適応的実験計画法を用いた水溶性金属加工液の設計効率化

抄録

Metalworkingfluids (MWFs) are industrial lubricants used in the processing of materials such as metals, with purposes of lubrication and chip removal. They prevent the overheating and degradation of workpieces, reduce tool wear and friction, and contribute to improved production efficiency and quality. MWFs which require a wide range of performance characteristics are created by blending various raw materials. The formulation adjustment of MWF is so sensitive that even a minor change may occur the stability issues. Because these adjustments impact product performance, they require careful handling. The objective of this study was to explored improving the efficiency of the MWFs development process by applying machine learning-based experimental design. Machine learning models to predict two key physical properties of MWFs: product stability and antifoaming were constructed with four ensemble learning algorithms: Random Forest, Gradient Boosting, LightGBM, and XGBoost. Based on the predictions from these models, evaluating 20 proposal formulations from the models, two samples in the proposal achieved all goals. Furthermore, the one of the two samples showed the best antifoaming index in all training samples.