An explainable fault diagnosis system and method for discrete event systems

Abstract

The application provides a kind of explainable fault diagnosis system and method of discrete event system, adopts the calculation mode of combining off-line construction fault diagnosis device and on-line fault diagnosis, while fusing field expert knowledge, according to the possible mode of system fault given by field expert, customizes limited linear time logic formula template, takes the historical operation log of system as sample, encodes fault diagnosis problem as satisfiability problem, obtains the explanation of each type of fault behavior of system: FLTL formula with the help of SAT solving technique.These FLTL formulas can form on-line fault diagnosis device on the one hand, and also provide the explanation and analysis of system fault, which can provide decision support for later system fault recovery and system improvement;Thus, the defect that existing DES fault diagnosis technology is difficult to provide explainable fault diagnosis is solved.

Description

A fault diagnosis system and method for interpretable discrete event systems Technical Field

[0001] This invention relates to the field of fault diagnosis for discrete event systems, and more specifically, to a fault diagnosis system and method for interpretable discrete event systems. Background Technology

[0002] Discrete Event Systems (DES) are dynamic systems whose states are discrete and whose evolution is driven by events occurring at irregular intervals. DES are a very common type of system, used to model flexible manufacturing systems, computer software systems, communication systems, intelligent transportation systems, and more. With the development of next-generation information technology, new DES systems are becoming increasingly large-scale and complex in structure; this inevitably leads to system errors during operation due to hardware failures, software failures, and design flaws. How to quickly and accurately diagnose various system faults and protect the functional safety of the system is a significant challenge. System fault diagnosis includes two basic tasks: 1. Providing an accurate fault type diagnosis for the system experiencing a fault; 2. Explaining and describing the system fault to provide decision support for further system fault recovery and system improvement.

[0003] Existing DES system fault diagnosis technologies are mainly divided into the following two categories: 1. Model-based Diagnosis (MBD); 2. Data-based Diagnosis (DBD).

[0004] The basic principle of MBD diagnostic technology is as follows: First, a complete model of the system is established, that is, a system model that fully describes the normal and abnormal behaviors of the system. Then, an MBD diagnostic tool is constructed by combining the system model and system observation mapping. The advantage of MBD diagnostic technology is that, as long as the system is diagnosable, the MBD diagnostic tool can ensure accurate diagnosis of system faults within a certain time delay. The main disadvantage of MBD diagnostic technology is that the construction of the MBD diagnostic tool depends on a complete model of the system; however, obtaining a complete model of the system in practical applications is very difficult. This, to some extent, limits the widespread application of MBD diagnostic technology in real-world systems.

[0005] To address this problem, DBD diagnostic technology was proposed. The basic principle of DBD diagnostic technology is to reduce the system fault diagnosis problem of DES (Device System Error Correction) to a classification problem, using system operation records as learning samples, and obtaining a DBD diagnostic tool through machine learning and other artificial intelligence methods. DBD diagnostic technology no longer relies on a complete model of the system; it only requires a sufficient amount of historical system operation records. Since system operation records are often easier to obtain than a complete system model in many practical applications, DBD diagnostic technology can be widely applied to real-world systems.

[0006] Existing DBD diagnostic techniques have largely fulfilled the first basic task of fault diagnosis: providing an accurate fault type diagnosis for a system experiencing a failure. However, a review of domestic and international literature reveals that most existing DBD diagnostic techniques can only provide a qualitative diagnosis of the system fault type, without offering explanations or descriptions of the system fault, thus hindering decision support for subsequent system fault recovery and system improvement. Furthermore, it is noteworthy that CJ Christopher et al., in their paper "Formulating Event-Based Critical Observations in Diagnostic Problems" presented at the IEEE CDC 2015 international conference, proposed a fault diagnosis technique based on critical observations (DES). The "critical observations" mentioned in the paper are essentially a special form of system fault explanation. Subsequently, Jiang Cheng et al., in their paper "Fault Diagnosis in Unknown Discrete Event Systems via Critical Tree" presented at the CDCC 2019 conference, employed a critical tree-based DES fault diagnosis technique, significantly improving the search speed for critical observations and the accuracy of fault diagnosis, thereby enhancing the work of CJ Christopher et al. Although CJ Christopher et al. and Jiang Cheng et al. conducted preliminary research and exploration on interpretable fault diagnosis of DES in the early stage, it is difficult to use key observations to characterize some complex fault modes because key observations are only a fault interpretation that considers the sequence of events.

