arabia.ni.comarabia.ni.com/sites/default/files/predictive maintenance vibration... · 7 as we look...

5

The above diagram is pretty familiar to us all; degradation is inevitable and a given; we know that if we operate something, it will degrade over time and eventually fail So, condition monitoring helps us understand where a piece of equipment lies, at any time, on the curve so that we can take the most appropriate action i.e. preventative maintenance, predictive maintenance or do nothing and let the item run to failure. From my perspective, the factors that help you decide which of these actions to take depends on 3 things: Safety, cost, and inconvenience. Whenever anything fails, one or more of those factors come into play to some varying degree. Just to illustrate this simply, in an airplane if the engine fails, it’s a major safety issue, with some element of cost (for maintenance and diversion action etc) and yes, the passengers are inconvenienced if it involves a change to their schedule, so while all 3 elements are present, it’s safety that’s the main issue. If, on that same airplane, the overhead passenger reading light doesn’t work or the in flight entertainment system doesn’t work, inconvenience is the major factor as opposed to safety or cost (it would only be unsafe if the passengers got unruly because they were so disgruntled!). Finally, if the cost of an unplanned shutdown on a production line (car plant or diaper making machine) is so high that the company wants to avoid such an event as much as possible, then cost is the driver as opposed to safety or inconvenience. The point is, a CM system can serve and satisfy many different purposes. So, I suggest one needs to look at any degradation and eventual failure from those perspectives and answer the question “ Do I need a condition monitoring system to look for evidence of the specific failure, and if I do, for what reason (cost, safety, inconvenience) and will it be to perform preventative, or predictive maintenance (a run to failure decision suggests you don’t need a CM system). With each maintenance philosophy, there’s pos and cons which is where the cost benefit analysis really starts. -A preventative maintenance strategy is completely dependent on time, not machine performance, and is often based on manufacturer’s recommended maintenance plans. This means that repairs be performed when not really needed, causing undue downtime, yet equipment deterioration that is happening faster than anticipated by the regular scheduled inspections can still cause unexpected or, as I like to call them, surprise failures.

-Run to failure is a maintenance approach that is completely dependent on machine performance, only repairing when the machine fails. It extends the amount of machine uptime but can significantly increase repair costs because of the increased risk of a catastrophic failure. I suggest that the run to failure option is reserved for the innocuous and trivial events such as the overhead reading light failing on the airplane; anything that involves unscheduled downtime, increased maintenance and associated increased costs will usually warrant a predictive type philosophy to be adopted.

-Predictive maintenance, the strategy we have been discussing, attempts to extend machine uptime through acceptable performance by only performing repairs when on the cusp of dropping below acceptable operating levels.

There are many measurements that can be made to monitor the condition of the asset. These measurements, change over time as the asset degrades or wears.

6

7

As we look at condition monitoring, and even at factory acceptance or design verification testing we see several sensory measurements which provide a view of the health of the machine. National Instruments offers a platform which incorporates hardware to make direct measurements from analog and digital sensors, as well as communicates with control systems to simultaneously note the operational state of the machine being monitored or tested. There are several sensory measurements to be made in Asset Health Monitoring. Vibrations are the most common, and offer the most information about the health of machine components. A multi-month prediction based on changing vibration levels at the specific frequencies of bearing, gearbox and other mechanical component provides ample time to schedule production, maintenance staff, and the availability of spare parts. In turn, the time to repair is reduced since everything is in place for the repair.

Many operators of rotating machines are working to optimize their equipment maintenance process. These operators want to move from manual vibration data collection to an automated continuous monitoring system. In this way, the vibration analysts and other machinery experts are able to spend more time review sensory vibration readings as compared to collecting the readings. By bringing data from multiple machines together in one place, it becomes possible to conduct comparative analysis, comparing one pump’s vibration pattern to another. Further, collaboration is easier with an on-line monitoring system, and it is easier to integrate vibration sensory data with other maintenance records.

