Estimation of Remaining Useful Life Based on Switching Kalman Filter Neural Network Ensemble

Abstract : The proposed method is an extension of an existing Kalman Filter (KF) ensemble method. While the original method has shown great promise in the earlier PHM 2008 Data Challenge, the main limitation of the KF ensemble is that it is only applicable to linear models. In prognostics, degradation of mechanical systems is typically non-linear in nature, therefore limiting the applications of KF ensemble in this area. To circumvent this problem, this paper propose to approximate non-linear functions with piecewise linear functions. When estimating the RUL, the Switching Kalman Filter(SKF) is able to choose the most probable degradation mode and thus make better predictions. The implementation of the proposed SKF ensemble method is illustrated by implementing on NASAs C-MAPSS Dataset as well as the PHM2008 Data Challenge Dataset. The results show the effectiveness of the SKF in detecting the switching point between various degradation modes as well as the improved accuracy of the SKF ensemble method compared to other available methods in literature.

[1]  Abhinav Saxena,et al.  Damage propagation modeling for aircraft engine run-to-failure simulation , 2008, 2008 International Conference on Prognostics and Health Management.

[2]  Anders Krogh,et al.  Neural Network Ensembles, Cross Validation, and Active Learning , 1994, NIPS.

[3]  Enrico Zio,et al.  A Kalman Filter-Based Ensemble Approach With Application to Turbine Creep Prognostics , 2012, IEEE Transactions on Reliability.

[4]  Jianbo Yu,et al.  A similarity-based prognostics approach for Remaining Useful Life estimation of engineered systems , 2008, 2008 International Conference on Prognostics and Health Management.

[5]  Sharad Singhal,et al.  Training feed-forward networks with the extended Kalman algorithm , 1989, International Conference on Acoustics, Speech, and Signal Processing,.

[6]  Sébastien Borguet,et al.  Coupling principal component analysis and Kalman filtering algorithms for on-line aircraft engine diagnostics , 2009 .

[7]  Giorgio Valentini,et al.  Ensemble methods : a review , 2012 .

[8]  L. Peel,et al.  Data driven prognostics using a Kalman filter ensemble of neural network models , 2008, 2008 International Conference on Prognostics and Health Management.

[9]  D. Opitz,et al.  Popular Ensemble Methods: An Empirical Study , 1999, J. Artif. Intell. Res..

[10]  Zhi-Hua Zhou,et al.  Ensemble Methods: Foundations and Algorithms , 2012 .

[11]  Khanh Le Son,et al.  Remaining useful life estimation on the non-homogenous gamma with noise deterioration based on Gibbs filtering: A case study , 2012, 2012 IEEE Conference on Prognostics and Health Management.

[12]  Lee A. Feldkamp,et al.  Decoupled extended Kalman filter training of feedforward layered networks , 1991, IJCNN-91-Seattle International Joint Conference on Neural Networks.

[13]  F.O. Heimes,et al.  Recurrent neural networks for remaining useful life estimation , 2008, 2008 International Conference on Prognostics and Health Management.

[14]  Al Mathami,et al.  Probabilistic Controllled Airpsace Infringement Tool , 2012 .