Fault multi-time diagnosis reasoning method and system based on tensor decomposition

Abstract

The invention discloses a fault multi-diagnosis reasoning method and system based on tensor decomposition, and belongs to the technical field of fault prediction and health management. The method comprises the following steps: step 1, constructing an initial fault knowledge graph according to expert experience; 2, performing vectorization on entities and edges in the initial fault knowledge graph by using tensor decomposition to obtain a fault vector knowledge graph; and step 3, obtaining a to-be-detected fault phenomenon # imgabs0 #, calculating the similarity of all entities in the fault vector knowledge graph of the to-be-detected fault phenomenon, and regarding all entities which are directly connected with the entity # imgabs2 # with the highest similarity with the to-be-detected fault phenomenon # imgabs1 # through edges in the fault vector knowledge graph as a one-hop entity set connected with the to-be-detected fault phenomenon. According to the invention, comprehensive fault reasons for fault phenomena can be diagnosed.

Description

Technical Field

[0001] The invention relates to a fault multiple diagnosis reasoning method and system based on tensor decomposition, belonging to the technical field of fault prediction and health management. Background Art

[0002] The aerospace field has always been a key area of ​​development for each country, and it is also a reflection of a country's scientific and technological innovation capabilities. With the continuous development and improvement of my country's economic strength, the aerospace industry is booming. However, with the continuous improvement of equipment functions and effectiveness, the structure and function of avionics equipment have become more complex, and the difficulty of maintenance and repair of avionics equipment has also increased significantly. Therefore, at this stage, in order to ensure the normal operation of aviation equipment and improve the reliability of aviation equipment, it has become one of the important research topics to be able to locate and solve faults in a timely and accurate manner when faults occur, and to be able to predict faults in a timely manner before they occur.

[0003] Combining faults with knowledge graph technology and creatively constructing fault knowledge graphs is actually using the advantages of artificial intelligence technology to overcome the limitations of traditional fault analysis. This integration is not only an innovation in fault analysis methods, but also an innovative attempt at knowledge management and intelligent decision-making. The fault knowledge graph integrates multiple data sources such as existing fault information, expert knowledge, and equipment information, and presents them visually in the form of a graph. Through the structured method of knowledge graphs, knowledge graphs can assist operators in quickly and accurately retrieving and obtaining relevant information, and can discover potential connections and common features between different fault cases. Because the existing information is incomplete and insufficient, the relationship between two entities may be missing or ignored in the construction of the knowledge graph. Therefore, the use of knowledge reasoning is crucial in the knowledge graph. Through knowledge reasoning, the existing knowledge graph relationship can be enriched, making the process of establishing the fault knowledge graph more intelligent and efficient. However, due to the difficulty and high cost of acquiring knowledge about avionics equipment, some electronic equipment can only acquire a small amount of knowledge. The knowledge graph constructed based on the existing information will always have less information and weaker correlations (i.e., the "edges" shown in the graph are not rich enough). Summary of the invention

[0004] In order to overcome the shortcomings of the prior art, the present invention provides a fault multiple diagnosis reasoning method and system based on tensor decomposition, which decomposes the fault mode of each fault phenomenon into multiple fault elements, queries multiple fault causes for each fault element, and vectorizes the entities in the fault knowledge graph through tensor decomposition to obtain a fault vectorized knowledge graph. The fault vectorized knowledge graph can diagnose comprehensive fault causes for the fault phenomenon.

[0005] To achieve the above-mentioned object of the invention, the present invention provides a fault multiple diagnosis method based on tensor decomposition, which comprises the following steps: Step 1: Based on expert experience, sort out the fault mode of the system fault phenomenon, decompose the fault mode into multiple fault elements, and sort out multiple fault causes for the fault elements; take the fault phenomenon, fault element and fault cause as entities, and the relationship between the fault phenomenon and the fault element, and the relationship between the fault element and the fault cause as edges to build the initial fault knowledge graph; Step 2: Use tensor decomposition to vectorize the entities and edges in the initial fault knowledge graph to obtain a fault vector knowledge graph; Step 3: Obtain the fault phenomenon to be detected , calculate the similarity of all entities in the fault vector knowledge graph of the fault phenomenon to be detected, and the similarity of all entities in the vector knowledge graph with the fault phenomenon to be detected The most similar entity All entities directly connected by edges are regarded as a set of one-hop entities connected to the fault phenomenon to be detected, denoted as .

