A Fault Diagnosis Reasoning Method and Device Based on a Time Fault Propagation Graph

Abstract

The present application provides a fault diagnosis and reasoning method and device based on a time fault propagation graph, belonging to the field of avionics system health management. The method includes: generating an assumption after obtaining a system abnormal event; updating the assumption: when diagnosing an intermediate event, different update operations are performed on the assumption according to different situations where the new event and a certain assumption meet the preset conditions, where the preset conditions are that the fault mode of the new event and a certain assumption is related based on the node connection relationship and the time is consistent, the intermediate event is an event other than the first event, and the connection relationship includes an AND connection relationship and an OR connection relationship; adding an assumption; evaluating the effect of each assumption based on an evaluation index to obtain a fault diagnosis and reasoning result, which can improve the accuracy and rationality of fault diagnosis work for complex systems with multiple faults occurring.

Description

Technical Field

[0001] This application belongs to the field of avionics system health management, and particularly relates to a fault diagnosis and reasoning method and device based on a timed failure propagation graph. Background Art

[0002] With the development of technology, the complexity of systems has been continuously increasing, and at the same time, the occurrence of faults in each subsystem or component has become more and more complex. Fault diagnosis technology can timely detect and report problems existing in the system to the operation and maintenance personnel, facilitating the maintenance personnel to formulate effective maintenance work plans and improving the stability and reliability of the system.

[0003] For avionics systems, the fault diagnosis method based on a directed graph requires no large amount of monitoring data, and can use information such as the extracted system connection relationship and causal propagation path to implement fault modeling, providing an effective way for fault diagnosis. Among them, the timed failure propagation graph (TFPG) can not only express the complex causal relationship of system fault propagation, but also consider the fault propagation time constraint of the dynamic system, making the abstract fault model better reflect the fault propagation situation of the actual system.

[0004] For simple and small systems, usually the timed failure propagation graph is also relatively simple, and there is often only one directed edge connection between nodes, without considering the comprehensive influence of multiple different nodes. However, for the timed failure propagation model of complex products such as avionics systems, multiple faults may occur. In such cases, it is necessary to check the nodes in a differentiated manner and design a comprehensive reasoning algorithm. To meet this actual need, a diagnostic reasoning algorithm based on a timed failure propagation graph is proposed to improve the accuracy and rationality of fault diagnosis work. Summary of the Invention

[0005] To solve the problems in the related art, this application provides a fault diagnosis and reasoning method and device based on a timed failure propagation graph, and the technical solutions are as follows:

[0006] In a first aspect, a fault diagnosis and reasoning method based on a timed failure propagation graph is provided, and the method includes:

[0007] Generating a hypothesis: When the first event arrives, a new hypothesis is established;

[0008] Update Hypothesis: When diagnosing intermediate events, different update operations are performed on a hypothesis according to different situations where a new event and a hypothesis meet preset conditions. Herein, the preset conditions are that the failure mode of the new event and a hypothesis is related based on the node connection relationship and the time is consistent. Herein, the intermediate event is an event other than the first event, and the connection relationship includes an AND connection relationship and an OR connection relationship;

[0009] Add Hypothesis: If the new event does not meet the preset conditions with all existing hypotheses in the hypothesis library, a new hypothesis is generated;

[0010] Evaluate the effect of each hypothesis based on evaluation indicators to obtain a fault diagnosis reasoning result.

[0011] Wherein, the generation of the hypothesis includes:

[0012] Relevance Analysis: Calculate the cut set set of the differential nodes according to the differential nodes and connection relationship in the first event. The cut set set includes multiple cut sets, and each cut set is a combination of failure modes that may cause the activation of the differential nodes. The connection relationship is extracted from the fault diagnosis model;

[0013] Time Consistency Analysis: Determine whether the activation time of the differential nodes is consistent with the propagation time of the failure modes included in each cut set to the differential nodes. When the activation time of the differential nodes is consistent with the propagation time of the failure modes included in each cut set to the differential nodes, generate a hypothesis based on the pre-constructed basic hypothesis structure, determine the attributes of the differential nodes, the failure modes that occur in the hypothesis, and multiple alarm sets. The propagation time of the failure modes to the differential nodes is extracted from the fault diagnosis model.

[0014] Wherein, the update of the hypothesis includes:

[0015] Relevance Analysis: Calculate the cut set set of the differential nodes according to the differential nodes and connection relationship in the intermediate event, and determine whether at least one cut set is a subset of the diagnosis set. When at least one cut set is a subset of the diagnosis set, perform time consistency analysis. The diagnosis set is the set of failure modes that occur in the hypothesis in the basic hypothesis structure;

[0016] Update the hypothesis according to the time consistency analysis result, and update the attributes of the nodes and each alarm set that have a connection relationship with the differential nodes in the intermediate event.

[0017] Wherein, the addition of the hypothesis includes:

[0018] Relevance Analysis: Calculate the cut set set of the differential nodes according to the differential nodes and connection relationship in the intermediate event;

[0019] Time consistency analysis: Determine whether the activation time of the difference node is consistent with the propagation time from the fault mode included in each cut set to the difference node. When the activation time of the difference node is consistent with the propagation time from the fault mode included in each cut set to the difference node, add an assumption based on the pre-constructed basic assumption structure, determine the fault mode in which the assumption occurs, and determine the node attributes and each alarm set in the new assumption according to the correlation and time consistency between this new event and all previously activated difference nodes and each added new assumption.

[0020] Among them, the method further includes:

[0021] Extract the information in the fault diagnosis model. The fault diagnosis model is an optimized time fault propagation graph. The connection relationship of the nodes included in the optimized time fault propagation graph contains AND connection relationships and OR connection relationships; the information extracted from the fault diagnosis model includes: fault mode nodes, difference nodes, connection relationships, and propagation time information.

