Distant touch hydrodynamic imaging with an artificial lateral line

Coombs, Douglas L. Jones, and Chang Liu Yingchen Yang, Jack Chen, Jonathan Engel, Saunvit Pandya, Nannan Chen, Craig Tucker, Sheryl

doi:10.1073/pnas.0609274103 published online Nov 28, 2006; PNAS

This information is current as of November 2006.

www.pnas.org#otherarticlesThis article has been cited by other articles:

E-mail Alerts. click hereat the top right corner of the article or

Receive free email alerts when new articles cite this article - sign up in the box

Rights & Permissions www.pnas.org/misc/rightperm.shtml

To reproduce this article in part (figures, tables) or in entirety, see:

Reprints www.pnas.org/misc/reprints.shtml

To order reprints, see:

Notes:

Distant touch hydrodynamic imagingwith an artificial lateral lineYingchen Yang*, Jack Chen*, Jonathan Engel*, Saunvit Pandya*, Nannan Chen*, Craig Tucker*, Sheryl Coombs†,Douglas L. Jones‡, and Chang Liu*§

*Micro and Nanotechnology Laboratory, and ‡Coordinated Science Laboratory, University of Illinois at Urbana-Champaign, 208 North Wright Street,Urbana, IL 61801; and †Department of Biology, Bowling Green State University, Bowling Green, OH 43403

Communicated by Thomas J. Hanratty, University of Illinois, Urbana, IL, October 23, 2006 (received for review June 27, 2006)

Nearly all underwater vehicles and surface ships today use sonarand vision for imaging and navigation. However, sonar and visionsystems face various limitations, e.g., sonar blind zones, dark ormurky environments, etc. Evolved over millions of years, fish usethe lateral line, a distributed linear array of flow sensing organs,for underwater hydrodynamic imaging and information extrac-tion. We demonstrate here a proof-of-concept artificial lateral linesystem. It enables a distant touch hydrodynamic imaging capabilityto critically augment sonar and vision systems. We show that theartificial lateral line can successfully perform dipole source local-ization and hydrodynamic wake detection. The development ofthe artificial lateral line is aimed at fundamentally enhancinghuman ability to detect, navigate, and survive in the underwaterenvironment.

dipole localization � hot wire anemometer � micromachining � wakedetection � neuromast

A lateral line is a spatially distributed system of flow sensorsfound on the body surface of fish (1) and aquatic amphib-

ians (2). It is comprised of arrays of neuromasts, which can beclassified into two types: superficial neuromasts and canalneuromasts (Fig. 1). A superficial neuromast is situated on thesurface of the fish and responds in proportion to fluid velocity(3, 4). In contrast, a canal neuromast is packaged in fluid-filledcanals located beneath the surface of the skin and are commonlydescribed as a detector of outside water acceleration that isproportional to the pressure gradient (1, 3–6). As an integratedflow sensing system, such lateral lines form spatial-temporalimages of nearby sources based on their hydrodynamic signatures(1, 3, 5) and provide mechanosensory guidance for many dif-ferent behaviors, including synchronized swimming in schools,predator and obstacle avoidance, prey detection and tracking,rheotaxis, and holding station behind immersed obstacles instreams (4, 7, 8). This ‘‘distant touch’’ sense complements othersensory modalities, including vision and hearing, to increasesurvivability in unstructured environments. To date, there hasnever been an engineering equivalent of the fish lateral linesystem for underwater vehicles and platforms. The goal of thepresent research is to build an artificial lateral line that mimicsthe functional organization and imaging capabilities of thebiological one. The artificial lateral line can facilitate fundamen-tal studies of biological systems and provide unprecedentedsensing and control functions to underwater vehicles and plat-forms. Specifically, we envision that the distant touch hydrody-namic imaging capability of the artificial lateral line can providea new sense in addition to sonar and vision. In this article, wedemonstrate the functions of an artificial lateral line under twobiologically relevant scenarios: (i) localizing a moving target withflapping part (7, 9, 10) and (ii) imaging a hydrodynamic trail forprey capture (11, 12).

