Weld Seam Trajectory Planning Using Generative Adversarial Networks

Reliable electricity transmission in battery cells and modules is indispensable for energy storages. However, common joining technologies for such devices such as bolting or soldering suffer from several drawbacks, including force-dependent resistance or low dynamic strength. Laser beam welding shows potential to overcome these disadvantages. Besides excellent joint properties, it is applicable to small assembly spaces and has potential for the implementation of lightweight construction. In addition, laser beam welding allows users to precisely adjust the weld seam’s electrical conductivity and mechanical strength by an adaption of the weld seam trajectory. For industrial purposes, low costs and short development cycles are crucial. These short development cycles require a fast and easy design-to-production process. Therefore, an adapted Machine Learning method (Generative Adversarial Networks) is presented to simplify and accelerate the weld seam trajectory planning for laser beam welding. The algorithm predicts a suitable weld seam trajectory to achieve the desired electrical conductivity and tensile strength. For the algorithm used, feasibility was demonstrated using a dataset of the Modified National Institute of Standards and Technology (MNIST) database.