Non-stationary assessment of structural operational measurements using recurrence quantification analysis

Abstract

The vibration signals are usually assumed to be stationary for the sake of simplicity, but actually many signals are not definitely stationary, and ignoring the non-stationary characteristics of signals may affect the follow-up analysis directly. Therefore, proper evaluation of the non-stationarity of the analyzed signal is an inevitable and very important preprocessing procedure. In this paper, a recurrence quantification analysis-based signal non-stationarity assessment method using the recurrence plot, a time–time plane, is presented. The aiming of this method is to quantify the non-stationarity of structural operational vibration signals, and extract the non-stationary sensitive features of structural operational vibration signal from its recurrence plot. Hence various recurrence patterns in the recurrence plot are studied, of which different patterns indicate that the signals have different characters. Drift and disruption represent the signal non-stationarity. The drifted pattern means that the signal contains a trend or drift, and the disrupted pattern means that sudden changes occurred in the signal. Meanwhile, variation rules of local recurrence density in the recurrence plots caused by those non-stationarity patterns, are discussed. Afterwards, based on the variation of local recurrence density and Taguchi’s quality loss function, a new non-stationarity measurement, named recurrence loss value, is proposed. Then the comparison between the existing measurement recurrence trend and the proposed measurement recurrence loss value is conducted. Results show that the recurrence trend is mainly sensitive to the drift, and the recurrence loss value is not only sensitive to the disruption but also effective to the drift. Finally, the proposed measurement is further verified by assessing the non-stationarity of structural operational vibration signals. Those examples have demonstrated that the recurrence loss value has a promising feasibility in the practical engineering application.