Abstract Prediction of welding deformations using Computational Welding Mechanics in fields like shipbuilding and steel construction is limited by numerical reasons. Parallelization does not scale well with non-linear mechanical problems, therefore residual deformations on large structures could only be predicted with simplified methods. A multitude of methods have been proposed in literature but few of them have found an industrial application. In this work, a method is proposed which is more readily implementable in a general purpose FE code. Equivalent loads are calibrated using an inverse analysis procedure starting from reference deformation data derived from testing or accurate numerical simulations. The developed inverse analysis procedure allows speeding the calibration phase. The model has been validated on a steel connection consisting of around 140 welds. Preliminary results have been reported. It was found that the direction of deformation is predicted as well as the order of magnitude. The realized assembly includes scatter in the welding parameters due to manual welding and also some exception in the number of runs utilized. Because of the FE model is linear elastic, the welding sequence could not be optimized. However, production problems could be identified beforehand, and clamps be introduced at the expense of increasing residual stresses.