Development of the Artificial Lateral LineThe artificial lateral line consists of a monolithically integratedarray of microfabricated flow sensors (13), with the sizes of

individual sensors and spacings between them matching those ofthe biological counterpart. The sizes of individual sensors andintersensor spacings are �50 �m to 2 mm across biologicalspecies (14). The small dimensions of the sensors are importantto ensure that they pose minimum interference with the flowfield and with each other. An engineering equivalence of thelateral line sensing organ has not been made before. The makingof miniaturized sensors in dense arrays is beyond the capabilitiesof conventional machining and commercial f low sensors. For-tuitously, advancements made in the area of microengineering inthe past two decades now allow us to build sensors that match thesizes and mimic the functions and organization of the lateral line.Our artificial lateral line consists of a linear array of flow sensors,much like the array of neuromasts along the trunk of many fish.In the present version, each individual sensor within the array isbased on the thermal hot wire anemometry (HWA) principle(15). The HWA sensors exhibit high sensitivity, small dimen-sions, and consequently, reduced interference to flow field.

We used a recently developed surface micromachining tech-nique to fabricate arrays of miniaturized HWAs (13) that havedimensions on the same order of magnitude (tens to hundredsof microns tall) as biological neuromasts (Fig. 2). The hot wiredoes not reside in the substrate plane (i.e., the bottom of theboundary layer) but rather is elevated above the substrate by twoprongs, just as some superficial neuromasts in fish are elevatedabove the skin surface by papillae (14). The sensor is first madein-plane by using photolithography. It is then assembled out-of-plane by using a 3D magnetic assembly method (16). Theresultant elevation corresponds to the design length of theprongs defined by photolithography. Because of the use of aphotolithography process, micromachined HWA sensors canhave a prong length from 50 �m to 2 mm. For the presentapplication, we focused on a wire length of 400 �m and anelevation of �600 �m. The hot wire consists of a nickel filamentthat is sandwiched by two layers of polyimide, which serve aspassivation and structural support. The nickel hot wire exhibitsa temperature coefficient of resistance (�) of 4,100 ppm/°C.

The largest array we have constructed so far consists of 16HWA sensors with 1-mm spacing (Fig. 2B). Each sensor ismonolithically integrated with dedicated complementary metal–oxide–semiconductor circuitry for on-chip signal conditioning,noise-f loor reduction, and parallel data acquisition. Upon com-pletion of this sensor array, the velocity sensitivities of theindividual sensors were characterized by moving the array inquiescent water at various speeds. Under constant temperaturemode with overheat ratio of 0.1, our measurements indicate athreshold of 200 �m/s and a bandwidth of 1 KHz (17). The sensor

Author contributions: Y.Y. and C.L. designed research; Y.Y. and J.C. performed research;Y.Y., J.C., J.E., S.P., N.C., C.T., S.C., and D.L.J. contributed new reagents/analytic tools; Y.Y.analyzed data; and Y.Y. and C.L. wrote the paper.

The authors declare no conflict of interest.

Abbreviation: HWA, hot wire anemometry.

§To whom correspondence should be addressed. E-mail: changliu@uiuc.edu.

© 2006 by The National Academy of Sciences of the USA

www.pnas.org�cgi�doi�10.1073�pnas.0609274103 PNAS � December 12, 2006 � vol. 103 � no. 50 � 18891–18895

NEU

ROSC

IEN

CEEN

GIN

EERI

NG

array was also calibrated in a water channel under the sameconditions with inflow velocity up to 0.25 m/s, and a typicalnonlinear trend for each sensor was observed (17). Furthercharacterization showed that the drifts of zero outputs for all ofthe sensors were negligibly small over a testing period of 2 h.

Application on Dipole Source LocalizationThe developed artificial lateral line was used for hydrodynamictesting under biologically relevant scenarios. We first establishedthe performance and functionality of our artificial lateral line byrecording the spatial-temporal response to a nearby dipolesource. The dipole is simple and yet ubiquitous in the underwaterworld. When a fish swims, its tail beat causes a dipolar near-fieldflow in addition to a wake behind it (10, 18, 19). Certainpredators can accurately localize and attack a prey that is nearby(i.e., at a distance equivalent to one or two fish body lengthsaway) (9, 10, 20) solely by using the lateral line system to measurethe dipole field associated with the prey.