[0007] In order to overcome the shortcomings of existing fault diagnosis technologies for DES and improve existing fault diagnosis technologies, this invention proposes an interpretable fault diagnosis system and method for DES. Summary of the Invention

[0008] Therefore, to overcome the shortcomings of existing DES fault diagnosis techniques, this method employs Finite Linear Temporal Logic (FLTL) formulas for fault interpretation. FLTL formulas support common fault modes and can more deeply characterize the essence of system faults, thus better fulfilling the two fundamental tasks of fault diagnosis: providing accurate system fault diagnosis and further offering interpretable system fault analysis, providing decision support for system fault recovery and system improvement. The specific technical solution is as follows:

[0009] A fault diagnosis system for an interpretable discrete event system includes an offline construction module and an online diagnosis module, wherein,

[0010] The offline construction module is configured to construct an interpretable fault diagnostic tool by integrating domain expert knowledge based on the system's existing historical operation records in the system's offline mode.

[0011] The online diagnostic module is configured to dynamically observe the real-time operation records of the system and match the observation sequence with the interpretation of system faults to obtain the fault diagnosis results of the system.

[0012] A fault diagnosis method for an interpretable discrete event system includes the following steps:

[0013] S10. Based on the existing system operation logs, construct a fault diagnostic tool in offline mode; S20. Based on the fault diagnostic tool constructed in step S10, perform online fault diagnosis on the system faults.

[0014] Furthermore, step S10 specifically includes the following steps:

[0015] S11. Preprocess the existing system historical operation logs, including log collection, cleaning, normalization, symbolization and annotation, and finally obtain a set of labeled system operation sequences.

[0016] S12. Domain experts provide possible failure modes for various system failure behaviors, such as action disorder, action deadlock, action mutual exclusion, action starvation, security violation, and liveness violation.

[0017] S13. For each possible mode of fault behavior obtained in step S12, customize its corresponding finite linear temporal logic formula template.

[0018] Furthermore, step S10 also includes the following steps:

[0019] S14. For each type of system fault, using the standard system operation sequence set obtained in step S11 as a learning sample, search for its fault explanation from the FLTL formula template in step S13. If the expert cannot provide the fault mode or cannot find a fault explanation that matches the learning sample from the formula template, then search for the fault explanation from all valid FLTL formulas.

[0020] Furthermore, step S20 also includes the following steps:

[0021] S21. Dynamically observe the operation of the system to obtain the real-time operation record of the system;

[0022] S22. Perform data cleaning and symbolization operations on the real-time system operation records in step S21 to obtain an unlabeled real-time system operation sequence.

[0023] S23. Check the satisfiability of the FLTL formula for each type of system fault in the real-time running sequence of the system in turn; collect the set of system fault types corresponding to the satisfied FLTL formulas, and label the system fault diagnosis as the set of fault types.

[0024] Technical effects:

[0025] 1. A computational mode combining offline fault diagnosis and online fault diagnosis is adopted. This mode completes most of the computation of the diagnostic system offline, thereby significantly reducing the computational load required for online fault diagnosis and ensuring the real-time requirements of fault diagnosis.

[0026] 2. Integrating domain expert knowledge, a customized Finite Linear Temporal Logic (FLTL) template is developed based on the possible failure modes provided by domain experts. Simulation experiments show that in scenarios with a small amount of system operation log data, incorporating expert knowledge can significantly improve the accuracy of fault diagnosis.

[0027] 3. Using the system's historical operation logs as samples, the fault diagnosis problem is encoded as a Satisfiability Problem (SAT). With the aid of SAT solving techniques, explanations for various fault behaviors of the system are obtained: FLTL formulas. These FLTL formulas can form an online fault diagnostic tool and also provide explanations and analyses of system faults, offering decision support for subsequent system fault recovery and system improvement; thus overcoming the deficiency of existing DES fault diagnosis technology in providing interpretable fault diagnoses. Attached Figure Description

[0028] The invention will be further understood from the following description taken in conjunction with the accompanying drawings. The components in the drawings are not necessarily drawn to scale, but rather the emphasis is on illustrating the principles of the embodiments. In different views, the same reference numerals designate corresponding parts.

