Thermal Inspection of Electrical Substation’s Equipments Using YOLOv8

The thermal inspection in the elements of an electrical substation is a standard procedure that integrate a predictive maintenance plane. It has the objective of detect functional anomalies on typical substation’s components as power transformers, disconnect switches, circuit breakers, bus-bars and general connections between those equipment. Those inspections allow the operation team to search for hotspots on those critical equipment and connections, those anomalies can indicate malfunction or bad connection that, if not treated properly, will probably cause damage to the equipment and unscheduled shut down of the substation plant, some of the worse scenarios for a power electrical company. The rules of this procedure are defined by the regulatory sector of electricity companies, which classify hotspots into 4 levels of criticality according to the temperature measured. This study has the objective of produce an image detection method, using YOLOv8, that possibilities to flag and classify hotspots on equipment of an electrical substation yard without the human visual analysis. To training and adjusting the model it was used an inspection thermal image dataset collected from transmission substations power plants located at the Espírito Santo (Brazil) state. The result was a precision of 88,1% on the best configuration of the model. To evaluate the model, the results were compared to the simpler model Yolov5n, showing that the proposed image detection model has increased the precision in 17,62%, bringing a fresh perspective on the use of machine learning for thermal image identification and classification solutions.