Modeling and control of batch‐to‐batch polymer processes

The modeling and control of a typical cyclic polymer process, such as injection molding or thermoforming, are considered. The purpose of control is to achieve a specified product quality for a sequence of parts. First, conventional feedback controllers are compared to statistically based controllers with respect to random noise disturbances. It is shown that conventional controllers may not react quickly to load disturbances without magnifying background noise and reducing product yield. It is demonstrated that statistically based controllers are able to differentiate between noise and load disturbances, proving them superior when part quality tolerance is tight and noise level relatively large. Next, the responsiveness of conventional controllers (proportional, integral, and proportionalintegral) is compared with several statistically based controllers (CUSUM, Western Electric runs rules, and simple Shewhart) when subjected to load disturbances. Three load disturbances were modeled; steps, ramps, and sinusoids. Again, statistically based methods generally prove superior to and, at worst, comparable to conventional controllers.