APPLICATION OF MACHINE LEARNING METHODS TO STRESS ANALYSIS OF THIN PLATES WITH HOLE

This paper analyzes the stress problem of a thin plate with holes using machine learning combined with computational mechanics, in which data-driven neural networks rely on input data and make predictions by learning patterns in the data. Physically informed neural network improves the accuracy and generalization ability by embedding the equilibrium equations. The deep energy method constructs the loss function based on the principle of minimum potential energy, which has significantly better computational efficiency and accuracy, and gives its Von Mises stress and error cloud maps under bi-directional uniform and non-uniform stretching with an error of no more than 5%. The intersection with machine learning strongly contributes to the innovation of computational mechanics research paradigm and continues to expand its depth and application scope.