Artificial Intelligence-Based Error Detection and Adaption of Shot Curve Parameters in the Hot Chamber High-Pressure Die Casting Process

The high-pressure die casting process has become one of the most important production processes today due to recent developments in the automotive industry. However, there are still large quantities of defective parts and waste of raw material, especially during the first cycles in manufacturing. In that case, it is necessary to identify defective parts and combine them with relevant process parameters to improve the part quality. To influence the process parameters, there is a very short time frame of a few milliseconds. Through this research a newly developed methodology is capable to visually identify quality-relevant process parameters at the die casting shot curve. A supervised learning algorithm helps to determine the part quality under assessment of the porosity by Artificial Intelligence. For this purpose, an adapted Convolutional Neural Network [CNN] evaluates the acquired end-of-line X-ray images of the generated casting parts. Parts are labelled according to their degree of porosity. A further step correlates the porosity degree with the real-time machine time-series data sets under the use of an adapted ResNet18-Convolutional Neural Network. The CNN is capable of inferencing the part-porosity based on the acquired process datasets. Captured data sets are converted into shot-curve images that are assessed by Artificial Intelligence. Responsible areas for insufficient part quality are identified under the use of an occlusion sensitivity map. This method allows to manually adjust the die casting process parameter settings, with prediction accuracies of up to 80 % achieved in this research.