[0022] Construct a basic assumption structure for fault diagnosis. The basic assumption structure includes the attributes of each target node, the fault mode in which the assumption occurs, and multiple alarm sets. The target nodes include fault mode nodes and difference nodes.

[0023] Among them, the attributes of the fault mode node include the earliest time and the latest time when the fault occurs;

[0024] The attributes of the difference node include the observation state, the actual activation time, the assumed state, the earliest time and the latest time when activation is expected.

[0025] Among them, optimizing the time fault propagation graph includes:

[0026] Extract the AND attributes and OR attributes of the difference nodes in the fault diagnosis model;

[0027] Set up virtual nodes according to the extracted AND attributes and OR attributes. The virtual nodes have AND connection relationships or OR connection relationships with the existing difference nodes;

[0028] Optimize the time fault propagation graph based on the connection relationship between the virtual nodes and the existing difference nodes.

[0029] Among them, the evaluation indicators include: credibility and robustness.

[0030] In a second aspect, there is provided a fault diagnosis reasoning device based on a time fault propagation graph. The device includes:

[0031] A generation module, configured to generate an assumption: when the first event arrives, establish a new assumption;

[0032] Update module for updating hypotheses: When diagnosing intermediate events, different update operations are performed on a hypothesis according to different situations where a new event and a certain hypothesis meet preset conditions. The preset conditions are that the fault mode of the new event and a certain hypothesis is related based on the node connection relationship and the time is consistent. The intermediate event is an event other than the first event, and the connection relationship includes an AND connection relationship and an OR connection relationship;

[0033] Addition module for adding hypotheses: When a new event does not meet the preset conditions with all existing hypotheses in the hypothesis library, a new hypothesis is generated;

[0034] Evaluation module for evaluating the effect of each hypothesis based on evaluation metrics to obtain a fault diagnosis inference result.

[0035] The fault diagnosis inference method and device based on a time fault propagation graph provided by this application can achieve automatic fault detection and isolation of the system. The inference method comprehensively considers various situations, including: multi-fault situations with "AND" logic on the propagation path, situations where multiple abnormal events are triggered simultaneously, situations where events are false alarms, etc. The output of the inference is a fault hypothesis, which can provide interval estimates of the fault mode and the fault occurrence time, support the alarm collection of the hypothesis, the set of false alarms, the set of missing alarms, and the set of expected alarms. The hypothesis is evaluated in combination with rationality and robustness evaluation metrics to assist decision-makers in determining the alarm cause in a timely manner, and then formulating corresponding maintenance measures, which can improve the reliability, maintainability, and safety of the product and is beneficial to reducing the maintenance support cost. Description of the Drawings

[0036] Figure 1 It is a flowchart of the fault diagnosis inference method based on a time fault propagation graph provided by this application;

[0037] Figure 2 It is a schematic diagram of the TFPG of the dual-redundancy integrated display unit in the case provided by this application;

[0038] Figure 3 It is a schematic diagram of a diagnostic display result provided by this application;

[0039] Figure 4 It is another schematic diagram of a diagnostic display result provided by this application. Detailed Embodiments

[0040] The following further details this application through specific embodiments and the drawings.

[0041] In the related art, for an avionics system, the fault diagnosis method based on a directed graph requires no large amount of monitoring data, and can use information such as the extracted system connection relationship and causal propagation path to implement fault modeling, providing an effective way for fault diagnosis. Among them, TFPG can not only express the complex causal relationship of system fault propagation, but also consider the fault propagation time constraint of a dynamic system, making the abstract fault model better reflect the actual system's fault propagation situation.

[0042] For a simple and small system, usually the time fault propagation graph is also relatively simple, and there is often only one directed edge connection between nodes, without considering the comprehensive influence of multiple different nodes. However, for the time fault propagation model of complex products such as avionics systems, it is necessary to distinguish the "AND / OR" attributes of different nodes. If there is a propagation path with "AND" logic, there may be multiple faults occurring, and this kind of situation requires a differentiated time consistency check of the nodes and a comprehensive reasoning algorithm design. To meet this practical need, a diagnostic reasoning algorithm based on the time fault propagation graph is proposed for an avionics system, so as to improve the accuracy and rationality of fault diagnosis work.

[0043] This application provides a fault diagnosis reasoning method based on a time fault propagation graph, as Figure 1 shown, the method includes:

[0044] Step 110, generating an assumption: When the first event arrives, a new assumption is established.

[0045] Step 120, updating an assumption: When diagnosing an intermediate event, different update operations are performed on the assumption according to different situations where the new event and a certain assumption meet the preset conditions. Among them, the preset conditions are that the fault mode of the new event and a certain assumption is related based on the node connection relationship and is time-consistent. Among them, the intermediate event is an event other than the first event, and the connection relationship includes an AND connection relationship and an OR connection relationship, that is, the connection relationship includes an AND connection relationship and an OR connection relationship.

[0046] Step 130, adding an assumption: If the new event does not meet the preset conditions with all the existing assumptions in the assumption library, a new assumption is generated.

[0047] Step 140, evaluating the effect of each assumption based on evaluation indicators to obtain the fault diagnosis reasoning result.

[0048] Among them, the evaluation indicators include: credibility and robustness.

[0049] Among them, generating an assumption includes:

[0050] Correlation analysis: Calculate the cut set set of the differential nodes according to the differential nodes and connection relationships in the first event. The cut set set includes multiple cut sets, and each cut set is a combination of fault modes that may cause the activation of the differential nodes. The connection relationships are extracted from the fault diagnosis model;

[0051] Time consistency analysis: Determine whether the activation time of the differential nodes is consistent with the propagation time from the fault modes included in each cut set to the differential nodes. When the activation time of the differential nodes is consistent with the propagation time from the fault modes included in each cut set to the differential nodes, generate hypotheses based on the pre-constructed hypothesis basic structure, determine the attributes of the differential nodes, the fault modes that are assumed to occur, and multiple alarm sets. The propagation time from the fault modes to the differential nodes is extracted from the fault diagnosis model.

