Unmanned aerial vehicle operator training method based on knowledge graph and digital twin technology

Abstract

The application provides a kind of unmanned aerial vehicle operation personnel training method based on knowledge graph and digital twin technology, method includes: the key knowledge in unmanned aerial vehicle knowledge graph is injected into unmanned aerial vehicle digital twin, and fusion twin is obtained;Based on the current actual operation data of the training personnel executed on the fusion twin, generate the training report of the training personnel;Unmanned aerial vehicle knowledge graph is updated based on historical actual operation data.The method provided by the application obtains and fuses unmanned aerial vehicle digital twin and knowledge graph, constructs a knowledge-driven high-fidelity training environment, generates a refined ability evaluation report based on the actual operation data of the students, forms an experience precipitation-simulation verification-practical optimization-experience iteration closed-loop training system, effectively solves many pain points in the traditional training mode, can significantly improve the training efficiency, guarantee training safety, and realize the quantitative evaluation and continuous improvement of the actual combat ability of the students.

Description

Technical Field

[0001] This invention relates to the field of drone training technology, and in particular to a method for training drone maintenance personnel based on knowledge graph and digital twin technology. Background Technology

[0002] With the increasingly widespread application of drones in logistics delivery, agricultural plant protection, power line inspection, and emergency rescue, the market demand and importance of drone maintenance personnel have surged. Currently, training for drone maintenance personnel mainly relies on traditional methods combining oral instruction from professionals with hands-on training.

[0003] However, this model has significant drawbacks. On the one hand, experience transfer is fragmented, with expert knowledge relying heavily on oral instruction and lacking systematic accumulation, making it difficult for new trainees to fully grasp the concepts and for historical experience to be efficiently reused. On the other hand, practical training has strong limitations; operations in high-risk or extreme scenarios cannot be reproduced due to safety and cost issues, resulting in weak emergency response capabilities among trainees. Therefore, current drone operation and maintenance personnel training methods suffer from significant bottlenecks: low efficiency, high cost and risk, and inability to meet the needs of complex scenario training. Summary of the Invention

[0004] This invention provides a training method for drone operation and maintenance personnel based on knowledge graph and digital twin technology, which solves the shortcomings of existing drone operation and maintenance personnel training methods, such as low efficiency, high cost and risk, and inability to meet the training needs of complex scenarios.

[0005] This invention provides a training method for drone operation and maintenance personnel based on knowledge graph and digital twin technology, including: Obtain digital twins and knowledge graphs of drones; Key knowledge from the UAV knowledge graph is injected into the UAV digital twin to obtain a fused twin; Based on the actual operational data of the trainee performing on the fused twin, a training report for the trainee is generated; The drone knowledge graph is updated based on historical operational data.

[0006] According to the present invention, a training method for drone operation and maintenance personnel based on knowledge graph and digital twin technology is provided, wherein the key knowledge includes fault knowledge and / or component structure knowledge; The step of injecting key knowledge from the UAV knowledge graph into the UAV digital twin to obtain a fused twin includes: The fault knowledge is injected into the UAV digital twin to obtain a fault twin; The structural knowledge of the components is injected into the digital twin of the UAV to obtain a structural display twin; The fault twin and / or the structural display twin are used as the fused twin.

[0007] According to the present invention, a training method for drone operation and maintenance personnel based on knowledge graph and digital twin technology includes the following steps for updating the drone knowledge graph: Acquire the initial knowledge graph, the historical operational data, and industry rules; Static entity relationships are extracted from the historical actual operation data, and the initial knowledge graph is updated based on the static entity relationships to obtain a static knowledge graph; Based on the industry rules, knowledge reasoning is performed on the static knowledge graph to obtain the UAV knowledge graph.

[0008] According to the present invention, a training method for drone operation and maintenance personnel based on knowledge graph and digital twin technology is provided, wherein the drone knowledge graph is constructed by performing knowledge reasoning on the static knowledge graph based on the industry rules, including: Event data is extracted from the static knowledge graph; Retrieve the target industry rules that match the event data from the industry rules; Based on the target industry rules, knowledge reasoning is performed on the event data to construct the UAV knowledge graph.

[0009] According to the present invention, a training method for drone operation and maintenance personnel based on knowledge graph and digital twin technology is provided, wherein generating a training report for the trainee based on the current actual operation data performed by the trainee on the fused twin includes: Reference operation data for the current actual operation data is obtained from the UAV knowledge graph; The training report is generated by comparing the reference operation data with the current actual operation data. The reference operation data includes fault reference handling data and / or component reference maintenance data.

[0010] According to the present invention, a training method for drone operation and maintenance personnel based on knowledge graph and digital twin technology, wherein the step of comparing the reference operation data and the current actual operation data to generate the training report includes: By comparing the reference operating data with the current actual operating data, the deviation result is obtained; By reverse querying the knowledge points related to the deviation results from the UAV knowledge graph, the knowledge weaknesses can be identified. Based on the knowledge gaps and the UAV knowledge graph, a knowledge gap topology graph is constructed. The training report is generated based on the aforementioned knowledge gap topology map.

[0011] According to the present invention, a training method for drone operation and maintenance personnel based on knowledge graph and digital twin technology, wherein generating the training report based on the knowledge gap topology graph further includes: Acquire a 3D scene; Acquire the flight operation data of the trainee on the digital twin of the UAV in the three-dimensional scene; Based on the flight operation data, a capability radar map is generated; The training report is generated based on the knowledge gap topology map and the capability radar map.

[0012] According to the present invention, a method for training UAV operation and maintenance personnel based on knowledge graph and digital twin technology is provided, wherein generating a capability radar chart based on the flight operation data includes: Based on the flight operation data, the flight stability score and flight energy efficiency are evaluated. The capability radar map is generated based on the flight stability score and flight energy efficiency.

[0013] This invention also provides a training device for drone operation and maintenance personnel based on knowledge graph and digital twin technology, comprising: The acquisition unit acquires the drone's digital twin and drone knowledge graph; The fusion unit injects key knowledge from the UAV knowledge graph into the UAV digital twin to obtain a fused twin; The report generation unit generates a training report for the trainee based on the actual operation data performed by the trainee on the fused twin. The drone knowledge graph is updated based on historical operational data.

