An Improved Statistical Modeling Strategy by Spectroscopy for Online Monitoring and Diagnosis of Batch Processes

Recently, the use of spectroscopic techniques for online process monitoring has been introduced, which are deemed to be able to provide a rich source of chemical information about operation conditions within a process system. This paper presents an improved statistical analysis and modeling strategy using spectra data for online fault detection and diagnosis of batch processes. The general principle of the proposed method is that the systematic chemical information in the successful batches can be regarded as the linear combination of some underlying unobserved and independent source spectra and the mixing coefficients as the contribution of each source to the external spectra measurements. Accordingly, it employs independent component analysis (ICA) algorithm to separate those sources and identify their time-varying effects on observed spectra throughout the batch duration and thus formulates the statistical monitoring model. Moreover, in combination with contribution plots, the actual cause of the disturbances can be diagnosed. The proposed method yields more chemical statistical meanings, results in easy model interpretation and can be readily put into online application without data estimation. Its effectiveness is successfully illustrated when applied to a case study of a two-step conversion reaction.