Bridge Structural Damage Identification Based on Parallel Multi-head Self-attention Mechanism and Bidirectional Long and Short-term Memory Network

To address the health monitoring challenges of bridge structures and identify suitable damage features, this paper initially denoises the input data using variational mode decomposition combined with wavelet threshold denoising. The multi-head self-attention mechanism is combined in parallel with the bidirectional long and short-term memory network to fully utilize the former’s capability to get the global features of the data and the latter’s capability to obtain the temporal features of the data. Ultimately, these two features are horizontally spliced to construct the parallel multi-head self-attention mechanism and bidirectional long and short-term memory (PMABL) network model. This paper uses the steel truss structure and IASC-ASCE benchmark datasets to assess the model. Experimental results demonstrate that the PMABL model surpasses existing models, achieving higher damage identification accuracy and better recognition ability for damage patterns with similar characteristics.