State-based fault diagnosis of discrete-event systems with partially observable outputs☆
Introduction
In the last three decades, modeling, control, and diagnosis of discrete-event systems (DES) [1], [2], [3], [4], [5], [6], [50], [51], [52] have been extensively studied by researchers and engineers from different domains. Typical applications include but are not restricted to aerospace systems, transportation systems manufacturing systems [43], [44], [45], [46], and real-time scheduling and reconfigurations [55], [56], [57], [58]. Informally, a DES is discrete in time and in state space, and event-driven rather than time-driven.
Fault diagnosis plays an vital role in maintaining the performance and enhancing the reliability of DES, especially the safety-critical systems. The problem of fault diagnosis has attracted considerable attention from industry and academia, and it has been widely investigated in the literature [6], [7], [8], [9], [10], [11], [12], [13], [14], [15], [16], [17], [18], [19], [20], [21], [22], [23], [24], [25], [26], [27], [28], [29], [30], [31], [32], [33], [34], [35], [36], [37], [38], [39], [40], [41], [42]. Initially, the problem of fault diagnosis is studied in [7] and [9] where the concept of diagnosability is introduced and formalized in the DES setting. Later on, problems of modular [10], decentralized [11], [12], [14], state-based [13], distributed [15], hierarchical [16], [17], and robust approaches [18] to diagnosis have also been discussed. Moreover, the method to diagnosis is investigated for timed DES in [19] and [20]. In [21], the diagnosis technique that utilizes model checking is introduced. In [22], [23], [24], diagnosability is extended to stochastic, fuzzy, and bi-fuzzy DES settings, respectively. In [25], a broader spectrum of diagnoser-based approaches regarding the degree of reasoning performed offline is proposed. In [26], the diagnoser is constructed in the form of the symbolic observation graph (SOG), which combines symbolic and enumerative representations as a basis of efficient verification. In [27], the abstraction-based verification of diagnosability is presented for the purpose of reducing the computational cost in the case of complex DES. Most of the aforementioned diagnosis methods follow the event-based framework [7], i.e., the diagnosis is deployed based on the observation of event sequences. The system faults are characterized by pre-specified faulty events and thus the event set is partitioned into faulty and non-faulty parts. To get a general understanding on the literature, the reader can refer to the survey paper [28], where the state of the art of techniques and tools relevant to fault diagnosis of DES is well reviewed.
In the context of DES, automata and Petri nets (PN) are the most used modeling formalisms. The pioneer works [7], [9] on fault diagnosis are based on automata. In [7], a model called “diagnoser” is introduced to verify diagnosability by examining the existence of indeterminate cycles, and diagnosis is deployed online by a mapping relation between the online observations and the states of the diagnoser. However, the construction of a diagnoser suffers from the state explosion problem. To tackle this issue, an efficient algorithm for verification of diagnosability [29], [30] is proposed and the computational complexity is polynomial with respect to the number of system states and linear with respect to the number of failure types. Besides, a series of work [31], [32], [33], [34], [35], [36], [37], [38], [39] concerning diagnosis and diagnosability of DES modeled by PN ensues owing to graphical and mathematical representations of PN. Various diagnosis methods of DES using PN, such as P-invariants [31], the basic reachability graph [35], and the integer linear programming [36], [37], [38], are proposed. In [32], the issues of diagnosability and online diagnosis of DES using interpreted PN are addressed. In [33], online diagnosis of DES modeled by partially observed PN is investigated based on the capture and analysis of observation sequences. In [34], a general setting that markings and transitions are partially observable is considered and the diagnosis problem is studied by transforming partially observed PN into equivalent labeled PN. For more details, the reader can refer to the survey papers [4], [40] and the references therein.
In [13], a state-based method for fault diagnosis of a DES with fully observable outputs is introduced. The system to be diagnosed is modeled as a nondeterministic finite-state Moore automaton. It is assumed that the state set of the system can be partitioned according to the condition. The diagnosis problem is to decide that the state belongs to the normal or faulty partition when the last measurement is received via sensor readings. As the extension and reinforcement of [13], this work investigates the problems of diagnosis and diagnosability of a DES with partially observable outputs by considering more practical cases (partial observation) and the algorithm optimization (efficiency). The diagnoser in this research consists of two parts: a state estimator and a failure decision-maker for better expansibility and flexibility. In practice, a state estimator often needs to be renovated according to accessible information (e.g., output signals), while a failure decision-maker is fixed. The diagnosis is implemented online, i.e., the system is operational and the diagnosis decision is updated after a new output signal is observed. As the basis of online diagnosis, diagnosability analysis is performed. The notion of diagnosability is given and a polynomial-time algorithm is designed for verifying diagnosability without constructing a diagnoser. To demonstrate the proposed algorithm, a pump-valve-controller system [13] is provided.
