proposing a novel method in diagnosing power transformer failures based on the analysis of morphological components

Power transformers are very important in electrical energy distribution systems. in establishing a power plant and even in distribution networks, the cost of power transformers is significant. any damage or failure in transformers can cause irreparable and heavy losses. therefore, this equipment is always important in periodic visits. in periodic visits, various parameters of this equipment are checked. one of the ways used in periodic visits is to use infrared images to evaluate the health of this equipment. however, the analysis of these images by human experts is always error-prone and expensive. in this article, a new method is proposed in machine vision to diagnose transformer faults. in the proposed method, infrared images are pre-processed and then features based on multi-resolution transformations such as morphological component analysis, wavelet transform and bandlet transform are extracted and dimension reduced with the help of independent component analysis. the given one-dimensional features are classified using random forest classification, support vector machine and k-nearest neighbor. the classification accuracy obtained in the random forest bin class as the best classifier in fault detection is equal to 98.81%, as well as 91.51% sensitivity and 84.54% negative news rate, as well as 91.86% negative news rate compared to other the numbers showed their superiority. in the f criterion, this value has reached 0.99, which shows the efficiency of the proposed method.