[0014] The present invention also provides an electronic device, including a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the program to implement the drone operation and maintenance personnel training method based on knowledge graph and digital twin technology as described above.

[0015] The present invention also provides a non-transitory computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the drone operation and maintenance personnel training method based on knowledge graph and digital twin technology as described above.

[0016] The present invention also provides a computer program product, including a computer program that, when executed by a processor, implements the drone operation and maintenance personnel training method based on knowledge graph and digital twin technology as described above.

[0017] The UAV operation and maintenance personnel training method based on knowledge graph and digital twin technology provided by this invention constructs a knowledge-driven, high-fidelity training environment by acquiring and integrating UAV digital twins and knowledge graphs. It generates a refined capability assessment report based on trainees' practical data, forming a closed-loop training system of experience accumulation, simulation verification, practical optimization, and experience iteration. This effectively solves many pain points in traditional training models, significantly improves training efficiency, ensures training safety, and enables quantitative assessment and continuous improvement of trainees' practical abilities. Attached Figure Description

[0018] To more clearly illustrate the technical solutions in this invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are some embodiments of this invention. For those skilled in the art, other drawings can be obtained from these drawings without creative effort.

[0019] Figure 1 This is a flowchart illustrating the training method for drone maintenance personnel based on knowledge graph and digital twin technology provided by the present invention. Figure 2 This is a schematic diagram of the structure of the drone maintenance personnel training device based on knowledge graph and digital twin technology provided by the present invention; Figure 3 This is a schematic diagram of the architecture of the drone operation and maintenance personnel training system based on knowledge graph and digital twin technology provided by the present invention; Figure 4 This is a schematic diagram of the structure of the electronic device provided by the present invention. Detailed Implementation

[0020] To make the objectives, technical solutions, and advantages of this invention clearer, the technical solutions of this invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some, not all, of the embodiments of this invention. All other embodiments obtained by those skilled in the art based on the embodiments of this invention without creative effort are within the scope of protection of this invention.

[0021] To address the aforementioned issues, this invention provides a training method for drone maintenance personnel based on knowledge graph and digital twin technologies, enabling efficient, low-cost, low-risk training of drone maintenance personnel that meets the needs of complex scenario training. Figure 1This is a flowchart illustrating the drone maintenance personnel training method based on knowledge graph and digital twin technology provided by the present invention, as shown below. Figure 1 As shown, the method includes: Step 110: Obtain the UAV digital twin and the UAV knowledge graph; the UAV knowledge graph is updated based on historical actual operation data.

[0022] Here, a drone digital twin refers to a high-fidelity, dynamic, virtual model of a specific drone model in the physical world. This model not only includes the drone's three-dimensional geometry and material physical properties, but also integrates its operational dynamics, avionics system behavior logic, and rules for interacting with the environment. In essence, a drone digital twin can provide trainees with an interactive, controllable, and immersive virtual learning environment with realistic physical feedback.

[0023] In addition, the drone knowledge graph is used to describe the concepts, entities, attributes and their interrelationships in the field of drone operation and maintenance. It can be built based on historical actual operation data and transforms scattered and unstructured operation and maintenance experience, such as fault phenomena, causes, troubleshooting steps, required tools, maintenance skills, etc., into a machine-understandable knowledge base.

[0024] Specifically, to obtain a digital twin of a drone, a detailed 3D model of the drone can be constructed using modeling software such as CAD / CAE. Based on a physics engine, such as Unity3D or Unreal Engine, the physical parameters of each component of the drone, such as mass, inertia, and aerodynamic characteristics, can be set. Furthermore, the drone's flight control algorithms and mathematical models of sensors, such as digital models of IMU, GPS, and motor speed sensors, can be integrated to ensure that its behavior in the virtual environment is highly consistent with the physical entity. For example, a 3D simulation scene library containing scenarios such as high-voltage power line environments and mountainous terrain can be built, and a high-fidelity digital twin of the equipment can be calibrated and constructed based on sensor data from the physical drone, such as motor speed and battery voltage.

[0025] Simultaneously, a drone knowledge graph can be obtained by integrating and processing historical operational data, and then updating the initial knowledge graph based on the processed knowledge. Here, in the non-initial training phase, the initial knowledge graph can refer to the latest version of the knowledge graph. In the initial training phase, the initial knowledge graph can be obtained by collecting a large amount of drone knowledge. Its construction steps may include: collecting and integrating more than 100,000 drone component relationship data and data covering dozens of failure modes, storing them using graph database technology to construct the initial knowledge graph for the chef training phase.

[0026] After obtaining the initial knowledge graph, the current UAV knowledge graph can be obtained by acquiring historical actual operation data and defining entities and relationships such as "fault-cause-parameter" and "maintenance plan-applicable fault-required tools" in the historical actual operation data.

[0027] Here, historical operational data is multi-source and heterogeneous, and may include, but is not limited to: historical maintenance records, flight control logs, sensor data, manufacturer maintenance manuals, industry airworthiness standards, and typical fault case libraries. Sensor data may include motor vibration frequencies and battery temperature profiles. It is understood that the current operational data of trainees in each training round within a historical time period can be used as historical operational data.

[0028] Step 120: Inject the key knowledge from the UAV knowledge graph into the UAV digital twin to obtain a fused twin.

[0029] Here, key knowledge refers to the core information extracted from the UAV knowledge graph that is directly related to a specific training task, such as fault diagnosis or emergency response. It's important to understand that key knowledge is not the entirety of the knowledge graph, but rather a subset selected based on the training objectives, such as the causal chain of a specific fault, standard operating procedures, fault evolution paths, and related maintenance cases. Its purpose is to empower the digital twin, enabling it to simulate complex, knowledge-driven scenarios.

[0030] In addition, the fused twin here refers to the enhanced virtual model obtained by injecting key knowledge into the UAV digital twin. Thus, the fused twin is no longer a simple model that can only perform flight simulation, but a training agent that can dynamically reproduce specific faults, intelligently guide trainees' operations, and understand trainees' operational intentions.

