Impact of short-duration acceleration records on the ability of signal processing techniques to derive accurate bridge frequencies

This paper envisions a scenario in which unmanned aerial vehicles gather data from low-cost and flexible wireless sensor networks, i.e., accelerometers. However, flight duration, coupled with limited sensor battery time, is a substantial technical limitation. In order to assess the impact of these constraints on bridge monitoring, this paper analyses the extraction of bridge dynamic features from short-duration acceleration records. The short acceleration record is simulated using the theoretical response of a simply supported beam subjected to a moving load. Estimated frequencies are obtained in free vibration and compared with the natural frequencies calculated from formula. Given that short records limit the resolution in the frequency domain, the error in the prediction of frequencies will typically decrease as the duration of the signal increases. Signal processing techniques for extracting dynamic features include the Fast Fourier Transform, Frequency Domain Decomposition, Continuous Wavelet Transform and Hilbert-Huang Transform. This paper carries out an assessment of the accuracy of these signal processing techniques in extracting frequencies as a function of the duration of the measurements. Edge effects and loss of resolution are shown to remain key issues to be addressed when the duration of the signal is too short.