Improving the Accuracy of Structural Health Monitoring Using Synchronization Property of Vibration Signals from Multiple Positions

In recent years, the combination of phase-based motion estimation (PME) and deep learning has been proven to be a very promising structural health monitoring (SHM) method. PME is a machine vision-based vibration measurement technique, which treats each pixel in a video as an independent vibration sensor (i.e., one pixel is one sensor). Thus, PME is possible to extract millions of vibration signals from a single video to meet the demand for large amounts of data required for deep learning. In addition, the vibration signals provided by PME from multiple positions are perfectly synchronized in time domain. Previous studies have demonstrated that synchronized vibration signals can considerably augment the accuracy of SHM compared to asynchronous signals. However, how to fully leverage this temporal synchronization to facilitate SHM remains a challenge. To address this issue, this article proposes a data preprocessing method, which utilizes the temporal synchronization among multiple signals to augment the data, thereby considerably improving the accuracy of SHM. The effectiveness of the proposed method was validated by detecting misalignment of the gear shaft and loosening of the bolted joints. The results showed that the recognition accuracy of the deep learning model was considerably improved using the proposed method.