Minimizing feature set for transformer fault classification with infrared imaging and ACO-KNN

Transformers are critical components in power transmission and distribution systems. However, their performance may deteriorate over time due to multiple factors. To detect such issues, diagnostic techniques like vibration analysis and infrared imaging are commonly used. Among these, infrared imaging stands out as a non-contact method requiring only an infrared camera. Nevertheless, interpreting thermal images can be difficult, as visual differences between faulty and healthy conditions are often subtle and indistinguishable to the human eye. This paper proposes a novel approach for detecting and classifying nine transformer conditions using infrared thermal images combined with Ant Colony Optimization (ACO) and the K-Nearest Neighbors (KNN) algorithm. Features are extracted from the R, G, and B channels separately, and four statistical indicators are computed to generate a comprehensive feature matrix. ACO is used to optimize the feature selection based on classification accuracy. The method addresses the challenge of achieving high diagnostic accuracy while meeting the constraints of online implementation. Experimental results show that the proposed method can accurately distinguish between one healthy state and eight types of short-circuit faults, outperforming conventional techniques.