A sub-space artificial neural network for mold cooling in injection molding

The applications of artificial intelligence (AI) have considerably expanded over recent years. A new class of industrial systems is beginning to evolve that incorporates using high volume data and advanced analytics to better optimize product quality while reducing energy consumption. Artificial neural networks (ANN) when combined with advanced modeling and control, begins to form an AI platform that can be further enhanced for factories of the future. This paper provides a demonstration of such initial work that can be further developed for future systems in a generic way. When considering polymer processing such as plastic injection molding, the mold cavity temperature (MCT) profile directly relates to part quality and part reject rates. Therefore, it is desirable to optimize the mold cooling process using real time control of MCT as it directly affect part quality. However, MCT is affected by a number of interacting nonlinear dynamic parameters that are often neglected due to the challenge of quantifying such parameters. Advanced model based control algorithms are often used for providing improved control of complex systems. However, they depend on good model formulations that are analytically insufficient. An online intelligent system identification approach for the mold cooling process is developed and tested. An ANN is designed to adjust online sub-space parameters that govern a mold cooling model. Results demonstrate that this online ANN approach can be used to accurately predict the dynamic behavior of mold cavity surface temperature. This is key to many industrial systems where their states are not directly observable and uncertainties are unknown. The methodology can be readily adapted for different operating conditions as in this case of polymer processing and has good potential for its integration with advanced model based control schemes and cloud computing approaches for the next generation of machines.