Mathworks CAE simulation suite – case in point from automotive and aerospace.
1 like78 views
This document discusses predictive maintenance and how to develop predictive maintenance algorithms using MATLAB. It defines predictive, preventative, and reactive maintenance. It then outlines the steps to develop a predictive algorithm, including acquiring sensor data, preprocessing the data, identifying condition indicators, training a model, and deploying the model. It provides examples of developing algorithms for fault classification and remaining useful life estimation using sensor data from a triplex pump.
Mathworks CAE simulation suite – case in point from automotive and aerospace.
- 1. 1© 2020 The MathWorks, Inc. Predictive Maintenance with MATLAB: Case in Point from Industry Dr Bulat Khusainov Education Technical Specialist
- 2. 2 Agenda: 1. What is Predictive Maintenance? Who is benefiting by doing it? 2. How can you develop a predictive maintenance algorithm using MATLAB? 3. How can you get started quickly?
- 3. 3 Predictive maintenance Preventive maintenance Reactive maintenance Types of maintenance
- 6. 6 Time Machine health Reactive maintenance Preventive maintenance Time Machine health x Failure ? ?
- 7. 7 Time Machine health Reactive maintenance x Failure Preventive maintenance Time Machine health x Still usable condition
- 8. 8 Time Machine health Reactive maintenance x Failure Preventive maintenance Time Machine health x Still usable condition Predictive maintenance Time Machine health x now Predicted failure Optimum time to do maintenance
- 9. 9 Condition Monitoring Remaining Useful Life Estimation A Predictive Maintenance Algorithm Answers These Questions Why is my machine behaving abnormally? How much longer can I operate my machine ? Anomaly Detection Is my machine operating normally? For example: I need help. One of my cylinders is blocked. I will shut down your line in 15 hours.
- 10. 10 Predictive Maintenance Success Stories Online engine health monitoring ▪ Real-time analytics integrated with enterprise service systems ▪ Predict sub-system performance (oil, fuel, liftoff, mechanical health, controls Pump Health Monitoring System ▪ Spectral analysis and filtering on binary sensor data and neural network model prediction ▪ More than $10 million projected savings Production machinery failure warning ▪ Reduce waste and machine downtime ▪ MATLAB based HMI warns operators of potential failures ▪ > 200,000 € savings per year
- 11. 11 Link to user story
- 12. 12 Transocean uses MATLAB tools to transition from preventative maintenance to CPM* for a critical deep sea drilling component *Condition and Performance Monitoring Challenges • Minimize unplanned downtime of the component • Use as-close-to-real-time data for CPM • Deploy CPM solution to components in other locations Solution ▪ Monitoring data from the drill is collected in a PI archive. MATLAB is used to create an app that takes in the data, preprocesses the data, and generates a quality indicator. ▪ The app is deployed onto MATLAB Production Server for real-time use; results are stored back on the PI archive. Results Able to move to CPM for their component(s) with improved decision-making capabilities through faster access to data and quicker analytics deployment. Interfacing directly with data in PI gives you increased performance at the cost of having to write/deal with low-level code (“getting under the hood”). It was noted that using MATLAB parallel tools (which entails a higher-level interface) was not only easier but also it provided “hands-down” faster performance than interfacing with PI directly. Link to Expo Talk Drilling data is stored in an OSISoft PI Archive. MATLAB Production Server is used in the application specific processing
- 13. 13 Condition Monitoring Remaining Useful Life Estimation Predictive Maintenance Toolbox for Developing Algorithms Why is my machine behaving abnormally? How much longer can I operate my machine ? Anomaly Detection Is my machine operating normally?
- 14. 14 Workflow for Developing a Predictive Maintenance Algorithm Acquire Data Preprocess Data Identify Features Train Model Deploy & Integrate Machine Learning
- 15. 15 Agenda: 1. What is Predictive Maintenance? Who is benefiting by doing it? 2. How can you develop a predictive maintenance algorithm using MATLAB? 3. How can you get started quickly?
- 16. 16 Condition Monitoring Remaining Useful Life Estimation Develop Predictive Maintenance Algorithm: Use cases Why is my machine behaving abnormally? How much longer can I operate my machine ? Anomaly Detection Is my machine operating normally?
