Analysis Of Recurrence Characteristics For The Monitoring Signals Of Bridge SHM

In recent years, with development of the bridge health monitoring, operational vibration signals based structural damage identification, load identification and other aspects of research have already been the hottest research area in the civil engineering. However, in the application of signal processing methods on the structural operational vibration signal, especially for signal of large bridges, most traditional methods were influenced by the non-stationary property and uncertainty factors. It made the results of the analysis unstable and caused the deviation and error in the recognition process. Therefore, an attempt to using some novel methods for non-stationary signals and the corresponding feature extraction is the purpose of this study.This paper reviewed a large amount of references and materials for the recurrence plot and its recurrence quantification analysis, and also discussed its theoretical significance, research and application background, the existing achievements and problems thoroughly. At the same time an overview of the basic theory and algorithms of the recurrence plot, and recurrence quantification analysis is introduced. At the Recurrence plot theory, the recurrence matrix was constructed through the phase space reconstruction method, and then the matrix showed in a graphical form as the object of analysis. The recurrence quantification analysis was based on statistical characteristics of the characteristic patterns in the recurrence plot, which was used for qualitative and quantitative analysis of the signals.In this paper, the discrete dyadic wavelet transform based phase space reconstruction theory is proposed, and it was used for the selection of reconstruction parameters for operational vibration signals in the recurrence plot method. And the association between discrete dyadic wavelet transform and the phase space reconstruction were explained. The discrete dyadic wavelet transform can be understood as signal’s projection into the high-pass or low pass filter vector after the reconstruction. This unified the physical sense of the discrete dyadic wavelet transform and the phase space reconstruction theory proposed by Packard et al. In other words, that meant there were no different between the physical natures of the two methods. And for the structural operational vibration signal, wavelet analysis is a very common and efficiency analytical method. On the other hand, phase space reconstruction is unified with it, which meant the phase space reconstruction based recurrence plot theory can also suit for vibration signal analysis.This paper proposes an optimal recurrence threshold selection method which is based on surrogate data and the quality loss function. Different signals have in nature different threshold values. So in order to determine the optimal threshold for recurrent plot-based vibration signal analysis of civil engineering structures under operational conditions, the surrogate technique and quality loss function are proposed to generate reliable data and to achieve the optimal discrimination power point where the threshold is optimum accordingly. Examples demonstrated the optimal recurrence threshold selection method is not only applicable for recurrence plot analysis of the general signal but also for the operational vibration signal. This method can improve analytical accuracy and stability of the results.According to the non-stationarity of operational vibration signals, a novel measure is proposed based on the recurrence quantification analysis, named Recurrence Loss Value (RLV). And the comparison between traditional index Recurrence Trend and RLV was made. In order to distinguish the non-stationarity of different signals, graphical and statistical methods were used in the recurrence plot. The traditional measure is just suit for some special signal which was included significant trend, but the proposed measure had more extensive scope. In addition, this measure can also be used for significant trend contained signal. For reinforcing proposed measurement, the measurements are used for operational vibration signals. The results showed that recurrence loss value have a good applicability for operational vibration signals. In order to evaluate the non-stationary degree of massive vibration signals, based on the concept of "local stationarity", this paper proposed an operational way to evaluate the non-stationary degree of vibration signals. Multi-variable sample matrix was established using the two non-stationary measures. The Multi-variable sample matrix was reduced to one dimension vector by Principal Component Analysis. And then based on the reference of white noise, the novel non-stationary feature DNS (Degree of Non-the stationary) is proposed by the cluster analysis. This feature was used for fast, accurate and comprehensive evaluation of non-stationarity of the operational vibration signals. Numerical Simulation and the measured ambient cable vibration signals verified the validity of this approach.The proposed operational non-stationarity assessment method is used for the operational vibration signals of Zhanjiang Bay Bridge, in order to test whether the data meet the assumption of stationarity. In the process of analyzing data, a part of signals cannot obtain the realistic results. Their non-stationarity was assessed. And the results of assessment showed that some of these signals did not satisfy the assumption of stationarity. Hence, the stationarity based modal parameter identification methods were unsuitable. Then, the signals needed further processing or other more suitable analysis method, and even deleting some poor ones to avoid none-identification and false identifying. And this can reduce workload and improve accuracy.Finally, based on unthreshold recurrence plot, the proposed delay matrix was constructed by multi-variable, and the entropy of singular was introduced to obtain a novel damage index, recurrence singular entropy. This index is sensitive to transient signal and can be used for monitoring and damage detection in structural health monitoring system. Numerical experimentation verified this index’s validity, and it can effectively identify the various damage levels of the simulated simply supported beam.