High-pressure die casting is a common industrial process for the production of complexly shaped products made of light metal alloys. The recording of various process data during production is more or less state of the art. Parts-related quality data is also recorded digitally at some foundries. However, the added value of this data has so far been very low. In this context, machine learning offers the possibility of training prediction models with this data and, for example, predicting the quality of cast components during production almost in real time based solely on recorded process data. This has the potential, for example, to reduce labour-intensive and cost-intensive quality inspections, create more confidence in the company's own products and increase process transparency.
The aim of this thesis is to determine the capabilities of machine learning prediction models for predicting component properties (strength, ductility, surface properties, weldability) in aluminium high-pressure die casting. For this purpose, a literature research on the basics of machine learning and aluminium high-pressure die casting is to be carried out first. In the experimental part, tests are to be carried out to determine the properties of die-cast components. Prediction models (classification and regression models) are to be trained using machine learning on the basis of the quality data determined and the associated component-related production data. Finally, the capabilities of the various forecasting models are to be critically assessed on the basis of selected evaluation metrics.
