Inferential-learning control of quality properties in semi-batch reactors

A strategy for controlling product quality properties, subject to batch-to-batch and within batch disturbances is presented. The methodology is based on readily available measurements, information extracted from existing databases and a few simple identification experiments. This approach extends the mid-course correction control strategies to overcome model error when the process is affected by constant, correlated and uncorrelated disturbances. Adaptive PLS (Partial Least Squares) methods are used to extract the information and to update the models used for prediction and control. The methodology is illustrated with a few emulsion polymerisation examples.