 ¼ �Φ2πλ1l

;C2 ¼ �Φ2πλ2l

;C3 ¼ Φ ln r12πλ1l

þΦ ln r2=r1ð Þ2πλ2l

þ Φ2πr2lh

þ Rmi−juΦþ T∞;

C4 ¼ Φ ln r22πλ2l

þ Φ2πr2lh

þ T∞

where λ1 is the heat conductivity coefficient of aluminum, λ2 is the heat conduc-tivity coefficient of titanium, l is the length of the chassis heat sink, h is the con-vective heat transfer coefficient between the outside wall of the JB and seawater,T∞ is the temperature of seawater, and Rmi-ju is the contact thermal resistance be-tween the curved base of the chassis heat sink and interior wall of the vessel. Theinterface that between the curved base and the interior wall of the vessel cannotmeet continuous conditions when deformation of the vessel happens in deep seaor influenced by the machining precision of the heat sinks and vessel. Therefore,the contact thermal resistance Rmi-ju is introduced. Φ, r1, r2, λ1, l, and T∞ areall known. Hence, the exact temperature field models of two simplified structurescan be established when Rmi-ju and h are obtained. The temperature range of theactual structure is also obtained.

Calculation Model for Contact Thermal Resistance Rmi-ju

Thermal contact conductance is indispensable for the heat dissipation of theJB. The heat transfer coefficient can be obtained via assembly tests. However, theobtained value will be conservative because surface pattern parameters and contactconditions vary considerably under diverse circumstances. An appropriate and ef-fective established model can predict the contact heat resistance Rmi-ju of the JB.The possible minimum contact heat resistance can then be determined by analyz-ing the model.

The geometric structure of rough surfaces is random, and roughness featuresare identified at a large number of length scales in accordance with the Gaussiandistributionmodel (Majumdar& Bhushan, 1990). Amodel is then established byimplementing random simulations to analyze roughness features using the MonteCarlo method. A numerical model is developed to obtain the contact thermal re-sistance on the macroscopic scale according to the calculation of each single ther-mal resistance on the microscopic scale.

Machined surfaces have asperities that are isotropically and homogeneouslydistributed (Dyachenko et al., 1963). The surface model can be establishedunder the assumption that the asperities are circular cones with equal apex anglesthat rest on a base plane (Leung et al., 1998). Two contact surfaces are equivalentto a rough surface and a smooth plane with a rigid contact (Hsieh, 1974). Char-

March/A

acteristic parameters of the equivalentrough surface are obtained as follows:

σ ¼ffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiσ21 þ σ22

qð8Þ

m ¼ 11m1

þ 1m2

ð9Þ

H ¼ min H1;H2f g ð10Þ

where σ1 and σ2 are the roughnessvalues of the two surfaces, m1 and m2

are the slopes of the circular cones, andH1 and H2 are the hardness values ofthe two surfaces.

A random number matrix is gen-erated using Monte Carlo method tosimulate the peaks of the rough surfacedistribution. The distances betweenthe peaks and the contact plane underthe applied pressure are described asfollows:

Δi ¼ zi � d ð11Þ

where Δi is the distance between thepeaks of circular cones and the contactplane, zi is the summit height of thecircular cones measured from the baseplane, and d is the distance betweenthe contact plane and the base plane.

Figure 7 shows that Δi is onlymeaningful when it is positive, whichimplies the contact between the circu-lar cone on the rough surface and therigid plane. The following parametersare obtained when contact occurs:

ri ¼ Δi=m ð12Þ

ai ¼ πr2i ð13Þ

fi ¼ Hai ð14Þ

where ri is the radius of the singlecontact area, ai is the contact area,


and fi is the applied pressure of thesingle contact area. The distributionof circular cones on the rough surfaceand characteristic parameters on thecontact surface are obtained whenthe characteristic parameters σ andm of the rough surface are provided.

The contact thermal resistance Riof each circular cone is obtained withthe use of a disk model of a singlepoint contact (Gibson, 1976). Figure 8shows the disk model of the singlepoint contact.

The single contact thermal resis-tance is obtained as follows:

Ri ¼ 1� ri=bið Þ1:5=4riλs ð15Þ

where Ri is the single contact thermalresistance, ri is the radius of the singlecontact area, and bi is the radius of thesingle heat channel. λs is the equivalent


thermal conductivity, which satisfiesthe following equation:

1=λs ¼ 1=λ1 þ 1=λ2 ð16Þ

where λ1 and λ2 are the thermal con-ductivities of the two surfaces.

The total contact thermal resis-tance is the sum of the parallel con-nections of all single point thermalresistances. Therefore, the total con-tact thermal resistance is obtained asfollows:

Rmi�ju ¼ 1XN1

1=Ri

ð17Þ

Figure 9 shows that the predictionof the contact thermal resistance by

l

the present computation method con-siderably agrees better with the experi-mental data than those in the literature(Weilan, 1995). This finding con-firms the effectiveness of the method.The curve clusters in Figure 10 showsthe relationship between contact ther-mal resistance and the applied pressureunder various circumstances. The ma-terials of the two surfaces are aluminumand titanium, and the parameter σvaries from 0.025 to 50 μm. Surfaceswith different surface roughness valuesare obtained through various machin-ing processes. Differences in surfacecharacteristics are also observed. Thespecific relationships among surfaceroughness values, machining pro-cesses, and surface characteristics arepresented in Table 1. Heat transfer ef-ficiency is also related to themachiningtolerance between the interior wall ofthe vessel and the curved base of theheat sink. Improving machining toler-ance can result in better contact andfitting, which will increase heat trans-fer efficiency.

The effects of the characteristic pa-rameters, applied pressure, and heat

FIGURE 7

Contact between the equivalent rough surface and the base plane.

FIGURE 8

Disk model for a single point contact.

FIGURE 9

Comparison of the computing model with the experimental results.

conductivity of the rough surfaces oncontact thermal resistance are quanti-tatively analyzed. The method enablesthe determination of the least possiblecontact heat resistance in the JB.The contact thermal resistance Rmi-juobtained in real situations of the JB is6 ×10−3 K/W.

Computation of the ConvectiveHeat Transfer Coefficient h

The flow velocity of seawater incoastal areas in South China Sea is ap-proximately 2 m/s and almost neg-ligible at a water depth of 2,000 m(Yu, 2010). Either forced or naturalconvection therefore occurs betweenthe exterior wall of the JB and sea-water depending on the surroundingenvironment.

The convection heat transfer coef-ficient h between the exterior walland seawater is obtained iteratively.The Grashof number Gr is used asthe criterion to determine the heattransfer condition in natural convec-tion and is given as follows:

Gr ¼ gαvΔtl3

ν2ð18Þ

where g is the gravitational accelera-tion, αν is the expansion coefficientof seawater, Δt is the temperature dif-ference between the exterior wall of theJB and seawater, l is the length of theJB, and v is the viscosity of seawater.

The intensity of the natural con-vection heat transfer for a horizontalsealed cylinder is determined through

March/A

the Nusselt number Num, which isgiven as follows:

Num ¼ C GrPrð Þnm ð19Þ

whereC and n are parameters providedby the experiments and Pr is the Prandtlnumber of seawater.

The convective heat transfer coeffi-cient h is obtained through the Nusseltnumber Num, as follows:

h ¼ Nuλd

ð20Þ

Heat power q is given as follows:

q0 ¼ hAΔt ¼ hπdLΔt ð21Þ

The calculated convective heattransfer coefficient h is inaccurate ifa large difference between the realheat power q and the calculated heatpower q exists. Then, Δt is modi-fied to recalculate h. This recycled pro-cess ends when an appropriate h isobtained.

Finally, we obtain the convectiveheat transfer coefficient h1 in naturalconvection as 383 W/(m2 K) and theconvective heat transfer coefficient h2 inforced convection as 3874 W/(m2 K)when the flow velocity of seawater is2 m/s. The corresponding convectiveheat transfer coefficient is between h1

FIGURE 10

Relationship between contact thermal resistance and applied pressure.

TABLE 1

Relationships between surface roughness values and various machining processes.

Surface Roughness Values
Machining Processes Surface Characteristics
σ = 50, 25
rough turning, rough planning, rough milling, etc. obvious tool mark
σ = 12.5, 6.3, 3.2
finish turning, finish milling, rolling, etc. tool mark is faintly visible
σ = 1.6, 0.8, 0.4
finish grinding, end milling, etc. processing traces is invisible, processing directionis faintly recognizable
σ = 0.2, 0.1
supergrinding, polishing, etc. dim glossy surface
σ = 0.05, 0.025
superfinish bright or mirror glossy surface

and h2 when the seawater flow condi-tion is between the two cases.

Software SimulationAnalysis

The models of parameters Rmi-ju

and h have been previously established,so all parameters needed in the modelof the temperature range of the JB areobtained. The temperature fields ofthe JB under different seawater flowrates are established according to thetemperature range model of the actualstructure of the JB. The length, exter-nal diameter, inner diameter, and wallthickness of the vessel are 876, 310,278, and 16 mm, respectively. Thetotal heat generated inside the vesselis 1,188 W.

Finite element models of the tem-perature field in the JB were built inthe following extreme cases to verifymodel accuracy: (1) seawater was al-most still, and heat convection betweenthe exterior wall of the JB and seawaterwas purely natural convection; and(2) the velocity of seawater was 2 m/s,which is the maximum flow rate in theSouth China Sea. The 3D model forthe JB was established using ANSYSsoftware. Vessel deformation causedby external water pressure could reduceheat transfer efficiency; hence, pressureload was added in the software simula-tion by an equivalent way. Finally, heatsources, boundary conditions, contacttypes, and physical property parameterswere set to calculate the temperaturefield based on a practical situation. Fig-ure 11 shows the contours of the tem-perature distribution of the vessel indifferent environments: Figure 11(a) issimulated at 1 atm, and Figure 11(b) issimulated under 20 MPa pressure.

The models for the temperaturerange of the JB are established whenthe convection heat transfer coefficient


is equal to h1 or h2. The curves in Fig-ure 12 show the relationship betweenthe temperature model and the simula-tion results of the JB. In Figures 12(a)and 12(b), the JB is under 1 atm, andh = h1 and h = h2, respectively. In Fig-ures 12(c) and 12(d), the JB is undera pressure of 20 MPa, and h= h1 andh= h2, respectively.

Figure 12 shows that the simulatedtemperature is within a narrow rangedefined by the two temperature models.A sudden change in temperature is ob-served in all the figures. The existenceof contact thermal resistance Rmi-ju

causes discontinuity in temperature dis-tribution. The following conclusionsare drawn after comparing the figures:the internal temperature of the vesseland the contact thermal resistance be-

l

tween heat sink curved bases and the in-terior cylinder wall increase, the heattransfer efficiency of the system is re-duced, and the cooling performance de-teriorates when the vessel is deformedby external water pressure under the sit-uation in which the convective heattransfer coefficient h is the same value.Thus, contact thermal resistance isessential, particularly when good heatdissipation performance is required.Therefore, the effect of contact thermalresistance should be considered in de-signing the JB and should be reducedto improve the heat dissipation effect.

Experimental ResultsA series of laboratory experiments

was conducted in a water tank and a

FIGURE 11

Contours of temperature distribution in the JB.

sealed pressure case to verify the theo-retical model for the temperature fieldand validate the heat dissipation per-formance of the JB. The JB and therelated instruments were placed in awater tank and a sealed pressure casefilled with simulated seawater, respec-tively, as shown inFigure 13. Six temper-ature sensors were installed to monitorthe temperature at different positionsin the JB. Temperature was obtainedthrough the data acquisition moduleof the lower computer in real time andtransmitted by Ethernet to an upper

computer. The computer communica-tion interface and monitoring softwareshown as Figure 14 was designed tomonitor data changes, display historicalcurves, and control the switches of themultichannel power output.

The temperature of the simulatedseawater was 24°C during the courseof the experiments, and its velocitywas 0 m/s. Therefore, natural convec-tion heat transfer was observed be-tween the exterior wall of the JB andthe simulated seawater. The powerconsumption of the external loads

March/A

reached the maximum capacity thatcould be supplied by the power chan-nels. The temperature monitored in-side the JB gradually increased as theJB started working and eventually sta-bilized after approximately 2 h. In thewater tank test, the heat source tem-perature remained at 33.4°C, whichwas assumed as the highest tempera-ture inside the JB. The temperatureof the outside surface of the heat sinkremained at 32.1°C, whereas the tem-perature of the interior wall of the JBremained at 29.6°C. In the sealed

FIGURE 12

Temperature inside the JB at different radial distances: (a) h = h1 at 1 atm; (b) h = h2 at 1 atm; (c) h = h1 under 20MPa pressure; (d) h = h2 under 20MPapressure.



pressure case test, the correspondingtemperatures in the aforementionedpositions are 34.8°C, 33.4°C, and30.8°C. Figure 15 depicts the compari-sons of the time series of the temperaturedata obtained from the experiments andthe dynamic simulation results fromANSYS in three different positions at1 atm during the experiments and thesimulations. The experimental resultsagree well with the simulated resultsin each position when the external pres-sure of the vessel is 1 atm.However, theexperimental temperatures are slightlyhigher than the simulated values whena water pressure of 20 MPa is loadedonto the outside surfaces of the JB. Webelieve that the effects of the internalradiation and convection of the vessel,which are disregarded in the softwaresimulation but actually exist during thetests, reduce the actual temperature.

