Download as pdf or txt
Download as pdf or txt
You are on page 1of 46

Aircraft Engine Health Monitoring using MATLAB

A Data-based Approach

Peeyush Pankaj Russell Graves


Application Engineering Application Engineering

© 2020 The MathWorks, Inc.


1
AGENDA

▪ Why Predictive Maintenance ?

▪ Predictive Maintenance Workflow


– DEMO: Condition Monitoring of a Turbofan Engine

▪ How Can You Get Started Quickly ?

2
Why Perform Predictive Maintenance?

▪ Example: Mid-air engine failure due to


loose spinner cone
https://www.youtube.com/watch?v=01KSBM-GAKw

▪ Your equipment can cost millions of


dollars

▪ Failures can be dangerous

▪ Maintenance also very expensive

3
Reactive maintenance

Types of maintenance Preventive maintenance

Predictive maintenance

4
Reactive maintenance
Machine
health

x
Failure
P O
Time

Preventive maintenance Challenges:


Machine • Expense
health • Downtime
• Unscheduled Maintenance
x Still usable
condition

Time

Predictive maintenance
Machine
health Uptime & Safety
Optimum time to
now do maintenance Maintenance Cost & Time
x Predicted Unscheduled Maintenance
failure
Time 5
Predictive Maintenance Success Stories

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
6
Demo: Condition monitoring of a Turbofan Engine

Background *
128 engines continuous operation data
8 datasets, each containing 7 failure modes
32 sensors data (flow, temperature, pressure, altitude, Mach, etc..)
3 Flight Classes – Short, Medium & Long

Objective: Develop AI models for


• Fault Classification
• Remaining Useful Life Estimation

* NASA N-CMAPSS Dataset 7


Objective #1: Detect & Classify Faults

8
Objective #2: Predict Remaining Useful Life

9
Collaboration is key to success in Predictive Maintenance

Domain Expert Data Scientist

10
User Stories

11
Predictive Maintenance Workflow
Access and Explore Develop Predictive Integrate Analytics with
Preprocess Data
Data Models Systems

Files Working with Model Creation e.g. Desktop Apps


Messy Data Machine Learning

Databases Data Reduction/ Parameter Enterprise Scale


Transformation Optimization Systems

Sensors Feature Model Embedded Devices


Extraction Validation and Hardware

12
Predictive Maintenance Workflow

Access and Explore


Data

Operational • OSISoft PI
Technology
Challenges
▪ File I/O – Text, Spreadsheet, etc.
Data Sources

Streaming • Apache Kafka


data • Azure EventHub ▪ Web Sources – RESTful, JSON, etc.
▪ Repositories – HDFS, SQL, NoSQL, etc.
Data • AWS S3
stores • Azure BLOB ▪ Real-Time Sources – Sensors, GPS, etc.
▪ How to do visualization of data
• Desktop
Files Files ▪ Generating Synthetic Data

13
Modeling Physical Systems
with Simscape
▪ Foundation library of physical elements
– Mechanical, hydraulic, electrical…
▪ Use with other Simscape add-on products
to model more complex systems
A B

T P

14
Simscape Library

15
Resources: Aircraft systems Modeling & Simulation

Electric Aircraft Model in Simscape

Aileron Actuator Development


with Model-Based Design

Hybrid Electric Aircraft Model in Simscape Webinar


Airplane Wing Landing Gear Actuation Webinar
16
Simscape
Key Features
REQUIREMENTS
▪ Foundation physical modeling blocks in
SYSTEM
more than 10 different physical domains
– Mechanical, electrical, hydraulic,…
▪ Custom components with Simscape language
Control Mechanical Electrical
▪ Solver technology for systems of DAEs
– Symbolic equation reduction, event handling
IMPLEMENTATION
– Specialized solvers for real-time simulation
▪ Use of add-on products without Embed. HIL
purchasing add-on product license Software System

▪ Support for C-code generation


– Test embedded controllers without hardware prototypes
INTEGRATION AND TEST

18
Access Data

datastore

19
DEMO: Data Exploration

20
Predictive Maintenance Workflow

Access and
Preprocess &Explore
Analyze
Data
Data Challenges
▪ Data clean up
– Poorly formatted files
Operational • Working with
OSISoft PI
Messy Data – Irregularly sampled data
Technology
– Redundant data, outliers, missing data etc.
Data Sources

Streaming • Data Reduction/


Apache Kafka ▪ Data specific processing
data • Transformation
Azure EventHub – Signals: Smoothing, resampling, denoising, Wavelet
transforms, etc.
Data • AWS S3 – Images: Image registration, morphological filtering,
stores • Azure BLOB deblurring, etc.
Feature Data preparation accounts for about 80% of the
Extraction work of data scientists - Forbes
• Desktop
Files Files

21
DEMO: Data Pre-processing

22
Predictive Maintenance Workflow

A condition indicator can be any feature that is useful -


Develop
Access Forecast
and Explore • for distinguishing normal from faulty operation or
Models
Data • for predicting remaining useful life