A vibrating sphere was used to function as a dipole source (Fig.3A). Theoretical prediction of pressure gradient felt by a canallateral line in response to a nearby dipole has been made in thepast (21, 22) and verified by neuro-physiological studies (9, 20).The spatial distribution of pressure gradient amplitude resem-

bles a ‘‘Mexican hat’’ profile; the magnitude is the highest at theprojected center of the dipole and diminishes gradually withincreasing lateral distance from the center. Our artificial lateralline was able to record a profile well matched to the theoreticalprediction (Fig. 3B), which was accomplished despite the dif-ferences between transfer functions of the biological canalneuromasts and HWAs. It should be noted that HWA sensorssubjected to periodic oscillating flow sense water particle dis-placement (23), not velocity or pressure gradient. However, afternormalization they all have the same spatial-distribution profile(21, 22). The HWA sensor poses another challenge; namely, itcannot discern the flow polarity. The amount of heat convectedfrom the sensing filament of a HWA depends on the magnitudeof water particle displacement for oscillating flows, not on itspolarity. As a result, the HWA sensor provides a rectifiedreading of a complex oscillatory flow field (24, 25). To recoverthe signal rectification, we developed a derectification tech-nique. It is based on reconstruction of Fourier series expansionof a signal, with initial phase angles among terms obeyinganalytically derived relationships, and amplitudes recovered bysignal squaring method.

Inspired by fish behaviors (9, 26, 27), we demonstrate that thespatially distributed response from the lateral-line array can beused to identify the exact location of a moving dipole source. Wefind that the location of a dipole source is encoded in the location

Fig. 1. Schematics of lateral line neuromasts. (A) A superficial neuromast ofa clawed frog (6). (B) A lateral line neuromast in a canal of ruffe (6). (C) A lateralline periphery of a teleost fish (4). Superficial neuromasts are represented bydots, and canal neuromasts are represented by dots inside shaded strips.

Fig. 2. Schematic representation artificial lateral line. (A) Schematic of anindividual microfabricated, out-of-plane HWA sensor used to build artificialneuromasts. The hot wire is elevated above the substrate surface by a pre-scribed distance. (B) Scanning electron micrograph of the artificial lateral linewith 16 HWA sensors spaced 1 mm apart.

18892 � www.pnas.org�cgi�doi�10.1073�pnas.0609274103 Yang et al.

and amplitude of the apex. A signal-processing algorithm basedon maximum-likelihood analysis was developed (28). It com-pares the pattern of the signal received by the array with theexpected pattern at all positions and selects the best match as theestimate of the actual dipole location. The algorithm can predictthe location of the dipole even when the apex lies outside of thelength of the lateral line. We illustrate the capability of thisdipole tracking ability by using three representative cases. In alltests, the strength of the dipole source was maintained constant.The strength level does not affect the testing results much, aslong as a proper signal-to-noise ratio at measurement locationsis achieved. The dipole was confined to move within an area thatcovers two body lengths along the artificial lateral line, and onebody length away from it, where body length represents thelength of the sensor array. In the first case, the dipole source istranslated stepwise parallel to the artificial lateral line (path 1 ofFig. 3E). The lateral output recorded at each step has identicalamplitude with shifting apex (Fig. 3C). In the second case, thedipole source is moved perpendicularly away from the lateralline. The profiles of lateral line output flatten out when thedipole source fades into the distance (Fig. 3D). In a third case,the dipole traverses in a complex path in the plane of the sensorarray (path 3, Fig. 3E). It is evident that for the three represen-

tative paths (Fig. 3E) predictions are accurate in most locationsand generally become less accurate with increased distance fromthe sensor array.

Application on Hydrodynamic Wake CharacterizationIn addition to localizing a swimming prey in a dipolar near field,following the wake behind the prey can allow a predator to trackand eventually localize the prey, starting at a much greaterdistance, i.e., a few to a few tens of prey-body length (11, 12).A functional lateral line is indispensable for following wake(11, 12).

The wake behind a swimming fish contains organized vortices(10, 29–31). Following this inspiration, we generated a turbulentwake by using a vertically mounted circular cylinder placed inwater flow (Reynolds number � 5,000) (Fig. 4). The wakeconsists of alternately shed large-scale vortices known as aKarman Street (32) (Fig. 4A). We found that by using anartificial lateral line one can identify the signature of a wake andthe general direction of the source. Our artificial literal line wasexposed to the wake to record the spatial distribution of localvelocity f luctuations. To cover the desired size of the field of view(3.5D wide and 6D deep, with D being the diameter of thecylinder), we traversed the sensor array across the wake andstitched multiple images (Fig. 4A).

