Page 1
[email protected] http://www.flll.jku.at/staff/franciscoFrancisco Serdio
IEEE SMC 2014 – San Diego, October 5-8, 2014
Reducing False Positives for Residual-Based On-line Fault Detection
by Means of Filters
Francisco Serdio Fernández
Department of Knowledge-Based Mathematical Systems
Johannes Kepler University Linz, Austria
Page 2
[email protected] http://www.flll.jku.at/staff/franciscoFrancisco Serdio
IEEE SMC 2014 – San Diego, October 5-8, 2014
Page 3
[email protected] http://www.flll.jku.at/staff/franciscoFrancisco Serdio
IEEE SMC 2014 – San Diego, October 5-8, 2014
IntroductionApproachResultsConclusions
Page 4
[email protected] http://www.flll.jku.at/staff/franciscoFrancisco Serdio
IEEE SMC 2014 – San Diego, October 5-8, 2014
Fault Detection
Operator Monitoring Tools
System
Page 5
[email protected] http://www.flll.jku.at/staff/franciscoFrancisco Serdio
IEEE SMC 2014 – San Diego, October 5-8, 2014
Operators
There are systems where the operators are only in charge of the monitoring
Page 6
[email protected] http://www.flll.jku.at/staff/franciscoFrancisco Serdio
IEEE SMC 2014 – San Diego, October 5-8, 2014
Operators
There are systems where the operators are also in charge of the operation of the system
Page 7
[email protected] http://www.flll.jku.at/staff/franciscoFrancisco Serdio
IEEE SMC 2014 – San Diego, October 5-8, 2014
Too many false positives Operator… Gets distracted Loses confidence on the system Could even decide to do not use the system anymore
False positives
Jacob Bortman. Invited speaker at PHM 2013, New Orleans, 2013
Page 8
[email protected] http://www.flll.jku.at/staff/franciscoFrancisco Serdio
IEEE SMC 2014 – San Diego, October 5-8, 2014
Fight the false positives !!
Page 9
[email protected] http://www.flll.jku.at/staff/franciscoFrancisco Serdio
IEEE SMC 2014 – San Diego, October 5-8, 2014
IntroductionApproachResultsConclusions
Page 10
[email protected] http://www.flll.jku.at/staff/franciscoFrancisco Serdio
IEEE SMC 2014 – San Diego, October 5-8, 2014
Filters
Use filters in the residual space
Page 11
[email protected] http://www.flll.jku.at/staff/franciscoFrancisco Serdio
IEEE SMC 2014 – San Diego, October 5-8, 2014
Filters
There are some from other domains Engineering, Image processing, Econometrics, etc
We used… Moving Average Modified Moving Average Moving Median Gaussian
Page 12
[email protected] http://www.flll.jku.at/staff/franciscoFrancisco Serdio
IEEE SMC 2014 – San Diego, October 5-8, 2014
Well known in image processing Used to smooth noisy pictures
Moving Average
C. Glasbey and R. Jones, Fast computation of moving average and related filters in octagonal windows, Pattern Recognition Letters, vol. 18, pp. 555–565, 1997.
Page 13
[email protected] http://www.flll.jku.at/staff/franciscoFrancisco Serdio
IEEE SMC 2014 – San Diego, October 5-8, 2014
Well known in econometrics It’s a particular exponential moving average
Modified Moving Average
D.S. Grebenkov and J.Serror, Following a trend with an exponential moving average: Analytical results for a Gaussian mode, Physica A: Statistical Mechanics and its Applications, vol. 394, pp. 288–303, 2014.
Page 14
[email protected] http://www.flll.jku.at/staff/franciscoFrancisco Serdio
IEEE SMC 2014 – San Diego, October 5-8, 2014
Well known in image processing Used to deal with impulse noise
Moving Median
A. M. Kizrak and F. Örzen, A new median filter based fingerprint recognition algorithm, Procedia Computer Science, vol. 3, pp. 859–865, 2011.
Page 15
[email protected] http://www.flll.jku.at/staff/franciscoFrancisco Serdio
IEEE SMC 2014 – San Diego, October 5-8, 2014
Applications in electronics and signal processing Applications in image processing
Gaussian
ISO 11562:1996. Geometrical Product Specifications (GPS) surface texture: profile method metrological characteristics of phase correct filters.
Page 16
[email protected] http://www.flll.jku.at/staff/franciscoFrancisco Serdio
IEEE SMC 2014 – San Diego, October 5-8, 2014
All filters together
Page 17
[email protected] http://www.flll.jku.at/staff/franciscoFrancisco Serdio
IEEE SMC 2014 – San Diego, October 5-8, 2014
Testing Environment
We tested a real scenario engine test bench We used artificial faults
100 faults 50 runs * 2 faults / run 5 fault intensities 5%, 10%, 20%, 50%, 100%
Page 18
[email protected] http://www.flll.jku.at/staff/franciscoFrancisco Serdio
IEEE SMC 2014 – San Diego, October 5-8, 2014
IntroductionApproachResultsConclusions
Page 19
[email protected] http://www.flll.jku.at/staff/franciscoFrancisco Serdio
IEEE SMC 2014 – San Diego, October 5-8, 2014
Results
Page 20
[email protected] http://www.flll.jku.at/staff/franciscoFrancisco Serdio
IEEE SMC 2014 – San Diego, October 5-8, 2014
Results
Linear Regression Moving Median Filter Window size = 8
Page 21
[email protected] http://www.flll.jku.at/staff/franciscoFrancisco Serdio
IEEE SMC 2014 – San Diego, October 5-8, 2014
Results
Page 22
[email protected] http://www.flll.jku.at/staff/franciscoFrancisco Serdio
IEEE SMC 2014 – San Diego, October 5-8, 2014
Results
PLS Regression Moving Median Filter Window size = 8
Page 23
[email protected] http://www.flll.jku.at/staff/franciscoFrancisco Serdio
IEEE SMC 2014 – San Diego, October 5-8, 2014
Results
Page 24
[email protected] http://www.flll.jku.at/staff/franciscoFrancisco Serdio
IEEE SMC 2014 – San Diego, October 5-8, 2014
Results
Fuzzy Systems SparseFIS Moving Median Filter Window size = 4
Page 25
[email protected] http://www.flll.jku.at/staff/franciscoFrancisco Serdio
IEEE SMC 2014 – San Diego, October 5-8, 2014
IntroductionApproachResultsConclusions
Page 26
[email protected] http://www.flll.jku.at/staff/franciscoFrancisco Serdio
IEEE SMC 2014 – San Diego, October 5-8, 2014
Conclusions
Filtering the Residual Space Reduces False Positives Increases True Positives
Median Filters Appear to be the more promising
Other domains… Could bring ideas to ours
Page 27
[email protected] http://www.flll.jku.at/staff/franciscoFrancisco Serdio
IEEE SMC 2014 – San Diego, October 5-8, 2014
Thanks a lot for your attention!