Superiority of the structure is veri-fied by experimental results, simulation

FIGURE 13

(a) Experiments in a water tank. (b) Experi-ments in a sealed pressure case.

FIGURE 14

Human machine interface of the JB.

FIGURE 15

Comparison of the experimental and simulated results at 1 atm.

results, and theoretical model analysis.The heat dissipation structure demon-strates good performance. The temper-ature increases by 10°C when the JBis fully loaded and exhibits a naturalconvection heat transfer with seawater.However, the temperature increase by7°C when the JB is fully loaded and ex-hibits a forced convection heat transferwith seawater in which the seawatervelocity is 2 m/s. The good agreementamong the temperature field model,simulated results, and experimentaldata indicates the accuracy of the theo-retical model and simulation analysis.

The electronic components and in-struments are highly reliable in coolenvironments, so the reliability of theentire system of the JB is guaranteed.Seawater flow velocity varies betweenstationary to 2 m/s considering thechanges in sea conditions. Convectionheat transfer coefficient can exhibit the

corresponding changes between h1 andh2. The theoretical model and simula-tion results in Figure 16 show that themaximum temperature of the JB in dif-ferent sea conditions is obtained.

ConclusionThe heat dissipation mechanism

of the JB is established by analyzing itsthermal resistance network accordingto current studies on cooling methodsfor submarine equipment and the struc-tures of OBSEA and NEPTUNE JB.The internal structure inside the JB con-sisting of circumferentially equi-spacedchassis heat sinks and an adaptive elas-tic support structure is designed. Thestructure demonstrates excellent heatdissipation performance.

The discontinuous model of thetemperature range of the JB is estab-lished. The model comprehensively

March/A

considers the deep-sea ambient pressure,machining precision of the workpieces,flow rate of seawater, and thermal con-tact resistances between the curved baseof the heat sinks and the interior wallof the JB. Structural parameters areanalyzed to evaluate their effects onheat dissipation in the JB and optimizethe structure. 3D dynamic simulationsof heat dissipation were conducted todescribe the temperature field in theJB under different ambient pressures.Water tank tests and sealed pressurecase tests of the JB connected to the lab-oratory setup of the Experimental Un-derwater Observatory Network SystemofChinawere run continuously for 120and 24 h, respectively. Experimentaldata curves are consistent with the com-puted results of the theoretical modeland simulated results. The maximumtemperature increase inside the JB is be-tween 7°C and 10°C based on the dif-ferent sea conditions in South ChinaSea. The findings also validate the ex-cellent heat dissipation performance inthe JB. We believe that the structuredesigned in the present work is reason-able and has high thermal reliability.This work proposes a design to guaran-tee long-term reliable operation of a ca-bled ocean observatory network.

AcknowledgmentsThisworkwas supported byTheNa-

tional High Technology Research andDevelopment ProgramofChina (Grants2012AA09A402 and 2012AA09A408)and was supported by the National Nat-ural Science Foundation of China underGrant 51409229, by the Science Fundfor Creative Research Groups of theNational Natural Science Foundationof China under Grant 51221004, andby the Zhejiang Provincial Natural Sci-ence Foundation of China under GrantLQ14E070002.

FIGURE 16

Maximum temperature of heat sources in the JB at different sea conditions.


Corresponding Author:Canjun YangState Key Laboratory of Fluid Powerand Mechatronic Systems, ZhejiangUniversity Hangzhou 310027, ChinaEmail: [email protected]

ReferencesAguzzi, J., Mànuel, A., Condal, F., Guillen,

F., Nogueras, M., del Rio, J.,… Palanques, A.

2011. The new seafloor observatory (OBSEA)

for remote and long-term coastal ecosystem

monitoring. Sensors. 11(6):5850-72. http://

dx.doi.org/10.3390/s110605850.

Barnes, C.R., & Tunnicliffe, V. 2008.

Building the World’s First Multi-node Cabled

Ocean Observatories (NEPTUNE Canada

and VENUS, Canada): Science, Realities,

Challenges and Opportunities. In: Proceedings

of MTS/IEEE Kobe Techno-Ocean, Oceans

2008. pp.1213-20. Kobe: IEEE.

Chave, A.D., Waterworth, G., Maffei, A.R., &

Massion, G. 2004. Cabled ocean observatory

systems.Mar Technol Soc J. 38(2):30-43. http://

dx.doi.org/10.4031/002533204787522785.

Chen, Y.H., Yang, C.J., Li, D.J., & Jin, B.

2012. Design and application of a junction

box for cabled ocean observatories.Mar Technol

Soc J. 46(3):50-63. http://dx.doi.org/10.4031/

MTSJ.46.3.4.

del Río, J., Aguzzi, J., Costa, C., Menesatti, P.,

Sbragaglia, V., Nogueras, M., … Manuel, A.

2013. A new colorimetrically-calibrated auto-

mated video-imaging protocol for day-night

fish counting at the OBSEA coastal cabled

observatory. Sensors. 13(11): 14740-53. http://

dx.doi.org/10.3390/s131114740.

Dyachenko, P.E., Tolkacheva, N.N., Andreev,

G.A., & Karpova, T.M. 1963. The Actual

Contact Area between Touching Surface.

Moscow: Acad Sca. Press.

Favali, P., Beranzoli, L., & Santis, A.D. 2015.

Seafloor Observatories: A new vision of the

Earth from the Abyss. UK: Chichester. 669 pp.

Gibson, R.D. 1976. The contact resistance

for a semi-infinite cylinder in a vacuum. Appl


Energ. 2(1):57-65. http://dx.doi.org/10.1016/

0306-2619(76)90039-8.

Howe, B.M., Chan, T., Sharkawi, M.E.,

Kenney, M., Kolve, S., Liu, C.C., … Siani, C.

2006. Power System for the MARS Ocean

Cabled Observatory. Available from http://

www.interactiveoceans.washington.edu/file-s/

howe_mars_power_ssc06.pdf.

Hsieh, C.K. 1974. A Critical Evaluation of

Surface Geometrical Parameters for a Nominally

Flat Surface Model. J Tribol. 96(4):638-9.

http://dx.doi.org/10.1115/1.3452512.

Kawaguchi, K., Kaneda, Y., & Araki, E.

2008. The DONET: A real-time seafloor re-

search infrastructure for the precise earthquake

and tsunami monitoring. In: Proceedings of

Oceans 2008. pp. 1-4. Kobe: IEEE. http://dx.

doi.org/10.1109/oceanskobe.2008.4530918.

Kays, W.M., Crawford, M.E., & Weigand, B.

2012. Convective heat and mass transfer,

4th ed. New York: Tata McGraw-Hill Edu-

cation. 576 pp.

Leung, M., Hsieh, C.K., & Goswami, D.Y.

1998. Prediction of thermal contact conduc-

tance in vacuum by statistical mechanics.

J Heat Trans. 120(1):51-7. http://dx.doi.org/

10.1115/1.2830064.

Majumdar, A., & Bhushan, B. 1990. Role of

fractal geometry in roughness characterization

and contact mechanics of surfaces. J Tribol.

112(2):205-16. http://dx.doi.org/10.1115/

1.2920243.

McDonald, G., & Naiman, M. 2002. Heat-

transfer advances for submerged oceano-

graphic systems. In: Proceedings of Oceans

2002. pp. 2045-9. Mississippi: IEEE. http://

dx.doi.org/10.1109/oceans.2002.1191946.

Petitt, R.A., Jr., Bowen, A., Elder, R.,

Howland, J., & Naiman, M. 2004. Power

system for the new Jason ROV. In: Pro-

ceedings of Oceans 2004. pp. 1727-31. Kobe:

IEEE. http://dx.doi.org/10.1109/oceans.2004.

1406385.

Priede, I.G., Solan,M.,Mienert, J., Person, R.,

Weering, T., Pfannkuche, O., … Klages, M.

2004. ESONET-European Sea Floor Obser-

l

vatory Network. In: Proceedings of Oceans

2004. pp. 2155-63. Kobe: IEEE. http://dx.doi.

org/10.1109/oceans.2004.1406479.

Weilan, G. 1995. A Method for Calculation

of Contact Heat Resistance between Two Con-

tacting Surfaces. J Aerospace Power. 3:233-6.

Woodroffe, A.M., Pridie, S.W., & Druce, G.

2008. The NEPTUNE Canada Junction

Box-Interfacing science instruments to sub-sea

cabled observatories. In: Proceedings of Oceans

2008. pp. 1-5. Kobe: IEEE. http://dx.doi.org/

10.1109/oceanskobe.2008.4531021.

Yu, X.H. 2010. The Research and Imple-

mentation of Heat Dissipation System for

Seafloor Junction Box. Master thesis, Zhejiang

University. 59 pp.

T E C H N I C A L N O T E

Study of a Novel Underwater CableWith Whole Cable Monitoring
A U T H O R SJiawang ChenChunying XuYing ChenOcean College, Zhejiang University
Jian JiInstitute of Zhejiang MarineDevelopment Research

Dongxu YanOcean College, Zhejiang University

A B S T R A C T
This paper describes a novel concept for detecting and tracking underwater ca-
bles that differs from common detection methods, including visual, acoustic andmagnetic detection, which are based on a survey platform, such as remotely oper-ated vehicles and autonomous underwater vehicles, and have high costs and re-quire long-time monitoring. The proposed detection method can be performed inlocations that are difficult to observe using cameras without the use of magneticfields. The concept relies on low-cost flex sensors that provide underwater, real-time, whole cable detection. Building off of this concept, a detection system andan error analysis system were incorporated, which includes the characteristics ofthe flex sensors, the mounting structure for the flex sensors, and the underwatercable as well as the 2-D cable bend model and a water tank test. The detected dataset, which was rendered on an observation monitor, provided visual feedback re-garding any bending or warping problems. To evaluate the detection accuracy,mean absolute error and similarity analyses were performed using the observationaldata and sensor data obtained from the water tank test. The results of these analysesshowed strong agreement between the fitted curve to the detected data and the ob-servational curve, indicating acceptable error. These results demonstrate the feasi-bility of the proposed system for effective underwater cable detection. Thus, theresults of this investigation clearly reveal the potential of the novel sensing tech-nique for practical underwater cable detection andmonitoring applications. However,the 2-D cable bending detection model is not suitable for 3-D bending or torsion,which will be considered in future research.Keywords: underwater cable detection, 2-D cable bend model, flex sensor

they are exposed to flow scouring, mak-ing them apt to bend or warp, which

Introduction

Underwater cable connections arekey pieces of equipment in topside andsubsea facilities for electro-hydrauliccontrol channel underwater productionunits, which provide underwater com-munications, power, and pulling forcesto marine equipment, marine drag sys-tems, and mooring systems (Szyrowskiet al., 2013a, 2013b). However, under-water cables must operate under roughmarine conditions, and at the seafloor,

could result in underwater device fail-ure. Moreover, excessive bending, oreven breakage, presents a hazard forrescuing personnel and environmentaldamage. A succession of cable damag-ing events that disrupted system per-formance can be noted in Talling et al.(2013) and Carter et al. (2014, 2012).As a result of the multiple functionsthat cables must serve, in addition tothe dynamic underwater environmentin which they are installed and op-erated, their stability and reliabilityfor underwater systems are of utmostimportance.

A common method to reduce therisk of failure is frequent detection andpreventative maintenance. There arethree main common methods of under-water power and telecommunicationcable detection and tracking: visualtracking, acoustic detection, and mag-netic detection (Szyrowski et al., 2013a,2013b; El-Fakdi & Carreras, 2013).

Common visual-based systems forunderwater cable detection on surveyplatforms typically rely on remotelyoperated vehicles (ROVs) or autono-mous underwater vehicles (AUVs)equipped with mounted video cameras

March/A

(Kuhn et al., 2013; Ortiz et al., 2011;Jordán et al., 2011; Ortiz & Antich,2014; Lee et al., 2012; Sakagami et al.,2013). The vision detection process in-cludes cable location search and track-ing. Cable search is the early work ofunderwater cable detection that is per-formed to find the cable position.Cable tracking is the process of mov-ing along the cable while performingthe necessary measurements (Bagnitskyet al., 2011).

