Dynamic updating method for digital test model of aviation equipment
By combining event-driven and data-driven hierarchical collaborative update methods, the problem of dynamic updating of digital test models for aviation equipment was solved, achieving high-fidelity and adaptive model updates, and improving the practical value and reliability of digital testing.
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- BEIHANG UNIV
- Filing Date
- 2026-01-09
- Publication Date
- 2026-04-17
AI Technical Summary
The existing methods for updating digital test models of aviation equipment lack intelligent decision-making mechanisms, making it difficult to achieve high-fidelity, adaptive, and sustainable dynamic updates in complex and dynamic actual operating environments. Moreover, the update process is often lagging or lacks physical interpretability.
A hierarchical collaborative update system combining event-driven and data-driven approaches is adopted. Through finite state machines and deep reinforcement learning, it achieves rapid local response and globally consistent updates, integrates event logic and data patterns, dynamically selects update methods, and constructs a self-optimizing closed-loop mechanism.
It improves the update adaptability and reliability of digital test models for aviation equipment, ensures that the model maintains high fidelity under unknown or sudden factors, takes into account real-time performance and consistency, and enhances the effectiveness and practical value of digital testing.
Smart Images
Figure CN121879871A_ABST
Abstract
Description
Technical Field
[0001] This invention belongs to the fields of electronic engineering and computer science and technology, and specifically relates to a method for dynamically updating digital test models of aviation equipment. Background Technology
[0002] The development, testing, and maintenance of aviation equipment are increasingly reliant on high-fidelity digital testing technologies. By constructing digital test models of aviation equipment, its structure, performance, and multi-scenario mission behavior can be simulated in virtual space, significantly reducing the cost and risk of physical testing, shortening the development cycle, and providing support for predictive maintenance and mission optimization. The fidelity of the digital model directly determines the credibility and engineering application value of the digital test.
[0003] However, aviation equipment undergoes continuous dynamic evolution during actual service, and its state is affected by a variety of complex factors, including controllable operation, environmental loads, performance degradation, and maintenance activities. This causes digital models to gradually deviate from the actual state of the equipment. Currently, the methods for updating digital models mainly have two limitations: one is rule-based or event-driven methods, which can achieve rapid response by utilizing prior knowledge, but rely on manually defined logic and are difficult to adapt to the influence of unknown or sudden factors, lacking flexibility and adaptability; the other is methods that rely entirely on data-driven methods, which can extract patterns from data through machine learning, but lack physical interpretability, are susceptible to data quality interference, and the update process is often lagging, making it difficult to guarantee the real-time performance and overall consistency of the model.
[0004] Current technologies have not effectively addressed the collaborative utilization of prior knowledge and real-time monitoring data, lacking a dynamic update system capable of intelligent decision-making based on actual conditions and integrating dual-driven mechanisms. Furthermore, model updates often employ a "one-size-fits-all" approach, failing to distinguish between the local and global nature of parameter changes, and lacking a closed-loop mechanism for continuous self-optimization. Therefore, achieving high-fidelity, adaptive, and sustainable dynamic updates of digital test models for aerospace equipment in complex and dynamic real-world operating environments has become a key technical challenge for improving the effectiveness and practical value of digital testing. Summary of the Invention
[0005] To address the aforementioned technical issues, this invention provides a dynamic update method for digital test models of aviation equipment. This method effectively integrates event logic and data patterns, forming a hierarchical collaborative update system that combines "local rapid response" with "global consistent update." This ensures that the digital test model of aviation equipment maintains high fidelity throughout its entire lifecycle, thereby enhancing the practical value and reliability of digital testing.
[0006] To achieve the above objectives, the present invention adopts the following technical solution:
[0007] A method for dynamically updating a digital test model of aviation equipment includes:
[0008] Step 1: Analyze and classify the factors that cause the digital test model of aviation equipment to deviate from the actual operating state of aviation equipment, and divide the factors into known events and other potential influencing factors;
[0009] Step 2: Classify the known events into control events, state events, and maintenance events, and use a finite state machine to describe the triggering and state transition logic of the known events, construct a quantitative correlation between the known events and the parameters of the digital experimental model, and realize event-driven updating of the parameters of the digital experimental model;
[0010] Step 3: Based on the variational autoencoder, monitor whether there are still other potential influencing factors after the event is updated. If so, use deep reinforcement learning to update the parameters of the data-driven digital experiment model.
