Identification of Full-Field Dynamic Loads on Structures Using Computer Vision and Unsupervised Machine Learning

Abstract

Real-world structures, such as civil and aerospace structures, are subjected to various dynamic loads which are spatially local and distributed. Assessment of operational performance, prediction of the dynamic responses, and prognosis of the remaining service life of the structure therefore requires accurate, high-resolution measurements, and modeling of the dynamic loads. This is extremely difficult, if not impossible, with the current state of the art. First, dynamic loads on structures usually come from a wide spectrum of sources, some of which are extremely challenging to accurately measure, such as the traffic loads on a bridge. Also, it is impractical to instrument a dense array of force measurement devices on the structure due to the high cost, the effect of mass-loading, and modification of the structure’s surface. On the other hand, digital video cameras are non-contact measurement device that are relatively low-cost, agile, and able to provide high spatial resolution, simultaneous, pixel measurements. This study develops a novel method for identification of the high-resolution, full-field loads on the structure from the video of the operational structures by leveraging advanced computer vision and unsupervised learning techniques. Impact and wind loads were applied on a cable structure to experimentally validate the method. The non-contact, remote, simultaneous sensing capability of the proposed technique should enable truly high-resolution, full-field force estimation that was previously not feasible.