In every new car project, the material mix is defined during the development phase. With respect to the car body, the definition of the material mix concerns different metal grades and thicknesses, as well as coatings. With an ever-increasing number of new metal grades and thinner sheet metal thicknesses, the possibilities for selecting and defining materials according to thickness for multi-sheet joints are increasing. Production planning engineering is responsible for ensuring producibility. While a considerable number of multi-sheet joints are well known with respect to their weldability, new material combinations in terms of alloys, thicknesses, and coatings frequently appear, so they must be checked for suitability for spot welding. Checking for weldability requires considerable experimental effort and typically necessitates the procurement of third-party labs.

Based on an existing dataset containing information on material combinations, physicochemical and technical properties, material thickness, coating presence/absence, and linking this with weldability information (yes/no), a labelled dataset was created. This feature vector comprises 176 parameters and contains all the information required to train a machine learning model that can predict the weldability of new material combinations with 99.9% accuracy. The core module (the AI/machine learning module) is part of an existing application with a web front end that allows users to import lists of welding points for analysis. This initial list is then automatically transformed into a material-thickness combination list, to which the machine learning model is applied. The result, indicating whether or not the welding points can be welded, is then provided as output. The entire workflow is controlled via the web front end, including the transformation of the various data sets and the execution of the machine learning model.