[0031] Specifically, firstly, key knowledge relevant to the current training subject is extracted from the UAV knowledge graph. For example, if the training subject is handling motor overheating faults, the extracted key knowledge includes possible causes of motor overheating, associated sensor parameters, corresponding standard operating procedures, and handling reports of similar historical cases. Then, this key knowledge can be injected into the simulation engine of the UAV digital twin through data interfaces or predefined protocols. For example, the fault logic "IMU module temperature drift causing attitude angle error" from the knowledge graph can be injected into the twin, enabling it to dynamically simulate abnormal attitude data during simulation. Similarly, maintenance operation specifications defined in the knowledge graph, such as "the magnetic ring clearance tolerance is controlled within ±0.05mm when replacing a brushless motor," can be injected into the twin as rules for subsequent assessment of the trainee's operational accuracy.

[0032] It should be noted that by integrating key knowledge from the knowledge graph with the digital twin of the drone, an intelligent, dynamic training environment that conforms to real operation and maintenance logic is created. This allows trainees to not only perform routine flight operations, but also to conduct safe and frequent drills in near-real high-risk or extreme scenarios. This significantly improves trainees' emergency response capabilities and practical experience, and solves the problem of insufficient practical training in high-risk scenarios in traditional training.

[0033] Step 130: Generate a training report for the trainee based on the current actual operation data performed by the trainee on the fused twin.

[0034] Here, "current operational data" refers to the real-time recorded behavioral data stream generated by the interaction between the trainee and the fused twin during the training process. This can include flight control data, such as stick inputs and flight trajectory stability; fault diagnosis data, such as the use of a virtual multimeter and the order in which parameters are viewed; maintenance operation data, such as tool selection, component disassembly and assembly sequence, and screw torque deviation; and operation time, etc. It is understandable that current operational data can provide objective and detailed raw evidence for assessing trainees' abilities. Furthermore, it can serve as historical operational data for updating the knowledge graph later. In addition, the training report is a structured assessment result generated after comprehensive analysis of the current operational data, aiming to comprehensively and quantitatively reflect the trainee's knowledge mastery and practical skills level.

[0035] Specifically, during training, every step of the trainee's operation on the fused digital twin can be captured, forming current operational data. For example, when a trainee performs a "battery thermal runaway handling" drill, the time, sequence, and key parameters, such as battery temperature changes, can be recorded from the detection of the alarm to the execution of landing and power-off. After training, the current operational data can be compared with the Standard Operating Procedures (SOPs) and expert experience stored in the knowledge graph. For example, the knowledge graph automatically verifies whether the trainee's operation steps conform to the SOP specifications, and the digital twin records the deviation between the trainee's screw tightening torque value and the standard value. Finally, all analysis results can be integrated to generate a visualized training report. It should be noted that the training report may include: a capability radar chart reflecting the trainee's comprehensive abilities, a topology map indicating the trainee's knowledge gaps, statistical data on the success rate of simulated tasks and the timeliness of risk handling, and corrective tips for erroneous operations.

[0036] The method provided in this invention constructs a knowledge-driven, high-fidelity training environment by acquiring and integrating a UAV digital twin and a knowledge graph. Based on the trainees' practical data, it generates a refined competency assessment report, forming a closed-loop training system of experience accumulation, simulation verification, practical optimization, and experience iteration. This effectively solves many pain points in traditional training models, significantly improves training efficiency, ensures training safety, and enables quantitative assessment and continuous improvement of trainees' practical abilities.

[0037] Based on any of the above embodiments, the key knowledge includes fault knowledge and / or component structure knowledge; Step 120 includes: The fault knowledge is injected into the UAV digital twin to obtain a fault twin; The structural knowledge of the components is injected into the digital twin of the UAV to obtain a structural display twin; The fault twin and / or the structural display twin are used as the fused twin.

[0038] Here, fault knowledge refers to the structured information extracted from the UAV knowledge graph used to describe and reproduce UAV faults. Specifically, it may include, but is not limited to: the characterization of fault phenomena, such as abnormal fuselage shaking or loss of image transmission signals; the underlying mechanisms and causes of faults, such as propeller dynamic balance failure or image transmission module overheating; fault evolution paths, such as abnormal battery voltage drop ultimately leading to motor shutdown; and fault-related diagnostic logic and standard handling procedures. Fault knowledge can inject faults into digital twins, enabling them to simulate various fault scenarios that may occur in the real world.

[0039] Component structural knowledge refers to the set of information extracted from the UAV knowledge graph that describes the physical structure and assembly relationships of the UAV. Specifically, it may include, but is not limited to: the UAV's component decomposition tree and its hierarchical relationships, the geometric constraints and connection relationships of each component, standard disassembly and assembly procedures, and the specifications and torque parameters of the tools required for operation. Therefore, component structural knowledge can support in-depth structural cognition and virtual maintenance operation training.

[0040] Specifically, based on a pre-defined training subject, such as "IMU temperature drift fault diagnosis," corresponding fault knowledge can be retrieved from the UAV knowledge graph. It should be noted that this fault knowledge may contain a causal chain: "sudden drop in ambient temperature -> IMU heating module failure -> IMU internal temperature below normal operating range -> gyroscope zero-point drift -> attitude calculation error -> UAV horizontal drift." This knowledge chain, representing fault knowledge, can be transformed into simulation parameters and logical rules, and injected into the dynamics and sensor models of the UAV digital twin. After injection, the twin becomes a fault twin. At the start of training, this fault twin will simulate the phenomenon of horizontal drift. When trainees view IMU temperature data through a virtual terminal, they will discover temperature anomalies, thus guiding them to perform correct fault diagnosis.

[0041] Furthermore, the structural knowledge of the components is injected into the digital twin of the drone to obtain a structural display twin. Specifically, firstly, based on a preset training subject, such as "replacing the gimbal shock absorber ball," the structural knowledge of the gimbal components is extracted from the knowledge graph. It should be noted that the structural knowledge of the gimbal components can include the assembly relationships, screw types, and disassembly sequence between the gimbal, shock absorber plate, and shock absorber ball. This structural knowledge can then be bound to the 3D model of the drone's digital twin. After injection, the twin becomes a structural display twin. Trainees can select a disassembly mode on this twin, where the model will demonstrate the standard disassembly steps in animated form; trainees can also perform virtual operations themselves. If an incorrect screwdriver type is selected, prompts can be provided based on the injected component structural knowledge, thereby training trainees' practical skills and operational standards.

