Transformer-Based Bolt Looseness Detection with Data Augmentation by Variational Autoencoders

Bolt looseness would lead to hazardous damage to engineering structures. Neural network-based algorithms can effectively monitor health conditions using the impedance signal. On the other hand, the impedance data is difficult to obtain, and the data has an imbalanced distribution. In this research, we propose a data augmentation method based on the Variational Autoencoder (VAE) model with channel attention and spatial attention mechanisms. This model takes advantage of the VAE model to generate additional data to resolve the issue of an imbalanced sample. The new dataset is then fed into the Transformer model for training and fault identification. Experimental results demonstrate that this method performs well in enhancing and classifying imbalanced data, increasing accuracy from 81.82% to 87.01%.