Scholarlink

<p>Large metallic components are fabricating through wire arc direct energy deposition (WA-DED) which is most preferred method among all additive manufacturing (AM) methods due to its exceptionally high deposition rate. However, the high deposition rate introduces defects including cracking, porosity and other defects all of which significantly compromise the structural integrity and overall performance of WA-DED components. Accurate detection of these defects within the component are therefore paramount importance. Conventionally, such defects have been identified through non-destructive testing (NDT) methods. However, these NDT methods are associate with limitations such as dependency on skilled human intervention and specific kind of NDT tool or instrument for the identification of each kind of defects. These limitations overcome thorough deep learning techniques, whereas, the current research collected the acoustic signals are acquired using a high sensitivity microphone, then perform slitting using wire cut EDM of fabricated samples to identified the corresponding defects and then perform preprocessing to eliminate the noise of the acoustic signals. The resulting data is analysed by the time and frequency domain and the feature extraction by Fast Fourier Transform (FFT). The CNN and several machine learning (ML) models are trained validated using k-fold cross validation techniques. Among all evaluated models, the CNN framework outperforms, and its accuracy is 99% and the ROC-AUC value is 1.00. These results show the potential of deep learning-based acoustic monitoring applied in defect detection in real time fabrication of part through WA-DED, offering a viable alternative to post-process inspections and thereby enhancing overall manufacturing efficiency.</p> Graphic abstract <p></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

Deep learning and machine learning approaches for acoustic-based defect monitoring in wire arc direct energy deposition

  • 期刊论文

  • 发表日期：

    2026年3月27日

查看全文
  • Rupendra Singh Tanwar,
  • Suyog Jhavar,
  • Debtanay Das,
  • N Chethan Kumar

摘要

Large metallic components are fabricating through wire arc direct energy deposition (WA-DED) which is most preferred method among all additive manufacturing (AM) methods due to its exceptionally high deposition rate. However, the high deposition rate introduces defects including cracking, porosity and other defects all of which significantly compromise the structural integrity and overall performance of WA-DED components. Accurate detection of these defects within the component are therefore paramount importance. Conventionally, such defects have been identified through non-destructive testing (NDT) methods. However, these NDT methods are associate with limitations such as dependency on skilled human intervention and specific kind of NDT tool or instrument for the identification of each kind of defects. These limitations overcome thorough deep learning techniques, whereas, the current research collected the acoustic signals are acquired using a high sensitivity microphone, then perform slitting using wire cut EDM of fabricated samples to identified the corresponding defects and then perform preprocessing to eliminate the noise of the acoustic signals. The resulting data is analysed by the time and frequency domain and the feature extraction by Fast Fourier Transform (FFT). The CNN and several machine learning (ML) models are trained validated using k-fold cross validation techniques. Among all evaluated models, the CNN framework outperforms, and its accuracy is 99% and the ROC-AUC value is 1.00. These results show the potential of deep learning-based acoustic monitoring applied in defect detection in real time fabrication of part through WA-DED, offering a viable alternative to post-process inspections and thereby enhancing overall manufacturing efficiency.

Graphic abstract