The development of the electric power industry is accompanied by the improvement of diagnostic tools for the state of equipment in energy systems. Some of the electrical network equipment is significantly worn out and requires increased attention in order to determine the residual resource. The synthesis of intelligent technologies and generally accepted diagnostic methods is the next step towards the future of the electric power industry. The aim of the project is to develop the principles for the functioning of an automated processing system for results of thermal imaging diagnostics of electrical equipment. The paper examines the criteria for evaluating defects in electrical equipment based on the heating temperature. An algorithm for automating the processing of the results of thermal imaging diagnostics of electrical equipment is also being developed on the basis of artificial neural networks. The software implementation of the detection of electrical installation elements in infrared images is performed using the YOLOv5 architecture. Testing and evaluation of the trained neural network are performed using thermal imaging diagnostics data of working electrical equipment. The neural network model trained as part of the study demonstrates confident detection of electrical installation parts based on the results of detecting thermograms from a test sample. Based on the results of the analysis of regulatory documentation, an approach to determining the degree of defect development has been clearly laid out. In addition to using thermal images of electrical grid equipment, the current load and ambient temperature are also recorded to select a suitable formula for calculating temperature excess or excessive temperature of an electrical installation unit or contact connection. The developed algorithm for automating the processing of the results of thermal imaging diagnostics of electrical installations based on the YOLOv5 neural network reflects the main processes necessary for the functioning of the system. A custom dataset was generated and marked up, including thermograms of real-life electrical installations, on the basis of which a neural network model was trained. Using a test sample, we were able to calculate the values of metrics to evaluate the quality of YOLOv5 model learning. The developed system has been tested on thermograms of electrical equipment. Its use makes it possible to identify not only the developed defects, but also the initial stage of the occurrence of defects in an automated mode.
Kosenko et al. (Wed,) studied this question.