DOMAIN ADAPTATION NETWORK BASED ON PSEUDO-LABEL GENERATION USING ACOUSTIC EMISSION FOR BOLT LOOSENESS DIAGNOSIS

Acoustic emission (AE) technique is widely used in non-destructive testing for structural and material damage evaluation. Its main advantages include high sensitivity to minor damage and no need for active excitation. For bolted-joint structures, it is shown that the friction between contact surfaces can also produce AE and changes in the AE signals are primarily influenced by bolt preload, which makes AE detection a potential tool to indicate interface contact conditions. In recent years, deep learning (DL) has developed rapidly in pattern recognition and damage detection. However, traditional DL methods assume the data distribution of the training and testing datasets are precisely the same. In fact, for most engineering structures, there are obvious differences in the distribution of datasets obtained under different working conditions. Therefore, it is difficult to obtain accurate prediction results using traditional DL methods. Domain adaptation (DA) refers to training a model on the source domain and applying it to the target domain. It usually assumes some data from the target domain can be obtained during the training process. Therefore, this paper proposes a novel DA method for bolt looseness using the AE technique. Firstly, a superposition loss function is proposed, which takes into account the ordinal characteristics of bolt looseness. Then, the Deep Coral method is used to improve the generalization ability of the model. Finally, the model is retrained by accessing the target domain dataset and generating pseudo labels. To verify the proposed method, the “ORION-AE” dataset is used. This dataset contains five measurement campaigns, each of which contains seven bolt preload torques. Different measurement campaigns are used as the target dataset in turn, and the remaining four measurement campaigns are used as training dataset. The results show that the proposed method achieves good diagnostic accuracy under different working conditions.