[0029] Figure 1 is a schematic block diagram of the structure of a fault diagnosis system for an interpretable discrete event system according to an embodiment of the present invention.

[0030] Figure 2 is an explanation of the FLTL formula for the r-th type of fault in this embodiment of the invention. r The search algorithm.

[0031] Figure 3 is an example of the syntax tree for FLTL formulas in an embodiment of the present invention.

[0032] Figure 4 is a schematic diagram comparing the fault diagnosis accuracy of the embodiments of the present invention with that of the prior art fault diagnosis algorithm.

[0033] Figure 5 is a schematic diagram comparing the interpretability of the fault diagnosis algorithm of the present invention and the prior art. Detailed Implementation

[0034] To make the objectives, technical solutions, and advantages of this invention clearer, the invention will be further described in detail below with reference to its embodiments. It should be understood that the specific embodiments described herein are merely illustrative of the invention and do not limit the scope of protection of the invention.

[0035] It should be noted that when an element is referred to as being "fixed to" another element, it can be directly attached to the other element or there may be an intervening element. When an element is referred to as being "connected to" another element, it can be directly connected to the other element or there may be an intervening element. The terms "vertical," "horizontal," "left," "right," and similar expressions used herein are for illustrative purposes only and do not represent the only possible implementation.

[0036] Unless otherwise defined, all technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the art to which this invention pertains. The terminology used herein in the description of the invention is for the purpose of describing particular embodiments only and is not intended to be limiting of the invention. The term "and / or" as used herein includes any and all combinations of one or more of the associated listed items.

[0037] In this invention, "first" and "second" do not represent a specific quantity or order, but are merely used to distinguish names.

[0038] As shown in Figures 1, 2, 3, 4, and 5, an interpretable discrete event system fault diagnosis system according to an embodiment of the present invention includes an offline construction module and an online diagnosis module, wherein...

[0039] The offline construction module is configured to construct an interpretable fault diagnostic tool by integrating domain expert knowledge based on the system's existing historical operation records in the system's offline mode.

[0040] The online diagnostic module is configured to dynamically observe the real-time operation records of the system and match the observation sequence with the interpretation of system faults to obtain the fault diagnosis results of the system.

[0041] A fault diagnosis method for an interpretable discrete event system includes the following steps:

[0042] S10. Based on the existing system operation logs, construct a fault diagnostic tool in offline mode; S20. Based on the fault diagnostic tool constructed in step S10, perform online fault diagnosis on the system faults.

[0043] Preferably, step S10 specifically includes the following steps:

[0044] S11. Preprocess the existing system historical operation logs, including log collection, cleaning, normalization, symbolization and annotation, and finally obtain a set of labeled system operation sequences.

[0045] S12. Domain experts provide possible failure modes for various system failure behaviors, such as action disorder, action deadlock, action mutual exclusion, action starvation, security violation, and liveness violation.

[0046] S13. For each possible mode of fault behavior obtained in step S12, customize its corresponding finite linear temporal logic formula template.

[0047] Preferably, step S10 further includes the following steps:

[0048] S14. For each type of system fault, using the standard system operation sequence set obtained in step S11 as a learning sample, search for its fault explanation from the FLTL formula template in step S13. If the expert cannot provide the fault mode or cannot find a fault explanation that matches the learning sample from the formula template, then search for the fault explanation from all valid FLTL formulas.

[0049] Preferably, step S20 further includes the following steps:

[0050] S21. Dynamically observe the operation of the system to obtain the real-time operation record of the system;

[0051] S22. Perform data cleaning and symbolization operations on the real-time system operation records in step S21 to obtain an unlabeled real-time system operation sequence.

[0052] S23. Check the satisfiability of the FLTL formula for each type of system fault in the real-time running sequence of the system in turn; collect the set of system fault types corresponding to the satisfied FLTL formulas, and label the system fault diagnosis as the set of fault types.

