A preview of this full-text is provided by IOP Publishing.
Content available from Measurement Science and Technology
This content is subject to copyright. Terms and conditions apply.
Measurement Science and Technology
Meas. Sci. Technol. 35 (2024) 075103 (15pp) https://doi.org/10.1088/1361-6501/ad3be0
Improved DBO-VMD and optimized
DBN-ELM based fault diagnosis for
control valve
Dengfeng Zhang1, Chi Zhang1,∗, Xiaodong Han2and Cunsong Wang1
1Institute of Intelligent Manufacturing, Nanjing Tech University, Nanjing 210009, People’s Republic of
China
2Institute of Telecommunication and Navigation Satellites, China Academy of Space Technology,
Beijing, People’s Republic of China
E-mail: zc18336113545@163.com and zhdf@njtech.edu.cn
Received 5 February 2024, revised 2 April 2024
Accepted for publication 8 April 2024
Published 16 April 2024
Abstract
Control valves play a vital role in process production. In practical applications, control valves
are prone to blockage and leakage faults. At the small control valve openings, the vibration
signals exhibit the drawbacks of signicant interference and weak fault characteristics, which
causes subpar fault diagnosis performance. To address the issue, a diagnostic model based on
optimized variational mode decomposition (VMD) and improved deep belief network-extreme
learning machine (DBN-ELM) is proposed. Firstly, good point set population initialization,
nonlinear convergence factor, and adaptive Gaussian–Cauchy mutation strategies are applied in
the dung beetle optimization algorithm (DBO) to escape local optima. Then, the improved DBO
(IDBO) is used to optimize VMD parameters to obtain a series of modal components. Next, the
generalized dispersion entropy (GDE) is formed by the combination of generalized Gaussian
distribution and rened composite multiscale uctuation-based dispersion entropy. The
maximum correlation coefcient modal components are applied to extract GDE. Finally, the
IDBO is applied to optimize the parameters of the DBN-ELM network to improve the
classication performance of control valve faults. The comparative experiment results
demonstrate that the proposed model can extract effective features and the diagnostic accuracy
reaches 99.87%.
Keywords: control valve, fault diagnosis, generalized Gaussian distribution (GGD),
rened composite multiscale uctuation-based dispersion entropy
1. Introduction
As important actuactors, the control valves are widely applied
in process industries such as petrochemical production, chem-
ical industry, and environmental protection [1]. The faults of
control valves not only affect their performance, but also slow
down the progress of chemical engineering production, and
even lead to safety accidents [2]. Multiple types of faults occur
in the control valves including blockage, leakage, stickiness of
the valve stem, and spring failure [3,4]. The most prevalent
∗Author to whom any correspondence should be addressed.
faults are blockage and leakage [5]. For example, the valve
leakages account for 54% of UF6 gas accidents [6]. Therefore,
develop a fault diagnosis system for the control valve that
holds signicant application value.
Currently, the model-based and data-driven methods for
control valve fault diagnosis have become dominant [7]. The
model-based methods detect features by establishing fault
mechanism models. Andrade et al [8] developed a nonlinear
autoregressive neural network model based on the decision
trees models for the fault diagnosis. Han et al [2] proposed a
canonical variate analysis model and utilized the square of the
residual Mahalanobis distance as a diagnosis index to improve
the model performance. However, many important parameters
1 © 2024 IOP Publishing Ltd