ReviewConstruction of health indicators for condition monitoring of rotating machinery: A review of the research
Introduction
Mechanical equipment has been widely used in the energy, aerospace, and manufacturing industries. Hence, it is vital to ensure proper working conditions of mechanical equipment to eliminate property and personnel loss caused by accidental breakdown. As a fertile research field that addresses this issue, CM has attracted increasing attention from scholars and industrial practitioners over the last two decades. As a major component of predictive maintenance, CM involves observing the parameters that reflect the condition of machinery (vibration, current, and temperature) to identify significant changes that indicate developing faults. Thus, the use of CM enables maintenance to be scheduled or other actions can be taken to prevent damage and catastrophes. Moreover, CM has a unique benefit in that conditions that would shorten the average lifespan can be addressed before they develop into major failures.
As important mechanical components that transmit rotation, bearings and gears are vital in mechanical equipment. The CM of bearings and gears is also of great importance and requires the development of an efficient technique to monitor their working health conditions. The requirements of such a CM technique for bearings and gears include two aspects: 1) detection of the incipient faults of the bearings and gears to prevent unexpected fatal damage and 2) tracking and characterizing the health condition evolution trend over time, which can provide a clear and quantifiable view of the object status at a specific time.
Over the past decades, numerous studies have been performed to detect incipient faults from signals collected by vibration sensors based on signal processing techniques. The underlying principle of these methods is to exploit the spectrum-based approach, such as envelope spectrum analysis (Brown & Paper, 1989) which extracts the fault features in the frequency domain after implementing an advanced signal processing method. Denoising has been applied as a preprocessing method for weak incipient faults caused by intense background noise or interference from other mechanical parts, such as self-adaptive noise cancellation (SANC) (Ho & Randall, 2000), time-synchronous averaging (TSA) (Randall, Sawalhi, & Coats, 2011) and cepstrum pre-whitening (CPW) (Borghesani, Pennacchi, Randall, Sawalhi, & Ricci, 2013). Meanwhile, bandpass filtering, as a robust interference elimination strategy, has been developed worldwide, both in academia and industry. The main issue is how to determine the filter parameters (bandwidth and center frequency) to acquire the optimal filtered signal that contains informative fault features; the most representative method for this purpose is spectral kurtosis, which was first introduced by Antoni (Antoni & Randall, 2006) to select the proper demodulation band for the application of envelope-based techniques. The main idea is to supervise the parameter selection of the bandpass filter using kurtosis because of its excellent characteristic of indicating the impulsiveness of the filtered signal. Impulsive noise always excites high-level kurtosis and consequently makes the kurtosis-based demodulation method ineffective for signals that contain impulsive noise (Kang, et al., 2021). To solve this problem, the cyclostationary processes have been introduced to characterize the fault features of rotating mechanisms (Jérôme Antoni et al., 2004, Bouillaut and Sidahmed, 2001, Capdessus et al., 2000, McCormick and Nandi, 1998; Robert B Randall, Antoni, & Chobsaard, 2001), and recently some novel cyclostationary index (Mauricio et al., 2020, Schmidt et al., 2020) was proposed to weaken the impulsive noise interference and to quantify the fault features more effectively to obtain the optimal demodulation band. Here, fault detection is implemented based on the combination of fault-indicator-guided bandpass filtering and envelope spectrum analysis. In addition, sparse representation theory has been introduced and modified to denoise and reconstruct pure fault signals (Zhang et al., 2020, Zhao et al., 2020, Zhou et al., 2020; Haoxuan Zhou, Wen, Zhang, Huang, & Dong, 2021). The sparse constrained optimization theory realizes the reconstruction and extraction of the fault-related components in the signal from the perspective of optimization based on the sparse characteristics of the fault signal and then detects the fault by spectrum or envelope spectrum analysis. Time-frequency-based signal processing methods such as empirical modal decomposition (EMD) wavelet package transformation (WPT), which is utilized in fault detection of rotating machinery (RM), can be found in the literature (Li et al., 2020, Minhas et al., 2021, Wang and Shao, 2020). Time-frequency methods such as short time–frequency transformation (STFT) are introduced to extract the time–frequency ridgeline and estimate the speed when the rotation speed is time-varying (Peeters et al., 2019, Wang et al., 2020, Wang et al., 2020).
Signal processing methods such as spectrum analysis and envelope spectrum analysis can realize fault detection of RM, such as bearings and gears, and the spectrum of the vibration signal is applied as the input for prognosis using other machine learning methods. However, the single (envelope) spectrum method is not the best approach for characterizing the evolution trend of the health condition over time, because the mechanism of the degradation information that may appear in the spectrogram is not completely clear. While, it is important to characterize the health state of RM using a health indicator (HI) or index (a parameter obtained from the monitoring signal such as vibration) to achieve incipient fault detection, degradation assessment, and prognosis. It is worth noting that academia and industry introduced a series of well-known HIs built on time-domain vibration signals into the CM of RM many years ago, including kurtosis and root-mean-square (RMS). As mentioned above, the kurtosis index has been widely used and developed based on envelope spectrum analysis.
