Fault Diagnosis of Less Oil Equipment Based on Domain Adversarial Transfer Learning Convolutional Neural Network

• Published in: 2024 IEEE International Conference on High Voltage Engineering and Applications (ICHVE) pp. 1 - 4
• Main Authors: Wang, Shuai, Mu, Hai-Bao, Liu, Yan-Qi, Zhou, Jin-Ming, Lin, Hao-Fan, Zhang, Guan-Jun
• Format: Conference Proceeding
• Language: English
• Published: IEEE 18.08.2024
• Subjects: Accuracy; convolutional neural networks; domain adversarial transfer learning; Fault diagnosis; Feature extraction; High-voltage techniques; less oil equipment; Neural networks; Oils; Transfer learning
• ISSN: 2474-3852
• DOI: 10.1109/ICHVE61955.2024.10676075

Machine learning has made significant progress in main equipment fault diagnosis based on DGA data. However, due to characteristics such as less oil content, rapid fault progression, limited DGA data samples, and difficulty in obtaining fault samples, there is insufficient assessment methods for less oil equipment. This paper proposes a novel fault recognition method, domain adversarial transfer learning convolutional neural networks (DATLCNN). The proposed DATLCNN consists of a feature extractor based on residual convolutional neural networks (RCNN), a label classifier, and a domain discriminator. Through domain adversarial transfer learning, DATLCNN transfers the diagnostic knowledge learned by RCNN from main equipment DGA data to less oil equipment, achieving high-precision fault diagnosis for such equipment. Experimental results indicate that compared to other methods, DATLCNN achieves an accuracy of 86.6% with limited on-site less oil equipment DGA samples, demonstrating the effectiveness and accuracy of the proposed method.