Intuitive and Less-Supervised Structural Damage Detection Using Phase-Based Vibration Imaging

Abstract

Numerous structural vibration monitoring techniques have been developed to identify structural operational conditions or structural damage. However, a high-level supervision from experts is usually required especially for applications on complicated structures. Cameras, as a noncontact sensor, have emerged as a powerful tool for measuring the full-field vibration of complex structures due to its advantages over the traditional contact sensors, such as flexible positioning, simultaneous multi-points tracking, and high-spatial resolution. However, extracting key information from the large amount of data from video sequences can be challenging. In this paper, we propose an improved phase-based optical flow method to efficiently estimate the full-field displacement, enabling visualization of the vibration energy distribution in a lesssupervised manner. The method consists of several steps. First, multiple Gabor filters are applied to extract the local phases at different scales and in different directions. The phase variations and spatial gradients are then used to estimate displacement components at each pixel. Sub-pixel level accuracy is achieved by rejecting unreliable components using confidence measures and then integrating them into the motion. The user only needs to provide inputs for two thresholds of confidence measures. Second, based on the extracted full-field displacements, the structural vibration is transformed into a series of images using feature extraction techniques, such as mean absolute deviation, which scales the amplitude of vibration level, and/or frequency filtering which visualizes operational deflection shapes in near real-time. Finally, a nonlinearity weighted local area mapping algorithm is proposed to detect anomalies present in a structure, which is done by calculating the difference in the full-field feature map between healthy and tested states. To validate our approach, we conducted on experiment on an air compressor and compared the results with those measured by accelerometers. The results showed that the proposed vibration imaging technique can clearly identify damage and indicate its potential for various structural dynamics applications.