Scholarlink

<p>Rapid detection of defects in metal additive manufacturing (AM) components remains a challenge. In this paper, laser-induced breakdown spectroscopy (LIBS) technology was used to establish a rapid identification of metal AM defects and defect-free control groups. The corresponding spectral acquisition of metal AM components with defects and without defects was carried out, and the research elements (Fe, Cr, Mn, and Ti) and corresponding spectral lines were obtained in combination with the NIST database. The spectral lines with features of importance greater than the average value are selected by random forest (RF). The selected spectral lines were used as the input variables of the k-nearest neighbor (KNN) model and the backpropagation neural network (BPNN) model. The classification performance and verification results of KNN, RF–KNN, and RF–BPNN models were compared. The results showed that the RF–BPNN model exhibited the best accuracy, sensitivity, and specificity in the training set, test set and validation set, with accuracies of 99.4, 97.2, and 96.67%, respectively. This indicates that LIBS combined with RF–BPNN can be used for the detection of defects in metal AM components.</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

Rapid Identification of Defects in Metal Additive Manufacturing Components by Libs Combined with BP-Neural Network and Random Forest Method

  • 期刊论文

  • 发表日期：

    2025年7月9日

查看全文
  • Shanping Gao,
  • Xiaomei Lin,
  • Yixiang Huang,
  • Zongxu Chen,
  • Huijin Chen

摘要

Rapid detection of defects in metal additive manufacturing (AM) components remains a challenge. In this paper, laser-induced breakdown spectroscopy (LIBS) technology was used to establish a rapid identification of metal AM defects and defect-free control groups. The corresponding spectral acquisition of metal AM components with defects and without defects was carried out, and the research elements (Fe, Cr, Mn, and Ti) and corresponding spectral lines were obtained in combination with the NIST database. The spectral lines with features of importance greater than the average value are selected by random forest (RF). The selected spectral lines were used as the input variables of the k-nearest neighbor (KNN) model and the backpropagation neural network (BPNN) model. The classification performance and verification results of KNN, RF–KNN, and RF–BPNN models were compared. The results showed that the RF–BPNN model exhibited the best accuracy, sensitivity, and specificity in the training set, test set and validation set, with accuracies of 99.4, 97.2, and 96.67%, respectively. This indicates that LIBS combined with RF–BPNN can be used for the detection of defects in metal AM components.