Knowledge graph-based aircraft air conditioning system fault diagnosis method and device

Abstract

The invention discloses an aircraft air conditioning system fault diagnosis method and device based on a knowledge graph. The method comprises the following steps: standardizing and defining collected multi-source data of the aircraft air conditioning system to obtain standardized multi-source data; preprocessing and fusing the standardized multi-source data to obtain fused data; constructing a knowledge graph according to the model of the aircraft air conditioning system; annotating the fused data to obtain annotated operation and maintenance data; constructing an entity recognition model according to a bidirectional gating circulation unit of a self-attention mechanism and a conditional random field method; performing identification and relation extraction on the operation and maintenance data with the labels according to an entity identification model to obtain entity data and relation data; performing entity alignment according to the entity data and the relational data, and then extracting triple data to construct an aircraft air conditioning system knowledge graph; the fault diagnosis model is constructed by combining the knowledge graph of the aircraft air conditioning system with the convolutional neural network and the bidirectional gating circulation unit of the self-attention mechanism, so that the accuracy of a fault diagnosis result is improved.

Description

Technical Field

[0001] The present invention relates to the field of fault diagnosis, and particularly to a method and device for fault diagnosis of an aircraft air conditioning system based on a knowledge graph. Background Art

[0002] The aircraft air conditioning system is a key device to ensure a suitable environment in the cabin and cockpit during high-altitude flight, and its performance is directly related to flight safety and passenger comfort. However, due to the complexity of the air conditioning system and high-altitude flight, the system is prone to failures, and the types and causes of failures are diverse, so it is necessary to troubleshoot the failures of the aircraft air conditioning system in a timely manner. Traditional methods for fault diagnosis of air conditioning systems usually rely on manual experience or rule-based diagnostic systems.

[0003] The applicant has found through research that the existing methods for fault diagnosis of air conditioning systems at least have the following defects:

[0004] The unstructured data in the operation and maintenance data of the aircraft air conditioning system is not fully utilized, and it is easy to lose key information such as rich fault content, fault status, and maintenance methods, resulting in low diagnostic efficiency and insufficient accuracy.

[0005] The information disclosed in this background art section is only intended to enhance the overall understanding of the present invention and should not be regarded as an admission or any form of implication that this information constitutes prior art already known to those of ordinary skill in the art. Summary of the Invention

[0006] The purpose of the present invention is to efficiently and accurately identify the faults of the aircraft air conditioning system.

[0007] The present invention provides a method for fault diagnosis of an aircraft air conditioning system based on a knowledge graph, including the steps of:

[0008] S11. Collect multi-source data of the aircraft air conditioning system; and perform normalized definition on the multi-source data to obtain normalized multi-source data;

[0009] S12. Perform preprocessing and fusion on the normalized multi-source data to obtain pre-training fusion data for fault diagnosis of the aircraft air conditioning system;

[0010] S13. Construct a knowledge graph according to the model of the aircraft air conditioning system; the knowledge graph includes entities, entity attributes, and relationships between the entities;

[0011] S14. Perform annotation on the pre-training fusion data for fault diagnosis of the aircraft air conditioning system to obtain operation and maintenance data of the aircraft air conditioning system with annotations;

[0012] S15. Construct an entity recognition model based on the knowledge graph, the bidirectional gated recurrent unit of the self-attention mechanism, and the conditional random field method;

[0013] S16. Identify entity data from the annotated aircraft air conditioning system operation and maintenance data according to the entity recognition model; extract fault logic representation relationship data from the annotated aircraft air conditioning system operation and maintenance data according to the entity recognition model;

[0014] S17. After entity alignment based on the entity data and the fault logic representation relationship data, extract triple data; construct an aircraft air conditioning system knowledge graph according to the triple data; the triple data includes: entity triples, attribute triples, and time triples;

[0015] S18. Construct a fault diagnosis model based on the aircraft air conditioning system knowledge graph and in combination with a convolutional neural network and a bidirectional gated recurrent unit; the fault diagnosis model is used for fault diagnosis of the aircraft air conditioning system.

[0016] Preferably, in the embodiment of the present invention, the normalization definition of the multi-source data includes:

[0017] Structured data, semi-structured data, and unstructured data, where

[0018] The structured data includes an aircraft proprietary vocabulary definition library; the semi-structured data includes aircraft maintenance records; the unstructured data includes: aircraft maintenance manuals, fault isolation manuals, technical standards, civil aviation professional field books, airworthiness regulations, and expert experience.

[0019] Preferably, in the embodiment of the present invention, the preprocessing and fusion of the normalized multi-source data includes:

[0020] The preprocessing includes: removing stop words, text normalization, and delimiter processing;

[0021] The fusion includes: information extraction, information mapping, and information fusion.

[0022] Preferably, in the embodiment of the present invention, constructing the knowledge graph includes:

[0023] Construct entities of the knowledge graph; the entities include: fault location, fault form, fault cause, and maintenance measures;

[0024] Describe the characteristics of the entities;

[0025] Define the relationships between the entities, including subordination relationships, causal relationships, and dependency relationships.

[0026] Preferably, in the embodiment of the present invention, the annotation of the pre-trained fusion data for aircraft air conditioning system fault diagnosis includes:

[0027] Annotating the start of the entity segment, the middle of the entity segment, the end of the entity segment, and the characters through the BIEO annotation method.

[0028] Preferably, in the embodiment of the present invention, the obtaining of the entity data by identifying the annotated aircraft air conditioning system operation and maintenance data according to the entity recognition model includes the steps of:

[0029] S161. Converting the annotated aircraft air conditioning system operation and maintenance data into a plain text format, adding a domain vocabulary, and extracting sentence vectors through a pre-trained language model;

[0030] S162. Inputting the sentence vectors into a bidirectional gated recurrent unit improved by a self-attention mechanism to obtain respective feature functions; the feature functions include latent features;

[0031] S163. Performing label constraint on the basis of each of the feature functions through the conditional random field method to obtain the entity data.

[0032] Preferably, in the embodiment of the present invention, the obtaining of the fault logic representation relationship data by performing relationship extraction on the annotated aircraft air conditioning system operation and maintenance data according to the entity recognition model includes:

[0033] Inputting the annotated aircraft air conditioning system operation and maintenance data into a bidirectional gated recurrent unit to obtain the front and back information relationship of the aircraft air conditioning system operation and maintenance data;

[0034] Calculating the fault logic representation relationship data according to the aircraft air conditioning system operation and maintenance data with the front and back information relationship through a self-attention mechanism.

