Study of inner porosity detection for Al-Mg alloy in arc welding through on-line optical spectroscopy: Correlation and feature reduction
Study of inner porosity detection for Al-Mg alloy in arc welding through on-line optical spectroscopy: Correlation and feature reduction
复制标题
在线光谱法检测铝镁合金电弧焊内部气孔的研究：相关性和特征约简
DOI：
10.1016/j.jmapro.2019.02.016
复制
发表时间：
2019-03-01
影响因子：
6.2
通讯作者：
Wen, Guangrui
中科院分区：
文献类型：
--
作者：
Zhang, Zhifen;Zhang, Linjie;Wen, Guangrui
电弧焊铝合金内部气孔通常是不可见的、瞬态的，物理信息微弱，难以实时检测。本研究采用电弧光谱和后显微表征技术，研究了脉冲气体钨极电弧焊（GTAW）中铝镁（Al-Mg）合金内部气孔在线检测的关键技术。使用光谱仪、电荷耦合器件摄像头和麦克风开发了机器人 GTAW 监控系统。在使用扫描电子显微镜和能量色散光谱进行后测试和微观表征之前，设计了不同程度孔隙率的实验。发现了铝合金GTAW 两种内部气孔的产生机制。通过光谱主成分分析，在特征提取之前选择金属和氢成分的线谱。利用统计值定量研究线谱主成分系数（H I 和 Mg I）与内部孔隙率之间的相关性。然后，提出了一种用于内部孔隙度检测的改进特征参数，例如第一主成分的 H I 谱的绝对系数，并在不同孔隙度水平下进行了实验验证。最后，通过所提出的主成分分析和 t 分布随机邻域嵌入实现了特征缩减和可视化。还讨论了微观分析和光谱特征之间的相关性。
Inner porosity of aluminum alloy in arc welding is usually invisible, transient and difficult to detect in real-time with weak physical information. This study investigated the key technologies for on-line inner porosity detection for aluminum-magnesium (Al-Mg) alloy in pulsed gas tungsten arc welding (GTAW) using arc optical spectroscopy and post-micro-characterization. A monitoring system for robotic GTAW was developed using a spectrometer, charge-coupled device camera, and microphone. Experiments for different degrees of porosity were designed before post-testing and microscopic characterization using scanning electron microscopy and energy dispersive spectroscopy. Two types of generation mechanism of inner porosity were discovered for Al alloy in GTAW. The line spectrums of metal and hydrogen components were selected before feature extraction by means of principal component analysis of optical spectroscopy. The correlation between the principal component coefficient of line spectrums (H I and Mg I) and inner porosity was quantitatively investigated using statistical values. Then, an improved feature parameter, e.g., the absolute coefficient of the H I spectrum from the first principal component, for inner porosity detection was proposed and experimentally validated at different levels of porosity. Finally, feature reduction and visualization were achieved by means of the proposed principal component analysis and t-Distributed Stochastic Neighbor Embedding. The correlation between microscopic analysis and spectrum features is also discussed.