Strain signal denoising in bridge SHM: A comparative analysis of MODWT and other techniques

Abstract

Accurate denoising of strain signals is critical for early damage detection in bridge structural health monitoring (SHM). However, signals denoising methods often struggle with the non-stationary and broadband noise encountered in real-world environments. This study provides the first comprehensive comparison of various denoising techniques specifically tailored for bridge strain signals, emphasizing the maximal overlapping discrete wavelet transform (MODWT) for its capacity to handle complex noise profiles. We rigorously compare MODWT with time-domain (moving average filter, finite impulse response filter, empirical mode decomposition), frequency-domain (bandpass filter, Fourier mode decomposition), and other wavelet-based (discrete wavelet transform) approaches. Uniquely, this study employs three datasets from two distinct bridge types (masonry arch and steel bowstring) and evaluates performance using both expert assessments and quantitative metrics (signal-to-noise ratio, peak signal-to-noise ratio, root mean square error, and correlation coefficient). Our findings demonstrate that MODWT exhibits a distinct advantage in high-intensity white noise environments, a common scenario in real-world bridge monitoring, offering valuable guidance for engineers in selecting appropriate denoising strategies. The results not only validate MODWT as a promising preprocessing technique but also offer critical insights into the limitations of existing methods, paving the way for the development of more adaptive and robust denoising solutions in bridge SHM.