Special situation disposal-oriented flight training simulation and evaluation method and system

Abstract

The invention provides a flight training simulation and evaluation method and system for special situation disposal, and belongs to the technical field of aviation flight training, and the system comprises a special situation scene intelligent generation module, a digital twin simulation module, a multi-modal data acquisition module, an intelligent evaluation module, a self-adaptive training scheduling module and a visual redisk module. By intelligently generating a special situation scene, collecting multi-modal data in real time, performing multi-dimensional quantitative evaluation based on a deep learning algorithm, and dynamically adjusting a training scheme according to an evaluation result, accurate training and scientific evaluation of the special situation disposal ability of the pilot are realized. According to the method, the problems of lack of systematicness of special situation scene generation, high subjectivity of an evaluation process, non-fine training feedback and the like in traditional flight training are solved, the scientificity and efficiency of special situation disposal training are improved, the special situation disposal ability of a pilot can be accurately enhanced, and a guarantee is provided for aviation safety.

Description

Technical Field

[0001] This invention relates to the field of aviation flight training technology, and in particular to a method and system for flight training simulation and evaluation oriented towards handling special situations. Background Technology

[0002] In-flight emergency handling capabilities are a crucial factor in ensuring flight safety, directly determining a pilot's effectiveness in responding to emergencies such as engine failure, hydraulic malfunctions, and severe weather. Therefore, emergency handling training is a key component of the aviation flight training system. Currently, the mainstream emergency handling training in the industry uses full-motion flight simulators as the core platform, supplemented by instructors manually designing training scenarios, monitoring the training process, and completing evaluations. While this model can initially simulate real flight environments, it remains the primary means of cultivating pilots' emergency handling capabilities.

[0003] However, existing training models for handling emergencies have some significant limitations: First, the generation of training scenarios relies heavily on the instructor's personal experience, lacking a systematic design based on historical flight data and expert experience databases. This results in a limited range of scenario types and insufficient levels, making it difficult to cover complex real-world scenarios such as single emergencies and related malfunctions. Second, the evaluation process is primarily based on the instructor's subjective judgment, only able to provide qualitative evaluations of the handling results. It cannot quantify the standardization and rationality of the handling process based on multi-dimensional data such as stick operation, heart rate, and eye movement, resulting in insufficient objectivity in the evaluation. Third, post-training debriefing relies solely on instructor recollections or simple video playback, lacking in-depth analysis of key decision points and operational sequences. The feedback is vague and fails to guide pilots in targeted improvements. Fourth, the training program is fixed and uniform, unable to be dynamically adjusted according to the different pilots' skill gaps, leading to a waste of training resources and hindering personalized instruction.

[0004] Consequently, the aforementioned problems directly lead to inconsistent training effectiveness in handling emergencies, failing to effectively address pilots' skill gaps. Some pilots lack experience in dealing with complex combinations of emergencies, while others repeat training in scenarios they already know, reducing training efficiency and failing to meet the aviation industry's demand for high-precision emergency response training. Insufficient training may even lead to inappropriate responses to actual emergencies, thus hindering flight safety.

[0005] Therefore, there is an urgent need for a training simulation and evaluation method and system for handling special situations that integrates intelligent scenario generation, quantitative assessment, and personalized scheduling. Summary of the Invention

[0006] To overcome the shortcomings of existing technologies, the purpose of this invention is to provide a method and system for flight training simulation and evaluation oriented towards emergency response. This method solves the problems of lack of systematic generation of emergency scenarios, strong subjectivity in the evaluation process, and imprecise training feedback in traditional flight training. It improves the scientific nature and efficiency of emergency response training, can accurately enhance pilots' emergency response capabilities, and provides a guarantee for aviation safety.