8

The Center for Intelligent Maintenance Systems, www.imscenter.net, offers a systematic approach to prognostics design and implementation. First, the specific assets of interest and the return on investment (business needs) should be articulated. These business needs serve as the benchmarks or goals of the implemented system. Once the assets of interest are identified, the critical components of the assets should be identified. Critical components can be identified using FMEA (failure modes and effects analysis), historical maintenance records, or by subject matter experts having experience with the specific class of assets. Once the critical components of the specific asset is identified, the next step is to identify sensory information that will indicate or track the components’ degradation over time. With sensors, components, and assets identified, the next step is to select the method for the prognostics system. Data-Drive models use empirical sensory data to model normal operating conditions of the asset as well as know failure modes of the asset. Model based (physics of failure) prognostics, use physical models to represented normal or expected behavior of the asset. In many cases, it is desirable to use a hybrid method, leveraging empirical data for data driven models, and physics based models to represent normal or expected behavior. In the process of prognostics design, implementation and feedback using trials and pilot deployments. In this manner, the prognostics system can be tuned to best meet the business needs.

9

http://www.imscenter.net/

A common plant asset where prognostics can be used is the electric motor driven machine. Electric motor driven machines include pumps, fans, Heating and Ventilation or (HVAC) blowers and many other machines. This class of machines is the most commonly used machine in the world. Did you know there are more induction motors on planet earth than there are people? Electrical motor machines are often driven by induction motors. Their rotating shafts are supported by roller element bearings. Gearboxes provide speed changing mechanics to adapt the speed of the shaft for the driven component such as the pump or the fan. Fans and pumps have vanes and blades.

To test or monitor an electrical machine, we may use several sensors. These include accelerometers for vibration, tachometers or eddy current probes for speed, pressure and cavitations sensors for blade or vane monitoring, and even electrical power sensors to monitor the motor itself.

Software that analyzes the data will show Vibration levels at specific fault frequencies measured with a Sound and Vibration tailored FFT. Integration from acceleration to velocity is often used prior to data logging and trending. As with any machine monitoring or test application, data logging at specific times or events creates a series of data files for off-line analysis. With proper labeling of sensory data recordings, we can even conduct analysis across multiple files. In most cases, dynamic signal acquisition hardware with 24bit resolution, simultaneous sampling, power for sensors, and anti-aliasing filters are desired.

10

11

While the computer-based machinery monitoring system is capable of monitoring many sensors and signals, we will focus on vibration for the majority of this presentation. Let’s consider some vibration sources of a generic operating machine. Some vibration sources are due to machine design: slot frequency/EM related, gears, coupling, mechanical resonance, and bearings are some examples. Other sources result from potential problems: mechanical looseness, unbalance, and alignment. All these vibration sources combine to create a measurable vibration signal. The vibration sensor (accelerometer, proximity probe, velocity probe, etc.) produces a multi-tone signature where each tone represents the vibration from a specific mechanical source.

For example in rotating machinery there are several common failure modes which can be monitored for by vibration sensors and vibration signal processing.

12

Sensors, measure small changes in health and performance of a given asset

13

14

The Fourier Transform is a signal processing tool used to identify repetitions in a data set. Since a machine rotates, its vibration signals repeat each rotation. The Fourier Transform identifies these repetitive signals and allows the maintenance team to monitor the repetitive vibration from individual mechanical components such as bearings and gears. Using a carefully selected data acquisition system to acquire the data, the next step in machinery monitoring for vibration is to perform frequency analysis. This graphic shows the typical vibration frequency of several machine vibration phenomenon. While the machine being monitored or under test may be rotating at 30 to 200 hz, its bearing fault frequencies may be 10 or 20 times as high as rotational frequency. Similarly gear fault frequencies may be 20 to 50 times as high as rotational speed. The workhorse analysis tool for vibration monitoring is the Fast Fourier Transform, or FFT. This tool allows us to focus attention on particular frequency space to completely understand the vibration characteristics of the mechanical components, including amplitude, frequency, phase, harmonics, sidebands, and so forth.

15

To enlist predictive maintenance techniques, you need information from the machine. So here is a large industrial machine, this could be an air compressor creating the drive for pneumatics used to assembly aircraft. The typical sensor measurements performed on a machine consist of an accelerometer, proximity probe, tachometer, and other process variables. Combining machine vibration with operation parameters helps round out the sensory inputs for asset health management applications.

