Comparison of Statistical & Mechanistic Models for Batch Process Monitoring

On-line process monitoring is paramount in every industry. It protects against a variety of problems such as to minimize product quality variability, for real time release of the product to the customers, for maintenance of equipment and instruments, and sometimes even for process safety.

Two advanced methods for real time process monitoring will be presented and compared. Both methods use and exploit the information in the on-line measurements that are available around the process. The example that will be used is an industrial Batch process (Nylon-6,6), but the same methods can be applied to continuous processes as well.

The first method is based on Mechanistic models. Mechanistic models are the classical chemical engineering models that incorporate mass and energy balances, kinetics, thermodynamics, etc. These models in the form of Kalman Filters, can be used in real time with the clever use of few process measurements (temperature, pressure, etc) to predict certain states such as product quality,.

The second method is based on multivariate statistical techniques (Multi-way Principal Component Analysis, and Partial Least Squares) that analyze the information contained in the variety of on-line measurements available around a process, and compares this against the normal process behavior that has been identified from a historical database of past successful periods. This result in an efficient real time monitoring scheme.

Both methods will be compared and contrasted, highlighting their advantages and disadvantages.