In this paper, microphone array technology was used to monitor the dynamic pulsed GMAW process. At first, the splash sound signal is successfully separated out based on FastICA blind signal separation algorithm, and its frequency domain energy distribution is mainly concentrated in the high frequency band of 6000−8000 Hz. Through time and frequency domain analysis, it is found that the short-time energy of 500−1000 Hz band of observed signal and the short-time energy of splash signal can identify the burn-through defects well, furthermore the ratio of them can be used as a robust feature because of its high sensitivity and anti-interference performance. Since the splash sound signal is a characteristic signal separated by microphone array, it is shown that the abundant dynamic information provided by microphone array can better assist in the identification and monitoring of welding defects compared to a single microphone. In order to solve the serious imbalance problem between positive and negative samples, the logistic regression and BP neural network model are improved based on the machine learning method. The experimental results show that the recognition accuracy of the optimization models have been greatly improved, even could reach 99.6 %.

在本文中，麦克风阵列技术用于监测动态脉冲GMAW过程。首先，基于FastICA盲信号分离算法成功地将飞溅声信号分离出来，其频域能量分布主要集中在6000-8000 Hz的高频段。通过时域和频域分析发现，观察信号的500-1000 Hz频段的短时能量和飞溅信号的短时能量可以很好地识别穿通缺陷，而且它们的比率可以是由于其高灵敏度和抗干扰性能而被用作坚固的功能。由于飞溅声信号是由麦克风阵列分隔的特征信号，结果表明，与单个麦克风相比，麦克风阵列提供的丰富的动态信息可以更好地帮助识别和监视焊接缺陷。为了解决正负样本之间的严重不平衡问题，基于机器学习方法改进了逻辑回归和BP神经网络模型。实验结果表明，优化模型的识别精度得到了很大的提高，甚至达到了99.6％。