[0052] Among them, updating hypotheses includes:

[0053] Correlation analysis: Calculate the cut set set of the differential nodes according to the differential nodes and connection relationships in the intermediate event, and determine whether at least one cut set is a subset of the diagnosis set. When at least one cut set is a subset of the diagnosis set, perform time consistency analysis. The diagnosis set is the set of fault modes that occur in the hypotheses in the hypothesis basic structure;

[0054] Update the hypotheses according to the results of the time consistency analysis, and update the attributes of the nodes and each alarm set that have connection relationships with the differential nodes in the intermediate event.

[0055] When performing time consistency analysis, there are the following four situations:

[0056] When the cut set is equal to the diagnosis set, the assumed state of d is consistent with the activation state and satisfies the propagation time constraint; when the cut set is equal to the diagnosis set, both d and f in the assumed FM satisfy the propagation time constraint; when the cut set is included in the diagnosis set, both d and f in the cut set satisfy the propagation time constraint; other situations.

[0057] Among them, adding hypotheses includes:

[0058] Correlation analysis: Calculate the cut set set of the differential nodes according to the differential nodes and connection relationships in the intermediate event;

[0059] Time consistency analysis: Determine whether the activation time of the differential nodes is consistent with the propagation time from the fault modes included in each cut set to the differential nodes. When the activation time of the differential nodes is consistent with the propagation time from the fault modes included in each cut set to the differential nodes, add hypotheses based on the pre-constructed hypothesis basic structure, determine the fault modes that are assumed to occur, and determine the node attributes and each alarm set in the new hypotheses according to the correlation and time consistency between this new event and all previously activated differential nodes and each added new hypothesis.

[0060] Furthermore, the method further includes:

[0061] Extracting information from the fault diagnosis model, where the fault diagnosis model is an optimized time fault propagation graph, and the connection relationships of the nodes included in the optimized time fault propagation graph contain AND connection relationships and OR connection relationships; the information extracted from the fault diagnosis model includes: fault mode nodes, difference nodes, connection relationships, and propagation time information;

[0062] Constructing an assumed basic structure for fault diagnosis, where the assumed basic structure includes the attributes of each target node, the assumed fault mode that occurs, and multiple alarm sets, and the target nodes include fault mode nodes and difference nodes.

[0063] Among them, the attributes of the fault mode node include the earliest time and the latest time when the fault occurs;

[0064] The attributes of the difference node include the observed state, the actual activation time, the assumed state, the earliest time and the latest time when it is expected to be activated.

[0065] Among them, optimizing the time fault propagation graph includes:

[0066] Extracting the AND attribute and the OR attribute of the difference nodes in the fault diagnosis model;

[0067] Setting up virtual nodes according to the extracted AND attribute and OR attribute, and there is an AND connection relationship or an OR connection relationship between the virtual nodes and the existing difference nodes;

[0068] Optimizing the time fault propagation graph based on the connection relationship between the virtual nodes and the existing difference nodes.

[0069] This application also provides another fault diagnosis reasoning method based on a time fault propagation graph, and this method includes the following steps:

[0070] Step 1: Extract information from the fault diagnosis model.

[0071] For the operation of the diagnostic algorithm, it is necessary to extract the information in the time fault propagation graph as the input of the algorithm, including fault mode nodes, difference nodes, connection relationships, and propagation time information, and store them in a mathematical expression form of a set or a matrix.

[0072] (1) Fault mode node set F

[0073] F = {f1, f2, f3... f m}.

[0074] (2) Difference node set D

[0075] The differential nodes include ordinary differential nodes and constructed "AND" connection nodes and "OR" connection nodes. The set of differential nodes is represented as: D = {d1, d2, d3... d n1 , o1,... o n2 , a1... a n3}. The number of ordinary differential nodes is denoted by n1, the number of "OR" connection nodes is denoted by n2, the number of "AND" connection nodes is denoted by n3, and the total number of differential nodes is denoted by n, where n1 + n2 + n3 = n.

[0076] (3) Adjacency matrix A

[0077] A is an (n + m) × (n + m) - dimensional matrix that records the connection relationships between nodes v ∈ (F ∪ D). If there is a connection between node v i and node v j , then A ij = 1; otherwise A ij = 0.

[0078] (4) Reachability matrix A *

[0079] A * is an (n + m) × (n + m) - dimensional matrix that records whether there is a path between two nodes. If there exists a path from v i to v j , then A * ij = 1; otherwise A * ij = 0.

[0080] (5) Minimum propagation time matrix t min

[0081] t min is an (n + m) × (n + m) - dimensional matrix that is associated with the adjacency matrix A. The element t ij min represents the minimum time required for a fault to propagate from the parent node v i to the child node v j (when A ij = 1). If A ij = 0, then t ij min = ∞.

[0082] (6) Maximum propagation time matrix t max

[0083] t max is an (n + m) × (n + m) - dimensional matrix that is associated with the adjacency matrix A. The element t ij maxIndicates that the fault propagates from the parent node v i to the child node v j The maximum time required (A ij = 1). If A ij = 0, then t ij max = ∞.

[0084] (7) Minimum reachable time matrix A min

[0085] A min is an (n + m) × (n + m) dimensional matrix, associated with the reachability matrix A * . The element A ij min represents the minimum time for the fault to propagate from one node v i to another node v j (A ij * = 1). If A ij * = 0, then A ij min = ∞.