[0035] Preferably, in the embodiment of the present invention, the entity alignment includes:

[0036] Calculating the similarity between the entities by using the cosine similarity through the following formula:

[0037]

[0038] In the formula, represents the similarity of the word vectors X a and the word vector X b ; is the similarity of the entity names; , are the word vectors of different entities; is the transpose of the word vector The The larger the value, the more similar the entity names are.

[0039] Preferably, in the embodiments of the present invention, the fault diagnosis model includes:

[0040] An entity set is obtained through a knowledge graph representation learning model according to the aircraft air conditioning system knowledge graph;

[0041] An optimal feature splicing is obtained through the convolutional neural network according to the entity set;

[0042] The optimal feature splicing is input into the bidirectional gated recurrent unit to obtain a feature vector set ;

[0043] The feature vector set is input into a fully connected layer to obtain the fault diagnosis result of the aircraft air conditioning system; In the formula, represents the feature representation vector corresponding to the first step; represents the feature representation vector corresponding to the second step; represents the feature representation vector corresponding to the

[0044] On the other hand of the present invention, there is also provided a fault diagnosis device for an aircraft air conditioning system based on a knowledge graph, including:

[0045] A multi-source data acquisition unit, configured to acquire multi-source data of the aircraft air conditioning system; and perform a standardized definition on the multi-source data to obtain standardized multi-source data;

[0046] An aircraft air conditioning system fault diagnosis pre-training fusion data generation unit, configured to perform pre-processing and fusion on the standardized multi-source data to obtain aircraft air conditioning system fault diagnosis pre-training fusion data;

[0047] A knowledge graph construction unit, configured to construct a knowledge graph according to the model of the aircraft air conditioning system; the knowledge graph includes entities, entity attributes, and relationships between the entities;

[0048] An aircraft air conditioning system operation and maintenance data annotation unit, configured to annotate the aircraft air conditioning system fault diagnosis pre-training fusion data to obtain annotated aircraft air conditioning system operation and maintenance data;

[0049] An entity recognition model construction unit, configured to construct an entity recognition model according to the knowledge graph in combination with a bidirectional gated recurrent unit and a conditional random field method of a self-attention mechanism;

[0050] The entity data and fault logic characterization relationship data acquisition unit is used to identify entity data from the labeled aircraft air conditioning system operation and maintenance data according to the entity recognition model; and extract fault logic characterization relationship data from the labeled aircraft air conditioning system operation and maintenance data according to the entity recognition model.

[0051] The aircraft air conditioning system knowledge graph construction unit is used to perform entity alignment based on the entity data and the fault logic characterization relationship data and extract triple data; construct an aircraft air conditioning system knowledge graph according to the triple data; the triple data includes: entity triples, attribute triples, and time triples.

[0052] The fault diagnosis model construction unit is used to construct a fault diagnosis model according to the aircraft air conditioning system knowledge graph in combination with a convolutional neural network and a bidirectional gated recurrent unit; the fault diagnosis model is used for fault diagnosis of the aircraft air conditioning system.

[0053] Compared with the prior art, the present invention has the following beneficial effects:

[0054] In the present invention, by analyzing the mechanism of the aircraft air conditioning system, the multi-source data collected is defined in a standardized manner to obtain standardized multi-source data, thereby generating the proprietary terms of the aircraft air conditioning system; the standardized multi-source data is preprocessed and fused to obtain the pre-training fusion data for aircraft air conditioning system fault diagnosis; a knowledge graph is constructed according to the model of the aircraft air conditioning system; and the pre-training fusion data for aircraft air conditioning system fault diagnosis is labeled to obtain the labeled aircraft air conditioning system operation and maintenance data; an entity recognition model is constructed according to the knowledge graph in combination with the bidirectional gated recurrent unit and conditional random field method of the self-attention mechanism. The entity recognition model adds an entity-level masking strategy and a phrase-level masking strategy, which can effectively learn the information of semantic dependencies and also increase the generalization and adaptability of the entity recognition model; the self-attention mechanism is introduced into the entity recognition model to increase the weight ratio of feature allocation, so as to identify the contribution degree of different types of feature information in the aircraft air conditioning system operation and maintenance data to the fault type detection, and at the same time, potential features can be effectively identified; through the bidirectional gated recurrent unit and conditional random field method, the global dependency relationship of the aircraft air conditioning system operation and maintenance data can be captured, thereby improving the correlation between the aircraft air conditioning system operation and maintenance data; the "entity, relationship, entity" triple data is formed according to the entity data identified by the entity recognition model and the extracted fault logic representation relationship data, and the aircraft air conditioning system knowledge graph constructed by the triple data has stronger reliability; taking the aircraft air conditioning system knowledge graph as the underlying database, a fault diagnosis model of the aircraft air conditioning system is constructed through a convolutional neural network and a bidirectional gated recurrent unit; the present invention constructs a fault diagnosis model by combining the aircraft air conditioning system knowledge graph with a convolutional neural network and a bidirectional gated recurrent unit of the self-attention mechanism, improves the accuracy of the fault diagnosis result, and solves the problems of poor interpretability and difficulty in reasoning when using a convolutional neural network model alone.

[0055] In addition, the present invention stores the aircraft air conditioning system operation and maintenance data in the underlying graph database in the form of triples with the graph as the core, realizing the visualization of the aircraft air conditioning system operation and maintenance data.

[0056] The above description is only an overview of the technical solution of the present invention. In order to be able to understand the technical means of the present invention more clearly and implement it according to the content of the specification, and at the same time to make the above and other purposes, technical features and advantages of the present invention more understandable, one or more preferred embodiments are listed below and described in detail in conjunction with the accompanying drawings as follows. Brief Description of the Drawings

[0057] In order to more clearly illustrate the technical solution of the present invention, the drawings required for the embodiments will be briefly introduced below. Obviously, the drawings in the following description are only some embodiments of the present invention. For those of ordinary skill in the art, other drawings can be obtained based on these drawings without creative efforts.

