Design of thermal imaging-based health condition monitoring and early fault detection technique for porcelain insulators using Machine learning

The inspection of insulator faults is an important task to prevent catastrophic failures in the operation of an electric substation. Manual inspection of overhead power line insulators can be very dangerous owning to the presence of high voltage in power sub-stations. Hence, in this paper, we present an infrared thermal (IRT) camera based non-invasive computer vision system for automatic monitoring and visual inspection of overhead on line power insulators. In the proposed work, initially, an optimal threshold method is applied to segment the region of interest (ROI) in IRT images. Subsequently, various geometrical, morphological, intensity and statistical features are computed from the segmented ROI, which are eventually utilized as an input to Gaussian kernel support vector machine to classify the different type of faults in insulator images. Computer vision based automatic inspection of insulators can play an important role from environment as well as human safety point of view. Timely inspection of insulators can ensure the environmental safety through prevention of fire that may be caused due to the insulator failures leading to the sudden breakdown of high-power lines. The proposed system achieved the true positive rate (TPR), and false negative rate (FNR) of 97.3%, and 2.66%, respectively. Whereas, the system obtained the Positive Predictive Value (PPV) and False Discovery Rates (FDR) of about 97% and 3%, respectively with the accuracy of 0.97 on the receiver operating characteristics (ROC) curve. The performance of the proposed method is compared with other existing state of the art methods and found that our method outperformed over them. Hence, we recommend proposed system to detect the severity of faults in IRT images long before any catastrophic failures take place at power sub-stations. [Display omitted] •Segmented insulator images into two regions such as faulty and non-faulty regions.•Computed geometrical, and statistical features from the segmented faulty region to build a Gaussian Kernel based Support Vector Machine (GK-SVM) classifier.•Employed GK-SVM classifier to classify faulty insulator into three classes viz., less faulty, medium faulty and critical faulty.•Compared the performance of the proposed system with other recent state of the art methods.