Weighted and constrained possibilistic C-means clustering for online fault detection and isolation
- First Online:
- 392 Downloads
In this paper, a new weighted and constrained possibilistic C-means clustering algorithm is proposed for process fault detection and diagnosis (FDI) in offline and online modes for both already known and novel faults. A possibilistic clustering based approach is utilized here to address some of the deficiencies of the fuzzy C-means (FCM) algorithm leading to more consistent results in the context of the FDI tasks by relaxing the probabilistic condition in FCM cost function. The proposed algorithm clusters the historical data set into C different dense regions without having precise knowledge about the number of the faults in the data set. The algorithm incorporates simultaneously possibilistic algorithm and local attribute weighting for time-series segmentation. This allows different weights to be allocated to different features responsible for the distinguished process faults which is an essential characteristic of proper FDI operations. A set of comparative studies have been carried out on the large-scale Tennessee Eastman industrial challenge problem and the DAMADICS actuator benchmark to demonstrate the superiority of the proposed algorithm in process FDI applications with respect to some available alternative approaches.
KeywordsFault detection and isolation Possibilistic clustering Feature weighting
Unable to display preview. Download preview PDF.
- 9.Widodo A, Yang BS (2007) Support vector machine in machine condition monitoring and fault diagnosis. Mech Syst Fault Diagn 21:2560–2574 Google Scholar
- 10.Palade V, Patton RJ, Uppal FJ, Quevedo J, Daley S (2002) Fault diagnosis of an industrial gas turbine using neuro-fuzzy methods. In: IFAC, 2002 Google Scholar
- 12.Vasko KT, Toivonen HTTT (2002) Estimating the number of segments in time series data using permutation tests. IEEE Int Conf Data Mining 466–473 Google Scholar
- 14.Berkhin P (2002) Survey of clustering data mining techniques. Accrue Software, Inc., Fremont Google Scholar
- 16.Mardina K, Kent J, Bibby J (1980) Multivariate analysis. Academic Press, San Diego Google Scholar
- 18.Wettschereck D, Aha DW, Mohri T (1997) A review and empirical evaluation of feature weighting methods for a class of lazy learning algorithms. Artif Intell 11:273–314 Google Scholar
- 20.Srinivasan R, Qian MS (2007) State-specific key variables for monitoring multi-state processes. Chem Eng Res Des 85:1630–1644 Google Scholar
- 28.Kelly PM (1994) An algorithm for merging hyperellipsoidal clusters. Technical Report LA-UR-94-3306, Los Alamos National Laboratory, Los Alamos, NM Google Scholar
- 29.Syfert M, Patton R, Bartys M, Quevedo J (2003) Development and application of methods for actuator diagnosis in industrial control systems (Damadics): a benchmark study. In: Proceedings of the IFAC symposium safe process, pp 939–950 Google Scholar