[0007] To achieve the above objectives, the present invention provides the following solution: On the one hand, this invention provides a flight training simulation and evaluation method for handling special situations, including the following steps: Acquire training scenario data output by the special situation scenario intelligent generation module, wherein the training scenario data includes an ordered sequence of special situation evolution nodes generated based on the historical flight accident database and the expert experience database; Based on the training scenario data, a high-fidelity flight dynamics model is run through the digital twin simulation module to simulate the development process of special situations, thereby generating a real-time simulation data stream containing aircraft state parameters, environmental simulation data, and special situation evolution node triggering information. Based on the real-time simulation data stream, the pilot's joystick operation data, eye tracking data, and heart rate variability physiological data are synchronously collected through the multimodal data acquisition module to generate a time-aligned multimodal training dataset. Based on the special situation evolution node trigger information and the multimodal training dataset, the multi-dimensional fusion evaluation engine in the intelligent evaluation module is used to quantitatively analyze the pilot's operational standardization, decision quality, situational awareness and workload before and after each special situation evolution node, so as to generate a multi-dimensional node evaluation report. Based on the multi-dimensional node evaluation report, the adaptive training scheduling module identifies the pilot's skill deficiencies and dynamically adjusts the training strategy based on reinforcement learning algorithms to generate personalized subsequent training scenario configuration schemes. Based on the personalized subsequent training scenario configuration scheme and the historical training data stored in the data management module, the training process is reproduced in three dimensions and key decision points are marked through the visualization review module to generate a visual interactive review analysis report. The visualized, interactive debriefing analysis report is fed back to the special scenario intelligent generation module and the intelligent evaluation module to update the expert experience base and optimize the evaluation model parameters, forming an adaptive closed-loop training process.

[0008] Preferably, the step of acquiring training scene data includes: Retrieve predefined typical special situation scenario data for engine failure and hydraulic system malfunction from the basic special situation library unit; According to the training objective, multiple typical special situation scenario data are combined through the special situation combination unit, and the suddenness and complexity parameters of the special situation are set through the difficulty adjustment unit. Based on the combined special situation data and difficulty parameters, a complete training scheme containing the ordered special situation evolution node sequence is automatically generated by the scenario generator to output the training scenario data.

[0009] Preferably, the step of performing quantitative analysis based on the special situation evolution node trigger information and the multimodal training dataset includes: Based on the joystick operation data, the operation standardization evaluation unit uses rule engine technology to check its conformity with the standard operating procedure in order to generate an operation standardization score. Based on the special situation evolution node trigger information and pilot operation data, the decision quality assessment unit uses a deep learning algorithm to evaluate the timing of the decision and the rationality of the plan, so as to generate a decision quality score. Based on the eye-tracking data and operational data, the situational awareness assessment unit analyzes the pilot's situational awareness level to generate a situational awareness score. Based on the aforementioned heart rate variability physiological data, the pilot's cognitive workload status is monitored by the workload assessment unit to generate a workload score; Based on the operational standardization score, decision quality score, situational awareness score, and workload score, the multi-dimensional fusion evaluation engine performs weighted fusion to generate the multi-dimensional node evaluation report.

[0010] On the other hand, the present invention also provides a flight training simulation and evaluation system for performing the above-described method, comprising: The special situation scenario intelligent generation module is configured to generate training scenario data containing an ordered sequence of special situation evolution nodes; The digital twin simulation module is configured to run high-fidelity simulations and output real-time simulation data streams. The multimodal data acquisition module is configured to synchronously acquire data and generate a multimodal training dataset; The intelligent assessment module is configured to perform quantitative analysis and generate multi-dimensional node assessment reports. The adaptive training scheduling module is configured to generate personalized training scenario configuration schemes based on evaluation reports. The visual debriefing module is configured to generate interactive debriefing analysis reports. The data management module is configured to store all scenario, collection, evaluation, and review data, and supports historical data query and feedback.

[0011] According to specific embodiments provided by the present invention, the present invention discloses the following technical effects: (1) This invention sets up an intelligent special situation scenario generation module to generate multi-level training scenarios based on flight data analysis and expert experience database. Combined with basic special situation database, special situation combination and difficulty adjustment functions, it solves the problem that traditional special situation training scenarios rely on instructor experience and lack systematicity and diversity. At the same time, it combines a digital twin simulation module to establish a high-fidelity aircraft and environment model to realize real-time simulation of the special situation development process, providing pilots with a more realistic training environment and improving the scenario coverage and realism of special situation training.

[0012] (2) This invention uses a multimodal data acquisition module to acquire pilot operation data, physiological data and environmental data in real time. Combined with an intelligent evaluation module, it uses deep learning algorithms and rule engine technology to quantitatively evaluate the handling process from multiple dimensions such as operational standardization, decision quality, situational awareness and workload. This changes the traditional training model that relies on subjective human evaluation, reduces evaluation bias, and makes the evaluation of pilots' emergency handling ability more objective and accurate, providing data support for subsequent training improvement.

