Eliminating the Initial State for the Generalized Likelihood Ratio Test

Fault detection based on comparing a batch of data with a model of the system using the generalized likelihood ratio test is considered. Careful treatment of the initial state of the model is quite important, in particular for short batch sizes. There are two standard approaches to this problem. One is based on a parity space, where the influence ofinitial state is removed by projection, and the other on using prior information obtained by Kalman filtering past data. A new idea of anti-causal Kalman filtering in the present data batch is introduced and compared to the previous methods. An efficient parameterization of incipient faults is given. It is shown in simulations of torque disturbances on a DCmotor that efficient fault profile parameterization and using smoothed estimates of the initial state increase performance considerably