Challenges
Operational • OSISoft PI
Technology
Model Creation e.g. ▪ Lack of data science expertise
Machine Learning /
Deep Learning
▪ Feature Extraction – How to transform data to best represent
Data Sources

Streaming • Apache Kafka the system?


data • Azure EventHub – Requires subject matter expertise
Parameter
– No right way of designing features
Optimization
Data • AWS S3
stores • Azure BLOB ▪ Feature Selection – What attributes or subset of data to use?
– Entails a lot of iteration – Trial and error
Model
• Desktop – Difficult to evaluate features
Files Validation
Files
▪ Model Development
– Many different models
– Model Validation and Tuning

▪ Time required to conduct the analysis

23
DEMO: AI Modeling

24
Signal-Based Condition Indicators

Time-domain Frequency-domain Time-frequency


features features domain features

Mean Power bandwidth Spectral entropy


Standard deviation Mean frequency Spectral kurtosis
Skewness Peak values

...
Root-mean square Peak frequencies
Kurtosis Harmonics

...
...

Learn more about Condition Indicators


25
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

26
Diagnostic Feature Designer

27
Machine deterioration
profile
Condition indicator
Current condition
Failure condition

What is RUL?

Time
Remaining useful life (RUL)

[ Number of days ]
[ Miles ]
[ Cycles ]

Learn Further: MATLAB RUL Estimator models


28
RUL Estimator Models

Similarity model Degradation model


Safet y
threshold
Healthy state Failure
Survival model
Healthy state Failure

Healthy state Failure

29
RUL Methods and when to use them
Requirement: Need to know what constitutes failure data

30
DEMO: RUL Modeling

31
Predictive Maintenance Workflow

Access and Explore


Operationalize Preprocess & Analyze Develop Forecast Operationalize
Data
Models Data Models Models

Operational • OSISoft PI Working with

Dashboards
Dashboards Model Creation e.g.

Desktop App
Desktop App

Technology Messy Data


Machine Learning /
Deep Learning
Tableau Tableau
PowerBI Spotfire PowerBI
Data Sources

Streaming Spotfire
• Apache Kafka Data Reduction/
data • AzureQlik
EventHub Transformation Qlik
Parameter
Optimization

Enterprise App
Enterprise App

Data • AWS S3

Web App
• Azure BLOB
Web App

stores
Feature
Extraction
{• RESTful API } Model { RESTful API }
Desktop
Files Files Validation

32
Predictive Maintenance Workflow

Challenges
Operationalize ▪ End user: Operators, Analysts, Administrative
Models Staff, customers etc.

▪ Different target platforms:


Dashboards

Desktop App

Cluster or Cloud environment


– Standalone desktop applications
Tableau
Spotfire PowerBI – Server based Web and enterprise systems
Qlik
– Embedded hardware
Enterprise App

▪ Different Interfaces: C++, Java, Python, .NET


Web App

etc.
{ RESTful API }
▪ Need to translate analytics to production
environment

33
Operationalize Models
Simulink
Compiler

MATLAB
Coder MATLAB
Compiler
Products Compiler
MATLAB SDK

Embedded Hardware Enterprise Systems

Excel Hadoop/ Java MATLAB


Add-in
C/C++ ++ Python .NET Production
C, C++ HDL PLC CUDA .exe Spark DLL JAR py DLL Server
MATLAB Deployed
Web App Archive

MATLAB
MATLAB Web Production
App Server Server

34
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

Data Cleaning, Desktop/


▪ Predictive maintenance BIG DATA Outlier
Frequency domain
& Time-Frequency
WebApp/ MPS/
Handling Detection & Em bedded
– Saves money for equipment operators Signal Filters
Analysis/ ML Apps Hardw are

– Increases reliability and safety of equipment Datastores


Tall arrays
LIVE tasks
ML Apps
PdM Tlbx → DFD
Coder & Compiler
Products

– Creates opportunities for new services that equipment


manufacturers can provide

35
MathWorks can help you get started TODAY

▪ Examples

▪ Documentation

▪ Tutorials & Workshops

▪ Tech Talk Series

37
MathWorks Services

▪ Consulting
– Integration
– Data analysis/visualization
– Unify workflows, models, data
www.mathworks.com/services/consulting/

▪ Training
– Classroom, online, on-site
– Data Processing, Visualization, Deployment, Parallel Computing

www.mathworks.com/services/training/

38
© 2016 The MathWorks, Inc. MATLAB and Simulink are registered trademarks of The MathWorks, Inc. See www.mathworks.com/trademarks
for a list of additional trademarks. Other product or brand names may be trademarks or registered trademarks of their respective holders.

© 2020 The MathWorks, Inc.


