Knowledge graph-based aircraft fuel system fault isolation and maintenance decision-making method

Abstract

The invention relates to an aircraft fuel system fault isolation and maintenance decision-making method based on a knowledge graph, and the method comprises the steps: employing an equipment tree analysis method to disassemble an aircraft fuel system into a plurality of maintenance units of different levels, and obtaining system knowledge data; obtaining fault knowledge data and maintenance knowledge data of each maintenance unit; constructing a fault isolation and maintenance knowledge graph according to the obtained system knowledge, fault knowledge and maintenance knowledge; and obtaining a to-be-matched fault early warning result output by the fault early warning system, searching a matched entity in the fault isolation and maintenance knowledge graph, and outputting a corresponding fault isolation and maintenance decision. According to the invention, accurate fault isolation suggestions and maintenance decisions can be quickly generated according to fault information, the maintenance work efficiency can be greatly improved, and accurate maintenance of equipment is realized.

Description

Technical Field

[0001] This invention relates to the field of aircraft fuel system fault detection technology, and in particular to a knowledge graph-based method for aircraft fuel system fault isolation and maintenance decision-making. Background Technology

[0002] As the aircraft's power source, the fuel system is responsible for delivering fuel from the fuel tank to the engine, providing a stable fuel supply to the engine throughout all phases of flight. The fuel system comprises multiple components such as fuel tanks, pumps, valves, and pipelines, making its structure complex and requiring high levels of expertise in fault diagnosis and maintenance.

[0003] Existing knowledge graph-based fault detection methods mostly extract triples from collected text data using large language models. When applied to complex systems such as aircraft fuel systems that contain multiple levels of components, this approach can lead to problems such as entity granularity mismatch, overly coarse relational semantics, and missing fault reasoning logic. Consequently, the generated knowledge graphs are insufficient to meet the precise requirements for fault isolation in aircraft fuel systems. Summary of the Invention

[0004] The technical problem to be solved by the present invention is to provide a knowledge graph-based method for fault isolation and maintenance decision-making in aircraft fuel systems. This method can quickly generate accurate fault isolation suggestions and maintenance decisions based on fault information, which can significantly improve maintenance efficiency and achieve precise equipment maintenance.

[0005] The technical solution adopted by this invention to solve its technical problem is: to provide a knowledge graph-based method for aircraft fuel system fault isolation and maintenance decision-making, including: The aircraft fuel system was broken down into multiple maintenance units at different levels using the device tree analysis method, resulting in system knowledge data. Acquire fault knowledge data and maintenance knowledge data for each maintenance unit; Construct a fault isolation and maintenance knowledge graph based on the acquired system knowledge, fault knowledge, and maintenance knowledge; Obtain the fault warning results to be matched from the fault warning system, search for matching entities in the fault isolation and maintenance knowledge graph, and output the corresponding fault isolation and maintenance decisions.

[0006] Furthermore, the system knowledge data includes equipment structural relationships, signal measurement point relationships, and equipment functional relationships.

[0007] Furthermore, the fault knowledge data includes fault mode information, fault causal relationships, and fault correlation relationships, which are obtained by analyzing the collected aircraft fuel system equipment fault data using FMEA analysis, FTA analysis, and fault correlation analysis methods.

[0008] Furthermore, the maintenance knowledge data includes maintenance task analysis and maintenance cases, which are obtained by statistically analyzing the maintenance knowledge required for aircraft fuel system fault isolation by combining the equipment design knowledge and system maintenance knowledge of aircraft fuel system equipment manufacturers.

[0009] Furthermore, the search matching entity includes: The first entity that completely matches the fault warning result to be matched is output as the matching entity; If the first entity does not exist, the second entity with the highest similarity to the fault warning result to be matched is searched and output as the matching entity.

[0010] Furthermore, the search matching entity includes: The first entity that completely matches the fault warning result to be matched is output as the matching entity; If the first entity does not exist, the second entity with the highest similarity to the fault warning result to be matched is searched and output as the matching entity.

[0011] Furthermore, the second entity with the highest similarity is obtained using the longest common subsequence similarity matching method.