With the rapid development of technologies based on envelope spectrum analysis, numerous HIs (such as the negative entropy, Gini index, smoothness index, and second-order cyclostationary index series) have been proposed, which can also be used to characterize the current state of health of RM and the evolution of faults. Simultaneously, with the increasing complexity of the application scenarios of RM and equipment, the HIs construction method based on the knowledge of the fault mechanism cannot be effectively applied in some cases in which the fault mechanism is not clear or the fault mechanism is poorly generalized in extreme cases. Scholars are targeting the burgeoning field of machine learning, where powerful intelligence tools such as deep learning (Goodfellow et al., 2016, LeCun et al., 2015) offer an intelligent, failure-mechanism-free, data-driven approach to HIs construction, and some model and data-driven HIs construction methods have been proposed (Chen et al., 2020, Sun and Liu, 2022, Xiang et al., 2022) to achieve more intelligent construction processes.
The proposed HIs are related to different technical approaches, and HIs derived from the same technical approach have different application characteristics and scopes. Scholars have conducted reviews of CM of RM, which partly include HIs, to elucidate and standardize the current state of research on HIs construction. For instance, Mitchell et al.(Lebold, McClintic, Campbell, Byington, & Maynard, 2000) reviewed the statistical features (indices or indicators) used to measure the vibration level compared to a threshold value that indicates a failed gearbox condition. Preprocessing of each statistical feature, such as the “residual signal,” was also discussed in this article. Lei et al.(Yaguo Lei, Lin, Zuo, & He, 2014) reviewed the CM and fault diagnosis of planetary gearboxes and comprehensively reviewed numerous techniques, including structural and dynamic characteristics, monitoring, and diagnosis of RM. Goyal et al.(Goyal, Vanraj, Pabla, & Dhami, 2016) reviewed the HIs of a fixed gearbox and addressed the HIs domain-wise. Wang et al.(Dong Wang, Tsui, & Miao, 2018) reviewed the vibration-based HI of rolling bearings and gears from mechanical signal processing, modeling, and machine learning. Antoni et al.(J. Antoni & Borghesani, 2019) briefly reviewed common statistic HIs and categorized them into two groups dedicated to characterizing the non-Gaussian and non-stationary properties of the monitored vibration signal. These review works on RM provided a rough technical blueprint for HI-based monitoring. Nevertheless, there are still some aspects that previous articles have not comprehensively summarized, and comprehensive and detailed reviews of these HIs are still lacking.
- (1)
Many reviews of HI methods associated with fault mechanisms are incomplete and describe only part of the work from a technical perspective rather than providing readers with a comprehensive blueprint for technological development.
- (2)
There is still a lack of review work on intelligent HIs construction methods, especially in terms of deep learning techniques, and most review works focus on HIs derived from fault mechanisms.
- (3)
Some new and exciting approaches for constructing HIs associated with failure mechanisms have been proposed in the past, and these works have not been effectively included in the previous reviews.
Therefore, it is imperative to review the relevant research to help readers understand the current state-of-the-art techniques related to HIs construction of RM and to facilitate the design of an effective solution to address some practical challenges quickly.
To overcome the aforementioned limitations, this article attempts to provide a comprehensive review of HIs construction for RM. First, in contrast to the existing reviews, the concept of HIs of RM is introduced briefly to create a distinctive picture of the actual application and construction principles of HIs for readers. Second, this paper differs from the existing reviews in that they mainly focus on specific application objects such as bearings or gears, whereas this article categorizes the current methods in terms of practical techniques, providing a more comprehensive framework and enabling the inclusion of research work related to different objects. Specifically, the current work is categorized into statistical parameter-based construction methods, signal processing-based construction methods, and machine learning-based construction methods. Finally, the above classification can more effectively encompass the research work with different application objects and show the reader the development of HIs construction methods of RM from the past to the present, which may better inspire the readers of this paper.
The main contributions of this article are as follows:
- (1)
The basic concepts and theories of HIs construction for RM are introduced and categorized into statistic parameter-based methods, signal preprocessing-based methods, and machine learning-based methods, presenting a comprehensive overview of the current research from the perspective of the techniques employed.
- (2)
Recent studies on deep learning approaches that utilize the network training process based on healthy monitoring data are reviewed, which would provide the readers with a new and more comprehensive framework of the existing research for HIs construction for RM.
