Application of Symbolic Regression in Polymer Processing

Modeling and simulation is essential in polymer processing for predicting process characteristics and designing processing machines. Traditional models are based on analytical approaches. Over the last decades numerical simulation techniques have grown significantly with the rising computational power. With the ongoing digitalization the available data increased significantly and data-based modeling techniques have become popular also for production systems. Utilizing the available data powerful models, for instance, decision trees and artificial neural networks, can be trained. The prediction accuracy is strongly governed by the quality of the underlying training data. In this work, a hybrid approach is presented combining analytical, numerical and data-based approaches efficiently to overcome the limitations of the individual techniques. As a result, explicit symbolic regression models are obtained, which are optimized on the basis of a numerically derived dataset. The power of this approach is demonstrated by a selected use-case. These highly accurate models may be implemented into any further application.