FMES data generation method and system for airborne system safety analysis

Abstract

The present invention belongs to the technical field of aviation airborne system safety analysis, and specifically relates to a method and system for generating FMES data for airborne system safety analysis. The method comprises: S1: acquiring FMEA data of the airborne system, performing data processing, and extracting data features using the BERT model; S2: using sinusoidal similarity to measure the FEMA structured data feature vectors and calculate the similarity of the FEMA structured data features; S3: obtaining the FEMA structured data feature similarity matrix and performing a first agglomerative hierarchical clustering analysis; S4: performing a second clustering analysis on the tree-like clusters of the FEMA structured data features based on fault tree analysis (FTA); and S5: generating structured Failure Mode and Effect Summary (FMES) data based on the second clustering results. The present invention proposes a complete framework for automatically generating FMES data, and improves the quality of the automatic generation through secondary clustering.

Description

Technical Field

[0001] The present invention belongs to the technical field of aviation airborne system safety analysis, and in particular relates to a FMES data generation method and system for airborne system safety analysis. Background Art

[0002] With the rapid development of the modern aviation industry, the functionality and performance of aviation equipment have significantly improved. Simultaneously, the structural complexity of airborne systems has also increased exponentially. Safety, as a core quality characteristic of aviation equipment, has attracted considerable attention. Currently, aviation equipment manufacturers generally adopt systems engineering thinking, using system safety analysis methods to design and verify the safety of airborne systems. Representative standards and guidelines include MIL-STD-882D, GJB900A, and SAEARP4761.

[0003] System safety analysis runs throughout the entire lifecycle of aircraft development. Its core approach is to achieve the following goals through continuous iteration: clarifying system safety requirements, optimizing system safety design, and conducting system safety assessments (SSAs). Specific methods encompass a range of mature analysis techniques, including functional hazard analysis (FHA), fault tree analysis (FTA), failure mode and effects analysis (FMEA), failure mode and effects summary (FMES), common mode analysis (CMA), and common cause analysis (CCA). These methods work together to provide strong theoretical support for airborne system safety design and verification, ensuring the reliability and safety of aviation equipment in complex operating environments.

[0004] Failure Modes and Effects Analysis (FMEA) is a bottom-up, systematic approach for identifying failure modes and failure effects. FMEA can be initiated at any level. FMEAs can be quantitative or qualitative and can be performed on all types of systems (e.g., electrical, electronic, or mechanical). Common FMEA effects can be used in the FMEA Summary (FMES). The FMEA Summary (FMES) summarizes lower-level failure modes with the same effects and detection mechanisms as those in the FMEA. The FMEA Summary (FMES) can be used to support other analysis techniques in the Safety Safety Assessment (SSA) process, such as Fault Tree Analysis (FTA), Dependency Diagrams (DD), Markov Model Analysis (MA), or Model-Based Safety Analysis (MBSA). In the FMEA Summary (FMES), the failure effects derived from the FMEA analysis are refined into failure modes in the FMEA Summary (FMES), while the Effects column presents higher-level failure effects. The same failure effects and detection mechanisms are integrated and classified as a single mode in the Failure Mode and Effect Summary (FMES), providing an efficient and consistent basis for subsequent systematic safety analysis. Summary of the Invention

[0005] In response to the shortcomings of the existing technology, the present invention provides a method and system for generating FMES data for airborne system safety analysis, provides a complete framework for the automatic generation of structured failure mode and effect summaries (FMES), and proposes a method for secondary clustering based on agglomerative hierarchical clustering combined with fault tree analysis (FTA) model information in the clustering stage, thereby improving the quality of automatic generation of failure mode and effect summaries (FMES) data.

[0006] To achieve the above objectives, the present invention discloses the following technical solutions:

[0007] A method for generating FMES data for airborne system safety analysis, comprising:

[0008] S1: Obtain FMEA data of the airborne system, perform data processing, and use the BERT model to extract data features;

[0009] S11: Extract FMEA data from the airborne system, perform data conversion and preprocessing, and obtain a FEMA structured dataset The FMEA data is the failure mode and effects analysis data of the airborne system;

[0010] S12: Use the BERT model to extract the FEMA structured dataset in step S11 The data feature vector v i ;

[0011] S2: Use the sinusoidal similarity to measure the FEMA structured data feature vector v in step S1 i , calculate the similarity of FEMA structured data features as:

[0012]

[0013] Among them, S ij is the similarity of FEMA structured data features; sin(v i ,v j ) is the data feature vector v i and data feature vector v j The sinusoidal similarity of v i ·v j is the dot product of the data feature vector; ∥∥ is the Euclidean norm of the vector; v i is the i-th data feature vector; v j is the jth data feature vector;

[0014] Perform linear normalization to construct the FEMA structured data feature similarity matrix S′ of the airborne system ij ;

[0015] S3: Get the FEMA structured data feature similarity matrix in step S2 and perform the first agglomerative hierarchical clustering analysis; each FEMA structured data feature similarity matrix starts as a separate cluster, and gradually merges adjacent clusters until they are merged into one cluster, obtaining a tree cluster set of FEMA structured data features in, because but

[0016] S4: Based on the fault tree analysis FTA, the tree clustering of the FEMA structured data features in step S3 is performed Conduct secondary cluster analysis;

[0017] S41: Construct a logical expression for fault tree analysis (FTA); use a logical expression based on logic gates and bottom events to represent a fault tree;

[0018] S42: extracting the event relationship of the FEMA structured data, and iteratively calculating the minimum cut set F based on the logical expression in step S41;

[0019] S43: Determine whether the secondary clustering condition is met based on the minimum cut set F in step S42, and perform the first clustering result in step S3 based on the association relationship. Perform secondary clustering and obtain the second clustering results in

[0020] S5: Based on the second clustering result of step S4, generate structured failure mode and effect summary FMES data;

[0021] According to the second clustering results, each cluster automatically generates structured failure mode and effect summary FMES data, FMES = {fds1, fds2, ..., fds i}, used for airborne system safety analysis.

