Fiber-optic acoustic wave microcrack identification of transmission tower concrete pile foundation based on synchronous compression time–frequency analysis and multi-scale CNN

Abstract

Early damage detection in concrete pile foundations of transmission towers remains a significant challenge due to the extremely weak acoustic signals emitted by microcracks and the overwhelming presence of complex background noise. To overcome this limitation, this study presents a novel hybrid approach that integrates synchronous compression transform (SCT) with a multi-scale convolutional neural network (MCNN). Distributed fiber-optic acoustic sensing is employed to capture structural response signals, which are inherently non-stationary. The SCT technique is then applied to achieve high-resolution time–frequency representation, effectively enhancing the fine-scale features associated with microcracking while suppressing noise interference. These refined time–frequency representations are converted into 2D images and fed into the MCNN model for intelligent pattern recognition. To enhance sensitivity to subtle fracture signatures, the MCNN architecture is designed to extract discriminative features across multiple scales, enabling robust detection even under severe noise conditions—with signal-to-noise ratios as low as below 0 dB. Extensive experimental validation was conducted using simulated crack scenarios in concrete pile foundations. The proposed method achieved a microcrack recognition accuracy of 96.3%, a recall rate exceeding 94%, and an F1 score above 95%, demonstrating outstanding precision, robustness, and consistency. The results confirm that the SCT-MCNN framework significantly improves the sensitivity and reliability of early damage identification, offering a promising solution for high-precision, real-time structural health monitoring of complex civil infrastructure.