<![CDATA[Thermal monitoring of high-voltage electrical equipment is an important aspect of maintaining the reliability and operational safety of electrical power systems. Conventional classification approaches generally require labeled data and are limited in representing transitional thermal conditions. Therefore, this study proposes an unsupervised learning approach using K-Means and Fuzzy C-Means (FCM) clustering methods to analyze thermal behavior patterns in high-voltage electrical equipment based on thermal image features. The proposed model utilizes two main features extracted from thermal images, namely the percentage of white regions (% white) and non-white regions (% non-white), where white regions represent high-temperature areas. A total of 12 thermal images were used in the clustering process. Experimental results showed that the K-Means algorithm converged after only 2 iterations, whereas FCM required 53 iterations to achieve convergence . Both methods successfully identified dominant thermal patterns corresponding to Normal, Warning, and Hazardous conditions. The most extreme thermal condition was observed in data sample 6, which had a white-region percentage of 83.2647% and was consistently classified as Hazardous by both K-Means and FCM . In addition, FCM demonstrated superior capability in representing transitional thermal conditions through membership values. Data sample 3, with a white-region percentage of 56.5476%, was classified as Hazardous by K-Means but categorized as Warning by FCM with a dominant membership value of 0.702543 . These results indicate that FCM provides more flexible thermal behavior representation compared with hard clustering approaches. Overall, the proposed clustering-based approach demonstrates significant potential for real-time thermal condition assessment and predictive maintenance applications in high-voltage electrical equipment.]]>
Copyrights © 2026
