1. A performance comparison of deep learning and shallow machine learning in acoustic emission monitoring of aluminium alloy pulsed laser welding.
- Author
-
Zeng, Da, Wu, Di, Luo, Zhongyi, Dong, Jinfang, Huang, Hongxing, Yang, Fangyi, Zhang, Peilei, and Ye, Xin
- Subjects
ALUMINUM alloy welding ,LASER welding ,CONVOLUTIONAL neural networks ,ALUMINUM alloys ,MACHINE learning ,DEEP learning - Abstract
The penetration depth is one of the important indicators in aluminium alloy laser welding, which is closely related to the welding quality. In situ monitoring penetration depth can provide a judging basis for the process optimization so as to reduce the occurrence of lack of penetration or burn-through defects. In this paper, the acoustic emission (AE) technique combing with various machine learning (ML) algorithms, including the shallow machine learning (SML) and deep learning (DL) algorithms, was proposed to predict the weld penetration. Furthermore, the class activation map (CAM) method was applied to visualise and explain the prediction mechanism of convolutional neural network (CNN). The comparison results demonstrated that the combination of ML and AE monitoring can achieve very accurate penetration prediction, and the classification performance of DL algorithm is stronger than that of SML algorithm. The constructed ResNet weld penetration status recognition model achieved the highest accuracy of up to 97.7% on the test set. In addition, the CAM method can effectively reveal the important feature areas related to penetration in AE signal, which provides valuable information for the understanding and optimization of DL algorithm. In summary, our work proposes a new approach for in situ penetration monitoring of aluminium alloy pulse laser welding, which can improve the welding quality and efficiency. [ABSTRACT FROM AUTHOR]
- Published
- 2024
- Full Text
- View/download PDF