Machine learning-based prediction of quality characteristics in joining processes to minimize destructive testing: A case study on crimp connections

In quality control of joining processes, destructive tests are used to verify that materials and components meet certain quality requirements. However, due to their destructive nature, these tests are costly and can only be carried out on a sample basis, which means that indirect quality control methods are required for the produced parts in between these samples. This prevents a direct assessment of the quality of each part produced. This work presents a machine learning-based approach to consistently predict quality characteristics normally determined by metallographic analysis and mechanical testing across the entire production batch. In a case study focused of crimp connections, supervised learning methods are applied to predict quality characteristics such as crimp height and pull-out force from the force-displacement curves generated during the crimping process. The method improves transparency of quality control and achieves deviations as low as 0.06% relative to specific measured quality characteristics. Additionally, it can be transferred to other joining processes such as welding or soldering.