[0058] Figure 1 It is the step diagram of the aircraft air conditioning system fault diagnosis method based on the knowledge graph described in the present invention; Figure 2 It is the structural schematic diagram of fusing the normalized multi-source data described in the present invention; Figure 3 It is the structural schematic diagram of the BIEO annotation described in the present invention; Figure 4 It is the structural schematic diagram of the entity recognition model based on ERNIE-SABiGRU-CRF described in the present invention; Figure 5 It is the structural schematic diagram of the ERNIE layer network described in the present invention; Figure 6 It is the structural schematic diagram of the SABiGRU model described in the present invention; Figure 7 It is the structural schematic diagram of the fault diagnosis model based on KG-CNN-BiGRU described in the present invention; Figure 8 It is the structural schematic diagram of the confusion matrix of the fault diagnosis result described in the present invention; Figure 9 The structural schematic diagram of the aircraft air conditioning system fault diagnosis device based on the knowledge graph described in the present invention. Specific Embodiments

[0059] The following combines the accompanying drawings to describe in detail the specific embodiments of the present invention, but it should be understood that the protection scope of the present invention is not limited by the specific embodiments.

[0060] Unless otherwise clearly stated, throughout the specification and claims, the term "comprising" or its variations such as "including" or "having" etc. will be understood to include the stated elements or components, and does not exclude other elements or other components.

[0061] In this article, the terms "first", "second", etc. are used to distinguish two different elements or parts, and are not used to limit a specific position or relative relationship. In other words, in some embodiments, the terms "first", "second", etc. can also be interchanged with each other.

[0062] Embodiment 1

[0063] In order to efficiently and accurately identify the faults of the aircraft air conditioning system, as Figure 1 shown, in the embodiment of the present invention, a method for diagnosing faults in an aircraft air conditioning system based on a knowledge graph is provided, including the steps:

[0064] S11. Collect multi-source data of the aircraft air conditioning system; and perform normalized definition on the multi-source data to obtain normalized multi-source data;

[0065] Due to the professional and strict characteristics of aircraft air conditioning system data, it is necessary to standardize the definition of professional terms based on complete expert knowledge, which can prevent the entity recognition model from separating proprietary vocabulary during knowledge extraction, resulting in a decrease in entity recognition accuracy. In the embodiments of the present invention, the multi-source data of the aircraft air conditioning system collected is deeply analyzed, the multi-source data is standardized, and a custom vocabulary library for the aircraft air conditioning system is designed.

[0066] Among them, the standardization of multi-source data includes:

[0067] Structured data, semi-structured data, and unstructured data, where

[0068] Structured data includes the aircraft proprietary vocabulary definition library; semi-structured data includes aircraft maintenance records; unstructured data includes: aircraft maintenance manuals, fault isolation manuals, technical standards, civil aviation professional field books, airworthiness regulations, and expert experience.

[0069] In actual application scenarios, structured data is recorded in a database; semi-structured data is recorded in XML or JSON, and unstructured data is recorded in text or pictures.

[0070] The custom vocabulary library for the aircraft air conditioning system can be designed with reference to Table 1 below:

[0071] Table 1:

[0072] S12. Preprocess and fuse the standardized multi-source data to obtain pre-training fusion data for aircraft air conditioning system fault diagnosis;

[0073] The preprocessing of the standardized multi-source data includes: deleting stop words, text normalization, and delimiter processing;

[0074] In the embodiments of the present invention, in order to make the entity recognition model pay more attention to the words and sentence structures in the text, these texts need to be cleaned; mainly including special symbols in the text, such as: "()", "!", "…" and other symbols, and the format is unified by removing redundant spaces; in the manually described maintenance text, there are usually words with high frequency but no actual meaning, such as "of", "and", "and", etc. Removing these words can reduce the feature dimension of the text to a certain extent.

[0075] In order to break the barriers between the operation and maintenance data of the aircraft air conditioning system, in the embodiments of the present invention, the preprocessed standardized multi-source data is fused, specifically as Figure 2As shown, it includes: information extraction, information mapping, and information fusion; information extraction refers to extracting the required information from text data or tabular data; information mapping refers to finding the data type that matches this group of data from another group of data; the purpose of information fusion is to merge the mapped data into a group of data.

[0076] First, train the semantic vectors of multi-source data through the Word2Vec word vector model to capture the semantic relationships between words and output the fixed-dimensional vectors of each word. ; Then calculate the importance of each word in a specific document and suppress high-frequency but meaningless words (such as stop words); it can be calculated through the following formula:

[0077]

[0078] Among them, is the word appearance times in the document, is the word inverse document frequency, which can be expressed as , is the total number of documents, is the word document number.

[0079] Finally, combine the semantic vectors of words with their importance weights, input the calculated TF-IDF scores, and form a comprehensive feature vector , integrate the comprehensive feature vectors from different sources into a unified feature space, and generate a new text set.

[0080] S13. Construct a knowledge graph according to the model of the aircraft air conditioning system; the knowledge graph includes entities, entity attributes, and the relationships between the entities;

[0081] Constructing a knowledge graph includes:

[0082] Construct the entities of the knowledge graph; the entities include: fault location (Position), fault form (Form), fault reason (Reason), and maintenance measure (Method);

[0083] Describe the characteristics of the entities;

[0084] Define the relationships between the entities, including subordination relationships, causal relationships, and dependency relationships.

[0085] In the embodiment of the present invention, a top-down method is used to construct the knowledge graph. First, construct the knowledge graph relationship and ontology model to describe the relationships between various faults and their related components, phenomena, and solutions; then build a data schema graph to describe the organizational structure of the data model, the relationships between elements, and the storage and flow of data.

[0086] The entities of the constructed knowledge graph are shown in Table 2 below:

[0087] Table 2:

[0088] Among them, the category of the fault location is defined as the location where the fault occurs, such as: distribution system, air-conditioning refrigeration system, heating system, equipment cooling system, temperature control system, pressurization system, etc.; the category of the fault form is defined as the form in which the fault occurs, such as: less cabin air supply, abnormal closing of the flow control valve, high outlet temperature of the high-pressure compressor, etc.; the category of the fault cause is defined as the cause of the fault, such as: poor sealing of the connecting pipeline between the flow sensor and the flow regulator, blockage of the heat exchanger, damage to the blades of the air cycle machine ACM, overheat trip switch of the air-conditioning refrigeration system, etc.; the category of the maintenance measure is defined as the troubleshooting method for the occurrence of the fault, such as: first check the appearance state of the connection pipeline joint between the flow sensor and the flow regulator and whether the installation and fixation do not meet the requirements of defects, etc.