[0042] Ultimately, different types of fusion twins can be flexibly provided to trainees based on task requirements. When learning fault diagnosis theory, only a fault twin can be provided; when learning component recognition or maintenance procedures, only a structural display twin can be provided; and in a complete comprehensive training task "from fault diagnosis to component replacement", the two can be combined. Trainees first complete the diagnosis on the fault twin, and then the system switches to the structural display twin mode for trainees to complete subsequent virtual maintenance operations.

[0043] The method provided in this invention refines key knowledge into fault knowledge and component structure knowledge, and correspondingly divides the fused twin into fault twin and structure display twin, thereby achieving modularization and specialization of training functions. This allows training content to more accurately align with specific job skill requirements. Whether it is cognitive training that focuses on cultivating logical diagnostic abilities or skills-based training that focuses on cultivating standardized operating abilities, a highly matched virtual training environment can be obtained, thus significantly improving the relevance and effectiveness of training.

[0044] Based on any of the above embodiments, the steps for constructing the UAV knowledge graph include: Obtain the historical operational data and industry rules; Static entity relationships are extracted from the historical actual operation data, and a static knowledge graph is constructed based on the static entity relationships. Based on the industry rules, knowledge reasoning is performed on the static knowledge graph to construct the UAV knowledge graph.

[0045] Based on any of the above embodiments, the steps for updating the UAV knowledge graph include: Acquire the initial knowledge graph, the historical operational data, and industry rules; Static entity relationships are extracted from the historical actual operation data, and the initial knowledge graph is updated based on the static entity relationships to obtain a static knowledge graph; Based on the industry rules, knowledge reasoning is performed on the static knowledge graph to obtain the UAV knowledge graph.

[0046] Specifically, the first step is to acquire the initial knowledge graph, historical operational data, and industry rules.

[0047] Here, industry rules refer to a set of predefined, universally applicable logical constraints, expert experience, or industry standards. For example, it could be a physical rule like "if the motor current exceeds the rated value by 20% for 5 consecutive seconds, it can be considered an overload state," a standard operating procedure rule like "compass calibration must be performed after replacing the flight control module," or an expert experience rule like "the probability of GPS satellite loss increases significantly in environments with strong electromagnetic interference." Understandably, industry rules can be used to logically complete and verify the consistency of the data map in subsequent steps.

[0048] In detail, an initial knowledge graph can be loaded from a local database or file, historical actual operation data can be aggregated from multiple sources such as operation and maintenance log system and maintenance work order system through data interface, and industry rules pre-written by domain experts and stored in the rule base can be loaded.

[0049] The second step is to extract static entity relationships from historical operational data, and update the initial knowledge graph based on these static entity relationships to obtain a static knowledge graph.

[0050] Here, static entity relationships refer to objective factual relationships that can be directly extracted from historical operational data without complex reasoning. These relationships are relatively fixed. For example, extracting the event "Maintenance worker Zhang San replaced motor number 3 of drone SN001 on October 23, 2025" from a maintenance work order includes multiple static entity relationships such as (Zhang San, executes, replacement operation), (replacement operation, acts on, motor number 3), and (motor number 3, belongs to, drone SN001). It can be understood that the extracted static entity relationships can transform unstructured or semi-structured raw data into structured graph knowledge. Furthermore, the static knowledge graph here refers to the intermediate product obtained after integrating these extracted static entity relationships into the initial knowledge graph. It contains richer and more realistic operational facts than the initial knowledge graph, but may still lack deep logical connections.

[0051] In detail, knowledge extraction techniques can be used to process historical operational data. For unstructured data, such as maintenance report text, natural language processing techniques can be used to identify entities such as "fault phenomenon," "part name," and "operation action" through named entity recognition technology, and then relationship extraction techniques can be used to identify the relationships between them. For structured data, such as flight control logs, scripts can be written to extract, transform, and load log entries into events and relationships in the knowledge graph. For example, a log entry of "Motor 3 Current Spike" can be transformed into a relation triple such as (Motor 3, Occurrence, Current Spike). It should be noted that all newly extracted entities and relationships are added to the initial knowledge graph. If an entity or relationship already exists, its attributes may be updated, ultimately forming a larger and more detailed static knowledge graph.

[0052] It should be noted that this step, through the automated processing and transformation of historical operational data, greatly enriches the content of the knowledge graph, solidifies the practical operational experience of trainees and real events that occurred on the equipment into structured knowledge, and significantly improves the coverage and authenticity of the knowledge graph.

[0053] The third step involves performing knowledge reasoning on the static knowledge graph based on industry rules to obtain the UAV knowledge graph.

[0054] Here, knowledge reasoning refers to the process of automatically discovering implicit knowledge, performing logical verification, and generating new knowledge based on existing facts in a static knowledge graph, using predefined industry rules. Its role is to allow the knowledge graph to evolve from simply knowing what happened to understanding why it happened and the deep connections between them, thereby endowing the knowledge graph with a certain degree of reasoning ability.

[0055] Specifically, industry rules can be applied to static knowledge graphs using a knowledge reasoning engine. For example, a static knowledge graph might contain the facts: (Event A, occurs during a high-voltage line inspection mission) and (Event A, includes GPS signal loss). After applying the industry rule "In environments with strong electromagnetic interference, the probability of GPS signal loss increases significantly," the reasoning engine can automatically infer a new piece of knowledge: (Event A, the inferred cause is electromagnetic interference). As another example, a static knowledge graph might contain the facts: (Component X, replacement cycle, 6 months) and (Component X, last replacement date, March 1, 2025). After applying time calculation rules, the reasoning engine can infer a new piece of knowledge: (Component X, current status, replacement recommended). It should be noted that this inferred new knowledge will be added back to the graph, ultimately forming a complete and logically consistent UAV knowledge graph for use in training methods.

[0056] This step introduces industry-rule-based logical reasoning to achieve deep reasoning of knowledge, making the knowledge graph not just a collection of facts, but an intelligent system capable of analysis, diagnosis, and prediction.

