Knowledge graph-based sensor precision unqualified influence parameter positioning method

Abstract

The present application relates to the technical field of mechanical detection, and more particularly to a sensor precision unqualified influence parameter positioning method based on a knowledge graph, which introduces multi-process domain feature extraction based on historical alarm data and a knowledge graph based on modeling of these features, realizes hidden causal relationship discovery and causal relationship proportion evaluation, and on this basis, constructs an alarm knowledge graph, fuses an expert experience knowledge graph, obtains a parameter positioning knowledge graph, analyzes data characteristics of real-time alarm data and positions nodes on the corresponding sensor knowledge graph, obtains root cause positioning results through alarm causal relationships to discover parameters affecting sensor precision unqualification. The technical scheme realizes alarm root cause positioning for real-time alarm data, can strengthen the supervision of ground tests of process domains with greater influence, and has important significance and value for ensuring the success rate of test flights.

Description

Technical Field

[0001] This invention relates to the field of mechanical testing technology, and in particular to a knowledge graph-based method for locating parameters affected by sensor inaccuracy. Background Technology

[0002] Aircraft mission sensors are responsible for collecting and summarizing data from various systems on the aircraft. This data can be used to analyze the aircraft's operational status. During each stage before flight testing, extensive system-level testing is required. Data collected by mission sensors is used as the basis for the next stage of research and development, and to test system strength, reliability, and performance. However, in practical applications, sensor accuracy can vary due to factors such as individual sensor differences, environmental changes, noise interference, and aging. Using test data from each stage to predict whether the accuracy of each system's sensors will meet standards during flight testing can improve the success rate of each system's flight tests. The test data from each stage is also known as multi-process domain data. When the aircraft is in different process domains, its sensor accuracy will change due to different environments and operating conditions, significantly affecting test accuracy. Currently, there is no unified and reliable method to assess the impact of data from different process domains, which also affects the success rate of aircraft flight tests to some extent. Summary of the Invention

[0003] The purpose of this invention is to address the problem that varying sensor accuracies across different process domains of an aircraft significantly impact the final test accuracy. This invention proposes a knowledge graph-based method for locating the parameters affecting sensor accuracy failures. First, an expert experience knowledge graph is constructed based on ground testing expert knowledge. A data knowledge graph is then built by training a model using historical aircraft mission sensor accuracy failure alarm data (hereinafter referred to as alarm data). Second, the experience knowledge graph and the data knowledge graph are merged to form the knowledge graph for locating the parameters affecting sensor accuracy failures. A corresponding knowledge graph needs to be established for all sensors whose accuracy needs to be assessed. Finally, a real-time analysis method is established based on the knowledge graph. By analyzing real-time alarm data, root cause analysis results are provided, indicating in which process domain the sensor parameter error led to the final accuracy failure. This method enables root cause location of alarms based on real-time alarm data, thereby strengthening the control of ground testing in process domains with significant impact, which is of great significance and value in ensuring the success rate of flight tests.

[0004] The technical solution adopted to achieve the above objectives is as follows:

[0005] A knowledge graph-based method for addressing the impact of sensor inaccuracy on parameter localization, characterized by the following steps:

[0006] S1. Establish a BiLSTM network, which includes an input layer, a hidden layer, and an output layer;

[0007] S2 is based on multi-process domain data from aircraft mission sensors input into the input layer, including expert experience data, historical warning data, and real-time warning data;

[0008] S3, In the hidden layer, a sensor knowledge graph is constructed and stored using multi-process domain data; wherein, the construction of the sensor knowledge graph includes: constructing an experience knowledge graph for different sensors based on expert experience; constructing an alarm knowledge graph for different sensors based on historical alarm data; and fusing the experience knowledge graph and the alarm knowledge graph to obtain the sensor knowledge graph.

[0009] S4 calls upon the sensor knowledge graph to participate in the online analysis of real-time alarm data, in order to locate the root cause of the parameter impact of sensor inaccuracy, and outputs the location result by the output layer.

