Novelty detection approach for the monitoring of structural vibrations using vision-based mean frequency maps

Abstract

Damage detection is an important part of modern-day engineering. Early damage detection is facilitated by Structural Health Monitoring methods, which may be employed using numerous modalities, such as vibration, guided waves, thermography, or computer vision. These methods produce information that can then be interpreted to detect and localize damage or quantify its extent. Novelty Detection (ND) is a data interpretation approach that enables damage detection without prior knowledge of damage-related influences on gathered data. ND can be conveniently performed using computer vision methods, which allow continuous, non-contact monitoring of large structures with the possibility of relying on such quantifiers as deflection, vibration, or strains.In this work, we present an ND method for monitoring the structural changes in rotary machinery equipment using vision-based data. The proposed technique detects changes in the characteristic frequencies of vibrations due to damage. Using the optical flow calculated for the videos acquired using a high-speed camera, the maps of mean frequency can be estimated and used for evaluating the differences between the reference data set and the data obtained during the monitoring. The Optimal Baseline Selection is used to compensate for the varying operational conditions under which the structure is monitored. The approach was tested on the air compressor working under variable pressure, and the damage introduced to the structure was successfully detected.