[0013] (3) The present invention uses an adaptive training scheduling module to dynamically adjust the training scheme and plan personalized training paths based on the evaluation results and the reinforcement learning algorithm, thereby avoiding the waste of resources in the traditional fixed training scheme. At the same time, the visualization review module provides three-dimensional reproduction of the training process and analysis of key decision points, and the data management module realizes the storage, query and statistics of training process data, making training feedback more refined and training adjustments more targeted, effectively improving the efficiency of special situation training, helping pilots to accurately strengthen their special situation handling capabilities, and providing a guarantee for aviation safety. Attached Figure Description

[0014] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the drawings used in the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0015] Figure 1 This is an overall architecture diagram of an intelligent flight training simulation and evaluation system for handling special situations, as described in this invention. Figure 2 A flowchart illustrating the workflow of the intelligent generation module for special scenarios provided by this invention; Figure 3 A schematic diagram illustrating the composition of the digital twin simulation module provided by this invention; Figure 4 A schematic diagram of the evaluation dimensions of the intelligent evaluation module provided by the present invention; Figure 5 This is an example diagram of the interface of the visualization review module during the training process provided by the present invention. Detailed Implementation

[0016] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0017] To make the above-mentioned objects, features and advantages of the present invention more apparent and understandable, the present invention will be further described in detail below with reference to the accompanying drawings and specific embodiments.

[0018] like Figure 1 As shown, this invention provides an intelligent flight training simulation and evaluation system for handling special situations, comprising: The special situation scenario intelligent generation module is configured to generate training scenario data containing an ordered sequence of special situation evolution nodes, which is used to generate multi-level training scenarios based on flight data analysis and expert experience base, and output the generated training scenarios to the digital twin simulation module. The digital twin simulation module is configured to run high-fidelity simulation and output real-time simulation data stream. It is used to receive the training scenario output by the special situation intelligent generation module, build a high-fidelity aircraft model and environment model based on the scenario, simulate the development process of special situations in real time, and synchronize the state data in the simulation process to the multimodal data acquisition module and intelligent evaluation module. The multimodal data acquisition module is configured to synchronously acquire data and generate a multimodal training dataset. It is used to receive the synchronization signal from the digital twin simulation module, acquire pilot operation data, physiological data and environmental data in real time, and output the acquired data to the intelligent evaluation module and the data management module respectively. The intelligent evaluation module is configured to perform quantitative analysis and generate multi-dimensional node evaluation reports. It receives data output from the multi-modal data acquisition module, performs multi-dimensional quantitative evaluation of the disposal process based on deep learning algorithms, and outputs the evaluation results to the adaptive training scheduling module, the visualization review module, and the data management module. The adaptive training scheduling module is configured to generate a personalized training scenario configuration scheme based on the evaluation report. It receives the evaluation results output by the intelligent evaluation module, dynamically adjusts the training scheme according to the evaluation results, feeds back the adjusted training scheme to the special scenario intelligent generation module, and synchronizes it to the data management module. The visualization review module is configured to generate interactive review analysis reports, which are used to receive the evaluation results of the intelligent evaluation module and the data collected by the multimodal data acquisition module, and provide three-dimensional reproduction and analysis functions of the training process; The data management module is configured to store all scenario, acquisition, evaluation, and review data, and supports historical data querying and feedback. It stores data acquired by the multimodal data acquisition module, evaluation results from the intelligent evaluation module, and training schemes from the adaptive training scheduling module. It supports data query, statistics, and analysis functions, and can feed back historical data to the special scenario intelligent generation module and the intelligent evaluation module for later retrieval. The data management module is a well-known technology in this field, and further details regarding existing technologies will not be elaborated upon.

[0019] The special situation scenario intelligent generation module includes: The basic emergency database unit is used to store typical emergency scenario data. It includes emergency scenario data such as engine failure, hydraulic system failure, electrical system failure, and severe weather conditions, and can provide emergency scenario data to the scenario generator in real time.

[0020] The special situation combination unit supports the generation of multiple special situations and transmits the combined special situation data to the scenario generator. The difficulty adjustment unit is used to set the suddenness and complexity of the emergency situation and sends the difficulty parameters to the scenario generator. The scenario generator receives scenario data from the basic scenario library unit, combination data from the scenario combination unit, and difficulty parameters from the difficulty adjustment unit. It automatically generates a complete training plan based on the training objective and outputs the training plan to the digital twin simulation module and the data management module.

