Robust and efficient feature-based method for structural health monitoring of large structures

It is crucial to identify potential natural patterns and employ them for measuring displacements of large structures. However, detecting and matching unique features over a natural pattern is challenging due to the random order, intensity and movement of the pixels. This study presents a feature-based methodology that employs the KAZE feature detection and descriptor algorithm due to its superior capability in dealing with complex natural features experiencing complicated scale-space changes. The developed methodology identifies a natural pattern in the reference image and determines the strongest interest point (SIP) with the help of the KAZE algorithm. This reference key point, i.e. the SIP, is used to determine the corresponding matches in the other images to obtain the displacements and the natural frequencies. The work conducts simulation and indoor experiments for validation and applies the developed approach to a utility-scale wind turbine. The developed KAZE-based methodology accurately determined the vibrational characteristics from the natural patterns of the tower and one of the blades of the wind turbine. In addition, a comparative study is demonstrated with BRISK, SIFT, and SURF algorithms to validate the superior accuracy and robustness of the implemented approach. The present study also includes an extensive vibrational study of the blade by creating multiple regions of interest (ROIs) around the available natural patterns. A parallel computation approach was developed to process the ROIs simultaneously, which reduces the execution time significantly by more than 50%.