The reminder of the paper is arranged as follows. Section 2 provides necessary concepts, terminologies, and particularly recalls the definition of the output projection function. In Section 3, an online diagnosis framework for detecting the occurrences of faults and localizing the cause of faults is introduced. Different from [13], the diagnoser in this research consists of two parts: a state estimator and a failure decision-maker. Section 4 gives the notion of diagnosability of a DES with partially observed outputs. In Section 5, an efficient algorithm for verification of diagnosability is designed and the computational complexity is polynomial with respect to the number of system states. Furthermore, a model reduction technique for detecting cycles that violate diagnosability condition efficiently is discussed. In Section 6, the developed algorithm is applied to a pump-valve-controller system. Further discussions on the comparison between state-based and event-based diagnosis approaches and the advantages of the proposed approach over the existing work are presented in Section 7. Finally, we draw the conclusion in Section 8. For the better readability, the related proofs have been moved to Appendix.
Section snippets
System model
In [13], a discrete-event system (DES) to be diagnosed is modeled as a nondeterministic finite-state Moore automaton (FSMA)where
- •
Q is the finite state set;
- •
Σ is the finite set of events;
- •
δ: Q × Σ → 2Q (2Q denotes the power set of Q) is the partial transition function;
- •
q0 ∈ Q is the initial state;
- •
Λ is the finite output set;
- •
λ: Q → Λ is the output map, which assigns each state in Q with an output.
Diagnosis framework
In this section, the framework for diagnosis of discrete-event systems (DES) with partially observable outputs is presented. The following hypotheses are necessary for the system under investigation:
- 1.
The state set of the system can be partitioned based on the condition (failure or normal status) of the system;
- 2.
The system is live, i.e., (∀q ∈ Q)(∃σ ∈ Σ) δ(q, σ)!;
- 3.
The faulty states are permanent, i.e., the faulty states only transit to those corresponding to the same failure mode.
The proposed
Diagnosability
In this section, the notion of diagnosability of a DES with partially observable outputs is given, which is the extension of that in [13]. Different from [7], diagnosabillity in [13] and this work is defined and analyzed with respect to the initialization of diagnosis at any moment. Before introducing diagnosability, the following definitions are needed. Definition 6 Let x ∈ 2Q. Then x is said to be Fi-certain if . Definition 7 Let x ∈ 2Q. Then x is said to be Fi-uncertain if and i.e., x indicates[Fi-certain]
[Fi-uncertain]
Verification of diagnosability
This section presents a novel algorithm for verifying diagnosability of a DES with partially observable outputs, that is an extension of the one proposed in [13]. Given a random initialization, it is obvious that the first observed output l0 must belong to Λo. Thus we can verify whether the system is diagnosable or not for each first observed output l0. If the answer is YES, then the system is diagnosable. Otherwise, the system is not diagnosable. In comparison, a diagnoser in [13] is used for
An example: pump-valve-controller system
In [13], a pump-valve-controller system consists of a pump, a valve, a DES controller, together with a flowmeter (a sensor), as depicted in Fig. 8. There are two failure modes: stuck-closed (F1) and stuck-open (F2) for the valve. The DES controller opens the valve (VE), then turns on the pump (PE). After a while it shuts down the pump (PD) and closes the valve (VD). This process repeats. The DES models of the three individual components in the system can be found in [13]. The output of the
Discussion
In this section, we have further discussions on
- 1.
the comparisons between state-based and event-based approaches;
- 2.
the advantages of the proposed approach over the existing work.
Conclusions and future work
In this work, we present a state-based method for online passive diagnosis of discrete-event systems (DES) with partially observable outputs. The diagnoser is made up of two parts: a state estimator and a failure decision-maker to perform specific tasks. Moreover, no information about the state or condition (failure status) of the system is required when starting diagnosis; thus there is no need to initialize the system and diagnosis simultaneously. The issue of diagnosability is also studied
Declaration of Competing Interest
The author(s) declare(s) that there is no conflict of interest (such as personal or professional relationships, affiliations, knowledge or beliefs) in the subject matter or materials discussed in the manuscript entitled “State-Based Fault Diagnosis of Discrete-Event Systems with Partially Observable Outputs”, which we wish to be considered for publication in Information Sciences.
