+ All Categories
Home > Documents > Prof. Dr. Olga Fink, ETH Zurich2bf51729-d6b1-4476-b50d...Can only use Healthy data for training! 1...

Prof. Dr. Olga Fink, ETH Zurich2bf51729-d6b1-4476-b50d...Can only use Healthy data for training! 1...

Date post: 19-Mar-2020
Category:
Upload: others
View: 1 times
Download: 0 times
Share this document with a friend
25
SNSF (Swiss National Science Foundation) Professor for intelligent maintenance systems at ETH Zürich Before, she was heading the research group “Smart Maintenance” at the Zurich University of Applied Sciences (ZHAW) Holds Ph.D. in civil engineering from ETH Zurich, and Diploma degree in industrial engineering from Hamburg University of Technology Gained valuable industrial experience as reliability engineer for railway rolling stock and as reliability and maintenance expert for railway systems Research focuses on Data‐Driven Condition‐Based and Predictive Maintenance, amongst others Prof. Dr. Olga Fink, ETH Zurich 1
Transcript
Page 1: Prof. Dr. Olga Fink, ETH Zurich2bf51729-d6b1-4476-b50d...Can only use Healthy data for training! 1 fault Olga Fink 06.03.2019 8 Michau, G., T. Palmé, and O. Fink (2017): Deep Feature

• SNSF (Swiss National Science Foundation) Professor for intelligent

maintenance systems at ETH Zürich

• Before, she was heading the research group “Smart Maintenance” at

the Zurich University of Applied Sciences (ZHAW)

• Holds Ph.D. in civil engineering from ETH Zurich, and Diploma degree

in industrial engineering from Hamburg University of Technology

• Gained valuable industrial experience as reliability engineer for railway

rolling stock and as reliability and maintenance expert for railway

systems

• Research focuses on Data‐Driven Condition‐Based and Predictive

Maintenance, amongst others

Prof. Dr. Olga Fink, ETH Zurich

1

Page 2: Prof. Dr. Olga Fink, ETH Zurich2bf51729-d6b1-4476-b50d...Can only use Healthy data for training! 1 fault Olga Fink 06.03.2019 8 Michau, G., T. Palmé, and O. Fink (2017): Deep Feature

||Intelligent Maintenance Systems

Prof. Dr. Olga Fink

06.03.2019

Deep learning and artificial intelligence for predictive

maintenance applications

Olga Fink 2

Page 3: Prof. Dr. Olga Fink, ETH Zurich2bf51729-d6b1-4476-b50d...Can only use Healthy data for training! 1 fault Olga Fink 06.03.2019 8 Michau, G., T. Palmé, and O. Fink (2017): Deep Feature

||Intelligent Maintenance Systems

Examples of what AI can do today

06.03.2019Olga Fink

Recognize ObjectsSpot cancer in tissue slides

Hold Interviews at Press Conferences

Teach Itself How to Code

Write film scriptsPaint paintings that

are sold

Source: www.businessinsider.com

3

Page 4: Prof. Dr. Olga Fink, ETH Zurich2bf51729-d6b1-4476-b50d...Can only use Healthy data for training! 1 fault Olga Fink 06.03.2019 8 Michau, G., T. Palmé, and O. Fink (2017): Deep Feature

||Intelligent Maintenance Systems

06.03.2019Olga Fink 4

In which areas does AI excel today?

Artificial Narrow Intelligence

Artificial General Intelligence

ArtificialSuperintelligence

Page 5: Prof. Dr. Olga Fink, ETH Zurich2bf51729-d6b1-4476-b50d...Can only use Healthy data for training! 1 fault Olga Fink 06.03.2019 8 Michau, G., T. Palmé, and O. Fink (2017): Deep Feature

||Intelligent Maintenance Systems

Application Fields of Deep Learning and Artificial Intelligence

25.02.2019Olga Fink

Predictive

Maintenance

Computer Vision Natural Language Processing Speech Recognition

Olga Fink 5

Page 6: Prof. Dr. Olga Fink, ETH Zurich2bf51729-d6b1-4476-b50d...Can only use Healthy data for training! 1 fault Olga Fink 06.03.2019 8 Michau, G., T. Palmé, and O. Fink (2017): Deep Feature

||Intelligent Maintenance Systems

06.03.2019Olga Fink

Challenges in Predictive Maintenance

6

Few occurring faults with very specific characteristics + Large variety of different fault types Not sufficient examples for the algorithms to learn from