Fig. 3. Characterization of dipolar near field and localization of the dipole source. (A) Analytical model (21, 22) of pressure contours (blue lines) and a lineararray of lateral line canal neuromasts (in orange). (B) Comparison of experimental (green lines) and analytical (21, 22) (red line) results on displacement amplitudeof water particles. (C) Time-elapsed spatial profiles of displacement amplitude with step-by-step translation of the dipole source along the artificial lateral linefollowing path 1 diagramed in E. (D) Displacement profiles under step-by-step translation following path 2 indicated in E. (E) Comparison of actual paths (solidline with filled circles) and predicted ones (dashed line with empty circles).

Yang et al. PNAS � December 12, 2006 � vol. 103 � no. 50 � 18893

NEU

ROSC

IEN

CEEN

GIN

EERI

NG

The distribution of rms velocity fluctuation (Fig. 4B) showsthe that lateral line array is capable of capturing the main featureof the wake (33): dramatic dual peaks with a valley in betweenin the near wake, and decreasing intensity of the peaks furtherdown stream.

The velocity fluctuation was determined by using a secondmethod that accentuates peak features with even greater con-trast. This method uses fast Fourier transform to obtain spectraldistribution of a signal. Then the flow velocity amplitude at acharacteristic frequency associated with the wake-generatingsource, e.g., the vortex shedding frequency from the cylinder, isextracted (Fig. 4C). As a result, two clearly defined peaksoccurred along the entire field of view. We conjecture that thisalgorithm lowers the background signal by rejecting broadbandnoise in the fluid.

DiscussionThe current study illustrates the potential of biomimetic sensingof artificial lateral line-equipped underwater vehicles. In a neardistance, for example, the capability of a lateral line on dipolesource localization enables a fish to capture a prey or evade apredator by solely discerning the dipolar near field generated by

others. By demonstrating the same capability of our artificiallateral line, it is convincing that with further engineering ad-vancement, manmade underwater vehicles will be able to imagehydrodynamic events from surroundings, like fish do. Suchhydrodynamic events can be caused by aquatic animals or othervehicles. With advanced algorithms and training methods, thiscapability can even be extended to general water disturbancesother than an idealized dipole source.

From a distance away, fish can still rely on their lateral linesto track a target by detecting the target’s hydrodynamic trail. Ourartificial lateral line demonstrates this capability, and the trail isnot limited to a wake generated by a cylinder only. Instead, it canbe any kinds of trails, e.g., a wake behind a propeller-drivensubmarine. However, different trails might have different fea-tures, thus they may require different data processing techniquesto obtain a sharp definition of the features, based on measure-ments of the artificial lateral line. For example, for a complexwake with no dominant feature frequencies, a rms evaluation ofthe fluctuating velocities might be a good measure (see Fig. 4B).Whereas with a dominant frequency, the wake might be betterdefined by extracting this frequency component based on spec-tral analysis (see Fig. 4C), which is especially essential when theaquatic environment is contaminated with hydrodynamic noises.

However, the artificial lateral line discussed in the foregoingdiffers from its biological counterpart in many aspects. Struc-turally, a real lateral line consists of numerous superficial andcanal neuromasts, with each neuromast having a bundle of haircells capsulated in a cupula (34) to function as a flow sensor. Theartificial lateral line, on the other hand, has only a limitednumber of superficially placed HWAs to serve this purpose. Forinformation collection, aquatic animals acquire firing frequen-cies from each neuromast (3) to represent the strength of localf low; whereas the artificial lateral line records magnitude ofvoltage output from individual HWAs. For decision making,aquatic animals use a back-propagation learning algorithm (35);comparatively, the artificial lateral line uses the minimum mean-squared error method (28). There is no doubt that in many waysa real lateral line, after millions of years of evolution, is superiorto the artificial lateral line presented herein. Nonetheless, thisartificial lateral line enables biological behaviors to be realized,such as localizing a vibration source and detecting a hydrody-namic wake.

Future work should involve both close studies of biology andfurther development of engineering. Biological studies areneeded at different levels, including the organism level, sensorlevel, and cellular level. These studies may involve differentaspects, including behavior studies, neurophysiology studies,f luid mechanics, and morphological studies. Significant engi-neering efforts are also necessary to improve the performanceand prove functions in a noisy environment. Engineering devel-opment in the areas of advanced materials, sensor developmentand integration, signal processing, control, and robotics isneeded in the future.

