Fault detection properties of global, local and time evolving models for batch process monitoring.

This paper discusses alternative methods for batch process monitoring. Two alternative methods are investigated and compared to an existing one (the benchmark). A description of the models is given and the performance is discussed by means of fault detection performance indices. The performance indices used are the overall type I error and the action signal time. In order to evaluate the performance of the models in terms of the overall type I error and action signal time, six different batch process data sets have been used. The data sets comprise four industrial data sets, one simulated data set and a laboratory spectral data set.