Quality Prediction for Rotational Molding Processes Using a Bayesian Network

Rotational molding (RM) is a polymer manufacturing process that is challenging to optimize due to the complex interactions among heat transfer, material flow, and mold kinematics. As manufacturers strive to reduce scrap rates and improve efficiency, conventional trial‐and‐error approaches are becoming increasingly inadequate, motivating the development of predictive modeling methods. However, developing accurate predictive models is hindered by the scarcity of data, caused by long production cycles and costly experimental trials. Consequently, process parameter selection often relies on operator experience. This study proposes a Bayesian network (BN) framework for product‐quality prediction in RM that integrates heterogeneous information sources, including experimental measurements, simulation results, process settings, and domain knowledge. The framework was developed and validated using data from the production of Type IV hydrogen‐tank liners. By modeling relationships between process variables probabilistically, the BN provides both quality predictions and associated uncertainty estimates, making it well suited to data‐sparse manufacturing environments. Benchmarking against established machine‐learning methods shows that the BN achieves competitive predictive performance and provides the most accurate estimates of wall‐thickness variability among the models considered. The results demonstrate that BNs offer an interpretable, data‐efficient approach that combines complementary knowledge sources while delivering reliable uncertainty estimates without compromising predictive accuracy.