In summary, the proof-of-concept biomimetic study shows thatthe developed artificial lateral line is able to localize an underwatervibrating source in a near distance and is able to detect a hydro-dynamic wake for long-distance tracking. The accomplishments areof significant importance in various underwater applications. Beingequipped with such an artificial lateral line system, a submarine isexpected to be capable of detecting and tracking other movingunderwater targets and to be capable of collision avoidance. Anartificial lateral line is especially indispensable when vision andsonar are limited, such as in dark or murky environments, in sonarblind-zones, or when the sonar system is in passive mode forconcealment purposes. Similarly, by applying the technique to otherautonomous underwater vehicles, this augmented hydrodynamic-imaging capability will enable safer and more flexible navigationperformance.

Fig. 4. Wake signatures for tracking a source. (A) Schematic showingexperimental set-up. (B) The pattern of rms water velocity in the wake of acylinder. (C) The pattern of peak water velocity at vortex shedding frequencyin the wake of a cylinder. Both rms and peak water velocities were normalizedby free-stream inflow velocity.

18894 � www.pnas.org�cgi�doi�10.1073�pnas.0609274103 Yang et al.

MethodsFabrication. The fabrication process of the artificial lateral lineinvolves surface micromachining, a technique commonly usedin microelectromechanical systems to create freestanding can-tilevers (36). The micromachining process is performed on asilicon wafer with preexisting analog integrated circuitry.Surface micromachining is used to define the HWA, includinga nickel-iron alloy support prong and a nickel-polymer com-posite hot-wire sensing element. The planar surface microma-chining process is followed by a magnetically assisted assemblystep that rotates the cantilevers out of plane. With this process,many devices can be bent out of plane in parallel with highyield (13). Further, the sensor elements can be monolithicallyintegrated with signal processing electronics, as the process isperformed at room temperature and would not cause incom-patibility with elements of electronics circuits. At the end ofthe process, the hot wire may be encapsulated by a conformallydeposited, 2-�m-thick Parylene film for waterproofing andfurther structure strengthening.

Dipole Experiments. A minishaker (model 4010; B&K, Norcross,GA) was fixed to a motorized three-axis linear stage system(model 8MT175; Standa, Vilnius, Lithuania) equipped withcomputerized motion control through step motors. A sphere of3 mm in diameter was attached to the minishaker through a12-gauge needle that served as the dipole source. An acceler-ometer (model 352B10; PCB, Depew, NY) was attached to thebase of the needle to measure the acceleration of the dipole. The

artificial lateral line, the packaged HWA array, was mounted ona test fixture rigidly attached to the base of the water tankthrough a suction cup. For all experiments, arrayed HWAsensors were aligned parallel to the vibration axis of the sphere(i.e., the dipole source), which was operated in sinusoidal modeby the minishaker at a frequency of 75 Hz with displacementamplitude of 0.4 mm.

Wake Experiments. A desktop water tunnel (model 501; ELD Inc.,Lake City, MN) with a test section of 150 � 150 mm was used.A cylinder of 25 mm in diameter was vertically fixed in a uniformcurrent at a speed of 0.2 m/s. The corresponding Reynoldsnumber was �5,000. The artificial lateral line was exposed in thewake behind the cylinder, with arrayed HWA sensors perpen-dicular to the inflow and to the axis of the cylinder. The printedcircuit board holding the substrate where sensors resided wastilted at an angle of attack of 5° to suppress flow separation fromthe leading edge on the sensor side. To correct sensitivitydifference among sensors and the nonlinear response of eachindividual sensor, the sensor array was calibrated under the samecondition without the cylinder.

We thank Professor Marcelo H. Garcıa and Mr. Andrew R. Waratukeof the Hydrosystem Laboratory, University of Illinois at Urbana–Champaign, for their generosity and kindness for making their facilitiesavailable for our experiments. This work was supported by the Bioin-spired Concepts program, which is funded by the Air Force Office ofScientific Research, and the BioSenSE program, which is funded by theDefense Advanced Research Projects Agency.