Underwater visual detection has acritical limitation: in most cases, thereis little or no ambient light, resulting


in poor visibility. To solve this lim-itation, light sources are required toilluminate the underwater cable. How-ever, images of underwater objectsilluminated by a light source containblurring from the spatial attenuationof higher frequencies and backscatterfrom the more predominant lowerfrequencies (Szyrowski et al., 2013a).Ortiz et al. (2011) identified under-water image characteristics, includingblurring, low contrast, nonuniform il-lumination and instability caused bythe motion of the vehicle, thereby in-creasing the complexity of detection.The clarity of images directly affectsdetection judgment because detectionis based on distinguishing the cablefrom the surrounding environment.Moreover, when a cable has rested onthe seabed for a long period of time,detection is increasingly difficult dueto the rocks and sediment that coverit. Wirth et al. (2008) and Ortiz et al.(2011) adopted a probabilistic ap-proach based on particle filters whichhandle the ambiguities, and they con-cluded that the approach is suitable foronline cable detection.

Inzartsev & Pavin (2008) notedthat visual detection alone was notsufficient for underwater cable detec-tion because video imaging does notprovide consistently reliable data forcable recognition.

Acoustic methods are able to pene-trate the seabed, enabling the detectionof buried utilities (Szyrowski et al.,2013a). Zhou et al. (2015) showedhow sonar detection is based on the re-flected sonar signal of a submarinecable, which includes a sonar shallowprofile. Capus et al. (2010) measuredthe impact of different seabed typeson the cable detection ability and con-cluded that the rougher the sediment,the higher the reverberation level. Thegrazing angle (which is the angle at


which a sonar pulse meets and movesacross the seabed) plays a critical rolein determining the reverberation noiselevel. For rough sediments, a lowergrazing angle provides better results,and vice versa. Brown et al. (2011) con-cluded that detection was also affectedby the cable curvature, tank wall re-turns, ambient noise sources, and dis-turbances in the sediment surface froman experiment in a water tank.

Improving postdetection filter-ing and noise suppression are use-ful methods to minimize these issues(Capus et al., 2010). Mandhouj et al.(2012) noted that most methods re-ducing the noise apply different fil-tering and are often classified in twocategories: methods acting in thespatial domain and those acting inthe transformed domain. Isaacs andGoroshin (2010) noted that the clas-sical approach to cable detection insonar imagery is low-pass filtering,followed by edge detection and the ap-plication of the Hough transform. Inaddition, Villar et al. (2014) indicatedthat the efficiency of online image pro-cessing from acoustic data is also a keyissue for acoustic. These data should beprocessed in an efficient time so thatthe detection system is able to detectand recognize a predefined target.

Acoustic detection cannot be usedfor underwater cables in shallowwater, especially those in coastal re-gions. Interfacing reverberations fromthe air/sea and sea/bottom interfacesand complex seafloor topographicalprofiles present a difficult acoustic en-vironment. Moreover, man-made re-fuse or structures in coastal regionsmay cause a high false alarm ratewhen using the acoustic sensor alone(Pei et al., 2010).

Magnetic detection is an effectivemethod for underwater buried targetdetection and can augment acoustic

l

means of detection, effectively reduc-ing the number of false alarms (Peiet al., 2010). Wang et al. (2011) de-scribed cable detection using magneticfield technology that was based on twobasic principles: an electric current canproduce a magnetic field (electromag-netism) and a changing magnetic fieldwithin a coil of wire induces a volt-age across the ends of the coil (electro-magnetic induction). Induction coilsare based on Faraday’s induction law.Szyrowski et al. (2013b) noted thattwo types of approaches could befound in practical applications of mag-netic detection. The first method is theactive magnetic detection method,which relies on a submarine cable con-ductor operating at a specific AC sig-nal, allowing for the detection of theparameters of the surrounding electro-magnetic signals. This method oftenrequires onshore access to the cable,and the attenuation is large and isthus not suitable for long-distance sub-marine cable detection (Zhou et al.,2009). The second method employspassive magnetic detection, which re-lies on the physical magnetic proper-ties of the submarine cable, does notrequire the signal of the submarinecable to be strengthened, and is suit-able for the detection of complex wa-ters (Zhou et al., 2015).

Underwater cable projects oftenuse a variety of detection equipmentand technologies. Data from visual,acoustic, and magnetic sensors areoften processed together. However,all of the methods described hereinare based on survey devices, such asROVs or AUVs, which are high costand cannot detect the whole cablein real time (El-Fakdi & Carreras,2013). The limitations demonstratethat there is still a great need for amethod that can detect underwatercables with real time and low cost.

To address this need, this paper de-scribes a novel method of using flexsensors to measure the bending anglesof underwater cables that prevent acci-dents caused by cable bending, warp-ing, or breaking at a lower cost thanmost state-of-the-art methods.

To detect the bend state, differentsensors have been investigated: a con-ductive ink-based flex sensor, a novelskin curvature sensor (Kaniusas et al.,2004), an elastomer film embeddedwith a micro-channel of a conductiveliquid (Majidi et al., 2011), strain sen-sors (Lorussi et al., 2005), and capaci-tive or magnetic devices (Fahn & Sun,2010). In this study, the conductiveink-based flex sensors (Spectra SymbolInc.) are adopted because they areflexible, light-weight, and extremelylow-cost sensors that can be appliedto measure the bend state of objectsand have been used for data gloves,medical devices, and so on (Saggioet al., 2009a, 2009b; Saggio, 2014).

The rest of this paper is organized asfollows. First, the concept of underwa-ter cable detection based on flex sensorsis proposed. Second, the characteristicsof flex sensors are discussed. Third, onthe basis of the flex sensors, the mount-ing structure between flex sensors andunderwater cables is presented. Fourth,a 2-D visual bending model is estab-lished to provide guidelines for thesystem performance. Last, a watertank experiment with error analysis isperformed to demonstrate the feasibil-ity of the approach.

A Novel Conceptfor UnderwaterCable Detection

The detection system is illustratedin Figure 1 and uses an ROV ’s umbil-ical cable as an example. Flex sensors(Spectra Symbol Inc.) were arranged

to sense the underwater flexible cablebend angles. Data acquired from thesensors are conditioned and sent to acomputer via a serial port controlledby MCU (stm32), which has three12-Bit ADCs, 112 fast I/O ports,and provides an interface to the com-puter. TheMCU reads all of the sensorangles at a rate of 100 Hz and com-municates with the computer via aRS-232C serial interface at speeds of9,600 baud. The resulting signalswere used to build a 2-D computerbend model including position andbend. The real-time image displaysthe deformation of the underwatercable, providing the underwater cableworking state, and gives the user a vi-sual feedback.

Characteristics of theFlex Sensors

This paper adopted the light-weightand flexible 4.5-inch Spectra SymbolFlex sensors, as shown in Figure 2.

March/A

The flex sensor is a unipolar device,that is, the resistance increases as theflex angle increases in one directionand decreases or is unchanged whenbent in the other direction.

To obtain the characteristics of theflex sensors, a proper measurementsetup was made (Figure 3). The flexsensor was attached to the hingewings, which mounted on the steppermotor axis. One wing of the hinge wasfixed to the central axis. The other onerotated with the motor. The bendingangle was controlled by the motor,and the resistive response of the sensorwas read by a digital multimeter.

The sensor was moved from 0° to90° by increments of 10°. The relation-ship between the bend angle and theresistance of the flex sensor is shownin Figure 4.

Figure 4 shows that the relationshipbetween resistance and bend angle ofthe flex sensor has a nearly linear be-havior only after approximately 30°and some degree of nonlinearity be-tween 0 and 30°. According to thecharacteristics of the flex sensor, thesampling circuit of each flex sensorcan be seen in Figure 5.

The output signal Vout accordingto Eq. 1:

Vout ¼ Vcc � RRf þ R

ð1Þ

wherein R is the resistance value of flexsensor and Rf is the reference resistancethat the value is constant. According toSaggio (2014), in order to assure output

FIGURE 1

System diagram: detection system using anROV’s umbilical cable as an example.

FIGURE 2

Spectra Symbol Flex sensor: length: 4.5 inches, height: ≦ 0.43 mm (0.017 inch), life cycle:> 1 million.


signal Vout with the largest dynamic,the value of Rf could be obtained byEq.2:

Rf ¼ffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiRmax � Rmin

pð2Þ


wherein Rmax and Rmin are providedby the flex sensor under maximumand flat, respectively.

According to the sampling circuit,the relationship between the bendangle and computer sampling valuecan be obtained, as shown in Figure 6.Based on the relationship, the com-puter gets the AD sampling value fromthe sampling circuit and converts it toa bend angle.

Flex Sensors andUnderwater CableMounting Structure

To sense shape effectively, severalflex sensors should be employed. Inthis paper, an experimental under-water cable was divided into five equalsegments. Each segment had two flexsensors placed along the axial sym-metric to provide bidirectional benddetection because the flex sensors areunipolar, that is, one flex sensor hada negative response with respect tothe other (Figure 7). One end of eachsensor is fixed on the tube, while theother end can slide freely along the

l

tube (longitudinal direction) whenthe underwater cable bends, butmovement in any other direction wasrestricted.

2-D Bend Model of anUnderwater Cable

The bend angles of each under-water cable segment were obtainedfrom flex sensor data. The length ofeach underwater cable segment isshort, allowing it to be modeled as acircular arc without twist. In particu-lar, the length of each cable segmentis equal to the sensor length, suchthat the bend angle of the sensor repre-sents the cable segment bend angle.The flex sensors were numberedaccording to the position along theunderwater flexible cable for build-ing model, as shown in Figure 7. Forexample, sensor #12 represents thesecond flex sensor of the first cablesegment.

The underwater cable bend direc-tion must be determined for eachcable segment. Because the flex sensorswere defined relative to a 0 degree

FIGURE 3

Diagram of measurement setup for the sen-sor’s characterization.

FIGURE 4

The characterization of a flex sensor: resis-tance vs. bending angle.

FIGURE 5

The sampling circuit of flex sensor.

FIGURE 6

The relationship between the bend angle and sampling value.

bend, the resistance increases (positive)as the flex angle increases in the cali-brated direction (an inward bend inFigure 2) and is negative or unchangedwhen bent in the opposite direction. Ifthe signs of both sensors in the samesegment are opposite, then the posi-tive sign of sensor data is regarded asthe bend direction of the cable. Other-wise the sensor data are regarded asinvalid.

Assuming that the underwater flex-ible cable was divided into n segments,each with an equal arc length s, from

the arc length formula, each segmentarc radius (R1, R2,…, Rn) can be deter-mined. Figure 8 shows the diagram ofthe cable shape reconstruction for 2-Dbends.

The first segment start point l1 isknown and fixed. Assuming that thedirection of the first segment arc radiusis the same as the direction of unit vec-tor t, the center of first arc O1 is:

O1;x

O1;y

� �¼ l1;x

l1;y

� �± R1⋅ t ð3Þ

where ‘+’ corresponds to a concave arcand ‘−’ corresponds to a convex arc.

When the arc segment is concave,the first segment end point is obtainedby

liþ1;x

liþ1;y

� �¼ Oi;x

Oi;y

� �þ Ri

cos θi þWð Þsin θi þWð Þ

� �

¼ Oi;x

Oi;y

� �þ sθi

cos θi þWð Þsin θi þWð Þ

� �

ð4Þ

wherein w is determined by

w ¼Xi�1

j¼1θj ð5Þ

where θj is positive for a convex arc andnegative for a concave arc.

March/A

The arc center is obtained accord-ing to the tangent slope and is equalbetween the segments junction:

Oi;x

Oi;y

� �¼ li;x

li;y

� �±

Ri

Ri�1

li;x � Oi�1;x

li;y � Oi�1;y

� �

ð6Þ

When the concavity or convexity ofi −1 is the same as that of the i segment,Eq. 6 is ‘−’ and is otherwise ‘+’.

Via Eq. 3 through Eq. 6, the 2-Dbend shape of the underwater cable isdetermined.

Water Tank Test andErrors Analysis

To evaluate the detection systemproposed in this paper, a series of testswere performed in a tank measuring1.5 m × 80 cm × 40 cm. The tank wasequipped with a tape rule for unifiedcoordinate systems of physical rubberand model curve. The cable was con-structed from crylonitrile-butadienerubber that was 25 cm in diameterand 65 cm in length. The rubber wasdivided into five segments; each seg-ment had two flex sensors, as shownpreviously. The lengths of the cable seg-ment and flex sensor were equal, suchthat the sensor bend angles representedthe rubber segment bend, as previouslydiscussed. The rubber with flex sensorswas waterproof and enclosed in bellowsfor use in underwater environments. Inthis paper, five different bend positionswere tested (Figure 9).

The differences between two curvesconsist of degree of similarity andmag-nitude of the errors. So a similaritycomparison and mean absolute errorwere used for evaluating the dif-ferences between fitted curves to thedetected data and the observationalcurves in this study. The similarity cal-culation made use of the cosine value

FIGURE 7

The flex sensors and underwater flexible cablemounting structure.