[0022] Preferably, in step S11, FMEA data is extracted from the airborne system, and data conversion and preprocessing are performed to obtain a FEMA structured data set. Specifically:

[0023] S111: Convert the extracted FMEA data into structured data and build a structured data set, specifically:

[0024] fd=[fm,fr,fe]

[0025]

[0026] Among them, fd is FMEA structured data; fm is the failure mode data of FMEA data; fr is the failure cause data of FMEA data; fe is the failure impact data of FMEA data; It is a structured dataset for FEMA;

[0027] The failure impact data fe of the FMEA data is formed by connecting the local impact data e1 and the final impact data e2, specifically:

[0028] fe=e1⊕e2

[0029]

[0030] Among them, fe is the failure impact data of FMEA data; e1 is the local impact data; e2 is the final impact data; ⊕ is the semi-additive combination; It is the FEMA structural failure impact dataset;

[0031] S112: Using population statistics to identify FEMA structured datasets The missing values ​​in the data are deleted or filled.

[0032] Preferably, in step S12, the BERT model is used to extract the FEMA structured dataset in step S11. The data feature vector v i , specifically:

[0033] S121: FEMA structured dataset Perform data segmentation operations;

[0034] S122: Determine the input word embedding vector of the BERT model:

[0035] E i =T i +S i +P i

[0036] Among them, E i is the i-th word embedding vector; T i is the i-th word vector; P i is the embedding vector for the i-th position; S i is the i-th segment embedding vector;

[0037] S123: Stacking Transformer layers to build a BERT model; the calculation method for each Transformer layer is:

[0038] H (l) =Transform(H (l-1) )

[0039] Among them, H (l) is the output vector of the lth Transformer layer, H (0) is the word embedding vector E i ; Transformer is a data processing method based on deep learning models; l is the layer number of the BERT model;

[0040] S124: Use the average pooling method to represent the data vector and obtain the FEMA structured data feature vector:

[0041]

[0042] Among them, v i is the FEMA structured data feature vector; n is the number of tokens in the text.

[0043] Preferably, linear normalization is performed in step S2 to construct the FEMA structured data feature similarity matrix S′ of the airborne system. ij , specifically:

[0044] S21: Get the FEMA structured data feature v obtained in step S1 i , construct the FEMA structured data feature vector group V as:

[0045] V=[v1,v2,…v i ,…v N ] T

[0046] Where V is the FEMA structured data feature vector; is the i-th FEMA structured data feature vector; N is the dimension of the FEMA structured data feature vector;

[0047] S22: Use the sinusoidal similarity measurement method to calculate the similarity S of FEMA structured data features ij ;

[0048] S23: Construct an N×N dimensional FEMA structured data feature similarity matrix S and perform linear normalization to obtain:

[0049]

[0050] Among them, S′ ij is the normalized result of the similarity of FEMA structured data features; S is the N×N dimensional FEMA structured data feature similarity matrix.

[0051] Preferably, in step S3, each FEMA structured data feature similarity matrix starts as a separate cluster, and adjacent clusters are gradually merged until they are merged into one cluster, specifically:

[0052] S31: Initialize the FEMA structured data feature similarity matrix, let D ij =S′ ij =d(v i ,v j ),i≠j;

[0053] S32: Merge the closest data feature clusters to determine the two data feature clusters with the smallest distance C i ,C j , perform the merge operation, specifically:

[0054]

[0055] Among them, C i is the i-th data feature cluster; C j is the jth data feature cluster; To find two data feature clusters C a and C b Minimize the distance function; C a is the ath data feature cluster; C b is the bth data feature cluster; a is the first number of the data feature cluster; b is the second number of the data feature cluster;

[0056] S33: Update the distance between data feature clusters, and set the data feature cluster C i ,C j Merge into data feature cluster C k =Ci ∪C j , and update the distance between data feature clusters, and use the average value of the distance between sample pairs within the data feature cluster to calculate the distance between data feature clusters:

[0057]

[0058] Among them, d(C a ,C b ) is the data feature cluster C a and C b distance between them; d(x,y) is the data feature cluster C a The element distance in; x is the data feature cluster C a The elements in; y is the data feature cluster C b Elements in

[0059] S34: Repeat the iteration to update the distance until a tree cluster is formed:

[0060]

[0061] in, It is a tree-shaped cluster; is a set of node clusters; is the set of edges between nodes;

[0062] S35: Tree cluster leaf node Taking each FEMA structured data feature similarity matrix as an independent cluster, by setting the interception height in the tree cluster, the first clustering result of the FEMA structured data feature tree cluster set is obtained as follows:

[0063]

[0064] in, It is a tree cluster set of FEMA structured data features; because but

[0065] Preferably, the logical expression for constructing the fault tree analysis FTA in step S41 specifically includes: an AND gate logical expression, an OR gate logical expression, a NOT gate logical expression, an XOR gate logical expression, a NAND gate logical expression, a NOR gate logical expression and a voting gate logical expression;

[0066] The AND gate logic expression states that all input events must occur simultaneously for a system failure to occur. The specific expression is:

[0067] F1=A·B

[0068] Among them, F1 is the output result of the AND gate; A is the first event of airborne system safety; B is the second event of airborne system safety;

[0069] The OR gate logic expression indicates that a system failure will occur as long as any one of the input events occurs. The specific expression is:

[0070] F2=A+B

[0071] Among them, F2 is the output result of the OR gate;

[0072] The NOT gate logic expression inverts the input event. If the input is false, the output is true. The specific expression is:

[0073]

[0074] Among them, F3 is the output result of the NOT gate;