Just as a reference point, here is a motor drive pump with accelerometers mounted at each of the four bearing locations (two on the motor, and two on the pump).

16

Some machines are much more complex, as indicated by this diagram from Prognost, a supplier of reciprocating machine monitoring solutions.

17

By using a collection of wear indications, coupled with statistical / prognostics algorithms, the process of prognostics is born. By tracking the degradation of the asset, and comparing its degradation pattern to either modeled or historical degradation patterns, it becomes possible to predict both the time and the type of functional failure of the asset.

19

Prognostics is a component of the ISO 13374 Standard for Condition Monitoring Applications. Prognostics tools participate in State Detection, Health Assessment, Prognostics, and Advisory Generation. The end goal of a well defined and implemented prognostics application is automated health assessment, prognostics (how long to failure), and advisory generation (given risks, when should action be taken).

20

There are many common vibration sensory data calculations that are made, trended, and used for threshold alarming.

21

When we put together all of these data acquisition system components, we are able to create a Asset Monitoring Systems Architecture. The National Instruments Asset (Condition) Monitoring Architecture has five major components. 1) The Data Acquisition and Analysis Nodes (DAANs) which gather vibration and process data on

an event basis. Vibration data is analyzed in the node using order analysis to integrate to velocity from acceleration, and to calculate overall, 1x , and 2x amplitude and phase. Time Waveform data files with calculations and descriptive meta data are sent to the Plant Server computer

2) The plant server computer is a Windows 2008 server computer with enough drive space to store the data files, which are essentially the database for analysis. The data file format is TDMS, which is an open standard which can easily be loaded into excel. The Plant Server performs several background functions and often hosts two application functions.

1) The server polls each of the DAANs for new data every few minutes, and when new data is found, it retrieves the data files from the CompactRIO DAAN.

2) When new data arrives from a CompactRIO, the plant server publishes calculated condition indicating values as OPC (www.opcfoundation.org) tags. This is the interface to a plant historian.

3) When new data arrives, it is added to the data index, the NI DataFinder. 4) Optionally, data aging services reduce data to specific time slots per day, such as one

data file per shift or one data file per day. 3) The enterprise is a customer supplier plant historian. This historian subscribes to the NI OPC

server to receive vibration tag updates. 4) The Administration Client configures and commissions DAANs 5) The Visualization and Analysis Client provides off-line viewing of trends, time waveforms,

Spectrum with harmonic and sideband cursors, and Waterfall Plots

23

http://www.opcfoundation.org/

1) The Data Acquisition and Analysis Nodes (DAANs) which gather vibration and process data on an event basis. Vibration data is analyzed in the node using order analysis to integrate to velocity from acceleration, and to calculate overall, 1x , and 2x amplitude and phase. Time Waveform data files with calculations and descriptive meta data are sent to the Plant Server computer

BOP Systems • Accounts for 89-90% of measurements • Periodic acquisition once an hour

(default setting, can be configured) • Support for accelerometers, velocity, general purpose voltage, temperature, and 4-20mA sensors • 10kS/s default across all vibration channels • Time, RPM & force triggering Critical Equipment Systems • Periodic acquisition once every hour

(default setting, can be configured) • Burst logging/ streaming of waveform data during run up and cost down events

until systems reaches steady state • Support for accelerometers, velocity, and proximity probes (including

Keyphasor sensors) • Can utilize Buffered Outputs from Bently Nevada system • Time, RPM & force triggering • 10kS/s default across all vibration channels

• MIMOSA hierarchy • Store & Forward Architecture

27

CompactRIO systems are designed for extreme ruggedness, reliability, and I/O flexibility. With 50 g shock ratings and a wide -40 to 70 °C operating temperature on some models, CompactRIO systems and are ideal for automotive, industrial automation, and advanced control applications. CompactRIOs are also available with conformal coating options for an even higher level of ruggedness. CompactRIO also offers high levels of processing performance with up to a 1.33 GHz dual core Intel i7 processor at low power consumption levels, with a wealth of I/O availability. These flexible systems offer the processing power you need for advanced control applications, high-speed data transfer and logging, and processing-intensive applications such as rapid control prototyping and advanced motion. <build> Finally, LabVIEW allows you target the CompactRIO directly from the project, and allows you program the embedded processor, FPGA, and interface with the I/O all from the same environment. • CompactRIO leverages the LVRIO architecture, and offers high levels of ruggedness and

performance

29

In addition, there are a wide range of modules available for all types of sensory measurements.

