A Survey of Fault Diagnosis and Fault-Tolerant Techniques—Part I: Fault Diagnosis With Model-Based and Signal-Based Approaches
Summary (4 min read)
Fig. 2. Schematic diagram of fault tolerant control
- Real-time fault diagnosis can detect whether the system is faulty, and tell where the fault occurs and how severe the malfunction is.
- A three-part survey paper [10] [11] [12] on fault diagnosis was presented in 2003, respectively from the viewpoint of quantitative mode-based methods, qualitative model-based methods and process history based methods.
- In [14] [15] [16] , comprehensive fault diagnostic methods were reviewed respectively from the data-driven perspective.
- For fault-tolerant control, an early review paper was presented by [27] in 1991, which introduced the basic concepts of fault-tolerant control and analyzed the applicability of artificial intelligence (e.g., neural network and expert systems) to fault-tolerant control systems.
- The first-part and second-part survey papers aim to review the existing fault diagnosis methods and applications within a framework by using the upto-date references.
II. MODEL-BASED FAULT DIAGNOSIS METHODS
- Model-based fault diagnosis was originated by Beard [39] in 1971 in order to replace hardware redundancy by analytical redundancy, and comprehensive results were documented in some well-written books (i.e., see [40, 41] ).
- In model-based methods, the models of the industrial processes or practical systems are required to be available, which can be obtained by using either physical principles or systems identification techniques.
- Based on the model, fault diagnosis algorithms are developed to monitor the consistency between the measured outputs of the practical systems and the model predicted outputs.
- Model-based fault diagnosis methods are reviewed following the four categories: deterministic fault diagnosis methods, stochastic fault diagnosis methods, fault diagnosis for discrete-events and hybrid systems, and fault diagnosis for networked and distributed systems, which are classified in terms of types of the models used.
- 𝑅 𝜔 are the system state, control input, measured output, unexpected actuator fault, component/parameter fault, sensor fault, process disturbance and measurement noises, respectively.
A. Deterministic fault diagnosis methods
- It is indicated from (3) that the residual signal is subjected to both fault signals and disturbance signals (including modelling errors, process disturbances and measurement noises).
- Recently, eigenstructure assignment based fault diagnosis approaches have been applied to vehicles [44] , gas turbine engines [45] , spacecraft [46] and wind turbine systems [47] .
- Recent results on robust fault isolation are developed for nonlinear systems [55, 56] , and various applications such as for aircraft engine [57] , robotic manipulators [58] , and lithium-ion batteries [59] .
- Actually the above observer techniques may be integrated or combined in order to solve engineering-oriented problems.
- Another well-known model-based fault diagnosis is parity relation approach, which was developed in the early of 1980s [80, 81] .
B. Stochastic Fault diagnosis methods
- In parallel with the development of the fault diagnosis for deterministic systems, stochastic approaches were also developed for fault diagnosis in the early 1970s.
- Further researches have led to a couple of modified Kalman filter techniques for fault diagnosis, such as extended Kalman filters, unscented Kalman filters, adaptive Kalman filters, and augmented state Kalman filters.
- Adaptive Kalman filters can be employed to tune process noise covariance matrix, or measurement noise covariance matrix in order to obtain satisfactory fault diagnosis [100, 101] .
- The augmented state Kalman filters can be utilized to simultaneously estimate system states and fault signals [102] .
- The parameter estimation based fault diagnosis methods are very straightforward if the model parameters have an explicit mapping with the physical coefficients.
C. Fault diagnosis for discrete-events and hybrid systems
- In industrial processes, the signals of some dynamic systems switch from one value to another rather than changing their values continuously.
- This kind of systems is called discrete-event systems.
- Fault diagnosis of discrete-event systems was initialized by [117] in 1990s, and the underlying theory of fault diagnosis for discrete-event systems was proposed.
- On the other hand, Petri net has intrinsically distributed nature where the notions of state and action are local, which has been an asset to reduce the computational complexity of solving fault diagnosis problems [122] .
- Monitoring and fault diagnosis for hybrid systems entails challenges due to the fact that the continuous dynamics and discrete event changes are mutually dependent and interacted.
D. Fault diagnosis for networked and distributed systems
- The rapid developments in network technologies have much stimulated the real-time control and monitoring via communication channels, that is called networked control and monitoring, which have valuable advantages such as cost effectiveness, less weight and power requirements, easier for installation and maintenance as well as resources sharing [133] .
- In [136] , least-square filters and Kalman filters were integrated for fault detection, isolation and estimation for network sensing systems.
- In order to monitor the guaranteed quality of service (QoS) of the router and the whole topology, sliding mode observer techniques were employed in [138, 139] for anomaly detection in the transmission control protocol (TCP).