[0086] (8) Maximum reachable time matrix A max

[0087] A max is an (n + m) × (n + m) dimensional matrix, associated with the reachability matrix A * . The element A ij max represents the maximum time for the fault to propagate from one node v i to another node v j (A ij * = 1). If A ij * = 0, then A ij max = ∞.

[0088] Step 2: Construct the basic structure of the fault diagnosis hypothesis.

[0089] The output of the diagnostic reasoning algorithm is a series of fault hypotheses. Therefore, it is necessary to pre - specify the basic structure of the hypothesis, including the attributes of each node, the fault modes that occur, and each alarm set.

[0090] The basic structure of the hypothesis: h = (F, D, FM, SA, FA, MA, PA).

[0091] (1) The fault node includes two attributes: the earliest time and the latest time.

[0092] a) Earliest time f.tearl: It is the earliest time estimate of the occurrence of the fault mode f derived according to the time consistency principle and propagation relationship due to the change of the node state in the model.

[0093] b) Latest time f.tlat: It is the latest time estimate of the occurrence of the fault mode f derived according to the time consistency principle and propagation relationship due to the change of the node state in the model.

[0094] (2) The differential node attributes include the observed state, activation time, assumed state, earliest time, and latest time.

[0095] a) Observed state d.oState: It represents whether an alarm actually occurs at the differential node.

[0096] If the differential node d is activated (i.e., an alarm occurs), then d.oState = 1; otherwise, d.oState = 0.

[0097] b) Activation time d.time: It is the time when the differential node is activated, the time when the alarm occurs at d.

[0098] If d is activated at time t, then d.time = t.

[0099] c) Assumed state d.pState: Corresponding to the observed state, each differential node has an assumed state.

[0100] If there is a predecessor node of d that is activated and d has not been activated yet, and according to the propagation path, it is expected that d will be activated, then the assumed state of d is ON, that is, d.pState = 1.

[0101] d) Earliest time d.tearl: It represents the earliest time estimate of the occurrence of an alarm at the differential node d.

[0102] The default value of d.tearl is 0. If a certain node d has an alarm at time t, then d.tearl = d.time = t.

[0103] e) Latest time d.tlat: It represents the latest time estimate of the occurrence of an alarm at the differential node d.

[0104] The default value of d.tlat is 0. If a certain node d has an alarm at time t, then d.tlat = d.time = t.

[0105] (3) Diagnosis set FM: The set contains at least one fault mode inferred in the hypothesis.

[0106] (4) Consistent alarm set SA: It is the set of activated nodes that are related to the connection relationship in FM in the hypothesis and are time-consistent.

[0107] (5) False alarm set FA: It is the set of nodes that are activated when they should not occur under the current hypothesis h, including two categories: a) Nodes that generate alarms and are not related to FM; b) Nodes that are related to FM but do not meet the time consistency.

[0108] (6) Missing alarm set MA: It is the set of nodes that should occur but are not activated under the current hypothesis h;

[0109] (7) Expected alarm set PA: It is the set of nodes that are related to the connection relationship of hypothesis h, have not occurred at the current moment, and may be activated later according to the propagation path.

[0110] Step 3: Conduct correlation analysis and time consistency analysis for each event.

[0111] The process of generating a fault hypothesis requires conducting correlation analysis and time consistency analysis for each abnormal event. The event includes the activation node identifier d and the activation time t, denoted as <d, t>. Correlation analysis includes the direct connection or indirect connection between nodes, the requirements for activating different nodes, and the correlation between the activated different nodes and the hypothesis. Time consistency analysis determines whether there is a contradiction between the occurrence of the event and the time constraint of the propagation of the fault effect.

[0112] 1) Correlation analysis

[0113] (1) Correlation between nodes

[0114] Use the adjacency matrix A and the reachability matrix A * to make a correlation judgment. If A ij = 1, then node i is directly related to node j. Node i is called the parent node of node j, and node j is called the child node of node i. If A ij * = 1, then node i is indirectly related to node j. Node i is called the subsequent node of node j, and node j is called the previous node of node i.

[0115] (2) Correlation between different nodes and fault mode nodes

[0116] When generating possible fault hypotheses, it is necessary to find the fault modes related to the different nodes included in the event. The "OR connection relationship" results in more than one path from the fault mode to the different nodes, while the "AND connection relationship" requires multiple paths to reach the different nodes to activate the alarm. Calculate the cut set set of the different nodes according to the different nodes and connection relationships in the event. The cut set refers to the set of fault modes that can activate the node. The occurrence of all faults in the set can lead to the occurrence of the abnormal event.

[0117] Derive layer by layer backward from the event node d according to the adjacency matrix. When encountering an "AND" node, increase the order of the cut set (the number of cut set elements). When encountering an "OR" node, increase the number of cut sets until all elements in the cut set are failure modes. Finally, merge and eliminate the duplicate elements and duplicate cut sets in the cut set.

[0118] (3) Correlation between the difference node and the hypothesis

[0119] The correlation analysis between the difference node and the hypothesis is to compare each cut set of the difference node with the diagnostic set of each hypothesis, and divide the comparison results into the following three categories:

[0120] Category 1: A certain cut set is equal to a certain diagnostic set;

[0121] Category 2: A certain cut set is included in a certain diagnostic set;

[0122] Category 3: None of the cut sets is a subset of the diagnostic set;

[0123] When at least one cut set is a subset of the diagnostic set, there is a correlation between the difference node and the hypothesis.

[0124] 2) Temporal consistency analysis

[0125] When the correlation condition is satisfied, perform temporal consistency analysis on the event <d, t>. Temporal consistency analysis is to judge whether there is a contradiction between the occurrence of the event and the time constraint of the failure influence propagation.

