Scholarlink

<p>Due to the inherently complex and nonlinear relationship between process parameters and layer geometry in wire-arc directed energy deposition (DED), accurately predicting deposition layer geometry and optimizing process parameters remain significant challenges. This study proposed a forward and reverse control framework based on machine learning and Non-dominated Sorting Genetic Algorithm II (NSGA-II), which was capable of forwardly predicting layer geometry and reversely optimizing process parameters to improve forming quality. Six machine learning models were trained and tested through experimental data, and the backpropagation neural network (BPNN) was selected to ensure the accuracy of the multi-objective optimization. The framework’s feasibility was validated through experiments. The reverse prediction of process parameters demonstrated that the actual layer geometry closely matched the target values, with the mean absolute percentage errors (MAPE) of 3.3% for layer height and 2.21% for layer width. By incorporating interlayer temperature (100–250&#xa0;°C) into the model, the BPNN achieved high predictive accuracy on the test set, with a MAPE of 3.07% for layer height and 3.57% for layer width. Finally, the effectiveness of temperature parameter coupling optimization was verified through dynamic adjustment to process parameters. The experimental group reduced the thermal accumulation effect. The maximum deviation of surface flatness was reduced from 3.21&#xa0;mm in the control group to 0.78&#xa0;mm in the experimental group. Additionally, the interlayer bonding was robust, with no observable defects. These results confirm the potential of the proposed intelligent control system to enhance process stability and form quality in DED.</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

Machine learning-assisted layer geometry prediction and multi-objective optimization in wire-arc directed energy deposition

  • 期刊论文

  • 发表日期：

    2026年7月16日

查看全文
  • Lei Wang,
  • Kui Zhang,
  • Xiaotian Zhang,
  • Xiaopeng Li,
  • Yong Peng,
  • Yong Huang,
  • Kehong Wang

摘要

Due to the inherently complex and nonlinear relationship between process parameters and layer geometry in wire-arc directed energy deposition (DED), accurately predicting deposition layer geometry and optimizing process parameters remain significant challenges. This study proposed a forward and reverse control framework based on machine learning and Non-dominated Sorting Genetic Algorithm II (NSGA-II), which was capable of forwardly predicting layer geometry and reversely optimizing process parameters to improve forming quality. Six machine learning models were trained and tested through experimental data, and the backpropagation neural network (BPNN) was selected to ensure the accuracy of the multi-objective optimization. The framework’s feasibility was validated through experiments. The reverse prediction of process parameters demonstrated that the actual layer geometry closely matched the target values, with the mean absolute percentage errors (MAPE) of 3.3% for layer height and 2.21% for layer width. By incorporating interlayer temperature (100–250 °C) into the model, the BPNN achieved high predictive accuracy on the test set, with a MAPE of 3.07% for layer height and 3.57% for layer width. Finally, the effectiveness of temperature parameter coupling optimization was verified through dynamic adjustment to process parameters. The experimental group reduced the thermal accumulation effect. The maximum deviation of surface flatness was reduced from 3.21 mm in the control group to 0.78 mm in the experimental group. Additionally, the interlayer bonding was robust, with no observable defects. These results confirm the potential of the proposed intelligent control system to enhance process stability and form quality in DED.