Effect of Training Data Ratio and Normalizing on Fatigue Lifetime Prediction of Aluminum Alloys with Machine Learning

It is critical to evaluate the estimation of the fatigue lifetimes for the piston aluminum alloys, particularly in the automotive industry. This paper investigates the effect of different normalization methods on the performance of the fatigue lifetime estimation using Extreme Gradient Boosting (XGBoost), as a supervised machine learning method. For this purpose, the dataset used in this study includes various physical and experimental inputs related to an aluminum alloy and the corresponding fatigue lifetime outputs. Furthermore, before fitting the XGBoost model, different fatigue lifetime preprocessing methods were utilized and evaluated using metrics such as Root Mean Square Error (RMSE), Determination Coefficient (R2), and Scatter Band (SB). The results indicate that modeling fatigue lifetime with logarithmic values as a preprocessing method excels when XGBoost is trained with 100% of the data. However, other normalization methods demonstrate superior accuracy in estimating test data with a 20% test and 80% train set split.