[0006] To achieve the above-mentioned object of the invention, the present invention also provides a system, which includes a storage medium and one or more processors, wherein the storage medium stores a computer program, and the computer program is called by one or more processors to implement the above-mentioned tensor decomposition-based fault multiple diagnosis method.

[0007] Compared with the prior art, the present invention provides a fault multiple diagnosis reasoning method and system based on tensor decomposition, which has the following beneficial effects: by decomposing the fault mode of each fault phenomenon into multiple fault elements, multiple fault causes are queried for each fault element, and the entities in the fault knowledge graph are vectorized by tensor decomposition to obtain a fault vectorized knowledge graph, and the comprehensive fault causes for the fault phenomenon can be diagnosed through the fault vectorized knowledge graph. BRIEF DESCRIPTION OF THE DRAWINGS

[0008] Figure 1 It is a flow chart of the fault multiple diagnosis and deduction method based on tensor decomposition provided by the present invention. DETAILED DESCRIPTION

[0009] The technical solutions in the embodiments of the present invention will be clearly and completely described below in conjunction with the drawings in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, rather than all the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without making creative work are within the scope of protection of the present invention.

[0010] Figure 1is a flow chart of a fault multiple diagnosis method based on tensor decomposition provided by the present invention, such as Figure 1 As shown, the tensor decomposition-based fault multiple diagnosis method provided by the present invention includes the following steps: Step 1: Based on expert experience, sort out the fault mode of the system fault phenomenon, decompose the fault mode into multiple fault elements, and sort out multiple fault causes for the fault elements; take the fault phenomenon, fault element and fault cause as entities, and the relationship between the fault phenomenon and the fault element, and the relationship between the fault element and the fault cause as edges to build the initial fault knowledge graph; Step 2: Use tensor decomposition to vectorize the entities and edges in the initial fault knowledge graph to obtain a fault vector knowledge graph; Step 3: Obtain the fault phenomenon to be detected , calculate the similarity of all entities in the fault vector knowledge graph of the fault phenomenon to be detected, and the similarity of all entities in the vector knowledge graph with the fault phenomenon to be detected The most similar entity The entities directly connected by the edge are regarded as the one-hop entity set connected to the fault phenomenon to be detected, denoted as , M is a positive integer.

[0011] The tensor decomposition-based fault multiple diagnosis and inference method provided by the present invention also includes the following steps: Step 4: Query the fault vector knowledge graph for the one-hop entity set Each entity in The entities directly connected by the edge minus the return flow entity is used as the two-hop entity set of the fault phenomenon to be detected. , In the formula, Representation and Entity The entity set obtained by subtracting the backflow entities from the entities directly connected by the edges.

[0012] By analogy, query the fault vector knowledge graph with n-1 hop entity set E n Each entity in the is directly connected to the entities through the edge and is reduced to the entities in the backflow as the n-hop entity set of the fault phenomenon to be detected. After multiple reasonings, the comprehensive fault causes for the fault phenomenon can be diagnosed through the fault vectorization knowledge graph.

[0013] In the present invention, using tensor decomposition to vectorize entities and edges in the initial fault knowledge graph to obtain a fault vector knowledge graph also includes: S1-1: The two entities and relations in the initial fault knowledge graph constitute a fault triple, and their vector triples are , where h represents the vector of the head entity and r represents the vector of the relationship. A vector representing the tail entity; S1-2: Extract all entities to form entity codes, extract all relations in a relationship to form relationship codes, the size of the entity set is , the size of the relationship set is ; S1-3: Constructing tensors ,if If true, set , otherwise it is 0; then the tensor To break it down: , where is the embedding matrix of all entities, mapping the vectors of all entities to the entity space ; is the embedding matrix of all relations, mapping the vector of relations to the relation space ; represents tensor modular product; S1-4: From the embedding matrix Take any two rows from the header as the embedding vector of the head entity and tail entity embedding vector ; From the relation embedding matrix Take a row as the relation embedding vector ; S1-5: Traverse the vector triples of fault triples in the initial fault knowledge graph Set D, calculate the vector triple according to the following formula The forward loss function : ; S1-6: Construct negative samples by randomly replacing vector triplets Generate a pseudo triplet vector from the vector of the head or tail entity in , , ; S1-7: Traversing the false triple vectors in the initial fault knowledge graph gather , calculate the pseudo triple vector according to the following formula Negative loss function : , In the formula, in the formula, is the head entity embedding vector of the fake triplet; is the tail body embedding vector of the false triple; S1-8: Use the following formula to calculate the total loss L: , In the formula, represents the positive part; D is the set of real triples, is the set of negative triples, is a hyperparameter; S1-9: Determine whether the total loss L is the minimum. If the total loss L is the minimum, use the vector value of the test fault triplet Verification is performed and the entities and relationships in the initial fault knowledge graph are assigned fixed vectors to obtain the optimized core tensor. , entity embedding matrix and the relation embedding matrix , and then obtain the initial fault vector knowledge graph; if the total loss L does not reach the minimum, the gradient descent method is used to update the vector h of the head entity, the vector t of the tail entity and the vector value r of the relationship, and the updated vector values ​​of the head entity, tail entity and relationship replace the vector values ​​before the update, and then return to step S1-4.

