Improving the Capability of Detecting Damages in the Early State by Advanced Frequency Estimation

Detecting damage in the early state is crucial in assessing structural integrity. Most current vibration-based damage detection methods use frequency shifts to assess the damage, observed as a change of the positions on which the peaks in the spectrum are located. However, accurate estimation of the natural frequencies can be challenging due to the raw frequency resolution obtained for short signals. We propose in this paper a signal post-processing algorithm that permits obtaining a spectrum with significantly enhanced resolution, without being necessary to increase the length of the signal. The super-resolution is obtained by overlapping numerous spectra calculated for the signal cropped iteratively. The spectral peaks are distributed in accordance with a pseudosinc function, which is asymmetrical, but the estimated frequencies are close to the real one. By interpolation, we improve the estimate. Moreover, by applying a correction term we find the true frequency. The algorithm is implemented in a Python application that can be linked to any virtual instrument developed in LabVIEW. The algorithm is tested for signals with known frequencies, in the absence and presence of noise and for real-world signals. It provides accurate results that permit observing the occurrence of damage in the very early state.