Wavelet-based denoising of structural health monitoring strain measurements

Since the structural assessment of steel bridges relies on the data collected from structural health monitoring (SHM), measurement singularities should be mitigated for more precise assessment. The presence of noise in measured data may obscure significant details and resulting inaccurate structural conditions; therefore, the signal-denoising process is essential. This study focuses on the noise treatment of the measurements recorded by strain gauges implemented through the SHM system of the Southern Danube Railway Bridge (SDRB) in Budapest. In the study, a wavelet-based denoising approach is investigated by implementing five wavelet mother functions (Haar, Daubechies, Symlets, Coiflet, and Biorthogonal) conjugated by two universal (hard and soft) thresholding methods, which in turn use four threshold techniques (Sqtwolog, Rigrsure, Minimaxi, and Heursure) to eliminate the noise and preserving the critical signal details. Three evaluation metrics, signal-to-noise ratio (SNR), square root of the error (RMSE), and the correlation coefficient (CC), are used to estimate the denoising performance. Further smoothness refinement of the denoised signal is achieved by investigating two filtering techniques (lowpass and Savitzky-Golay). The performance of filters is examined by estimating the smoothness of the filtered signal. The study revealed that the wavelet mother functions have comparable results, and the hard thresholding method implementing the Rigrsure threshold technique and lowpass filter shows higher performance. Further validation of the proposed approach is achieved by comparing its performance with the contemporary denoising methods. To ensure the adaptability of the proposed strategy, it is applied to a measurement recorded from a different structural system.