FIGURE 8

Diagram of the cable shape reconstruction for2-D bends: O1, O2, … , On are the centers ofeach cable segment; l1, l2, … , ln representthe start point of cable segment; R1, R2, … ,Rn are the radii; and t is the radius unit vectorof the first segment.


of the vector angle model (Eq. 7), as itreflected the directional differencesbetween two curves.

sim X ; Yð Þ ¼

Xni¼1

XiYiffiffiffiffiffiffiffiffiffiffiffiffiffiffiXni¼1

X 2i

s ffiffiffiffiffiffiffiffiffiffiffiffiffiXni¼1

Y 2i

s ð7Þ

sim (X, Y ) is the similarity and X andY represent two images: curves fromthe observed image and detected data,respectively. The mean absolute errorwas calculated according to Eq. 8. Themean absolute error does not exhibitthe offset of the positive and negativevalues, better reflecting the errors. Thesame number of data points from theobservational curves and the detecteddata was calculated.

ava error X ;Yð Þ ¼

Xni¼1

��Xi � Yi��

nð8Þ

where n is the experimental number(505 in this study). The data of the


fitted curve of the observed image andthe detected data must be normalizedbefore calculation. Figure 9 shows thecomparison between the measureddata and the fitted curve of the physicalpicture.

The mean absolute error and simi-larity values between the detected dataand the fitted curve of rubber are shownin Table 1.

From Figure 10 and Table 1, strongagreement is demonstrated betweenthe detected data and the observationaldata. The error corresponding to thefront part of the curve could be additive,resulting in larger cumulative errors forthe end of the curve, thus making itimportant to improve the accuracy atthe front of the curve.

There are three main reasons for theerrors. First, the resistance and bendangle relationship of the flex sensor hasa nonlinearity at small bend angles. Sec-ond, the data extracted from the phys-ical picture results in larger errors inthe error analysis. Third, the measure-ment errors related to small misalign-ments of the sensors are due to sensorinstabilities when the cable moved.

Conclusions andFurther Research

This paper described a novel methodfor underwater cable detection based

l

F

Cfi

FIGURE 9

Five bend states of rubber in the tank.

TABLE 1

The mean absolute error and similarity betweenthe detected data and fitted curve of the rubber.

Position
Mean AbsoluteError (cm) Similarity
Down
1.4092 0.98485
Up
0.98389 0.99162
S1
1.6036 0.94422
S2
0.37699 0.99507
Flat
0.48044 0.97618
IGURE 10

omparison between the detected data and thetted curve of rubber corresponding to Figure 9.

on flex sensors. Through real-timedetection of underwater cable bendstates, underwater accidents that resultfrom underwater cable bending will beforecasted and prevented. Comparedwith common methods, the advan-tages of the proposed technique in-clude the following: (1) The sensor isa separate device as well as the camera,but does not require another platform.(2) Flex sensors can be arranged alongthe whole cable to realize whole cablereal-time detection with low cost.The number of the sensors dependson the user. The more sensors, themore accurate. (3) The novel detectionmethods for cables can be deployedin marine or terrestrial environmentswith poor visual conditions; for exam-ple, towlines, cables of anchor, andmooring system. A water tank experi-ment showed that the proposed systemis feasible as it could accurately detectthe underwater cable bending. Never-theless, further research is still required—particularly for developing a 3-Dmodel for underwater cable bendingand torsion.

AcknowledgmentsThis study was funded by the Pro-

ject (No. 51004085) supported bythe National Science Foundation ofChina and the Program for ZhejiangLeading Team of S&T Innovation(No. 2010R50036). The authors thankthe anonymous reviewers for their de-tailed and valuable comments, whichstrengthened this manuscript.

Corresponding Author:Jiawang ChenOcean College, Zhejiang University,Hangzhou 310058, ChinaEmail: [email protected]

ReferencesBagnitsky, A., Inzartsev, A., Pavin, A.,

Melman, S., & Morozov, M. 2011. Side scan

sonar using for underwater cables & pipelines

tracking by means of AUV. In Underwater

Technology (UT), 2011 IEEE Symposium on

and 2011 Workshop on Scientific Use of

Submarine Cables and Related Technologies

(SSC) (pp. 1-10). Tokyo, Japan: IEEE. http://

dx.doi.org/10.1109/ut.2011.5774119.

Brown, K., Capus, C., Pailhas, Y., Petillot, Y.,

& Lane, D. 2011. The application of bio-

inspired sonar to cable tracking on the sea-

floor. EURASIP J Adv Sig Pr. 2011:484619.

http://dx.doi.org/10.1155/2011/484619.

Capus, C., Pailhas, Y., Brown, K., & Lane, D.

2010. Detection of buried and partially buried

objects using a bio-inspired wideband sonar.

In: OCEANS 2010 IEEE, 24–27May. Sydney:

IEEE. 1-6. http://dx.doi.org/10.1109/oceanssyd.

2010.5603675.

Carter, L., Gavey, R., Talling, P.J., & Liu, J.T.

2014. Insights into submarine geohazards

from breaks in subsea telecommunication

cables. Oceanography. 27(2):58-67. http://

dx.doi.org/10.5670/oceanog.2014.40.

Carter, L., Milliman, J.D., Talling, P.J.,

Gavey, R., & Wynn, R.B. 2012. Near-

synchronous and delayed initiation of long

run-out submarine sediment flows from a

record-breaking river flood, offshore Taiwan.

Geophys Res Lett. 39(12):12603. http://

dx.doi.org/10.1029/2012GL051172.

El-Fakdi, A., & Carreras, M. 2013. Two-step

gradient-based reinforcement learning for

underwater robotics behavior learning. Robot

Auton Syst. 61(3):271-82.

Fahn, C.S., & Sun, H. (2010). Development

of a fingertip glove equipped with magnetic

tracking sensors. Sensors. 10(2):1119-40. http://

dx.doi.org/10.3390/s100201119.

Inzartsev, A.V., & Pavin, A.M. 2008. AUV

cable tracking system based on electromagnetic

and video data. In: OCEANS 2008-MTS/

IEEE Kobe Techno-Ocean (pp. 1-6). Kobe,

Japan: IEEE. http://dx.doi.org/10.1109/

oceanskobe.2008.4531082.

March/A

Isaacs, J.C., & Goroshin, R. 2010. Automated

cable tracking in sonar imagery. In: OCEANS

2010, 20–23 Sept, Seattle, WA: IEEE. 1-7.

http://dx.doi.org/10.1109/oceans.2010.

5664414.

Jordán, M.A., Berger, C., & Bustamante, J.L.

2011. Design of a vision-based sensor of

position and rate for the guidance of autono-

mous underwater vehicles. Sensors. 5:6.

Kaniusas, E., Pfutzner, H., Mehnen, L.,

Kosel, J., Varoneckas, G., Alonderis, A., …

Marquardt, B. 2004. Magnetoelastic skin

curvature sensor for biomedical applications.

In Sensors, 2004. Proceedings of IEEE

(pp. 1484-1487). Vienna, Austria: IEEE.

Kuhn, V.N., Drews, P.L.J., Jr, Gomes, S.C.P.,

Cunha, M.A.B., & da Costa Botelho, S.S.

2013. Automatic control of a ROV for inspec-

tion of underwater structures using a low-cost

sensing. J Braz Soc Mech Sci. 1-14.

Lee, D., Kim, G., Kim, D., Myung, H., &

Choi, H.T. 2012. Vision-based object detec-

tion and tracking for autonomous navigation

of underwater robots. Ocean Eng. 48:59-68.

http://dx.doi.org/10.1016/j.oceaneng.2012.

04.006.

Lorussi, F., Scilingo, E.P., Tesconi, M.,

Tognetti, A.,&Rossi, D.D. 2005. Strain sensing

fabric for hand posture and gesture monitoring.

IEEE T Inf Technol B. 9(3):372-81.

Majidi, C., Kramer, R., & Wood, R.J. 2011.

A non-differential elastomer curvature sensor

for softer-than-skin electronics. Smart Mat St.

20(10):105017. http://dx.doi.org/10.1088/

0964-1726/20/10/105017.

Mandhouj, I., Amiri, H., Maussang, F., &

Solaiman, B. 2012. Sonar image processing

for underwater object detection based on high

resolution system. In: SIDOP 2012: 2nd

Workshop on Signal and Document Process-

ing (Vol. 845, pp. 5-10). CEUR Workshop

Proceedings.

Ortiz, A., & Antich, J. 2014. Bayesian visual

tracking for inspection of undersea power

and telecommunication cables. J Marit Res.

6(2):83-94.


Ortiz, A., Antich, J., & Oliver, G. 2011.

A particle filter-based approach for tracking

undersea narrow telecommunication cables.

Mach Vision Appl. 22(2):283-302. http://

dx.doi.org/10.1007/s00138-009-0199-6.

Pei, Y.H., Yeo, H.G., Kang, X.Y., Pua, S.L.,

& Tan, J. 2010. Magnetic gradiometer on an

AUV for buried object detection. In: OCEANS

2010 (pp. 1-8). Seattle, WA: IEEE. http://

dx.doi.org/10.1109/oceans.2010.5664272.

Sakagami, N., Ishimaru, K., Kawamura, S.,

Shibata, M., Onishi, H., & Murakami, S.

2013. Development of an Underwater Robotic

Inspection System using Mechanical Contact.

J Field Robot. 30(4):624-40. http://dx.doi.org/

10.1002/rob.21463.

Saggio, G. 2014. A novel array of flex sensors

for a goniometric glove. Sensor Actuat A-Phys.

205:119-25. http://dx.doi.org/10.1016/j.sna.

2013.10.030.

Saggio, G., Bisegna, P., Latessa,G.,&Bocchetti,

S. 2009a. Mechanical modeling of bend sen-

sors exploited to measure human joint move-

ments. In World of Wireless, Mobile and

Multimedia Networks & Workshops, 2009.

WoWMoM 2009. IEEE International Sympo-

sium on a (pp. 1-4). Kos: IEEE. http://dx.doi.

org/10.1109/wowmom.2009.5282407.

Saggio, G., Bocchetti, S., Pinto, C.A.,

Orengo, G., & Giannini, F. 2009b. A novel

application method for wearable bend sensors.

In Applied Sciences in Biomedical and Com-

munication Technologies, 2009. ISABEL

2009. 2nd International Symposium on

(pp. 1-3). Bratislava: IEEE. http://dx.doi.org/

10.1109/isabel.2009.5373625.

Szyrowski, T., Sharma, S. K., Sutton, R., &

Kennedy, G. A. 2013a. Developments in

subsea power and telecommunication cables

detection: Part 1-Visual and hydroacoustic

tracking. Underwater Technol. 31:123-32.

http://dx.doi.org/10.3723/ut.31.123.

Szyrowski, T., Sharma, S.K., Sutton, R., &

Kennedy, G.A. 2013b. Developments in

subsea power and telecommunication cables

detection: Part 2—Electromagnetic detection.

Underwater Technol. 31:133-43. http://

dx.doi.org/10.3723/ut.31.133.


Talling, P.J., Paull, C.K., & Piper, D.J. 2013.

How are subaqueous sediment density flows

triggered, what is their internal structure

and how does it evolve? Direct observations

from monitoring of active flows. Earth-Sci

Rev. 125:244-87. http://dx.doi.org/10.1016/

j.earscirev.2013.07.005.

Villar, S.A., Acosta, G.G., Sousa, A.L., &

Rozenfeld, A. 2014. Evaluation of an efficient

approach for target tracking from acoustic

imagery for the perception system of an

autonomous underwater vehicle. Int J Adv

Robot Syst. 11:24.

Wang, P., Goddard, K., Lewin, P., & Swingler,

S. 2011. Electromagnetic field application

to Underground Power Cable Detectiom.

In: 17th International Symposium on high

voltage engineering. 22–26 August 2011,

Hannover, Germany: Leibniz Universitat

Hannover.

Wirth, S., Ortiz, A., Paulus, D., & Oliver, G.

2008. Using particle filters for autonomous

underwater cable tracking. In: IFACWorkshop

on Navigation, Guidance and Control of

Underwater Vehicles. Killaloe, Ireland: Inter-

national Federation of Automatic Control.

161-6.

Zhou, X.,Wang, Y., &Zhou, Y. 2009. Research

on modeling and experimentation on detect-

ing buried depth of optical fiber submarine

cables based on active detection technique.

In: Electronic Measurement & Instruments,

2009. ICEMI’09 (pp. 1-91). Beijing, China:

IEEE. http://dx.doi.org/10.1109/ICEMI.

2009.5274300.