- Very recently, model-based detection and monitoring approaches were addressed in [140, 141] for monitoring potential intermittent connections or faulty nodes for controller area networks (CANs).
- Moreover, applications on unmanned airships [145] and power networks [146] are reported as well.
A. Time-domain signal based methods
- By analyzing the changes of the measured root-mean-square current characteristics between healthy conditions and the situations under single/dual transistor short circuit or open circuit, a fault diagnosis method was developed for power converters of switched reluctance motors.
- In [148] , the absolute value of the derivative of the Park's vector phase angle was used as a fault indicator, which was employed for diagnosing multiple open-circuit faults in two converters of permanent magnet synchronous generators (PMSG) drives for wind turbine applications.
- In [151] , it was shown that, under balanced supply voltage, the phase angle, the magnitude of the negative and zero-sequence currents can be considered as reliable indicators of stator faults in the induction motors.
- Taking advantages of the periodicity of the geared faults, the CK algorithm can identify the position of the local gear fault in the gearbox.
- It is noted that, when converting signals into images, the added noise acts as illumination variation.
B. Frequency-domain signal based methods
- Frequency-domain signal based method is to detect changes or faults by using spectrum analysis tool such as discrete Fourier transformation (DFT).
- One of the most powerful frequency-domain methods for diagnosing motor faults is motor-current signature analysis (MCSA), which utilizes the spectral analysis of the stator current to sense rotor faults associated with broken rotor bars and mechanical balance.
- Recent development of current based spectrum signature analysis for fault diagnosis can be found in [156, 157] .
- In [158] , an acoustic fault detection method was addressed for gear box on the basis of the improved frequency domain blind de-convolution flow.
- Recently in [159] , Fourier spectrum and the demodulated spectra of amplitude envelope were employed to detect and locate multiple gear faults in planetary gearboxes.
C. Time-Frequency signal based methods
- For machines under an unloaded condition, or unbalanced supply voltages, varying load, or load torque oscillations, the measured signals are generally transient and dynamic under the concerned time section.
- Therefore, analysis of the stationary quantities in some cases finds difficult to monitor or detect faults via either a pure time-domain or frequencydomain method.
- STFT method allows determining signal frequency contents of local sections as the signal changes in time, which has been widely applied to detect both stator and rotor faults in inductor motors [160] .
- It is noticed that STFT and WT may suffer some uncertain limitations.
- On the basis of the instantaneous frequencies resulting from the intrinsic-mode functions of the signal being analyzed, HHT method is not constrained by the uncertain limitations with respect to the time and frequency resolutions suffered by some time-frequency techniques (e.g., STFT and WT), which has shown quite interesting performance in terms of fault severity evaluation [163] .
IV. CONCLUSION
- In the first-part survey paper, fault diagnosis techniques and their applications have been comprehensively reviewed from model-based and signal-based perspectives, respectively.
- Specifically, model-based fault diagnosis is reviewed following the categories of fault diagnosis approaches for deterministic systems, stochastic fault diagnosis methods, discrete-event and hybrid system diagnosis approaches, and networked and distributed system diagnosis techniques, respectively.
- Meanwhile, signal-based fault diagnosis is surveyed following the classifications of time-domain, frequency-domain, and time-frequency-domain approaches, respectively.
- The overview on knowledge-based fault diagnosis, hybrid and active fault diagnosis is to be carried out in the second-part review paper, which will complete the whole overview on the fault diagnosis techniques and their applications.
Did you find this useful? Give us your feedback
Citations
1,240 citations
1,173 citations
905 citations
805 citations
760 citations
Cites background from "A Survey of Fault Diagnosis and Fau..."
...In general, the fault diagnosis methods can be classified into model-based, signal-based, knowledgebased and hybrid/active approaches [14]....
[...]
References
9,627 citations
"A Survey of Fault Diagnosis and Fau..." refers background in this paper
...[13], a structured and comprehensive overview of the research on anomaly detection is provided, which is referred to the problem of finding patterns in data that do not conform to...
[...]
3,826 citations
"A Survey of Fault Diagnosis and Fau..." refers background in this paper
...model or frequency-domain input–output model, which is well revisited by the books [40], [41], [82]....
[...]
3,313 citations
2,455 citations
2,416 citations
"A Survey of Fault Diagnosis and Fau..." refers background in this paper
...A number of survey papers were written, for instance, [2]–[38], which are presented in Table I, categorized...
[...]
...of analytical redundancy for model-based fault detection and diagnosis in the early survey paper [2]....
[...]
Related Papers (5)
Frequently Asked Questions (22)
Q2. Why is time frequency decomposition needed for real-time monitoring and fault diagnosis?
Due to the time-varying frequency spectrumof the transient signals, suitable time-frequency decompositiontools are needed for real-time monitoring and fault diagnosis.
Q3. What is the basic fault diagnosis problem?