[0089] S14. Label the pre-trained fusion data for aircraft air-conditioning system fault diagnosis to obtain labeled aircraft air-conditioning system operation and maintenance data;

[0090] In the embodiment of the present invention, labeling the pre-trained fusion data for aircraft air-conditioning system fault diagnosis includes: labeling the start of the entity segment, the middle of the entity segment, the end of the entity segment, and the characters through the BIEO labeling method. Among them, B means Begin, indicating the start of the entity segment, I means Inside, indicating the middle of the entity segment, E means End, indicating the end of the entity segment, and O means Other, indicating that the character is not any entity; there are mainly four extracted entities, namely fault location (Position), fault form (Form), fault cause (Reason), and maintenance measure (Method), so there are a total of 13 labels: B-Position, I-Position, E-Position; B-Form, I-Form, E-Form; B-Reason, I-Reason, E-Reason; B-Method, I-Method, E-Method; Other. For the convenience of labeling, take the first letters P, F, R, M of Position, Form, Reason, Method for simplification; the BIEO labeling can be referred to Figure 3 As shown, "blockage of the heat exchanger" represents the fault cause, where "heat" represents the start of the fault cause entity (such as Figure 3 "B-R" in Figure 3 ), "exchanger blockage" represents the inside of the fault cause entity (such asFigure 3 ("E-R" in Figure 3 ), "prone to" represents a non-entity type (such as Figure 3 "O" in Figure 3 ). "High high-pressure compressor outlet temperature" represents a fault form, where "high" represents the start of the fault result entity (such as Figure 3 "B-F" in Figure 3 ), "compressor outlet temperature" represents the interior of the fault result entity (such as Figure 3 "I-F" in Figure 3 ), and the ending "high" represents the end of the fault result entity (such as Figure 3 "E-F" in Figure 3 ).

[0091] S15. Construct an entity recognition model based on the knowledge graph, the bidirectional gated recurrent unit with self-attention mechanism, and the conditional random field method;

[0092] In the embodiment of the present invention, an entity recognition model (ERNIE-SABiGRU-CRF) is constructed. As Figure 4 shown, among them, the input layer is a knowledge-enhanced representation layer (i.e., the ERNIE layer), which performs word embedding and position embedding processing on the input text to obtain the embedding vector Z 1 of the word, Z 2 ,..., Z n ; then comes the self-attention bidirectional gated recurrent unit layer (i.e., the SABiGRU layer), which extracts enhanced context information through the stacked structure of the bidirectional gated recurrent unit (i.e., the BiGRU layer) and the Self-Attention layer to obtain richer hidden states h 1 of the word, h 2 ,..., h n ; finally, there is the conditional random field layer (Conditional Random Field, CRF as Figure 4 the CRF layer in Figure 4 ) for sequence labeling, and finally outputs the BIEO tags B, I, E, O to complete entity recognition. Compared with the traditional pre-trained language model, the ERNIE pre-trained model mainly adds an entity-level masking strategy and a phrase-level masking strategy, which can effectively learn semantic dependency information and improve the generalization and adaptability of the entity recognition model.

[0093] The ERNIE model uses multiple layers of Transformer as the basic decoder, specifically as Figure 5As shown, the input sentence is connected with the unique tokens [CLS] and [SEP], where [CLS] stands for Classification and is used to indicate the start of the sentence, and [SEP] stands for Separate and is used to indicate the end of the sentence. The input text "The bearing of the air circulation machine is worn." is decomposed into individual characters, and each character represents its semantic information through word embeddings. The position relationship of the characters is encoded by combining position embeddings. At the same time, sentence embeddings are added to represent the semantics of the entire sentence. These embeddings are combined together by addition to form a comprehensive representation of each character. The combined embedding vectors are input into the Transformer encoder to generate the context representation S of each character. i , which is used for subsequent natural language processing tasks. ERNIE represents the sentence vector through the stacking of a word embedding layer, a position embedding layer, and a sentence embedding layer, as shown by the following formula :

[0094]

[0095] In the formula, is the vector obtained by the aircraft air conditioning system operation and maintenance data through the word embedding layer, is the vector obtained by the aircraft air conditioning system operation and maintenance data through the position embedding layer, is the vector obtained by the aircraft air conditioning system operation and maintenance data through the sentence embedding layer.

[0096] The stacked sentence vector is finally transformed into a vector form and input into the SABiGRU layer.

[0097] The SABiGRU layer, specifically the SABiGRU model shown in Figure 6 , consists of a bidirectional gated recurrent unit BiGRU and a self-attention mechanism and is used to process sequence data. Among them, x i,1 , x i,2 , …, x i,n , each element represents a word or character in the sequence; y i,1 , y i,2 , …, y i,n , each y i,j represents the context information at the j-th position in the sequence, which combines the forward and backward GRU outputs. By capturing the context dependencies at each position in the sequence, the hidden states are input into the Self-Attention layer to obtain the word embedding vectors Z 1 , Z 2 , …, Z n ; and y 1 , y 2 , …, yn Each y i contains the context information at the i-th position in the sequence; k 1 k 2 …, k n represents the key vector calculated by the Self-Attention layer, e 1 e 2 …, e n represents the context representation after being processed by the attention mechanism. Each e i contains the context information at each position in the sequence and can better capture the dependency relationships between different positions in the sequence.

[0098] The input sequence is processed by BiGRU to obtain the context representation of each word; the BiGRU network consists of two independent GRU units, mainly including the reset gate and the update gate :

[0099] Reset gate:

[0100] Update gate:

[0101] Current memory content:

[0102] Output value:

[0103] In the formula, is the weight matrix of the candidate hidden state; is the weight matrix of the reset gate; is the weight matrix of the update gate; the operator represents the element-wise product operator; is the hidden state of the neural network; is the hidden state of the previous time step, is the input of the current time step, is the concatenation of the feature vectors, is the sigmoid function.

[0104] One GRU unit processes data in the forward time series, and the other GRU unit processes data in the reverse time series. Using the BiGRU network can simultaneously capture the forward and backward information relationships of the text data of the aircraft air conditioning system so that the network can better understand the text information. The specific calculation method is as follows:

[0105] ; ; ;

[0106] In the formula, and correspond to the weights of the forward and backward states of the BiGRU respectively; is the bias term, is the operator of the GRU network.

[0107] The self-attention mechanism is calculated through the following formula:

[0108]

[0109]

[0110] In the formula, Q is the query vector; K is the label vector; V is the vector of the information corresponding to the label; is the transpose of the K vector; is the intermediate function; is the length of the K vector.

[0111] The values of the three parameters Q, K, and V are generated by the input values themselves. If the input is defined as , represents the input value, , , are the bias matrices of the three respectively, and their calculation formulas are as follows:

[0112] Q:

[0113] K:

[0114] V:

[0115] Through the self-attention mechanism, the dependency relationships between positions in the sequence can be captured, and long-distance dependency problems can also be better handled.