30

31

The core input module used in the Balance of Plant and in the Turbine Monitoring system is the 9232 module.

To capture vibration and process sensory data, triggered by events, an on-line always on monitoring system is desired. An open and modular system allows for vibration and related sensor types to be digitized at the same time. With onboard intelligence and storage, the correct data recording settings can be achieved to yield the desired FMAX and Lines of Resolution. When wireless is required, there are many options for configuring a powerful wireless data acquisition system. Using the NI Value based CompactRIO chassis with specific I/O modules, AND by leveraging the previously mentioned reference design, it is quick and easy to deploy a ready to run application that is remotely configurable. The reference design also produces live data streams as well as captured TDMS data files. With CompactRIO, it is easy to use wired or wireless Ethernet for remote monitoring. Shown here are several radio Ethernet solutions National Instruments customers have used.

33

In many cases, an industrial data acquisition device can be placed very close to the rotating machinery. This helps reduce the cable runs, and lowers the installation costs. By incorporating wifi or similar radios, communications cabling is also avoided. In the panel shown here, a industrial CISCO radio transmits time waveform data files over 300 feet to the maintenance building.

34

National Instruments continues to invest in new measurements, both in hardware and software. Shown here is work being done in Electro Magnetic Signature Analysis for monitoring high frequency “noise” caused by electrical degradation of electrical power equipment. NI is also adapting its Camera input hardware for asset monitoring, including the use of thermal cameras. NI is also integrating motor current signature analysis (MCSA) for electric motor component degradation detection.

37

To be “intelligent about what is recorded and saved”, National Instruments (NI) has implemented an architecture we call “Store and Forward”. Store and Forward is a decision based data logger. It has several parts. Store and forward functionality is important to allow the embedded sensory data acquisition system to operate in the absence of network connectivity. Init: Creates file structure and then starts acquisition Check Trig: Monitors trigger conditions for waveform features or user input New File: Opens a new TDMS file with a unique name Acq & Save: Acquires data and saves data to TDMS file Close File: Closes the TDMS file Manage Drive: Looks at drive space and delete files if necessary.

40

Sensor data is measured (the green waveform on the left) and features are extracted from these time waveforms. These features form patterns of normal and faulty behavior, and often of specific faults under specific operating conditions.

41

It is also important to record the working condition of the asset. Sensory measurements will vary based on working conditions. This allows sensory measurements to be sorted into working condition “groups”

42

The key to finding data, is to label data as much as possible. The CompactRIO smart logging functions, label the data as it is recorded in order to compare like machines, machine components, and recordings from a similar time period.

43

• Once Data arrives on the Host Computer, a number of additional and more advanced algorithms can go to work.

• Sound & Vibration Suite, extensive development throughout the years of analysis algorithms

• Through the Tools Network, NI offers the Watchdog Agent® Prognostics Toolkit for LabVIEW, NI also offers the

• Systems Identification toolkit for asset model development, and the • Machine Learning toolkit for evaluation of large field sensory data sets.

44

There are many common vibration sensory data calculations that are made, trended, and used for threshold alarming.

45

For reference, here is a pictorial representation of the prognostics process with an automated diagnostics goal component.

47

To summarize the Logistic regression method, the Watchdog Agent® help files offer this diagram. By periodically exercising this flow diagram, it is possible to detect new anomalies while checking for specific faults. If fault-specific features are not a match to real-time monitoring data, yet an anomaly has been detected, it is likely that a new fault has been detected. It is then possible to collect the new fault-2-specific feature set, and set up another logistic regression to check for fault 2. This process can be repeated.