Page 7: Prof. Dr. Olga Fink, ETH Zurich2bf51729-d6b1-4476-b50d...Can only use Healthy data for training! 1 fault Olga Fink 06.03.2019 8 Michau, G., T. Palmé, and O. Fink (2017): Deep Feature

||Intelligent Maintenance Systems

Learning the “health” indicator

“Health Indicator”

06.03.2019Olga Fink 7

Learning

features

automatically

Michau, G., T. Palmé, and O. Fink (2017): Deep Feature Learning Network for Fault Detection and Isolation, Annual conference of the PHM society, October 2017

Page 8: Prof. Dr. Olga Fink, ETH Zurich2bf51729-d6b1-4476-b50d...Can only use Healthy data for training! 1 fault Olga Fink 06.03.2019 8 Michau, G., T. Palmé, and O. Fink (2017): Deep Feature

||Intelligent Maintenance Systems

Generator Health Monitoring

320 monitoring sensors:

Partial discharge

Rotor shaft voltage

Rotor flux

Stator end winding vibration

Stator Water Temperature

275 days of recorded operation,

60 000 observations

1 faultCan only use Healthy data for training!

06.03.2019Olga Fink 8

Michau, G., T. Palmé, and O. Fink (2017): Deep Feature Learning Network for Fault Detection and Isolation, Annual conference of the PHM society, October 2017

Page 9: Prof. Dr. Olga Fink, ETH Zurich2bf51729-d6b1-4476-b50d...Can only use Healthy data for training! 1 fault Olga Fink 06.03.2019 8 Michau, G., T. Palmé, and O. Fink (2017): Deep Feature

||Intelligent Maintenance Systems

Generator Health Monitoring

HealthyAbnormal behavior 100 days before!

06.03.2019Olga Fink 9

Time [days]

No

rmali

zed

dis

tan

ceto

the

train

ing

set

Page 10: Prof. Dr. Olga Fink, ETH Zurich2bf51729-d6b1-4476-b50d...Can only use Healthy data for training! 1 fault Olga Fink 06.03.2019 8 Michau, G., T. Palmé, and O. Fink (2017): Deep Feature

||Intelligent Maintenance Systems

06.03.2019Olga Fink

Challenges in Predictive Maintenance

10

Few occurring faults with very specific characteristics + Large variety of different fault types Not sufficient examples for the algorithms to learn from

Page 11: Prof. Dr. Olga Fink, ETH Zurich2bf51729-d6b1-4476-b50d...Can only use Healthy data for training! 1 fault Olga Fink 06.03.2019 8 Michau, G., T. Palmé, and O. Fink (2017): Deep Feature

||Intelligent Maintenance Systems

06.03.2019Olga Fink

Challenges in Predictive Maintenance

11

Few occurring faults with very specific characteristics + Large variety of different fault types Not sufficient examples for the algorithms to learn from

Classification (that is typically applied in fault isolation) not suitable for distinguishing between the fault types

Page 12: Prof. Dr. Olga Fink, ETH Zurich2bf51729-d6b1-4476-b50d...Can only use Healthy data for training! 1 fault Olga Fink 06.03.2019 8 Michau, G., T. Palmé, and O. Fink (2017): Deep Feature

||Intelligent Maintenance Systems

Integrated fault diagnostics

“Health Indicator”

06.03.2019Olga Fink 12

Michau, G., T. Palmé, and O. Fink (2017): Deep Feature Learning Network for Fault Detection and Isolation, Annual conference of the PHM society, October 2017

Page 13: Prof. Dr. Olga Fink, ETH Zurich2bf51729-d6b1-4476-b50d...Can only use Healthy data for training! 1 fault Olga Fink 06.03.2019 8 Michau, G., T. Palmé, and O. Fink (2017): Deep Feature

||Intelligent Maintenance Systems

Water Temperature Shaft Voltage Rotor Flux Rotor Flux

Integrated fault diagnostics: Generator case study

At no additional cost!

