Machine learning-based analysis of fillet radius effects on stress distribution in torsionally loaded shafts: A review study

The geometry of fillets and especially, of fillet radius, is extremely important in managing the stress concentration and fatigue life of torsionally loaded shafts. Finite element analysis (FEA) and conventional analytical formulations are also common methods of analysis of stress concentration factors, but they are computationally intensive when studying large design spaces. Machine learning (ML) is a relatively new effective surrogate modeling technique that can be used to predict the peak von Mises stress and stress distribution with great precision. The current review outlines the developments in the recent years in the field of using ML-based methods to analyze the performance of fillet radius in torsionally loaded shafts, data generation strategies, model architectures, evaluation practices, and existing issues. The article emphasizes the fact that the geometry of fillets plays a more significant role in reducing stress in elastic environments compared to the choice of materials and presents the prospective research perspectives of physics-informed learning and uncertainty quantification.