Fault detection and classification in kinematic chains by means of PCA extraction-reduction of features from thermographic images

Abstract

Kinematic chains are essential elements configurable in different topologies according to the requirements of industry. Their main components are the rotating machines and mechanical parts in which diverse faults can appear. Nowadays, infrared imaging analysis has gained attention for monitoring kinematic chains, however, the approaches for detecting and classifying faults still can be improved. Therefore, this work presents a methodology that uses a low-cost infrared measurement system and combines adequate techniques, such as infrared images preprocessing and segmenting, extraction of statistical indicators, generation of a high-dimensional matrix of features, features reduction, and categorization, for accurately detecting and classifying a wide variety of fault conditions in kinematic chains. This approach was applied to a configurable kinematic chain under the following conditions: healthy motor, misalignment, unbalance, one and two broken rotor bars, bearing faults on the outer race, healthy gearbox, and gearbox wearing. The obtained results validate the effectiveness of the proposed methodology.