Evaluation of a batch process by means of batch statistical process control and system identification

Abstract: Batch processes play an important role in the production of high quality specialty chemicals. Examples include the production of polymers, pharmaceuticals and formulated products. In this master thesis, the study of transformation of materials, by batch distillation and mixing is studied. The study is done by means of batch statistical process control and system identification methods in order to build soft sensors that can predict product quality and end-point but also to use the batch trajectory features for early fault detection. In contrast to a continuous process, a batch process is a finite duration process, from initialization to completion. The physical state of the process is derived from measured variables, for example, temperatures, ressures and flows and comes from on-line measurements of the on-going process. Since there are many variables, in terms of inputs and outputs, multivariate data analysis is a suitable choice for extracting systematic information which is used to find a relationship among the variables but also to visualize the batch trajectories and deviations from normal batch evolution. The results suggest that the end point can be predicted during distillation and mixing and it seems like it is possible to separate normal batches from different batches by means of batch statistical process control strategies. However, estimating product purity during distillation was not possible due to limited variation in the output data. Instead, system identification methodologies were a better choice. Product quality after mixing was poorly estimated with system identification tools due to the lack of variability within the time average of the different variables used, but was better predicted with batch statistical process control.