Hybrid Genetic-Fuzzy Systems
for Improved Performance
in Residual-Based Fault Detection
Francisco Serdio Fernández
Dep...
NABIC 2014 – Porto, July 30,31 - August 1, 2014
francisco.serdio@jku.at
Introduction
Approach
Results
Conclusions...
Operator Monitoring Tools
System
NABIC 2014 – Porto, July 30,31 - August 1, 2014
francisco.serdio@jku.at
Fault Detecti...
Expert Knowledge
 Allow to detect faults of small sizes
 Difficult or impossible for many systems
 Limited for syste...
Expert Knowledge
J. Korbicz, J.M. Koscielny, Z. Kowalczuk and W. Cholewa (Eds.). Fault Diagnosis: Models,
Artificial Int...
FD without Expert Knowledge
 Discover dependencies between variables
 Represent the ground truth of the systems
 Sta...
Running Fault Detection System
NABIC 2014 – Porto, July 30,31 - August 1, 2014
francisco.serdio@jku.at
http://www.flll....
FD with Residual-based approaches
Analytical Redundancy graphically
Moving from the signal space to the regression line ...
FD with Residual-based approaches
 More information regarding Fault Detection in
F. Serdio, E. Lughofer, K. Pichler, T....
Problems - Low Quality Models
 Fault detection performance decreases
NABIC 2014 – Porto, July 30,31 - August 1, 2014
f...
NABIC 2014 – Porto, July 30,31 - August 1, 2014
francisco.serdio@jku.at
Introduction
Approach
Results
Conclusions...
Our Approach
 Build more residual generators
 Would increase the Fault Detection Performance
 Using Genetic Fuzzy Sy...
NABIC 2014 – Porto, July 30,31 - August 1, 2014
francisco.serdio@jku.at
Our Approach
http://www.flll.Francisco Serdio j...
Genetic Fuzzy Systems
 Codification of an Individual
 Represent an individual  Fuzzy System
NABIC 2014 – Porto, July...
NABIC 2014 – Porto, July 30,31 - August 1, 2014
francisco.serdio@jku.at
Genetic Fuzzy Systems
 Initial population
 “...
 We have extended the Random Convex Crossover
Dumitru Dumitrescu, Beatrice Lazzerini, Lakhmi C Jain, and Anca Dumitrescu...
Crossover
 Applied to μ, σ, β, ω separately
 Parents A, B  Offspring X, Y
 Example with centers μ
1. Select random...
 80% for training, 20% for validation
 Trains the Fuzzy Systems of the individual
 Asses the quality of the Fuzzy Sys...
Testing Environment
 We tested a real scenario  engine test bench
 We used artificial faults
 100 faults  50 runs ...
Introducing faults in the data
NABIC 2014 – Porto, July 30,31 - August 1, 2014
francisco.serdio@jku.at
http://www.flll....
NABIC 2014 – Porto, July 30,31 - August 1, 2014
francisco.serdio@jku.at
Introduction
Approach
Results
Conclusions...
NABIC 2014 – Porto, July 30,31 - August 1, 2014
francisco.serdio@jku.at
Results
http://www.flll.Francisco Serdio jku.at...
Introduction
Approach
Results
Conclusions
NABIC 2014 – Porto, July 30,31 - August 1, 2014
francisco.serdio@jku.at...
 Can be build by Genetic Fuzzy Systems
 Improve the Fault Detection performance
 There is room for improvement
NABIC...
Thanks a lot for your attention!
NABIC 2014 – Porto, July 30,31 - August 1, 2014
francisco.serdio@jku.at
http://www.fll...
of 25

NABIC 2014

F. Serdio, A.-C. Zavoianu, E. Lughofer, K. Pichler, T. Buchegger and H. Efendic, Hybrid Genetic-Fuzzy Systems for Improved Performance in Residual-Based Fault Detection, World Congress on Natural and Biologically Inspired Computing, NaBIC 2014, Porto, Portugal, 2014, pp. 91-96.
Published on: Mar 3, 2016
Published in: Presentations & Public Speaking      
Source: www.slideshare.net


