Machine learning in prediction of residual stress in laser shock peening for maximizing residual compressive stress formation

Abstract

Laser Shock Peening (LSP) is an advanced technique for enhancing surface properties, drawing significant interest for its ability to induce beneficial residual stresses in materials. Traditional LSP design processes, reliant on manual parameter selection, often result in imprecise control over the stress distribution, necessitating multiple iterations and high costs. This study introduces a machine learning (ML)-based approach, utilizing the Random Forest (RF) algorithm, to automate and optimize the design of LSP parameters for nickel-aluminium bronze surfaces. Our findings demonstrate the RF model’s capability to accurately predict and optimize residual stress distributions, achieving compressive stresses up to 472 MPa with a notable reduction in design iterations. The model forecasts both uniform and non-uniform stress patterns, particularly identifying areas susceptible to Residual Stress Holes (RSH) with improved precision. With an Absolute Percentage Error (APE) of only 6.2 %, our approach significantly outperforms traditional ML algorithms, offering a novel method for efficiently designing complex residual stress fields in LSP applications.

Keywords

• Laser shock peening
• Machine learning
• Random forest regression
• Residual stress