- 17. 17 Condition Monitoring Develop Predictive Maintenance Algorithm for - Why is my machine behaving abnormally? Predictive maintenance algorithm Identify fault type Determine what needs to be fixed Condition indicators Use case: Fault Classification
- 18. 19
- 19. 20 Time Sensor data Predictive Maintenance Algorithm Preprocess Data Identify Condition Indicators Train Model Deploy & Integrate Triplex pump Temperature Pressure Flow Collect data using sensorshealthy + faulty *For simplification purposes, healthy and faulty operation are represented by single measurements. In a realistic scenario, there may be hundreds of measurements for both types of operations. Acquire Data Sensor Data Generated Data
- 20. 21 Acquire Data Predictive Maintenance Algorithm Preprocess Data Identify Condition Indicators Train Model Deploy & Integrate Triplex pump Temperature Pressure Flow Collect data using sensorshealthy + faulty Time Sensor data *Quality data->Robust Algorithms
- 21. 22 Acquire Data Predictive Maintenance Algorithm Preprocess Data Identify Condition Indicators Train Model Deploy & Integrate Triplex pump Temperature Pressure Flow Collect data using sensorshealthy + faulty Time Sensor data
- 22. 23 Time Sensor data Acquire Data Predictive Maintenance Algorithm Preprocess Data Identify Condition Indicators Train Model Deploy & Integrate Triplex pump Temperature Pressure Flow Collect data using sensorshealthy + faulty Operating temperature: 30°F Fluid viscosity: Low Operating temperature: 90°F Fluid viscosity: High
- 23. 24 Acquire Data Predictive Maintenance Algorithm Preprocess Data Identify Condition Indicators Train Model Deploy & Integrate Triplex pump Time Sensor data What if real failure data is not available?
- 24. 25 Time Sensor data Acquire Data Predictive Maintenance Algorithm Preprocess Data Identify Condition Indicators Train Model Deploy & Integrate Triplex pump Time Sensor data Inject faults1Seal leakage 2Blocked inlet 3Worn bearing Mathematical model of the triplex pump (Digital Twin) Refine model 1 2 3 Time Synthetic data
- 25. 26 240 measurements of flow and pressure with a duration of 1.2 seconds Flow Pressure Health condition (fault code) • Healthy (0) • Blocked inlet (010) • Worn bearing (001) • Seal Leakage (100) • Blocked inlet, worn bearing (011) • Seal leakage, worn bearing (101) • Seal leakage, blocked inlet (110) • Seal leakage, blocked inlet, worn be Learn further: Data Ensembles
- 26. 27 Predictive Maintenance Algorithm Preprocess Data Identify Condition Indicators Train Model Deploy & Integrate Acquire Data Sensor Data Generated Data Challenges ▪ Data clean up – Poorly formatted files – Irregularly sampled data – Redundant data, outliers, missing data etc. ▪ Data specific processing – Signals: Smoothing, resampling, denoising, Wavelet transforms, etc. – Images: Image registration, morphological filtering, deblurring, etc.
- 27. 28 0.5 1 1.5 2 2.5 3 3.5 4 Time (s) 6.5 7 7.5 8 8.5 9 9.5 10 Raw data Pressure(bar) Spikes to sensor’s maximum value Offset in time
- 28. 29
- 29. 30 0.5 1 1.5 2 2.5 3 3.5 4 Time (s) 6.5 7 7.5 8 8.5 9 9.5 10 Raw data Pressure(bar) Time (s) Preprocessed data Pressure(bar) 0 0.2 0.4 0.6 0.8 1 1.2 6.95 7 7.05 7.1 7.15 7.2 7.25 7.3 7.35
- 30. 31 Predictive Maintenance Algorithm Preprocess Data Identify Condition Indicators Train Model Deploy & Integrate Acquire Data Sensor Data Generated Data • What are Condition Indicators? • Why are they important? • How to identify significant ones?
