Cross-process-chain detection of material and process anomalies in imbalanced data for technical plastic assemblies

The goal of the ProData project is to analyze cross-processacross injection molding and assembly processes to identify anomalies and determine the causes of errors, serving as the basis for “hands-on” development of data competencies in the field of injection molding and the transfer of these competencies to the next generation of scientists. The focus here is on working with real production data from series production and, in particular, on strategies for handling imbalanced data (an “imbalanced data problem” occurs whenever one class in the dataset (here: defect-free components) is much more heavily represented than the other (here: defective components)), since real-world application scenarios often involve highly uneven class distributions due to high repetition rates and automation technology. To this end, a real-world process chain was established between the Department of Plastics Engineering at the University of Kassel (injection molding machines) and the RIF Institute for Research and Transfer e.V. (assembly station), whose machine, process, material, and test data are being evaluated across process chains for the first time. The RIF subproject addresses, from a data-driven perspective, the establishment of a shared data analysis environment, the integration of the available data, and the execution and coaching of heterogeneous data analyses—both for individual process steps and across process chains.