[0057] Therefore, the knowledge graph construction method provided in this embodiment of the invention systematically solves the problem of constructing a high-quality UAV operation and maintenance knowledge graph through a three-stage process of basic knowledge graph construction, filling in real historical operation data, and logical rule reasoning based on industry rules. This ensures that the knowledge injected into the digital twin is accurate, comprehensive, and has deep logic, providing core knowledge assurance for achieving efficient and intelligent training.

[0058] Based on any of the above embodiments, and based on the industry rules, knowledge reasoning is performed on the static knowledge graph to construct the UAV knowledge graph, including: Event data is extracted from the static knowledge graph; Retrieve the target industry rules that match the event data from the industry rules; Based on the target industry rules, knowledge reasoning is performed on the event data to construct the UAV knowledge graph.

[0059] Here, event data refers to dynamic information units or combinations of information extracted from static knowledge graphs that can characterize changes in the state of a drone or its operation and maintenance.

[0060] Specifically, firstly, a static knowledge graph can be traversed by executing pre-defined query statements, such as SPARQL queries, to identify and extract event data that conforms to specific patterns. For example, a query task can be set up to extract all nodes and their associated information where "sensor readings have exceeded their normal thresholds three times consecutively within the past 24 hours," forming a set of event data. Alternatively, event data consisting of a sequence of events can be extracted, such as "a drone experienced a transmission module failure immediately after completing a flight mission in a high-humidity environment."

[0061] Then, the inference engine can use an efficient pattern matching algorithm to compare the event data with the conditional parts of all rules in the rule base. For example, if the event data extracted in the previous step is "the current value of motor A is 25A", and there is a rule in the rule base that is "if the current value of a certain motor is >20A", then this rule will be successfully matched and become the target industry rule for this inference. If the event data is more complex, such as an event sequence, the inference engine will search for complex event processing rules that can match the pattern of that sequence.

[0062] It should be noted that a precise rule matching mechanism bridges the gap between general knowledge and specific application scenarios, ensuring that subsequent reasoning is based on correct logic and expert experience, thus guaranteeing the reliability and authority of the reasoning results.

[0063] Finally, after the inference engine matches the target industry rule, the conclusion of that rule can be executed. It should be noted that executing the conclusion usually means creating new knowledge in the knowledge graph, i.e., adding new entity nodes or new relation edges. For example, for the target industry rule "If the current value of a motor is >20A, then the motor state is 'overload'", the inference engine can add a new relation to the entity node "Motor A" in the graph, pointing to a node representing the "overload" state, thus generating a new triple (Motor A, State, Overload) in the graph.

[0064] It should also be noted that if the conclusion of the target industry rule is more complex, such as "generate a high-risk warning and associate it with the motor and the corresponding flight mission", the warning node can be created in the graph and multiple associations can be established accordingly through the inference engine.

[0065] Understandably, the process of reasoning about event data based on target industry rules is iterative. Newly generated knowledge, such as the "overload" state, may serve as new event data to trigger other industry rules, such as "if a motor is in the 'overload' state, the recommended maintenance operation is 'check the bearing'", thus forming a reasoning chain and continuously enriching and deepening the knowledge graph.

[0066] The method provided in this invention not only improves the efficiency and intelligence of knowledge graph construction by procedurally and automatically integrating expert experience and industry standards into the knowledge graph, but more importantly, it ensures that the final generated knowledge graph can contain a large number of deep causal relationships, fault diagnosis conclusions and decision suggestions derived from basic data, so as to provide theoretical knowledge support for subsequent high-fidelity fault simulation and intelligent training and evaluation.

[0067] Based on any of the above embodiments, step 130 includes: Reference operation data for the current actual operation data is obtained from the UAV knowledge graph; The training report is generated by comparing the reference operation data with the current actual operation data. The reference operation data includes fault reference handling data and / or component reference maintenance data.

[0068] Here, reference operation data refers to the benchmark dataset stored in the UAV knowledge graph, used to evaluate the correctness and quality of trainees' operations. It is typically derived from industry standard operating procedures, expert historical best practices, or a large amount of successful case data, providing a quantifiable benchmark for trainees' current operational data. Depending on the training task, reference operation data can be further subdivided into fault handling reference data and / or component maintenance reference data.

[0069] Specifically, before the start of a training task or during the evaluation phase, a targeted query can be initiated into the UAV knowledge graph based on the nature of the current training task. For example, if the training task is "troubleshooting image transmission loss," the query will retrieve reference operational data associated with the task. If the task is troubleshooting, the corresponding fault reference handling data can be obtained; if the task is "replacing the gimbal motor," the corresponding component reference maintenance data can be obtained. For example, for the "replacing the gimbal motor" task, the reference operational data obtained from the knowledge graph might be a set of serialized data containing the following information: {Step 1: Select a T5 screwdriver; Step 2: Remove the 4 fixing screws counterclockwise; ...; Step N: Calibrate the gimbal's horizontal angle}.

[0070] Furthermore, the training report can be compiled by comparing the trainee's current actual operation data captured in real time with the reference operation data obtained in the previous step. This comparison can be performed through at least one of the following methods: step sequence comparison, key parameter comparison, and logical path comparison.

[0071] For step sequence alignment, sequence alignment algorithms, such as dynamic time warping, can be used to align the student's operation step sequence with the standard sequence of reference operation data, thereby identifying redundant operations, missing operations, or operations with incorrect order.

[0072] For key parameter comparison, the deviation can be calculated by extracting the parameter values ​​of trainees at key operation nodes, such as the torque for tightening screws, the time taken to diagnose faults, and the stability parameters of flight attitude, and comparing them with the standard values ​​or standard ranges in the reference data.

[0073] In addition, for logical path comparison, in the fault diagnosis task, the logical path of the trainee viewing data and making judgments can be compared with the optimal diagnosis path stored in the knowledge graph to analyze the efficiency and accuracy of the diagnosis approach.

[0074] It should be noted that after the comparison is completed, the results can be quantified by integrating deviation and matching information from all dimensions, and a structured training report can be automatically generated. This report may include: operational compliance score, comparison with standard operation time, a detailed list of incorrect operation steps and corrective suggestions, knowledge weaknesses identified through the trainee's operation path, and a comprehensive capability radar chart that intuitively displays the trainee's performance in multiple dimensions such as "operational standardization", "diagnostic efficiency", and "emergency response".