[0014] Optionally, the vectors of fault triples and false triples in the initial fault knowledge graph are perturbed, and the perturbation loss is calculated by the following formula: , In the formula, is the disturbance amount, and determines the total loss Is it the minimum? If the total loss is minimized, the test fault triples are used for verification. If the verification is qualified, the entities and relationships in the initial fault knowledge graph are assigned fixed vectors, and the optimized entity embedding matrix, relationship embedding matrix and normal vector matrix are obtained, and then the initial fault vector knowledge graph is obtained; if the total loss L is not minimized, the updated values ​​of the head entity vector h, the tail entity vector t and the relationship vector r are replaced by the updated values ​​before the update, and then return to step S1-4.

[0015] The present invention quantifies all entities and relationships in the initial fault knowledge graph through the above technical scheme, and assigns a fixed vector to each entity and relationship in the initial fault knowledge graph, thereby laying a foundation for diagnosing comprehensive fault causes for fault phenomena and enhancing the applicability of the quantified knowledge graph.

[0016] In the present invention, using tensor decomposition to vectorize entities and edges in the initial fault knowledge graph to obtain a fault vector knowledge graph also includes: S1-10: Select any two entities in the initial fault vector knowledge graph and , ; Calculate their embedding vectors respectively as and ; Choose any relationship in the relationship , calculate its embedding vector , which forms a fault embedding vector triple , the fault embedding vector triplet is calculated according to the following formula: Tensor: , In the formula, , represents tensor modular product; S1-11: Determine the tensor Is it close to 1? If the tensor , For the minimum setting value, the entity and entities Connected, the connected edges are ; If the tensor ,entity and entities Not connected; S1-12: Repeat steps S1-10 and S1-11 to traverse all entities in the initial fault vector knowledge graph to obtain a first fault vector knowledge graph.

[0017] The present invention uses the above technical solution to make the connection of entities in the initial fault vector knowledge graph more comprehensive.

[0018] In the present invention, using tensor decomposition to vectorize entities and edges in the initial fault knowledge graph to obtain a fault vector knowledge graph also includes the following steps: S1-13: Get new fault symptoms , calculate the new fault phenomenon Similarity with all entities e in the knowledge graph of the first fault vector , if the similarity If it is greater than or equal to the threshold, the fault phenomenon As an entity, and connected to the entity connected to entity e through an edge, the vector of entity e is adjusted according to the similarity as the fault phenomenon The vector of the edge connecting entity e and its connected entities is adjusted according to the similarity as the connection failure phenomenon and the edge vector of the entity connected to entity e; if the similarity less than a threshold, sort out a new fault mode for the new fault phenomenon based on expert experience, decompose the new fault mode into multiple new fault elements, and sort out multiple new fault causes for the new fault elements; take the new fault phenomenon, the new fault element and the new fault cause as entities, and the relationship between the new fault phenomenon and the new fault element, and the relationship between the new fault element and the new fault cause as edges to construct a new initial fault knowledge graph, and then use tensor decomposition to vectorize the entities and edges of the new initial fault knowledge graph to obtain a second fault vector knowledge graph, and fuse the first fault vector knowledge graph and the second fault vector knowledge graph to obtain a fault vector knowledge graph.

[0019] In the present invention, using tensor decomposition to vectorize entities and edges in the initial fault knowledge graph to obtain a fault vector knowledge graph also includes the following steps: Step 1-14: Generate a mapping table by deriving the correspondence between entities and vectors, relations and vectors from the fault vector knowledge graph.

