Data-driven wear prediction method for complex engineering structures

Predicting the evolution of wear in metallic structural components is vital for accurately estimating the lifetime of engineering equipment. However, this remains a significant challenge due to the prohibitively large number of cycles required for traditional experiments or simulations. To address this, we established a data-driven approach to predict metal wear evolution during dynamic mechanical interactions. Our methodology involves two main steps: developing a high-fidelity finite element (FE) model to accurately simulate the wear, and then training a deep learning model that uses applied loads and historical wear data to predict future wear evolution. We selected contact wire clips in a high-speed railway system as a practical example, where the accuracy of our numerical model was successfully validated by experimental results in calculating wear distribution. The subsequent deep learning model demonstrated high accuracy ( <mml:math xmlns:mml="http://www.w3.org/1998/Math/MathML" id="m1"> <mml:mrow> <mml:msup> <mml:mi>R</mml:mi> <mml:mn>2</mml:mn> </mml:msup> <mml:mo>&gt;</mml:mo> <mml:mn>0.95</mml:mn> </mml:mrow> </mml:math> ) in predicting future wear depth at distinct positions against ground truth data. This presented approach offers a wide range of applications for predicting the wear evolution of equipment in various engineering fields.