A Hybrid Physical-Data-Driven Approach for Residual Stress Prediction in Peening Scenario  
Author Butong Li


Co-Author(s) Junjie Zhu; Xufeng Zhao


Abstract Studying the distribution and history of residual stresses through different laser impact strategies is essential to assess alloy components' structure, failure mechanisms, and alloy properties. Commercial modelling methods, including finite element methods, have been introduced to predict residual stress, while the analysis is computationally expensive due to complex engineering conditions. Here we combine the machine learning algorithm and an experimentally validated efficient FEM model to predict the residual stresses of engineering peening process. Multiple machine learning algorithms were used to build the predictive models, including ensemble learning and neural networks-based algorithms. Subsequently, the ensemble learning algorithms are able to make highly accurate predictions with different process parameters. Moreover, data-based machine learning methods are an excellent alternative to FEM models in view of the significant reduction in computational cost.


Keywords Ensemble learning, residual stress, hybrid data-driven model, finite element method
    Article #:  RQD28-135

Proceedings of 28th ISSAT International Conference on Reliability & Quality in Design
August 3-5, 2023