Convolutional neural networks-based valve internal leakage recognition model. (June 2021)

• Record Type: Journal Article
• Authors: Zhu, Shen-Bin; Li, Zhen-Lin; Li, Xiang; Xu, Hao-hao; Wang, Xi-ming
• Is Part Of: Measurement. Volume 178(2021)
• Publication Date: 2021-06
• Subjects: Valve -- Internal leakage -- Acoustic emission -- Leakage recognition -- Convolutional neural networks; Weights and measures -- Periodicals; Measurement -- Periodicals; Measurement; Weights and measures; Periodicals; 530.8
• DOI: 10.1016/j.measurement.2021.109395
• Languages: English
• ISSNs: 0263-2241

Abstract

Highlights: CNN is proposed to diagnose the valve internal leakage. The collected multiple working condition datasets ensures the sample diversity. CNN shows stronger learning ability than ANN, RBF-SVM and K-Medoids. The maximum prediction error of CNN is less than 3%. The paper has a guiding significance for diagnosing other fluid leakage. Abstract: The internal leakage signal of a valve is generally weak, and it is vulnerable to complex background noise. Due to the limitations of traditional internal leakage diagnosis methods and models, it is difficult to effectively evaluate a valve state under complex working conditions. Recognising this challenge, convolutional neural networks (CNN) is proposed to recognise valve internal leakage, which uses the power spectral density images of internal leakage and non-leakage signals under multiple working conditions as input. The experimental results show that the proposed models effectively recognise a internal leakage or non-leakage signal, and the maximum prediction error is less than 3%, which can be used a new method for valve leakage diagnosis.