[0075] The method provided in this invention generates a training report by comparing reference operation data obtained from the UAV knowledge graph with the current actual operation data. This allows trainees to clearly recognize their shortcomings and areas for improvement, and also provides training managers with a reliable basis for optimizing courses and providing targeted guidance, thereby constructing a complete training closed loop of learning, practice, evaluation, and improvement.

[0076] Based on any of the above embodiments, the training report is generated by comparing the reference operation data and the current actual operation data, including: By comparing the reference operating data with the current actual operating data, the deviation result is obtained; By reverse querying the knowledge points related to the deviation results from the UAV knowledge graph, the knowledge weaknesses can be identified. Based on the knowledge gaps and the UAV knowledge graph, a knowledge gap topology graph is constructed. The training report is generated based on the aforementioned knowledge gap topology map.

[0077] Specifically, firstly, the reference operation data and the current actual operation data are compared to obtain the deviation results. After comparing the reference operation data and the current actual operation data, each identified deviation, such as erroneous operation or omitted step, can be used to obtain the deviation results.

[0078] Next, the deviation results are correlated with the UAV knowledge graph. That is, by comparing the results, a reverse query can be performed in the UAV knowledge graph to identify the knowledge points in the reference operation data that support the correct operation. For example, the reference operation data requires "immediately check the cooling duct after the motor overheat alarm." The knowledge graph contains the relationship: (motor overheating) - [one of the main causes] -> (cooling duct blockage). If the trainee misses this step, the knowledge relationship "(motor overheating) - [one of the main causes] -> (cooling duct blockage)" is marked as a knowledge weakness for that trainee.

[0079] Understandably, by performing such attribution analysis on all operational deviations, a set of knowledge weaknesses consisting of multiple knowledge graph entities or relationships is ultimately obtained.

[0080] Then, based on the knowledge weaknesses and the UAV knowledge graph, a knowledge weakness topology graph can be constructed. For example, the set of knowledge weaknesses can be used as input, and these knowledge weaknesses can be used as the core to traverse and expand the complete UAV knowledge graph, extracting the local network structure containing these points and edges, thereby constructing the knowledge weakness topology graph.

[0081] During the construction process, the graph can be enhanced. For example, it can be weighted according to the number of times a knowledge gap is triggered and marked on the graph with different colors or node sizes. Frequently occurring gaps represent the learner's core weaknesses. For instance, if a learner's weaknesses include "not understanding ESC response delay" and "not understanding the impact of PID parameters on response," the topology graph will clearly show that both knowledge points point to the higher-order concept node of "flight control stability," thus revealing that the learner has systemic problems in the entire flight control parameter tuning knowledge domain.

[0082] Finally, the knowledge gap topology map is embedded as the core content into the training report, and analytical text is automatically generated around the map.

[0083] The method provided in this invention obtains knowledge weaknesses by reverse querying knowledge points related to the deviation results from the UAV knowledge graph; based on the knowledge weaknesses and the UAV knowledge graph, a knowledge weakness topology graph is constructed. This allows for the accurate diagnosis of which knowledge structure the trainee lacks, based on the ability to evaluate trainee operational errors. This achieves a shift from standardized assessment to precise empowerment, greatly improving the relevance and effectiveness of training.

[0084] Based on any of the above embodiments, generating the training report based on the knowledge gap topology map further includes: Acquire a 3D scene; Acquire the flight operation data of the trainee on the digital twin of the UAV in the three-dimensional scene; Based on the flight operation data, a capability radar map is generated; The training report is generated based on the knowledge gap topology map and the capability radar map.

[0085] Specifically, firstly, a 3D scene is acquired, which can be pre-constructed. For example, a specific 3D scene can be loaded from a scene library or generated programmatically based on a preset training subject. For instance, if the training subject is "maintaining flight path under complex weather conditions," the server will construct a mountainous scene including random crosswinds and updrafts; if the subject is "manual return to base under GPS denial," it will construct a city center scene with tall buildings and blocked GPS signals. These scenes are all built based on a physics engine, which can realistically simulate aerodynamic effects and the impact of the environment on sensors.

[0086] Then, when the trainee begins to control the drone digital twin to perform tasks in the constructed 3D scene, the simulation engine will record all the above-mentioned flight operation data in real time at a high frequency and transmit it to the training system's backend server for storage and analysis. It should be noted that the flight operation data here is a specific subset of the actual operation data, specifically referring to a series of data streams directly related to flight control generated when the trainee controls the drone digital twin in the 3D scene.

[0087] Next, the acquired flight operation data can be calculated and processed, transforming the raw data into scores for various capability dimensions. For example, the "course accuracy" score can be obtained by calculating the root mean square error between the actual flight trajectory and the preset course; the "attitude stability" score can be obtained by analyzing the variance of pitch and roll angular velocities during flight; and the "energy management efficiency" score is assessed based on the total power consumed to complete the mission. These standardized scores are then plotted on the corresponding axes of a radar chart, forming a closed polygon. The shape and area of ​​this polygon visually reflect the trainee's overall flight skill level and the balance of their various capabilities.

[0088] Finally, the knowledge gap topology map and the capability radar map generated in this embodiment can be integrated and used together as the core content of the final training report.

[0089] Based on any of the above embodiments, generating a capability radar chart based on the flight operation data includes: Based on the flight operation data, the flight stability score and flight energy efficiency are evaluated. The capability radar map is generated based on the flight stability score and flight energy efficiency.

[0090] Specifically, statistical analysis can be performed on attitude angular velocities in flight operation data, such as pitch and roll angular velocities, as well as linear acceleration data, for example, calculating their standard deviation or variance throughout the flight. A smaller variance indicates smoother attitude and velocity changes, which can be mapped to a higher stability score. Furthermore, the remote controller's stick input data can be analyzed to deduct points for high-frequency, large-amplitude operations.

[0091] In addition, to assess flight energy efficiency, the total power consumed to complete the mission can first be obtained from the battery model of the UAV's digital twin, for example, by integrating current and voltage. Then, combined with mission completion data, such as total flight distance or number of work sites covered, an energy consumption value per unit of mission workload can be calculated. The lower this value, the higher the flight energy efficiency score.