Zhou, X.J., Zhang, Y., Zuo, M.J., & Yuan,

Z.D. 2015. Research on submarine cable

detection technology based on information

fusion. In: 2015 International Industrial

Informatics and Computer Engineering

Conference. Atlantis Press. http://dx.doi.org/

10.2991/iiicec-15.2015.121.

l

P A P E R

Effect of Low Temperature and High Pressureon Deep-Sea Oil-Filled Brushless DC Motors
A U T H O R SMinjian CaiShijun WuCanjun YangState Key Laboratory of FluidPower and Mechatronic Systems,Zhejiang University
A B S T R A C T

NomenclatureT0 Normal tP0 Atmosphμ Dynamicμ0 Dynamicρ DensityP PressureT Temperatα Pressure-vη Kinetic viffiffiffiffiffiTa

pTaylor nu

Mv Viscous dMC Viscous dCM MomentRi Rotor diaC Air-gap clL Rotor lenB Clearanceω Rotor angRe Reynoldsk CoefficienPel Input elecPir Iron pow

The principle of electronic commutationmakes brushless DCmotors suitable fordeep-sea application by sealing themotor in an oil-filled housing. However, if the oil-filled actuator is not designed properly, it will malfunction in the deep sea in spite ofits excellent performance on land. In this paper, oil viscosity variations with pressureand temperature are reviewed because both factors vary with the operating depth,and a practical approach to estimate the viscous torque is advanced based on a vis-cous drag model of the selected motor and the properties of the oil. An experimentalrig that can simulate the deep sea environment was developed, by which a series ofexperiments to study themotor efficiency and dynamic performance in different con-ditions were conducted. The values of viscous torque obtained in the experimentsagreed well with our estimation. The low efficiency in the 2°C experimental groupconfirmed the influence of temperature, while the dramatic difference in the dynamicperformance of the motor when filled with different oils verified the importance ofanalyzing the properties of the oil and of making a deliberate selection.Keywords: deep sea, high pressure, low temperature, oil-filled motor, oil properties
benthos of the deep ocean. However,
Introduction

emperatureeric pressureviscosityviscosity at given temperature T0 and pressure P0

ureiscosity coefficientscositymberrag momentrag moment acting on cylinder flankcoefficientmeterearancegthbetween housing and end face of the motorular velocitynumbert relating to the motor geometrytric powerer loss

With various instruments and un-dersea vehicles unveiling the mysteryof thousands of meters under the sea,we can now gain a better understand-ing of the geology, topography, and

because of the complicated and extremeenvironment, it is still a challenge todesign a reliable and efficient systemfor use in such an environment. Formost deep-sea equipment, electric actu-ators play a vital role in accomplishingspecific tasks such as adjusting a pan-tiltmechanism for camera orientation, dis-placement of a robot arm, and drivinga pump in a fluid system. An efficientactuator not only contributes to pro-longed battery life but also reducesheat dissipation and enhances reliabilityof the whole system.

Motors have been widely usedin subsea applications because oftheir efficiency and versatility. Tanet al. (2010, 2012) have been usingan oil-filled stepper motor to drive amicro-metering pump for underwaterpH sensor calibration. ConventionalDC motors are also widely used in un-dersea mechatronic systems becauseof their outstanding speed adjust-ment feature and large output torque.


Pme Mechanical powerPj Joule power lossR Phase resistanceU Input voltageI Input current

Krishfield et al. (2008) used a DCmotor to drive the traction unit inice-tethered profilers. The motor runsin air within the thick pressure casewith its torque transferred via a mag-netic coupler. Saegusa et al. (2006)employed a DC motor to drive apump to adjust the sampling ratein their multibottle fluid sampler.Seewald et al. (2002) developed a gas-tight sampler for hydrothermal fluidsthat used a DC electric motor to ac-tuate the sampler valve. In these twocases, motors are filled with oil to com-pensate for the ambient pressure so asto avoid the requirement of beinghoused in a thick-walled metallic pres-sure vessel, thus protecting them fromhigh-pressure seawater and preventingfatal damage to the motor if even asmall amount of seawater seeps in.Nevertheless, in an insulated oil envi-ronment, the oil film in the brusheswill affect the commutation of themotor to some extent.

The brushless DC (BLDC) motor,more and more widely used due to itshigh reliability, efficiency, and controlperformance, has been used success-fully in the design of thrusters andmanipulators of underwater vehicles(Bhangu et al., 2005; Lewandowskiet al., 2008). The electronic commuta-tion of a BLDC motor eliminates thepotential for poor electrical contacts,which makes it suitable for a varietyof oceanic apparatus such as position-ing systems of undersea laser spectrom-eters and underwater robotics (Kunzet al., 2008; Peng et al., 2014; Whiteet al., 2005). However, oil-filled


motor-based drive systems cannotarbitrarily be designed because the oilproperties change dramatically withthe pressures and temperatures of thedeep ocean. The viscous drag intro-duced by the cold and compressed oilaffects the efficiency and load matchingof themotor. Up to now, little is knownabout exactly how the deep sea envi-ronment affects oil-filled motors.

Viscous drag produced by the rotormay not be so obvious in ordinary en-vironments, but it becomes significantwhen an oil-filled motor works in thecold deep sea if an unsuitable kindof insulating oil is used. Therefore,study of the viscous drag force in oil-filled motors is valuable for thoseworking in the deep-sea engineeringfield. Some previous research has beencarried out to minimize the impact ofviscous drag on the motor rotor.Ahmed and Toliyat (2007) presenteda design procedure for submersiblemotors and corroborated their designconsideration of coupled-field analysisneeds based on their simulation re-sults. Zou et al. (2012) designed anoil-filled BLDC motor that takes intoaccount the effect of high pressure.They analyzed the composition of thetotal loss and drew a comparison be-tween the results for normal pressureand high pressure, which highlightedthe need to consider the pressure ef-fect. These studies provide useful in-formation for oil-filled motor design;however, they have emphasized theneed for comprehensive considerationof the relevant parameters but ne-glected some specific details such

l

as the properties of the oil and theimportant environmental factor, i.e.,temperature.

In this paper, we analyzed the vis-cous drag torque of a BLDC motorin different compensating oils andstudied the actual effects on it of tem-perature and pressure. A simplifiedanalytical model to calculate the po-tential viscous moment is describedin the Analysis and Modeling section,and the effects of both temperatureand pressure on the insulating oil arediscussed and taken into consider-ation. In the Experimental Setup andMethod section, a simulation envi-ronment that was developed to providedifferent temperatures and pressuresis described. A series of experimentsunder different conditions were carriedout to study viscous drag loss and tran-sient response of the motor. The re-sults and discussion are presented inthe Results and Discussion section.

Analysis and ModelingBecause the viscosity behavior of

the compensating liquid under lowtemperatures and high pressures isone of the key motivations for thispaper, before delving into this lineof research, we would like to reviewsome proposed theories on the viscos-ity properties of commonly used in-dustrial oils. Based on this knowledgeof the oil properties, a quantitative an-alytical model was created to estimatethe initial viscous drag torque in actualdeep sea environments.

Oil ViscosityThe concept that oils get thicker

when the temperature drops or thepressure rises is well known, whichmeans that, in the deep sea environ-ment, an oil’s viscosity will increase

markedly due to the degradation ofboth of these factors.

The general form for the pressureand temperature dependence of viscos-ity has been known for over half a cen-tury (Bair et al., 2001), and researchershave long focused on revealing the ac-curate relationship of viscosity, tem-perature, and pressure of assorted oils(Bair et al., 2001; Gold et al., 2001).A general method of predicting theoil viscosity is the Barus equation(Gold et al., 2001).

μ ¼ μ0 exp αPð Þ: ð1Þ

where the temperature dependencecan be described by

μ0 ¼ Q expG

T þ P

� �: ð2Þ

and the pressure-viscosity coefficient αis additionally taken into account by

α P;Tð Þ ¼ 1a1 þ a2T þ b1 þ b2Tð ÞP

ð3Þ

where Q, G, P, a1, a2, b1, and b2 needto be determined separately by experi-mental data for each oil, and then theBarus equation can represent the fluidbehavior in a wide pressure and tem-perature range. Some other simplifiedmethods to calculate the viscosity of aspecific fluid are usually adopted in en-gineering; however, for accuracy, wewill use the above equation to obtainthe oil viscosity at a specific condition.

A typical hydraulic oil viscosityvariation versus temperature and pres-sure is illustrated in Figure 1. It showsthat, with the pressure increase, theviscosity increases gradually, and withthe drop in temperature, especiallybelow 20°C, the viscosity could in-crease dramatically.

Usually, considering the generaldeep sea environment (T ≥ 0°C andP < 110 MPa), the viscosity of mostoil is more sensitive to temperaturethan pressure. This implies that weshould pay more attention to the tem-perature factor when dealing with anoil-filled actuator used in a cold envi-ronment. As for the oil-filled motorused at such conditions, it is necessaryto take both temperature and pressureinto consideration when evaluating themotor’s actual output torque.

In the present work, two typicaloils, namely hydraulic oil (a mineraloil) and silicone oil (a synthetic oil),were used in our experiments. Hydrau-lic oil usually has an antiwear propertyand good incompressibility, while sili-cone oil provides compatibility withother organic materials and hydropho-bic property. The ISO viscosity grade(VG) of hydraulic oil is 22, whichdenotes that its dynamic viscosity is22 mm2/s at 40°C and atmosphericpressure. Because it is hard to findtwo different kinds of oil with equalviscosity, the silicone oil we used is50 cSt, which should have a viscosity

March/A

of 50 mm2/s at 25°C and atmosphericpressure. As a result, viscosities of thetwo oils were measured to be about41 mPa∙s and 47 mPa∙s, respectively,at 25°C and atmospheric pressure.

The pressure and temperature de-pendence of the dynamic viscosity μis derived from equations (1) and (3) as

μ ¼ μ0 expP

AT þ BTP

� �ð4Þ

where AT and BT are the temperature-dependent parameters shown in equa-tion (3). Because our work focusedon two typical temperatures, namely25°C and 2°C, we obtained the cor-responding values of AT and BT byexperiments.

Some relevant properties of the oilsconcerned in our research are listedin Table 1.

Modeling of ViscousDrag Moment

Along with the experiments carriedout in this work, a quantitative ap-proach to determine the effective outputtorque was studied. The motor used in

FIGURE 1

Viscosity variation with temperature and pressure.


our research is shown in Figure 2a, andits main specifications and parametersare listed in Table 2. Because it is dif-ficult to establish a precise model thattakes into account detailed structuressuch as bearings and the hall sensorunit, a simplified model, i.e., a pairof coaxial cylinders filled with oil inthe gap, was developed to study theviscous load of the oil-filled BLDCmotor (Figure 2b).

As shown in Figure 2b, the viscousdrag of the rotor consists of two parts:one produced by the flank of the rotorand the other produced by the endsurface. Then, the fluid drag forceacting on the rotor surfaces can be


calculated by identifying the flowstate between the stator and rotor inadvance, which can be determined bythe aid of dimensionless Taylor num-ber

ffiffiffiffiffiTa

pintroduced as

ffiffiffiffiffiTa

p¼ RiωC

η

ffiffiffiffiCRi

rð5Þ

where η is the kinetic viscosity of thefluid.

The flow pattern in the annulus re-gion will make a transition from lami-nar flow to transition flow or turbulentflow when

ffiffiffiffiffiTa

pis larger than 41.3,

which leads to inapplicability of themodeling derivation if it is based on

l

the assumption of laminar flow (Deng,2007; Qi et al., 2010). Bilgen andBoulos (1973) concluded a general re-lationship for the moment coefficientthat is defined in equation (6) as a func-tion of the geometry parameters andReynolds number when the flow pat-tern varies.

CM ¼ MC= 0:5πρω2R 4i L

� � ð6Þ

Re ¼ ρωRiC=μ ð7Þ

CM ¼ K C=Rið ÞnR βe ð8Þ

whereMC denotes the viscous drag mo-ment exerted by the cylinder and con-stant K and exponents n and β can bedetermined by the flow regime.

In fact, for small motors, the Taylornumber is usually below the criticalvalue due to the small physical dimen-sions. In our case, the Taylor numberwas checked below the critical numberfor both oils (about 0.1 in silicone oil,

TABLE 1

Main characteristics.

Oil Type

μ0 at 25°C
μ0 at 2°C α at 25°C α at 2°C
mPa s
mPa s Pa−1 Pa−1
Hydraulic oil
41.6 140.3 12:01eþ 07ð Þ þ 5:1e� 1ð ÞP
14:49eþ 07ð Þ þ 1:6e� 02ð ÞP

Silicone oil
46.9 71.3 11:15eþ 08ð Þ � 1:2e� 01ð ÞP
19:43eþ 07ð Þ þ 2:8e� 01ð ÞP

TABLE 2

Main specifications.

Specification
Value
Nominal power
12 W
Rated current
661 mA
Rated speed
7740 rpm
Rated torque
11.9 mN m
Phase resistance
12.4 ohm
FIGURE 2

The BLDC motor (a) and its simplified model (b).

giving a speed of 6,000 rpm). Hence,for Newtonian fluids whose viscousstresses are linearly proportional tostrain rate, we can obtain the viscousload function as equation (9) with abasic knowledge of fluid mechanicsby separating the rotor surface intothe disk and flank (White, 2011).

