Feature selection based on visual analytics for quality prediction in aluminium die casting

Digital transformation is regarded as a key enabler for quality management in the modern casting industry. By uncovering parameters that affect the casting part quality, a continuous process improvement could be enabled. In particular, a focus is laid on the explicability of the identified dependencies due to the acceptance of data-driven decisions by workers as well as to initiate tailored improvement measures. For this purpose, in this paper two visual analytics approaches are discussed for uncovering quality-affecting parameters with the claim for a high degree of explicability - on the one hand a method-feature-heatmap that visualizes a multiple backward feature elimination and on the other hand a graph-based visualization of class association rules. An automotive industry use case demonstrates the approaches in the context of gravity die casting of aluminium knuckles.