To implement predicting and controlling of welding quality are significant during pulsed gas tungsten arc welding (GTAW) process. In this paper, a multi-sensor system has been developed to synchronously obtain arc voltage, welding current, arc power, arc sound and weld pool images during pulsed GTAW process. The convolutional neural network (CNN) is designed to extract the visual feature of weld pool images. Besides, the time-frequency domain features of arc voltage, welding current, arc power, arc sound are also extracted. These features constituted a 19-dimensional feature vector. The long short-term memory (LSTM) network is used to fuse the extracted 19-dimensional features and learn time series information from the fused features. Further, the LSTM network can predict the different welding states 0−2 s in advance: normal penetration, lack of fusion, sag depression, burn through and misalignment. Finally, the proposed hybrid network model, CNN-LSTM, is verified to be effective with high accuracy and robustness.

在脉冲气体钨极氩弧焊（GTAW）过程中，实现焊接质量的预测和控制非常重要。本文开发了一种多传感器系统，以在脉冲GTAW过程中同步获取电弧电压，焊接电流，电弧功率，电弧声和焊池图像。卷积神经网络（CNN）用于提取焊缝池图像的视觉特征。此外，还提取了电弧电压，焊接电流，电弧功率，电弧声的时频特征。这些特征构成19维特征向量。长短期记忆（LSTM）网络用于融合提取的19维特征并从融合的特征中学习时间序列信息。此外，LSTM网络可以提前0-2 s预测不同的焊接状态：正常熔深，缺乏融合，凹陷松弛，烧穿和错位。最后，验证了所提出的混合网络模型CNN-LSTM是有效的，具有较高的准确性和鲁棒性。