[0011] Step 4: Establish a dynamic update mechanism that dynamically selects event-driven, data-driven, or hybrid update methods based on event knowability and data credibility, and performs update verification and iterative optimization.
[0012] Furthermore, in step 2, control events include flight command adjustments and mission mode switching; state events include structural fatigue, performance degradation, and changes in environmental loads; and maintenance events include component replacement and system upgrades. The digital test model includes an aviation equipment model, a test scenario model, and a test mission model.
[0013] Furthermore, in step 2, the quantitative correlation between known events and digital experimental model parameters is constructed as follows: the quantitative correlation is represented by an influence weight matrix. When a known event occurs, the update amount of model parameters triggered by the known event is calculated based on the category identifier vector of the event and the corresponding weight matrix.
[0014] Furthermore, step 3 specifically includes: using a variational autoencoder to extract and reconstruct features from the physical test data and the output data of the digital test model; setting a threshold for anomaly detection through reconstruction error; when an anomaly is detected, constructing a deep reinforcement learning environment with model parameter correction as the action and the degree of agreement between the model output and the actual data as the reward; and iteratively optimizing the model parameters through the policy gradient method to obtain the update amount of the model parameters.
[0015] Furthermore, in step 4, the update method is dynamically selected based on event knowability and data credibility. Specifically, three decision thresholds are set: high, medium, and low. If event knowability is greater than or equal to the high threshold and data credibility is less than the medium threshold, then the event-driven update method is adopted. If event knowability is less than the low threshold and data credibility is greater than or equal to the medium threshold, then the data-driven update method is adopted. In other cases, a hybrid event- and data-driven update method is adopted.
[0016] Furthermore, when using a hybrid-driven update approach, the event-driven model parameter update quantity and the data-driven model parameter update quantity are calculated in parallel, and the fusion weight is calculated based on the real-time event knowability and data credibility. The two are then weighted and fused to obtain the final model parameter update quantity.
[0017] Furthermore, step 4 also includes a hierarchical update strategy: online real-time fine-tuning of a subset of model parameters that are highly sensitive and change frequently; and monitoring the cumulative model deviation, triggering a process of globally reconstructing the model parameters using the latest data and event history when the deviation exceeds a preset threshold.
[0018] Furthermore, in step 4, performing update verification and iterative optimization specifically includes: after each update, comparing the model output with the latest physical test data and calculating the comprehensive fidelity score; storing the decision information, fusion weights, and fidelity score of this update as empirical data for periodically optimizing the decision logic of the dynamic update mechanism and adjusting the decision threshold.
[0019] In a second aspect, the present invention provides an electronic device, comprising: one or more processors; and a memory for storing one or more programs; wherein, when the one or more programs are executed by the one or more processors, the one or more processors implement the aforementioned method for dynamically updating a digital test model of aviation equipment.
[0020] Thirdly, the present invention provides a computer-readable storage medium storing executable instructions thereon, which, when executed by a processor, enable the processor to implement the aforementioned method for dynamically updating a digital test model of aviation equipment.
[0021] The beneficial effects of this invention are as follows:
[0022] 1. Integrating dual drivers to enhance update adaptability and reliability. By quantitatively assessing event knowability and data credibility, the system can intelligently select event-driven, data-driven, or hybrid-driven update strategies, effectively combining prior rules with real-time data. This overcomes the shortcomings of single-driven modes in terms of flexibility or interpretability, and significantly enhances the model's adaptability to unknown or sudden factors.
[0023] 2. Achieve hierarchical collaborative updates, balancing real-time performance and consistency. A hierarchical strategy combining local online rapid learning and global dynamic reconstruction is adopted. This allows for real-time fine-tuning of high-frequency and sensitive parameters to ensure rapid response, while triggering global reconstruction when the model experiences a systematic shift, ensuring the overall accuracy and stability of the model during long-term operation. Attached Figure Description
[0024] Figure 1 This is a flowchart of a method for dynamically updating a digital test model of aviation equipment according to the present invention. Detailed Implementation
[0025] The present invention will be further described below with reference to the accompanying drawings and embodiments.