49

Behavior of an asset or asset component, is typically defined by several extracted features of condition indicators. This collection of condition indicators define a state or stage of degradation. With Data Driven models, it is possible to compare the current health state of the asset to a normal or health state using comparison algorithms such as statistical pattern recognition. Here, the degradation level or health of the asset is defined by the similarity measure with normal behavior as defined by the normal baseline signature.

50

The health radar chart is a method available to visualize the condition of multiple components within a single asset.

51

The Self Organizing Map is a useful tool for mapping an assets current condition to known faults.

52

One tool in the LabVIEW environment on which the Watchdog Agent® toolkit is built, is curve fitting. If we take the degradation or health curve produced by the SPR or MQE anomaly detection algorithms, it is possible to fit the curve using polynomial, spline, exponential, and other curve fitting tools to predict where the curve is trending. With specific thresholds set on health limits, it is possible to predict the number of time units left until a failure threshold is met.

53

Here is a summary of the modeling process for a single asset operating condition. A prognostics model is developed (Data Driven, Model Driven, or Hybrid) for each operating regime and failure mode. The prognostic system is trained using test and live sensory data arriving from data collection devices and control systems. Leveraging the Prognostics algorithms in the Watchdog Agent™ toolset for LabVIEW, www.ni.com/watchdogagent and NI’s platforms for data collection and signal processing, health assessment, fault diagnosis, and performance prediction are assisted and automated, lessoning the requirement for full time human intervention in data analysis. A variety of graphics are available to indicate the condition of the asset, with hierarchical drill downs to system level components. Together, National Instruments and you can develop and deploy cost effective and powerful condition monitoring systems leveraging traditional mathematical functions as well as prognostics capabilities.

54

Sentinel is an excellent case study of turn-around from build it yourself to buy into the packaged cRIO platform. It took specific handholding from NIC, including domain expertise, and we are able to close a growing OEM for vibration monitoring services in the arena of reciprocating compressors.

56

SM Instruments is deploying CompactRIO condition monitoring systems for Korea Electric Power. These systems are for on-shore and off-shore Wind Energy. SM Instruments recently has obtained GL certification for the Green-Eye brand condition monitoring systems based on CompactRIO.

57

The vision for Fleet Wide Asset Monitoring, is the use of a wide range of sensory data, with signal processing, to provide an automated diagnostics and prognostics system. To make the information available, Fleet Drill down dashboards will help asset owners and maintainers quickly navigate their fleet of assets, and identify developing maintenance needs. With machine learning and prognostics tools, high level operational and maintenance dashboards are possible.

58

For a specific asset, drill downs shows the individual health of mechanical components of the asset.

59

National Instruments Asset (Condition) Monitoring Architecture has five major components. 1) The Data Acquisition and Analysis Nodes (DAANs) which gather vibration and process

data on an event basis. Vibration data is analyzed in the node using order analysis to integrate to velocity from acceleration, and to calculate overall, 1x , and 2x amplitude and phase. Time Waveform data files with calculations and descriptive meta data are sent to the Plant Server computer

2) The plant server computer is a Windows 7 or Windows 2008 server computer with enough drive space to store the data files, which are essentially the database for analysis. The data file format is TDMS, which is an open standard which can easily be loaded into excel. The Plant Server performs several background functions and often hosts two application functions.

1) The server polls each of the DAANs for new and when new data is found, it retrieves the data files from the CompactRIO DAAN.

2) When new data arrives from a CompactRIO, the plant server publishes calculated values as OPC (www.opcfoundation.org) tags. This is the interface to a plant historian.

3) When new data arrives, it is added to the data index, the NI DataFinder. 4) Optionally, data aging services reduce data to specific time slots per day, such as

one data file per shift or one data file per day. 5) The Administration Client configures and commissions DAANs

3) The Visualization and Analysis Client provides off-line viewing of trends, time waveforms, Spectrum with harmonic and sideband cursors, and Waterfall Plots

4) The enterprise is a customer supplier plant historian. This historian subscribes to the NI OPC server to receive vibration tag updates.

