Scholarlink

<p>The formation of the weld interface during magnetic pulse welding is a critical factor influencing the mechanical properties of solid-state bonded joints. However, conventional tensile testing methods complicate the engineering workflow and introduce variability into the results. In this study, we address the challenge of predicting the mechanical properties of welded joints between 1&#xa0;mm thick A1060 aluminum and T2 copper. We propose two distinct research methodologies to tackle this issue: one involves an automated weld seam extraction technique, while the other integrates long short-term memory networks with attention mechanisms and convolutional neural networks. The primary highlights of these approaches are as follows. The automatic weld seam extraction system employs an image processing module equipped with autonomous measurement capabilities to quantitatively characterize weld seam dimensions through digital image analysis, which subsequently serves as input for predictive modeling. The predictive model utilizes convolutional neural networks to extract features, long short-term memory networks to account for input sequences to model temporal information, and an attention mechanism to emphasize critical data points. The validity of this methodology was assessed through a validation framework that included computational simulation modeling and experimental device construction. Finally, the developed model was interpreted using the DeepShap method to analyze the impact of each feature on model output and explore interactions between different features affecting prediction results.</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

Prediction model for 1 mm thickness A1060 aluminum and T2 copper magnetic pulse welding joints based on CNN-LSTM-AM deep learning algorithm and DeepSHAP interpretability analysis

  • 期刊论文

  • 发表日期：

    2025年10月11日

查看全文
  • Mingjian Yuan,
  • Gongcheng Peng,
  • Ruru Xiao,
  • Bing Zeng,
  • Wenxuan Xu,
  • Guangyao Li,
  • Tianyu Gao,
  • Junjia Cui,
  • Hao Jiang

摘要

The formation of the weld interface during magnetic pulse welding is a critical factor influencing the mechanical properties of solid-state bonded joints. However, conventional tensile testing methods complicate the engineering workflow and introduce variability into the results. In this study, we address the challenge of predicting the mechanical properties of welded joints between 1 mm thick A1060 aluminum and T2 copper. We propose two distinct research methodologies to tackle this issue: one involves an automated weld seam extraction technique, while the other integrates long short-term memory networks with attention mechanisms and convolutional neural networks. The primary highlights of these approaches are as follows. The automatic weld seam extraction system employs an image processing module equipped with autonomous measurement capabilities to quantitatively characterize weld seam dimensions through digital image analysis, which subsequently serves as input for predictive modeling. The predictive model utilizes convolutional neural networks to extract features, long short-term memory networks to account for input sequences to model temporal information, and an attention mechanism to emphasize critical data points. The validity of this methodology was assessed through a validation framework that included computational simulation modeling and experimental device construction. Finally, the developed model was interpreted using the DeepShap method to analyze the impact of each feature on model output and explore interactions between different features affecting prediction results.