A Modular Edge-/Cloud-Solution for Automated Error Detection of Industrial Hairpin Weldings using Convolutional Neural Networks

The traction battery and the electric motor are the most important components of the electrified powertrain. To increase the energy efficiency of the electric motor, wound copper wires are being replaced by coated rectangular copper wires, so-called hairpins. Hence, to connect the hairpins conductively, they must be welded together. However, such new production processes are unknown compared with classic motor production. Therefore, this research aims to integrate Industry 4.0 techniques, such as cloud and edge computing, and advanced data analysis in the production process to better understand and optimize the manufacturing processes. Welding defects are classified with the help of a convolutional neural network (CNN) (predictive analysis) and, depending on the defect, a recommended course of action for reworking (prescriptive analysis) is given. However, the application of such complex algorithms as neural networks to large amounts of data requires huge computing resources. Therefore, a modular combination of an edge and cloud architecture is proposed in this paper. Furthermore, a pure cloud solution is compared with the edge solution.