Predictive Maintenance with MATLAB

© 2021 The MathWorks, Inc.

1
Agenda

▪ Predictive Maintenance Concepts

▪ MATLAB Examples
– Anomaly Detection
– Condition Monitoring
– Remaining Useful Life Estimation

▪ Useful Resources

▪ Q&A
2
Why Perform Predictive Maintenance?

▪ Example: faulty braking system leads to

wind turbine disaster
https://youtu.be/7nSB1SdVHqQ

▪ Wind turbines cost millions of dollars

▪ Failures can be dangerous

▪ Maintenance also very expensive and

dangerous

3
Types of Maintenance

▪ Reactive – Do maintenance once there’s a

problem
– Problem: unexpected failures can be expensive and
potentially dangerous

▪ Scheduled – Do maintenance at a regular rate

– Problem: unnecessary maintenance can be wasteful;
may not eliminate all failures

▪ Predictive – Forecast when problems will arise

– Problem: difficult to make accurate forecasts for
complex equipment

5
What Does a Predictive Maintenance Algorithm Do?
Helps make maintenance decisions based on large volumes of complex data

Is my machine
Anomaly
operating
Detection
normally?

Why is my
Condition
machine behaving
Data abnormally?
Monitoring Decision

How much longer Remaining

can I operate my Useful Life
machine? Estimation

6
Predictive Maintenance Algorithm Development Workflow

Machine Learning

Identify
Acquire Preprocess Condition Train Deploy &
Data Data Indicators Model Integrate

7
Triplex Reciprocating Pump

▪ Three-phase pump commonly used

for drilling and servicing oil wells Component
– Three plungers try to ensure a uniform Failure
flow

Outlet
▪ Condition monitoring to detect:
– Seal leak
8 fault codes
– Inlet blockage
(1, 10, 111, etc.)
– Bearing degradation Crankshaft

Inlet

8
Triplex Reciprocating Pump

▪ Three-phase pump commonly used

Pressure & Flow
for drilling and servicing oil wells Component Sensor
– Three plungers try to ensure a uniform Failure
flow

Outlet
▪ Condition monitoring to detect:
– Seal leak
8 fault codes
– Inlet blockage
(1, 10, 111, etc.)
– Bearing degradation Crankshaft

Inlet
▪ Analyze fault present in system using
only pressure and flow sensor data
9
 Baker Hughes Develops Predictive Maintenance Software
for Gas and Oil Extraction Equipment Using Data Analytics
and Machine Learning
Challenge
Develop a predictive maintenance system to reduce
pump equipment costs and downtime
Solution
Use MATLAB to analyze nearly one terabyte of data
and create a neural network that can predict machine Truck with positive displacement pump.
failures before they occur
Results “MATLAB gave us the ability to convert previously unreadable
▪ Savings of more than $10 million projected data into a usable format; automate filtering, spectral analysis,
and transform steps for multiple trucks and regions; and
▪ Development time reduced tenfold
ultimately, apply machine learning techniques in real time to
▪ Multiple types of data easily accessed
predict the ideal time to perform maintenance.”
- Gulshan Singh, Baker Hughes

Link to user story

10
Predictive Maintenance Algorithm Development Workflow

Is my machine Anomaly Component

operating normally? Detection Failure?

Why is my machine
Condition
Data behaving
Monitoring
Decision
abnormally?

How much longer Remaining

can I operate my Useful Life
machine? Estimation

Identify
Acquire Preprocess Condition Train Deploy &
Data Data Indicators Model Integrate

11
Preprocess Data

Data does not fit in memory?

▪ Scale with Datastore and Tall array

12
Predictive Maintenance Algorithm Development Workflow

Is my machine Anomaly Component

operating normally? Detection Failure?

Why is my machine
Condition
Data behaving
Monitoring
Decision
abnormally?

How much longer Remaining

can I operate my Useful Life
machine? Estimation

Identify
Acquire Preprocess Condition Train Deploy &
Data Data Indicators Model Integrate

13
Identify Condition Indicators

▪ Signal statistics
– Mean
– Signal-to-noise ratio
– Peak frequency
– … Condition Indicators/
Health Indicators/
▪ Data fusion and Features
transformation
– Principal component
analysis
– Regression
– Normalization
– …

14
Predictive Maintenance Algorithm Development Workflow

Is my machine Anomaly Component

operating normally? Detection Failure?

Why is my machine
Condition
Data behaving
Monitoring
Decision
abnormally?

How much longer Remaining

can I operate my Useful Life
machine? Estimation

Identify
Acquire Preprocess Condition Train Deploy &
Data Data Indicators Model Integrate

15
Scenario 1: Fault Classification

Is my machine Anomaly Component

operating normally? Detection Failure?

Why is my machine
Condition
Data behaving
Monitoring
Decision
abnormally?

