Part Geometry and Parameter-Based Anticipatory Failure Investigation in Rotational Molding: A Hybrid Machine Learning Algorithm

Rotational molding (RM) is a widely used plastic manufacturing pro-cess that presents various advantages for producing complex and seamless prod-ucts. However, determining the optimal production parameters remains challeng-ing as this often relies on trial and error and expert intuition. This study introduces a hybrid machine learning algorithm to systematically identify and optimize process parameters for RM by leveraging historical production data and shape similarity analysis. The developed method integrates hierarchical clustering to classify new products based on shape similarity using histogram of oriented gradients for fea-ture extraction. Subsequently, an artificial neural network algorithm predicts the possibility of having failures based on historical RM data. The proposed approach was validated through two case studies demonstrating high accuracy in classifica-tion and failure prediction. The silhouette coefficient validated the robustness of the clustering, ensuring reliable shape classification while the ANN achieved an accuracy of 98.32%, significantly reducing dependency on manual expertise. The proposed hybrid approach streamlines production planning, preventing failures, and enhances operational efficiency in RM by automating parameter optimization based on objective historical analysis.