[0075] The XOR gate logic expression has only one input event, and the system failure will occur. Event A or Event B must occur. The specific expression is:

[0076] F4=A⊕B

[0077] Among them, F4 is the output result of the XOR gate;

[0078] The logic expression of the NAND gate is the inverse of the gate. The output is true only when all input events do not occur. The specific expression is:

[0079]

[0080] Among them, F5 is the output result of the NAND gate;

[0081] The logic expression of the NOR gate is the reverse of the OR gate. When all input events do not occur, the system failure occurs. The specific expression is:

[0082]

[0083] Among them, F6 is the output result of the NOR gate;

[0084] The voting gate logic expression is that when a certain number of events occur among all the input events, the output event occurs. The specific expression is:

[0085] F7=AtLeast(k,{E1,E2,…,E n})

[0086] Among them, F7 is the output result of the voting gate; k is the minimum number of events that need to occur; E is the bottom event; AtLeast is the number of events among all bottom events {E1, E2, ..., E n} at least k occur.

[0087] Preferably, in step S42, the event relationship of the FEMA structured data is extracted, and the minimum cut set F is iteratively calculated based on the logical expression in step S41, specifically:

[0088] For the AND gate, the top event occurs if and only if all input events occur, the top event G = E1·E2, then the AND gate cut set C1 = {E1, E2};

[0089] For the OR gate, the top event occurs when any of the input events occurs, the top event G = E1 + E2, then the OR gate cut set C2 = {E1} or {E2};

[0090] For the NOT gate, the top event occurs when the input event does not occur, the top event Then the non-gate cut set C3 = {E1};

[0091] For the XOR gate, the top event occurs when exactly one of the input events occurs, and the top event G = E1⊕E2, then the XOR gate cut set C4 = {E1} or {E2};

[0092] The NAND gate is the reverse operation of the AND gate. The top event occurs when all the input events do not occur at the same time. Then the NAND gate cut set C5={E1,E2};

[0093] The NOR gate is the reverse operation of the OR gate. The top event occurs when all the input events do not occur. Then the NOR gate cut set C6 = {E1, E2};

[0094] The voting gate indicates that from multiple input events, a certain number of events must occur before the top event occurs. The cut set is the combination of k events among the n events, and the voting gate cut set is C7 = {E1, E2}, {E1, E3}, {E2, E3};

[0095] Through the above cut set event calculation method, iterative calculation is performed to obtain the minimum cut set F of the fault tree of the fault tree analysis FTA:

[0096]

[0097] Among them, F is the minimum cut set; C i ={E k …} is the minimum cut set consisting of the bottom events; m is the total number of cut sets consisting of the bottom events.

[0098] Preferably, in step S43, it is determined according to the minimum cut set F whether the secondary clustering condition is satisfied, specifically:

[0099] Get the first clustering result obtained in step S3 in The failure mode of fd1 is represented as fd1.fm; for the cluster Clu i ,Clu j , the relationship between the minimum cut set F is:

[0100]

[0101] Among them, fd i .fm is FMEA structured data fd i Failure mode; fd j .fm is FMEA structured data fd j failure modes;

[0102] Then we can determine the cluster Clu i ,Clu j The secondary clustering condition is met and secondary clustering can be performed.

[0103] Preferably, the generation of structured failure mode and effect summary (FMES) data in step S5 is specifically as follows:

[0104] The second clustering result Clus i Composed of FMEA structured data fd data, Clus i ={fd1,fd2,…,fd i}, one cluster corresponds to a failure mode and effect summary FMES data, the structure is fds = [fms, frs, pfc], where:

[0105] fms=fd1.fe

[0106]

[0107] Among them, fms is the first component of the Failure Mode and Effect Summary FMES data; frs is the second component of the Failure Mode and Effect Summary FMES data; pfc is the third component of the Failure Mode and Effect Summary FMES data;

[0108] Get the final failure mode and effect summary FMES data FMES={fds1,fds2,…,fds i}.

[0109] A second aspect of the present invention proposes an FMES data generation system for an FMES data generation method for airborne system safety analysis, which includes: an FMEA data preprocessing module, an FMEA data feature engineering module, a failure impact similarity matrix construction module, a first agglomerative hierarchical clustering module, a second clustering module based on FTA information, and a structured FMES data generation module;

[0110] The FMEA data preprocessing module is used to process the original FMEA data;

[0111] The FMEA data feature engineering module is used to convert FMEA data into vector data that retains data features and can be calculated, and extract data features;

[0112] The failure impact similarity matrix construction module is used to construct a FEMA structured data feature similarity matrix;

[0113] The first agglomerative hierarchical clustering module is used to perform the first cluster analysis of FMEA data;

[0114] The secondary clustering module based on FTA information performs a second clustering analysis on the first clustering result data in order to mine and fuse the FTA logic information;

[0115] The structured FMES data generation module automatically generates the structured failure mode and effect summary FMES data required for airborne system safety analysis based on the second cluster analysis results.

[0116] Compared with the prior art, the present invention has the following beneficial effects:

[0117] (1) The present invention performs data processing on the standardized and vectorized failure mode and effects analysis (FMEA) data. By preprocessing and extracting features from the original FMEA data, the data is standardized and vectorized, ensuring that the data format is unified and suitable for subsequent analysis, thereby improving the standardization and adaptability of data processing.

[0118] (2) The present invention combines the two-layer clustering framework of fault tree analysis (FTA) information and adopts the first agglomerative hierarchical clustering method to preliminarily classify the failure impact data. Then, the structured information of the fault tree analysis (FTA) is introduced to perform secondary clustering based on the onboard system logic. The causal relationship information of the fault tree analysis (FTA) is used to improve the accuracy and quality of the clustering results.

