Application of multiway ICA for on-line process monitoring of a sequencing batch reactor

Abstract

Multiway principal component analysis has been shown to be a powerful monitoring tool in many industrial batch processes. However, it has the shortcomings that all batch lengths should be equal, the measurement variables must be normally distributed and that future values of the current batch must be estimated to allow on-line monitoring. In this work, it is shown that multiway independent component analysis (MICA) can be used to overcome these drawbacks and obtain better monitoring performance. The on-line MICA monitoring of batch processes is based on a new unfolding method and independent component analysis (ICA). ICA provides better monitoring performance than PCA in cases with non-Gaussian data because it is not based on the assumption that the latent variables are normally distributed. The MICA algorithm does not require any estimation of future batch values and can also be applied to non-equal batch length data sets. This article describes the application of on-line MICA monitoring of a sequencing batch reactor (SBR). It is successfully applied to an 80L SBR for biological wastewater treatment, which is characterized by a variety of disturbance sources with non-Gaussian characteristics. The SBR poses an interesting challenge from the point of process monitoring characterized by non-stationary, batchwise, multiscale, and non-Gaussian characteristics. The results of the bench-scale SBR monitoring clearly showed the power and advantages of MICA monitoring in comparison to conventional monitoring methods.