- (3)
Future challenges and potential directions of HIs construction for RM are identified, potentially providing valuable insights to potential newcomers and seasoned researchers that can inform future works.
This paper thoroughly summarizes and discusses the research on HIs construction methods for rolling bearings and gears. The remainder of this paper is organized as follows. To enable readers to understand the concept of HIs and related issues quickly, Section 2 provides a brief overview of HIs used in CM of RM. Section 3 reviews the statistical parameters used as HIs of RM, such as traditional statistical parameters, sparsity-based parameters, and complexity-based parameters, as well as the performance of these parameters. Section 4 summarizes the signal preprocessing-based HIs construction method. Section 5 outlines the machine learning techniques used for HIs construction and provides a comprehensive review of the HIs constructed from data driving. Section 6 discusses the pros and cons of typical HIs, challenges and future research prospects. Finally, the conclusions are drawn in Section 7.
The notational conventions are as follows. The time-domain monitoring signal is denoted as X, and is its discrete form. N is the discrete signal length, X. If there are no special instructions, the monitoring signals mentioned in the article are vibration signals, because they are currently the most widely used monitoring signals in CM and RM prognosis.
Section snippets
A brief overview of HIs
Reliability has been achieved in previous research by consistently monitoring faults using strategies and techniques capable of preventing or minimizing the effects of these faults (Salameh, Cauet, Etien, Sakout, & Rambault, 2018). CM involves assessing the current health status of equipment in real-time using machine status signals (such as temperature and vibration). As the main procedure in prognostics and health management (PHM), CM is preliminary and crucial to maintaining machine
Statistical parameter-based HIs
This section briefly introduces and reviews the statistical parameters from four perspectives. The first one is the commonly used traditional statistical parameters, including dimensional and dimensionless parameters. A high-order moment (or cumulant) is introduced because most dimensionless parameters are deduced from it. The second one is the recently well-researched sparsity-based HIs, which include entropy, the Gini index, and Lp/Lq. the third one is the complexity-based HIs that measure
Signal preprocessing-based HIs
For the original monitoring signals, such as vibration, it is unavoidable to receive interference from the signal components not correlated with the fault by constructing time-domain indicators describing the fault-related characteristics from the statistical characteristics to characterize the health state of the RM. However, the domain space characteristics of the original time-domain are limited, and there may be some fault-related components or features that cannot be well highlighted.
Machine learning-based HIs
Machine learning methods involve thinking more from the perspective of data; therefore, regardless of whether the object is a bearing or a gear, the differences in the nature of the method are relatively small compared to those mentioned in Section 4 related to HIs construction for fault mechanisms. Thus, this paper continues to divide and review this research content in terms of technical and methodological perspectives.
Common statistical parameters derived from the time or frequency domain
Discussion and research prospects
The previous sections have reviewed the literature on HIs construction for RM, from basic concepts to well-developed techniques. The characteristics and connections of the above methods are summarized in Fig. 10.
The figure shows the development of ideas and the characteristics of each technique of the research covered in the paper. The three primary techniques can be divided into two categories regarding whether they require expert experience and fault mechanisms, which are also referred to in
Conclusion
The CM of bearings and gears is a vital process in RM health management. The most common method of achieving health state monitoring involves utilizing HIs. This paper provided a comprehensive review of research on HIs construction for RM, from fundamental statistical parameter-based HIs to machine learning-based HIs. Influential research was also analyzed in great detail. Finally, recommendations for future research were presented. This systematic review will enhance technical understanding of
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
This research was supported in part by the National Key Research and Development Program of China (No. 2020YFB1710002), the National Natural Science Foundation of China (No. 51775409), and the Equipment Pre-research Fund of China (No. 61420030301).
References (220)
- et al.
Envelope analysis of rotating machine vibrations in variable speed conditions: A comprehensive treatment
Mechanical Systems and Signal Processing
(2017) - et al.
Rolling element bearing fault identification using a novel three-step adaptive and automated filtration scheme based on Gini index
ISA Transactions
(2020) Cyclic spectral analysis of rolling-element bearing signals: Facts and fictions
Journal of Sound and Vibration
(2007)The infogram: Entropic evidence of the signature of repetitive transients
Mechanical Systems and Signal Processing
(2016)- et al.
Cyclostationary modelling of rotating machine vibration signals
Mechanical Systems and Signal Processing
(2004) - et al.
A statistical methodology for the design of condition indicators
Mechanical Systems and Signal Processing
(2019) - et al.
The spectral kurtosis: Application to the vibratory surveillance and diagnostics of rotating machines
Mechanical Systems and Signal Processing
(2006) - et al.
The relationship between kurtosis- and envelope-based indexes for the diagnostic of rolling element bearings
Mechanical Systems and Signal Processing
(2014) - et al.