[0126] (1) For two difference nodes d i , d j ∈D, if the two nodes are temporally consistent at time t, then on the premise of meeting the correlation requirement (A ij * =1), satisfy any of the following constraints:

[0127] Constraint 1: d i is not activated, and d j is not activated;

[0128] Constraint 2: d i is activated, d j is not activated, but t is earlier than the latest time estimate when d j is activated

[0129] Constraint 3: d i is activated, d j is activated, and the interval time conforms to the model propagation time interval

[0130] (2) If the child node is an "OR" node, it will be activated when any of its parent nodes propagates a fault to it. For the two nodes to be consistent, under the premise of meeting the correlation requirement ()(A ij * = 1), it needs to meet any of the following constraints:

[0131] Constraint 1;

[0132] Constraint 2;

[0133] Constraint 3:

[0134] Constraint 4 - O: d i Not activated, d j Activated, there is another path reaching

[0135] (3) If the child node is an "AND" node, it will be activated only when all of its parent nodes propagate a fault to it. For the two nodes to be consistent, under the premise of meeting the correlation requirement (A ij * = 1), it needs to meet any of the following conditions:

[0136] Constraint 1;

[0137] Constraint 2 - A: d i Activated, d j Not activated, d j There is at least one parent node that has not propagated a fault to

[0138] Constraint 3 - A: d i Activated, d j Activated, d j Meet the latest time constraint on any path that propagates to it, and meet the earliest time constraints for all propagation paths:

[0139]

[0140] Step 4: Generate and update fault hypotheses based on the analysis results

[0141] 1) Generate hypotheses: When the first event arrives, since the hypothesis library is empty, new hypotheses can be directly generated.

[0142] Correlation analysis: Calculate the cut - set set of d;

[0143] Time consistency analysis: Judge whether both d and f in its cut - set meet the propagation time constraint (Condition 1);

[0144] If d and one of its cut - sets meet the propagation time constraint, generate a hypothesis; when the number of cut - set elements of the differential node is greater than 1, generate multiple hypotheses;

[0145] If d does not satisfy the propagation time constraint with any of its cut sets, then classify this event as an error alarm generation hypothesis.

[0146] 2) Update hypotheses: When diagnosing intermediate events, update the existing hypotheses in the hypothesis library according to the information of the intermediate events.

[0147] Correlation analysis: Determine the category according to the correlation between the differential node and the hypothesis, and conduct time consistency analysis according to different categories;

[0148] Time consistency analysis:

[0149] (1) Category 1: A certain cut set is equal to a certain diagnostic set;

[0150] The hypothesis state of d is consistent with the activation state and satisfies the propagation time constraint. Update SA, PA, and MA in the hypothesis, and update the attributes of the previous nodes of d.

[0151] Both d and f in the FM of the hypothesis satisfy the propagation time constraint. Update SA, PA, and MA in the hypothesis, update the attributes of the previous nodes of d, and update the attributes of f in the FM.

[0152] (2) Category 2: A certain cut set is included in a certain diagnostic set;

[0153] Both d and f in the cut set satisfy the propagation time constraint. Update SA, PA, and MA in the hypothesis, update the attributes of the previous nodes of d, and update the attributes of f included in the cut set.

[0154] (3) Other cases

[0155] Update FA in the hypothesis.

[0156] 3) Add hypotheses: If the intermediate event cannot be explained by the existing hypotheses, new hypotheses need to be added.

[0157] Correlation analysis: Calculate the cut set set of d;

[0158] Time consistency analysis: Judge whether both d and f in its cut set satisfy the propagation time constraint (Condition 1);

[0159] If d satisfies the propagation time constraint with one of its cut sets, generate a hypothesis; when the number of cut set elements of the differential node is greater than 1, generate multiple hypotheses;

[0160] In addition, determine each element in the new hypothesis structure according to the correlation and time consistency between all the nodes activated before the current node and each generated new hypothesis.

[0161] Step Five: Diagnosis Hypothesis Evaluation and Selection

[0162] To determine which hypothesis can best explain the current abnormal state of the system, the hypotheses are evaluated from two aspects.

[0163] 1) Rationality. Rationality can be measured using credibility, which is expressed as the ratio of the number of alarms supporting the hypothesis to the total number of alarms. The main focus is on the fault hypothesis that can explain the current observed state.

[0164]

[0165] 2) Robustness. Robustness represents the property that the hypothesis remains unchanged, reflecting the degree of influence of the fault difference nodes that are expected to occur on the current hypothesis, and providing assistance in determining when to take actions to repair the system.

[0166]

[0167] Taking the TFPG of the dual-redundancy hot standby integrated display unit in this application as an example, a fault diagnosis and reasoning method based on the time fault propagation graph is described as follows:

[0168] (1) Extract the information in the fault diagnosis model of the avionics system.

[0169] The TFPG model is as Figure 2 shown, and the information required for fault diagnosis is extracted from it.

[0170] There are 6 fault mode nodes, F = {f1, f2, f3, f4, f5, f6}.

[0171] (2) There are 6 difference nodes, including 3 original difference nodes, 2 OR connection nodes, and 1 AND connection node, D = {d1, d2, d3, o1, o2, a1}.

[0172] (3) Adjacency matrix A

[0173] Table 1 Adjacency matrix A

[0174]

[0175]

[0176] (4) Reachable matrix A *

[0177] Table 2 Reachable matrix A *

[0178] 1 2 3 4 5 6 7 8 9 10 11 12 1 0 0 0 0 0 0 1 0 1 1 0 1 2 0 0 0 0 0 0 1 0 1 1 0 1 3 0 0 0 0 0 0 1 0 1 1 0 1 4 0 0 0 0 0 0 0 1 1 0 1 1 5 0 0 0 0 0 0 0 1 1 0 1 1 6 0 0 0 0 0 0 0 1 1 0 1 1 7 0 0 0 0 0 0 0 0 1 0 0 1 8 0 0 0 0 0 0 0 0 1 0 0 1 9 0 0 0 0 0 0 0 0 0 0 0 0 10 0 0 0 0 0 0 1 0 1 0 0 1 11 0 0 0 0 0 0 0 1 1 0 0 1 12 0 0 0 0 0 0 0 0 1 0 0 0

[0179] (5) Minimum propagation time matrix t min

[0180] The propagation time between unconnected nodes in the model is labeled as 9999.