[0020] In the present invention, fusing the first fault vector knowledge graph and the second fault vector knowledge graph to obtain the fault vector knowledge graph includes the following steps: S1-15: embedding matrix of all entities in the knowledge graph of the first fault vector And the embedding matrix of all entities in the second fault vector knowledge graph Concatenate the first fault vector knowledge graph into an embedding matrix of all relations The embedding matrix of all relations between the first fault vector knowledge graph Phase splicing; S1-16: Take any entity from the first fault vector knowledge graph , select any entity from the second fault vector knowledge graph ; Calculate their embedding vectors respectively as and ,in, , ; from Take a relationship , calculate its embedding vector , which forms a fault embedding vector triple , the fault embedding vector triplet is calculated according to the following formula: Tensor: , In the formula, , represents tensor modular product; S1-17: Determine the tensor Is it close to 1? If the tensor , For the minimum setting value, the entity and entities Connected, the connected edges are ; If the tensor ,entity and entities Not connected; S1-18: Repeat steps S1-16 and S1-17 to traverse all entities in the initial fault vector knowledge graph to obtain the fault vector knowledge graph.

[0021] The present invention can fuse multiple fault vector knowledge graphs greater than or equal to 2 through the above technical solution without having to re-quantize the spliced ​​fault knowledge graphs.

[0022] The present invention expands the fault-based knowledge graph through the above technical solution, making the fault phenomenon, fault elements and fault causes more comprehensive, which facilitates more accurate fault diagnosis.

[0023] In the present invention, the mapping table records the correspondence between entities and vectors, and between relations and vectors in the fault vector knowledge graph.

[0024] To achieve the above-mentioned object of the invention, the present invention also provides a system, which includes a storage medium and one or more processors, wherein the storage medium stores a computer program, and the computer program is called by one or more processors to implement the above-mentioned tensor decomposition-based fault multiple diagnosis method.

[0025] To achieve the above-mentioned object of the invention, the present invention also provides a computer program product, which uses computer language to compile the above-mentioned tensor decomposition-based multiple fault diagnosis and deduction method into a computer program called and executed by one or more processors.

[0026] The preferred embodiments of the present invention disclosed above are only used to help explain the present invention. The preferred embodiments do not describe all the details in detail, nor do they limit the invention to only specific implementation methods. Obviously, many modifications and changes can be made according to the contents of this specification. This specification selects and specifically describes these embodiments in order to better explain the principles and practical applications of the present invention, so that technicians in the relevant technical field can well understand and use the present invention. The present invention is only limited by the claims and their full scope and equivalents.

Claims

1. A fault multiple diagnosis method based on tensor decomposition, characterized in that: The steps include: Step 1: Based on expert experience, sort out the fault mode for the system fault phenomenon, decompose the fault mode into multiple fault elements, and sort out multiple fault causes for each fault element T; take the fault phenomenon, fault element and fault cause as entities, and the relationship between the fault phenomenon and the fault element, and the relationship between the fault element and the fault cause as edges to build the initial fault knowledge graph; Step 2: Use tensor decomposition to vectorize the entities and edges in the initial fault knowledge graph to obtain a fault vector knowledge graph; Step 3: Obtain the fault phenomenon to be detected , calculate the similarity of all entities in the fault vector knowledge graph of the fault phenomenon to be detected, and the similarity of all entities in the vector knowledge graph with the fault phenomenon to be detected The most similar entity All entities directly connected by edges are regarded as a set of one-hop entities connected to the fault phenomenon to be detected, denoted as .

2. The fault multiple diagnosis method based on tensor decomposition according to claim 1 is characterized in that: The following steps are also included: Step 4: Query the fault vector knowledge graph for the one-hop entity set Each entity in Entities directly connected by edges are added as the two-hop entity set of the fault phenomenon to be detected: ， In the formula, Representation and Entity The entity set obtained by subtracting the backflow entities from the entities directly connected by the edges.

