Multivariate monitoring for time-derivative non-Gaussian batch process

Abstract

This research is an application of process monitoring on a pilot-scale sequencing batch reactor (SBR) using a batchwise multiway independent component analysis method (MICA) for denoising effect, which can extract meaningful hidden information from non-Gaussian data. Three-way batch data of SBR are unfolded batch wise, and then a multivariate monitoring method is used to capture the non-Gaussian and nonlinear characteristics of normal batches. It is successfully applied to an 80 L SBR for biological wastewater treatment, which is characterized by a variety of error sources with non-Gaussian characteristics. In the monitoring result, multiway principal component analysis (MPCA) can detect the abnormal batches with a false alarm rate of 47.5%, whereas MICA charts show less false alarm rate of 4.5%. The results of this pilot-scale SBR monitoring system using simple on-line measurements clearly demonstrated that the MICA monitoring technique showed lower false alarm rate and physically meaningful robust monitoring results.