Research On Infrared Thermal Imaging Technology Application In Electrical Equipment Fault Recognition

Monitoring the thermal condition of electrical equipment is essential for keeping the dependability of electrical system. Electrical equipment with aging factors will lead to overheating situation, the thermal dissipation will eventually lead to equipment failure. Additionally, after the appear of equipment failure need to spend a lot of maintenance cost, manpower and can also be catastrophic, which causing injures or even deaths. Therefore, the recognition process of equipment conditions as normal or fault is a vital step towards keeping dependability and stability of the system. Nowadays a celerity, dependable, non-contact and cost effective infrared themogyaphic inspection system is being utilized widely for detecting defects occur in equipment.This paper presents a method for obtaining electrical equipments' infrared temperature heat map utilizing infrared camera, and analyzed to evaluate the thermal state of the device. The image datas are captured by infrared camera without suspendding the running operation of the system. The infrared thermal imager which was used is produced by Japan's NEC.After obtain the thermal images, initially, to achieve the interest regions of the images, the images which were manually segmented. Then, preprocessing and image enhancement. During image enhancement, this paper adopted three methods, histogram equalization, the two-dimensional discrete wavelet transform (DDWT) and two-dimensional empirical mode decomposition(BEMD). In turn, to extract the different characteristics between fault position and the related components, from segmented regions, the first order histogram and gray level co-occurrence matrix feature as well as the characteristics, a total of 22 characteristics. Principle component analysis is applied for the better characteristics selection, a selection of 15 from 22 feature characteristics, and using discriminant analysis, the optimum features were chosen. Finally, ten characters features were slected from 22 characters feature as the inputs of the state classification system. On the final state classification, two methods were using: discriminant analysis and artificial neural network method. The performances of discriminant analysis classifier were in contrast to the performances of the neural network. The comparison results showed that discriminant analysis classifier produced better performance. The superior performance with accuracy 82.6%.