61


1) The Data Acquisition and Analysis Nodes (DAANs) which gather vibration and process data on an event basis. Vibration data is analyzed in the node using order analysis to integrate to velocity from acceleration, and to calculate overall, 1x , and 2x amplitude and phase. Time Waveform data files with calculations and descriptive meta data are sent to the Plant Server computer

BOP Systems • Accounts for 89-90% of measurements • Periodic acquisition once an hour

(default setting, can be configured) • Support for accelerometers, velocity, general purpose voltage, temperature, and 4-20mA sensors • 10kS/s default across all vibration channels • Time, RPM & force triggering Critical Equipment Systems • Periodic acquisition once every hour

(default setting, can be configured) • Burst logging/ streaming of waveform data during run up and cost down

events until systems reaches steady state • Support for accelerometers, velocity, and proximity probes (including

Keyphasor sensors) • Can utilize Buffered Outputs from Bently Nevada system • Time, RPM & force triggering • 10kS/s default across all vibration channels

• MIMOSA hierarchy • Store & Forward Architecture

62

The plant server computer is a Windows 7 or Windows 2008 server computer or cluster of computers that acts as the central pillar for the solution, where a lot of the hard work happens. The computers should have enough drive space to store the data files, which are essentially the database for analysis. The data file format is TDMS, which is an open standard which can easily be loaded into excel. The Plant Server performs several background functions and often hosts two application functions.

1) The server polls each of the DAANs for new data every two minutes, and when new data is found, it retrieves the data files from the CompactRIO DAAN.


3) When new data arrives, it is added to the data index, the NI DataFinder. 4) Optionally, data aging services reduce data to specific time slots per day, such as one data file per shift or one data file

per day. 1) The enterprise is a customer supplier plant historian. This historian subscribes to the NI OPC server to receive vibration tag

updates. 2) The Administration Client configures and commissions DAANs

3) Web-based configuration and management client

(Dashboards for

Server Features: Systems Management System health & status monitoring of CompactRIO systems Deployment image management for managing larger systems User and device authentication

Data Analytics Vibration-specific analytics

Data Management Data aging and storage rules Alarming Integration into 3rd party historians (OSIsoft PI Server, OPC UA) • Functionality is extensible using SDK

• Web-based thin client • to remotely configure and commission CompactRIO systems • View dashboard of active alarms • Configure new alarm rules

63








one data file per shift or one data file per day. 3) The enterprise is a customer supplier plant historian. This historian subscribes to the NI

OPC server to receive vibration tag updates. 4) The Administration Client configures and commissions DAANs 5) The Visualization and Analysis Client provides off-line viewing of trends, time

waveforms, Spectrum with harmonic and sideband cursors, and Waterfall Plots

64








one data file per shift or one data file per day. 3) The enterprise is a customer supplier plant historian. This historian subscribes to the NI

OPC server to receive vibration tag updates. 4) The Administration Client configures and commissions DAANs 5) The Visualization and Analysis Client provides off-line viewing of trends, time

waveforms, Spectrum with harmonic and sideband cursors, and Waterfall Plots

65


To help summarize, this graphic describes the fleet wide monitoring efforts now underway at several power generation plants in the Americas. Many data acquisition and analysis nodes are being deployed to initially capture vibration and temperature time waveforms based on time or speed events. Often, a business wireless tcp/ip network is used for data transfer to a business server in the PdM work area at the plant. This business server is accessible from other computers on the network for configuration and data analysis. This overall architecture is reflective of the Big Analog Data™ solution I showed earlier. As we add more sensors, we have more analog data to work with. Yet intelligent analytics can reduce the data while “in-motion” to those features which indicate condition and degradation rates. With modern server technology, we have the resources at hand to fuse time waveform data and condition indicators with operational and maintenance notes to create the systems envisioned by EPRI, and its power generation members. With the right data, the right analytics, automation of diagnostics and prognostics yields advisory generation of the future.

66

Instep Software LLC, provides an advanced pattern recognition tool for the power generation industry. PRiSM™ uses condition indicators and operational data stored in the plant historian to build data driven models of normal operational behavior under each of the operating conditions the asset experiences. During the on-line monitoring process, the condition indicators are compared to expected values of the models to provide alerts when parameters deviate from expected normal.

68

arabia.ni.comarabia.ni.com/sites/default/files/predictive maintenance vibration... · 7 as we look...

Documents