Full length article
Deep branch attention network and extreme multi-scale entropy based single vibration signal-driven variable speed fault diagnosis scheme for rolling bearing

https://doi.org/10.1016/j.aei.2022.101844Get rights and content

Abstract

In view of the difficulty in measuring the speed signal and integrating the vibration and speed information flexibly in actual variable speed bearing fault diagnosis, a single vibration signal-driven variable speed intelligent fault diagnosis scheme for rolling bearings is developed to guarantee the reliability and safety of the equipment in this paper. In the proposed fault diagnosis scheme, the extreme multi-scale entropy (EMSEn) of the raw vibration signal is employed as the alternative characterization parameter of the speed information, and an intelligent diagnosis model named deep branch attention network (DBANet) is developed to integrate the vibration and speed information more flexibly. The developed DBANet contains 2 parallel and relatively independent forward propagation channels, and the attention mechanism is introduced into the deep architecture at branch level to adjust the importance of different branches, which endow the model with the ability of fusing the vibration and speed information autonomously. The effectiveness of the proposed method is verified by experiments, and the experimental results show that, compared with the methods relying on external information fusion, the suggested DBANet can integrate the vibration and speed information more flexibly. Besides, in the case of no speed signal, the proposed diagnosis scheme can achieve more outstanding results compared with the methods of using other multi-scale entropy features as the alternative characterization parameter of the speed information.

Introduction

Rolling bearing is the basic component of various mechanical equipment, and it directly affects the overall operation safety and performance of the equipment [1], [2], [3]. In the long-term operation process, once the rolling bearing fails, it will inevitably reduce the reliability of the whole equipment, and even lead to unexpected accidents [4], [5], [6]. Therefore, it is of great significance to study advanced fault diagnosis approach for rolling bearings [7], [8], [9], [10], [11].

The key of bearing fault diagnosis is to extract the fault-related features from the collected vibration signals, and various advanced signal processing technologies, such as wavelet transform (WT) [12], [13], empirical mode decomposition (EMD) [14], [15] and variational mode decomposition (VMD) [16], [17], et al, have been applied in past studies. Nevertheless, bearings always work with variable rotational speed in actual scenarios due to torque ripple, vibrational load, speed fluctuation and time-varying meshing stiffness, et al. factors [18]. For the above reason, the order-tracking technology with the help of tachometer has become the mainstream approach of variable speed fault diagnosis for a long time [19], [20], [21]. In recent years, to get rid of the dependence on tachometers, scholars have explored the variable speed fault diagnosis methods that do not rely on tachometers [22], [23], [24], [25]. On the whole, these methods have made certain achievements, but still suffer from the non-stationary characteristics of the signal and the resolution of the time–frequency representation, resulting in limited actual performance.

In addition to the methods based on fault characteristic frequency analysis, intelligent diagnosis approaches also play an important role in variable speed fault diagnosis [26], [27], especially the deep learning (DL)-based methods emerging in recent years [28], [29], [30], [31]. Compared with the signal processing-based methods and the conventional machine learning-based methods, the dl-based method is less affected by the non-stationary characteristics of the signal, and can automatically abstract the fault features during model optimization, which greatly reduces the dependence on artificial feature extraction, and makes the entire diagnosis procedure more objective. Nevertheless, in spite of the achievements the dl-based approaches have made, problem remains. Firstly, although the deep neural networks enjoy excellent feature abstraction ability, the method only relying on a single vibration signal still cannot ensure a satisfactory diagnosis accuracy in the case of speed fluctuation, and the reality that the speed signal is difficult to measure is still one of the main factors that limit the diagnosis performance. Secondly, most of the existing diagnostic models are single input mode. Even if the rotational speed information is available, the vibration and rotational speed information still need to be integrated externally in advance, and cannot be integrated deeply and automatically through the model, resulting in insufficient information utilization.

