Investigation of the sensitivity of pattern recognition techniques for improving batch reactor performance

This study investigates the use of pattern recognition techniques such as neural networks, to diagnose the potential occurrence of problems in batch reactor product quality. The ability to diagnose small changes in heat transfer coefficients and lower reactions rates in a batch reactor that is under PI control is demonstrated. In this manner it is shown that potential occurrence of these problems can be detected and provide early warning before any significant change in product quality such as conversion is detected.