A Machine learning-based model to predict residual stress in aluminum shell formed by shot peening

Abstract

Uncertainties and errors caused in the experimental procedure and finite element modeling (FEM) of the shot peening can impact the residual stress (RS) magnitude and distribution significantly. In the present work, a machine learning-based model is used to predict the RS distribution in an Al 2024 shell formed by the shot peening process. The experimental test is performed to measure the induced RS at three points around the center of the shell. FEM is performed to capture the RS diagram considering single-shot and multi-shot scenarios. FEM validation with experimental results is also carried out. In the next step, K-nearest neighbors (KNN), random forest (RF), and XGBoost algorithms predicted the RS profile considering data with 0%, 5%, 10%, and 15% noise. The results show that the KNN algorithm indicates the highest accuracy in estimating the location and value of the maximum negative residual stress (MNRS), which is about 97.6%. However, this model is influenced by the applied random noise and cannot estimate the RS profile correctly. On the other hand, although the RF model has a 5% higher mean error in predicting the value and location of the MNRS, it has accurately forecasted the RS diagram.