06.03.2019Olga Fink 13

Michau, G., T. Palmé, and O. Fink (2017): Deep Feature Learning Network for Fault Detection and Isolation, Annual conference of the PHM society, October 2017

Integrated!

Page 14: Prof. Dr. Olga Fink, ETH Zurich2bf51729-d6b1-4476-b50d...Can only use Healthy data for training! 1 fault Olga Fink 06.03.2019 8 Michau, G., T. Palmé, and O. Fink (2017): Deep Feature

||Intelligent Maintenance Systems

06.03.2019Olga Fink

Challenges in Predictive Maintenance

14

Few occurring faults with very specific characteristics + Large variety of different fault types Not sufficient examples for the algorithms to learn from

Classification (that is typically applied in fault isolation) not suitable for distinguishing between the fault types

Page 15: Prof. Dr. Olga Fink, ETH Zurich2bf51729-d6b1-4476-b50d...Can only use Healthy data for training! 1 fault Olga Fink 06.03.2019 8 Michau, G., T. Palmé, and O. Fink (2017): Deep Feature

||Intelligent Maintenance Systems

06.03.2019Olga Fink

Challenges in Predictive Maintenance

15

Few occurring faults with very specific characteristics + Large variety of different fault types Not sufficient examples for the algorithms to learn from

Classification (that is typically applied in fault isolation) not suitable for distinguishing between the fault types

Varying and evolving operating conditions Even healthy

system conditions are not always representative

Page 16: Prof. Dr. Olga Fink, ETH Zurich2bf51729-d6b1-4476-b50d...Can only use Healthy data for training! 1 fault Olga Fink 06.03.2019 8 Michau, G., T. Palmé, and O. Fink (2017): Deep Feature

||Intelligent Maintenance Systems

06.03.2019Olga Fink 16

Example Gas Turbines

Only a short observation period

Winter

16Olga Fink 25.02.2019

Page 17: Prof. Dr. Olga Fink, ETH Zurich2bf51729-d6b1-4476-b50d...Can only use Healthy data for training! 1 fault Olga Fink 06.03.2019 8 Michau, G., T. Palmé, and O. Fink (2017): Deep Feature

||Intelligent Maintenance Systems

06.03.2019

Example Gas Turbines

Olga Fink

Extend the

observation period

17Olga Fink 25.02.2019

Page 18: Prof. Dr. Olga Fink, ETH Zurich2bf51729-d6b1-4476-b50d...Can only use Healthy data for training! 1 fault Olga Fink 06.03.2019 8 Michau, G., T. Palmé, and O. Fink (2017): Deep Feature

||Intelligent Maintenance Systems

06.03.2019Olga Fink

Solution: Using the fleet experience! Transfer the experience

Transfer of experience with respect to the healthy

operating conditions Enlarge the set of

representative «healthy data»

Transfer the experience with respect to faulty system

conditions

But: single units are operated differently, have different configurations and

environmental conditions

Challenge: If fleet units too similar no additional experience added

If fleet units too different faulty system conditions recognized as healthy

18

Page 19: Prof. Dr. Olga Fink, ETH Zurich2bf51729-d6b1-4476-b50d...Can only use Healthy data for training! 1 fault Olga Fink 06.03.2019 8 Michau, G., T. Palmé, and O. Fink (2017): Deep Feature

||Intelligent Maintenance Systems

Using the fleet experience

06.03.2019Olga Fink

Michau, Gabriel, Thomas Palmé, and Olga Fink. 2018. “Fleet PHM for Critical Systems: Bi-Level Deep Learning Approach for Fault Detection.” In

European Prognognostics and Health Management Conference. Utrecht.

