An edge-enabled lightweight intelligent framework for structural health monitoring with strain-temperature sensing using wavelet scattering and 1D convolutional autoencoders

Abstract

Due to the ageing of civil infrastructure, efficient and reliable structural health monitoring (SHM) systems have become increasingly necessary. At the same time, the growing volume and complexity of sensing data demand advanced signal processing approaches that can transform raw measurements into damage-sensitive features and interpretable health indicators (HIs). The integration of artificial intelligence (AI) techniques into SHM is not only beneficial for automation but also essential for scalable analysis of non-stationary and noisy time-series data. Furthermore, lightweight processing frameworks explicitly designed for edge or in-situ deployment offer advantages including reduced latency, real-time monitoring, and scalability. This study presents a generalizable, deep learning–based intelligent signal processing framework for the monitoring of civil infrastructure using strain–temperature sensing. The proposed framework was validated using synthetic damage data generated from long-term data collected from road infrastructure monitored with fibre Bragg grating (FBG) sensors, as well as experimental fatigue test data. In this approach, strain and temperature data were first collected through a sensor network. A wavelet scattering network (WSN) was applied for efficient feature extraction, producing scattering coefficients that significantly improved computational efficiency and robustness to variability. These features were processed by a one-dimensional convolutional autoencoder (1D-CNNAE), optimized using Bayesian methods, to construct automated health indicators (HIs) that capture structural condition. The framework contributes to structural health monitoring signal processing by integrating multi-stage data compression, interpretability through environmental variable fusion, and computational efficiency suitable for practical deployment a computationally lightweight design suitable for future edge deployment. Case studies demonstrate that the resulting HIs reliably distinguish operational variability from structural degradation. More broadly, the methodology exemplifies how intelligent signal processing pipelines can enable real-time monitoring and support timely maintenance decisions across infrastructure systems.