Deep Learning Method and Infrared Imaging as a Tool for Transformer Faults Detection

Abstract

In behaviour recognition, the development of the DL (Deep Learning) method introduced massive improvements in the field of artificial intelligence, where DL represents an upgrade of the present ANN (artificial neural network) architecture. Deep Learning as a comprehensive new field of artificial intelligence completely covers the neural networks architecture that is devised to carry out certain forms of identification, such as behaviour, forms of things, trends, similarities in complex forms, etc. Regarding thermography in energy, the cases used to illustrate this are photographs of active energy components in the plant. Failures that are seen with thermography cannot be recognized by other methods. However, an expert needs to do segmentation of focusing and classification of failures. The need for daily sampling and expert work is growing. With the DL method, it can be done in real time any time. One of the popular network architectures for using DL in image analysis is the recognition algorithm-CNN (convolution neural network). Traditional artificial intelligence methods require determining factors and computations, leading to training algorithm. Machine learning has important features as well as the right weight to make decisions about new input data. This work presents DL as a flexible and adaptive method for the analysis of thermal images of energy facilities, as well as a tool used for the construction and implementation of an efficient fault analysis on the 10/0.4 kV service transformer.

A UAV Intelligent System for Greek Power Lines Monitoring

“…Another challenge lies in removing the noise and artifacts that usually impact the collected imagery in the field. Although several deep learning thermal-image-based defect detection approaches in electrical equipment and energy distribution networks have been developed, very limited ones focus on the recognition of power lines instead of other key components such as power transformers, circuit breakers, surge arresters, cutout switch bus fuse connections, and insulations [24][25][26][27]. He et al [28] used the temperature information of infrared images to diagnose the fault of power transmission lines, applying the cellular automaton technique for separating the regions of interest and the background, the Hessian matrix for detecting the image transmission lines, and thresholding temperature information for deciding the power lines' defects.…”

Techniques of infrared thermography for condition monitoring of electrical power equipment

“…The major issue with K-means is linked to its vulnerability to outliers. Mlakic et al [41] applied modern machine learning tools for power asset monitoring inspired to present a fault identification technique in transformers using deep learning tools for the analysis of IR images of 10/0.4 kV distribution transformers. The authors applied the AlexNet CNN-based learning algorithm in Matlab for processing the raw image datasets.…”

Linear discriminate analysis and k-nearest neighbor based diagnostic analytic of harmonic source identification

“…Generally, the main goal of LDA is to discriminate against the observed features by maximizing the posterior probability [53]. KNN is a common machine learning algorithm that has commonly applied in many engineering applications such as feature selection, pattern recognition, and fault identification [54,55]. Among the modern algorithms, KNN can usually perform faster to achieve the results.…”