Non-contact modal parameters identification using a K-cluster algorithm

Abstract

Non-contact structural health monitoring is a promising field for assessing civil structures, such as bridges. Not having to access the structure avoids different issues: the closure of the structure, the use of special equipment to access it, and others. This study uses digital image processing, machine learning, and parallel computing to detect the vibration of a flexible structure. If a structure is too stiff, a reinforced concrete short-span bridge or a multi-story building, it is hard to identify its natural frequencies without some sort of target panel or target feature. Instead, if the structure is flexible, it is possible to identify its displacement and its natural frequencies, but it is a challenge with high computational cost. This study presents an unsupervised machine-learning algorithm to identify a structure, its displacement, and its natural frequencies. The algorithm was deployed on a simple supported beam using a commercially available camera and an inexpensive GPU.