Pseudo-sample based contribution plots: innovative tools for fault diagnosis in kernel-based batch process monitoring

• This article explores the potential of Kernel-Principal Component Analysis (K-PCA) for batch process monitoring
• The idea of pseudo-sample projection is exploited for diagnostic purposes
• The proposed approach is found to enable a better fault diagnosis than bilinear ones when dealing with non-linear batch data
• It may also represent a valid alternative to model batch processes, whose physics and/or chemistry are not completely known

This article explores the potential of kernel-based methods for fault diagnosis in batch process monitoring by combining Kernel-Principal Component Analysis and three common techniques which permit analyzing batch data by means of bilinear models: variable-wise unfolding, batch-wise unfolding and landmark feature extraction. Gower's idea of pseudo-sample projection is exploited to develop novel tools, the pseudo-sample based contribution plots, for diagnostic purposes. The results show that, when the datasets under study are affected by severe non-linearities, the proposed approach performs better than classical ones.