Artificial neural network in prediction of mixed-mode I/II fracture load

•A data-driven approach is presented for mixed mode fracture prediction.•388 experimental data is collected and used to train and test the ANN models.•ANN models are optimized with LM, BR and BFGS algorithms.•The relative importance of ANN input parameters is investigated.•All three models of ANN-LM, ANN-BR, and ANN-BFGS are shown to be successful. The current research demonstrates the application of artificial neural network (ANN) in predicting the fracture under mixed-mode I/II loadings. To this end, based on the analysis of the relative importance of different input factors, the crack parameters of modes I and II stress intensity factors (KI and KII), T-stress, mode I fracture toughness (KIc), and ultimate tensile strength (σu) are selected and used as input data to the ANN model. Subsequently, a large number of empirical data are used, different ANN models are trained and built with the help of Levenberg-Marquardt (LM), Bayesian regularization (BR), and Broyden-Fletcher-Goldfarb-Shanno (BFGS) optimization algorithms. Finally, the trial-and-error procedure is used to determine the optimal number of hidden layers and neurons. The amounts of mean absolute percent error (MAPE) for the optimized models with LM, BR, and BFGS algorithms are equal to 8.4%, 5.1%, and 6.3%, respectively. All three models (i.e., ANN-LM, ANN-BR, and ANN-BFGS), estimate the new testing data successfully with approximately 91%, 95%, and 93% accuracy, respectively. This paper shows the effectiveness and the potential wide range application of the data-driven based fracture predictions in comparison with the traditional physics-based criteria. [Display omitted]