The use of fuzzy sets to create a dataset for a neural network assessing the hazard of glass insulator defects

The article considers the formation of a calculation model for assessing the hazard of 1–2 mm air inclu-sions in glass parts of PS-70E (U70) insulators. The model includes fuzzy sets for the following input variables: defect size, distance from the part axis to the defect center, position of the defect center. The hazard of defects was assessed using the Mamdani algorithm based on the obtained model. The calculation results were verified based on the calculation of electric field distribution patterns. The article shows that the calculation model allows for the correct classification of defects for the considered examples in terms of the hazard of electrical strength failures. The developed model can be improved by expanding its generalizing ability to other types of defects and boundaries of glass parts. Given the ex-pected performance accuracy, the model can be used to form training data for a neural network able to assess the hazard of potential defects in terms of electrical strength failures based on the images of glass parts.