A neuro-statistical method for fault detection in stochastic systems

Author

Chowdhury, Fahmida ; Lobo, Evarist ; Pei, Xiaoqin ; Rajasekaran, T.

Author_Institution

Dept. of Electr. & Comput. Eng., Univ. of Southwestern Louisiana, Lafayette, LA, USA

Volume

1

fYear

1998

fDate

1-4 Sep 1998

Firstpage

283

Abstract

We present techniques for residual-generation as the basis for statistical hypothesis testing for fault detection in stochastic systems. If a system model is not available, we perform system identification using an ARMAX or NARMAX structure. We propose that a Kalman filter is used to estimate the ARMAX model, and a feedforward neural network is used to estimate the NARMAX model. The test of hypothesis can be done directly on the residual if the system is single-output. For multi-output systems, we show how a neuron can be used to implement a Chi-squared test

Keywords

Kalman filters; autoregressive moving average processes; discrete time systems; fault diagnosis; feedforward neural nets; parameter estimation; statistical analysis; stochastic systems; ARMAX; Chi-squared test; Kalman filter; NARMAX; discrete time systems; fault detection; feedforward neural network; identification; statistical hypothesis testing; stochastic systems; Autoregressive processes; Covariance matrix; Extraterrestrial measurements; Fault detection; Feedforward neural networks; Neural networks; Neurons; Stochastic systems; System identification; System testing;

fLanguage

English

Publisher

ieee

Conference_Titel

Control Applications, 1998. Proceedings of the 1998 IEEE International Conference on

Conference_Location

Trieste

Print_ISBN

0-7803-4104-X

Type

conf

DOI

10.1109/CCA.1998.728425

Filename

728425