[0116] For the operation and maintenance data of the aircraft air conditioning system, different types of feature information have different contribution degrees to the detection of fault types. In order to help the entity recognition model effectively discover potential features and thus improve the accuracy and precision of the fault diagnosis model, the weight ratio of feature allocation can be increased, and it is specifically calculated through the following formula:

[0117] ; ; ; ; ; ;

[0118] In the formula, is the input data; is the output data of the vector after passing through the BiGRU layer; is the hidden layer state; is the operator of the BiGRU network; are the weight coefficients of different features; is the calculation result after passing through the first self-attention mechanism; and are the weight coefficient matrices of the first layer and the second layer respectively; is the hyperbolic tangent function; is to the result after inputting into the BiGRU network for learning; is the final calculation result.

[0119] The CRF layer can maximize the log-likelihood of the label distribution of the entire sequence through the conditional random field model, learn the transition probabilities between labels, and consider the dependencies between labels in the entire sequence, especially the relationships between adjacent labels. Given the input sequence , predict the corresponding output label sequence , where the elements in Z are feature vectors; is the output result after the data is input into the SABiGRU layer; the label Y is the corresponding class label; is the output probability of the corresponding label; The probability distribution of the predicted output label sequence Y corresponding to the input sequence Z is as follows:

[0120]

[0121] In the formula, is the transition feature function, describing the relationship between adjacent labels and ; is the state feature function, describing the relationship between the label and the input sequence ; and are the weight parameters of the feature function; is the normalization factor, and the specific calculation formula is as follows:

[0122]

[0123] Then, through normalization processing, it is ensured that the sum of the probabilities of all possible label sequences is 1.

[0124] S16. Identify the entity data from the labeled aircraft air-conditioning system operation and maintenance data according to the entity recognition model; extract the fault logic representation relationship data from the labeled aircraft air-conditioning system operation and maintenance data according to the entity recognition model;

[0125] In the embodiment of the present invention, the entity recognition model constructed according to step S15 is used to recognize the annotated aircraft air conditioning system operation and maintenance data to obtain entity data, including the steps of:

[0126] S161. Convert the annotated aircraft air conditioning system operation and maintenance data into a pure text format, add a domain vocabulary, and extract sentence vectors through a pre-trained language model;

[0127] S162. Input the sentence vectors into a bidirectional gated recurrent unit improved by a self-attention mechanism to obtain respective feature functions; the feature functions include latent features;

[0128] S163. Perform label constraint on the basis of each of the feature functions through the conditional random field method to obtain the entity data.

[0129] The entity recognition model constructed according to step S15 performs relationship extraction on the annotated aircraft air conditioning system operation and maintenance data to obtain fault logic representation relationship data, including:

[0130] Input the annotated aircraft air conditioning system operation and maintenance data into a bidirectional gated recurrent unit to obtain the front and back information relationships of the aircraft air conditioning system operation and maintenance data;

[0131] Calculate the fault logic representation relationship data according to the aircraft air conditioning system operation and maintenance data with the front and back information relationships through a self-attention mechanism.

[0132] S17. After entity alignment is performed according to the entity data and the fault logic representation relationship data, triple data is extracted; a knowledge graph of the aircraft air conditioning system is constructed according to the triple data; the triple data includes: entity triples, attribute triples, and time triples;

[0133] In the embodiment of the present invention, entity alignment refers to determining whether two entities refer to the same object. If they represent the same object, the entities are merged. The cosine similarity method can be used to measure the similarity between entities, and the similarity between two entities is judged by setting a threshold, thereby completing entity alignment, including the following formula:

[0134]

[0135] In the formula, represents the similarity of word vectors X a and word vectors X b ; is the similarity of entity names; is the transpose of the word vector . The larger the value of the two entity names, the more similar the entity names are.

[0136] After entity alignment, triple data is extracted. The triple data includes: entity triples, attribute triples, and time triples. Entity triples represent the relationships between different entities, in the form of "entity - relationship - entity", such as: "cabin - failure form - low air supply volume"; Attribute triples: represent the attributes or characteristics of entities, in the form of "entity - attribute - attribute value", such as: "pressure sensor - output signal - no signal"; Time triples: involve the relationships of time or events, in the form of "entity - relationship - time", such as, "heat exchanger - maintenance time - 2020".

[0137] The extracted triple data is stored in the Neo4j graph database to achieve visualization. In practical applications, when storing the triple data, the Cypher statement "LOAD CSV" can be used to batch import the triple data into the Neo4j platform. The basic usage is "LOAD CSV FROM 'file: / / / xxx.csv' AS row", where FROM: specifies the path of the CSV file; AS row: is used to reference the fields in the row. By representing entities such as fault locations, fault forms, fault causes, and maintenance measures in the form of nodes, and representing the relationships between each entity with edges, a clear and understandable knowledge graph of the aircraft air - conditioning system can be constructed.

[0138] S18. Construct a fault diagnosis model based on the knowledge graph of the aircraft air - conditioning system in combination with a convolutional neural network and a bidirectional gated recurrent unit; the fault diagnosis model is used for fault diagnosis of the aircraft air - conditioning system.

[0139] In the embodiment of the present invention, the constructed fault diagnosis model, such as Figure 7 the fault diagnosis model based on KG - CNN - BiGRU shown, includes:

[0140] Obtain an entity set through a knowledge graph representation learning model (i.e., TransD) according to the knowledge graph of the aircraft air - conditioning system;

[0141] Obtain the optimal feature splicing through the convolutional neural network according to the entity set;

[0142] Input the optimal feature splicing into the bidirectional gated recurrent unit to obtain a set of feature vectors ;

[0143] Input the set of feature vectors into the fully - connected layer to obtain the fault diagnosis result of the aircraft air - conditioning system.

[0144] In practical applications, such as obtaining an entity set through TransD , represents the set quantity, entity vector , represents the th entity in the matrix, , represents the number of faults, represents the entity dimension; the specific calculation formula is as follows:

[0145]

[0146]

[0147]

[0148]

[0149] In the formula, is the result obtained after the entity vector is convolved through the filter , r is the rth entity set, , is the window size, is the number of channels; is the convolution activation function; is the convolution operation; is the bias; is the result of feature splicing; is the optimal data obtained after the final pooling method; represents the optimal feature splicing, represents the number of convolution kernels, represents the optimal data of the feature splicing result corresponding to the convolution kernel p.

