Fault Diagnosis of Electrical Equipment through Thermal Imaging and Interpretable Machine Learning Applied on a Newly-introduced Dataset

In this study, an interpretable, fully automated pipeline for condition monitoring of electrical equipment using thermal imaging is proposed. A wider array of defects in comparison with other thermography surveys is investigated. While many fault conditions led to significant heat dissipation, a number of fault conditions result in even less heat dissipation than that of healthy equipment, implying a challenging segmentation. To overcome this problem, a pre-processing step is applied which divides data into two distinct categories according to the equipment's thermal state, namely 'cold' and 'hot' states. Afterwards, Random Forest and AdaBoost classifiers are utilized for segmentation using a sliding window approach, with regard to Interpretable Machine Learning. Moreover, a new dataset of infrared images of transformer and 3-phase induction motors is created. The proposed method has been evaluated on the very same dataset, achieving state-of-the-art results.