[0181] Table 3 Minimum propagation time matrix t min

[0182] 1 2 3 4 5 6 7 8 9 10 11 12 1 0 ∞ ∞ ∞ ∞ ∞ ∞ ∞ ∞ 20 ∞ ∞ 2 ∞ 0 ∞ ∞ ∞ ∞ ∞ ∞ ∞ 30 ∞ ∞ 3 ∞ ∞ 0 ∞ ∞ ∞ ∞ ∞ ∞ 50 ∞ ∞ 4 ∞ ∞ ∞ 0 ∞ ∞ ∞ ∞ ∞ ∞ 20 ∞ 5 ∞ ∞ ∞ ∞ 0 ∞ ∞ ∞ ∞ ∞ 30 ∞ 6 ∞ ∞ ∞ ∞ ∞ 0 ∞ ∞ ∞ ∞ 50 ∞ 7 ∞ ∞ ∞ ∞ ∞ ∞ 0 ∞ ∞ ∞ ∞ 0 8 ∞ ∞ ∞ ∞ ∞ ∞ ∞ 0 ∞ ∞ ∞ 0 9 ∞ ∞ ∞ ∞ ∞ ∞ ∞ ∞ 0 ∞ ∞ ∞ 10 ∞ ∞ ∞ ∞ ∞ ∞ 0 ∞ ∞ 0 ∞ ∞ 11 ∞ ∞ ∞ ∞ ∞ ∞ ∞ 0 ∞ ∞ 0 ∞ 12 ∞ ∞ ∞ ∞ ∞ ∞ ∞ ∞ 0 ∞ ∞ 0

[0183] (6) Maximum propagation time matrix t max

[0184] Table 4 Maximum propagation time matrix t max

[0185]

[0186]

[0187] (7) Minimum reachable time matrix A min

[0188] Table 5 Minimum reachable time matrix A min

[0189] 1 2 3 4 5 6 7 8 9 10 11 12 1 0 ∞ ∞ ∞ ∞ ∞ 20 ∞ 20 20 ∞ 20 2 ∞ 0 ∞ ∞ ∞ ∞ 30 ∞ 30 30 ∞ 30 3 ∞ ∞ 0 ∞ ∞ ∞ 50 ∞ 50 50 ∞ 50 4 ∞ ∞ ∞ 0 ∞ ∞ ∞ 20 20 ∞ 20 20 5 ∞ ∞ ∞ ∞ 0 ∞ ∞ 30 30 ∞ 30 30 6 ∞ ∞ ∞ ∞ ∞ 0 ∞ 50 50 ∞ 50 50 7 ∞ ∞ ∞ ∞ ∞ ∞ 0 ∞ 0 ∞ ∞ 0 8 ∞ ∞ ∞ ∞ ∞ ∞ ∞ 0 0 ∞ ∞ 0 9 ∞ ∞ ∞ ∞ ∞ ∞ ∞ ∞ 0 ∞ ∞ ∞ 10 ∞ ∞ ∞ ∞ ∞ ∞ 0 ∞ 0 0 ∞ 0 11 ∞ ∞ ∞ ∞ ∞ ∞ ∞ 0 0 ∞ 0 0 12 ∞ ∞ ∞ ∞ ∞ ∞ ∞ ∞ 0 ∞ ∞ 0

[0190] (8) Maximum reachable time matrix A max

[0191] Table 6 Maximum reachable time matrix A max

[0192] 1 2 3 4 5 6 7 8 9 10 11 12 1 0 ∞ ∞ ∞ ∞ ∞ 40 ∞ 40 40 ∞ 40 2 ∞ 0 ∞ ∞ ∞ ∞ 40 ∞ 40 40 ∞ 40 3 ∞ ∞ 0 ∞ ∞ ∞ 50 ∞ 50 50 ∞ 50 4 ∞ ∞ ∞ 0 ∞ ∞ ∞ 40 40 ∞ 40 40 5 ∞ ∞ ∞ ∞ 0 ∞ ∞ 40 40 ∞ 40 40 6 ∞ ∞ ∞ ∞ ∞ 0 ∞ 50 50 ∞ 50 50 7 ∞ ∞ ∞ ∞ ∞ ∞ 0 ∞ 0 ∞ ∞ 0 8 ∞ ∞ ∞ ∞ ∞ ∞ ∞ 0 0 ∞ ∞ 0 9 ∞ ∞ ∞ ∞ ∞ ∞ ∞ ∞ 0 ∞ ∞ ∞ 10 ∞ ∞ ∞ ∞ ∞ ∞ 0 ∞ 0 0 ∞ 0 11 ∞ ∞ ∞ ∞ ∞ ∞ ∞ 0 0 ∞ 0 0 12 ∞ ∞ ∞ ∞ ∞ ∞ ∞ ∞ 0 ∞ ∞ 0

[0193] 2) Construct the basic structure of the fault diagnosis hypothesis: h = (F, D, FM, SA, FA, MA, PA).

[0194] (1) F = {f1, f2, f3, f4, f5, f6}, and the initial values of the earliest time f.tearl and the latest time f.tlat of the fault nodes are 0.

[0195] Table 7 Fault node attributes and initial values

[0196]

[0197]

[0198] (2) D = {d1, d2, d3, o1, o2, a1}, and the initial values of the observation state, activation time, hypothesis state, earliest time, and latest time of the difference nodes are shown in the following table.