[0012] Furthermore, the second entity with the highest similarity to the fault warning result to be matched includes: Traverse the entities in the fault isolation and maintenance knowledge graph; Extract the longest common subsequence between the fault warning result to be matched and the current entity, and then calculate the similarity. The entity with the highest similarity to the fault warning result to be matched is output as the second entity.

[0013] Furthermore, the extraction of the longest common subsequence between the fault warning result to be matched and the current entity includes: Obtain the string length m of the fault warning result to be matched and the string length n of the current entity, respectively; Build size is The matching matrix, and based on the characters at corresponding positions of the two strings. and The similarities and differences are used to calculate the element values ​​of the matching matrix. ,in, ; To match the element in the bottom right corner of the matrix Starting from the beginning, backtrack to obtain all elements along the backtracking path before their values ​​changed, and arrange the obtained elements in order to obtain the longest common subsequence.

[0014] Furthermore, the element values ​​of the matching matrix The following formula is used to calculate:

[0015] in, For the matching matrix of the th Line number The element values ​​of the column, The first fault warning result to be matched Each element value For the current entity's first Each element value.

[0016] Furthermore, it also includes: Obtaining user questions; Use a Bayesian classifier to classify user questions; The classification results are used as question-and-answer keywords. Entities matching the question-and-answer keywords are searched from the fault isolation and maintenance knowledge graph, and the answers to the questions are output.

[0017] Furthermore, based on the obtained system knowledge data, fault knowledge data, and maintenance knowledge data, a fault isolation and maintenance knowledge graph is constructed, including: Define the ontology of the knowledge graph and determine the entity types and relation types; Entities, relationships, and attributes are extracted from system knowledge data, fault knowledge data, and maintenance knowledge data using a knowledge extraction model, which is built based on BiLSTM, MHA, and CRF models. Construct a fault isolation and maintenance knowledge graph based on the extracted entities, relationships, and attributes. Beneficial effects

[0018] By employing the above-mentioned technical solutions, this invention has the following advantages and positive effects compared with existing technologies: This invention disassembles the aircraft fuel system using equipment tree analysis technology, then extracts system knowledge, fault knowledge, and maintenance knowledge to construct a fault isolation and maintenance knowledge graph. This enables the rapid generation of accurate fault isolation suggestions and maintenance decisions based on fault information, significantly improving maintenance efficiency and achieving precise equipment maintenance. Furthermore, by combining equipment tree analysis technology, fault impact and hazard analysis technology, fault tree analysis technology, and fault correlation analysis technology to extract system knowledge, fault knowledge, and maintenance knowledge, this invention ensures that the generated knowledge graph meets the precise requirements for fault isolation in aircraft fuel systems. Finally, this invention extracts question-and-answer keywords using a Bayesian classifier and combines this with a similarity algorithm to quickly and accurately match answers from the fault isolation and maintenance knowledge graph. This allows maintenance personnel to quickly retrieve the necessary maintenance knowledge during on-site fault diagnosis, significantly improving maintenance efficiency and achieving precise equipment maintenance. Attached Figure Description

[0019] Figure 1This is a flowchart of an embodiment of the present invention; Figure 2 This is a flowchart illustrating an embodiment of the present invention; Figure 3 This is a diagram showing the results of aircraft fuel system fault knowledge according to an embodiment of the present invention; Figure 4 This is a diagram of the BiLSTM+MHA+CRF model according to an embodiment of the present invention; Figure 5 This is a knowledge graph result diagram of fault isolation and maintenance according to an embodiment of the present invention. Detailed Implementation

[0020] The present invention will be further illustrated below with reference to specific embodiments. It should be understood that these embodiments are for illustrative purposes only and are not intended to limit the scope of the invention. Furthermore, it should be understood that after reading the teachings of this invention, those skilled in the art can make various alterations or modifications to the invention, and these equivalent forms also fall within the scope defined by the appended claims.