[0053] The basic task of step S11 is to preprocess the system's historical operation logs, mainly including log collection, normalization, cleaning, symbolization, and annotation. For log collection, methods such as SNMPTrap and Syslog are selected based on the specific characteristics of the system to be diagnosed. If the system involves logs from multiple different components, the collected logs may exist in various formats; in this case, a log conversion configuration file needs to be first defined, and then the various logs are converted into a unified format for further processing. After completing log collection and normalization, appropriate data cleaning tools (such as DataWrangler, Google Refine, IBM Datastage, etc.) need to be selected based on the specific application scenario to clean the logs, including removing conflicting data, merging duplicate data items, filling missing values, and deleting invalid data. Finally, to facilitate subsequent formalization operations, the logs also need to be symbolized and annotated. An example of symbolization and annotation for wireless router system logs is provided below.

[0054] Example 1:

[0055] Logs from the wireless router system were collected using Syslog. After preprocessing, system logs in a unified format were obtained. The table below shows a log segment:

[0056] Table 1: Log fragments from the wireless routing system

[0057]

[0058]

[0059] The above log fragments involve 2 DHCP activity records, 3 IPv6 service activity records, 2 TCP activity records, and 1 cloud service record. The symbols d, i, t, and c represent the four service types: DHCP, IPv6, TCP, and cloud service, respectively. Subscripts are used to represent different actions under specific network service types. Finally, the log fragments are symbolized into the following sequence based on log timestamp order: {d1, i1}{d2, i2}{i3}{t1}{c1}{t1}. Furthermore, since the above logs do not involve faults, the sequence is labeled "n" to indicate that the current log records normal system behavior.

[0060] In step S11, after the log preprocessing described above, a standardized system operation record set will be obtained. For ease of explanation in subsequent steps, its formal definition is given below. Let AP represent the basic event set (also called the atomic proposition set) of the log, ∑=2 AP Let Δ represent the power set of AP (i.e., the set of all subsets), Δ = {f1, f2, ..., f...} K , n} represents the log tag set, where tag f r Let r represent the r-th type of fault, r∈{1,2,...,K}, and label n represent no fault occurring. A standard log entry can be represented as a two-tuple (l, t), where l∈∑ * Let be the sequence of events in the log, and t∈Δ be the log label. Therefore, the standardized system operation record set L obtained after system log preprocessing can be represented as a finite set of two tuples:

[0061] L={(l i ,ti)|l i ∈∑ * , t i ∈Δ,i∈{1,2,…,M}}. (1)

[0062] The basic task of step S12 is to integrate the knowledge of system domain experts regarding system faults into the automatic fault diagnosis system to improve the accuracy and interpretability of fault diagnosis. For a typical DES (Disorderly Execution System), common system fault modes include the following: 1. Actions Disorder: A system fault caused by a disruption in the triggering order of certain events; 2. Actions Mutual Exclusion: A fault caused by the combined triggering of certain mutually exclusive events; 3. Actions Starvation: A fault caused by the prolonged lack of response to certain request events; 4. Absence of Safety: A fault caused by the occurrence of certain "bad" events; 5. Absence of Liveness: A fault caused by the absence of "good" events.

[0063] For a specific type of failure in a particular system, the failure mode may fall into one or more of the five common modes mentioned above, or it may be another failure mode. Generally speaking, system domain experts have a certain understanding of the operating rules of the system and have some prior knowledge of system failures and their handling. Therefore, they will refer to the failure modes provided by experts to generate explanations of system failures and fault diagnostic tools.

[0064] The basic task of step S13 is to provide a formal characterization of the failure modes provided by the domain expert in step S12. The main technical solution adopted is to characterize the system failure modes using Linear Temporal Logic (LTL) formulas. Considering that the standard log obtained in step S11 is a finite sequence, this invention will consider the finite semantics of the LTL formula, i.e., FLTL. To help readers understand, the syntax and semantics of the FLTL formula are briefly described below.

[0065] Let AP be a finite set of atomic propositions. Then the FLTL formula is recursively defined as: 1) for any atomic proposition p∈AP, p is an FLTL formula; 2) for any FLTL formula... and ψ, It is also an FLTL formula, in which ∧ represents logical NOT; ∧ represents logical AND; operations Indicates the next point in time. Established; Operation Formula This holds true until the formula ψ holds true. To improve the simplicity of FLTL formulas, some operators are added as syntactic sugar operators: logical truth... Logical false Logical OR operation Logical implication operation Logical equivalence operations Final calculation Global operations Given a log sequence of length N, l = σ1σ2…σ N ∈∑ N , let l i σ represents the i-th character of sequence l. i l[i…] represents a suffix starting with the i-th character: σ i σ i+1 …σ N The semantics of the basic FLTL operators are shown in Table 2. The semantics of other syntactic sugar operators can be derived from their relationship with the basic operators.

