Control-Oriented Gray-Box Modeling for Thermoset Injection Molding

Cavity pressure control can enhance the repeatability of injection molding processes. While extensive research has focused on thermoplastic cavity pressure control, there is a notable gap in models and control strategies for thermoset injection molding. This study aims to develop a model structure for thermoset injection molding suitable for integration into a model-based control scheme. The modeling approach is intended to be as generalizable as possible and sufficiently flexible to adapt to various process conditions. At the same time, it should be easy to parameterize or to train. To address this challenge, we first derive a first-principles process model. In the second step, we integrate a feed-forward artificial neural network into this model, which learns parameters and source terms from past injection molding cycles, resulting in a gray-box model. The neural network outputs replace the initial model parameters with functions of system inputs, states, and time. We validate both models against experimental data from a thermoset injection molding machine using a fat-plate mold geometry and a phenolic resin compound. We identify limitations of the proposed approach and suggest potential solutions.