Abstract: Data preprocessing completed in a data management system plays an important role in structural health monitoring. The Romanowski guideline (t-test) is selected to eliminate the gross error through comparison. The wavelet denoising method can effectively remove the noise (or error) caused by environmental factors, and yield more reliable monitoring data. However, a consensus has not been achieved on how to select the threshold function, the threshold and the number of decomposition layers. Using the 39 groups of monitoring data obtained in the health monitoring of the Yangtze River tunnel, we analyze the effects of various combinations of thresholds and threshold functions on the results of the wavelet denoising method. It is shown that the denoising effect via 4-5 decomposition layers is the best using the combination of Rigrsure threshold and hard threshold function. A satisfactory result is achieved when the above research are applied to denoise the corresponding period monitoring data. Based on these results, it is shown that the denoised data can effectively prevent the generation of false alarm in the early warning compared to the original data. Finally, a method based on the combination of the wavelet denoising and least squares method is proposed to predict the monitoring data.

Key words: tunnel, health monitoring, gross error reject, wavelet denoising, early warning, data forecasting