Mv ¼ πR4i μωB

þ 4πR3i LμωC

¼ kμω ð9Þ

whereMv denotes the viscous dragmo-ment on the rotor and k is a constantrelated to the motor geometry. Thisfunction shows a linear relationshipbetween viscous torque and the dy-namic viscosity μ by regarding therotor speedω as constant. Substitutingthe corresponding viscosity depen-dence into equation (9), we can thenobtain viscous moment correlationfor hydraulic oil and silicone oil as

Mv ¼ kωμ0 expP

AT þ BTP

� �ð10Þ

where the relevant parameters are listedin Table 1.

Consequently, once we obtain thegeometry parameter k by measuringviscous moment in a known condi-tion, we can use equation (10) to cal-culate the viscous torque in variousenvironments given the actual workingtemperature and pressure values. Inour study, the results for a velocity of2,000 rpm are listed in Table 3.

We can see that the potential vis-cous load torque increases sharplywhen the temperature is low or thepressure is high if we fill the motorwith hydraulic oil, and the low temper-ature seems to have the worst impacton the start-up process. The siliconeoil, however, seems to have much lessimpact on the motor output.

Experimental Setupand MethodExperimental Setup

An experimental rig was designedto simulate the deep sea environment.The whole test rig consisted of threeparts: a high-pressure reservoir, a tem-perature regulationmodule, and a con-trol and data acquisition unit (Figure 3a).The main structural features of the testrig are presented in Figure 3b.

Pressure of up to 60 MPa in thetest apparatus was supplied by a high-pressure test pump. The temperatureregulation system mainly compriseda commercial recirculating cooler, acooling jacket, and a special thermo-couple that penetrated into the pressurereservoir. The cooler pumps recirculatewater into the cooling jacket and re-ceive it at the outlet, making the fast-flowing water transfer heat betweenthe pressure reservoir and coolingwater. By comparing the set tempera-ture and the monitored temperaturefrom thermocouple, the recirculatingcooler finally regulated the inner watertemperature in a closed-loop mannerto an accuracy of ±0.2°C.

The BLDCmotor was encapsulatedin an oil-filled vessel that was located in

March/A

the high-pressure reservoir. A rubbersleeve in the oil-filled vessel and twogroups of holes distributed aroundthe metallic housing (as shown in Fig-ure 3b) ensured the pressure balance.The oil-filled vessel was filled carefully,and the motor was kept running longenough to remove all air bubbles in themotor. Two surface-mounted thermalresistors attached to the motor housingsurface were used to measure the oiltemperature, by which we could guar-antee that the motor ran in the ex-pected temperature condition.

The motor and its driver weresupplied by a 24-V regulated powersupply. The motor speed was set by avirtual control panel on a PC, whichalso recorded the electric current andtemperature values synchronously.Both of the current and temperaturevalues acquired by the driver boardwere calibrated, apart from a few un-avoidable measurement errors.

MethodTo have an intuitive understanding

of the viscous torque caused by the oilin a deep sea environment and to verifythe estimated approach introducedabove, the motor was set to operate

TABLE 3

Calculated results of initial viscous moment at 2,000 rpm.

Condition
In Hydraulic Oil (mN m)
pril 2016 Vol

In Silicone Oil (mN m)

25°C and 0 MPa (tested)
17.32 20.21
25°C and 20 MPa
32.55 24.13
25°C and 40 MPa
43.63 29.04
25°C and 60 MPa
51.67 35.23
2°C and 0 MPa
58.29 26.59
2°C and 20 MPa
91.02 32.48
2°C and 40 MPa
142.59 38.85
2°C and 60 MPa
221.05 45.63
The torque calculated here is the viscous drag torque at the starting stage that neglects temperature rising.

ume 50 Number 2 87

in different conditions. The power ofthe motor without oil and load wasfirst measured at room temperature(25°C) and atmospheric pressure.The vessel was then successively filledwith two kinds of oils, and a constanttemperature environment was provided.For each kind of oil, the experimentswere divided into two groups: one at25°C and the other at 2°C. Both exper-iments were undertaken in the samepressure series to study the influenceof different depths.


According to the energy conser-vation principle, the input electricalpower is converted into mechanicalpower Pme with associated joule powerloss Pj and iron loss Pir.

Pel ¼ Pme þ Pir þ Pj ð11Þ

Because the motor runs with no exter-nal load, the mechanical power Pme isreplaced by Pv, which can be rewrittenin detail as

UI ¼ ωMv þ Pir þ I 2R: ð12Þ

l

The joule power loss dependson the winding R and input current,while mechanical power Pme is relatedto the angular velocity and torque ex-erted on the rotor. Previous literaturehas indicated that the iron loss becomeslarger with deterioration of the mag-netic property caused by mechanicalstress (Fujisaki et al., 2007; Fujisaki& Satoh, 2004; Takahashi & Miyagi,2011). However, from a differentperspective, the iron loss increasedby the compressive stress is relativelysmall (about 1 W/kg with a stressof 100 MPa and a flux density of0.5 T) compared with the other terms(Takahashi & Miyagi, 2011). So weregarded the iron loss term as a con-stant (0.3 W) during experimental dataprocessing. By subtracting Pir and Pj,we could determine Mv, which indi-cates the viscous term.

As discussed in the Analysis andModeling section, the viscous dragtorque depends on the inner geometryof the motor, velocity, and oil viscos-ity. Because the viscosity is closely re-lated to the temperature and the heatproduced by the motor varies withthe operating condition, the calcu-lation of the real-time viscous dragtorque is then complicated. To sim-plify the analysis, all viscous torqueresults from our calculation focusedonly on the starting stages, while theexperimental results provided bothinitial and steady state data. The fol-lowing assumptions were made inthis study.1. In the starting stage, the oil tem-

perature stayed constant, and theoil temperature in the motor wasregarded as uniform.

2. The influence of pressure andtemperature on the motor ’ smagnetic material was neglectedcompared with that of the com-pensating oil.

FIGURE 3

Experimental apparatus. (a) Low-temperature and high-pressure environment. (b) Inner schematicof the high-pressure reservoir.

Results and DiscussionThe viscous drag moment de-

creases as the motor temperature risesafter start-up, and then it reachesa steady value when the motor is inthermal equilibrium. What we areconcerned about is the initial viscoustorque that influences motor’s start-up performance and the steady viscouspower loss that matters in the continu-ous operation. The following results ofthe experiments revealed the influenceof both temperature and pressureon the oil-filled BLDC motor andhighlighted the importance of analyz-ing the properties of compensating oil.

Start-Up PerformanceThe influence of pressure on the

motor filled with two kinds of oil isshown in Figure 4. As predicted bythe calculation, the viscous torque atthe starting stage increased graduallyin both cases as the pressure climbed.It is noteworthy that, when themotor was compensated in hydraulicoil, part of the difference betweentested result and calculated value at

60 MPa is due to the power limita-tion of the motor. At this condition,the motor could not provide largeenough torque to overcome the in-creased viscous drag moment if itwere set to run at a faster speed, andit would have a problem at start-up ifit were connected to an extra load.Benefiting from low sensitivity to pres-sure, the power margin of the motor insilicone oil was larger compared withthat in hydraulic oil. Nevertheless,the problem still exists when the pres-sure continues to increase if we donot analyze the potential viscous loadbeforehand.

The viscous drag influenced by lowtemperature was so significant that themotor would actually not be able torun to the set speed immediately asin normal environment. As shown inFigure 5, the current data recorded inthe first 2 min after giving the motora step velocity signal confirm the exis-tence of deterioration in the transientprocess.

From the current responses in thefirst 20 s, as shown in Figures 5a and

March/A

5b, we can see how the pressure andtemperature (especially the tempera-ture) deteriorate the motor’s start-upperformance. In the case of hydraulicoil at 2°C, as a result of the influenceof low temperature on viscosity,the motor had almost reached thelocked rotor current at the pressureof 20 MPa. The time of overcurrentlasted longer to allow the motor toovercome the heavy viscous torqueafter pressurizing to 60 MPa. Thecurrent stayed above its rated valuefor almost 1 min and then droppedbelow this value as the oil temperaturerose. For all three low-temperature ex-periments, the motor consumed muchmore current than when it was run-ning at room temperature, until theinner oil temperature reached a stablelevel. As can be foreseen, such over-current will easily cause damage tothe electronic components and themotor windings, particularly in theintermittent operation mode.

The silicone oil fared better becauseof its low sensitivity to temperature. Itcould successfully start up even at 2°Cand high pressure of up to 60 MPa.Moreover, the current curve showsthat the initial viscous moment wasmuch less that it could reduce thepossibility of malfunction duringstart-up. However, the differences ofthe steady current values in the twocases are not that apparent.

Figure 5c shows the velocity andtemperature details of the motor filledwith hydraulic oil after it was started.It can be seen that the motor failed toaccelerate immediately to its set speeddue to the heavy viscous load andalmost stalled. Benefitting from theheat produced by itself, the motorthen slowly speeded up to overcomethe velocity-dependent viscous dragmoment. This resulted in a rapid cur-rent decrease when the motor reached

FIGURE 4

Comparison between experimental and calculated results of viscous torque when compensatedwith hydraulic oil and silicone oil at room temperature.



the set speed because the viscous mo-ment did not increase anymore.

Viscous Power LossThe results of the steady viscous

power loss Pme are presented in Fig-ure 6. According to equation (10),the steady viscous loss should increaseexponentially with the pressure, butthese data showed a linear trend. Thiscan mainly be ascribed to the differentfinal oil temperature, which offsetspart of the loss. The significant dif-ferences between the results at 25°Cand 2°C show that ambient tempera-ture also had a remarkable impact onmotor efficiency. Specifically, whenthe motor was filled with hydraulicoil, the loss in the low-temperaturecondition would increase by 70%and 57% at 40 and 60 MPa, respec-tively, compared with those measuredat room temperature. Moreover, it hasalmost reached its rated power at 2°Cand 60MPa in this case. This indicatesthat the motor’s speed will decrease orsimply fail when driving a load. On theother hand, the steady power loss forthe silicone oil case is only slightlylower. With the motor running, therising temperature did not reducedistinct viscous torque as it did in hy-draulic oil. The steady viscous loss wasabout 7.6 W at the worst condition.Consequently, the low temperatureand high pressure directly led to a scar-city of the available output torque anda low efficiency.

Lower Viscosity Silicone OilTo reduce power dissipation and

prevent excessive overcurrent, one sim-ple way is to replace the thick oil with alow-viscosity oil that has good fluidity,even in the deep sea environment. Be-cause silicone oil has a wide rangeof viscosity options and its viscosity-temperature coefficient is low, we

FIGURE 5

Current variation comparison between 25°C and 2°C when giving a step speed signal of 2,000 rpm.(a) Hydraulic oil. (b) Silicone oil. (c) Speed and current versus time when the motor was running inhydraulic oil at the experimental condition of 2°C and 60 MPa.

used silicone oil of 5 cSt (5 mm2/sat 25°C and atmospheric pressure) asa demonstrative case in our study.Table 4 shows the steady viscouspower loss of the motor in the threecases.

In addition, by benefiting from thelow viscosity at different conditions,the viscous moments in Figure 7 indi-cate a significant improvement in effi-ciency as well as a reduction in the

possibility of stalling. Thus, by ana-lyzing the viscous drag model and theproperties of different oils, we candetermine how the oil-filled actuatorwill behave and eventually implementan efficient and reliable design.

ConclusionIn this paper, we investigated the

influence of both the temperature

March/A

and pressure on an oil-filled BLDCmotor by the viscous drag modeland an experimental rig that cansimulate the deep sea environment.Exper iments va l idated that thepressure and temperature had amarked effect on the power loss anddynamic performance of the oil-filled motor.

Two typical types of oils (hydrau-lic oil and silicone oil) were usedin our study. From the two groupsof experiments, the steady powerloss increased as the pressure climbedor temperature dropped. In addi-tion, the temperature had a dramaticeffect on the start-up performanceif the motor is filled with hydraulicoil. When the ambient temperaturewas low (2°C), the viscosity of thehydraulic oil (ISO VG 22) increasedsharply, which overloaded the motor,and the resulting overcurrent lasteduntil the temperature rose. The sili-cone oil performed much betterbecause of its low temperature sensi-tivity. The results highlight the im-portance of analyzing the propertiesof compensating oil and indicatethat the temperature and pressureshould both be taken into consider-ation when using a compensating oilin the drive system designed for thedeep sea.

In addition to the experimentalwork that was conducted, a semi-experimental method to calculate thepotential starting torque was advanced,the results of which agreed well withthe experimental results. By deter-mining the viscous loss in an ordinaryenvironment, we can then obtain thepotential viscous torque using equa-tion (10) at specified conditions. Al-though our test object in this studywas a BLDC motor, the method canbe applied to other rotating electricalmachines.

FIGURE 6

Viscous power loss at different conditions when filled with (a) hydraulic oil and (b) silicone oil at aspeed of 2,000 rpm.


