Crack Growth Prediction Models for a Pre-defined Semi-elliptical Crack Embedded in a Cantilever Bar Using Supervised Machine Learning Algorithms

Any machine component or structure can fracture due to the presence of cracks. With the assistance of finite element tools, we can only dissect the stable crack growth that requires much computational time and is vulnerable. This work developed several ML models using supervised machine learning algorithms and compared their performance. These models have shown decent precision in detecting the crack growth behavior of a pre-defined semi-elliptical crack embedded in a cantilever bar. The correlation coefficient R squared (R²), mean squared error (MSE), root mean squared error (RMSE), and mean absolute error (MAE) were used to evaluate and compare the performance of the developed ML models. The accuracy of the crack growth forecast is found to be ~ 86.47%, ~ 93.68%, ~ 91.50%, ~ 92.04%, and ~ 94.64% for linear regression (LR), quadratic polynomial regression (QPR), decision tree (DT), random forest (RF), and k-nearest neighbor (KNN), respectively; among them, KNN had the best prediction accuracy.