Application of supervised machine learning methods in injection molding process for initial parameters setting: prediction of the cooling time parameter

The injection molding process is considered as one of the most used process in the plastics industry due to its reliability and its profitability; however nowadays, the injection industry marketplace becomes more and more competitive because of the excessive quality demand and the coast reduction requirement. Production workshops strive constantly to reduce coast and optimizing the process. One key optimization factor involves determining the optimal cooling time parameters during the initial setup phase. The cooling time parameter represents around 65% of the cycle time. The main goal of this study is to explore the implementation of various supervised machine learning methods for predicting the cooling time parameter and to compare their performance. Five algorithms, namely random forest, decision tree, KNN (K-nearest neighbors), XGBoost, and multiple regression, were employed in the analysis. The study aims to assess the effectiveness of these algorithms in predicting the cooling time parameter within the context of the injection molding process. To evaluate their efficiency, the study employed the following metrics: mean absolute error (MAE), root mean square error (RMSE), mean squared error (MSE), and mean average percentage error (MAPE). The dataset was collected from a real industrial workshop, encompassing 70 plastic components, 10 distinct material types, and 7 different types of machines. Despite the complexity and non-linearity among the process parameters, the study indicates that machine learning can still effectively capture and predict cooling time parameters. XGBoost, KNN, and random forest consistently demonstrate superior results across all metrics compared to decision tree and multiple regression, as example, the mean average percentage error (MAPE) of XGBoost is 14.76%, significantly outperforming the 23.96% MAPE associated with the decision tree. These outcomes validate that machine learning methods can play a significant role in predicting cooling time and contribute to the optimization of the overall process.