Local sensor fault detection using Bayesian decision theory

This paper describes an application of Bayesian likelihood ratio tests to the detection of faults within a sensor. In particular, it concentrates on detecting changes in sensor variance by analysing the residuals of an ARMA model of the sensor output. The proposed technique is implemented for a fault resulting from a change in the variance of one of the inputs of a sensor model. A Bayesian likelihood ratio test was implemented and used to detect the fault. A Monte Carlo study was used to determine the false alarm and missed detection probabilities. The cost (or importance) of occurrences of false alarms and missed detection was found to have an important bearing on the performance of the fault detection scheme, as was the size of data set used for the analysis at each step. The proposed scheme can be extended, most easily to detection of offsets in the residual. In addition, if other faults could be described in a probabilistic manner, it is probable that the technique could be extended to the detection of these.