Physical-data-driven fretting wear rate prediction model for GH4169 alloy milling surfaces

Abstract

This study provides an insight into the fretting wear properties of the high-temperature alloy GH4169 on milled machined surfaces, a critical manufacturing step that significantly alters the surface integrity of the material and consequently affects its wear behavior. The coefficient of friction, wear volume, wear rate, and wear mechanism were analyzed for milled surfaces with different fretting parameters. A finite element simulation-based fretting wear model is constructed for untreated surfaces, and transfer learning technology is integrated to develop an efficient physical-data dual-driven wear rate prediction model. The model initially captures the wear characteristics of the alloy using simulation data and is then fine-tuned through transfer learning to adapt to the surface state after milling. The study results demonstrate that the combination of finite element simulation and transfer learning method significantly enhances the accuracy of predicting the wear properties of GH4169 alloy. Compared to the model without transfer learning, the prediction accuracy of the model with transfer learning increased to 93.77 % on untreated surfaces and up to 93.43 % on milled machined surfaces. In addition, it was discovered that milling had a limited impact on the friction coefficient of the alloy. However, it significantly modified the fretting wear mechanism and wear resistance. This offers a new perspective for understanding the influence of milling on the wear behavior of materials.