Prognosis Based on Handling Drifts in Dynamical Environments: Application to a Wind Turbine Benchmark

• Published in: 2012 Eleventh International Conference on Machine Learning and Applications Vol. 2; pp. 1 - 6
• Main Authors: Chammas, A., Duviella, E., Leceouche, S.
• Format: Conference Proceeding
• Language: English
• Published: IEEE 01.12.2012
• Subjects: Benchmark testing; Clustering algorithms; Computational modeling; Degradation; drift; dynamical clus- tering; Heuristic algorithms; incipient fault; prognosis; Silicon; Wind turbines
• ISBN: 1467346519; 9781467346511
• DOI: 10.1109/ICMLA.2012.131

In this paper, we present a prognosis architecture that allows the computation of the Remaining Useful Life (RUL) of a failing process. A process subject to an incipient fault experiments slowly developing degradation. Sensor measurements and Condition Monitoring (CM) data extracted from the system allow to follow up the process drift. The prognosis architecture we propose makes use of a dynamical clustering algorithm to model the data in a feature space. This algorithm uses a sliding window scheme on which the model is iteratively updated. Metrics applied on the parameters of this model are used to compute a drift severity indicator, which is also an indicator of the health of the system. The architecture for prognosis is applied on a benchmark of wind turbine. The used benchmark has been constructed to serve as a realistic wind turbine model. It was used in the context of a global scale fault diagnosis and fault tolerant control competition. The benchmark also proposed a drifting fault scenario that we used to test our approach.