Scholarlink

<p>Addressing the challenges of heat input accumulation and forming size control associated with enhancing efficiency in double-wire double-arc additive manufacturing, this study proposes an external wire-assisted process. This approach dynamically suppresses overheating by introducing a cold external wire. To accurately correlate the complex process parameters—arc travel speed, double-wire feed speed, and external wire feed speed—with the resultant deposited dimensions (width and height), a bidirectional predictive model was developed using experimental data. A high-precision forward model (predicting dimensions from parameters) was developed using a Backpropagation Neural Network (BPNN), achieving average relative errors of only 2.04% (width) and 2.55% (height) within the training dataset. Additionally, an inverse model (predicting parameters from target dimensions) was formulated by optimizing the BPNN with a Genetic Algorithm (GA). Validation showed that the actual forming dimensions, determined using this inverse model, closely aligned with desired targets, exhibiting average relative errors of 2.5% for height and 6.97% for width. This advancement effectively solves the inverse problem of determining optimal process parameters for target dimensions, significantly improving the forming quality and process controllability of this efficient, low-heat-input additive manufacturing technique.</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

Bidirectional modeling of process parameters and bead dimensions in external wire-assisted double-wire double-arc additive manufacturing using a backpropagation neural network

  • 期刊论文

  • 发表日期：

    2026年3月11日

查看全文
  • Zeshi Jin,
  • Gaige Chang,
  • Liang Ke,
  • Xueli Chen,
  • Weidong Wang,
  • Dongqing Yang

摘要

Addressing the challenges of heat input accumulation and forming size control associated with enhancing efficiency in double-wire double-arc additive manufacturing, this study proposes an external wire-assisted process. This approach dynamically suppresses overheating by introducing a cold external wire. To accurately correlate the complex process parameters—arc travel speed, double-wire feed speed, and external wire feed speed—with the resultant deposited dimensions (width and height), a bidirectional predictive model was developed using experimental data. A high-precision forward model (predicting dimensions from parameters) was developed using a Backpropagation Neural Network (BPNN), achieving average relative errors of only 2.04% (width) and 2.55% (height) within the training dataset. Additionally, an inverse model (predicting parameters from target dimensions) was formulated by optimizing the BPNN with a Genetic Algorithm (GA). Validation showed that the actual forming dimensions, determined using this inverse model, closely aligned with desired targets, exhibiting average relative errors of 2.5% for height and 6.97% for width. This advancement effectively solves the inverse problem of determining optimal process parameters for target dimensions, significantly improving the forming quality and process controllability of this efficient, low-heat-input additive manufacturing technique.