Guided Wave-Based Damage Detection in Composite Plate With Temperature Variation: A Residual Attention Capsule Network Approach

The structural safety of carbon-fiber composite is usually threatened by the occurrence and accumulation of damage during the in-service period. For structural health monitoring (SHM), guided wave-based damage detection methods are widely adopted in industry. However, these methods are significantly affected by a variable temperature in-service environment, which would make it hard to extract useful damage information and may further result in false damage diagnosis. To address these issues, in this article, an efficient residual attention capsule (RAC) network is innovatively developed for guided wave-based damage detection in composite plate with the temperature variation. Considering the part-whole spatial relationships, the proposed RAC network has deep guided wave feature mining capability without damage index (DI) calculation under temperature variation. Synchronous squeeze-short time Fourier transform (S-STFT) method is adopted for wave signal processing, which can effectively convert a 1-D time-domain signal to a 2-D time-frequency signal with the advantage of time-frequency ambiguity phenomenon suppression. In addition, the guided wave propagation characteristic in composite structures is clearly analyzed for the wave mode analysis. The performance of the proposed RAC damage detection method is evaluated and verified with experimental data in the composite plate. The high accuracy results have been obtained and show greater potential than other state-of-the-art contrastive damage detection methods under a complex temperature environment.