Interpolating CTS specimens’ mode I and II stress intensity factors using artificial neural networks

• Trained with numerical results, artificial neural networks were capable of stress intensity factor prediction. • The developed solution reduced the root mean squared error by 90% when compared to analytical solutions. • Network performance increases with the number of hidden layers and total number of neurons. • The network requires at least 30 neurons evenly distributed in all its hidden layers to capture the data behaviour. • Backpropagation algorithms can vary in effectiveness depending on the total number of neurons in the network. Fracture mechanics parameters, such as the stress intensity factor (SIF), are fundamental for the analysis of fracture, fatigue crack growth and crack paths. SIFs of a cracked body can be determined either experimentally or numerically. Analytical solutions of SIF are very useful, but their determination from discrete values can be extremely complex when there are many independent variables. In this paper, artificial neural networks (ANN) are proposed to predict mode I and II stress intensity factors in a CTS specimen under mixed mode loading conditions. Trained with numerical data, the performance of different network architectures and backpropagation algorithms was assessed. Using at least 10 neurons, in the hidden layers, made it possible for the designed solution to match the performance of analytical solutions. Increasing the number of neurons, allowed the model performance to improve up to 90%, when compared with previous analytical solutions. This increases the quality of fracture and fatigue studies done with the CTS sample.