39
Safran Uses Simulink to Generate Failure Data To Train Neural
Networks To Detect Anomalies and Predict Failures in Factories

MATLAB Expo Talk Link

40
What Does Success Look Like?
SAFRAN Engine Health Monitoring Solution

▪ Monitor Systems
– Detect failure indicators
– Predict time to maintenance
– Identify components

▪ Improve Aircraft Availability


– On time departures and arrivals
– Plan and optimize maintenance
– Reduce engine out-of-service time

▪ Reduce Maintenance Costs


– Troubleshooting assistance Compiled Enterprise
Desktop
– Limit secondary damage Shared Integration
• Ad-hoc data analysis • Suite of MATLAB Analytics • Real-time analytics
• Analytics to predict failure • Shared with other teams • Integrated with
• Proof of readiness maintenance and service
systems
http://www.mathworks.com/company/events/conferences/matlab-virtual-conference/

41
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


42
Clockworks Analytics Reveals Building’s Hidden Energy Costs with
Analytics Engine and Digital Twin
Challenge
Improve building operational performance and energy
usage by identifying problems before they result in energy
waste and major repairs.
Solution
▪ Use MATLAB, modeling, and machine learning with
detailed diagnostic libraries for fault detection and
diagnosis.
Key Outcomes Clients with large building operations can identify waste
and reduce instances of equipment failure.
▪ Analytics engine identifies major, wasteful issues.
▪ Building operators address problems as soon as they “We wanted to be able to hire mechanical
occur. engineers with domain knowledge for the job.
Our engineers leverage the rich base of
▪ Continuous commissioning optimizes energy use and
MATLAB calculation tools without having to
issue resolution.
become traditional programmers.” —Siân
Kleindienst, chief scientist at Clockworks
Link feature story Analytics
43
Krones Develops Package-Handling Robot Digital Twin

Challenge
Increase the performance of an automated beverage-
packaging system by incorporating a dynamic tripod robot
into the design
Solution
Use Simulink and Simscape Multibody to create an
accurate digital twin that supports design optimization, fault
testing, and predictive maintenance The Krones Robobox T-GM package-handling robot.

Results “Simulations of the digital twin in Simulink enabled us


▪ Robot performance increased to obtain data and insights that would be either
▪ Product development time shortened impossible to get via hardware tests or simply too
▪ Testing time significantly reduced costly and time-consuming. Visualizing forces and
moments helped us to understand the effects of
individual components on a highly dynamic robot.”
- Benedikt Böttcher, Krones

Link to user story 44


Korea Institute of Energy Research Develops AI-Based
Predictive Maintenance Models for Offshore Wind Power

Challenge
Prevent costly downtime for offshore wind turbines by
identifying potential component failures before they occur
Solution
Use MATLAB to develop machine learning and deep
learning algorithms that use existing sensor data to predict
possible failures
KIER Wind Turbine Health Monitoring System.
Results
▪ Development time cut in half
▪ 90%+ prediction accuracy achieved “Despite having little previous experience with AI,
▪ Aggressive deadline met within a limited budget and a tight deadline, we
completed a diagnostics model in MATLAB capable
of detecting wind turbine component failure with over
90%+ accuracy.”
- Jung Chul Choi, Korea Institute of Energy Research

Link to user story 45


Mondi Implements Statistics-Based Health Monitoring and
Predictive Maintenance for Manufacturing Processes with
Machine Learning
Challenge
Reduce waste and machine downtime in plastics
manufacturing plants
Solution
Use MATLAB to develop and deploy monitoring and
predictive maintenance software that uses machine One of Mondi Gronau’s plastic production
machines, which deliver about 18 million tons of
learning algorithms to predict machine failures plastic and thin film products annually.

Results “MathWorks Consulting’s support is among the best I’ve seen;


▪ More than 50,000 euros saved per year the consultants are fast and exceptionally knowledgeable. We’ve
▪ Prototype completed in six months already seen a positive return on investment from cost savings,
▪ Production software run 24/7 and now we have more budget and time to complete more
machine learning projects that will provide similar benefits.”
- Dr. Michael Kohlert, Mondi

Link to user story 46


LG Energy Solution Uses Deep Learning for Predictive
Maintenance on Industrial Cutter

Challenge
Eliminate reliance on the site engineer’s opinion for scheduling
predictive maintenance, which can sometimes be too
conservative
Solution
Develop a condition monitoring system and deploy standalone
executable that acquires raw data from an NI device, makes
predictions, and displays the result.
Condition monitoring system using deep learning.
Key Outcomes
▪ Generated features and trained various AI models using
interactive apps
▪ Developed entire workflow from data acquisition to “Three advantages of MATLAB that led our project to
deployment success: app-based AI development workflow, compatibility
▪ Accelerated prototyping using MathWorks engineering
with third-party hardware, and short test cycle with rapid
support
prototyping.”
– Junghoon Lee, LG Energy Solution
» Explore MATLAB for predictive maintenance
47
Predictive Maintenance Applications in Auto

Link to User Story: Video Link


48

You might also like