[0119] (3) The present invention automatically generates structured Failure Mode and Effect Summary (FMES) data and calculates the failure probability by integrating clustering results with Failure Mode and Effect Analysis (FMEA) data; it realizes the automation of the entire process from data processing to Failure Mode and Effect Summary (FMES) generation, improves the efficiency of airborne system safety analysis, and reduces the complexity and error rate of manual intervention. BRIEF DESCRIPTION OF THE DRAWINGS

[0120] Figure 1 This is a control block diagram of the FMES data generation method for airborne system safety analysis of the present invention. DETAILED DESCRIPTION

[0121] The exemplary embodiments, features, and aspects of the present invention will be described in detail below with reference to the accompanying drawings. The same reference numerals in the accompanying drawings represent elements with the same or similar functions. Although various aspects of the embodiments are shown in the accompanying drawings, the drawings are not necessarily drawn to scale unless otherwise indicated.

[0122] The present invention provides a FMES data generation method for airborne system safety analysis, such as Figure 1 As shown, the FMEA data of the airborne system is obtained, data processing is performed, and data features are extracted using the BERT model; the FEMA structured data feature vector is measured using sinusoidal similarity to calculate the similarity of the FEMA structured data features; the FEMA structured data feature similarity matrix is ​​obtained and the first agglomerative hierarchical clustering analysis is performed; based on the fault tree analysis FTA, a second clustering analysis is performed on the tree cluster set of the FEMA structured data features; based on the second clustering results, the structured failure mode and effect summary FMES data is generated; which includes:

[0123] Step S1: Obtain FMEA data of the airborne system, perform data processing, and use the BERT model to extract data features.

[0124] Step S11: Extract FMEA data from the airborne system. FMEA data is the failure mode and effects analysis data of the airborne system. Extracting FMEA data from the airborne system ensures the integrity and accuracy of the data. The regulations involve the following subsystems and their FMEA data:

[0125] Flight control system: FMEA data focuses on flight control, autopilot, hydraulic system, etc. to ensure the maneuverability and safety of the aircraft in various situations.

[0126] Power system: FMEA data covers failure mode and effects analysis of components such as engines, fuel systems, and turbojets to ensure the reliability and performance of the power system.

[0127] Electrical and electronic systems: FMEA data focuses on the aircraft's power system, avionics equipment, communication system, etc. to ensure the normal operation and fault handling capabilities of the electrical and electronic systems.

[0128] Air pressure system and air conditioning system: FMEA data focuses on the cabin air pressure system to ensure the air quality and comfort in the cabin, while handling corresponding fault conditions.

[0129] Landing gear and braking system: FMEA data is used to conduct failure mode and effect analysis on the landing gear system, braking system, etc. to ensure the safety of the aircraft's landing and ground operation.

[0130] Fuel system: FMEA data is used to conduct failure mode and effect analysis on the fuel supply system, fuel transmission pipeline, etc. to ensure the safety and reliability of the fuel system.

[0131] Anti-icing system: FMEA data focuses on the failure modes and effects of the anti-icing system to ensure flight safety in adverse weather conditions.

[0132] Communication and Navigation Systems: FMEA data involves the aircraft’s communication equipment, navigation equipment, etc. to ensure that pilots can communicate and navigate accurately.

[0133] Perform data conversion and preprocessing to obtain FEMA structured data set Specifically:

[0134] Step S111: Convert the extracted FMEA data into structured data and build a structured data set, specifically:

[0135] fd=[fm,fr,fe];

[0136]

[0137] Among them, fd is FMEA structured data; fm is the failure mode data of FMEA data; fr is the failure cause data of FMEA data; fe is the failure impact data of FMEA data; It is a structured dataset of FEMA.

[0138] According to GBT7826, the failure impact data fe of FMEA data is composed of the local impact data e1 and the final impact data e2, specifically:

[0139] fe=e1⊕e2;

[0140]

[0141] Among them, fe is the failure impact data of FMEA data; e1 is the local impact data; e2 is the final impact data; ⊕ is the semi-additive combination; This is the FEMA structural failure impact dataset.

[0142] Step S112: Using overall statistics to identify FEMA structured data sets The missing values ​​in the data are deleted or filled.

[0143] Identify missing values: Use population statistics to identify missing values ​​in FMEA data. Common representations of missing values ​​include NaN, NULL, or blank values. Remove missing values: If the amount of missing data is small, consider simply deleting the rows or columns containing missing values. Fill missing values: For cases where numeric data is missing in the rare FMEA (Failure Mode and Effects Analysis) data, use the mode to fill missing values.

[0144] Step S12: Use the BERT model to extract the FEMA structured dataset in step S11 The data feature vector v i , specifically:

[0145] Step S121: FEMA structured dataset Perform data segmentation. The input of the BERT model is a vector composed of three parts: token embeddings, position embeddings, and segment embeddings.

[0146] Step S122: Determine the input word embedding vector of the BERT model as:

[0147] E i =T i +S i +P i ;

[0148] Among them, E i is the i-th word embedding vector; T i is the i-th word vector; P i is the embedding vector for the i-th position; S i is the embedding vector of the i-th segment.

[0149] Step S123: Stacking Transformer layers to build the BERT model. Transformer is a fixed processing method whose core structure is based on a deep learning model. Specifically, Transformer includes the following steps: multi-head self-attention calculation of the input data, then processing it through a feed-forward neural network, and finally outputting the result. The calculation method of each Transformer layer is:

[0150] H (l) =Transform(H (l-1) );

[0151] Among them, H (l)is the output vector of the lth Transformer layer, H (0) is the word embedding vector E i ; Transformer is a data processing method based on deep learning models; l is the layer number of the BERT model.

[0152] Step S124: Use the average pooling method to represent the data vector and obtain the FEMA structured data feature vector:

[0153]

[0154] Among them, v i is the FEMA structured data feature vector; n is the number of tokens in the text.

