Dynamic model-based fault diagnosis for (bio)chemical batch processes

To ensure a constant and satisfactory product quality, close monitoring of batch processes is an absolute requirement in the chemical and biochemical industry. Principal Component Analysis (PCA)-based techniques exploit historical databases for fault detection and diagnosis of the current batch run. To handle the dynamic nature of batch processes, dedicated techniques such as Batch Dynamic PCA (BDPCA [1]) and Auto-Regressive PCA (ARPCA [2]) have been developed. In this paper, the fault detection and diagnosis performance of BDPCA and ARPCA is compared with standard multi-way PCA (MPCA [3]) on an extensive dataset of a penicillin fermentation. For MPCA, an additional batch-wise normalization improves the detection of faults on the feed rate and dissolved oxygen (DO). ARPCA clearly outperforms MPCA and BDPCA in detection speed of drifts on the aeration rate, stirrer power and DO. Drifts on the feed rate are detected slightly faster by MPCA and BDPCA. Fault diagnosis performance is comparable for each model and allows for the correct fault diagnosis in most cases.