A study on the construction of die-casting production prediction model by machine learning with Taguchi methods

Abstract

ABSTRACT Technologies such as machine learning, big data, and Industry 4.0 have become the trends in the development of science and technology in various countries in recent years. This research hopes to establish a predictive model through data analysis to help die-casting plants determine whether there are defects in the castings and improve the production competitiveness of domestic die-casting plants. Data was taken from the domestic automobile industry and used actual production data as the basis for analysis. In this study, relevant parameters of die-casting manufacturing as independent variables were chosen and determined whether there were defects in the castings as strain numbers. Afterward, the researchers constructed Artificial Neural Network, Support Vector Machines, and Random Forests as three prediction models. Three prediction models with the Taguchi Methods are used to find the best parameter configuration of each model. AUC (Area Under Curve)- Receiver Operating-Characteristic (ROC) evaluates the strength and weaknesses of the three models and, in the end, finds the most suitable network prediction model.