Author:

Abstract:

The control and optimization of batch processes is a challenging problem faced by (bio)chemical industry. Traditionally, (full) factorial tests are executed to investigate the effect of the manipulated variables (MV) on the (quality of the) process. Due to their nature, these tests are very time-consuming for batch processes. This paper investigates whether suitable data-driven models for batch optimization and control can be identified from a more limited set of tests. Based on the results of two case studies, it is concluded that statistical inference models can predict the final quality of batches where the MV changes occur at time points not present in the training data, provided they fall inside of the time range used for training. Furthermore, the models provide accurate predictions for batches with multiple MV changes. This is a valuable result for industrial acceptance because it implies that fewer experiments are required for model identification. © 2013 Springer-Verlag.