[0155] Step S2: Use the sine similarity to measure the FEMA structured data feature vector v in step S1 i Perform linear normalization to construct the FEMA structured data feature similarity matrix S′ of the airborne system ij , specifically:

[0156] Step S21: Obtain the FEMA structured data feature v obtained in step S1 i , construct the FEMA structured data feature vector group V as:

[0157] V=[v1,v2,…v i ,…v N ] T ;

[0158] Where V is the FEMA structured data feature vector; is the i-th FEMA structured data feature vector; N is the dimension of the FEMA structured data feature vector.

[0159] Step S22: Calculate the similarity S of the FEMA structured data features using the sinusoidal similarity measurement method. ij for:

[0160]

[0161] Among them, S ij is the similarity of FEMA structured data features; sin(v i ,v j ) is the data feature vector v i and data feature vector v j The sinusoidal similarity of v i ·v j is the dot product of the data feature vector; ∥∥ is the Euclidean norm of the vector; v i is the i-th data feature vector; vj is the jth data feature vector.

[0162] Step S23: Construct an N×N dimensional FEMA structured data feature similarity matrix S and perform linear normalization to obtain:

[0163]

[0164] Among them, S′ ij is the normalized result of the similarity of FEMA structured data features; S is the N×N dimensional FEMA structured data feature similarity matrix.

[0165] Step S3: Obtain the FEMA structured data feature similarity matrix in step S2 and perform the first agglomerative hierarchical clustering analysis; each FEMA structured data feature similarity matrix starts as a separate cluster, and gradually merges adjacent clusters until they are merged into one cluster, specifically:

[0166] Step S31: Initialize the FEMA structured data feature similarity matrix, let D ij =S′ ij =d(v i ,v j ),i≠j.

[0167] Step S32: Merge the closest data feature clusters to determine the two data feature clusters with the smallest distance C i ,C j , perform the merge operation, specifically:

[0168]

[0169] Among them, C i is the i-th data feature cluster; C j is the jth data feature cluster; To find two data feature clusters C a and C b Minimize the distance function; C a is the ath data feature cluster; C b is the bth data feature cluster; a is the first number of the data feature cluster; b is the second number of the data feature cluster.

[0170] Step S33: Update the distance between data feature clusters, and set the data feature cluster C i ,C j Merge into data feature cluster C k =C i ∪C j , and update the distance between data feature clusters, and use the average value of the distance between sample pairs within the data feature cluster to calculate the distance between data feature clusters:

[0171]

[0172] Among them, d(C a ,C b ) is the data feature cluster C a and C b distance between them; d(x,y) is the data feature cluster C a The element distance in; x is the data feature cluster C a The elements in; y is the data feature cluster C b Elements in .

[0173] Step S34: Repeat the iteration to update the distance until a tree cluster is finally formed:

[0174]

[0175] in, It is a tree-shaped cluster; is a set of node clusters; is the set of edges between nodes.

[0176] Step S35: Tree cluster leaf nodes Taking each FEMA structured data feature similarity matrix as an independent cluster, by setting the interception height in the tree cluster, the first clustering result of the FEMA structured data feature tree cluster set is obtained as follows:

[0177]

[0178] in, It is a tree cluster set of FEMA structured data features; because but

[0179] Step S4: Based on the fault tree analysis FTA, the tree clustering of the FEMA structured data features in step S3 is performed. Conduct secondary cluster analysis.

[0180] Step S41: construct a logical expression for a fault tree analysis (FTA); use a logical expression based on logic gates and bottom events to represent a fault tree; the logical expression for the fault tree analysis (FTA) specifically includes:

[0181] The AND gate logic expression states that all input events must occur simultaneously for a system failure to occur. For example, a system failure may require multiple subsystem failures to occur simultaneously (such as two engines failing at the same time). The specific expression is:

[0182] F1=A·B;

[0183] Among them, F1 is the output result of the AND gate; A is the first event of airborne system safety; B is the second event of airborne system safety.

[0184] The OR gate logic expression states that a system failure will occur as long as any one of the input events occurs. For example, a device failure may be caused by multiple reasons, and the device will fail as long as one of the reasons occurs. The specific expression is:

[0185] F2=A+B;

[0186] Among them, F2 is the output result of the OR gate.

[0187] The NOT gate logic expression inverts the input event. If the input is true, the output is false. It is often used to indicate that a certain condition "does not occur." For example, a security system is only effective when "a door is not opened." The specific expression is:

[0188]

[0189] Among them, F3 is the output result of the NOT gate.

[0190] The XOR gate logic expression states that a system failure will occur if only one of the input events occurs. Event A or Event B must occur. If both events occur, output Event F will not occur. This is a relatively rare situation, but it may be used to represent a failure mode where two events are mutually exclusive. For example, if one of the two backup systems fails, both backups should not fail at the same time. The specific expression is:

[0191] F4=A⊕B;

[0192] Among them, F4 is the output result of the XOR gate.

[0193] The NAND gate is the inverse of the AND gate. The output is true only when all input events do not occur. If any of them does not occur, event F will occur. The NAND gate is used to combine multiple conditions in reverse order. For example, the system will be safe only when any of the multiple fault conditions does not occur. The specific expression is:

[0194]

[0195] Among them, F5 is the output result of the NAND gate.

[0196] The logical expression of the NOR gate is the reverse of the OR gate. The system failure occurs only when all input events do not occur. This situation is relatively rare, but it can be used to represent the normal working mode of the system under certain specific conditions. The specific expression is:

[0197]

[0198] Among them, F6 is the output result of the NOR gate.

[0199] The logic expression of the voting gate is that the output event occurs when a certain number of events occur among all possible input events. The specific expression is:

[0200] F7=AtLeast(k,{E1,E2,…,E n});

[0201] Among them, F7 is the output result of the voting gate; k is the minimum number of events that need to occur; E is the bottom event; AtLeast is the number of events among all bottom events {E1, E2, ..., E n} at least k occur.

