Closing the loop: Real-Time Error Detection and Correction in automotive production using Edge-/Cloud-Architecture and a CNN

One challenge faced by the automotive industry is the shift from combustion to electrically powered vehicles. This change strongly impacts on components such as the electric motor and the battery, and hence on production. In this context, the low level of expert knowledge is especially problematic. To meet these new challenges, this paper introduces a data-driven optimization of the production process by integrating a modular edge and cloud computing layer, and advanced data analysis. Defects are classified by a convolutional neural network (CNN) (predictive analytics) and corrected (depending on the defect type) by an automated rework (prescriptive analytics). The architecture of the CNN achieves an accuracy of 99.21% to predict the defect class. The automated rework process is selected through an implemented decision tree. The edge device communicates with a programmable logic controller (PLC) through a cyber physical interface. As an example of its practical application, the method is applied to hairpin welding of the stator of an electric motor with real production data.