1. Dijkgraaf S (1962) Biol Rev 38:51–105.2. Gorner P (1963) Z Vergl Physiol 47:316–338.3. Engelmann J, Hanke W, Mogdans J, Bleckmann H (2000) Nature 408:51–52.4. Coombs S (2001) Auton Robot 11:255–261.5. Coombs S, Georner P, Meunz H (1989) The Mechanosensory Lateral Line:

Neurobiology and Evolution (Springer, New York).6. van Netten SM (2006) Biol Cybern 94:67–85.7. Hassan ES (1985) Biol Cybern 52:23–36.8. Montgomery JC, Macdonald F, Baker CF, Carton AG (2002) Brain Behav Evol

59:190–198.9. Coombs S, Hastings M, Finneran J (1996) J Comp Physiol A 178:359–371.

10. Bleckmann H (1994) Reception of Hydrodynamic Stimuli in Aquatic andSemiaquatic Animals (Fischer, New York).

11. Pullman K, Grasso FW, Breithaupt T (2001) Proc Natl Acad Sci USA98:7371–7374.

12. Pohlmann K, Atema J, Breithaupt T (2004) J Exp Biol 207:2971–2978.13. Chen J, Liu C (2003) J Microelectromech Syst 12:979–988.14. Coombs S, Janssen J, Webb JF (1988) in Sensory Biology of Aquatic Animals,

eds Atema J, Fay RR, Popper AN, Tavolga WN (Springer, New York), pp553–593.

15. Bruun HH (1995) Hot-Wire Anemometry: Principles and Signal Analysis (OxfordUniv Press, New York).

16. Zou J, Chen J, Liu C, Schutt-Aine JE (2001) J Microelectromech Syst 10:302–309.

17. Yang Y, Pandya S, Chen J, Engel J, Chen N, Liu C (2006) in CANEUS 2006:Micro-Nano Technology for Aerospace, ed Dastoor M (Am Soc MechanicalEngineers, New York), pp 160–165.

18. Denton EJ, Gray JAB (1989) in The Mechanosensory Lateral Line: Neurobiologyand Evolution, eds Coombs S, Gorner P, Munz H (Springer, New York), pp229–245.

19. Kalmijn AJ (1997) Acta Physiol Scand Suppl 161:25–38.20. Blake BC, van Netten SM (2006) J Exp Biol 209:1548–1559.21. Kalmijn AJ (1985) in Sensory Biology of Aquatic Animals, eds Atema J, Fay RR,

Popper AN, Tavolga WN (Springer, New York), pp 83–130.22. Kinsler LE (2000) Fundamentals of Acoustics (Wiley, New York).23. Dubbelday PS (1989) Rev Sci Instrum 60:2745–2749.24. Dubbelday PS (1986) J Acoust Soc Am 79:2060–2066.25. Coombs S, Fay RR, Janssen J (1989) J Acoust Soc Am 85:2185–2193.26. Sand O (1981) in Hearing and Sound Communication in Fishes, eds Tavolga

WN, Popper AN, Fay RR (Springer, New York), pp 459–480.27. Hassan ES (1989) in The Mechanosensory Lateral Line: Neurobiology and Evolu-

tion, eds Coombs S, Gorner P, Munz H (Springer, New York), pp 217–228.28. Pandya S, Yang Y, Jones DL, Engel J, Liu C (2006) EURASIP J Appl Signal

Process 2006:1–8.29. Dehnhardt G, Mauck B, Hanke W, Bleckmann H (2001) Science 293:102–104.30. Bleckmann H, Breithaupt T, Blickhahn R, Tautz J (1991) J Comp Physiol A

168:749–757.31. Nauen JC, Lauder GV (2002) J Exp Biol 205:1709–1724.32. Zdravkovich MM (1997) Flow Around Circular Cylinders (Oxford Univ Press,

Oxford), Vol 1.33. Dong S, Karniadakis GE, Ekmekci A, Rockwell D (2006) J Fluid Mech 569:185–207.34. Flock A (1965) Acta Oto-Laryngol 199:(Suppl)1–90.35. Haykin S (1999) Neural Networks (Prentice–Hall, Upper Saddle River, NJ).36. Liu C (2005) Foundations of MEMS (Prentice–Hall, Upper Saddle River, NJ).

Yang et al. PNAS � December 12, 2006 � vol. 103 � no. 50 � 18895

NEU

ROSC

IEN

CEEN

GIN

EERI

NG