[0026] This invention relates to a method for dynamically updating digital test models of aviation equipment, applicable to aviation equipment capable of simultaneous digital and physical testing. One of the key aspects of realizing digital testing of aviation equipment is ensuring the fidelity of the digital test model. Factors leading to model inaccuracies include known events with clearly attributable causes, such as mission changes and maintenance operations, as well as unforeseen influencing factors such as implicit degradation and individual differences. Therefore, how to achieve continuous, reliable, and high-fidelity dynamic updates of aviation equipment digital test models has become an urgent problem to be solved. The method disclosed in this invention fully considers both predictable or observable "known events" and "unknown influencing factors" that are difficult to predict or accurately quantify in advance during model dynamic updates. Through event-driven and data fusion-driven update methods, it effectively integrates prior knowledge logic and real-time data patterns, providing support for high-fidelity digital testing of aviation equipment.
[0027] The flowchart of the present invention is as follows Figure 1 As shown, the specific steps include the following:
[0028] Step 1: Analysis of factors affecting digital testing, including analyzing and classifying the factors that cause the digital testing model of aviation equipment to deviate from the actual operating state of aviation equipment, and classifying the factors into known events and other potential influencing factors;
[0029] For a certain type of passenger aircraft in service, its high-fidelity digital test model is used for predictive maintenance, operational optimization, and modification evaluation. This digital test model includes an aviation equipment model, a test scenario model, and a test mission model. As the aircraft's actual flight hours and flight cycles increase, and as scheduled maintenance, repairs, and modifications are carried out, the digital model gradually deviates from the aircraft's actual state. Therefore, the system analyzes and classifies the factors leading to inaccuracies in the civil aircraft digital model, including known events such as control events, state events, and maintenance events, as well as influencing factors that are difficult to quantify or attribute in advance, such as the slow performance degradation of composite wing skin due to batch variations in manufacturing and the coupling performance drift of avionics systems in complex electromagnetic environments. The analysis results are output as a structured list of influencing factors, serving as the input source for subsequent updates.
[0030] Step 2, Event Classification and Description, includes classifying the known events into control events, state events, and maintenance events, and using a finite state machine to describe the event triggering and state transition logic, constructing a quantitative correlation between events and model parameters, and realizing event-driven model updates; specifically including:
[0031] Step 2.1: Classify the adjustment of flight phase commands and mission mode switching of civil aircraft such as "takeoff", "cruise" and "landing" as control events; structural fatigue, performance degradation and changes in environmental loads as state events, such as "airframe fatigue index exceeds threshold", "engine exhaust temperature baseline drift", "control surface efficiency slightly decreases"; component replacement and system upgrades as maintenance events, such as "engine replacement", "main landing gear actuator repair", "software version upgrade".
[0032] Step 2.2, Event Description; including: modeling the event-driven state transition process using a finite state machine, with the state transition function defined as follows: ,in For a finite set of states of digital test models of aviation equipment, Given a set of known events, when the event... In state The following is triggered and the preset conditions are met. When ), the model state is updated to t represents the current time; the state set S includes "normal cruise", "structural fatigue accumulation", "engine performance degradation" etc.; the event set E includes "execution of high load flight", "fatigue index exceeding threshold", "engine replacement"; if the event "engine replacement" is detected to be triggered under the state "engine performance degradation" and the new engine serial number is confirmed, the state will transition to "engine performance reset".
[0033] Step 2.3: Event-driven model update. Establish a quantitative correlation between events and model parameters (e.g., the "engine replacement" event is associated with parameters such as engine thrust curve, fuel consumption rate, and vibration characteristics in the digital model). This relationship is represented by the influence weight matrix W, which is learned from historical data. When an event (e.g., "engine replacement") occurs... When it occurs, the system automatically queries the event. Category identifier vector And calculate the parameter update amount. This automatically updates the old engine parameters in the digital model to the expected parameters for the new engine.
[0034] Step 3: Anomaly detection based on variational autoencoders, including monitoring whether other potential influencing factors still exist after the event update. If so, deep reinforcement learning is used to update the parameters of the data-driven digital test model for aerospace equipment. This step is used to handle deviations caused by unknown factors, specifically including:
[0035] Step 3.1, Anomaly Detection: Deploy a variational autoencoder and jointly train it using sensor data from the aircraft under normal operating conditions and simulation output data from the digital model at the corresponding time points. In online applications, the variational autoencoder extracts and reconstructs features from the physical test data of the aviation equipment and the output data of the digital test model. Anomaly detection thresholds are set based on the reconstruction error to identify inconsistencies caused by unknown factors.