Application of cepstrum pre-whitening for the diagnosis of bearing faults under variable speed conditions
Mechanical Systems and Signal Processing
(2013) - et al.
Cyclostationary approach and bilinear approach: Comparison, applications to early diagnosis for helicopter gearbox and classification method based on HOCS
Mechanical Systems and Signal Processing
(2001)
A smoothness index-guided approach to wavelet parameter selection in signal de-noising and fault detection
Journal of Sound and Vibration
Cyclostationary processes: Application in gear faults early diagnosis
Mechanical Systems and Signal Processing
Anomaly detection for wind turbines based on the reconstruction of condition parameters using stacked denoising autoencoders
Renewable Energy
Novel adaptation of the demodulation technology for gear damage detection to variable amplitudes of mesh harmonics
Mechanical Systems and Signal Processing
Hidden semi-Markov model-based methodology for multi-sensor equipment health diagnosis and prognosis
European Journal of Operational Research
A segmental hidden semi-Markov model (HSMM)-based diagnostics and prognostics framework and methodology
Mechanical Systems and Signal Processing
Bearing degradation process prediction based on the PCA and optimized LS-SVM model
Measurement
Automatic characteristic frequency association and all-sideband demodulation for the detection of a bearing fault
Mechanical Systems and Signal Processing
A recurrent neural network based health indicator for remaining useful life prediction of bearings
Neurocomputing
Optimisation of bearing diagnostic techniques using simulated and actual bearing fault signals
Mechanical Systems and Signal Processing
Condition assessment for the performance degradation of bearing based on a combinatorial feature extraction method
Digital Signal Processing
An adaptive method for health trend prediction of rotating bearings
Digital Signal Processing
Development of a gear vibration indicator and its application in gear wear monitoring
Mechanical Systems and Signal Processing
Ensemble of data-driven prognostic algorithms for robust prediction of remaining useful life
Reliability Engineering & System Safety
Residual life predictions for ball bearings based on self-organizing map and back propagation neural network methods
Mechanical Systems and Signal Processing
Analysing RMS and peak values of vibration signals for condition monitoring of wind turbine gearboxes
Renewable Energy
A geometrical investigation on the generalized l p /l q norm for blind deconvolution
Signal Processing
Sparse filtering with the generalized lp / lq norm and its applications to the condition monitoring of rotating machinery
Mechanical Systems and Signal Processing
Fault diagnosis of rotating machinery based on multiple ANFIS combination with GAs
Mechanical Systems and Signal Processing
Condition monitoring and fault diagnosis of planetary gearboxes: A review
Measurement
Machinery health prognostics: A systematic review from data acquisition to RUL prediction
Mechanical Systems and Signal Processing
A novel method using adaptive hidden semi-Markov model for multi-sensor monitoring equipment health prognosis
Mechanical Systems and Signal Processing
L-Kurtosis and its application for fault detection of rolling element bearings
Measurement
Modeling of the safe region based on support vector data description for health assessment of wheelset bearings
Applied Mathematical Modelling
An intelligent approach to machine component health prognostics by utilizing only truncated histories
Mechanical Systems and Signal Processing
Graph-based structural change detection for rotating machinery monitoring
Mechanical Systems and Signal Processing
Generalized Anomaly Detection Algorithm Based on Time Series Statistical Features
Implementing Industry
An Artificial Neural Network Approach for Early Fault Detection of Gearbox Bearings
IEEE Transactions on Smart Grid
A threshold self-setting condition monitoring scheme for wind turbine generator bearings based on deep convolutional generative adversarial networks
Measurement
Multisensor Feature Fusion for Bearing Fault Diagnosis Using Sparse Autoencoder and Deep Belief Network
IEEE Transactions on Instrumentation and Measurement
The measurement of the inequality of incomes
The Economic Journal
Wind Turbine Gearbox Anomaly Detection based on Adaptive Threshold and Twin Support Vector Machines
IEEE Transactions on Energy Conversion
Cited by (52)
An autoregressive model-based degradation trend prognosis considering health indicators with multiscale attention information
2024, Engineering Applications of Artificial IntelligenceNonlinear weight learning model for incipient fault detection and degradation modelling and its interpretability for fault diagnosis
2024, Mechanical Systems and Signal ProcessingAn orbit-based encoder–forecaster deep learning method for condition monitoring of large turbomachines
2024, Expert Systems with ApplicationsSensitivity analysis by differential importance measure for unsupervised fault diagnostics
2024, Reliability Engineering and System SafetyCompound fault diagnosis of wind turbine bearing under ultra-low speed operations using generalized sparse spectral coherence
2024, Mechanical Systems and Signal ProcessingHybrid augmented network with balance domain window for few-shot fault diagnosis under sharp speed variation
2024, Mechanical Systems and Signal Processing