Abstract: The real-time detection of porosity in welding process is an important problem to be solved in intelligent welding manufacturing. A new on-line detection method for porosity of aluminum alloy in robotic arc welding based on arc spectrum is proposed in this paper. First, k-shape and the improved k-means were used for the initial feature selection of the preprocessed arc spectrum to reduce the data dimension. Second, the secondary feature selection was carried out based on the importance of features to further reduce feature redundancy. Then, the optimal sample label library was established by combining the final characteristic parameters and the X-ray pictures of welds. Finally, an on-line detection method of porosity in gas tungsten arc welding of aluminum alloy based on light gradient boosting machine (LightGBM) was proposed. Compared with extreme gradient boosting(XGBoost) and categorical boosting (CatBoost), this method can achieve better detection performance. The new method proposed in this paper can be used to detect other welding defects, which is helpful to the development of intelligent welding technology.

Key words: porosity detection, robotic arc welding, arc spectrum

摘要： 焊接过程中气孔率的实时检测是智能焊接制造中需要解决的一个重要问题。提出了一种基于电弧谱的机器人弧焊铝合金孔隙率在线检测方法。首先，利用k-shape和改进的k-means对预处理后的弧谱进行初始特征选择，降低数据维数；其次，根据特征的重要度进行二次特征选择，进一步降低特征冗余；然后，将最终特征参数与焊缝X射线图像相结合，建立最优样品标签库。最后，提出了一种基于光梯度增强机（LightGBM）的铝合金钨极气体保护焊气孔率在线检测方法。与极端梯度增强（XGBoost）和分类增强（CatBoost）相比，该方法可以获得更好的检测性能。本文提出的新方法可用于其他焊接缺陷的检测，有助于智能焊接技术的发展。

关键词: 孔隙度检测，机器人弧焊，电弧谱