Estimation of Spot Weld Design Parameters using Deep Learning

In automotive production, each automobile has approximately 7,000 to 12,000 spot welds along with other kinds of connections. The position of the spot weld with respect to the flange and the distance between the spot welds as well as various other parameters usually vary for each part combination (spot weld design). If these properties are known, they can be used for automatic generation of spot welds during the design phase of the product development which is otherwise a cumbersome manual process. The spot weld design to be determined by the engineer depends on many factors (input parameters) such as loads and forces that might be applied to the structure, material combination, geometry of the parts, connection technology and its process parameters. Some of these parameters such as material combination and geometry of the parts are predefined by the designer or are results of the circumstances such as loads and forces applied at the connection. The remaining parameters such as connection technology, process parameters, spot weld distances and flange distance have to be chosen by the engineer. On the basis of existing designs and with help of machine learning techniques it may be possible to predict the spot weld design parameters like spot weld distance and flange distance. Within this work existing spot-weld designs are extracted from a vast amount of FEM simulation input data available in the Simulation Data Management (SDM) system LoCo of SCALE GmbH and applied as the basis for training and benchmarking new methods for estimating spot weld parameters.