How much longer Remaining

can I operate my Useful Life
machine? Estimation

Identify
Acquire Preprocess Condition Train Deploy &
Data Data Indicators Model Integrate

16
Scenario 1: Fault Classification

17
Scenario 2: Forecast Remaining Useful Life

Is my machine Anomaly When Will

operating normally? Detection Component
Fail?
Why is my machine
Condition
Data behaving
Monitoring
Decision
abnormally?

How much longer Remaining

can I operate my Useful Life
machine? Estimation

Identify
Acquire Preprocess Condition Train Deploy &
Data Data Indicators Model Integrate

18
RUL Methods and when to use them
Requirement: Need to know what constitutes failure data

System Data RUL Estimator Models

H F Similarity models

Safety
threshold

H F Degradation models

H F Survival models
H: Healthy state
F: Failure Details on model selection in the documentation 19
RUL Methods and when to use them
Requirement: Need to know what constitutes failure data

Details on model selection in the documentation

20
System Data for Degradation Model
Time 0
Health Indicator
Time 1
Health Indicator
Time 2
Health Indicator
Signal statistics +
Principal component
analysis

Safety threshold

Time N: Failure signals Health Indicator

21
Scenario 2: Forecast Remaining Useful Life

22
What if we also have domain experts of the equipment?
Leverage the engineering knowledge to enhance predictive maintenance

▪ Generate failure data from Simulink Build model

models Run
simulations
Simulink Model

▪ Identify root cause of fault via Sensor Data

Generated
Failure Data
parameter estimation
Fine tune
model
▪ Verify predictive maintenance
algorithm in new scenarios with a
Inject Failures
digital twin
Incorporate
failure modes

23
Predictive Maintenance Algorithm Development Workflow

Acquire
Data
Generate
d Data
Identify
Sensor Preprocess Condition Train Deploy &
Data Data Indicators Model Integrate

25
Scenario 3: Simulink for Fault Generation

26
What do your end users expect?

▪ Maintenance needs simple, quick AWS

Kinesis
information Azure
– Hand held devices, Alarms IoT Hub

Azure Azure Streaming Data

Blob SQL

▪ Operations needs a birds-eye view

– Integration with IT & OT systems

▪ Customers expect easy to digest Data Sources Analytics Platforms

Dashboards
Fleet & &
Inventory Analysis
information
Hand held Devices
– Automated reports

27
How do you deliver your analytics to different users?

Embedded Hardware Enterprise Systems

MATLAB
Standalone Web Excel Hadoop/ Production
C, C++ HDL PLC CUDA App Add-in C/C++ Java ++ Python .NET
Application Spark Server

MATLAB
Runtime
28
Predictive Maintenance with MATLAB and Simulink

Is my machine Anomaly
operating normally? Detection

Why is my machine
Condition
Data behaving
Monitoring
Decision
abnormally?

How much longer Remaining

can I operate my Useful Life
machine? Estimation

Identify
Acquire Preprocess Condition Train Deploy &
Data Data Indicators Model Integrate

MATLAB
Runtime

30
How can MATLAB and Simulink help you with predictive
maintenance?
“MATLAB gave us the ability to convert
▪ End-to-end solution
previously unreadable data into a usable
– Access and preprocess data
format; automate filtering, spectral analysis,
– Develop predictive models interactively
and transform steps for multiple trucks and
– Deploy models in production
regions; and ultimately, apply machine
▪ Modeling with Simulink
learning techniques in real time to predict
– Generate training data in the absence of
real failure data the ideal time to perform maintenance.”
– Verify algorithm in new scenarios with a
digital twin Gulshan Singh, Baker Hughes

▪ Get started quickly – we can help

– Reference examples
– MATLAB Tech Talks

31
MathWorks Engineering Support

Training Guided Evaluations

Onsite Workshops Consulting

Technical Support
32
Training Course: Predictive Maintenance with MATLAB

Topics included in this 2-day course:

▪ Importing and organizing data
▪ Creating custom visualizations
▪ Fault Detection/Classification
▪ Preprocessing to improve data quality, and
extract time and frequency domain features
▪ Estimating Remaining Useful Life (RUL)
▪ Interactive workflows with apps

See detailed course outline

33
 Achieve Results Faster with Predictive Maintenance Consulting
Our expert consultants can help you with the entire predictive maintenance workflow:
Data Preprocessing, Exploratory Analysis, Predictive Modeling, and Operational Deployment

Transparent Customized Return on

Approach Engagements Investment
You will have full access to all We’ll work with you on a Reduce development time and
our work throughout your project. customized project plan cost, learn faster, and improve
Your self-sufficiency is our goal. aligned to your business goals. quality and collaboration.

Request a free consultation: www.mathworks.com/pmp 34

Summary

MATLAB and Simulink enable engineers and data scientists to quickly

create, test and implement predictive maintenance programs

35
© 2021 The MathWorks, Inc.
36