CRediT authorship contribution statement
Deguang Wang: Conceptualization, Methodology, Software, Writing - original draft, Formal analysis, Visualization. Xi Wang: Investigation, Writing - review & editing, Supervision, Resources, Visualization. Zhiwu Li: Supervision, Funding acquisition, Writing - review & editing, Visualization.
References (58)
- et al.
On the history of diagnosability and opacity in discrete event systems
Annu. Rev. Control
(2018) - et al.
Petri nets and automatic control: a historical perspective
Annu. Rev. Control
(2018) - et al.
Decentralized fault prognosis of DES with guaranteed performance bound
Automatica
(2016) - et al.
A formal framework for the decentralized diagnosis of large scale discrete event systems and its application to telecommunication network
Artif. Intell.
(2005) - et al.
Robust diagnosis of discrete event systems against intermittent loss of observations
Automatica
(2012) - et al.
Fault diagnosis for discrete event systems: modelling and verification
Reliab. Eng. Syst. Saf.
(2010) Abstraction-based failure diagnosis for discrete event systems
Syst. Control Lett.
(2010)- et al.
Overview of fault diagnosis methods for discrete event systems
Annu. Rev. Control
(2013) New technique for fault detection using petri nets
Automatica
(1991)- et al.
Fault detection for discrete event systems using petri nets with unobservable transitions
Automatica
(2010)
Fault detection of DES by petri nets and integer linear programming
Automatica
Diagnosability of labeled petri nets via integer linear programming
Automatica
A framework for state attraction of discrete event systems under partial observation
Inf. Sci.
Petri net-based approach to short-term scheduling of crude oil operations with less tank requirement
Inf. Sci.
Deadlock recovery for flexible manufacturing systems modeled with petri nets
Inf. Sci.
On observability of discrete-event systems
Inf. Sci.
Supervisory control of state-tree structures with partial observation
Inf. Sci.
Priority-free conditionally-preemptive scheduling of modular sporadic real-time systems
Automatica
Supervisory control of a class of discrete-event processes
SIAM J. Control Optim.
Supervisory Control of Discrete-Event Systems, Monograph Series Communications and Control Engineering
Introduction to Discrete Event Systems
Control of Discrete-Event Systems: Automata and Petri Net Perspectives
Diagnasability of discrete-event systems
IEEE Trans. Autom. Control
Failure diagnosis using discrete event models
IEEE Trans. Control Syst. Technol.
Diagnosability of discrete event systems and its applications
Discrete Event Dyn. Syst.
A modular architecture for diagnosis of discrete event systems
Proc. 41st IEEE Conf. Dec. Cont.
Fault diagnosis in discrete-event systems: framework and model reduction
IEEE Trans. Autom. Control
Diagnosis of discrete event systems using decentralized architectures
Discrete Event Dyn. Syst.
Global and local consistencies in distributed fault diagnosis for discrete-event systems
IEEE Trans. Autom. Control
Cited by (18)
Non-interference assessment in colored net systems via integer linear programming
2024, Information SciencesAn approach for enforcing a class of GMECs on time Petri nets with uncontrollable transitions
2021, Information SciencesCitation Excerpt :Petri nets (PNs) are an essential model for discrete event systems (DESs) by providing both intuitive and explicit representations of system behavior [1]. To enforce control specifications on a DES, it is necessary to devise a supervisor that confines the state space of the system to solve a state-based control problem [2–5]. Generally, control specifications of a PN system are defined as a set of legal markings implemented by applying generalized mutual exclusion constraints (GMECs) [6].
Networked opacity for finite state machine with bounded communication delays
2021, Information SciencesCitation Excerpt :The interesting future directions include the following two aspects. In many applications, there are some faults in the system that need to be detected, and the fault diagnosis with state observation of networked systems is an interesting research topic [22]. Meanwhile, what we need to solve in reality is how to ensure the properties of networked opacity once the original system is not opaque.
Fault Diagnosis of Hierarchical Discrete-Event Systems Based on State-Tree Structures
2024, IEEE Systems JournalAsynchronous Fault Diagnosis of Stochastic Discrete-Event Systems in Industrial Applications
2024, IEEE Sensors Journal
- ☆
This work was supported in part by the National Natural Science Foundation of China under Grant Nos. 61703322, 61873342, 61673309, and 61603285, the Alexander von Humboldt Foundation, the National Key R&D Program of China under Grant 2018YFB1700104, and the Doctoral Students' Short-Term Study Abroad Scholarship Fund of Xidian University.