[0010] Preferably, in step S3, constructing the experience knowledge graph includes the following steps:

[0011] S31-1, based on expert experience, demonstrates the inherent logic between typical aircraft mission sensor failures, failure phenomena, and failure causes.

[0012] S31-2, typical faults, fault phenomena and fault causes are used as graph nodes, and nodes are manually connected according to the corresponding internal logic to obtain an experience knowledge graph;

[0013] S31-3, Update the experience knowledge graph regularly based on usage and accumulated experience.

[0014] Preferably, in step S3, constructing the alarm knowledge graph includes the following steps:

[0015] S32-1, Based on data feature analysis, feature extraction is performed on multi-process domain data, that is, data features of each process domain are extracted from the multi-process domain data of historical alarm data, so as to convert unstructured data into structured data used to construct alarm knowledge graph.

[0016] S32-2, Obtain feature types and causal relationships, that is, based on the feature extraction results, obtain the anomaly types, anomaly occurrence degrees and anomaly causes of aircraft mission sensors in different process domains;

[0017] S32-3, based on feature type and causal relationship, combined with the coupling effect of multiple parameter sensor accuracy failures in multiple process domains, quantifies the weight of the corresponding influence to obtain the causal relationship weight, which serves as the weight of the connection between nodes in the alarm knowledge graph and the basis for subsequent root cause localization.

[0018] S32-4: The anomaly type, anomaly occurrence degree, and anomaly cause obtained from feature extraction are used as nodes of the knowledge graph, and the causal relationship weight is used as the connection line between each node to complete the construction of the alarm knowledge graph.

[0019] Preferably, in step S3, constructing the alarm knowledge graph further includes steps S32-5, namely, revising the alarm knowledge graph by combining expert experience and the weight of causal relationships.

[0020] Preferably, in step S3, the corresponding fusion of the experience knowledge graph and the alarm knowledge graph involves merging the same nodes of the two knowledge graphs and redetermining the weight of causal relationships.

[0021] Preferably, in step S32-2, feature types and causal relationships are obtained by automatically extracting structured statements based on a BiLSTM network.

[0022] Preferably, in step S3, storing the sensor knowledge graph involves uniformly storing the sensor knowledge graphs corresponding to different aircraft mission sensors and establishing indexing rules for quickly selecting the target sensor knowledge graph.

[0023] Preferably, in step S4, the online analysis of real-time alarm data includes the following steps:

[0024] S41, reads real-time alarm data from the target sensor online;

[0025] S42, extract features from the multi-process and multi-process domain data in the real-time alarm data of the target sensor, and obtain the anomaly type of the target sensor based on the feature extraction results;

[0026] S43, retrieve the sensor knowledge graph corresponding to the target sensor from the stored sensor knowledge graph;

[0027] S44, based on the anomaly type of the target sensor and the corresponding sensor knowledge graph, the alarms are converged to the relevant nodes;

[0028] S45. Based on the alarm convergence results, an alarm cause-effect graph is constructed, and suspected paths are calculated according to the weights in the alarm cause-effect graph. All suspected paths are sorted to obtain the root cause path, so as to complete the parameter positioning affected by sensor inaccuracy.

[0029] The beneficial technical effects of this invention are as follows:

[0030] 1) This technical solution proposes a knowledge graph-based method for locating parameters affected by sensor accuracy defects. Based on expert experience and historical alarm data, an expert experience graph and a historical alarm graph are established. The expert experience graph drives the data to fuse and repair the traditional historical alarm graph, thereby improving the reliability of the sensor knowledge graph.

