A Self-Training-based Approach for Aluminum Alloy Casting Quality Prediction
Haonan Wang,Jun Wu,Quanzhi Sun,Weipeng Liu,Tao Peng,Renzhong Tang
DOI: https://doi.org/10.1109/case59546.2024.10711547
2024-01-01
Abstract:With the revolutionary advances of integrated structural components in current automotive industry, aluminum alloy castings are in a rising demand. Aluminum alloy casting quality prediction in high responsiveness and low operational cost is crucial for structural components manufacturers. Therefore, casting machine manufacturers are urged to incorporate advanced quality prediction functions in the next generation of intelligent casting machines. Acquiring ample quality inspection data is crucial for realizing such functions. However, these data are normally difficult, if not infeasible, to obtain due to practical issues, e.g. data proprietorship or privacy. To address this challenge, a self-training-based approach for aluminum alloy casting quality prediction is proposed. XGBoost is used as a base classifier in this study. This approach aims to train quality prediction model with scarce labeled data, and assigns pseudo-labels to unlabeled data that meet certain probability threshold. These unlabeled data are then selected and integrated into the labeled dataset, continuously updating the dataset and retraining prediction models. To verify the effectiveness of our approach, a casting machine manufacturer was collaborated to conduct a case study. The preliminary results demonstrate that our approach could achieve an accuracy, precision, recall and F1 score of 92%, 88%, 57%, and 0.69, respectively. The approach supports casting machine manufacturers to pre-train a casting quality prediction models with scarce labeled data, enabling timely quality defect identification.