To address the aforementioned issues, a single vibration signal-driven variable speed intelligent fault diagnosis scheme for rolling bearings is developed in this paper. In the proposed diagnosis scheme, the EMSEn [32] of the vibration signal is employed as the alternative characterization parameter of the speed signal, which can reduce the dependence on tachometer to a certain degree. In addition, to overcome the limitations of external information integration, the DBANet is developed as the classifier. The proposed DBANet contains 2 feature branches that can deal with the vibration and speed information respectively, and the attention mechanism is expanded to branch level to adjust the importance of different branch features, which contributes to the autonomous fusion of the vibration and speed information and can further improve the flexibility of variable speed fault diagnosis. Two variable speed bearing data sets are employed during the experiment, and the experimental results indicate that the developed DBANet can integrate the vibration and speed information more flexibly compared with the common external information fusion approaches. Besides, utilizing the EMSEn as the alternative characterization parameter of the speed information can achieve more outstanding results than other multi-scale entropy features.

The main contributions of this paper are summarized as follows:

  • Aiming at the problem that the speed signal is difficult to measure under variable speed conditions and the conventional fault diagnosis models are difficult to fuse the vibration and speed information flexibly, the single vibration signal-driven variable speed fault diagnosis scheme for rolling bearing based on DBANet and EMSEn is proposed.

  • In the suggested diagnosis scheme, to endow the model with the ability of fusing the vibration and speed information autonomously, the DBANet that contains 2 parallel and relatively independent forward propagation channels is developed, and the attention mechanism is introduced into the deep architecture at branch level to adjust the importance of different branches.

  • To reduce the dependence on the tachometer under variable speed conditions, the EMSEn of the vibration signal is employed as the alternative characterization parameter of the speed information. The effectiveness of the developed variable speed fault diagnosis scheme is verified by experiments.

The remainder of this paper is structured as follows: Section 2 presents a brief review of the EMSEn. The developed DBANet and the related single vibration signal-driven variable speed fault diagnosis scheme are given in Section 3 and Section 4, respectively. The experimental verification of the developed method is performed in Section 5. The conclusions are drawn in the final Section.

Section snippets

Brief review of the EMSEn

In practical variable speed scenarios, the installation of the tachometer often makes certain modifications to the equipment, such as setting coding disks and auxiliary brackets, which makes it is difficult to be promoted. As an effective measure of time series complexity, entropy and multi-scale entropy based algorithms [33], [26], [34] have been successfully applied in variable speed fault diagnosis as fault feature in past studies. Nevertheless, the classical multi-scale entropy algorithms

Dual branch fusion strategy for vibration and speed information

To make full use of the vibration and speed information, and improve the flexibility of information integration. Two separate forward propagation channels are constructed for vibration and speed information respectively, and the fusion of the vibration and speed information is embedded into the model by applying the branch structure in this work. Therefore, it is no longer necessary to integrate the features outside the model in advance. Moreover, in order to coordinate the contribution of the

The suggested single vibration signal-driven variable speed fault diagnosis scheme

In the case of speed signal is not available, to ensure the accuracy of variable speed fault diagnosis, this paper employs the EMSEn of the vibration signal as an alternative characterization parameter of the speed signal for auxiliary analysis. The original vibration signal and the calculated EMSEn are used as the inputs of different branches of the DBANet to realize the deep autonomous fusion of the information. The overall flow of the suggested variable speed fault diagnosis scheme is

Description of the experimental data

In this section, the variable speed rolling bearing data provided in literature [36] is employed. As shown in Fig. 6, the SpectraQuest mechanical fault simulation test bench (MFS-PK5M) used in the experiment is driven by a motor, and the speed can be controlled through the AC driver. Two ER16K ball bearings (The detailed parameters are shown in Table 1) are installed at both ends of the shaft, of which the left side is the healthy bearing and the right side is the test bearing. Bearings in

Conclusions

In this work, aiming at the problem that the speed signal is difficult to measure under variable speed conditions and the conventional fault diagnosis models are difficult to fuse the vibration and speed information flexibly, the single vibration signal-driven variable speed fault diagnosis scheme based on EMSEn and DBANet is proposed. In the suggested diagnosis scheme, the EMSEn of the vibration signal is calculated as the alternative characterization parameter of speed signal, and the DBANet

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.

Acknowledgements

This work is supported by the National Natural Science Foundation of China under Grant (52075310).

References (41)

Cited by (26)

View all citing articles on Scopus
View full text