[0031] 2) This technical solution proposes a knowledge graph-based method for locating parameters affected by sensor accuracy defects. It is implemented based on the establishment of a BiLSTM network. The BiLSTM network is used to automatically extract structured statements, realizing the automatic extraction of feature types and causal relationships. It can easily realize the establishment of sensor knowledge graph and real-time analysis of sensor accuracy defects, which is of great significance for quickly and accurately locating parameters affected by sensor accuracy defects. Attached Figure Description

[0032] Figure 1 This is a basic implementation flowchart of the technical solution;

[0033] Figure 2 This is a flowchart illustrating the preferred implementation of this technical solution. Detailed Implementation

[0034] To make the purpose, technical solution and advantages of the invention clearer, the technical solution of the invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are some embodiments of the invention, but not all embodiments.

[0035] Therefore, the following detailed description of the invention provided in the accompanying drawings is not intended to limit the scope of the claimed invention, but merely to illustrate selected embodiments of the invention. All other embodiments obtained by those skilled in the art based on the embodiments of the invention without inventive effort are within the scope of protection of the invention.

[0036] Example 1

[0037] This embodiment discloses a method for parametric localization based on knowledge graphs to address the impact of sensor accuracy deficiencies. As a basic implementation of this invention, it includes the following steps:

[0038] S1, Establish a BiLSTM network, which includes an input layer, hidden layers, and an output layer. Wherein:

[0039] Input layer: Structured information for inputting multi-process domain data;

[0040] Hidden layer: Using structured information, contextual features are learned through the BiLSTM structure, which can distinguish and score the "entities" and "relationships" to which it belongs. The score is the confidence level of the discrimination result.

[0041] Output layer: The result of this layer is to regress and classify the structured blocks of structured information, and establish the association between text and labels and between labels. The former concatenates individual strings into fault entities, while the latter maps each fault entity to the source data and can be used for the construction of knowledge graphs.

[0042] S2 is based on multi-process domain data from aircraft mission sensors input into the input layer, including expert experience data, historical warning data, and real-time warning data.

[0043] S3. In the hidden layer, a sensor knowledge graph is constructed and stored using multi-process domain data. The construction of the sensor knowledge graph includes: constructing an experience knowledge graph for different sensors based on expert experience; constructing an alarm knowledge graph for different sensors based on historical alarm data; and fusing the experience knowledge graph and the alarm knowledge graph to obtain the sensor knowledge graph.

[0044] S4 calls upon the sensor knowledge graph to participate in the online analysis of real-time alarm data, in order to locate the root cause of the parameter impact of sensor inaccuracy, and outputs the location result by the output layer.

[0045] Example 2

[0046] This embodiment discloses a method for parametric localization based on knowledge graphs to address the impact of sensor accuracy deficiencies. As a basic implementation of this invention, it includes the following steps:

[0047] S1. Establish a BiLSTM network, which includes an input layer, a hidden layer, and an output layer.

[0048] S2 is based on multi-process domain data from aircraft mission sensors input into the input layer, including expert experience data, historical warning data, and real-time warning data.

[0049] S3. In the hidden layer, a sensor knowledge graph is constructed and stored using multi-process domain data. The construction of the sensor knowledge graph includes: constructing an experience knowledge graph for different sensors based on expert experience; constructing an alarm knowledge graph for different sensors based on historical alarm data; and fusing the experience knowledge graph and the alarm knowledge graph to obtain the sensor knowledge graph.

[0050] Constructing an experience knowledge graph involves the following steps:

[0051] S31-1, based on the expert experience in the fields of aircraft manufacturing, testing, and flight testing, etc., demonstrates the internal logic among typical faults, fault phenomena, and fault causes of aircraft mission sensors. Further, there is also the evolution relationship between some faults and phenomena. Among them, the fault phenomena include various sensor data and physical changes of the airframe, etc.

[0052] S31-2, takes typical faults, fault phenomena, and fault causes as graph nodes, and manually connects the nodes according to the corresponding internal logic to obtain an empirical knowledge graph. Among them, the connecting lines of the nodes are the internal associations (internal logics) between the nodes.

[0053] S31-3, regularly updates the empirical knowledge graph according to usage and experience accumulation.