[0092] It should be noted that by reducing the complexity and continuous flight operation data to two core evaluation indicators with clear physical meaning, key quantitative inputs are provided for the subsequent generation of intuitive and easy-to-understand capability radar charts, making the evaluation process more focused and standardized.

[0093] Furthermore, the flight stability score and flight energy efficiency obtained from the previous assessment are used as values ​​for two important axes of the capability radar chart. For example, the capability radar chart can include five axes: flight stability, energy efficiency, flight path accuracy, obstacle avoidance, and emergency response. The system then plots the calculated flight stability score and flight energy efficiency score on their respective axes and connects them with the scores from other dimensions to ultimately form a polygon radar chart that comprehensively reflects the trainee's flight control capabilities.

[0094] Based on any of the above embodiments Figure 2 This is a schematic diagram of the drone maintenance personnel training device based on knowledge graph and digital twin technology provided by the present invention, as shown below. Figure 2 As shown, the device includes: Unit 210 is used to acquire the UAV digital twin and the UAV knowledge graph; The fusion unit 220 injects key knowledge from the UAV knowledge graph into the UAV digital twin to obtain a fused twin; The report generation unit 230 generates a training report for the trainee based on the current actual operation data performed by the trainee on the fused twin. The drone knowledge graph is updated based on historical operational data.

[0095] The device provided in this invention constructs a knowledge-driven, high-fidelity training environment by acquiring and fusing a UAV digital twin with a knowledge graph. It generates a refined competency assessment report based on trainees' practical data, forming a closed-loop training system of experience accumulation, simulation verification, practical optimization, and experience iteration. This effectively solves many pain points in traditional training models, significantly improves training efficiency, ensures training safety, and enables quantitative assessment and continuous improvement of trainees' practical abilities.

[0096] Based on any of the above embodiments, the key knowledge includes fault knowledge and / or component structure knowledge; The fusion unit is specifically used for: The fault knowledge is injected into the UAV digital twin to obtain a fault twin; The structural knowledge of the components is injected into the digital twin of the UAV to obtain a structural display twin; The fault twin and / or the structural display twin are used as the fused twin.

[0097] Based on any of the above embodiments, the device further includes a map updating unit, which is specifically used for: Acquire the initial knowledge graph, the historical operational data, and industry rules; Static entity relationships are extracted from the historical actual operation data, and the initial knowledge graph is updated based on the static entity relationships to obtain a static knowledge graph; Based on the industry rules, knowledge reasoning is performed on the static knowledge graph to obtain the UAV knowledge graph.

[0098] Based on any of the above embodiments, the map updating unit is further specifically used for: Event data is extracted from the static knowledge graph; Retrieve the target industry rules that match the event data from the industry rules; Based on the target industry rules, knowledge reasoning is performed on the event data to construct the UAV knowledge graph.

[0099] Based on any of the above embodiments, the report generation unit is specifically used for: Reference operation data for the current actual operation data is obtained from the UAV knowledge graph; The training report is generated by comparing the reference operation data with the current actual operation data. The reference operation data includes fault reference handling data and / or component reference maintenance data.

[0100] Based on any of the above embodiments, the report generation unit is further specifically used for: By comparing the reference operating data with the current actual operating data, the deviation result is obtained; By reverse querying the knowledge points related to the deviation results from the UAV knowledge graph, the knowledge weaknesses can be identified. Based on the knowledge gaps and the UAV knowledge graph, a knowledge gap topology graph is constructed. The training report is generated based on the aforementioned knowledge gap topology map.

[0101] Based on any of the above embodiments, the report generation unit is further specifically used for: Acquire a 3D scene; Acquire the flight operation data of the trainee on the digital twin of the UAV in the three-dimensional scene; Based on the flight operation data, a capability radar map is generated; The training report is generated based on the knowledge gap topology map and the capability radar map.

[0102] Based on any of the above embodiments, the report generation unit is further specifically used for: Based on the flight operation data, the flight stability score and flight energy efficiency are evaluated. The capability radar map is generated based on the flight stability score and flight energy efficiency.

[0103] In one embodiment, Figure 3 This is a schematic diagram of the architecture of the drone maintenance personnel training system based on knowledge graph and digital twin technology provided by the present invention, such as... Figure 3 As shown, the system architecture includes: a digital twin simulation platform construction module, a knowledge graph construction module, an integrated application module, and key technology support.

[0104] The knowledge graph construction module integrates multi-source data and performs dynamic knowledge reasoning on the integrated data based on industry rules to construct a drone knowledge graph.

[0105] In addition, the digital twin simulation platform building module includes 3D scene and equipment modeling. Specifically, it is used to build a 3D simulation scene library for scenarios such as high-voltage line environments and mountainous terrain, and to construct high-fidelity digital twins of equipment based on sensor data from physical drones, such as motor speed and battery voltage.

[0106] In addition, the knowledge graph and digital twin fusion application module is used to feed back into the knowledge graph to optimize decision-making logic based on simulation operation data, such as maintenance time and tool selection. At the same time, the knowledge graph rule base constrains the fault injection scope of the digital twin system.

[0107] Finally, the key technology support modules mainly include edge computing nodes running lightweight knowledge graph engines, such as Apache AGE, and WebGL to achieve twin scene rendering on the browser side, reducing reliance on VR devices. Additionally, an NLP engine parses student voice questions, such as "How to calibrate gimbal jitter," and uses the knowledge graph for precise retrieval; reinforcement learning simulates fault evolution paths and dynamically adjusts training difficulty.

[0108] Figure 4 An example is a schematic diagram of the physical structure of an electronic device, such as... Figure 4 As shown, the electronic device may include a processor 410, a communication interface 420, a memory 430, and a communication bus 440, wherein the processor 410, the communication interface 420, and the memory 430 communicate with each other through the communication bus 440. The processor 410 can call logical instructions in the memory 430 to execute a training method for UAV operation and maintenance personnel based on knowledge graph and digital twin technology. The method includes: acquiring a UAV digital twin and a UAV knowledge graph; injecting key knowledge from the UAV knowledge graph into the UAV digital twin to obtain a fused twin; generating a training report for the trainee based on the current actual operation data performed by the trainee on the fused twin; and updating the UAV knowledge graph based on historical actual operation data.

