A Study on High Pressure Die-Casting Defect Prediction Deep Learning Algorithm for Porosity Defect Detection based on Process Parameters and Thermal Image

Abstract

Existing analysis methods have limitations in identifying the exact cause of defects because several variables cause defects in a complex manner in the high-pressure die-casting process. However, as data processing speeds increase and analysis technologies advance, research activities are progressing on techniques to analyze complex manufacturing processes. In this study, numerical and image data were collected for the main variables that cause porosity defects in the die-casting process. Based on this, we intend to design a failure prediction algorithm using the HP-GAN(Hypothesis Pruning Generative Adversarial Network) algorithm and verify the algorithm. The HP-GAN algorithm is a combination of CNN(Convolutional Neural Network) and GAN algorithms. The raw data used in HP-GAN are line data derived from the die-casting equipment PLC and thermal images taken before and after spraying on the mold work surface through a thermal imaging camera. data, porosity defect data. To strengthen the algorithm, we used the Mean Squared Error (MSE) formula and the Gradient Decent Algorithm (GDA) to modify the weights of the algorithm to increase the prediction accuracy.