[0054] The construction of the alarm knowledge graph includes the following steps:

[0055] S32-1, extracts features from multi-process domain data according to data feature analysis, that is, extracts the data features of each process domain from the multi-process domain data of historical alarm data to convert unstructured data into structured data used for constructing the alarm knowledge graph. Among them, based on manual feature analysis, the data features of each process domain are extracted from the multi-process domain data of historical alarm data, and these features may be quantitative representations of internal and external factors. Some alarms correspond to the environmental reasons of the aircraft and its components, such as the influence caused by temperature and humidity during the manufacturing stage, so the mean and standard deviation of the temperature and humidity of this process are extracted; some alarms correspond to problems of the airframe itself, such as the wear of internal components of the sensor during the flight test stage, so its degradation features are extracted, etc. In this way, unstructured data can be converted into structured data used for constructing the knowledge graph to complete knowledge extraction. The structured data after feature extraction should conform to the Resource Description Framework (RDF), which is represented in the form of triples, that is, "<Subject (subject), Predicate (predicate), Object (object)>", corresponding to "<entity, relationship, entity>", but the information obtained in this step does not necessarily directly distinguish each structural information. To achieve efficient extraction of structural information, an automated method will be used to distinguish each structural block. Taking this method as an example, it can obtain something like "the degradation of the sensor during the flight test stage causes the measured lift speed during the in-flight climb stage to be lower than the actual value".

[0056] S32-2, obtains the feature types and causal relationships, that is, based on the feature extraction results, obtains the abnormal types, abnormal occurrence degrees, and abnormal causes that may occur in different process domains of aircraft mission sensors, and these features can be used as an abstract summary of the abnormalities of all aircraft of the same type.

[0057] S32-3, based on feature type and causal relationship, combines the coupled influence of multiple parameters from multiple process domains on sensor accuracy non-compliance, quantifies the weight of each influence to obtain the causal relationship weight, which serves as the weight of connections between nodes in the alarm knowledge graph and the basis for subsequent root cause localization. Sensor accuracy non-compliance may be caused by the coupled influence of multiple parameters from multiple process domains, therefore, the non-compliance phenomenon cannot be explained by the anomaly of a single parameter. However, the degree of influence of different parameters will vary depending on the type and degree of the anomaly, so it is necessary to quantify the weight of their influence as the weight of connections between nodes in the knowledge graph and the basis for subsequent root cause localization. This process can quantitatively determine the relationship between each parameter and sensor accuracy using methods such as principal component analysis, and then normalize these relationships to obtain the causal relationship weight of each node.

[0058] S32-4: The anomaly type, anomaly occurrence degree, and anomaly cause obtained from feature extraction are used as nodes of the knowledge graph, and the causal relationship weight is used as the connection line between each node to complete the construction of the alarm knowledge graph.

[0059] S4 calls upon the sensor knowledge graph to participate in the online analysis of real-time alarm data, in order to locate the root cause of the parameter impact of sensor inaccuracy, and outputs the location result by the output layer.

[0060] Example 3

[0061] This embodiment discloses a method for parametric localization based on knowledge graphs to address the impact of sensor accuracy deficiencies. As a basic implementation of this invention, it includes the following steps:

[0062] S1. Establish a BiLSTM network, which includes an input layer, a hidden layer, and an output layer.

[0063] S2 is based on multi-process domain data from aircraft mission sensors input into the input layer, including expert experience data, historical warning data, and real-time warning data.

[0064] S3. In the hidden layer, a sensor knowledge graph is constructed and stored using multi-process domain data. The construction of the sensor knowledge graph includes: constructing an experience knowledge graph for different sensors based on expert experience; constructing an alarm knowledge graph for different sensors based on historical alarm data; and fusing the experience knowledge graph and the alarm knowledge graph to obtain the sensor knowledge graph.

[0065] Constructing an experience knowledge graph involves the following steps:

[0066] S31-1, based on expert experience, demonstrates the inherent logic between typical aircraft mission sensor failures, failure phenomena, and failure causes.

