Scholarlink

<p>The machining performance of nickel-based superalloys, such as IN625, is affected by thermal gradients and cutting forces, leading to tool wear, poor surface quality, and dimensional inaccuracy. A physics-informed data-driven approach using ensemble and transfer learning is proposed to predict temperature fields and cutting forces. Finite element (FE) simulations were used to generate data for training machine learning (ML) models, including AdaBoost, SVR, Random Forest, and others. The best set of hyperparameter values for each ML model are derived by the grid search method with tenfold cross-validation. The estimated statistical parameters suggest that the AdaBoost regression model performed best, with 99.89% prediction accuracy for cutting temperature prediction on the training data, followed by Stochastic Gradient Boosting (99.69%), Random Forest (98.72%), and Gaussian Process Regression (98.57%). For cutting force prediction, the SVR model outperformed others with 100% prediction accuracy on the training data. On the validation data, AdaBoost demonstrated the best performance for predicting cutting temperature, and the Gaussian Process Regression model showed the best performance for predicting cutting force, with an average error of 4% and 7% between the model predictions and their experimental values, respectively. The results of feature importance analysis from various machine learning models formulated to predict cutting temperature and cutting force are aligned with the theoretical significance of the cutting variables on respective cutting responses. Accordingly, the presented results demonstrated the successful integration of FE model-based knowledge and prediction capability of ML approaches for the estimation of thermal fields and cutting forces in machining IN625 superalloy. Therefore,&#xa0;the presented architecture can be effectively implemented to derive the optimal machining conditions and also integrated with computer-aided manufacturing software as the cutting-edge technology for real-time process prediction and suggest the optimal machining conditions in the present smart manufacturing era&#xa0;in line with the emerging Industry 4.0 strategy.</p>

错误：搜索内容不能为空，请输入英文关键词
错误：关键词超出字数限制，请精简
高级检索

Physics-informed data-driven ensemble and transfer learning approaches for prediction of temperature field and cutting force during machining IN625 superalloy

  • 期刊论文

  • 发表日期：

    2025年3月12日

查看全文
  • Mondi Rama Karthik,
  • Thella Babu Rao

摘要

The machining performance of nickel-based superalloys, such as IN625, is affected by thermal gradients and cutting forces, leading to tool wear, poor surface quality, and dimensional inaccuracy. A physics-informed data-driven approach using ensemble and transfer learning is proposed to predict temperature fields and cutting forces. Finite element (FE) simulations were used to generate data for training machine learning (ML) models, including AdaBoost, SVR, Random Forest, and others. The best set of hyperparameter values for each ML model are derived by the grid search method with tenfold cross-validation. The estimated statistical parameters suggest that the AdaBoost regression model performed best, with 99.89% prediction accuracy for cutting temperature prediction on the training data, followed by Stochastic Gradient Boosting (99.69%), Random Forest (98.72%), and Gaussian Process Regression (98.57%). For cutting force prediction, the SVR model outperformed others with 100% prediction accuracy on the training data. On the validation data, AdaBoost demonstrated the best performance for predicting cutting temperature, and the Gaussian Process Regression model showed the best performance for predicting cutting force, with an average error of 4% and 7% between the model predictions and their experimental values, respectively. The results of feature importance analysis from various machine learning models formulated to predict cutting temperature and cutting force are aligned with the theoretical significance of the cutting variables on respective cutting responses. Accordingly, the presented results demonstrated the successful integration of FE model-based knowledge and prediction capability of ML approaches for the estimation of thermal fields and cutting forces in machining IN625 superalloy. Therefore, the presented architecture can be effectively implemented to derive the optimal machining conditions and also integrated with computer-aided manufacturing software as the cutting-edge technology for real-time process prediction and suggest the optimal machining conditions in the present smart manufacturing era in line with the emerging Industry 4.0 strategy.