[0202] In summary, a fault tree can be represented using a logical formula based on logic gates and bottom-level events. For example: F: Power failure of an onboard subsystem; E1: Battery exhaustion; E2: External power interruption; E3: Main processor failure; E4: Backup battery failure; E5: Overtemperature; The Fault Tree Analysis (FTA) expression is: F = ((E1⊕E2)·E4)+E3+E5).

[0203] Step S42: extract the event relationship of FEMA structured data, and iteratively calculate the minimum cut set F based on the logical expression in step S41, specifically:

[0204] For the AND gate, the top event occurs if and only if all input events occur, the top event G = E1·E2, then the AND gate cut set C1 = {E1, E2}.

[0205] For the OR gate, the top event occurs when any one of the input events occurs, the top event G = E1 + E2, then the OR gate cut set C2 = {E1} or {E2}.

[0206] For the NOT gate, the top event occurs when the input event does not occur, the top event Then the non-gate cut set C3 = {E1}.

[0207] For the XOR gate, the top event occurs when exactly one of the input events occurs, the top event G = E1⊕E2, then the XOR gate cut set C4 = {E1} or {E2}.

[0208] The NAND gate is the reverse operation of the AND gate. The top event occurs when all the input events are different. Then the NAND gate cut set C5 = {E1, E2}.

[0209] The NOR gate is the reverse operation of the OR gate. The top event occurs when none of the input events occur. Then the NOR gate cut set C6 = {E1, E2}.

[0210] The voting gate indicates that from multiple input events, a certain number of events must occur before the top event occurs. The cut set is a combination of k events among n events, and the voting gate cut set is C7 = {E1, E2}, {E1, E3}, {E2, E3}.

[0211] Through the above cut set event calculation method, iterative calculation is performed to obtain the minimum cut set F of the fault tree of the fault tree analysis FTA:

[0212]

[0213] Among them, F is the minimum cut set; C i ={E k …} is the minimum cut set consisting of the bottom events; m is the total number of cut sets consisting of the bottom events.

[0214] Step S43: Determine whether the secondary clustering condition is met based on the minimum cut set F in step S42, specifically:

[0215] Get the first clustering result obtained in step S3 in The failure mode of fd1 is represented as fd1.fm; for the cluster Clu i ,Clu j , the relationship between the minimum cut set F is:

[0216]

[0217]

[0218] Among them, fd i .fm is FMEA structured data fd i Failure mode; fd j .fm is FMEA structured data fd j failure mode.

[0219] Then we can determine the cluster Clu i ,Clu j If the secondary clustering condition is met, secondary clustering can be performed. Perform secondary clustering and obtain the second clustering results in

[0220] Step S5: Based on the second clustering result of step S4, generate structured failure mode and effect summary FMES data.

[0221] Automatically generate FMES information for each cluster, including FMES failure mode, failure probability and potential failure cause; generate structured FMES data, the second clustering result Clus i Composed of FMEA structured data fd data, Clus i ={fd1,fd2,…,fd i}, one cluster corresponds to a failure mode and effect summary FMES data, the structure is fds = [fms, frs, pfc], where:

[0222] fms=fd1.fe;

[0223]

[0224] Among them, fms is the first component of the failure mode and effects summary FMES data; frs is the second component of the failure mode and effects summary FMES data; pfc is the third component of the failure mode and effects summary FMES data.

[0225] Get the final failure mode and effect summary FMES data FMES={fds1,fds2,…,fds i}, used to support airborne system safety analysis.

[0226] The second aspect of the present invention proposes an FMES data generation system for an FMES data generation method for airborne system safety analysis, which includes: an FMEA data preprocessing module, an FMEA data feature engineering module, a failure impact similarity matrix construction module, a first agglomerative hierarchical clustering module, a secondary clustering module based on FTA information, and a structured FMES data generation module.

[0227] The FMEA data preprocessing module is used to process the original FMEA data; the FMEA data feature engineering module is used to convert the FMEA data into vector data that retains data features and can be calculated, and extract data features; the failure effect similarity matrix construction module is used to construct the FEMA structured data feature similarity matrix; the first agglomerative hierarchical clustering module is used to perform the first clustering analysis of FMEA data; the secondary clustering module based on FTA information performs a second clustering analysis on the first clustering result data to mine the fused fault tree analysis FTA logical information; the structured FMES data generation module automatically generates the structured failure mode and effect summary FMES data required for airborne system safety analysis based on the results of the second clustering analysis.

[0228] The beneficial effects of the present invention are as follows: the present invention provides a FMES data generation method for airborne system safety analysis, performs standardized and vectorized data processing and feature extraction on Failure Mode and Effects Analysis (FMEA) data, completes data standardization and vectorization, combines a two-layer clustering framework of Fault Tree Analysis (FTA) information, performs two cluster analyses, integrates clustering results with Failure Mode and Effects Analysis (FMEA) data, automatically generates structured Failure Mode and Effects Summary (FMES) data and calculates failure probability; realizes full process automation from data processing to Failure Mode and Effects Summary (FMES) generation, improves the efficiency of airborne system safety analysis, and reduces the complexity and error rate of manual intervention.

[0229] The embodiments described above are merely descriptions of preferred implementations of the present invention and are not intended to limit the scope of the present invention. Without departing from the spirit of the present invention, various modifications and improvements made to the technical solutions of the present invention by ordinary technicians in this field should fall within the scope of protection determined by the claims of the present invention.