[0067] S31-2, typical faults, fault phenomena and fault causes are used as graph nodes, and nodes are manually connected according to the corresponding internal logic to obtain an experience knowledge graph;

[0068] S31-3, Update the experience knowledge graph regularly based on usage and accumulated experience.

[0069] Building an alarm knowledge graph includes the following steps:

[0070] S32-1, Based on data feature analysis, feature extraction is performed on multi-process domain data, that is, data features of each process domain are extracted from the multi-process domain data of historical alarm data, so as to convert unstructured data into structured data used to construct alarm knowledge graph.

[0071] S32-2, Obtaining Feature Types and Causal Relationships: Based on the feature extraction results, this method obtains the anomaly types, anomaly severity, and causes of aircraft mission sensors in different process domains. Specifically, it uses a BiLSTM network to automatically extract feature types and causal relationships from structured statements. To achieve automatic extraction of feature types and causal relationships, this method uses a BiLSTM network to automatically extract structured statements. The BiLSTM network employs a bidirectional network structure, which can effectively extract information from preceding and following text, fully analyze the contextual relationships, distinguish the descriptions of various feature types and causal relationships within a text, and obtain key information.

[0072] S32-3, based on feature type and causal relationship, combined with the coupling effect of multiple parameter sensor accuracy failures in multiple process domains, quantifies the weight of the corresponding influence to obtain the causal relationship weight, which serves as the weight of the connection between nodes in the alarm knowledge graph and the basis for subsequent root cause localization.

[0073] S32-4, use the anomaly type, anomaly occurrence degree and anomaly cause obtained from feature extraction as nodes of the knowledge graph, and use the causal relationship weight as the connection line between each node to complete the construction of the alarm knowledge graph;

[0074] S32-5, This technical solution obtains an alarm knowledge graph through machine learning. However, this method has certain limitations, namely, the obtained alarm knowledge graph may contain some unreasonable factors. Therefore, it is necessary to revise the alarm knowledge graph by combining expert experience and the weight of causal relationships to ensure the reliability of its nodes and connections.

[0075] Furthermore, the corresponding fusion of the experience knowledge graph and the alarm knowledge graph involves merging the same nodes of the two knowledge graphs and redetermining the weight of causal relationships.

[0076] Furthermore, storing the sensor knowledge graph involves uniformly storing the sensor knowledge graphs corresponding to different aircraft mission sensors and establishing indexing rules for quickly selecting the target sensor knowledge graph. This ensures that the model (sensor knowledge graph) corresponding to the target sensor can be quickly selected during online analysis.

[0077] S4 calls upon the sensor knowledge graph to participate in the online analysis of real-time alarm data, in order to locate the root cause of the parameter impact of sensor inaccuracy, and outputs the location result by the output layer.

[0078] Example 4

[0079] This embodiment discloses a method for parametric localization based on knowledge graphs to address the impact of sensor accuracy deficiencies. As a basic implementation of this invention, it includes the following steps:

[0080] S1. Establish a BiLSTM network, which includes an input layer, a hidden layer, and an output layer.

[0081] S2 is based on multi-process domain data from aircraft mission sensors input into the input layer, including expert experience data, historical warning data, and real-time warning data.

[0082] S3. In the hidden layer, a sensor knowledge graph is constructed and stored using multi-process domain data. The construction of the sensor knowledge graph includes: constructing an experience knowledge graph for different sensors based on expert experience; constructing an alarm knowledge graph for different sensors based on historical alarm data; and fusing the experience knowledge graph and the alarm knowledge graph to obtain the sensor knowledge graph.

[0083] Constructing an experience knowledge graph involves the following steps:

[0084] S31-1, based on expert experience, demonstrates the inherent logic between typical aircraft mission sensor failures, failure phenomena, and failure causes.

