A wavelet-based despiking algorithm for large data of structural health monitoring

The last two decades have witnessed a rapid increase in the applications of long-term structural health monitoring technologies to the civil structures. A wealth of field data has been collected by the structural health monitoring systems. Nevertheless, the mining of information associated with structural condition from the large database is still a great challenge. In the structural health monitoring signals, spikes are commonly encountered anomalies that have large amplitudes and may show complex spatial and temporal patterns. They can introduce significant errors in data-based condition assessment of structures. Particularly, the long-term structural health monitoring data have an extraordinarily large volume. To remove the spikes in it, the algorithm is highly desired to be both automatic and efficient. An unsupervised and fast despiking method is proposed in this article on the theoretical cornerstone of the wavelet transform. This method is implemented by two steps, namely, spike detection and spike removal. The hypothesis testing and algorithm of searching wavelet modulus maxima chain are incorporated into the spike-detection procedure. Thus, the arrival time of the spikes can be identified fast. And then, the spikes are removed by a cross-scale maxima and minima search algorithm based on the maximum overlap discrete wavelet transform, retaining the unaffected information. The inverse transformation is not required in the spike-detection step, which improves the speed of the algorithm. The spike-removal algorithm removes spikes only from their occurrence frequency bands; thus, the unaffected signal components are intact after despiking. The proposed algorithm is demonstrated using three sets of structural health monitoring data collected from a real bridge, comparing with three other approaches, that is, the time-domain method, frequency filter and traditional wavelet method.