[0021] The embodiments of the present invention relate to a knowledge graph-based method for aircraft fuel system fault isolation and maintenance decision-making, such as... Figure 1 As shown, it includes: The aircraft fuel system was broken down into multiple maintenance units at different levels using the device tree analysis method, resulting in system knowledge data. Acquire fault knowledge data and maintenance knowledge data for each maintenance unit; Construct a fault isolation and maintenance knowledge graph based on the acquired system knowledge, fault knowledge, and maintenance knowledge; Obtain the fault warning results to be matched from the fault warning system, search for matching entities in the fault isolation and maintenance knowledge graph, and output the corresponding fault isolation and maintenance decisions.

[0022] System knowledge, fault knowledge, and maintenance knowledge, as the three sources of knowledge for aircraft fuel systems, can include the following components: System knowledge includes equipment structural relationships, signal measurement point relationships, and equipment functional relationships; Fault knowledge includes fault mode information, fault causal relationships, and fault correlation relationships, which are obtained by analyzing collected aircraft fuel system equipment fault data using FMEA analysis, FTA analysis, and fault correlation analysis methods; Maintenance knowledge includes maintenance task analysis and maintenance cases, which are obtained by statistically analyzing the maintenance knowledge required for aircraft fuel system fault isolation by combining the equipment design knowledge and system maintenance knowledge of aircraft fuel system equipment manufacturers.

[0023] More specifically, by combining equipment tree analysis technology, fault impact and hazard analysis technology, fault tree analysis technology, and fault correlation analysis technology, the three sources of knowledge about the aircraft fuel system can be extracted from knowledge data such as aircraft fuel system design data and maintenance manuals.

[0024] The following methods can be used to construct knowledge graphs: Define the ontology of the knowledge graph and determine the entity types and relation types; Entities, relationships, and attributes are extracted from system knowledge data, fault knowledge data, and maintenance knowledge data using a knowledge extraction model, which is built based on BiLSTM, MHA, and CRF models. Construct a fault isolation and maintenance knowledge graph based on the extracted entities, relationships, and attributes.

[0025] To address the issue of incomplete correspondence between the input fault warning results and entity names in the knowledge graph, a similarity-based intelligent search can be employed. When no perfectly matching entity is found, the output entities with high similarity to the input entity names are used, employing the longest common subsequence similarity method. A subsequence is a new sequence obtained by deleting several elements from a sequence. A common subsequence of two sequences is a sequence that is simultaneously a subsequence of both sequences. The longest common subsequence is the longest common subsequence among all possible subsequences of the two sequences. The longest common subsequence is the longest set of subsequences that are identical among all possible subsequences obtained by deleting several characters from each of the two sequences. The process of solving for the longest common subsequence between the strings "BDCABA" and "ABCBDAB" using dynamic programming is as follows: 1) Generate a 7×8 matrix based on the string length. 1) Calculate the value at each position in the matrix, and assign a value to each position in the matrix based on the similarity or difference of the corresponding characters at positions i and j of the two strings; 2) The length of the longest common subsequence between the two strings is the value at the bottom right corner of the matrix; 3) The length of the longest common subsequence between the two strings is the value at the bottom right corner of the matrix; 4) Start backtracking from the bottom right corner of the matrix to obtain a backtracking path; 5) Select the matrix positions before all values ​​on the backtracking path changed, and arrange the characters corresponding to these positions in order to obtain the longest common subsequence "BCBA" of the two strings.

[0026] like Figure 2 The image shown is a preferred embodiment 1 of this implementation.

[0027] This embodiment includes the following steps: Step 1: Combining equipment tree analysis technology, fault impact and hazard analysis technology, fault tree analysis technology, and fault correlation analysis technology, extract the three sources of knowledge (system knowledge, fault knowledge, and maintenance knowledge) of the aircraft fuel system from knowledge data such as aircraft fuel system design data and maintenance manuals.