Transcripts - NABIC 2014

  • 1. Hybrid Genetic-Fuzzy Systems for Improved Performance in Residual-Based Fault Detection Francisco Serdio Fernández Department of Knowledge-Based Mathematical Systems Johannes Kepler University Linz, Austria NABIC 2014 – Porto, July 30,31 - August 1, 2014 francisco.serdio@jku.at http://www.flll.Francisco Serdio jku.at/staff/francisco
  • 2. NABIC 2014 – Porto, July 30,31 - August 1, 2014 francisco.serdio@jku.at Introduction Approach Results Conclusions http://www.flll.Francisco Serdio jku.at/staff/francisco
  • 3. Operator Monitoring Tools System NABIC 2014 – Porto, July 30,31 - August 1, 2014 francisco.serdio@jku.at Fault Detection http://www.flll.Francisco Serdio jku.at/staff/francisco
  • 4. Expert Knowledge  Allow to detect faults of small sizes  Difficult or impossible for many systems  Limited for systems with simple equations  Represent the expert knowledge  Allow Pattern Recognition and Classification approaches  we know how a fault looks like J. Korbicz, J.M. Koscielny, Z. Kowalczuk and W. Cholewa (Eds.). Fault Diagnosis: Models, Artificial Intelligence, Applications. Springer-Verlag. Berlin Heidelberg. 2004. NABIC 2014 – Porto, July 30,31 - August 1, 2014 francisco.serdio@jku.at  System models  Expert systems  Fault Patterns http://www.flll.Francisco Serdio jku.at/staff/francisco
  • 5. Expert Knowledge J. Korbicz, J.M. Koscielny, Z. Kowalczuk and W. Cholewa (Eds.). Fault Diagnosis: Models, Artificial Intelligence, Applications. Springer-Verlag. Berlin Heidelberg. 2004. NABIC 2014 – Porto, July 30,31 - August 1, 2014 francisco.serdio@jku.at http://www.flll.Francisco Serdio jku.at/staff/francisco
  • 6. FD without Expert Knowledge  Discover dependencies between variables  Represent the ground truth of the systems  Starting point to produce residual signals  We move to the residual space  We can decide whether there is or not a fault NABIC 2014 – Porto, July 30,31 - August 1, 2014 francisco.serdio@jku.at  Identify systems (Sys Id)  Build models  Compute residuals  Manage residuals http://www.flll.Francisco Serdio jku.at/staff/francisco
  • 7. Running Fault Detection System NABIC 2014 – Porto, July 30,31 - August 1, 2014 francisco.serdio@jku.at http://www.flll.Francisco Serdio jku.at/staff/francisco
  • 8. FD with Residual-based approaches Analytical Redundancy graphically Moving from the signal space to the regression line we can graphically illustrate an untypical signal pattern NABIC 2014 – Porto, July 30,31 - August 1, 2014 francisco.serdio@jku.at http://www.flll.Francisco Serdio jku.at/staff/francisco
  • 9. FD with Residual-based approaches  More information regarding Fault Detection in F. Serdio, E. Lughofer, K. Pichler, T. Buchegger and H. Efendic, Data-Driven Residual-Based Fault Detection for Condition Monitoring in Rolling Mills. Proceedings of the IFAC Conference on Manufacturing Modeling, Management and Control, MIM '2013, St. Petersburg, Russia, 2013, pp. 1546-1551. (Winner of MIM 2013 Best paper award) F. Serdio, E. Lughofer, K. Pichler, T. Buchegger and H. Efendic, Residual-based Fault Detection using Soft Computing techniques for Condition Monitoring at Rolling Mills. Information Sciences, 259, pp. 304–330, 2014. F. Serdio, E. Lughofer, K. Pichler, T. Buchegger, M. Pichler and H. Efendic, Multivariate Fault Detection using Vector Autoregressive Moving Average and Orthogonal Transformation in the residual Space. Annual Conference of the Prognostics and Health Management Society, PHM 2013, New Orleans, LA, USA, 2013, pp. 548-555. F. Serdio, E. Lughofer, K. Pichler, T. Buchegger and H. Efendic, Fault Detection in Multisensor Networks based on Multivariate Time-series Models and Orthogonal Transformations. Information Fusion, (to appear), 2014. NABIC 2014 – Porto, July 30,31 - August 1, 2014 francisco.serdio@jku.at http://www.flll.Francisco Serdio jku.at/staff/francisco