[0036] Step 3.2: Once an anomaly is detected, the deep reinforcement learning module is activated. Its state is the deviation characteristics between the current digital experimental model output and the actual experimental data; its action is to suggest adjustments to the key parameters of the digital experimental model; and its reward is the improvement in the consistency between the adjusted digital experimental model output and the actual data. The parameter update strategy is iteratively optimized using the policy gradient method to find a set of parameter correction values. .
[0037] Step 4: Event and data fusion-driven model updates, including establishing a dynamic update mechanism that dynamically selects event-driven, data-driven, or hybrid-driven update methods based on event knowability and data reliability, and performs update verification and iterative optimization; specifically including:
[0038] Step (4.1) Integrate and update the decision. The decision module receives two inputs in real time: ① Event awareness For example, there are clear work order records for "engine replacement" events. Its value is determined by the confidence level and historical accuracy of the event detection system; ② Data credibility The decision-making module calculates the update path based on the sensor's health status, data noise level, and uncertainty in anomaly detection. The decision-making module selects the update path at each update moment according to the preset dynamic decision-making logic: (1) Then, an event-driven digital test model update method for aviation equipment is adopted; (2) (3) In other cases, the model update method driven by event and data fusion is adopted. These are adjustable decision thresholds: upper limit, median, and lower limit. Without loss of generality, the system can preset the upper limit, median, and lower limit as follows: The update path is dynamically selected based on the aforementioned logic.
[0039] Step 4.2, Layered Update and Global Reconstruction: For any update amount ΔP, firstly, online rapid micro-learning is performed on the most directly affected parameters at the local level. For highly sensitive and frequently changing parameter subsets, real-time fine-tuning is required. At the same time, the system monitors the cumulative deviation of all historical updates. When the deviation exceeds the threshold, it is determined that a qualitative change has occurred, triggering the global model dynamic update process. For example, if the aircraft has undergone multiple mixed updates and its overall aerodynamic characteristics have undergone systematic shifts, the global reconstruction process will be triggered, which may mean that the core aerodynamic parameters of the entire flight dynamics model need to be recalibrated using long-term accumulated data.
[0040] Step 4.3: Select the appropriate update strategy based on the actual situation:
[0041] When the decision is an event-driven update mode, the system calculates and applies the amount of model parameter change triggered by the known event based on the event-state transition logic in step 2 and the quantification relationship between events and model parameters. If the aircraft has just completed a fully documented engine replacement and the current flight data is stable, the system uses event-driven updates, directly applying knowledge graph-based computation. ;
[0042] When the decision-making process adopts a data-driven update mode, the system activates a parameter correction module based on deep reinforcement learning as the core update engine. Following step 3, it performs global optimization and adjustment of the model parameters to obtain the amount of model parameter change. If the aircraft reports unconfirmed abnormal vibration, but the vibration sensor data is of high quality and the anomaly is significant, the system will use data-driven updates and initiate deep reinforcement learning to find the optimal [system / mechanism]. ;
[0043] When the decision is a hybrid-driven update mode, the system runs two update channels in parallel: event-driven and data-driven, and generates the final update quantity through a weighted fusion mechanism. If the aircraft experiences a known severe turbulence event, the system initiates a hybrid-driven update, performing parallel calculations to update the structural load parameters triggered by the turbulence event. And parameter adjustments driven by data anomalies and according to the formula Calculate and merge weights According to the decision module and Calculated, i.e. .
[0044] Step 4.4, Verification and Iterative Optimization: After each update, the digital experimental model is validated using the latest flight data, and the fidelity score F is calculated. The entire "decision-update-result" data (i.e., decision, fusion weights, and comprehensive fidelity score) is stored as experience data. This experience data is periodically used to retrain the decision logic and optimize the thresholds to improve system intelligence.
[0045] In summary, this invention discloses a dynamic update method for a digital test model of aviation equipment, including a digital test influencing factor analysis module, an influencing event classification and description module, an anomaly detection module based on variational autoencoder, and an event and data fusion-driven update module for the digital test model of aviation equipment. By comprehensively considering various influencing factors, the dynamic fidelity of the digital test model of aviation equipment is improved, laying the foundation for improving the accuracy of digital tests.
[0046] In a second aspect, the present invention provides an electronic device, comprising: one or more processors; and a memory for storing one or more programs; wherein, when the one or more programs are executed by the one or more processors, the one or more processors implement the aforementioned method for dynamically updating a digital test model of aviation equipment.
[0047] Thirdly, the present invention provides a computer-readable storage medium storing executable instructions thereon, which, when executed by a processor, enable the processor to implement the aforementioned method for dynamically updating a digital test model of aviation equipment.
