Finding suitable feature extraction method for condition monitoring of electrical equipment

Abstract

The degradation of electrical equipment caused by excessive temperature rise leading to the failure of a total electrical system can be reduced by the thermal monitoring of the equipment. Manual analysis of thermal images is time-consuming, cost-effective and can cause injuries or health damages. Therefore, building an automated fault diagnosis system plus selecting the suitable features for developing that system is essential. As there are several feature extraction methods and applying all of them to identify suitable features is time-consuming and creates extra loads on the automated system, choosing one efficient method for feature extraction is necessary. This study actually shows the comparison among different texture feature extraction techniques and find the best one by using Machine Learning. After extracting different features using different methods from thermal images of electrical equipment, firstly, supervised learning was used along with Random Forest as a classifier and then training-testing data were used to train the machine and predict the segmented regions of the pictures. The study result shows that using Gray-Level Co-Occurrence Matrix as feature extracting method gave the most accuracy and less error in the performance analysis algorithm. Finally, the condition of the electrical equipment is also predicted whether it was faulty or normal in addition to which feature extracting method provides most accuracy.