[0199] Table 8 Difference node attributes and initial values

[0200] d d.oState d.time d.hState d.tearl d.tearl <![CDATA[d1]]> 0 - 0 0 0 <![CDATA[d2]]> 0 - 0 0 0 <![CDATA[d3]]> 0 - 0 0 0 <![CDATA[o1]]> 0 - 0 0 0 <![CDATA[o2]]> 0 - 0 0 0 <![CDATA[a1]]> 0 - 0 0 0

[0201] (3)FM = [], SA = [], FA = [], MA = [], PA = [].

[0202] 3) Perform correlation analysis and time consistency analysis for each event; generate and update fault hypotheses based on the analysis results; evaluate and select diagnostic hypotheses.

[0203] To trigger the diagnostic reasoning process, the simulation events <d = d2, t = 25> and <d = d3, t = 30> are used in this example, that is, d2 is activated at 25 seconds; d3 is activated at 30 seconds, and fault diagnosis reasoning is performed for each event.

[0204] (1) <d = d2, t = 25>

[0205] Correlation analysis: The parent node on the fault propagation path of d2 is an "OR" node, and the calculated fault cut set set of d2 is {[f4], [f5], [f6]}, that is, the possible faults are f4 or f5 or f6.

[0206] Time consistency analysis: Calculate according to the time consistency requirements. The calculated time interval estimate for the occurrence of fault f4 is [0, 5], while f5 and f6 are excluded because they do not meet the time consistency constraints.

[0207] Generate hypothesis: Since this event is the first event, according to the analysis results, a hypothesis is generated with f4.

[0208] Hypothesis evaluation: At the current moment (t = 25), since there are no false alarms, missing alarms, and expected alarms, both the credibility and robustness are 1 (the following formula is the index calculation process for Hypothesis 1).

[0209]

[0210]

[0211] Figure 3 The shown diagnostic display result (t = 25) includes partial information of the hypothesis structure and hypothesis evaluation indicators.

[0212] (2) <d = d3, t = 30>

[0213] Correlation analysis: There are both "AND" nodes and "OR" nodes on the fault propagation path of d3, and the cut set set of d2 can be obtained as follows:

[0214] {[f1,f4],[f2,f4],[f3,f4],[f1,f5],[f2,f5],[f3,f5],[f1,f6],[f2,f6],[f3,f6]}. The cases where the cut sets are all double faults conflict with the existing single-fault assumption.

[0215] Time consistency analysis: Not involved.

[0216] Update hypothesis: Since the correlation requirement is not met, the FA set of the existing hypothesis is updated. d3 is a false alarm.

[0217] Hypothesis evaluation: The credibility of the hypothesis is reduced to 0.5 (the following formula is the index calculation process of Hypothesis 1).

[0218]

[0219]

[0220] In addition, since d3 does not support any of the existing hypotheses, the hypothesis addition process will be entered.

[0221] Correlation analysis: There are both "AND" nodes and "OR" nodes on the fault propagation path of d3. The cut set set of d2 can be obtained as:

[0222] {[f1,f4],[f2,f4],[f3,f4],[f1,f5],[f2,f5],[f3,f5],[f1,f6],[f2,f6],[f3,f6]}.

[0223] Time consistency analysis: For each fault mode f in each cut set, the time consistency with d2 is judged. The cut set set that meets the consistency requirement is {[f1,f4],[f2,f4]}. Among them, the time estimate of f1 is [0,10] for both, the time estimate of f2 is [0,0] for both, and the time estimate of f4 is [0,5].

[0224] Add hypothesis: New Hypotheses 2-3 are generated. Since d1 on the propagation path of f1 / f2 and d3 is not activated, this node is classified as a missing alarm MA.

[0225] Hypothesis evaluation: The credibility of the hypothesis is 0.67 (the following formula is the index calculation process of Hypothesis 2, and the same for Hypothesis 3). The credibility of Hypotheses 2-3 is greater than that of Hypothesis 1, and the faults are isolated to f1 / f2 and f4.

[0226]

[0227]

[0228] The diagnostic display result at t = 30 is asFigure 4 as shown

[0229] This application also provides a fault diagnosis and reasoning device based on a time fault propagation graph, and the device includes:

[0230] A generation module, configured to generate an assumption: when the first event arrives, a new assumption is established;

[0231] An update module, configured to update an assumption: when diagnosing an intermediate event, update the existing assumptions in the assumption library according to the information of the newly added event. If the newly added event and a certain assumption meet the preset conditions, increase the credibility of this assumption; if the newly added event and a certain assumption do not meet the preset conditions, reduce the credibility of this assumption, where the preset condition is that the fault mode of the newly added event and a certain assumption is related based on the node connection relationship and the time is consistent, and the intermediate event is an event other than the first event, and the connection relationship includes an AND connection relationship and an OR connection relationship;

[0232] An addition module, configured to add an assumption: if the newly added event does not meet the preset conditions with all the existing assumptions in the assumption library, generate a new assumption;

[0233] An evaluation module, configured to evaluate the effect of each assumption based on an evaluation index to obtain a fault diagnosis and reasoning result.

[0234] The fault diagnosis and reasoning method and device based on the time fault propagation graph provided by this application can realize automatic fault detection and isolation of the system. The reasoning method comprehensively considers various situations, including: multi-fault situations where there is an "AND" logic on the propagation path, situations where multiple abnormal events are triggered simultaneously, situations where the event is a false alarm, etc. The output of the reasoning is a fault assumption, which can provide an interval estimate of the fault mode and the fault occurrence time, support the alarm collection of the assumption, the set of false alarms, the set of lost alarms, and the set of expected alarms. Combining rationality and robustness evaluation indicators to evaluate the assumption, assisting decision-makers to determine the alarm reason in a timely manner, and then formulating corresponding maintenance measures, can improve the reliability, maintainability, and safety of the product, and is beneficial to reducing the maintenance support cost.