[0150] The result after pooling is input into the BiGRU layer, and the output result of the BiGRU layer is , and it is input into the fully connected layer, and the fault diagnosis types and reasons of the aircraft air conditioning system are obtained through the fully connected layer.

[0151] In the formula, represents the feature representation vector corresponding to the first step; represents the feature representation vector corresponding to the second step; represents the feature representation vector corresponding to the th step in the sequence.

[0152] Next, the accuracy of the fault diagnosis result of this method is proved in combination with specific application scenarios. First, according to the fault isolation manual and relevant regulations, the faults of the aircraft air conditioning system are analyzed, and 6 types of faults are selected, namely: distribution system fault, air conditioning refrigeration system fault, heating system fault, equipment cooling system fault, temperature control system fault, and pressurization system fault; the specific content is as follows in Table 3:

[0153] Table 3:

[0154] Through the fault diagnosis model, the fault diagnosis confusion matrix is obtained as Figure 8 shown. The diagonal elements represent the accuracy of the fault information. Only the accuracy of the extraction of 2 types of fault information does not reach more than 95%. Other fault types have good results; thus, it also verifies that the fault diagnosis model can accurately identify the faults of the aircraft air conditioning system.

[0155] In summary, in the embodiment of the present invention, by analyzing the mechanism of the aircraft air conditioning system, the multi-source data collected is defined in a standardized manner to obtain standardized multi-source data, thereby generating the proper nouns of the aircraft air conditioning system; the standardized multi-source data is preprocessed and fused to obtain the pre-training fusion data for fault diagnosis of the aircraft air conditioning system; a knowledge graph is constructed according to the model of the aircraft air conditioning system; and the pre-training fusion data for fault diagnosis of the aircraft air conditioning system is labeled to obtain the labeled operation and maintenance data of the aircraft air conditioning system; an entity recognition model is constructed according to the knowledge graph in combination with the bidirectional gated recurrent unit and conditional random field method of the self-attention mechanism. An entity-level masking strategy and a phrase-level masking strategy are added to the entity recognition model, which can effectively learn the information of semantic dependencies and also increase the generalization and adaptability of the entity recognition model; the self-attention mechanism is introduced into the entity recognition model to increase the weight ratio of feature allocation, thereby identifying the contribution degree of different types of feature information in the operation and maintenance data of the aircraft air conditioning system to the fault type detection, and at the same time, potential features can be effectively identified; through the bidirectional gated recurrent unit and conditional random field method, the global dependency relationship of the operation and maintenance data of the aircraft air conditioning system can be captured, thereby improving the correlation between the operation and maintenance data of the aircraft air conditioning system; according to the entity data identified by the entity recognition model and the extracted fault logic representation relationship data, a triple data of "entity, relationship, entity" is formed. The knowledge graph of the aircraft air conditioning system constructed by the triple data is more reliable; the knowledge graph of the aircraft air conditioning system is used as the underlying database, and a fault diagnosis model of the aircraft air conditioning system is constructed through a convolutional neural network and a bidirectional gated recurrent unit; the present invention constructs a fault diagnosis model through the knowledge graph of the aircraft air conditioning system in combination with a convolutional neural network and a bidirectional gated recurrent unit of the self-attention mechanism, improves the accuracy of the fault diagnosis result, and solves the problems of poor interpretability and difficult reasoning when using a convolutional neural network model alone.

[0156] In addition, in the present invention, the operation and maintenance data of the aircraft air conditioning system is stored in the underlying graph database with a graph as the core, realizing the visualization of the operation and maintenance data of the aircraft air conditioning system.

[0157] Embodiment 2

[0158] Corresponding to the method embodiments, on the other hand of the embodiments of the present invention, there is also provided a fault diagnosis device for an aircraft air conditioning system based on a knowledge graph. Figure 9 FIG. shows a schematic structural diagram of a fault diagnosis device for an aircraft air conditioning system based on a knowledge graph provided by an embodiment of the present invention. The fault diagnosis device for an aircraft air conditioning system based on a knowledge graph is Figure 1 a device corresponding to the fault diagnosis method for an aircraft air conditioning system based on a knowledge graph in the corresponding embodiment, that is, implemented in the form of a virtual device Figure 1 the fault diagnosis method for an aircraft air conditioning system based on a knowledge graph in the corresponding embodiment. Each virtual module constituting the fault diagnosis device for an aircraft air conditioning system based on a knowledge graph can be executed by an electronic device, such as a network device, a terminal device, or a server. Specifically, the fault diagnosis device for an aircraft air conditioning system based on a knowledge graph in the embodiments of the present invention includes:

[0159] A multi-source data acquisition unit 01, configured to acquire multi-source data of the aircraft air conditioning system; and perform a normalized definition on the multi-source data to obtain normalized multi-source data;

[0160] In the embodiments of the present invention, performing a normalized definition on the multi-source data includes:

[0161] structured data, semi-structured data, and unstructured data, where

[0162] the structured data includes an aircraft proprietary vocabulary definition library; the semi-structured data includes aircraft maintenance records; the unstructured data includes: aircraft maintenance manuals, fault isolation manuals, technical standards, civil aviation professional field books, airworthiness regulations, and expert experience.

[0163] In an actual application scenario, the structured data is recorded in a database; the semi-structured data is recorded in XML or JSON, and the unstructured data is recorded in text or pictures.

[0164] An aircraft air conditioning system fault diagnosis pre-training fusion data generation unit 02, configured to perform preprocessing and fusion on the normalized multi-source data to obtain aircraft air conditioning system fault diagnosis pre-training fusion data;

[0165] In the embodiments of the present invention, performing preprocessing on the normalized multi-source data includes: deleting stop words, text normalization, and delimiter processing; performing fusion on the preprocessed normalized multi-source data includes: information extraction, information mapping, and information fusion.

[0166] A knowledge graph construction unit 03, configured to construct a knowledge graph according to the model of the aircraft air conditioning system; the knowledge graph includes entities, entity attributes, and relationships between the entities;

[0167] Constructing a knowledge graph includes:

[0168] Entities for constructing a knowledge graph; the entities include: fault location (Position), fault form (Form), fault reason (Reason), and repair measure (Method);

[0169] Describe the features of the entities;

[0170] Define the relationships between the entities, including subordination relationships, causal relationships, and dependency relationships.

[0171] The aircraft air-conditioning system operation and maintenance data annotation unit 04 is used to annotate the aircraft air-conditioning system fault diagnosis pre-training fusion data to obtain the annotated aircraft air-conditioning system operation and maintenance data;

[0172] In the embodiment of the present invention, annotating the aircraft air-conditioning system fault diagnosis pre-training fusion data includes: annotating the start of the entity segment, the middle of the entity segment, the end of the entity segment, and the characters through the BIEO annotation method.