[0109] Furthermore, the logical instructions in the aforementioned memory 430 can be implemented as software functional units and, when sold or used as independent products, can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present invention, or the part that contributes to the prior art, or a part of the 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 to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute all or part of the steps of the methods described in the various embodiments of the present invention. The aforementioned storage medium includes various media capable of storing program code, such as USB flash drives, portable hard drives, read-only memory (ROM), random access memory (RAM), magnetic disks, or optical disks.

[0110] On the other hand, the present invention also provides a computer program product, which includes a computer program that can be stored on a non-transitory computer-readable storage medium. When the computer program is executed by a processor, the computer can execute the UAV operation and maintenance personnel training method based on knowledge graph and digital twin technology provided by the above methods. The method includes: acquiring a UAV digital twin and a UAV knowledge graph; injecting key knowledge from the UAV knowledge graph into the UAV digital twin to obtain a fused twin; generating a training report for the trainee based on the current actual operation data performed by the trainee on the fused twin; and updating the UAV knowledge graph based on historical actual operation data.

[0111] In another aspect, the present invention also provides a non-transitory computer-readable storage medium storing a computer program thereon. When executed by a processor, the computer program implements a method for training UAV maintenance personnel based on knowledge graph and digital twin technology, as provided by the methods described above. The method includes: acquiring a UAV digital twin and a UAV knowledge graph; injecting key knowledge from the UAV knowledge graph into the UAV digital twin to obtain a fused twin; generating a training report for the trainee based on the current actual operation data performed by the trainee on the fused twin; and updating the UAV knowledge graph based on historical actual operation data.

[0112] The device embodiments described above are merely illustrative. The units described as separate components may or may not be physically separate. The components shown as units may or may not be physical units; that is, they may be located in one place or distributed across multiple network units. Some or all of the modules can be selected to achieve the purpose of this embodiment according to actual needs. Those skilled in the art can understand and implement this without any creative effort.

[0113] Through the above description of the embodiments, those skilled in the art can clearly understand that each embodiment can be implemented by means of software plus necessary general-purpose hardware platforms, and of course, it can also be implemented by hardware. Based on this understanding, the above technical solutions, in essence or the part that contributes to the prior art, can be embodied in the form of a software product. This computer software product can be stored in a computer-readable storage medium, such as ROM / RAM, magnetic disk, optical disk, etc., and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute the methods described in the various embodiments or some parts of the embodiments.

[0114] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, and not to limit them; although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some of the technical features; and these modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of the present invention.

Claims

1. A training method for drone operation and maintenance personnel based on knowledge graph and digital twin technology, characterized in that, include: Obtain digital twins and knowledge graphs of drones; Key knowledge from the UAV knowledge graph is injected into the UAV digital twin to obtain a fused twin; Based on the actual operational data of the trainee performing on the fused twin, a training report for the trainee is generated; The drone knowledge graph is updated based on historical operational data.

2. The method for training UAV maintenance personnel based on knowledge graph and digital twin technology according to claim 1, characterized in that, The key knowledge includes fault knowledge and / or component structure knowledge; The step of injecting key knowledge from the UAV knowledge graph into the UAV digital twin to obtain a fused twin includes: The fault knowledge is injected into the UAV digital twin to obtain a fault twin; The structural knowledge of the components is injected into the digital twin of the UAV to obtain a structural display twin; The fault twin and / or the structural display twin are used as the fused twin.

3. The method for training UAV maintenance personnel based on knowledge graph and digital twin technology according to claim 1, characterized in that, The steps for updating the drone knowledge graph include: Acquire the initial knowledge graph, the historical operational data, and industry rules; Static entity relationships are extracted from the historical actual operation data, and the initial knowledge graph is updated based on the static entity relationships to obtain a static knowledge graph; Based on the industry rules, knowledge reasoning is performed on the static knowledge graph to obtain the UAV knowledge graph.

4. The method of claim 3, wherein the method further comprises: The process of constructing the UAV knowledge graph by performing knowledge reasoning on the static knowledge graph based on the industry rules includes: Event data is extracted from the static knowledge graph; Retrieve the target industry rules that match the event data from the industry rules; Based on the target industry rules, knowledge reasoning is performed on the event data to construct the UAV knowledge graph.

5. The method of claim 1 to 4, wherein, The step of generating a training report for the trainee based on the current actual operation data performed by the trainee on the fused twin includes: Reference operation data for the current actual operation data is obtained from the UAV knowledge graph; The training report is generated by comparing the reference operation data with the current actual operation data. The reference operation data includes fault reference handling data and / or component reference maintenance data.

6. The method of claim 5, wherein the method further comprises: The process of comparing the reference operation data and the current actual operation data to generate the training report includes: By comparing the reference operating data with the current actual operating data, the deviation result is obtained; By reverse querying the knowledge points related to the deviation results from the UAV knowledge graph, the knowledge weaknesses can be identified. Based on the knowledge gaps and the UAV knowledge graph, a knowledge gap topology graph is constructed. The training report is generated based on the aforementioned knowledge gap topology map.

7. The method of claim 6, wherein the method further comprises: The generation of the training report based on the knowledge gap topology map also includes: Acquire a 3D scene; Acquire the flight operation data of the trainee on the digital twin of the UAV in the three-dimensional scene; Based on the flight operation data, a capability radar map is generated; The training report is generated based on the knowledge gap topology map and the capability radar map.

8. The method of claim 7, wherein the method further comprises: The generation of the capability radar map based on the flight operation data includes: Based on the flight operation data, the flight stability score and flight energy efficiency are evaluated. The capability radar map is generated based on the flight stability score and flight energy efficiency.

9. An unmanned aerial vehicle operator training device based on a knowledge graph and a digital twin technology, characterized in that, include: The acquisition unit acquires the drone's digital twin and drone knowledge graph; The fusion unit injects key knowledge from the UAV knowledge graph into the UAV digital twin to obtain a fused twin; The report generation unit generates a training report for the trainee based on the actual operation data performed by the trainee on the fused twin. The drone knowledge graph is updated based on historical operational data.

10. An electronic device comprising a memory, a processor, and a computer program stored on the memory and executable on the processor, characterized in that, When the processor executes the computer program, it implements the drone operation and maintenance personnel training method based on knowledge graph and digital twin technology as described in any one of claims 1 to 8.

Images