Claims

1. A method for generating FMES data for airborne system safety analysis, characterized by: It includes: S1: Obtain FMEA data of the airborne system, perform data processing, and use the BERT model to extract data features; S11: Extract FMEA data from airborne systems, perform data conversion and preprocessing, and obtain FEMA structured datasets The FMEA data is the failure mode and effects analysis data of the airborne system; Extract FMEA data from airborne systems, perform data conversion and preprocessing to obtain FEMA structured data sets Specifically: S111: Convert the extracted FMEA data into structured data and build a structured data set, specifically: fd=[fm,fr,fe] Among them, fd is FMEA structured data; fm is the failure mode data of FMEA data; fr is the failure cause data of FMEA data; fe is the failure impact data of FMEA data; It is a structured dataset for FEMA; The failure impact data fe of the FMEA data is formed by connecting the local impact data e1 and the final impact data e2, specifically: Among them, fe is the failure impact data of FMEA data; e1 is the local impact data; e2 is the final impact data; is a semi-additive combination; It is the FEMA structural failure impact dataset; S112: Using population statistics to identify FEMA structured datasets The missing values ​​in the , and take deletion or filling operations on the missing values; S12: Use the BERT model to extract the FEMA structured dataset in step S11 The data feature vector v i ; S2: Use the sinusoidal similarity to measure the FEMA structured data feature vector v in step S1 i , calculate the similarity of FEMA structured data features as: Among them, S ij is the similarity of FEMA structured data features; sin(v i ,v j ) is the data feature vector v i and data feature vector v j The sinusoidal similarity of v i ·v j is the dot product of the data feature vector; |||| is the Euclidean norm of the vector; v i is the i-th data feature vector; v j is the jth data feature vector; Perform linear normalization to construct the FEMA structured data feature similarity matrix S′ of the airborne system ij ; S3: Get the FEMA structured data feature similarity matrix in step S2 and perform the first agglomerative hierarchical clustering analysis; each FEMA structured data feature similarity matrix starts as a separate cluster, and gradually merges adjacent clusters until they are merged into one cluster, obtaining a tree cluster set of FEMA structured data features in, i=1,2,…,k, since but S4: Based on the fault tree analysis FTA, the tree clustering of the FEMA structured data features in step S3 is performed Conduct secondary cluster analysis; S41: Construct a logical expression for fault tree analysis (FTA); use a logical expression based on logic gates and bottom events to represent a fault tree; S42: extracting the event relationship of the FEMA structured data, and iteratively calculating the minimum cut set F based on the logical expression in step S41; S43: Determine whether the secondary clustering condition is met based on the minimum cut set F in step S42, and perform the first clustering result in step S3 based on the association relationship. Perform secondary clustering and obtain the second clustering results in S5: Based on the second clustering result of step S4, generate structured failure mode and effect summary FMES data; According to the second clustering results, each cluster automatically generates structured failure mode and effect summary FMES data, FMES = {fds1, fds2, ..., fds i }, used for airborne system safety analysis.

2. The FMES data generation method for airborne system safety analysis according to claim 1, characterized in that: In step S12, the BERT model is used to extract the FEMA structured dataset in step S11. The data feature vector v i , specifically: S121: FEMA structured dataset Perform data segmentation operations; S122: Determine the input word embedding vector of the BERT model: E i =T i +S i +P i Among them, E i is the i-th word embedding vector; T i is the i-th word vector; P i is the embedding vector of the i-th position; S i is the i-th segment embedding vector; S123: Stacking Transformer layers to build a BERT model; the calculation method for each Transformer layer is: H (l) =Transform(H (l-1) ) Among them, H (l) is the output vector of the lth Transformer layer, H (0) is the word embedding vector E i ; Transformer is a data processing method based on deep learning models; l is the layer number of the BERT model; S124: Use the average pooling method to represent the data vector and obtain the FEMA structured data feature vector: Among them, v i is the FEMA structured data feature vector; n is the number of tokens in the text.

3. The FMES data generation method for airborne system safety analysis according to claim 1, characterized in that: In step S2, linear normalization is performed to construct the FEMA structured data feature similarity matrix S′ of the airborne system. ij , specifically: S21: Get the FEMA structured data feature v obtained in step S1 i , construct the FEMA structured data feature vector group V as: V=[v1,v2,…v i ,…v N ] T Where V is the FEMA structured data feature vector; is the i-th FEMA structured data feature vector; N is the dimension of the FEMA structured data feature vector; S22: Use the sinusoidal similarity measurement method to calculate the similarity S of FEMA structured data features ij ; S23: Construct an N×N dimensional FEMA structured data feature similarity matrix S and perform linear normalization to obtain: Among them, S′ ij is the normalized result of the similarity of FEMA structured data features; S is the N×N dimensional FEMA structured data feature similarity matrix.

