Frederico H. R. Lopes, R. Zampolo, Rodrigo M. S. Oliveira, Victor Dmitriev
{"title":"Evaluation of transfer learning approaches for partial discharge classification in hydrogenerators","authors":"Frederico H. R. Lopes, R. Zampolo, Rodrigo M. S. Oliveira, Victor Dmitriev","doi":"10.1109/WCNPS56355.2022.9969682","DOIUrl":null,"url":null,"abstract":"Severe deterioration in the insulation system can interrupt the operation of high voltage electrical machines. Concerning hydrogenerators, unexpected interruptions result in important losses to both energy companies and consumers. Recent proposals for automatic partial discharge analysis, an effective approach to prevent failure in high voltage equipment, are mainly based on deep learning. Their performance, however, relies upon the availability of huge, and commonly expensive, datasets. Besides, if models are intended to be trained from scratch, significant computational resources are required. This work compares three fine-tuning strategies applied to a pre-trained convolutional neural network for partial discharge classification. We use phase-resolved partial discharge data, obtained during normal operation of hydrogenerators at Tucuruí (Pará, Brazil) power plant, to re-train the last layer of a deep classifier originally conceived to identify partial discharges in a different context. Our results demonstrate that effective transfer learning is achieved by using cross-validation and data augmentation techniques.","PeriodicalId":120276,"journal":{"name":"2022 Workshop on Communication Networks and Power Systems (WCNPS)","volume":null,"pages":null},"PeriodicalIF":0.0000,"publicationDate":"2022-11-17","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":"0","resultStr":null,"platform":"Semanticscholar","paperid":null,"PeriodicalName":"2022 Workshop on Communication Networks and Power Systems (WCNPS)","FirstCategoryId":"1085","ListUrlMain":"https://doi.org/10.1109/WCNPS56355.2022.9969682","RegionNum":0,"RegionCategory":null,"ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":null,"EPubDate":"","PubModel":"","JCR":"","JCRName":"","Score":null,"Total":0}
引用次数: 0
Abstract
Severe deterioration in the insulation system can interrupt the operation of high voltage electrical machines. Concerning hydrogenerators, unexpected interruptions result in important losses to both energy companies and consumers. Recent proposals for automatic partial discharge analysis, an effective approach to prevent failure in high voltage equipment, are mainly based on deep learning. Their performance, however, relies upon the availability of huge, and commonly expensive, datasets. Besides, if models are intended to be trained from scratch, significant computational resources are required. This work compares three fine-tuning strategies applied to a pre-trained convolutional neural network for partial discharge classification. We use phase-resolved partial discharge data, obtained during normal operation of hydrogenerators at Tucuruí (Pará, Brazil) power plant, to re-train the last layer of a deep classifier originally conceived to identify partial discharges in a different context. Our results demonstrate that effective transfer learning is achieved by using cross-validation and data augmentation techniques.