Abstract
Recently, infrared technology is increasingly used in condition monitoring of external insulations, e.g., bushing, reactor and potential and current transformers in substation. Even through the infrared technology can detect the failures of external insulation due to overheating in fast response time. However, the massive infrared images need to be manually analyzed by human for fault classification, which is a very time and consuming task. There are also technical trials in applying intelligent image recognitions technology for sorting out infrared images, but these smart technologies are mainly based on machine learning framework suitable for shape recognition and only very few types of faults can be automatically figured out. In this paper, an improved automatic fault diagnosis method was designed based on Mask Region convolutional neural network (Mask R-CNN) for infrared image segmentation combined with perceptual hash algorithm for fault characteristic recognition. This intelligent method consists of three steps, i.e., the normalization of infrared images according to grayscale, the fault region detection of infrared images by using Mask R-CNN and the collection of fault spectrum through the similarity recognition by perceptual hash. With the proposed joint algorithm on infrared images for external insulation with condition in known, it is confirmed that the accuracy of fault recognition reaches more than 90%. This automatic fault detection algorithm provides a desirable solution for the field application of infrared image-based diagnosis for external insulations.
Published Version
Talk to us
Join us for a 30 min session where you can share your feedback and ask us any queries you have
Schedule a call