4. The FMES data generation method for airborne system safety analysis according to claim 1, characterized in that: In step S3, each FEMA structured data feature similarity matrix starts as a separate cluster, and gradually merges adjacent clusters until it is merged into one cluster, specifically: S31: Initialize the FEMA structured data feature similarity matrix, let D ij =S′ ij =d(v i ,v j ),i≠j; S32: Merge the closest data feature clusters to determine the two data feature clusters with the smallest distance C i ,C j , perform the merge operation, specifically: Among them, C i is the i-th data feature cluster; C j is the jth data feature cluster; To find two data feature clusters C a and C b Minimize the distance function; C a is the ath data feature cluster; C b is the bth data feature cluster; a is the first number of the data feature cluster; b is the second number of the data feature cluster; S33: Update the distance between data feature clusters, and set the data feature cluster C i ,C j Merge into data feature cluster C k =C i ∪C j , and update the distance between data feature clusters, and use the average value of the distance between sample pairs within the data feature cluster to calculate the distance between data feature clusters: Among them, d(C a ,C b ) is the data feature cluster C a and C b distance between them; d(x,y) is the data feature cluster C a The element distance in; x is the data feature cluster C a The elements in; y is the data feature cluster C b Elements in S34: Repeat the iteration to update the distance until a tree cluster is formed: in, It is a tree-shaped cluster; is the set of node clusters; ε is the set of edges between nodes; S35: Tree cluster leaf node Taking each FEMA structured data feature similarity matrix as an independent cluster, by setting the interception height in the tree cluster, the first clustering result of the FEMA structured data feature tree cluster set is obtained as follows: in, It is a tree cluster set of FEMA structured data features; i=1,2,…,k, since but 5. The FMES data generation method for airborne system safety analysis according to claim 1, characterized in that: The logical expression for constructing the fault tree analysis FTA in step S41 specifically includes: an AND gate logical expression, an OR gate logical expression, a NOT gate logical expression, an XOR gate logical expression, a NAND gate logical expression, a NOR gate logical expression, and a voting gate logical expression; The AND gate logic expression states that all input events must occur simultaneously for a system failure to occur. The specific expression is: F1=A·B Among them, F1 is the output result of the AND gate; A is the first event of airborne system safety; B is the second event of airborne system safety; The OR gate logic expression indicates that a system failure will occur as long as any one of the input events occurs. The specific expression is: F2=A+B Among them, F2 is the output result of the OR gate; The NOT gate logic expression inverts the input event. If the input is false, the output is true. The specific expression is: Among them, F3 is the output result of the NOT gate; The XOR gate logic expression has only one input event, and the system failure will occur. Event A or Event B must occur. The specific expression is: Among them, F4 is the output result of the XOR gate; The logic expression of the NAND gate is the inverse of the gate. The output is true only when all input events do not occur. The specific expression is: Among them, F5 is the output result of the NAND gate; The logic expression of the NOR gate is the reverse of the OR gate. When all input events do not occur, the system failure occurs. The specific expression is: Among them, F6 is the output result of the NOR gate; The voting gate logic expression is that when a certain number of events occur among all the input events, the output event occurs. The specific expression is: F7=AtLeast(k,{E1,E2,…,E n }) Among them, F7 is the output result of the voting gate; k is the minimum number of events that need to occur; E is the bottom event; AtLeast is the number of events among all bottom events {E1, E2, ..., E n } at least k occur.

6. The FMES data generation method for airborne system safety analysis according to claim 1, characterized in that: In step S42, the event relationship of the FEMA structured data is extracted, and the minimum cut set F is iteratively calculated based on the logical expression in step S41, specifically: For the AND gate, the top event occurs if and only if all input events occur, the top event G = E1·E2, then the AND gate cut set C1 = {E1, E2}; For the OR gate, the top event occurs when any of the input events occurs, the top event G = E1 + E2, then the OR gate cut set C2 = {E1} or {E2}; For the NOT gate, the top event occurs when the input event does not occur, the top event Then the non-gate cut set C3 = {E1}; For the XOR gate, the top event occurs when exactly one of the input events occurs. Then the XOR gate cut set C4 = {E1} or {E2}; The NAND gate is the reverse operation of the AND gate. The top event occurs when all the input events do not occur at the same time. Then the NAND gate cut set C5={E1,E2}; The NOR gate is the reverse operation of the OR gate. The top event occurs when all the input events do not occur. Then the NOR gate cut set C6 = {E1, E2}; The voting gate indicates that from multiple input events, a certain number of events must occur before the top event occurs. The cut set is the combination of k events among the n events, and the voting gate cut set is C7 = {E1, E2}, {E1, E3}, {E2, E3}; Through the above cut set event calculation method, iterative calculation is performed to obtain the minimum cut set F of the fault tree of the fault tree analysis FTA: Among them, F is the minimum cut set; C i ={E k …} is the minimum cut set consisting of the bottom events; m is the total number of cut sets consisting of the bottom events.

7. The FMES data generation method for airborne system safety analysis according to claim 1, characterized in that: In step S43, the secondary clustering condition is determined based on the minimum cut set F, specifically: Get the first clustering result obtained in step S3 in The failure mode of fd1 is represented as fd1.fm; for the cluster Clu i ,Clu j , the relationship between the minimum cut set F is: Among them, fd i .fm is FMEA structured data fd i Failure mode; fd j .fm is FMEA structured data fd j failure modes; Then we can determine the cluster Clu i ,Clu j The secondary clustering condition is met and secondary clustering can be performed.

8. The FMES data generation method for airborne system safety analysis according to claim 1, characterized in that: The generation of structured failure mode and effects summary (FMES) data in step S5 is as follows: The second clustering result Clus i Composed of FMEA structured data fd data, Clus i ={fd1,fd2,…,fd i }, one cluster corresponds to a failure mode and effect summary FMES data, the structure is fds = [fms, frs, pfc], where: fms=fd1.fe Among them, fms is the first component of the Failure Mode and Effect Summary FMES data; frs is the second component of the Failure Mode and Effect Summary FMES data; pfc is the third component of the Failure Mode and Effect Summary FMES data; Get the final failure mode and effect summary FMES data FMES={fds1,fds2,…,fds i }.

9. A FMES data generation system for airborne system safety analysis, the system being used in the FMES data generation method for airborne system safety analysis according to claim 1, characterized in that: The system includes: an FMEA data preprocessing module, an FMEA data feature engineering module, a failure effect similarity matrix construction module, a first agglomerative hierarchical clustering module, a second clustering module based on FTA information, and a structured FMES data generation module; The FMEA data preprocessing module is used to process the original FMEA data; The FMEA data feature engineering module is used to convert FMEA data into vector data that retains data features and can be calculated, and extract data features; The failure impact similarity matrix construction module is used to construct a FEMA structured data feature similarity matrix; The first agglomerative hierarchical clustering module is used to perform the first cluster analysis of FMEA data; The secondary clustering module based on FTA information performs a second clustering analysis on the first clustering result data in order to mine and fuse the FTA logic information; The structured FMES data generation module automatically generates the structured failure mode and effect summary FMES data required for airborne system safety analysis based on the second cluster analysis results.

Images