[0028] The system knowledge statistical analysis is primarily based on the device tree analysis method. Device tree analysis is a step-by-step analysis process from the whole to the parts, targeting intelligent fault isolation and maintenance decision-making. It involves understanding the structure and function of the diagnostic object, and combining past fault cases and knowledge deduction to summarize the influence between devices during a fault. Because aircraft fuel systems exhibit holistic emergent and interconnected characteristics, the functions and structures of each component are mutually coupled, mutually stimulating, and mutually restrictive. Often, some components may appear normal in isolation, but the whole system can fail. Therefore, when conducting system analysis, it is crucial to focus on both detailed equipment breakdown, gradually disassembling the system into subsystems and components to study the mechanisms of fault occurrence and development, and monitoring the status of each subsystem from the part to the whole to study the evolution and coupling of faults. Equipment analysis aims to better uncover the connection between equipment and faults, analyze fault mechanisms, and clarify the physical structure of equipment for more precise fault location. Equipment analysis is not about being as detailed as possible, but rather about breaking down equipment to the smallest maintenance unit to facilitate maintenance work.

[0029] Fault knowledge mainly includes fault mode information, fault causal relationships, and fault correlations. A fault mode is the external manifestation of equipment failure and the direct cause of maintenance activities. Fault modes are forms of failure that can be observed through human senses or measuring instruments. When diagnosing equipment faults, it is first necessary to determine the fault mode through factual data, and then to judge its impact on the system based on the determined fault mode, proposing targeted maintenance measures and solutions. Aircraft fuel systems have numerous equipment types, most of which are complex, with many components and varying structures. Their fault types are also diverse, and fault types can be classified in different ways, such as by the location of the fault, the process of the fault, the equipment structure, etc. These factors contribute to the diversity of equipment anomalies and fault situations, making it difficult to list and analyze equipment fault modes. To address these issues, FMEA analysis, FTA analysis, and fault correlation analysis techniques are used in aircraft fault knowledge analysis. Based on various collected data on aircraft system and equipment faults, a multi-faceted analysis of fault knowledge is conducted. The results of aircraft fuel system fault knowledge are shown in the figure below. Figure 3 As shown.

[0030] Maintenance knowledge mainly includes maintenance task analysis and maintenance cases. It combines the equipment design knowledge and system maintenance knowledge of fuel system equipment manufacturers to conduct statistical analysis on the maintenance knowledge required for fuel system fault isolation.

[0031] Step 2: Using the three sources of knowledge about the aircraft fuel system as input, a knowledge graph for intelligent fault isolation and maintenance of a typical system is constructed using a Bi-directional Long Short-Term Memory (Bi-LSTM) network model, a Multi-Headed Attention (MHA) mechanism, and a Conditional Random Field (CRF) model. Combining the system / fault / maintenance knowledge obtained from the knowledge statistical analysis in the previous step, an intelligent fault isolation and maintenance knowledge graph is constructed. First, an ontology is constructed based on the three sources of knowledge to determine entity types and relation types; then, research is conducted on intelligent fault isolation and maintenance entity extraction and relation extraction techniques; finally, knowledge fusion and updates are performed to construct the knowledge graph.

[0032] More specifically, an ontology model is constructed based on the three sources of knowledge, defining entity categories, attributes, and relationships, and the designed ontology model is implemented into a machine-readable format (Web Ontology Language (OWL)) using ontology editing tools (such as Protege).

[0033] like Figure 4 As shown, the BiLSTM+MHA+CRF model is applied for entity extraction and relation extraction. The BiLSTM+MHA+CRF model consists of three layers: an input layer, a BiLSTM layer, an MHA layer, and a labeling layer. The input layer is mainly responsible for vectorizing the words in the window. It uses a trained word vector matrix to map each word in the input window into a distributed vector. To prevent overfitting, a dropout parameter is set. The BiLSTM layer automatically obtains sentence features. The vector sequence output by the input layer serves as the initial input value for the BiLSTM network at each time point. In the BiLSTM network layer, the sequential order of the input vector sequence serves as the input to the forward LSTM layer, and the reverse order serves as the input to the backward LSTM layer. The model concatenates the hidden state sequences output by the forward and backward LSTMs according to their positions. To automatically extract features, a linear transformation layer is added to the hidden layer to map the original hidden state sequence from m dimensions to k dimensions, where k represents the total number of labels in the label set. The MHA layer is used to obtain global features of the state sequence and the association strength between words; the CRF layer, also called the logistic regression layer, is used to label the sentence sequence. The CRF layer uses previously labeled information to label a new position. Finally, the softmax function is used for normalization to obtain the word labels.