[0066] Table 2: Semantics of FLTL Formulas

[0067]

[0068] Example 2: This example is a further description of the above examples. It should be understood that this example includes all the foregoing technical features and is further described in detail:

[0069] As an example, Table 3 provides the FLTL formula templates corresponding to the common system failure modes mentioned above (taking two related actions as an example).

[0070] Table 3: FLTL Formula Templates for Common System Faults

[0071]

[0072] In fact, apart from the fault modes described in the embodiments, any first-order fault mode can be characterized by LTL formulas, because it has been theoretically proven that LTL language has the same descriptive power as first-order logic.

[0073] The basic task of step S14 is to use the system standard log set L obtained in step S11 as a learning sample to learn the interpretation of various types of system faults, that is, to generate FLTL formulas to characterize the behavior of various faults. More specifically, for the r-th type of fault (r∈{1, 2, ..., K}), the standard logs are divided into the following two subsets:

[0074] L1={(l1, t1)|(l1, t1)∈L, t1=f r} (2)

[0075] L2=L\L1={(l2, t2)|(l2, t2)∈L, t2≠f r} (3)

[0076] The basic task of step S14 is to find a fault explanation for any fault type r: FLTL formula Υ r , making

[0077] For any (l1, t1) ∈ L1(4)

[0078] l2≠Υ r For any (l2, t2) ∈ L2(5)

[0079] Y satisfies formula (4-5) rThere may be many formulas; however, it is generally believed that the shorter the formula, the easier it is for humans to understand. Therefore, based on the consideration of interpretability, this invention will seek the shortest Y that satisfies formula (4-5). r .

[0080] It should be noted that although steps S12 and S13 aim to incorporate domain experts' prior knowledge of system faults into the automatic diagnostic system to improve the performance and interpretability of fault diagnosis, in some complex systems, domain experts may not be able to provide fault modes. In such cases, it is necessary to search among all valid FLTL formulas for Y that satisfies formula (4-5). r On the other hand, even if domain experts provide failure modes and customize corresponding FLTL formula templates based on these failure modes, it is still possible that Y satisfying formula (4-5) cannot be obtained in the FLTL template formula. r At this point, it is also necessary to search for Υ in all FLTL formulas. r .

[0081] Since the customized FLTL formula template in step S13 is essentially an FLTL formula, this invention incorporates the Y formula template. r The search, as Υ in all FLTL formulas r This involves solving a special case of the search for Υ. The following details the search for Υ among all valid FLTL formulas. r The technical solution.

[0082] The search Y used in this invention r The main technical solution is as follows: First, starting from n=1, the syntax tree of the FLTL formula of length n is encoded into the predicate logic formula Ω. n Then, the SAT solver is used to solve the predicate logic formula Ω. n If Ω n If it can be satisfied, then according to the predicate logic formula Ω n The model decoding yields the FLTL formula Υ r Otherwise, increment n by 1 and continue the search. Theoretically, it can be proven that the search algorithm is terminated, meaning there exists a finite n such that Ω... n This is acceptable. r The flowchart of the search algorithm is shown in Figure 2.

[0083] First, we introduce the generation of the syntax tree for FLTL formulas. The syntax tree of any FLTL formula can be constructed recursively as follows: the root is the operator of the last operation in the FLTL formula. If the operator is a binary operator, its first and second operands (sub-formulas) will similarly generate subtrees and be attached to the root node as its left and right subtrees, respectively. If the operator is a unary operator, its operands (sub-formulas) will similarly generate a tree and be attached to the root node as its subtree. If the operator is an atomic proposition, leaf nodes will be generated. Finally, each node in the syntax tree is labeled with a natural number 1, 2, ..., n, starting from the root and proceeding downwards. Therefore, the encoding of a child node in the syntax tree is always greater than the encoding of its parent node.

