Multi physics driven pipeline damage localization method based on Helical guided wave under noise environments

The structural health monitoring technology based on helical guided waves is an effective method for detecting damage in pipeline structures. However, environmental factors during service often introduce unpredictable noise, causing defect scattered signals to become weak or obscured by noise, thereby hindering accurate defect identification. To achieve damage localization under unknown noise conditions, this paper proposes a Multi-Channel and Multi-Branch Convolutional Denoise Autoencoder-Transformer model (MC2DAE-Transformer) driven by multiple physical priors. Firstly, a feature encoder is constructed that can extract multiple temporal and spatial features. Local and global information are fused and subsequently decoded to achieve denoising reconstruction of the noise signal. The propagation path similarity of HGW is introduced as prior knowledge to enhance the spatial correlation of the denoising model. And a loss term containing multiple physical information is introduced to improve the denoising performance of the model. Subsequently, Transformer module can extract the damage information contained in the denoised signal and map it to the damage coordinates to complete the damage localization. The experimental results show that MC2DAE can increase the signal-to-noise ratio (SNR) from −10 dB − 6 dB to an average SNR of 7.42 dB for noise signals. The average localization error of MC2DAE-Transformer is 7.43 mm, the standard deviation is 4.76 mm, and the relative error is 1.48%. Compared with other deep learning methods, the proposed method can achieve accurate damage localization under unknown noise, and has good stability and generalization. Our code will be available upon paper acceptance (https://github.com/red895866-rgb/MC2DAE-Transformer.git).