Prediction of Mixed-Mode I/II Fracture Load Using Practical and Interpretable Machine Learning Method

The purpose of this study is to create a useful and easy-to-understand machine learning-based model for predicting mixed-mode I/II fracture load. To this end, a database composed of specimen test configuration, material, ultimate stress, thickness, crack length, crack angle, mode I fracture toughness, mode I stress intensity, mode II stress intensity, T-stress, and mixed-mode I/II fracture load was used for the training and testing. As the problem is greatly dimensional, Gaussian Process Regression technique was chosen and optimized. The effect of variability in the input space on the output response was characterized using Monte Carlo technique. At the same time, the impact of the size of training set on the effectiveness of prediction model was also pointed out. The coefficient of determination (R2), Mean-Absolute-Error (MAE), and Root-Mean-Squared-Error (RMSE) were used as quality metrics during learning and determining the most suitable prediction model. Finally, the prediction capability was investigated and discussed. Besides, uncertainty investigation was conducted to quantify the confidence interval. For practical application, a Graphical User Interface (GUI) was developed and provided for interested users. This study showcases the effectiveness and wide-ranging applicability of data-driven fracture predictions, in contrast to conventional physics-based criteria.