- 31. 32 Identify condition indicators Time Condition indicator Time Raw data Healthy operation Faulty operation Healthy operation Faulty operation A condition indicator can be any feature that is useful - • for distinguishing normal from faulty operation or • for predicting remaining useful life
- 32. 33 Signal-Based Condition Indicators Time-domain features Frequency-domain features Time-frequency domain features Mean Standard deviation Skewness Root-mean square ... Kurtosis
- 33. 34 0.62 0.63 0.64 0.65 0.66 0.67 0.68 0.69 0.7 0.71 0.72 7 7.05 7.1 7.15 7.2 7.25 7.3 0.62 0.63 0.64 0.65 0.66 0.67 0.68 0.69 0.7 0.71 0.72 7 7.05 7.1 7.15 7.2 7.25 7.3 0.62 0.63 0.64 0.65 0.66 0.67 0.68 0.69 0.7 0.71 0.72 7 7.05 7.1 7.15 7.2 7.25 7.3 0.62 0.63 0.64 0.65 0.66 0.67 0.68 0.69 0.7 0.71 0.72 7 7.05 7.1 7.15 7.2 7.25 7.3 0.62 0.63 0.64 0.65 0.66 0.67 0.68 0.69 0.7 0.71 0.72 7 7.05 7.1 7.15 7.2 7.25 7.3 0.62 0.63 0.64 0.65 0.66 0.67 0.68 0.69 0.7 0.71 0.72 7 7.05 7.1 7.15 7.2 7.25 7.3 Blocked Inlet Blocked Inlet, Worn Bearing Seal Leakage Seal Leakage, Blocked Inlet Seal Leakage, Worn Bearing Worn Bearing Pressure(bar)Pressure(bar)Pressure(bar) Pressure(bar)Pressure(bar)Pressure(bar)Time (s) Time (s) 0.62 0.63 0.64 0.65 0.66 0.67 0.68 0.69 0.7 0.71 0.72 7 7.05 7.1 7.15 7.2 7.25 7.3 0.62 0.63 0.64 0.65 0.66 0.67 0.68 0.69 0.7 0.71 0.72 7 7.05 7.1 7.15 7.2 7.25 7.3 0.62 0.63 0.64 0.65 0.66 0.67 0.68 0.69 0.7 0.71 0.72 7 7.05 7.1 7.15 7.2 7.25 7.3 0.62 0.63 0.64 0.65 0.66 0.67 0.68 0.69 0.7 0.71 0.72 7 7.05 7.1 7.15 7.2 7.25 7.3 0.62 0.63 0.64 0.65 0.66 0.67 0.68 0.69 0.7 0.71 0.72 7 7.05 7.1 7.15 7.2 7.25 7.3 0.62 0.63 0.64 0.65 0.66 0.67 0.68 0.69 0.7 0.71 0.72 7 7.05 7.1 7.15 7.2 7.25 7.3 Blocked Inlet Blocked Inlet, Worn Bearing Seal Leakage Seal Leakage, Blocked Inlet Seal Leakage, Worn Bearing Worn Bearing Pressure(bar)Pressure(bar)Pressure(bar) Pressure(bar)Pressure(bar)Pressure(bar)Time (s) Time (s) Healthy Faulty Increasing fault severity
- 34. 35 0.62 0.63 0.64 0.65 0.66 0.67 0.68 0.69 0.7 0.71 0.72 7 7.05 7.1 7.15 7.2 7.25 7.3 Blocked Inlet Pressure(bar) 0.62 0.63 0.64 0.65 0.66 0.67 0.68 0.69 0.7 0.71 0.72 7 7.05 7.1 7.15 7.2 7.25 7.3 Blocked Inlet Pressure(bar) Blocked Inlet Healthy 7.225 7.23 7.235 7.24 7.245 7.25 MEAN No overlap Median 25th percentile 75th percentile Minimum Maximum
- 35. 36 7.21 7.215 7.22 7.225 7.23 7.235 7.24 7.245 7.25 MEAN Healthy Blocked Inlet Seal Leakage Worn Bearing Blocked Inlet & Worn Bearing Seal Leakage & Worn Bearing Seal Leakage & Blocked Inlet Fault type Overlap
- 36. 37 MEAN VARIANCE KURTOSIS Healthy Blocked Inlet Seal Leakage Worn Bearing Blocked Inlet & Worn Bearing Seal Leakage & Worn Bearing Seal Leakage & Blocked Inlet 7.21 7.215 7.22 7.225 7.23 7.235 7.24 7.245 7.25 SKEWNESS -2 -1.8 -1.6 -1.4 -1.2 -1 -0.8 -0.6 -0.4 -0.2 0 1 2 3 4 5 6 10-3 2 3 4 5 6 7 Fault type Fault type Fault type Fault type Outlier Outlier
- 37. 38 Healthy Blocked Inlet Seal Leakage Worn Bearing Blocked Inlet, Worn Bearing Seal Leakage, Worn Bearing Seal Leakage, Blocked Inlet Mean Variance
- 38. 40 Signal-Based Condition Indicators Time-domain features Frequency-domain features Time-frequency domain features Mean Standard deviation Skewness Root-mean square ... Kurtosis Learn more about Condition Indicators
- 39. 41 Three different vibration sources: - Bearing - Motor shaft - Disc • In time-domain, we observe the combined effect of different sources of vibration. DiscBearing Motor shaft Disc Bearing Motor shaft Machine with rotating components Time Amplitude Amplitude Frequency • Using frequency-domain analysis, we can distinguish different sources of vibration. Nominal Faulty
