16 / 2023-08-20 14:53:43

Transformer-based Bolt Looseness Detection with Data Enhancement by Variational Autoencoders

Bolt looseness, Variational autoencoder, Data augmentation, Transformer model

Special Sessions > SS10- Artificial Intelligence in Environmental Science

Bolt looseness would lead to hazard damage of glass curtain wall structures. Neural networks-based algorithms can effectively monitor the health conditions using the impedance signal. On the other hand, the impedance data of glass curtain wall is difficult to be obtained and the data has imbalanced data distribution. In this research, we propose an adaptive weighted average preprocessing and data augmentation method based on the Variational Autoencoder (VAE) model. This model takes advantage of VAE to generate additional data to resolve the issue of imbalance sample. The new dataset is then fed into the Transformer model for training and fault identification. Experimental results demonstrate that this method exhibits good effectiveness in enhancing and classifying imbalanced datasets, leading to an improved accuracy to 98%.