[0034] The specific algorithm is as follows: Inputs: Model parameters, loss function, training samples, training samples Output: 1) For each epoch cycle 2) For each batch loop 3) Function forward 4) BiLSTM forward training 5) MHA forward training 6) CRF forward training, calculating the loss function. 7) End Function 8) Function backward 9) Parameter update 10) CRF backpropagation and parameter update 11) MHA backpropagation and parameter update 12) BiLSTM backpropagation and parameter update 13) End function 14) End the batch loop 15) End the epoch loop 16) Return like Figure 5 As shown, the Neo4j graph database is used to store and visualize the fault isolation and maintenance knowledge graph. The knowledge fusion update aligns, associates and merges the newly extracted knowledge elements with the original knowledge graph, organically integrating them to expand the scope and content of the knowledge graph.

[0035] Step 3: Combining the established knowledge graph, construct an intelligent fault isolation model and a human-computer question-and-answer system based on intelligent search and Naive Bayes classifier.

[0036] More specifically, by combining the established knowledge graph, an intelligent fault isolation system is constructed for intelligent search. Text-based fault warning information is input into the intelligent fault isolation model to obtain fault isolation information. Then, a human-computer question-and-answer system based on a Naive Bayes classifier is constructed. The fault isolation information is input into the question-and-answer system to obtain maintenance decision suggestions.

[0037] To address the issue of incomplete correspondence between the input fault warning entities and the entity names in the knowledge graph, a similarity-based intelligent search is employed. When no perfectly matching entity is found, entities with high similarity to the input entity names are output, and the longest common subsequence similarity method is used. A subsequence is a new sequence obtained by deleting several elements from a sequence. A common subsequence of two sequences is a sequence that is simultaneously a subsequence of both sequences. The longest common subsequence is the longest common subsequence among all possible subsequences of two sequences. The longest common subsequence is the longest set of subsequences that are identical among all possible subsequences obtained by deleting several characters from each of the two sequences. The process of using dynamic programming to find the longest common subsequence between the strings "BDCABA" and "ABCBDAB" is as follows: 1) Generate a 7×8 matrix based on the string length. 1) Calculate the value at each position in the matrix, assigning a value to each position based on the similarity or difference of the corresponding characters at positions i and j of the two strings; 2) The length of the longest common subsequence between the two strings is the value at the bottom right corner of the matrix; 3) Backtrack from the bottom right corner of the matrix to obtain a backtracking path; 4) Select the matrix positions before all values ​​changed on the backtracking path, and arrange the characters corresponding to these positions in order to obtain the longest common subsequence "BCBA". In practical applications, the length of the longest common subsequence can be used as the similarity score; the longer the length, the higher the similarity between the two strings.

[0038] The above process can be represented as: Obtain the string length m of the fault warning result to be matched and the string length n of the current entity, respectively; Build size is The matching matrix, and based on the characters at corresponding positions of the two strings. and The similarities and differences are used to calculate the element values ​​of the matching matrix. ,in, ; To match the element in the bottom right corner of the matrix Starting from the beginning, backtrack to obtain all elements along the backtracking path before their values ​​changed, and arrange the obtained elements in order to obtain the longest common subsequence.

[0039] Matching matrix element values The following formula is used to calculate:

[0040] in, For the matching matrix of the th Line number The element values ​​of the column, The first fault warning result to be matched Each element value For the current entity's first Each element value.

[0041] Step 4: Construct a human-computer question-answering system based on a Naive Bayes classifier. Users input questions into the intelligent search and question-answering system, and the system outputs the answer to the question. The obtained answer is used as supplementary input into the fault isolation information. Intelligent question answering includes three steps: question classification, keyword extraction, and Neo4j query. The key lies in question classification, i.e., what type of result the user needs. The Naive Bayes classifier is applied as the search and question-answering method. The Naive Bayes classifier is the simplest classifier among probabilistic classifiers, possessing a fairly high classification accuracy in many cases, and is known for its high efficiency and good generalization ability. This classifier assumes that the attribute variables are conditionally independent given a class variable. Under the assumption of conditional independence, the Naive Bayes classifier has a simple star-shaped structure. Based on the Naive Bayes classifier structure, joint probabilities can be obtained to classify questions. After classifying questions using a Naive Bayes classifier, keywords for the questions and answers can be extracted through entity matching in the knowledge graph. Then, the Cypher language is used to retrieve the answers from the knowledge graph and output them to the user, thus completing the intelligent question answering process.