AcknowledgmentsThis work was supported by the

National Natural Science Founda-tion of China and the Science Fundfor Creative Research Groups of theNational Natural Science Founda-tion of China (Nos. 41106081 and51521064). The authors would liketo thank H. Sun for his help in thedesign of the control panel and K. Liand G. Ma for their help during theexperiments.


Corresponding Author:Shijun WuState Key Laboratory of Fluid Powerand Mechatronic Systems, ZhejiangUniversity, Hangzhou 310027, ChinaEmail: [email protected]

ReferencesAhmed, S., & Toliyat, H. 2007. Coupled

field analysis needs in the design of sub-

mersible electric motors. In: Electric Ship

l

Technologies Symposium, 2007. ESTS’07.

IEEE. pp. 231-7. Arlington, VA: IEEE.

http://dx.doi.org/10.1109/ests.2007.

372091.

Bair, S., Jarzynski, J., & Winer, W.O. 2001.

The temperature, pressure and time depen-

dence of lubricant viscosity. Tribol Int. 34(7):

461-8. http://dx.doi.org/10.1016/S0301-

679X(01)00042-1.

Bhangu, B.S., Snary, P., Bingham, C.M., &

Stone, D.A. 2005. Sensorless control of deep-

sea ROVs PMSMs excited by matrix con-

verters. In: Power Electronics and Applications,

2005 European Conference on. pp. 1-8.

Dresden, Germany: IEEE. http://dx.doi.org/

10.1109/epe.2005.219602.

Bilgen, E., & Boulos, R. 1973. Functional

dependence of torque coefficient of coaxial

cylinders on gap width and Reynolds numbers.

J Fluid Eng-T ASME. 95(1):122-6. http://

dx.doi.org/10.1115/1.3446944.

Deng, D. 2007. A numerical and experi-

mental investigation of Taylor flow instabil-

ities in narrow gaps and their relationship

to turbulent flow in bearings. Ph.D. thesis,

Akron University, 305 pp.

Fujisaki, K., Hirayama, R., Kawachi, T.,

Satou, S., Kaidou, C., Yabumoto, M., &

Kubota, T. 2007. Motor core iron loss anal-

ysis evaluating shrink fitting and stamping by

finite-element method. IEEE T Magn. 43(5):

1950-4. http://dx.doi.org/10.1109/TMAG.

2006.877041.

TABLE 4

Steady power loss comparison between three compensating oils at 2,000 rpm.

Experimental Condition
Viscous Loss for Hydraulic Oil(ISO VG 22) (W)
Viscous Loss for Silicone Oil(50 cSt) (W)

Viscous Loss for Silicone Oil(5 cSt) (W)

25°C and 0 MPa
2.47 3.62 0.43
25°C and 20 MPa
3.35 4.47 0.68
25°C and 40 MPa
4.43 5.03 0.83
25°C and 60 MPa
5.54 5.40 1.04
2°C and 20 MPa
5.77 6.30 0.95
2°C and 40 MPa
6.97 6.84 1.26
2°C and 60 MPa
7.96 7.50 1.61
FIGURE 7

Results of viscous torque tested at different conditions when compensatedwith silicone oil of 5 cSt(velocity = 2,000 rpm).

Fujisaki, K., & Satoh, S. 2004. Numerical

calculations of electromagnetic fields in silicon

steel under mechanical stress. IEEE T Magn.

40(4):1820-5. http://dx.doi.org/10.1109/

TMAG.2004.830509.

Gold, P.W., Schmidt, A., Dicke, H., Loos, J.,

& Assmann, C. 2001. Viscosity-pressure-

temperature behaviour of mineral and syn-

thetic oils. J Syn Lubr. 18(1):51-79. http://

dx.doi.org/10.1002/jsl.3000180105.

Kunz, C., Murphy, C., Camilli, R., Singh, H.,

Bailey, J., Eustice, R.,… Sohn, R. 2008. Deep

sea underwater robotic exploration in the ice-

covered arctic ocean with AUVs. In: Intelligent

Robots and Systems, 2008. IROS 2008. IEEE/

RSJ International Conference on. pp. 3654-60.

Nice, MN: IEEE. http://dx.doi.org/10.1109/

iros.2008.4651097.

Krishfield, R., Toole, J., Proshutinsky, A., &

Timmermans, M.L. 2008. Automated ice-

tethered profilers for seawater observations

under pack ice in all seasons. J Atmos Ocean

Tech. 25(11):2091-105. http://dx.doi.org/

10.1175/2008JTECHO587.1.

Lewandowski, C., Akin, D., Dillow, B.,

Limparis, N., Carignan, C., Singh, H., &

Sohn, R. 2008. Development of a deep-sea

robotic manipulator for autonomous sampling

and retrieval. In: 2008 IEEE/OES Autono-

mous Underwater Vehicles. pp. 14-9. Woods

Hole, MA: IEEE. http://dx.doi.org/10.1109/

auv.2008.5290533.

Peng, S., Yang, C., Fan, S., Zhang, S., Wang,

P., & Chen, Y. 2014. Hybrid underwater

glider for underwater docking: Modeling and

performance evaluation. Mar Technol Soc J.

48(6):112-24. http://dx.doi.org/10.4031/

MTSJ.48.6.5.

Qi, W., Zhou, J., & Li, J. 2010. Numerical

calculation of viscous drag loss of oil-filled

BLDC motor for underwater applications.

In: 2010 International Conference on In

Electrical Machines and Systems (ICEMS).

pp. 1739-42. Incheon, Korea: IEEE.

Saegusa, S., Tsunogai, U., Nakagawa, F., &

Kaneko, S. 2006. Development of a multi-

bottle gas‐tight fluid sampler WHATS II

for Japanese submersibles/ROVs. Geofluids.

6(3):234-40. http://dx.doi.org/10.1111/

j.1468-8123.2006.00143.x.

Seewald, J.S., Doherty, K.W., Hammar,

T.R., & Liberatore, S.P. 2002. A new gas-

tight isobaric sampler for hydrothermal fluids.

Deep-Sea Res Pt I. 49(1):189-96. http://

dx.doi.org/10.1016/S0967-0637(01)00046-2.

Takahashi, N., & Miyagi, D. 2011. Effect of

stress on iron loss of motor core. In: Electric

Machines & Drives Conference (IEMDC),

2011 IEEE International. pp. 469-74. Niagara

Falls, ON: IEEE. http://dx.doi.org/10.1109/

IEMDC.2011.5994642.

Tan, C., Ding, K., Jin, B., Seyfried, W.E., Jr,

& Chen, Y. 2012. Development of an in situ

pH calibrator in deep sea environments.

IEEE-ASME T Mech. 17(1):8-15. http://

dx.doi.org/10.1109/TMECH.2011.2168419.

Tan, C.Y., Jin, B., Ding, K., Seyfried, W.E.,

Jr, & Chen, Y. 2010. A long-term in situ

calibration system for chemistry analysis of

seawater. J Zhejiang Univ-Sc A. 11(9):01-708.

http://dx.doi.org/10.1631/jzus.A0900643.

White, F.M. 2011. Fluid Mechanics.

New York: McGraw-Hill, 896 pp.

White, S.N., Kirkwood, W., Sherman, A.,

Brown, M., Henthorn, R., Salamy, K., …

Brewer, P.G. 2005. Development and

deployment of a precision underwater posi-

tioning system for in situ laser Raman spectros-

copy in the deep ocean. Deep-Sea Res Pt I.

52(12):2376-89. http://dx.doi.org/10.1016/

j.dsr.2005.09.002.

Zou, J., Qi, W., Xu, Y., Xu, F., Li, Y., &

Li, J. 2012. Design of deep sea oil-filled

brushless DC motors considering the high

pressure effect. IEEE T Magn. 48(11):4220-3.

http://dx.doi.org/10.1109/TMAG.2012.

2204731.


May/June 2016Impacts of Data Collected and Lessons Learned from Ocean Observing Systems WorldwideGuest editors: Donna M. Kocak, Vembu Subramanian, Ian Walsh, Mairi Best, & Joshua Henson

July/August 2016Research Initiatives in Europe: Cooperation for Blue GrowthGuest editors: Andrea Caiti, Giuseppe Casalino, & Andrea Trucco

September/October 2016Bio-inspiration for Marine TechnologiesGuest editor: Jason D. Geder

November/December 2016Military Activities: Preserving the Legacy, Evaluating Threats and Mitigating ImpactsGuest editors: Geoff Carton & Lisa SymonsManuscripts due June 20, 2016

The MTS Journal includes technical papers, notes and

commentaries of general interest in

most issues.

Contact Managing Editor

Amy Morgante

([email protected])

if you have material you

would like considered.

Specifications for

submitting a manuscript

are on the MTS website

under the Publications menu.

www.mtsociety.org

E-mail:[email protected]

Or visit our homepage at

www.mtsociety.org

CALL FOR PAPERSUPCOMING MTS JOURNAL ISSUES

Statement of Ownership, Management, and Circulation(Required by 39 U.S.C. 3685)1. Publication Title: Marine Technology Society Journal2. Publication Number: 0025-33243. Filing Date: 3/18/164. Issue Frequency: Bi-monthly5. Number of Issues Published Annually: 66. Annual Subscription Price: $155 Domestic; $250 Foreign7- 8. Complete Mailing Address of Known Office of Publicationand Headquarters or General Business Office of Publisher (Notprinter): Marine Technology Society, 1100 H Street NW, LL-100, Washington, DC 200059. Publisher: Marine Technology Society, 1100 H Street NW, LL-100, Washington, DC 20005; Editor: Anni Vuorenkoski Dalgleish, MTS, 1100 H Street NW, LL-100, Washington, DC 20005; Managing Editor: Amy Morgante, 1100 H Street NW, LL-100, Washington, DC 2000510. Owner: Marine Technology Society, 1100 H Street NW, LL-100, Washington, DC 2000511. Known Bondholders, Mortgagees, and Other Security HoldersOwning or Holding 1 Percent or More of Total Amount of Bonds,Mortgages, or Other Securities: None12. Tax Status: Has Not Changed During Preceding 12 Months13. Publication Title: Marine Technology Society Journal14. Issue Date for Circulation Data Below: SEPT/OCT 2015

15. Extent and Nature of Circulation Average No. No. Copies of Copies Each Single Issue Issues During Published Preceding 12 Nearest to Months Filing Datea. Total Number of Copies (Net Press Run) 500 480b. Paid Circulation (By Mail and Outside the Mail) 1. Mailed Outside-Country Paid Subscriptions Stated on PS Form 3541 (include paid distribution above normal rate, advertiser’s proof copies, and exchange copies) 450 420 2. Mailed In-County Paid Subsriptions Stated On form 3541 (include advertiser’s proof and exchange copies) 0 0 3. Sales Through Dealers and Carriers, Street Vendors, Counter Sales, and Other Non-USPS Distribution 0 0 4. Paid Distribution by Other Classes of Mail Through the USPS 50 10c. Total Paid Distribution (sum of 15b, (1), (2), (3) and (4) 500 430

d. Free or Nominal Rate Distribution (By Mail and Outside the Mail) 1. Outside-country as Stated on Form 3541 0 0 2. In-County as Stated on Form 3541 0 0 3. Other Classes Mailed Through the USPS 0 0 4. Free Distribution Outside the Mail 25 0e. Total Free or Nominal Rate Distribution (Sum of 15d) 25 15f. Total Distribution (Sum of 15c and 15e) 525 445g. Copies not Distributed 25 35h. Total (sum of 15f and g) 550 480i. Percent Paid (15c divided by 15f times 100) 95.2% 96.6%16. Electronic Copy Circulationa. Paid Electronic Copies 65 54b. Total Paid Print Copies (Line 15c) + Paid Electronic Copies (Line 16a) 565 484c. Total Print Distribution (Line 15f) + Paid Electronic Copies (Line 16a) 590 499d. Percent Paid (Both Print & Electronic Copies) (16b divided by 16c times 100) 95.7% 97.0%17. Publication Statement of Ownership will be printed in the Vol. 50, No. 2 issue of this publication, MAR/APRIL 2016.I certify that 50% of all my distributed copies (electronic and print) are paid above a nominal price. I certify that the statements above made by me are correct and complete. Amy Morgante

STATEMENT OF OWNERSHIP

Check the Society website for future Journal topics.

www.mtsociety.org

MEMBERSHIP CATEGORIESMEMBER: $75 Any person with active professional inter-est in the marine field or a closely related field. Includesthe online version of the Journal.

ASSOCIATE: $75 Any person with an interest in themarine field and supporting MTS objectives. Includes theonline version of the Journal.

STUDENT: $25 Must be a high school Senior, full-timeundergraduate, or graduate student. Includes the onlineversion of the Jouranl.

PATRON: Any person supporting MTS objectives by contribu-ting $100 or more annually. (Contrubution above $75 duesis tax deductible.) Includes the online version of the Journal.

EMERITUS: $40 Any person who is retired from activeprofessional interest in the marine field. Includes theonline version of the Journal.