- 40. 42 Frequency-domain features: • Peaks • Peak frequencies 0 500 1000 1500 2000 2500 3000 3500 4000 4500 5000 0 0.005 0.01 0.015 0.02 Blocked Inlet 0 500 1000 1500 2000 2500 3000 3500 4000 4500 5000 Frequency (Cycles/min) 0 0.005 0.01 0.015 0.02 Blocked Inlet, Worn Bearing 0 500 1000 1500 2000 2500 3000 3500 4000 4500 5000 0 0.005 0.01 0.015 0.02 Seal Leakage 0 500 1000 1500 2000 2500 3000 3500 4000 4500 5000 0 0.005 0.01 0.015 0.02 Seal Leakage, Blocked Inlet 0 500 1000 1500 2000 2500 3000 3500 4000 4500 5000 0 0.005 0.01 0.015 0.02 Seal Leakage, Worn Bearing 0 500 1000 1500 2000 2500 3000 3500 4000 4500 5000 Frequency (Cycles/min) 0 0.005 0.01 0.015 0.02 Worn Bearing MagnitudeMagnitudeMagnitude
- 42. 45 Signal-Based Condition Indicators Time-domain features Frequency-domain features Time-frequency domain features Power bandwidth Peak values Peak frequencies Mean frequency Spectral entropy Spectral kurtosis Harmonics ... ... Mean Standard deviation Skewness Root-mean square ... Kurtosis Learn more about Condition Indicators
- 43. 46
- 44. 47 Diagnostic Feature Designer App Predictive Maintenance Toolbox R2019a ▪ Extract, visualize, and rank features from sensor data ▪ Use both statistical and dynamic modeling methods ▪ Work with out-of-memory data ▪ Explore and discover techniques without writing MATLAB code
- 45. 48
- 46. 49 Predictive Maintenance Algorithm Preprocess Data Identify Condition Indicators Train Model Deploy & Integrate Acquire Data Sensor Data Generated Data
- 47. 50 MODEL PREDICTION Machine Learning Workflow Train: Iterate till you find the best model using historical data Predict: Integrate trained models into applications MODELSUPERVISED LEARNING CLASSIFICATION REGRESSION PREPROCESS DATA SUMMARY STATISTICS PCAFILTERS CLUSTER ANALYSIS HISTORICAL DATA PREPROCESS DATA SUMMARY STATISTICS PCAFILTERS CLUSTER ANALYSIS NEW DATA Machine learning uses data and produces a program to perform a task
- 48. 51 Machine Learning Supervised Learning Classification Regression Unsupervised Learning Group and interpret data based only on input data Develop predictive model based on both input and output data Type of Learning Categories of Algorithms Condition Indicator Machine Learning: Types Regression
- 49. 52
- 50. 53 Video showing App in action
- 51. 54 Summary: Develop Predictive Maintenance Algorithm: Classification Learner App Diagnostic Feature Designer
- 52. 63 Acquire Data Predictive Maintenance Algorithm Preprocess Data Identify Condition Indicators Train Model Deploy & Integrate Edge device Cloud
- 53. 64 Deploy & Integrate analytics using MATLAB: MATLAB Runtime C, C++ HDL PLC Embedded Hardware C/C++ ++ Excel Add-in Java Hadoop/ Spark .NET MATLAB Production Server Standalone Application Enterprise Systems Python
- 54. 65 Feature Extraction Algorithm at the Edge Asset w. smart sensors ▪ 1 day ~ 1.3 GB ▪ 20 sensors/pump ~26 GB/day ▪ 3 pumps ~ 78 GB/day Pump flow sensor 1 sec ~ 1000 samples ~16kB Challenge: Data transmission cost is pretty high Solution: Extract only relevant information and send it to predictive model
- 55. 66 Video showing Codegen with MATLAB Coder
- 56. 67 Video showing Codegen with MATLAB Coder
- 57. 68 Code Deployment for Machine Learning ▪ SVM Classification ▪ Linear Classification ▪ Linear Regression ▪ Generalized Linear Regr. ▪ Decision trees ▪ Ensembles for Class. ▪ Ensembles for Regr. ▪ SVM Regression ▪ KNN Classification ▪ Gaussian Process Regr. ▪ Discriminant Analysis ▪ Non-tree Ensembles ▪ KNN with kd-tree Deploy trained models as standalone C/C++ code
- 58. 69 What do your end users want? Flexible Deployment ▪ Maintenance needs simple, quick information – Hand held devices, Alarms ▪ Operations needs a birds-eye view – Integration with IT & OT systems ▪ Customers expect easy to digest information – Automated reports Dashboards & Hand held Devices Fleet & Inventory Analysis Azure Blob Azure SQL Data Sources Analytics Platforms Azure IoT Hub AWS Kinesis Streaming Data
- 59. 70 Agenda: 1. What is Predictive Maintenance? Who is benefiting by doing it? 2. How can you develop a predictive maintenance algorithm using MATLAB? 3. How can you get started quickly?
- 60. 71 MathWorks can help you get started TODAY ▪ Examples ▪ Documentation ▪ Tutorials & Workshops ▪ Consulting ▪ Tech Talk Series
- 61. 72