[0042] It is understood that those skilled in the art can make equivalent substitutions or modifications to the technical solution and inventive concept of the present invention, and all such substitutions or modifications should fall within the protection scope of the appended claims.

Claims

1. A knowledge graph-based method for fault isolation and maintenance decision-making in aircraft fuel systems, characterized in that, include: The aircraft fuel system is broken down into multiple maintenance units at different levels using the device tree analysis method; Acquire system knowledge data, fault knowledge data, and maintenance knowledge data from each maintenance unit; Construct a fault isolation and maintenance knowledge graph based on the obtained system knowledge data, fault knowledge data, and maintenance knowledge data; Obtain the fault warning results to be matched from the fault warning system, search for matching entities in the fault isolation and maintenance knowledge graph, and output the corresponding fault isolation and maintenance decisions.

2. The method according to claim 1, characterized in that, The system knowledge data includes equipment structural relationships, signal measurement point relationships, and equipment functional relationships.

3. The method according to claim 1, characterized in that, The fault knowledge data includes fault mode information, fault causal relationships, and fault correlation relationships. It is obtained by analyzing the collected aircraft fuel system equipment fault data using FMEA analysis, FTA analysis, and fault correlation analysis methods.

4. The method according to claim 1, characterized in that, The maintenance knowledge data includes maintenance task analysis and maintenance cases, which are obtained by statistically analyzing the maintenance knowledge required for aircraft fuel system fault isolation by combining the equipment design knowledge and system maintenance knowledge of aircraft fuel system equipment manufacturers.

5. The method according to claim 1, characterized in that, The search matching entities include: The first entity that completely matches the fault warning result to be matched is output as the matching entity; If the first entity does not exist, the second entity with the highest similarity to the fault warning result to be matched is searched and output as the matching entity.

6. The method according to claim 5, characterized in that, The second entity with the highest similarity to the fault warning result to be matched includes: Traverse the entities in the fault isolation and maintenance knowledge graph; Extract the longest common subsequence between the fault warning result to be matched and the current entity, and then calculate the similarity. The entity with the highest similarity to the fault warning result to be matched is output as the second entity.

7. The method according to claim 6, characterized in that, The extraction of the longest common subsequence between the fault warning result to be matched and the current entity includes: Obtain the string length m of the fault warning result to be matched and the string length n of the current entity, respectively; Build size is The matching matrix, and based on the characters at corresponding positions of the two strings. and The similarities and differences are used to calculate the element values ​​of the matching matrix. ,in, ; To match the element in the bottom right corner of the matrix Starting from the beginning, backtrack to obtain all elements along the backtracking path before their values ​​changed, and arrange the obtained elements in order to obtain the longest common subsequence.

8. The method according to claim 7, characterized in that, The element values ​​of the matching matrix The following formula is used to calculate: in, For the matching matrix of the th Line number The element values ​​of the column, The first fault warning result to be matched Each element value For the current entity's first Each element value.

9. The method according to claim 1, characterized in that, Also includes: Obtaining user questions; Use a Bayesian classifier to classify user questions; The classification results are used as question-and-answer keywords. Entities matching the question-and-answer keywords are searched from the fault isolation and maintenance knowledge graph, and the answers to the questions are output.

10. The method according to claim 1, characterized in that, Based on the obtained system knowledge data, fault knowledge data, and maintenance knowledge data, a fault isolation and maintenance knowledge graph is constructed, including: Define the ontology of the knowledge graph and determine the entity types and relation types; Entities, relationships, and attributes are extracted from system knowledge data, fault knowledge data, and maintenance knowledge data using a knowledge extraction model, which is built based on BiLSTM, MHA, and CRF models. Construct a fault isolation and maintenance knowledge graph based on the extracted entities, relationships, and attributes.

Images