The basic event-driven fault diagnosis problem is toperform model-based inferring at run-time to determinewhether a given unobservable fault event has occurred or notin the past by using sequences of observable events [118].
Q4. What are the main types of fault diagnosis tools?
A variety of statistical tools, such as generalized likelihoods [93], χ2 testing [94], cumulative sum algorithms [95] and multiple hypothesis test [96], werefurther developed for testing Kalman-filter based residuals inorder to check the likelihood that a particular fault occurs.
Q5. What is the method for diagnosing motor faults?
One of the most powerfulfrequency-domain methods for diagnosing motor faults ismotor-current signature analysis (MCSA), which utilizes thespectral analysis of the stator current to sense rotor faultsassociated with broken rotor bars and mechanical balance.
Q6. What is the important method of fault diagnosis?
Another important stochastic fault diagnosis method isparameter estimation on the basis of system identificationtechniques (e.g., least-square error and its derived methods),which was initialized by [106].
Q7. What are the common models to represent hybrid systems?
Hybrid automata are the most common models to representhybrid systems, which can be utilized to design fault diagnosisalgorithms to detect and isolate faults [127, 128].
Q8. What is the general idea of the distributed fault diagnosis?
A general idea of the distributedfault diagnosis is to design local estimators or fault detectionfilters by intelligent agents according to the local sensing andcomputing resources, and a consensus strategy is utilized toensure the whole detection or estimation performance of allthe agents in the network.
Q9. What are the problems that make the network based fault diagnosis more difficult?
It is noted that the introduction of limited-capacitynetwork cables or wireless sensors into control and monitoringloops has unavoidably brought some unanticipated problemssuch as random communication delays, data dropout, andscheduling confusion, which make the network basedmonitoring and fault diagnosis more challenging comparedwith conventional point-to-point control and monitoringsystems.
Q10. How can a WVD method be used to diagnose rotors?
Very recently, via combining advanced notch FIR filters and the conventional WVD method, an improved WVD based fault diagnosis algorithm was proposed in [166], which can effectively minimize the cross terms and provide seamless high-resolution timefrequency diagrams enabling the diagnosis of rotor asymmetries and eccentricities in induction machines directly connected to the grid even in the worst cases.
Q11. What is the common method for determining signal frequency contents of local sections as the signal changes?
For instance, STFT method allows determining signal frequency contents of local sections as the signal changes in time, which has been widely applied to detect both stator and rotor faults in inductor motors [160].
Q12. What is the eigenstructure assignment approach for fault diagnosis?
eigenstructure assignment based fault diagnosisapproaches have been applied to vehicles [44], gas turbineengines [45], spacecraft [46] and wind turbine systems [47].
Q13. What is the eigenstructure of the observer-based fault detection filter?
Q14. What was the origin of the model-based fault diagnosis?
Model-based fault diagnosis was originated by Beard [39]in 1971 in order to replace hardware redundancy by analyticalredundancy, and comprehensive results were documented insome well-written books (i.e., see [40, 41]).
Q15. What is the main reason why the distributed structure is the mainstream in complex industrial processes?
This kindof decentralized or distributed structure has become the mainstream in complex industrial processes owing to its less use ofnetwork resources, cost effectiveness and convenience forexpansion.
Q16. What is the classification of fault diagnosis methods for discrete-event systems?
According to the model used, the fault diagnosis methods fordiscrete-event system can be roughly classified into automatabased fault method and Petri net based method.
Q17. What is the method for detecting gear faults?
Taking advantages of the periodicity of the geared faults, theCK algorithm can identify the position of the local gear faultin the gearbox.
Q18. What is the simplest way to solve the multi-object optimization problem?
the multi-object optimization problem describedby (4) can be reformulated by linear matrix inequality (LMI),which has been a popular method for fault diagnosis researchand applications owing to its wide applicability to a variety ofdynamic systems.
Q19. What is the classification of signal-based fault diagnosis methods?
Therefore signal-based fault diagnosis methods canbe thus classified into time-domain signal based approach,frequency-domain signal based approach and time-frequencysignal based method.
Q20. What was the first general fault detection and diagnosis procedure?
A generalfault detection and diagnosis procedure was first proposed in[92] by using residuals (or innovations) generated by Kalmanfilters with similar structure to observers, where the faultswere diagnosed by statistic testing on whiteness, mean andcovariance of the residuals.
Q21. What is the classification of the signal-based fault diagnosis?
signal-based fault diagnosis issurveyed following the classifications of time-domain,frequency-domain, and time-frequency-domain approaches,respectively.
Q22. What are some examples of a Kalman filter-based fault diagnosis?
Recent applicationexamples of Kalman filter-based fault diagnosis can be foundin [103-105] respectively for combustion engines, electronicsystems under mechanical shock, and permanent-magnetsynchronous motors.