Elsevier

Measurement

Volume 165, 1 December 2020, 108129
Measurement

An intelligent fault diagnosis approach based on Dempster-Shafer theory for hydraulic valves

https://doi.org/10.1016/j.measurement.2020.108129Get rights and content

Highlights

  • An intelligent fault diagnosis approach based on DS theory is proposed.

  • The conflict problem of the information source in the approach is solved.

  • The experimental results show effectiveness of proposed approach.

Abstract

Detecting faults in hydraulic valves are of great significance to improve the reliability and security of the whole hydraulic system. However, it is difficult to detect multiple faults in hydraulic valves using existing approaches due to closed structural components and complex hydraulic system itself. Therefore, an intelligent fault diagnosis approach based on Dempster-Shafer (DS) theory is proposed specifically for detecting several faults occurred in hydraulic valves. Actually, it is classified in the ensemble learning in terms of the information fusion theory. In this approach, signal segments containing fault information are selected to structure sample sets firstly. Then sample sets are simultaneously fed into the single classifier including long short-term memory networks (LSTM), convolutional neural network (CNN) and random forests (RF). Through learning spontaneously in these intelligent classification approaches, fault features are concluded and the probability of each type fault is respectively revealed. All probabilities are constructed as basic probability assignment (BPA) functions, which are further calculated in the information fusion process in terms of DS theory. Finally, the fault types are identified by the final fusion results. Experimental investigations are performed to validate performance of the present approach (taken a solenoid controlled pilot operated directional valve as an example). It is shown that the average accuracy ratio of proposed intelligent fault diagnosis approach is 98.5% for six fault types detection. The study does provide an effective access to detect faults in hydraulic valves.

Introduction

Hydraulic valves are the mechanical (or electrical) to fluid interface in hydraulic systems, so their performance should be under scrutiny, especially when system faults occur. Due to characteristics of the closed structure, it is necessary to explore fault diagnosis approaches to detect faults in hydraulic valves in terms of vibration signals on the valve surface [1].

Conventional fault diagnosis technology applied in hydraulic valves is usually dependent on engineering judgments. However, fault diagnoses in terms of engineering judgments in complicated cases are most difficult. Actually, once faults occur in objective valves, fault information is contained in measured vibration signals [2], though it is not easy to be observed especially for complex multiple fault modes or noisy working environment. Therefore, it is considered as a general and effective approach to analyze external measured vibration signals to reveal fault features and identify fault modes. So far, many signal processing approaches have been developed specially for analyzing fault features submerged in vibration signals, such as wavelet packet decomposition [3], local mean decomposition [4], empirical mode decomposition [5], Kalman Filter [6], etc. In these approaches, fault feature extraction is very dependent on referred features [7]. However, few referred fault information about hydraulic valves is in previous researches [8]. Therefore, instead of signal processing approaches, intelligent fault diagnosis approaches characterized by learning spontaneously fault features are applied which is suitable for diagnosing faults in hydraulic valves or other hydraulic components.

In the diagnosis fields of lacking fault features, various intelligent fault diagnosis approaches are developed rapidly. For example, a multi-layer feed-forward neural network was utilized to estimate health indicators of the valve based on the analysis toward the supply pressure, actuator vent blockage and diaphragm leakage [9]. Yan et␣al. found that some complex hydraulic pump faults could change the frequency information of the signal characteristics. When the fault features are not obvious, the convolutional neural network has great advantage in such pump faults identification compared to the signal processing approaches [10]. In addition, several other similar researches are carried out [11], [12], [13], [14]. However, it is noteworthy that each intelligent fault diagnosis approach has its own domain of competence, which is not suitable for all fault diagnosis tasks [15]. In some practical applications, the diagnosis results achieved by using a single intelligent fault diagnosis approach may be precarious, for example, the accuracy of some types of faults fluctuates greatly although the overall recognition accuracy is excellent. In order to improve the accuracy and stability of diagnostic results limited by a single intelligent fault diagnosis approach, ensemble learning approaches are proposed and developed as a new branch of intelligent fault diagnosis approaches.

The idea of common ensemble learning approaches is to combine multiple individual intelligent fault diagnosis approaches with bagging, boosting and other rules to form a strong learning model. Consequently, the performance of the learning is greatly improved by ensemble learning compared with a single weak learner (individual intelligent fault diagnosis approach). Several ensemble learning approaches applied in the mechanical system have been researched. Tian et␣al. presented a support vector machine ensemble based on bagging to establish a novel fault diagnosis system [16]. Three ensemble approaches were developed by Amozegar et␣al. using the bagging and boosting rules to diagnose faults of gas turbine engines [17]. In order to improve the accuracy of a final decision made by ensemble learning, some information fusion theories are combined in intelligent fault diagnosis approaches.

