MODEL BASED FAULT DETECTION AND ISOLATION BY FAULT PARAMETER ELIMINATION

This paper deals with model based fault detection and isolation of a pickling process within the steel industry. The model is based on the grey box methodology and reflects the physical behaviour of the process. Possible faults are included in the model as parameters, which are estimated on line. The estimation is based on minimizing a loss function using past data from a defined moving time window. The procedure of finding the faults starts by estimate all defined fault parameters. One fault parameter is removed from the set of prospective list of faults by removing the parameter with the smallest saliency. The saliency is defined as the quote between the parameter estimate and the corresponding element of the inverse of the hessian matrix. The parameter with the smallest saliency gives a measure of the relevance of the estimated parameters relative all estimated parameters. The procedure is repeated until all fault parameters are eliminated from the list. To isolate the faults, the Akaike's Information Criterion (AIC) is used to detect faults. This gives the threshold when a fault relevant parameter is removed from the list of prospective faults.