A

B

19

A B

Page 20: Prof. Dr. Olga Fink, ETH Zurich2bf51729-d6b1-4476-b50d...Can only use Healthy data for training! 1 fault Olga Fink 06.03.2019 8 Michau, G., T. Palmé, and O. Fink (2017): Deep Feature

||Intelligent Maintenance Systems

Fleet of Gas Turbines Plant 1: Healthy

06.03.2019 Olga Fink

20Olga Fink 2025.02.2019

Michau, Gabriel, Thomas Palmé, and Olga Fink. 2018. “Fleet PHM for Critical Systems: Bi-Level Deep Learning Approach for Fault Detection.” In

European Prognognostics and Health Management Conference. Utrecht.

Page 21: Prof. Dr. Olga Fink, ETH Zurich2bf51729-d6b1-4476-b50d...Can only use Healthy data for training! 1 fault Olga Fink 06.03.2019 8 Michau, G., T. Palmé, and O. Fink (2017): Deep Feature

||Intelligent Maintenance Systems

Fleet of Gas Turbines Plant 2: Fault

06.03.2019 Olga Fink

21Olga Fink 2125.02.2019

Michau, Gabriel, Thomas Palmé, and Olga Fink. 2018. “Fleet PHM for Critical Systems: Bi-Level Deep Learning Approach for Fault Detection.” In

European Prognognostics and Health Management Conference. Utrecht.

Page 22: Prof. Dr. Olga Fink, ETH Zurich2bf51729-d6b1-4476-b50d...Can only use Healthy data for training! 1 fault Olga Fink 06.03.2019 8 Michau, G., T. Palmé, and O. Fink (2017): Deep Feature

||Intelligent Maintenance Systems

06.03.2019Olga Fink

Challenges in Predictive Maintenance

22

Few occurring faults with very specific characteristics + Large variety of different fault types Not sufficient examples for the algorithms to learn from

Classification (that is typically applied in fault isolation) not suitable for distinguishing between the fault types

Varying and evolving operating conditions Even healthy

system conditions are not always representative

Page 23: Prof. Dr. Olga Fink, ETH Zurich2bf51729-d6b1-4476-b50d...Can only use Healthy data for training! 1 fault Olga Fink 06.03.2019 8 Michau, G., T. Palmé, and O. Fink (2017): Deep Feature

||Intelligent Maintenance Systems

06.03.2019Olga Fink 23

Areas of further research

AI in Mainte-nance

Interpre-tability

Self-Adapta-

bility

Robust-ness

Scala-bility

Auto-mation

Page 24: Prof. Dr. Olga Fink, ETH Zurich2bf51729-d6b1-4476-b50d...Can only use Healthy data for training! 1 fault Olga Fink 06.03.2019 8 Michau, G., T. Palmé, and O. Fink (2017): Deep Feature

Legal notice

24

©2019 Swiss Re. All rights reserved. You are not permitted to create any modifications or derivative works of this presentation or webinar or to use it for commercial or other public purposes without the prior written permission of Swiss Re.

The information and opinions contained in the presentation are provided as at the date of the presentation and are subject to change without notice. Although the information used was taken from reliable sources, Swiss Re does not accept any responsibility for the accuracy or comprehensiveness of the details given. All liability for the accuracy and completeness thereof or for any damage or loss resulting from the use of the information contained in this presentation is expressly excluded. Under no circumstances shall Swiss Re or its Group companies be liable for any financial or consequential loss relating to this presentation.

Page 25: Prof. Dr. Olga Fink, ETH Zurich2bf51729-d6b1-4476-b50d...Can only use Healthy data for training! 1 fault Olga Fink 06.03.2019 8 Michau, G., T. Palmé, and O. Fink (2017): Deep Feature

25


Recommended