Use of image recognition to investigate crack-pattern effects on seismic performance of an AGR core

This paper describes how motion-based multiple object tracking can be used to quickly visualise and interpret large multidimensional experimental data sets. These data sets were produced by a series of experimental tests investigating the response of the UK’s Advanced Gas Cooled Reactors (AGRs) to seismic inputs. At the University of Bristol, a quarter-scale single layer array (SLA) with the same layout as a layer of a real AGR was produced. This array was used to investigate the effect of cracking on the seismic resilience of AGRs. Experimental testing involved the placement of a set of cracked bricks to explore the effect of crack number and configuration on seismic performance. The bricks were tracked using a high-speed video system, and multiple object tracking was used to quantify the movements of each brick. Data challenges included processing the large video files efficiently and tracking many hundreds of points moving dynamically and simultaneously. The influence of cracking on response is discussed. Multiple object tracking of dynamic video has many applications toward interpretation of video in other disciplines such as surveillance.