[0048] The specific embodiments described above further illustrate the purpose, technical solution, and beneficial effects of the present invention. It should be understood that the above descriptions are merely specific embodiments of the present invention and are not intended to limit the present invention. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the protection scope of the present invention.
Claims
1. A method for dynamically updating a digital test model of aviation equipment, characterized in that, include: Step 1: Analyze and classify the factors that cause the digital test model of aviation equipment to deviate from the actual operating state of aviation equipment, and divide the factors into known events and other potential influencing factors; Step 2: Classify the known events into control events, state events, and maintenance events, and use a finite state machine to describe the triggering and state transition logic of the known events, construct a quantitative correlation between the known events and the parameters of the digital experimental model, and realize event-driven updating of the parameters of the digital experimental model; Step 3: Based on the variational autoencoder, monitor whether there are still other potential influencing factors after the event is updated. If so, use deep reinforcement learning to update the parameters of the data-driven digital experiment model. Step 4: Establish a dynamic update mechanism that dynamically selects event-driven, data-driven, or hybrid update methods based on event knowability and data credibility, and performs update verification and iterative optimization.
2. The method for dynamically updating a digital test model of aviation equipment according to claim 1, characterized in that, In step 2, control events include flight command adjustments and mission mode switching; state events include structural fatigue, performance degradation, and changes in environmental loads; and maintenance events include component replacement and system upgrades. The digital test model includes an aviation equipment model, a test scenario model, and a test mission model.
3. The method for dynamically updating a digital test model of aviation equipment according to claim 1, characterized in that, In step 2, the quantitative correlation between known events and digital experimental model parameters is constructed as follows: the quantitative correlation is represented by an influence weight matrix. When a known event occurs, the update amount of model parameters triggered by the known event is calculated based on the category identifier vector of the event and the corresponding weight matrix.
4. The method for dynamically updating a digital test model of aviation equipment according to claim 1, characterized in that, Step 3 specifically includes: using a variational autoencoder to extract and reconstruct features from physical experimental data and digital experimental model output data; setting a threshold for anomaly detection based on reconstruction error; when an anomaly is detected, constructing a deep reinforcement learning environment that takes model parameter correction as the action and the degree of agreement between the model output and the actual data as the reward; and iteratively optimizing the model parameters through the policy gradient method to obtain the model parameter update amount.
5. The method for dynamically updating a digital test model of aviation equipment according to claim 1, characterized in that, In step 4, the update method is dynamically selected based on the event knowability and data credibility. Specifically, three decision thresholds are set: high, medium, and low. If the event knowability is greater than or equal to the high threshold and the data credibility is less than the medium threshold, the event-driven update method is adopted. If the event knowability is less than the low threshold and the data credibility is greater than or equal to the medium threshold, the data-driven update method is adopted. In other cases, a hybrid event and data-driven update method is adopted.
6. The method for dynamically updating a digital test model of aviation equipment according to claim 5, characterized in that, When using a hybrid-driven update approach, the event-driven model parameter update quantity and the data-driven model parameter update quantity are calculated in parallel. The fusion weight is calculated based on the real-time event knowability and data credibility, and the two are weighted and fused to obtain the final model parameter update quantity.
7. The method for dynamically updating a digital test model of aviation equipment according to claim 1, characterized in that, Step 4 also includes a hierarchical update strategy: online real-time fine-tuning of a subset of model parameters that are highly sensitive and change frequently; and monitoring the cumulative model deviation. When the deviation exceeds a preset threshold, a process of globally reconstructing the model parameters using the latest data and event history is triggered.
8. The method for dynamically updating a digital test model of aviation equipment according to claim 1, characterized in that, In step 4, the execution of update verification and iterative optimization specifically includes: after each update, comparing the model output with the latest physical test data and calculating the comprehensive fidelity score; storing the decision information, fusion weights and fidelity score of this update as empirical data for periodically optimizing the decision logic of the dynamic update mechanism and adjusting the decision threshold.
9. An electronic device, characterized in that, include: One or more processors; Memory, used to store one or more programs; When one or more programs are executed by the one or more processors, the one or more processors implement the dynamic update method for a digital test model of aviation equipment as described in any one of claims 1-8.
10. A computer-readable storage medium, characterized in that, It stores executable instructions, which, when executed by a processor, enable the processor to implement the dynamic update method for a digital test model of aviation equipment as described in any one of claims 1-8.