Development of a workflow to build optimal machine learning models for stress concentration factor regression

In the design of shafts for drivetrains, it is important to have precise knowledge of the effective stress in critical notches. In nominal stress approaches, stress concentration factors are used to estimate the stress in the notch based on geometric properties of the shaft. They can be calculated using numerical methods like finite element method, which can be time consuming. Analytical equations have been developed for simple geometries, like shaft shoulders and round grooves, they are less accurate but much faster than numerical solutions. In this paper, machine learning is used to combine the advantages of both solutions. A process chain to develop models for the calculation of stress concentration factors is presented. It consists of methods to process data, creation and training of regression models and evaluation of the results. This toolbox allows different regression models to be used for different tasks without the need for major changes to the source code. The process is illustrated for shaft shoulders under tension and compression, bending and torsion. The resulting model is capable of calculating stress concentration factors with better accuracy than common analytical approaches while having comparable computation time.