[0085] S31-2, typical faults, fault phenomena and fault causes are used as graph nodes, and nodes are manually connected according to the corresponding internal logic to obtain an experience knowledge graph;

[0086] S31-3, Update the experience knowledge graph regularly based on usage and accumulated experience.

[0087] Building an alarm knowledge graph includes the following steps:

[0088] S32-1, Based on data feature analysis, feature extraction is performed on multi-process domain data, that is, data features of each process domain are extracted from the multi-process domain data of historical alarm data to convert unstructured data into structured data used to construct an alarm knowledge graph; furthermore, a feature extractor can be trained using historical alarm data, and the feature extractor can be used to perform corresponding feature extraction operations.

[0089] S32-2, obtaining feature types and causal relationships, that is, based on the feature extraction results, obtaining the anomaly types, anomaly occurrence degrees and anomaly causes of aircraft mission sensors in different process domains; wherein, feature types and causal relationships are obtained by automatically extracting structured statements based on BiLSTM network.

[0090] S32-3, based on feature type and causal relationship, combined with the coupling effect of multiple parameter sensor accuracy failures in multiple process domains, quantifies the weight of the corresponding influence to obtain the causal relationship weight, which serves as the weight of the connection between nodes in the alarm knowledge graph and the basis for subsequent root cause localization.

[0091] S32-4, use the anomaly type, anomaly occurrence degree and anomaly cause obtained from feature extraction as nodes of the knowledge graph, and use the causal relationship weight as the connection line between each node to complete the construction of the alarm knowledge graph;

[0092] S32-5, revising the alarm knowledge graph by combining expert experience and the weight of causal relationships.

[0093] Furthermore, the corresponding fusion of the experience knowledge graph and the alarm knowledge graph involves merging the same nodes of the two knowledge graphs and redetermining the weight of causal relationships.

[0094] Furthermore, storing sensor knowledge graphs involves uniformly storing the sensor knowledge graphs corresponding to different aircraft mission sensors in a storage device to obtain a graph database, and establishing indexing rules for quickly selecting target sensor knowledge graphs.

[0095] S4 invokes the sensor knowledge graph to participate in the online analysis of real-time alarm data, in order to pinpoint the root cause of sensor inaccuracy affecting parameters, and the output layer outputs the location results. Specifically, the online analysis of real-time alarm data includes the following steps:

[0096] S41, reads real-time alarm data from the target sensor online;

[0097] S42, extract features from the multi-process and multi-process domain data in the real-time alarm data of the target sensor, and obtain the anomaly type of the target sensor based on the feature extraction results; furthermore, a pre-trained feature extractor can be called to perform the corresponding feature extraction operation.

[0098] S43, retrieve the sensor knowledge graph corresponding to the target sensor from the stored sensor knowledge graph (graph database);

[0099] S44, based on the anomaly type of the target sensor and the corresponding sensor knowledge graph, the alarms are converged to the relevant nodes;

[0100] S45. Based on the alarm convergence results, an alarm cause-effect graph is constructed. This involves querying the graph database for connections between all relevant nodes to form the alarm cause-effect graph. Suspected paths are calculated based on the weights in the graph, and all suspected paths are sorted to obtain the root cause path, thus completing the parameter localization for sensor inaccuracy issues. Specifically, based on the alarm type extracted from the alarm features, a subgraph is queried in the graph database containing nodes surrounding the alarm (within a "one-hop" or low-hop range). Nodes representing all fault cause types are found within this subgraph, and a cause-effect graph is generated with these fault type nodes. Combining the weights of each edge, the root cause path of the anomaly occurring during the alarm process can be calculated.

