In this paper, some applications of Kullback Discrimination Information (KDI) to fault detection and model validation are developed for black-box type dynamical systems. Based on a general input-output model structure, the system is identified using data from distinct time intervals of finite but fairly large sets. Two models obtained are compared with the KDI defined by likelihood functions corresponding to the models, then the problem leads to model discrimination. An iterative scheme is derived for a feasible evaluation of the KDI with large data sets by using a Bayesian approach to likelihood functions. From the results several new criteria are introduced for model discrimination and they can be effectively used for batch fault detection and for model cross-validation in a thresholding approach. For a reasonable selection of a threshold value, statistical properties of the criteria are analized using asymptotic properties of model parameter estimates. Finally to confirm the effectiveness of the method, some simulation studies on fault detection and model validation are considered for a second order oscillator system.
[1]
N. Ishii,et al.
Computer classification of the EEG time series by Kullback information measure
,
1980
.
[2]
A. Willsky,et al.
A generalized likelihood ratio approach to the detection and estimation of jumps in linear systems
,
1976
.
[3]
M. Stone.
An Asymptotic Equivalence of Choice of Model by Cross‐Validation and Akaike's Criterion
,
1977
.
[4]
Lennart Ljung,et al.
Theory and Practice of Recursive Identification
,
1983
.
[5]
R. Gray,et al.
Distortion measures for speech processing
,
1980
.
[6]
Takashi Soeda,et al.
A method of predicting failure or life for stochastic systems by using autoregressive models
,
1980
.
[7]
R. Clark.
The dedicated observer approach to instrument failure detection
,
1979,
1979 18th IEEE Conference on Decision and Control including the Symposium on Adaptive Processes.
[8]
R. Shumway,et al.
Linear Discriminant Functions for Stationary Time Series
,
1974
.