Methodologies Developed for Dataset Preparation and the Interpretability of Machine Learning Algorithms Used for the Prediction of Crack Growth Rate

The recent development of systems with high computational capabilities has made data-driven methods increasingly useful also in the material science field. In this study, the information from existing literature on fatigue crack growth behaviour in additive manufactured components was used within the realm of machine learning methods. The crack growth behaviour in additive manufactured alloy is complex, and analytic models, such as the Paris law, don't consider the multiple influence factors. At first, we provide a methodology for dataset preparation that consists of converting the data of each feature into a format suitable for the machine learning algorithms. After the input features selection and data cleaning, several algorithms were trained using the experimental data of different types of materials. To enhance model interpretability, the Shapley values method was employed to analyse feature contributions. From an engineering perspective, the objective is to predict all points of a complete curve da/dN - ΔK, rather than isolated data points. The performance of the models was evaluated, and the most promising algorithms were identified for predicting crack growth rate in additive manufactured materials.