Combination of independent component analysis and multi-way principal component analysis for batch process monitoring

Multi-way principal component analysis (MPCA) has been successfully applied to the monitoring of batch and semi-batch process in fine chemical and biochemical industry. However, traditional MPCA is based on the assumption that the separated latent variables must be subject to Gaussian distribution, which sometimes cannot be satisfied. In the present work, a new method combined Independent Component Analysis (ICA) and Multi-way Principal component (MPCA) approach is proposed without assuming that the latent variables subject to Gaussian distribution. The approach is based on ICA method that finds independent variables as linear combination of MPCA latent variables. Combined ICA and MPCA method is capable of describing non- Gaussian distributed data precisely. This algorithm is evaluated on the penicillin fermentation benchmark process and is compared to the traditional MPCA. The method has significant benefit when the data does not subject to normal distribution.