[0084] Example 3: This example is a further description of the above examples. It should be understood that this example includes all the foregoing technical features and is further described in detail:

[0085] Figure 3 shows the syntax tree of the FLTL formula template described in Table 3.

[0086] Following the syntax tree generation method described above, for any FLTL formula, there exists a corresponding syntax tree. The following section first introduces how to encode the structure and semantics of an n-node FLTL syntax tree into a predicate logic formula Ω. n .

[0087] To facilitate the formalization of methods, the nodes in the syntax tree are divided into the following three categories: unary operator nodes. Binary operators and atomic proposition nodes make Set Boolean variable i∈{1, 2, ..., n}, The statement is true if and only if the operator corresponding to node i in the syntax tree is j; Boolean variable l i，k ,i∈{1,2,...n-1},k∈{2,...n},l i，k The statement is true if and only if the left child of node i is node k; Boolean variable. i∈{1, 2,…n-1}, k∈{2,…n}, The statement is true if and only if the right child of node i is node k.

[0088] Clearly, when node i is a binary operator, it has only one left child and one right child, i.e.

[0089]

[0090]

[0091] When node i is a unary operator, it has only one child node (here, l). i，k Encoding), that is

[0092]

[0093] When node i is an atomic proposition, it has no child nodes, i.e.

[0094]

[0095] Furthermore, since each node can only contain one type of operator, there is only one possible operator for each node i. Make If true, that is

[0096]

[0097] Let Ω str If we consider the conjunction (logical AND) of formulas (6), (7), (8), (9), and (10), then Ω str The structural constraints of a valid FLTL syntax tree with n nodes are fully encoded, meaning that any valid FLTL formula of length n corresponds to formula Ω. str A model.

[0098] Next, consider the semantic constraints of the FLTL formula. For any log (l, t) ∈ L and its arbitrary starting position m ∈ {1, 2, ..., |l|}, and any node i in the syntax tree of the FLTL formula, set a Boolean variable... Represents the sub-formula of sequence l[m…] denoted by node i. The satisfaction, that is

[0099] If and only if

[0100] Therefore, based on the semantics of the FLTL basic operators: atomic propositions, logical NOT, logical AND, X operator, and U operator, the following formulas are derived to characterize the value constraints of the basic operator nodes on the syntax tree with respect to the log sequence l:

[0101]

[0102]

[0103]

[0104]

[0105]

[0106] Similarly, the value constraint formulas for the FLTL syntactic sugar operator nodes on the syntax tree can also be derived from their relationship with the basic operators. Performing a conjunction (logical AND) operation on the above formulas yields the value constraint formulas for the FLTL formulas related to log l. Finally, the syntax tree of the FLTL formula of length n is encoded into the following predicate logic formula Ω. n :

[0107]

[0108] Based on the above encoding rules, it is easy to prove that there exists an FLTL formula Υ of length n. p Formula (4-5) is satisfied if and only if the predicate logic formula Ω n This is acceptable. Therefore, when Ω n When this condition is met, Ω can be obtained using the SAT solver. n The model (even if Ω) n Boolean variables that are true A set of assignments). Based on Boolean variables. The value of can be used to construct the syntax tree of the corresponding FLTL formula, thus obtaining the FLTL formula Υ. r That is, the explanation of the r-th type of fault (r∈{1,2,...,K}).

[0109] It is important to emphasize that when an expert-defined failure mode exists, a corresponding FLTL formula template can be customized. In this case, the length and structure of the FLTL formula are fixed, thus a significant portion of the Boolean variables... The value of Ω is also known. By importing the known Boolean variable value into formula (17), we can obtain the predicate logic formula corresponding to the FLTL formula template: Ω tmpl Therefore, the explanation for the r-th fault is searched from the FLTL formula template. r The above search for Υ from all FLTL formulas is the above. r A special case.

[0110] Example 4: This example is a further description of the above examples. It should be understood that this example includes all the foregoing technical features and is further described in detail:

[0111] Based on the above encoding rules, the predicate logic formula Ω corresponding to the template formula can be further obtained. tmpl The following uses a template formula. Let's take an example to illustrate. Based on the syntax tree structure of the formula tmpl (see Figure 3-e), it's clear that Boolean variables... The value is true, Boolean expression If this is also true, then the formula length n is 4. Therefore, we obtain the predicate logic formula Ω. tmpl :

[0112]

[0113] Regarding the SAT solver, this invention uses the open-source SAT solver glucose based on the DPLL algorithm. This solver incorporates numerous improvements and optimization strategies during the solution process and can theoretically handle the solution of predicate logic formulas containing billions of variables.