[0235] The above only expresses the implementation manners of this application, and the description is relatively specific and detailed, but it cannot be understood as a limitation on the scope of the invention patent. It should be noted that for those of ordinary skill in the art, without departing from the concept of this application, several deformations and improvements can still be made, and these all belong to the protection scope of this application.

Claims

1. A fault diagnosis and reasoning method based on a time fault propagation graph, characterized in that The method includes: Generating a hypothesis: When the first event arrives, a new hypothesis is established. Updating a hypothesis: Diagnosing intermediate events, and performing different update operations on the hypothesis according to different situations where the new event and a certain hypothesis meet the preset conditions. The preset conditions are that the failure mode of the new event and a certain hypothesis is related based on the node connection relationship and the time is consistent. The intermediate event is an event other than the first event, and the connection relationship includes an AND connection relationship and an OR connection relationship. Adding a hypothesis: If the new event does not meet the preset conditions with all existing hypotheses in the hypothesis library, a new hypothesis is generated. Evaluating the effect of each hypothesis based on evaluation metrics to obtain a fault diagnosis inference result.

2. The method according to claim 1, wherein The generating of the hypothesis includes: Correlation analysis: Calculating the cut set set of the differential nodes according to the differential nodes and connection relationships in the first event. The cut set set includes multiple cut sets, and each cut set is a combination of failure modes that may cause the activation of the differential nodes. The connection relationship is extracted from the fault diagnosis model. Time consistency analysis: Judging whether the activation time of the differential nodes is consistent with the propagation time from the failure modes included in each cut set to the differential nodes. When the activation time of the differential nodes is consistent with the propagation time from the failure modes included in each cut set to the differential nodes, a hypothesis is generated based on the pre-constructed hypothesis basic structure, and the attributes of the differential nodes, the failure modes that occur in the hypothesis, and multiple alarm sets are determined. The propagation time from the failure modes to the differential nodes is extracted from the fault diagnosis model.

3. The method according to claim 1, characterized in that The updating of the hypothesis includes: Correlation analysis: Calculating the cut set set of the differential nodes according to the differential nodes and connection relationships in the intermediate event, and judging whether at least one cut set is a subset of the diagnosis set. When at least one cut set is a subset of the diagnosis set, time consistency analysis is performed. The diagnosis set is the set of failure modes that occur in the hypothesis in the hypothesis basic structure. Updating the hypothesis according to the time consistency analysis result, and updating the attributes of the nodes and each alarm set that have a connection relationship with the differential nodes in the intermediate event.

4. The method according to claim 1, characterized in that, The adding of the hypothesis includes: Correlation analysis: Calculating the cut set set of the differential nodes according to the differential nodes and connection relationships in the intermediate event. Time consistency analysis: Judging whether the activation time of the differential nodes is consistent with the propagation time from the failure modes included in each cut set to the differential nodes. When the activation time of the differential nodes is consistent with the propagation time from the failure modes included in each cut set to the differential nodes, a hypothesis is added based on the pre-constructed hypothesis basic structure, and the attributes of the differential nodes, the failure modes that occur in the hypothesis, and multiple alarm sets are determined. And according to the correlation and time consistency between all the previously activated differential nodes and each newly added hypothesis before this new event, the node attributes and each alarm set in the new hypothesis are determined.

5. The method according to claim 1, characterized in that The method further includes: Extract the information in the fault diagnosis model, where the fault diagnosis model is an optimized time fault propagation graph, and the connection relationships of the nodes included in the optimized time fault propagation graph contain AND connection relationships and OR connection relationships; the information extracted from the fault diagnosis model includes: fault mode nodes, difference nodes, connection relationships, and propagation time information. Construct the basic structure of the hypothesis for fault diagnosis, where the basic structure of the hypothesis includes the attributes of each target node, the fault mode assumed to occur, and multiple alarm sets, and the target nodes include fault mode nodes and difference nodes.

6. The method according to claim 5, wherein The attributes of the fault mode node include the earliest time and the latest time when the fault occurs. The attributes of the difference node include the observed state, the actual activation time, the assumed state, the earliest time and the latest time when it is expected to be activated.

7. The method according to claim 5, wherein Optimize the time fault propagation graph, including: Extract the AND attribute and the OR attribute of the difference nodes in the fault diagnosis model; Set up virtual nodes according to the extracted AND attribute and OR attribute, and there is an AND connection relationship or an OR connection relationship between the virtual nodes and the existing difference nodes; Optimize the time fault propagation graph based on the connection relationship between the virtual nodes and the existing difference nodes.

8. The method according to claim 1, characterized in that, The evaluation metrics include: credibility and robustness.

9. A fault diagnosis and reasoning device based on a time fault propagation graph, characterized in that The device includes: A generation module, configured to generate a hypothesis: when the first event arrives, establish a new hypothesis; An update module, configured to update the hypothesis: when diagnosing intermediate events, update the existing hypotheses in the hypothesis library according to the information of the newly added events. If the newly added event and a certain hypothesis meet the preset conditions, increase the credibility of the hypothesis; if the newly added event and a certain hypothesis do not meet the preset conditions, decrease the credibility of the hypothesis, where the preset condition is that the fault mode of the newly added event and a certain hypothesis are related based on the node connection relationship and the time is consistent, and the intermediate event is an event other than the first event, and the connection relationship includes an AND connection relationship and an OR connection relationship; An addition module, configured to add a hypothesis: if the newly added event does not meet the preset conditions with all the existing hypotheses in the hypothesis library, generate a new hypothesis; An evaluation module, configured to evaluate the effect of each hypothesis based on the evaluation metrics to obtain the fault diagnosis reasoning result.

Images