Claims

1. A knowledge graph-based method for addressing the impact of sensor accuracy defects on parameter localization, characterized in that... Includes the following steps: S1. Establish a BiLSTM network, which includes an input layer, a hidden layer, and an output layer; S2 is based on multi-process domain data from aircraft mission sensors input into the input layer, including expert experience data, historical warning data, and real-time warning data; S3, In the hidden layer, a sensor knowledge graph is constructed and stored using multi-process domain data; wherein, the construction of the sensor knowledge graph includes: constructing an experience knowledge graph for different sensors based on expert experience; constructing an alarm knowledge graph for different sensors based on historical alarm data; and fusing the experience knowledge graph and the alarm knowledge graph to obtain the sensor knowledge graph. Constructing an experience knowledge graph involves the following steps: S31-1, based on expert experience, demonstrates the inherent logic between typical aircraft mission sensor failures, failure phenomena, and failure causes. S31-2, typical faults, fault phenomena and fault causes are used as graph nodes, and nodes are manually connected according to the corresponding internal logic to obtain an experience knowledge graph; S31-3, Update the experience knowledge graph regularly based on usage and accumulated experience; Building an alarm knowledge graph includes the following steps: S32-1, Based on data feature analysis, feature extraction is performed on multi-process domain data, that is, data features of each process domain are extracted from the multi-process domain data of historical alarm data, so as to convert unstructured data into structured data used to construct alarm knowledge graph. S32-2, Obtain feature types and causal relationships, that is, based on the feature extraction results, obtain the anomaly types, anomaly occurrence degrees and anomaly causes of aircraft mission sensors in different process domains; S32-3, based on feature type and causal relationship, combined with the coupling effect of multiple parameter sensor accuracy failures in multiple process domains, quantifies the weight of the corresponding influence to obtain the causal relationship weight, which serves as the weight of the connection between nodes in the alarm knowledge graph and the basis for subsequent root cause localization. S32-4, use the anomaly type, anomaly occurrence degree and anomaly cause obtained from feature extraction as nodes of the knowledge graph, and use the causal relationship weight as the connection line between each node to complete the construction of the alarm knowledge graph; S4 calls upon the sensor knowledge graph to participate in the online analysis of real-time alarm data, in order to locate the root cause of the parameter impact of sensor inaccuracy, and outputs the location result by the output layer.

2. The method for parameter localization based on knowledge graph-based sensor accuracy issues as described in claim 1, characterized in that, In step S3, constructing the alarm knowledge graph also includes steps S32-5, which involves revising the alarm knowledge graph by combining expert experience and the weight of causal relationships.

3. The method for parameter localization based on knowledge graph-based sensor accuracy issues as described in claim 1, characterized in that, In step S3, the corresponding fusion of the experience knowledge graph and the alarm knowledge graph involves merging the same nodes of the two knowledge graphs and redetermining the weight of causal relationships.

4. The method for parameter localization based on knowledge graph-based sensor accuracy issues as described in claim 1, characterized in that, In step S32-2, feature types and causal relationships are obtained by automatically extracting structured statements based on a BiLSTM network.

5. The method for parameter localization based on knowledge graph-based sensor accuracy issues as described in claim 1, characterized in that, In step S3, storing the sensor knowledge graph involves uniformly storing the sensor knowledge graphs corresponding to different aircraft mission sensors and establishing indexing rules for quickly selecting the target sensor knowledge graph.

6. The method for parameter localization based on knowledge graphs to address the impact of sensor accuracy defects as described in claim 1, characterized in that, In step S4, the online analysis of real-time alarm data includes the following steps: S41, reads real-time alarm data from the target sensor online; S42, extract features from the multi-process and multi-process domain data in the real-time alarm data of the target sensor, and obtain the anomaly type of the target sensor based on the feature extraction results; S43, retrieve the sensor knowledge graph corresponding to the target sensor from the stored sensor knowledge graph; S44, based on the anomaly type of the target sensor and the corresponding sensor knowledge graph, the alarms are converged to the relevant nodes; S45. Based on the alarm convergence results, an alarm cause-effect graph is constructed, and suspected paths are calculated according to the weights in the alarm cause-effect graph. All suspected paths are sorted to obtain the root cause path, so as to complete the parameter positioning affected by sensor inaccuracy.

Images