[0114] For each fault type r (r∈{1, 2, ..., K}), execute Υ once. r The search algorithm (its flowchart is shown in Figure 2) finally yields the FLTL formula set {Υ r |r∈{1,2,...,K}}. On one hand, this set of formulas can serve as an explanation of system faults: that is, the FLTL formula Υ r This is the r-th type of fault characteristic. The formula set {Υ} r |r∈{1, 2, ..., K}} is provided to domain experts to help them identify the underlying causes of system failures and further improve the system design. On the other hand, the formula set also constitutes a system fault diagnostic tool. The following briefly describes how to utilize the FLTL formula set {Υ r |r∈{1,2,...,K}} enables online fault diagnosis.

[0115] First, in step S21, the system's operation is dynamically observed to obtain a real-time operation record. Then, in step S22, the real-time operation record from step S21 is cleaned and symbolized; this process is similar to the corresponding operation in step S11 and will not be described further here. After these two steps, an unlabeled real-time system operation sequence is obtained: l = σ1σ2…σ |l| .

[0116] In step S23, the FLTL formula set Y for each type of system fault is checked sequentially for the real-time operating sequence l of the system. r Satisfactionability; collect the set of system fault types corresponding to the satisfied FLTL formulas, and label the system fault diagnosis as this fault type set. Specifically, the real-time operation record of the system is l = σ1σ2…σ|l| The fault diagnosis is as follows:

[0117]

[0118] This concludes the detailed introduction and explanation of the technical solution adopted in this invention. To verify the performance of the DES interpretable fault diagnosis system and method proposed in this invention, the following comparative experiments were conducted.

[0119] Example 5: This example is a further description of the above examples. It should be understood that this example includes all the foregoing technical features and is further described in detail:

[0120] First, a series of DES system models were randomly generated. Then, for each system model, a set of system execution sequences of a specified length was randomly generated. Next, the execution sequence sets of each system were randomly divided into two categories: a learning sample set (containing 80% of the sequences) and a test sample set (containing the remaining 20% ​​of the sequences). Fault diagnosis experiments were conducted on the above random datasets, testing four interpretable DES fault diagnosis algorithms: the Key Observation Algorithm (CO) proposed by CJ Christopher et al. at CDC2015, the Key Tree Algorithm (CT) proposed by Jiang et al. at CCDC 2019, the inventive algorithm without expert knowledge (FLTL_NK), and the inventive algorithm with expert knowledge (FLTL_K). Two metrics of the algorithms were compared:

[0121] 1) The average accuracy of system fault diagnosis, with the following parameters: number of system fault types K=5, number of log basic events |AP|=20, log sequence length N=200, and size of log sequence set M∈{100, 150, 200, 250, 300, 350, 400}. The four algorithms mentioned above were compared, and the experimental results are shown in Figure 4.

[0122] 2) Interpretability of fault diagnosis, i.e., the average length of the FLTL formula or the average length of key observations obtained by the algorithm, with the same parameters as above; the first three algorithms were compared (the FLTL_K algorithm was not included in this experiment because a FLTL template of a specified length was customized based on expert knowledge), and the comparison results are shown in Figure 5.

[0123] As shown in Figure 4, the proposed algorithms (FLTL_K, FLTL_NK) achieve higher fault diagnosis accuracy in all scenarios than existing interpretable fault diagnosis algorithms: CO and CT. It is noteworthy that expert knowledge contributes to the accuracy of fault diagnosis, especially when the amount of data is small, where the advantage of expert knowledge is particularly evident. This indicates that expert opinions are especially important in the early stages of system deployment when historical operational data is limited.

[0124] As shown in Figure 5, the fault interpretation (i.e., FLTL formula) generated by the FLTL_NK algorithm proposed in this invention is shorter than the fault interpretation (i.e., key observation) generated by the CO and CT algorithms in all cases. For human cognition, shorter things are often easier to understand; therefore, from this perspective, the interpretability of the fault diagnosis algorithm proposed in this invention is better than that of the CO and CT algorithms.