Actually, there are various information fusion theories applied in fault ensemble learning approaches, such as the voting [18] and sum. However, the simple information fusion theories are not effective in solving the uncertainty of the information source, which is inevitably involved in the process of decision support system. Therefore, the Dempster-Shafer (DS) theory as a novel information fusion theory is developed, which can create cooperation mechanism to handle uncertain and conflict problems in the decision support system [19], [20], [21]. Xiao proposed a decision making approach, which assembles DS theory and belief entropy. The approach utilizes belief entropy to obtain the basic probability assignments (BPA) functions [22]. Additionally, he also developed a novel reinforced belief (RB) divergence measure approach to measure the discrepancy between BPA functions. The measure approach provides a more convincing solution to measure conflicting evidence for DS theory [23]. Yang et␣al. proposed a fault diagnosis scheme in which independent BPA functions from two different sensors are fused by DS theory applied in three-phase induction motors [24]. Similarly, Gong et␣al. proposed a new fault diagnosis method based on DS theory applied in the main coolant system of nuclear power plant. In their method, BPA functions are constructed by the triangle fuzzy function of symptoms and the relationship between symptoms and faults, and then they are fused so that the final diagnosis result is obtained [25].

This paper aims to develop a novel intelligent fault diagnosis approach based on DS theory, applied in the fault diagnosis of hydraulic valves. The proposed approach not only solves the uncertainty of the information source, but also easily obtains the BPA function in DS theory. In this paper, signal segments containing fault information as sample sets input into three types of single classifiers, subsequently fault features and the probability of each type fault are respectively revealed which are constructed as BPA functions to be further calculated in the fusion, consequently the fault types are identified in terms of the final fusion results.

The remainder content of this paper is organized as follows. Theories of single intelligent fault diagnosis approach and DS theory have been briefly reviewed in Section␣II. The proposed approach is described in Section␣III. Concreteness of the proposed approach in application and experimental results are presented in Section␣IV. Finally, the conclusions are drawn in Section␣V.

Section snippets

Intelligent fault diagnosis approaches

In order to enhance the complementarily between intelligent fault diagnosis approaches, several single intelligent fault diagnosis approaches (also called basic classifiers) are incorporated into the ensemble. Considering the diversity of classifiers, three basic classifiers are selected according to their sensitivity to different fault data, which is carried out by a cut-and-try test with a standard sample containing valve fault information. The three basic classifiers are as following:

CNN is

An intelligent fault diagnosis approach based on ds theory

An intelligent fault diagnosis approach based on Dempster-Shafer (DS) theory is proposed specifically for detecting several faults occurred in hydraulic valves. In the proposed approach, there are three basic classifiers (RF, CNN and LSTM) selected as experts to analysis problems and acquire initial diagnostic results. Due to each expert has own view of diagnosis fault, some disagreement may occurred in the specific fault diagnosis, which is also called conflict problem of the information

An example of fault diagnosis application

In order to verify the performance of the proposed approach for hydraulic valve fault diagnosis, a hydraulic system fault simulation experimental table is built to collect sample data for experimental verification. The specific content of experimental research is discussed in the following section followed by the corresponding results. The proposed approach is executed with MATLAB R2018b and ran on a computer with Intel Xeon Silver 4216 CPU*2/NVIDIA P600 GPU/128 GB RAM.

Conclusion

An intelligent fault diagnosis approach based on DS theory is proposed, in order to detect several faults resulting in the internal leakage of hydraulic valves. Few fault diagnosis approach special for hydraulic valves is in previous researches. In this approach, three basic classifiers (RF, LSTM, CNN) are selected according to their sensitivity to different fault data, and then initial results of the each classifier are fused based on DS theory for obtaining final diagnosis result. The

Declaration of Competing Interest

The authors declare that they have no known competing financial interests or personal relationships that could have appeared to influence the work reported in this paper.

Acknowledgments

The authors are grateful to the support from the National Natural Science Foundation of China (Nos. 51805376, U1709208), the Zhejiang Special Support Program for High-level Personnel Recruitment of China (No. 2018R52034), the Zhejiang Provincial Natural Science Foundation of China (No. LY20E050028), the Wenzhou Basic Scientific Research Foundation of China (No.G20180021) and the Wenzhou Key Innovation Project of Science and Technology(2018ZG023).

References (33)

Cited by (48)

  • Fault diagnosis of pressure relief valve based on improved deep Residual Shrinking Network

    2024, Measurement: Journal of the International Measurement Confederation
View all citing articles on Scopus
View full text