Using Machine Learning to Establish the Relationship between Die Casting Parameters and the Casting Quality

Abstract

The die casting process is highly automated and computerized. Nowadays, the machines are able to display dozens of data each cycle of the process with few setting parameters. The daily data remain unexploited in the industry. In this experiment, there are six setting parameters for the machine, such as cylinder pressure, high speed, high-speed switching point, intensification pressure starting position, injection delay, and biscuit thickness. The purpose of this research is to use machine learning to build two models, model 1 establishes the relationship between die casting machine setting parameters and the machine response (displayed) parameters via polynomial regression and the R square is used to evaluate the model, and model 2 is built using support vector machine algorithm, predicts the quality of die castings based on machine response parameters. The two models are then combined, and it allows the foundry men to adjust the machine parameters to improve the quality of die-casting parts. The experimental results of model 1 show that R squared greater or equal to 0.5 means that the setting parameters and the reaction parameters have a certain correlation. After cross-validation of model 2, the model is stable and the accuracy rate can reach 74%, with a small amount of data under the circumstances, it has reached the applicable standard. This research results are based on two data sets provided by diecasters A and B to establish and verify the models. `