LIFE: $1000 One-time payment. Includes both the onlineversion of the Journal and the paper copy of the Journal.

CORPORATE Fortune 500: $2000 Any corporationendorsing the objectives, policies and activities of MTS.May appoint 16 representatives to the Society, who havethe same rights and privileges as Members. Includes theonline version of the Journal for each representative. Themain contact receives a subscription to the paper copy ofthe Journal.

CORPORATE Non-Fortune 500: $1000 Any corporationendorsing the objectives, policies and activities of MTS.May appoint 11 representatives to the Society, whohave the same rights and privileges as Members.Includes the online version of the Journal for each rep-resentative. The main contact receives a subscription tothe paper copy of the Journal.

BUSINESS: $550 Any business firm whose gross annualincome is less than $4 million endorsing the objectives,policies and activities of MTS. May appoint six represen-tatives to the Society, who have the same rights andprivileges as Members. Includes the online version ofthe Journal for each representative. The main contactreceives a subscription to the paper copy of theJournal.

INSTITUTION: $550 Any library, government unit orother qualified nonprofit organization endorsing theobjectives, policies and activities of MTS. May appointsix representatives to the Society, who have the samerights and privileges as Members. Includes the onlineversion of the Journal for each representative. The maincontact receives a subscription to the paper copy ofthe Journal.

MTS JOURNAL PRINT VERSION SUBSCRIPTION (Prices include shipping)Domestic Subscription

Member Rate: .....................................................Non-Member Rate: ..............................................$155

Canadian/International Subscription Member Rate: $55 Canadian/$125 InternationalNon-Member Rate: $175 Canadian/$250 International

Yes, I want to become a member of the Marine Technology Society.I wish to purchase a subscription to the print version of the MTS Journal. (MTS Journal online is a freemember benefit.)

I’m not ready to join at the present time, but I would like to receive news and information from MTS.Please keep my name on your mailing list. (Please fill in member information section and return to MTS.)

MEMBER INFORMATION (Please fill in the address where MTS publications and correspondence should be sent.)

Name: _______________________________________________Company (if applicable): _____________________________

Title:________________________________________________Email:____________________________________________

Phone:______________________________________________ FAX:_____________________________________________

Address:________________________________________________________________________________________________

_______________________________________________________________________________________________________

Please complete other side

Marine Technology Society Membership Applicationand MTS Journal Subscription Request

NON-MEMBER AND LIBRARY SUBSCRIPTIONS TO MTS JOURNAL ONLINEOnline only .......................................................$465 Print and Online (worldwide)............................. $499

$38

Optional Information:

n Male n Female What is your age? n Under 30 n 30-40 n 41-50 n 51-60 n Over 60

Marine Technology Society Membership Application (continued)

EDUCATIONAL INFORMATIONPlease fill out the following information about yourself:

Check your highest level of education:n High School Diploma n Associate (2 yr.) Degree n Four Year Degree n Graduate Degree n Doctorate

Check all that apply: n B.S. n B.E. n B.A. n M.S. n M.A. n M.E. n Ph.D. n Sc.D.

Do you have a P.E. license? n YES n NO

BUSINESS/PROFESSIONAL INFORMATION

Name of current employer: _________________________________________________________________________________

Your employer’s primary line of business at your location: ________________________________________________________

If you don’t work for an employer, please identify your business: ___________________________________________________

If military, rank: _________________________________________________________________________________________

Your principal jobfunction/responsibilities:

Your job title: Check areas of interest:

Four easy ways to join!Mail: Send application with check or credit card info to:

Marine Technology Society/1100 H Street NW, Suite LL-100/Washington, DC 20005Fax: Fax application to: 202-347-4302 (credit card payments only)Online: Apply online at www.mtsociety.orgPhone: Contact us at: 202-717-8705

MEMBERSHIP AND JOURNAL PAYMENTPayment Method: n Check Enclosed CK#____________ n Master Card n Visa n Diners Club n Am ExMake checks payable to the Marine Technology Society (U.S. funds only)

Card #: __________________________________________________________ Expiration Date: ________________________

Signature: ________________________________________________________ Date: _________________________________

TOTAL PAYMENT:

Membership: $_____________

Journal: $_____________

TOTAL: $_____________

__Ocean Economic Potential__Ocean Exploration__Ocean Observing Systems__Ocean Pollution__Oceanographic Instrumentaion__Offshore Structures__Physical Oceanography & Meteorology__Remote Sensing__Remotely Operated Vehicles__Renewable Energy__Ropes & Tension Members__Seafloor Engineering__Underwater Imaging__Unmanned Maritime Vehicles__Other (please specify)_______________________

__Arctic Technology__Buoy Technology__Cables & Connectors__Deepwater Field Development__Diving__Dynamic Postioning__Manned Underwater Vehicles__Marine Archaeology__Marine Education__Marine Geodetic Information Systems__Marine Law & Policy__Marine Materials__Marine Mineral Resources__Marine Security__Moorings

__Engineering Management__Science Management__Sales__Marketing__Administration__Policy Making, Regulatory__Public Affairs__Engineering Design__Mechanical Engineering__Software Engineering__Education/Teaching__Legal__Consulting__Retired__Other (please specify)_______________________

__President/CEO/COO__Owner/Partner__VP, Senior Manager__Project Manager, Engineering__Project Manager, Other__Corporate VP, Engineering__Engineering Director__Chief/Senior Engineer__Chief/Senior Scientist__Project Manager__Engineer__Operations VP__Scientist__Other (please specify)_______________________

CVV Number (Required): ______________ (Visa, MasterCard and Discover Card — The three digit number is located to the right of the regular number on the back of the card. American Express Card — the four digit number is located on the front of the card in either the upper-left or upper-right side of the regular number.)

C ORP OR AT E MEMBE RSABCO Subsea, Houston, Texas

AMETEK Sea Connect Products, Inc., Westerly,

Rhode Island

Bibby Subsea, Houston, Texas

BMT Scientific Marine Services Inc., Houston, Texas

C-MAR Group, Houston, Texas

Compass Publications, Inc., Arlington, Virginia

DeepSea Technologies, Inc., Houston, Texas

Deloitte & Touche, LLP, Houston, Texas

Delta SubSea, LLC, Montgomery, Texas

DNV GL, Houston, Texas

DOF Subsea USA, Inc., Houston, Texas

EMAS AMC, Houston, Texas

Fluor Offshore Solutions, Sugar Land, Texas

Forum Energy Technologies, Houston, Texas

Fugro Chance, Inc., Lafayette, Louisiana

Fugro Pelagos, Inc., San Diego, California

Fugro USA, Houston, Texas

Geospace Offshore Cables, Houston, Texas

GE Energy Power Conversion, Houston, Texas

Harris Corporation, Melbourne, Florida

Hydroid, LLC, Pocasset, Massachusetts

Innerspace Corporation, Covina, California

InterMoor, Inc., Houston, Texas

INTECSEA, Houston, Texas

Klein Marine Systems, Inc., Salem, New Hampshire

Kongsberg Maritime, Inc., Houston, Texas

L-3 MariPro, Goleta, California

Lockheed Martin Sippican, Marion, Massachusetts

MacArtney, Inc., Houston, Texas

Marine Cybernetics AS, Trondheim, Norway

Mitsui Engineering and Shipbuilding Co. Ltd., Tokyo,

Japan

Oceaneering Advanced Technologies, Hanover, Maryland

Oceaneering International, Inc., Houston, Texas

OceanWorks International, Houston, Texas

Oil States Industries, Inc., Arlington, Texas

Phoenix International Holdings, Inc., Largo, Maryland

Quest Offshore Resources, Sugar Land, Texas

Saipem America Inc., Houston, Texas

SBM Offshore, Houston, Texas

Schilling Robotics, LLC, Davis, California

Schottel, Inc., Houma, Louisiana

SEACON, El Cajon, California

Sebastion AS, Ulsteinvik, Norway

SonTek/YSI Inc., San Diego, California

South Bay Cable Corp., Idyllwild, California

Southwest Electronic Energy Group, Stafford, Texas

Stress Engineering Services, Inc., Houston, Texas

Technip, Houston, Texas

Teledyne Marine Systems, North Falmouth, Massachusetts

Teledyne Oil & Gas, Daytona Beach, Florida

Teledyne RD Instruments, Inc., Poway, California

Universal Pegasus International, Houston, Texas

Vencore, Inc., Stennis Space Center, Mississippi

Wartsila Corporation, Houston, Texas

Whitefield Plastics, Houston, Texas

Wood Group Kenny, Inc., Houston, Texas

Wood Group Mustang, Houston, Texas

BUSINE S S MEMBE RSAshtead Technology Offshore, Inc., Houston, Texas

ASV, LLC, Broussard, Louisiana

Atlantis Deep Sea Ltd, Valletta, Malta

Baker Marine Solutions, Mandeville, Louisiana

Bastion Technologies, Inc., Houston, Texas

BioSonics, Inc., Seattle, Washington

Blade Offshore Services Ltd., Gosforth, Newcastle

upon Tyne, UK

Braemar Engineering, Houston, Texas

CARIS, Alexandria, Virginia

C-LARs LLC, Bryan, Texas

DeepSea Power and Light, San Diego, California

Deepwater Rental and Supply, New Iberia, Louisiana

DPI Pte Ltd., Singapore

Energy Sales, Redmond, Washington

Falmat, Inc., San Marcos, California

GAMbIT Marine Solutions, Alexandria, Virginia

Global Marine Consultants and Surveyors, Ltd,

Alexandria, West Dunbartonshire, UK

GMC Inc., Houston, Texas

GRI Simulations Inc., Mount Pearl, Newfoundland

and Labrador, Canada

Horizon Marine, Inc., Marion, Massachusetts

INSPIRE Environmental, Newport, Rhode Island

Kongsberg Maritime Contros/Embient, Kiel, Germany

Laser Tools Co., Inc., Little Rock, Arkansas

Liquid Robotics, Sunnyvale, California

London Offshore Consultants, Inc., Houston, Texas

M3 Marine Expertise Pte Ltd., Singapore

Mactech Offshore, Inc., Broussard, Louisiana

Makai Ocean Engineering, Inc., Kailua, Hawaii

Maritime Assurance & Consulting Ltd., Aberdeen,

United Kingdom

Mirage Subsea Inc., Houston, Texas

North Pacific Crane Company, Seattle, Washington

Offshore Analysis & Research Solutions (OARS), LLC,

Austin, Texas

Poseidon Offshore Mining, Oslo, Norway

QPS, Portsmouth, New Hampshire

Remote Ocean Systems, Inc., San Diego, California

Rowe Technologies Inc., Poway, California

SeaLandAire Technologies, Inc., Jackson, Mississippi

Sonardyne, Inc., Houston, Texas

Soundnine, Inc., Redmond, WA

Sound Ocean Systems, Inc., Redmond, Washington

Stress Engineering Services, Inc., Houston, Texas

SURF Subsea, Inc., Magnolia, Texas

TALON Technical Sales, Inc., Houston, Texas

Technology Systems Corporation, Palm City, Florida

Teledyne SeaBotix, San Diego, California

Tension Member Technology, Huntington Beach,

California

Triton Submarines LLC, Vero Beach, Florida

Ultra Deep, LLC, Houston, Texas

Unitech International, Houston, Texas

VideoRay, LLC, Pottstown, Pennsylvania

INS T I T U T ION A L MEMBE RSCity of St. John’s, Newfoundland and Labrador, Canada

CLS America, Inc., Largo, Maryland

Consortium for Ocean Leadership, Washington, DC

Department of Business, Tourism, Culture and

Rural Development, St. John’s, Newfoundland

and Labrador, Canada

Fundação Homem do Mar, Rio de Janeiro, Brazil

Harbor Branch Oceanographic Institute, Fort Pierce,

Florida

International Seabed Authority, Kingston, Jamaica

Jeju Sea Grant Center, Jeju-City, South Korea

Marine Institute, Newfoundland and Labrador, Canada

Matthew Fontaine Maury Oceanographic Library,

NAVOCEANO, Stennis Space Center, Mississippi

Monterey Bay Aquarium Research Institute, Moss Landing,

California

Metanomy, Inc, 501(c)(3), Saint Petersburg, Florida

NOAA/PMEL, Seattle, Washington

OceanGate, Inc., Everett, Washington

Oregon State University College of Oceanic and

Atmospheric Sciences, Corvalis, Oregon

Schmidt Ocean Institute, Lake Forest Park, Washington

Stockton University, Port Republic, New Jersey

University of Massachusetts–Dartmouth, New Bedford,

Massachusetts

The Marine Technology Society gratefully acknowledges the critical support of the Corporate, Business, and Institutional members listed.Member organizations have aided the Society substantially in attaining its objectives since its inception in 1963.

Marine Technology Society Member Organizations

Postage for periodicalsis paid at Washington, DC

and additional mailing offices.1100 H Street NW, Suite LL-100Washington, DC 20005

the international, interdisciplinary society devoted to ... · program administrator editorial...

Documents