Super model-based process performance monitoring

For the assured and robust monitoring of batch processes, the issues of process dynamics and non-linearities must be addressed. Typical multivariate statistical projection based monitoring, tools such as principal component analysis (PCA) and partial least squares (PLS) are designed to deal with linear correlations between variables and thus an alternative approach is required for the modelling of batch data. One approach is through model-based multivariate techniques where mechanistic models of processes are combined with empirical modelling tools, In model-based PCA, the residuals generated from the difference between the first principle model and the data are monitored using PCA. This technique demonstrates improved fault detection ability over conventional PCA however it does not completely remove the non-linear and dynamic aspects of the data. A technique termed super-model-based-PCA((C)) (SMBPCA((C))) is proposed. It incorporates an additional residual modelling stage prior to the application of PCA. This approach exhibits superior fault detection over the other techniques examined. The technique is applied to a bench mark simulation of a 2-stage exothermic batch reactor.