Predictive Maintenance with MATLAB & Simulink

Mehernaz Savai

© 2015 The MathWorks, Inc.

1
Predictive Maintenance

What is Predictive Maintenance?

▪ Maintenance technique that reduces unnecessary
maintenance and eliminates unplanned downtime

What does a Predictive Maintenance solution do?

▪ Uses historical data + sensor data to predict Time-
to-Failure or Remaining Useful Life

▪ Relay this information to maintenance engineers,

operators, and plant managers

Source: Tensor Systems

2
Why is Predictive Maintenance Important?

▪ Improved operating efficiency ▪ New revenue streams ▪ Competitive differentiator

Source: GE Oil & Gas

3
Why is Predictive Maintenance Important?

▪ Improved operating efficiency ▪ New revenue streams ▪ Competitive differentiator

4
What should a Predictive Maintenance Algorithm do?
Turn large volumes of complex data into decisions

What is the condition of

my machine?

When will my machine

Data fail? Decision

How do I prevent this

failure?

Work with
1 Your domain experts
2 Your algorithms can
3
all of your data can do data science run anywhere

5
Predictive Maintenance Algorithm Workflow

Access and Explore Develop Predictive Integrate Analytics

Preprocess Data
Data Models with Systems

Business Data Data Reduction/ Model Creation Enterprise Systems

Transformation

Sensor Data Feature Extraction Model Validation Embedded Devices

6
Access and Preprocess Data

Access and Explore Develop Predictive Integrate Analytics

Preprocess Data
Data Models with Systems

Business Data Data Reduction/ Model Creation Enterprise Systems

Transformation

1
Sensor Data Feature Extraction Model Validation Embedded Devices

7
Access and Preprocess Data

Access and Explore

Preprocess Data
Data

Challenges
Business Data Data Reduction/
▪ I have too much data to handle easily Transformation

▪ I don’t have enough data

▪ I have no data
Sensor Data Feature Extraction
▪ I have too many data types and data
sources

▪ My data is too messy

8
Access and Preprocess Data

Access and Explore

Preprocess Data
Data

Business Data Data Reduction/

Transformation

Sensor Data Feature Extraction

9
Access and Preprocess Data

Access and Explore

Preprocess Data
Data

Business Data Data Reduction/

Transformation

Sensor Data Feature Extraction

10
Access and Preprocess Data

Access and Explore

Find out more:
Preprocess Data
Data
2 PM Session: Employing
Simscape to model electro-
Business Data mechanical systems in Simulink Data Reduction/
Transformation

Sensor Data Feature Extraction

11
Access and Preprocess Data
Work with all of your
1
data

Access and Explore

Databases Images HDFS Preprocess Data
Data

Business Data Data Reduction/

Transformation

Files Signals Videos

Sensor Data Feature Extraction

▪ Point and click tools to access variety of data sources

▪ High-performance environment for big data
▪ Built-in algorithms for data preprocessing

12
Build Predictive Models

Access and Explore Develop Predictive Integrate Analytics

Preprocess Data
Data Models with Systems

Business Data Data Reduction/ Model Creation Enterprise Systems

Transformation

1 2
Sensor Data Feature Extraction Model Validation Embedded Devices

13
Build Predictive Models

Develop Predictive
Preprocess Data
Models

Data Reduction/ Model Creation

Transformation Challenges
▪ I need to incorporate my domain
knowledge

▪ I need to extract and verify health

Feature Extraction Model Validation
indicators

▪ I lack machine learning experience

▪ I have deadlines to meet

14
Build Predictive Models

Develop Predictive
Preprocess Data
Models

Data Reduction/ Model Creation

Transformation

Feature Extraction Model Validation

15
Build Predictive Models
2
Your domain experts
can do data science

Apps MATLAB Develop Predictive

Preprocess Data
Models

Data Reduction/ Model Creation

Transformation

Feature Extraction Model Validation

Simulink

▪ Easy to use apps across multiple domains

▪ Documentation, examples, and videos to get started
▪ Automatic MATLAB code generation
16
Deploy and Integrate

Access and Explore Develop Predictive Integrate Analytics

Preprocess Data
Data Models with Systems

Business Data Data Reduction/ Model Creation Enterprise Systems

Transformation

1 2 3
Sensor Data Feature Extraction Model Validation Embedded Devices

17
Deploy and Integrate

Develop Predictive Integrate Analytics

Models with Systems
Challenges
▪ I have multiple end users – plant
managers, operations analysts,
Model Creation Enterprise Systems
maintenance staff, etc.

▪ I have to allow access through

different target platforms

Model Validation ▪ I need to scale to meet production Embedded Devices

needs

▪ I need to reduce bandwidth

consumption

18
Deploy and Integrate

Develop Predictive Integrate Analytics

Models with Systems

Model Creation Enterprise Systems

Model Validation Embedded Devices

19
Deploy and Integrate
Your algorithms can
3
run anywhere

Develop Predictive Integrate Analytics

Models with Systems

Model Creation Enterprise Systems

Model Validation Embedded Devices

▪ Royalty-free deployment
▪ Web services, apps, and cloud platforms
▪ Computation on smarter edge devices
▪ Automatic C/C++ code generation

20
Predictive Maintenance Algorithm Workflow

Access and Explore Develop Predictive Integrate Analytics

Preprocess Data
Data Models with Systems

Business Data Data Reduction/ Model Creation Enterprise Systems

Transformation

1 2 3
Sensor Data Feature Extraction Model Validation Embedded Devices

21
What does success look like?
Baker Hughes: Pump Health Monitoring System

Challenge
▪ As many as 20 trucks operate around the clock at a well site
▪ A truck with a pump failure must be immediately replaced

Solution
▪ Analyzed a terabyte of data collected at 50,000 samples/second
▪ Performed FFTs and spectral analysis to filter large movements
of the truck, pump, and fluids
▪ Best model was a neural network using pressure, vibration, and
timing sensor data of the valves and valve seats

Results
▪ Savings of more than $10 million projected
▪ Development time reduced tenfold

22
Key Takeaways

Work with
1 Your domain experts
2 Your algorithms can
3
all of your data can do data science run anywhere

23
Key Takeaways

Find out more:

Work with
1 Your domain experts
2 Your algorithms can
3
all of your data Booth: can do data science run anywhere
Data Analytics/
Machine Learning

24