3. The fault multiple diagnosis method based on tensor decomposition according to any one of claims 1-2, characterized in that: The fault vector knowledge graph is obtained by vectorizing the entities and edges in the initial fault knowledge graph using tensor decomposition, including: S1-1: The two entities and relations in the initial fault knowledge graph constitute a fault triple, and their vector triples are , where h represents the vector of the head entity and r represents the vector of the relationship. A vector representing the tail entity; S1-2: Extract all entities to form entity codes, extract all relations in a relationship to form relationship codes, the size of the entity set is , the size of the relationship set is ; S1-3: Constructing tensors ,if If true, set , otherwise it is 0; then the tensor To break it down: , where is the embedding matrix of all entities, mapping the vectors of all entities to the entity space ; is the embedding matrix of all relations, mapping the vector of relations to the relation space ; represents tensor modular product; represents the kernel tensor; S1-4: From the embedding matrix Take any two rows from the header as the embedding vector of the head entity and tail entity embedding vector ; From the relation embedding matrix Take a row as the relation embedding vector ; S1-5: Traverse the vector triples of fault triples in the initial fault knowledge graph Set D, calculate the vector triple according to the following formula The forward loss function : 。 4. The fault multiple diagnosis method based on tensor decomposition according to claim 3 is characterized in that: The fault vector knowledge graph obtained by vectorizing the entities and edges in the initial fault knowledge graph using tensor decomposition also includes: S1-6: Construct negative samples by randomly replacing vector triplets Generate a pseudo triplet vector from the vector of the head or tail entity in , , ; S1-7: Traverse the false triple vectors in the initial fault knowledge graph gather , calculate the pseudo triple vector according to the following formula Negative loss function : ， In the formula, Embedding vector for fake triplet head entity; is the tail embedding vector of the false triple; S1-8: Use the following formula to calculate the total loss L: ， represents the positive part; D is the set of real triples, is the set of negative triples, is a hyperparameter.

5. The fault multiple diagnosis method based on tensor decomposition according to claim 4 is characterized in that: The fault vector knowledge graph is obtained by vectorizing the entities and edges in the initial fault knowledge graph using tensor decomposition, including: S1-9: Determine whether the total loss L is the minimum. If the total loss L is the minimum, use the vector value of the test fault triplet Verification is performed and the entities and relationships in the initial fault knowledge graph are assigned fixed vectors to obtain the optimized core tensor. , entity embedding matrix and the relation embedding matrix , and then obtain the initial fault vector knowledge graph.

6. The fault multiple diagnosis method based on tensor decomposition according to claim 5 is characterized in that: The fault vector knowledge graph obtained by vectorizing the entities and edges in the initial fault knowledge graph using tensor decomposition also includes: S1-10: Select any two entities in the initial fault vector knowledge graph and , ; Calculate their embedding vectors respectively as and ; Choose any relationship in the relationship , calculate its embedding vector , which forms a fault embedding vector triple , the fault embedding vector triplet is calculated according to the following formula Tensor: ， In the formula, , represents tensor modular product; S1-11: Determine the tensor Is it close to 1? If the tensor , For the minimum setting value, the entity and entities Connected, the connected edges are ; If the tensor ,entity and entities Not connected; S1-12: Repeat steps S1-10 and S1-11 to traverse all entities in the initial fault vector knowledge graph to obtain a first fault vector knowledge graph.

7. The fault multiple diagnosis method based on tensor decomposition according to claim 6 is characterized in that: Using tensor decomposition to vectorize entities and edges in the initial fault knowledge graph to obtain a fault vector knowledge graph also includes the following steps: S1-13: Get new fault symptoms , calculate the new fault phenomenon Similarity with all entities e in the knowledge graph of the first fault vector , if the similarity If it is greater than or equal to the threshold, the fault phenomenon As an entity, and connected to the entity connected to entity e through an edge, the vector of entity e is adjusted according to the similarity as the fault phenomenon The vector of the edge connecting entity e and its connected entities is adjusted according to the similarity as the connection failure phenomenon and the edge vector of the entity connected to entity e; if the similarity less than a threshold, sort out a new fault mode for the new fault phenomenon based on expert experience, decompose the new fault mode into multiple new fault elements, and sort out multiple new fault causes for the new fault elements; take the new fault phenomenon, the new fault element and the new fault cause as entities, and the relationship between the new fault phenomenon and the new fault element, and the relationship between the new fault element and the new fault cause as edges to construct a new initial fault knowledge graph, and then use tensor decomposition to vectorize the entities and edges of the new initial fault knowledge graph to obtain a second fault vector knowledge graph, and fuse the first fault vector knowledge graph and the second fault vector knowledge graph to obtain a fault vector knowledge graph.

8. The fault multiple diagnosis method based on tensor decomposition according to claim 7 is characterized in that: The following steps are also included: Step 1-14: Generate a mapping table by deriving the correspondence between entities and vectors, relations and vectors from the fault vector knowledge graph.

9. A system, characterized in that: The method comprises a storage medium and one or more processors, wherein the storage medium stores a computer program, and the computer program is called by one or more processors to implement the tensor decomposition-based fault multiple diagnosis and inference method according to any one of claims 1 to 8.

Images