[0173] The entity recognition model construction unit 05 is used to construct an entity recognition model according to the knowledge graph, the bidirectional gated recurrent unit combined with the self-attention mechanism, and the conditional random field method;

[0174] In the embodiment of the present invention, an entity recognition model (ERNIE-SABiGRU-CRF) is constructed, and the ERNIE pre-training model is introduced. Compared with the traditional pre-training language model, the ERNIE pre-training model adds an entity-level masking strategy and a phrase-level masking strategy, which can effectively learn the information of semantic dependencies and increase the generalization and adaptability of the entity recognition model.

[0175] The entity data and fault logic characterization relationship data acquisition unit 06 is used to identify the entity data from the annotated aircraft air-conditioning system operation and maintenance data according to the entity recognition model; and extract the fault logic characterization relationship data from the annotated aircraft air-conditioning system operation and maintenance data according to the entity recognition model;

[0176] Identifying entity data from the annotated aircraft air-conditioning system operation and maintenance data according to the entity recognition model includes the steps of:

[0177] S161. Convert the annotated aircraft air-conditioning system operation and maintenance data into a pure text format, add a domain vocabulary, and extract sentence vectors through a pre-trained language model;

[0178] S162. Input the sentence vectors into a bidirectional gated recurrent unit improved by the self-attention mechanism to obtain each feature function; the feature function includes potential features;

[0179] S163. Perform label constraints on each of the above-mentioned feature functions through the conditional random field method to obtain the entity data.

[0180] Perform relationship extraction on the labeled aircraft air conditioning system operation and maintenance data using the entity recognition model constructed in step S15 to obtain fault logic representation relationship data, including:

[0181] Input the labeled aircraft air conditioning system operation and maintenance data into a bidirectional gated recurrent unit to obtain the front and back information relationships of the aircraft air conditioning system operation and maintenance data;

[0182] Calculate the fault logic representation relationship data based on the aircraft air conditioning system operation and maintenance data with front and back information relationships through a self-attention mechanism.

[0183] An aircraft air conditioning system knowledge graph construction unit 07, configured to perform entity alignment on the entity data and the fault logic representation relationship data and extract triple data; construct an aircraft air conditioning system knowledge graph according to the triple data; the triple data includes: entity triples, attribute triples, and time triples;

[0184] In the embodiments of the present invention, entity alignment refers to determining whether two entities refer to the same object. If they represent the same object, the entities are merged. The cosine similarity method can be used to measure the similarity between entities, and the similarity between two entities is judged by setting a threshold, thereby completing entity alignment, including the following formula:

[0185]

[0186] In the formula, represents the word vector X a and the word vector X b similarity; is the similarity of entity names; is the transpose of the word vector , and the value of the two entity names is larger, the more similar the entity names are.

[0187] A fault diagnosis model construction unit 08, configured to construct a fault diagnosis model according to the aircraft air conditioning system knowledge graph in combination with a convolutional neural network and a bidirectional gated recurrent unit; the fault diagnosis model is used for fault diagnosis of the aircraft air conditioning system.

[0188] In the embodiments of the present invention, the constructed fault diagnosis model, such as Figure 7 the fault diagnosis model based on KG-CNN-BiGRU shown, includes:

[0189] An entity set is obtained according to the aircraft air - conditioning system knowledge graph through a knowledge graph representation learning model (i.e., TransD).

[0190] An optimal feature splicing is obtained according to the entity set through the convolutional neural network.

[0191] The optimal feature splicing is input into the bidirectional gated recurrent unit to obtain a set of feature vectors ;

[0192] The set of feature vectors is input into the fully - connected layer to obtain the fault diagnosis result of the aircraft air - conditioning system.

[0193] It should be noted that the specific implementation manner and technical effect of the aircraft air - conditioning system fault diagnosis device based on the knowledge graph in the embodiments of the present invention can refer to Figure 1 the corresponding aircraft air - conditioning system fault diagnosis method based on the knowledge graph, which will not be elaborated here.

[0194] In several embodiments provided in the present application, it should be understood that the disclosed systems, devices, and methods can be implemented in other ways. Additionally, the couplings or direct couplings or communication connections shown or discussed among each other can be through some interfaces. The indirect couplings or communication connections of devices or units can be in electrical, mechanical, or other forms.

[0195] The units described as separate components may or may not be physically separated. The components shown as units may or may not be physical units, that is, they can be located in one place or distributed to multiple network units. Some or all of the units can be selected according to actual needs to achieve the purpose of the solution of this embodiment.

[0196] In addition, in each embodiment of the present application, each functional unit can be integrated in one processing unit, or each unit can exist physically alone, or two or more units can be integrated in one unit.

[0197] It should be understood that in the embodiments of the present application, the dependent claims, each embodiment, and features can be combined with each other to achieve the solution of the foregoing technical problems.

[0198] When the above-mentioned functions are implemented in the form of software functional units and sold or used as independent products, they can be stored in a computer-readable storage medium. Based on such an understanding, the technical solution of this application, in essence, or the part that contributes to the prior art or a part of this technical solution can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions for causing a computer device (which may be a personal computer, a server, or a network device, etc.) to execute all or part of the steps of the methods described in various embodiments of this application. The aforementioned storage medium includes: various media that can store program codes, such as USB flash drives, mobile hard disks, read-only memories (ROM, Read-Only Memory), random access memories (RAM, Random Access Memory), magnetic disks, or optical discs.

[0199] The above description of the disclosed embodiments enables those skilled in the art to implement or use this application. Various modifications to these embodiments will be obvious to those skilled in the art, and the general principles defined herein can be implemented in other embodiments without departing from the spirit or scope of this application. Therefore, this application will not be limited to the embodiments shown herein, but rather will be accorded the widest scope consistent with the principles and novel features disclosed herein.