  • 10. Problems - Low Quality Models  Fault detection performance decreases NABIC 2014 – Porto, July 30,31 - August 1, 2014 francisco.serdio@jku.at  Less residual generators http://www.flll.Francisco Serdio jku.at/staff/francisco
  • 11. NABIC 2014 – Porto, July 30,31 - August 1, 2014 francisco.serdio@jku.at Introduction Approach Results Conclusions http://www.flll.Francisco Serdio jku.at/staff/francisco
  • 12. Our Approach  Build more residual generators  Would increase the Fault Detection Performance  Using Genetic Fuzzy Systems  In channels without a good quality model NABIC 2014 – Porto, July 30,31 - August 1, 2014 francisco.serdio@jku.at  How?  Where? http://www.flll.Francisco Serdio jku.at/staff/francisco
  • 13. NABIC 2014 – Porto, July 30,31 - August 1, 2014 francisco.serdio@jku.at Our Approach http://www.flll.Francisco Serdio jku.at/staff/francisco
  • 14. Genetic Fuzzy Systems  Codification of an Individual  Represent an individual  Fuzzy System NABIC 2014 – Porto, July 30,31 - August 1, 2014 francisco.serdio@jku.at http://www.flll.Francisco Serdio jku.at/staff/francisco
  • 15. NABIC 2014 – Porto, July 30,31 - August 1, 2014 francisco.serdio@jku.at Genetic Fuzzy Systems  Initial population  “Smart” individuals  Random individuals http://www.flll.Francisco Serdio jku.at/staff/francisco
  • 16.  We have extended the Random Convex Crossover Dumitru Dumitrescu, Beatrice Lazzerini, Lakhmi C Jain, and Anca Dumitrescu. Evolutionary computation, volume 18. CRC press, 2000.  We used Single Point Mutation NABIC 2014 – Porto, July 30,31 - August 1, 2014 francisco.serdio@jku.at Genetic Fuzzy Systems  Crossover (rate 80%)  Mutation (rate 15%)  Selection  We used Random Selection  Replacement  We used Elitism http://www.flll.Francisco Serdio jku.at/staff/francisco
  • 17. Crossover  Applied to μ, σ, β, ω separately  Parents A, B  Offspring X, Y  Example with centers μ 1. Select random in [-0.2, 0.5] 2. Select random rules to cross 3. Create the new centers by NABIC 2014 – Porto, July 30,31 - August 1, 2014 francisco.serdio@jku.at  Avoid to be disruptive  Behavior http://www.flll.Francisco Serdio jku.at/staff/francisco
  • 18.  80% for training, 20% for validation  Trains the Fuzzy Systems of the individual  Asses the quality of the Fuzzy System  Mean Squared Error (MSE)  Uses training set  The last generation uses the validation set Trevor Hastie, Robert Tibshirani, and Jerome Friedman. The Elements of Statistical Learning: Data Mining, Inference and Prediction. Springer, 2nd edition, 2009. NABIC 2014 – Porto, July 30,31 - August 1, 2014 francisco.serdio@jku.at Fitness  Training and Test http://www.flll.Francisco Serdio jku.at/staff/francisco
  • 19. 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% NABIC 2014 – Porto, July 30,31 - August 1, 2014 francisco.serdio@jku.at http://www.flll.Francisco Serdio jku.at/staff/francisco
  • 20. Introducing faults in the data NABIC 2014 – Porto, July 30,31 - August 1, 2014 francisco.serdio@jku.at http://www.flll.Francisco Serdio jku.at/staff/francisco
  • 21. NABIC 2014 – Porto, July 30,31 - August 1, 2014 francisco.serdio@jku.at Introduction Approach Results Conclusions http://www.flll.Francisco Serdio jku.at/staff/francisco
  • 22. NABIC 2014 – Porto, July 30,31 - August 1, 2014 francisco.serdio@jku.at Results http://www.flll.Francisco Serdio jku.at/staff/francisco
  • 23. Introduction Approach Results Conclusions NABIC 2014 – Porto, July 30,31 - August 1, 2014 francisco.serdio@jku.at http://www.flll.Francisco Serdio jku.at/staff/francisco
  • 24.  Can be build by Genetic Fuzzy Systems  Improve the Fault Detection performance  There is room for improvement NABIC 2014 – Porto, July 30,31 - August 1, 2014 francisco.serdio@jku.at Conclusions  More residuals generators  Add operators to  Merge rules  Add / remove rules http://www.flll.Francisco Serdio jku.at/staff/francisco
  • 25. Thanks a lot for your attention! NABIC 2014 – Porto, July 30,31 - August 1, 2014 francisco.serdio@jku.at http://www.flll.Francisco Serdio jku.at/staff/francisco

Related Documents