Abstract

Fatigue assessment of notched components requires knowledge of elastic-plastic stress-strain responses at notch areas. Traditional elastic-plastic finite element analysis (FEA) is not computationally efficient, and approximation methods are not highly accurate. Therefore, the present study uses the integration of an artificial neural network (ANN) and finite element (FE) to predict elastic-plastic stress and strains at notch locations on the basis of an elastic FEA solution. To this end, two different Finite Element (FE) models were developed to generate hypothetical elastic and elastic-plastic stress and strain datasets for different hardening materials and load levels. The first FE model was based on an elastic deformation state, while the second one was under an elastic-plastic deformation state. The elastic stress data obtained from the elastic FE model was used as the input data, and the elastic-plastic stress-strain data from the nonlinear elastic-plastic FE model was utilized as the output data. Subsequently, the ANN was trained for the dataset to establish a relationship between the input and output data. The dataset fed to the ANN model was then divided into three groups: training, verification, and testing. The training data was used for the ANN learning algorithm to attain the desired accuracy by obtaining the hyperparameters of the model. Subsequently, the verification data was employed to evaluate chosen hyperparameters and make adjustments of the hyperparameters to determine optimum results, and in the last stage, the generalizability of the model was examined by the testing data. The predicted stress-strain results showed that the developed ANN model is able to accurately and efficiently predict elastic-plastic stress and strain for the notched body using only the elastic FEA solution. The developed ANN-FE methodology could efficiently estimate elastic-plastic stresses and strains for notch bodies with varying material hardening properties and load levels.