Claims

1. A method for fault diagnosis of aircraft air conditioning system based on knowledge graph, characterized in that: Includes steps: S11, collecting multi-source data of the aircraft air conditioning system; and performing normalized definition on the multi-source data to obtain normalized multi-source data; S12, preprocessing and fusing the normalized multi-source data to obtain aircraft air conditioning system fault diagnosis pre-training fusion data; S13. constructing a knowledge graph according to the model of the aircraft air conditioning system; the knowledge graph includes entities, entity attributes, and relationships between the entities; S14, labeling the aircraft air conditioning system fault diagnosis pre-training fusion data to obtain labeled aircraft air conditioning system operation and maintenance data; S15. Construct an entity recognition model based on the knowledge graph combined with a bidirectional gated recurrent unit of a self-attention mechanism and a conditional random field method; S16, identifying the labeled aircraft air conditioning system operation and maintenance data according to the entity recognition model to obtain entity data; performing relationship extraction on the labeled aircraft air conditioning system operation and maintenance data according to the entity recognition model to obtain fault logic representation relationship data; S17, performing entity alignment according to the entity data and the fault logic representation relationship data and extracting triple data; Constructing a knowledge graph of an aircraft air conditioning system according to the triple data; The triple data includes: entity triple, attribute triple and time triple; S18. Constructing a fault diagnosis model according to the aircraft air conditioning system knowledge graph and in combination with a convolutional neural network and a bidirectional gated recurrent unit; the fault diagnosis model is used for diagnosing faults of the aircraft air conditioning system.

2. The method for fault diagnosis of aircraft air conditioning system based on knowledge graph according to claim 1, characterized in that: The step of normalizing and defining the multi-source data includes: Structured data, semi-structured data and unstructured data, among which, The structured data includes an aircraft-specific vocabulary definition library; the semi-structured data includes aircraft maintenance records; the unstructured data includes: aircraft maintenance manuals, fault isolation manuals, technical standards, civil aviation professional books, airworthiness regulations, and expert experience.

3. The method for fault diagnosis of aircraft air conditioning system based on knowledge graph according to claim 2, characterized in that: The preprocessing and fusing of the standardized multi-source data includes: The preprocessing includes: deleting stop words, text normalization and delimiter processing; The fusion includes: information extraction, information mapping and information fusion.

4. The method for fault diagnosis of aircraft air conditioning system based on knowledge graph according to claim 3, characterized in that: Constructing the knowledge graph includes: Constructing entities of the knowledge graph; the entities include: fault location, fault form, fault cause and maintenance measures; Describing the characteristics of the entity; The relationships between the entities are defined, including subordinate relationships, causal relationships, and dependent relationships.

5. The aircraft air conditioning system fault diagnosis method based on knowledge graph according to claim 4 is characterized in that: The labeling of the aircraft air conditioning system fault diagnosis pre-training fusion data includes: The entity segment start, entity segment middle, entity segment end and characters are marked using the BIEO marking method.

6. The aircraft air conditioning system fault diagnosis method based on knowledge graph according to claim 1, characterized in that: The step of identifying the labeled aircraft air conditioning system operation and maintenance data according to the entity recognition model to obtain entity data comprises the following steps: S161, converting the annotated aircraft air conditioning system operation and maintenance data into a plain text format, adding the domain vocabulary, and extracting sentence vectors through a pre-trained language model; S162, inputting the sentence vector into a bidirectional gated recurrent unit improved by a self-attention mechanism to obtain each feature function; the feature function includes a potential feature; S163. Perform label constraints according to each of the feature functions using the conditional random field method to obtain the entity data.

7. The method for diagnosing aircraft air conditioning system faults based on knowledge graph according to claim 6, characterized in that: The extracting relationships from the labeled aircraft air conditioning system operation and maintenance data according to the entity recognition model to obtain fault logic representation relationship data includes: inputting the labeled aircraft air conditioning system operation and maintenance data into a bidirectional gated loop unit to obtain a before-after information relationship of the aircraft air conditioning system operation and maintenance data; The fault logic representation relationship data is calculated based on the aircraft air conditioning system operation and maintenance data with the previous and next information relationship through a self-attention mechanism.

8. The method for diagnosing aircraft air conditioning system faults based on knowledge graph according to claim 7, characterized in that: The entity alignment includes: The similarity between the entities is calculated using the cosine similarity formula: ； In the formula, Representing word vectors X a and word vectors X b similarity; is the similarity of entity names; For word vector The transpose of the two entity names The larger the value, the more similar the entity names are.

9. The aircraft air conditioning system fault diagnosis method based on knowledge graph according to claim 8, characterized in that: The fault diagnosis model comprises: According to the aircraft air conditioning system knowledge graph, an entity set is obtained by using a knowledge graph representation learning model; Obtaining optimal feature splicing through the convolutional neural network according to the entity set; The optimal features are concatenated and input into the bidirectional gated recurrent unit to obtain a feature vector set ; Inputting the feature vector set into a fully connected layer to obtain a fault diagnosis result of the aircraft air conditioning system; In the formula, Indicates the feature representation vector corresponding to the first step; Represents the feature representation vector corresponding to the second step; Indicates the first The feature representation vector corresponding to the step.

10. A knowledge graph-based aircraft air conditioning system fault diagnosis device, characterized in that: include: A multi-source data acquisition unit, used to collect multi-source data of the aircraft air conditioning system; and performing normalization definition on the multi-source data to obtain normalized multi-source data; an aircraft air conditioning system fault diagnosis pre-training fusion data generating unit, configured to pre-process and fuse the normalized multi-source data to obtain aircraft air conditioning system fault diagnosis pre-training fusion data; a knowledge graph construction unit, configured to construct a knowledge graph according to the model of the aircraft air conditioning system; the knowledge graph comprising entities, entity attributes and relationships between the entities; an aircraft air conditioning system operation and maintenance data labeling unit, configured to label the aircraft air conditioning system fault diagnosis pre-training fusion data to obtain labeled aircraft air conditioning system operation and maintenance data; An entity recognition model building unit, used to build an entity recognition model according to the knowledge graph combined with a bidirectional gated recurrent unit of a self-attention mechanism and a conditional random field method; an entity data and fault logic representation relationship data acquisition unit, configured to identify the labeled aircraft air conditioning system operation and maintenance data according to the entity recognition model to obtain entity data; and extract relationships from the labeled aircraft air conditioning system operation and maintenance data according to the entity recognition model to obtain fault logic representation relationship data; an aircraft air conditioning system knowledge graph construction unit, configured to perform entity alignment based on the entity data and the fault logic representation relationship data and extract triple data; Constructing a knowledge graph of an aircraft air conditioning system according to the triple data; The triple data includes: entity triple, attribute triple and time triple; A fault diagnosis model building unit is used to build a fault diagnosis model based on the aircraft air conditioning system knowledge graph and in combination with a convolutional neural network and a bidirectional gated recurrent unit; the fault diagnosis model is used to diagnose faults of the aircraft air conditioning system.

Images