[0125] This invention proposes an interpretable fault diagnosis system for DES (Device Execution System), employing a computational mode that combines offline fault diagnostic tool construction with online fault diagnosis. This computational mode completes most of the diagnostic system's computation offline, thereby significantly reducing the computational load required for online fault diagnosis and ensuring the real-time requirements of fault diagnosis.

[0126] This invention proposes an interpretable fault diagnosis method for DES (Device Execution System), which integrates prior knowledge of system faults from domain experts into the automatic diagnostic system. This technical solution helps improve the performance of the automatic diagnostic system; especially in scenarios with small amounts of system log data, incorporating expert knowledge can improve the accuracy of system fault diagnosis by 42% to 58%.

[0127] This invention proposes an interpretable fault diagnosis method for DES (Distributed Electronic System). This method employs Finite Linear Temporal Logic (FLTL) formulas as the interpretation of system faults. The powerful expressive capabilities of the LTL language allow it to characterize various common fault modes in a more concise manner. Simulation results show that, under the same conditions, the average length of the fault interpretation (FLTL formula) generated by this invention is only 51.8% of the average length of the fault interpretation (key observation) generated by existing technologies. Therefore, the fault diagnosis technology proposed in this invention significantly improves the interpretability of fault diagnosis.

[0128] The technical features of the above embodiments can be combined in any way. For the sake of brevity, not all possible combinations of the technical features in the above embodiments are described. However, as long as there is no contradiction in the combination of these technical features, they should be considered to be within the scope of this specification.

[0129] The embodiments described above are merely illustrative of several implementations of the present invention, and while the descriptions are relatively specific and detailed, they should not be construed as limiting the scope of the invention patent. It should be noted that those skilled in the art can make various modifications and improvements without departing from the concept of the present invention, and these all fall within the protection scope of the present invention. Therefore, the protection scope of this invention patent should be determined by the appended claims.

Claims

1. A fault diagnosis method for an interpretable discrete event system, characterized in that, A fault diagnosis system for interpretable discrete event systems includes an offline construction module and an online diagnosis module. The offline construction module is configured to construct an interpretable fault diagnostic tool by integrating domain expert knowledge and existing historical operation records of the system in offline mode. The online diagnosis module is configured to dynamically observe the real-time operation records of the system and match the observation sequence with the interpretation of system faults to obtain the system's fault diagnosis results. The fault diagnosis system implements the following steps: S10, constructing a fault diagnostic tool in offline mode based on existing system operation logs. S11. Preprocess the existing system historical operation logs, including log collection, cleaning, normalization, symbolization, and annotation; S12. Domain experts provide possible failure modes for various system fault behaviors, including action disorder, action deadlock, action mutual exclusion, action starvation, safety violation, and liveness violation; S13. For each possible failure behavior mode obtained in step S12, customize its corresponding finite linear temporal logic formula template; S14. For each type of system fault, using the standard system operation sequence set obtained in step S11 as a learning sample, search for its fault explanation from the FLTL formula template in step S13; For the... For this type of failure, the standard log is divided into the following two subsets: ； The basic task of step S14 is to find a fault explanation for any fault type r: the FLTL formula. , making ； S20. Based on the fault diagnostic tool constructed in step S10, perform online fault diagnosis on the system faults; S21. Dynamically observe the system operation to obtain the real-time operation record of the system; S22. Perform data cleaning and symbolization operations on the real-time operation record of the system in step S21 to obtain an unlabeled real-time system operation sequence; S23. Check the satisfiability of the FLTL formula for each type of system fault in the real-time system operation sequence in turn; collect the system fault type set corresponding to the satisfied FLTL formula, and label the system fault diagnosis as the fault type set; First, in step S21, dynamically observe the system operation to obtain the real-time operation record of the system; then, in step S22, perform data cleaning and symbolization on the real-time operation record of the system in step S21. After the above two steps, an unlabeled real-time system operation sequence will be obtained: In step S23, the FLTL formula set for each type of system fault is checked sequentially for the real-time operation sequence l of the system. Satisfactionability; collecting the set of system fault types corresponding to satisfied FLTL formulas, and labeling system fault diagnoses as this fault type set; specifically, real-time system operation records. The fault diagnosis is as follows: 。

Images