Predictive Maintenance and Condition Monitoring Tim Yeh Application Engineer 2018 The MathWorks, Inc. 1
Agenda What is Predictive Maintenance? Algorithm behind Remaining Useful Life (RUL) Estimation Methods Summary & Q/A 2
What is Predictive Maintenance? 3
Pump - detected I need help. One of my cylinders is blocked. I will shut down your line in 15 hours. 4
What do you expect from predictive maintenance? Maintenance cares about day-to-day operations Reduced downtime Operations & IT look at the bigger picture Improved operating efficiency Engineering groups get product feedback Better customer experience Upper management wants to drive growth New revenue streams Source: Tensor Systems 5
Types of Maintenance Reactive Do maintenance once there s a problem Scheduled Do maintenance at a regular rate Predictive Forecast when problems will arise Example Problem 6
What does a Predictive Maintenance algorithm do? Helps make maintenance decisions based on large volumes of complex data Data What is the condition of my machine? When will my machine fail? Condition Monitoring Predictive Maintenance Condition Monitoring Process of monitoring sensor data from machines (vibration, temperature etc.) in order to identify significant changes which can indicate developing faults Predictive Maintenance Technique that determines time-to-failure/remaining useful life (RUL) from sensor data & historical data in order to predict when maintenance should be performed 7
Machine Learning 8
Machine Learning Workflow Train: Iterate till you find the best model LOAD DATA PREPROCESS DATA SUPERVISED LEARNING MODEL FILTERS PCA CLASSIFICATION SUMMARY STATISTICS CLUSTER ANALYSIS REGRESSION Predict: Integrate trained models into applications NEW DATA PREDICTION 9
Example: Human Activity Learning Using Mobile Phone Data Machine Learning Data: 3-axial Accelerometer data 3-axial Gyroscope data 10
Deep Learning 11
Machine Learning or Deep Learning? Machine Learning Approach Sensor 1 Sensor 2 Sensor 25 Correlation Analysis Sensor a Sensor b Sensor c Feature Extraction Learning 1. Normal 2. Monitor 3. Maintain Output Sensor 1 Sensor 2 Sensor 25 Deep Learning Approach Feature Extraction & Learning 1. Normal 2. Monitor 3. Maintain Output 12
Object recognition using deep learning Training (GPU) Prediction Millions of images from 1000 different categories Real-time object recognition using a webcam connected to a laptop 13
Detection and localization using deep learning Regions with Convolutional Neural Network Features (R-CNN) Semantic Segmentation using SegNet 14
Deep Learning for Signals (Non Image) Convolutional Neural Networks (CNN) Convert the time series into a spectrogram/scalogram and use it as an input to a CNN. Long Short-Term Memory (LSTM) Neural networks which connect through time. 15
Deep Learning Enablers Increased GPU acceleration World-class models AlexNet PRETRAINED MODEL Caffe M O D E L S VGG-16 PRETRAINED MODEL GoogLeNet PRETRAINED MODEL ResNet PRETRAINED MODEL TensorFlow/Keras M O D E L S Labeled public datasets 16
Predictive Maintenance Customer Examples Pump Health Monitoring System Spectral analysis and filtering on binary sensor data and neural network model prediction More than $10 million projected savings Online engine health monitoring Real-time analytics integrated with enterprise service systems Predict sub-system performance (oil, fuel, liftoff, mechanical health, controls Production machinery failure warning Reduce waste and machine downtime MATLAB based HMI warns operators of potential failures > 200,000 savings per year 17
Common Challenges How do I get started with developing algorithms? Reference examples Workflow-based documentation How do I manage data and what if I don t have any data? Command line functions to organize data Examples showing Simulink models generating failure data How do I choose condition indicators and estimate the RUL? Functions for computing condition indicators Functions provided for estimating RUL 18
Condition Indicator Design Extract features using signal-based and model-based methods to determine machine health Capture time-varying dynamics (e.g. vibration data) by computing time-frequency moments Detect sudden changes in nonlinear systems using phase-space reconstruction methods Other methods available include statistical, spectral analysis, time-series modeling, and modal analysis 19
Remaining Useful Life (RUL) Estimation Methods Requirement: Need to know what constitutes failure data Lifetime data, no sensor data Use Survival methods Known failure threshold + some data Use Degradation methods Run-to-failure data Use Similarity methods 21
RUL Methods and when to use them Requirement: Need to know what constitutes failure data Details on model selection in the documentation 22
Predictive Maintenance of Turbofan Engine Sensor data from 200 engines of the same model Forecast Remaining Useful Life Have some combination of: Sensor data from the equipment Known failure thresholds Runtime to failure of similar equipment Use condition monitoring to determine when system starts degrading Can we predict how much longer we can use our equipment once degradation begins? Data provided by NASA PCoE https://ti.arc.nasa.gov/tech/dash/groups/pcoe/prognostic-data-repository 23
How does it looks like? Condition Monitoring Predictive Maintenance Decision 24
Failure Data Generation from Simulink Use simulation data that is representative of actual machine failures to train your algorithm Construct a model of your machine using Simulink and Simscape that captures its different failure modes 1. Build model Simulink Model 4. Run simulations Run multiple simulations of your model under different fault conditions Manage and label this simulated data using ensemble objects provided by the Predictive Maintenance Toolbox Sensor Data 2. Fine tune model Inject Failures Generated Failure Data 3. Incorporate failure modes 25
Key Takeaways Frequent maintenance and unexpected failures are a large cost in many industries MATLAB enables engineers and data scientists to quickly create, test and implement predictive maintenance programs Predictive maintenance Saves money for equipment operators Increases reliability and safety of equipment Creates opportunities for new services that equipment manufacturers can provide 26
Thank you! 27