Induction Motor Fault Detection and Classification using RCNN and SURF Based Machine Learning Algorithms and Infrared Thermography

Abstract

Induction motors in electrical industries face stress and potential faults. Preventive maintenance, including fault detection, is vital for safety and energy conservation. Infrared imaging, though underutilized, can monitor machine conditions effectively. In response to this gap, this paper presents a novel motor fault identification method employing infrared thermography (IRT) in combination with image processing and machine learning techniques, with a particular focus on energy efficiency. IRT is harnessed for early fault detection to promote energy conservation. The approach involves the extraction of color and texture features from the motor's infrared images using the Gabor filter and GNS (Global Neighbourhood Structure) map. The proposed method integrates the faster R-CNN (Region-based Convolutional NeuralNetwork) with the SURF (Speeded Up Robust Features) algorithm to enhance fault detection and classification accuracy. SURF serves as a feature descriptor for faster R-CNN, enabling object detection and fault classification based on the extracted features. Additionally, efficiency is assessed using the Finite Element Method (FEM) based on stator and rotor power, contributing to energy conservation through early fault detection in motors. Notably, the proposed motor fault classification is applicable under various loading conditions, consistently achieving accuracy rates exceeding 90%.