| 注册
首页|期刊导航|表面技术|机器学习驱动激光熔覆Stellite-6熔覆层工艺参数设计:显微硬度与耐腐蚀性能协同提升

机器学习驱动激光熔覆Stellite-6熔覆层工艺参数设计:显微硬度与耐腐蚀性能协同提升

韩基泰 崔宸 武美萍 缪小进 李鹏

表面技术2026,Vol.55Issue(20):131-149,19.
✕
表面技术2026,Vol.55Issue(20):131-149,19.DOI:10.16490/j.cnki.issn.1001-3660.2026.20.011

机器学习驱动激光熔覆Stellite-6熔覆层工艺参数设计:显微硬度与耐腐蚀性能协同提升

Machine Learning-Driven Process Parameter Design for Stellite-6 Cladding Layers by Laser Cladding:Synergistic Improvement of Microhardness and Corrosion Resistance

韩基泰 1崔宸 1武美萍 2缪小进 2李鹏1

作者信息

  • 1. 无锡学院 自动化学院,江苏 无锡 214105
  • 2. 江南大学 机械工程学院,江苏 无锡 214122
  • 折叠

摘要

Abstract

A data-driven framework integrating machine learning and laser cladding is proposed to achieve coordinated optimization of mechanical and electrochemical performance in Stellite-6 coatings deposited on 42CrMo steel.Unlike conventional trial-and-error parameter selection,the work aims to establish a quantitative mapping between processing parameters and coating performance by combining experimental design,statistical data augmentation,and hybrid optimization algorithms.A dataset consisting of 36 experimental conditions was constructed by systematically varying laser power(800-1 800 W)and scanning speed(350-600 mm/min),while maintaining constant powder feed rate(14 g/min)and overlap ratio(40%).Microhardness and corrosion current density were selected as dual performance indicators to capture the trade-off between strengthening and corrosion resistance.To overcome the limitation of small sample size,a multivariate Gaussian data augmentation strategy was introduced,generating 164 synthetic samples while preserving the statistical structure and physical correlations of the original dataset.Compared with linear interpolation and SMOTE approaches,Gaussian augmentation demonstrated superior capability in maintaining covariance consistency and avoiding artificial clustering,thereby improving model generalization.Four nonlinear regression models,including LSTM,BPNN,GBT,and SVM,were evaluated.Among them,the LSTM model achieved the best predictive performance,with R2 of 0.907 and RMSE of 9.65,indicating its strong ability to capture complex nonlinear interactions between process variables and coating responses.The superior performance of LSTM was attributed to its capability to model sequential dependencies and nonlinear coupling effects in limited datasets. To realize multi-objective optimization,a genetic algorithm(GA)was coupled with the trained LSTM model,enabling simultaneous maximization of microhardness and minimization of corrosion current density.Three representative parameter sets were obtained to prioritize hardness,corrosion resistance,and balanced performance,respectively.The experimentally validated prediction errors for both properties ranged from 2.3%to 5.5%,confirming the reliability of the proposed ML-driven optimization strategy.Microstructural characterization revealed that the optimized coatings consisted primarily of γ-(Co,Fe)solid solution and Cr7C3 and Cr23C6,with a refined grain structure and homogeneous carbide distribution.The evolution of microstructure was governed by laser energy density,which controlled melt pool temperature,lifetime,and Marangoni convection.Low heat input conditions promoted rapid solidification and high nucleation rates,resulting in fine equiaxed grains and dispersed carbides,thereby enhancing hardness through grain boundary strengthening and dispersion hardening.In contrast,high heat input increased dilution and prolonged solidification time,leading to grain coarsening and partial dissolution of carbides,which reduced hardness but improved corrosion resistance by forming a denser and more homogeneous passive film. A clear trade-off mechanism between hardness and corrosion resistance was identified:reduced laser energy input favored mechanical strengthening,while increased energy input enhanced metallurgical bonding and passivation behavior.The optimized coating exhibited a microhardness of 601HV0.2 and a corrosion current density as low as 2.439×10−6 A/cm2,demonstrating a significant improvement compared with the substrate.Wear tests further confirmed that the optimized coating showed reduced friction coefficient and shallower wear tracks,with the dominant wear mechanism transitioning from severe abrasive wear in the substrate to mild adhesive–abrasive wear in the coating. This work provides a systematic methodology for integrating machine learning with laser processing to achieve intelligent process design.The proposed framework not only improves prediction accuracy under limited data conditions but also enables efficient identification of optimal process windows for multi-performance targets.The findings highlight the critical role of energy density in governing microstructure and property relationships and offer a generalizable strategy for data-driven optimization of surface engineering systems.

关键词

激光熔覆/机器学习/显微硬度/电化学腐蚀/钴基合金

Key words

laser cladding/machine learning/microhardness/electrochemical corrosion/cobalt-based alloy

分类

矿业与冶金

引用本文复制引用

韩基泰,崔宸,武美萍,缪小进,李鹏..机器学习驱动激光熔覆Stellite-6熔覆层工艺参数设计:显微硬度与耐腐蚀性能协同提升[J].表面技术,2026,55(20):131-149,19.

基金项目

无锡市科技发展资金"太湖之光"科技攻关(基础研究)(K20221051)Wuxi Science and Technology Development Fund"Light of the Taihu Lake Lake"(Basic Research)(K20221051) (基础研究)

表面技术

1001-3660

访问量0
|
下载量0
段落导航相关论文