Scholarlink

<p>The current study introduces a novel physics-informed machine learning (PIML) method to determine optimized process parameters (PPs) for wire arc additive manufacturing (WAAM) without requiring extensive experimental trials. The focus is on predicting and optimizing the aspect ratio (AR, defined as the ratio of bead width to bead height) for mild steel and Inconel 718. Traditional machine learning methods often overlook the underlying physics of the process, leading to inaccurate predictions. To address this, the study incorporates physics-based variables (PBVs) such as volumetric energy density (VED), solidification time (Ts), surface tension force (F), and Marangoni number (Ma) into the training of three commonly used machine learning models (namely decision tree (DT), random forest (RF) and artificial neural networks (ANNs)) to improve prediction accuracy. Among the models tested, ANN achieved the lowest average absolute relative error (AARE%) of 3.33% for mild steel and 7.97% for Inconel 718 in forward prediction, demonstrating superior predictive capability over DT and RF. A key novelty of this work is the introduction of a double backward prediction framework, where optimized PBVs were first predicted from a target AR, followed by the determination of corresponding process parameters. The model’s predictions were validated through 12 experimental trials, confirming high reliability. The study establishes a robust framework for process optimization in WAAM, reducing the reliance on trial-and-error experimentation while ensuring defect-free deposition across different materials.</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

Optimizing multi-physics variables in wire arc additive manufacturing for weld bead aspect ratio: a machine learning approach

  • 期刊论文

  • 发表日期：

    2025年4月19日

查看全文
  • Prashant Veer,
  • Deepak Mudakavi,
  • Somashekara M Adinarayanappa

摘要

The current study introduces a novel physics-informed machine learning (PIML) method to determine optimized process parameters (PPs) for wire arc additive manufacturing (WAAM) without requiring extensive experimental trials. The focus is on predicting and optimizing the aspect ratio (AR, defined as the ratio of bead width to bead height) for mild steel and Inconel 718. Traditional machine learning methods often overlook the underlying physics of the process, leading to inaccurate predictions. To address this, the study incorporates physics-based variables (PBVs) such as volumetric energy density (VED), solidification time (Ts), surface tension force (F), and Marangoni number (Ma) into the training of three commonly used machine learning models (namely decision tree (DT), random forest (RF) and artificial neural networks (ANNs)) to improve prediction accuracy. Among the models tested, ANN achieved the lowest average absolute relative error (AARE%) of 3.33% for mild steel and 7.97% for Inconel 718 in forward prediction, demonstrating superior predictive capability over DT and RF. A key novelty of this work is the introduction of a double backward prediction framework, where optimized PBVs were first predicted from a target AR, followed by the determination of corresponding process parameters. The model’s predictions were validated through 12 experimental trials, confirming high reliability. The study establishes a robust framework for process optimization in WAAM, reducing the reliance on trial-and-error experimentation while ensuring defect-free deposition across different materials.