Dynamic process monitoring and fault detection in a batch fermentation process. comparative performance assessment between MPCA and BDPCA

A comparative study between two multivariate statistical techniques for batch process monitoring and fault diagnosis is presented. Such methods are Multiway Principal Component Analysis (MPCA) and Batch Dynamic Principal Component Analysis (BDPCA) and both are applied for monitoring the penicillin production process. Three faults are used to evaluate the detection performance and the effects of the unfolding arrangement and the pre-processing are tested. Fault diagnosis is covered using Support Vector Machines (SVM) as classification method. The study shows that BDPCA is better than MPCA in terms of diagnosis. Furthermore, monitoring and diagnosis performances seem to improve by dividing the process into batch and fed-batch stages.