[0021] In addition, refer to Figure 2 Starting with setting training objectives and parameters, the process first selects the training subject, then the incident type (if it is a single incident, a specific incident is called from the basic incident library; if it is a compound incident, the incident combination unit is started to select multiple incidents and set the trigger logic); then, difficulty parameters such as suddenness, complexity, and time pressure are configured. The scenario generator integrates all parameters and performs a rationality check on the incident logic. If the check passes, the final training scenario XML / JSON file is generated and output to the digital twin simulation module for execution. If the check fails, an error is displayed and the process is returned for adjustment, and the process ends.

[0022] In this embodiment, the digital twin simulation module adopts a high-precision flight dynamics model. This high-precision flight dynamics model is a common model known in the art, and will not be described in detail here. It supports real-time simulation of aerodynamic effects, system interconnection, and fault propagation. During the simulation of the development of special situations, the digital twin simulation module transmits the aircraft state parameters and environmental simulation data to the environmental data acquisition unit of the multimodal data acquisition module in real time, and sends the special situation evolution node information to the intelligent evaluation module to assist in the evaluation.

[0023] Specifically, refer to Figure 3 The digital twin simulation module achieves high-fidelity simulation of the development process of special situations through multi-unit collaboration: First, the special situation scenario parser analyzes the training scenario from the special situation scenario intelligent generation module and loads it into the fault injection and management unit; the fault injection and management unit triggers and manages the special situation state based on the scenario, sends a trigger command to the physical effect simulation unit, and the physical effect simulation unit calculates the chain physical effects caused by the special situation.

[0024] For high-fidelity aircraft models, there are flight dynamics models (used to simulate aerodynamic coefficients, engine characteristics, etc.), flight control system models (such as FCC, actuator systems, etc.), and aircraft system models (covering hydraulic, electrical, fuel, etc.). These sub-models undergo physical state changes under special circumstances, which then feed back the aircraft state to the real-time operating kernel. High-fidelity environment models include geographical models (terrain, airports, obstacles, etc.), air traffic models (ATC, other aircraft, etc.), and meteorological models (weather elements such as wind, rain, snow, and turbulence). The environmental state changes of each sub-model are also synchronized to the real-time operating kernel.

[0025] As a key component of the digital twin simulation module, the real-time runtime kernel receives aircraft states from various sub-models of the high-fidelity aircraft model, system states from the aircraft system model, and environmental states from various sub-models of the high-fidelity environment model. It also handles model scheduling and synchronization (coordinating the running steps of each sub-model to ensure simulation timing consistency), data distribution services (publishing simulation state data externally), and interacts with the multimodal data acquisition module and intelligent evaluation module through external interfaces. Ultimately, the real-time runtime kernel generates real-time simulation data streams containing aircraft states, environmental states, and fault information, which are then transmitted to the multimodal data acquisition module, intelligent evaluation module, and visualization review module, providing real-time and accurate simulation data support for multimodal data acquisition, intelligent evaluation, and visualization review.

[0026] In this embodiment, the multimodal data acquisition module includes: The operation data acquisition unit is used to record the operation data of the control stick, throttle, and steering surface, and transmits the operation data to the operation standardization assessment unit and data management module of the intelligent assessment module in real time. The physiological data acquisition unit is used to collect heart rate, eye movement, and electroencephalogram data, and send the physiological data to the workload assessment unit, situational awareness assessment unit, and data management module of the intelligent assessment module. The environmental data acquisition unit is used to receive aircraft status parameters and external environmental data output by the digital twin simulation module, record them, and then transmit them to the intelligent evaluation module and the data management module.

[0027] In addition, in this embodiment, the intelligent evaluation module includes: The operation standardization assessment unit is used to receive operation data from the multimodal data acquisition module, check the compliance of the operation procedure, and transmit the inspection results to the multidimensional fusion assessment engine. The decision quality assessment unit is used to combine the special situation evolution node information of the digital twin simulation module with the operation data of the multimodal data acquisition module to assess the timing of decisions and the rationality of solutions, and output the assessment results to the multi-dimensional fusion assessment engine. The context awareness assessment unit is used to receive eye-tracking data and operational data from the multimodal data acquisition module, analyze the level of context awareness, and send the analysis results to the multidimensional fusion assessment engine. The workload assessment unit is used to receive physiological data from the multimodal data acquisition module, monitor cognitive load status, and transmit the monitoring results to the multidimensional fusion assessment engine. The multi-dimensional fusion evaluation engine integrates the results of the above units, generates a multi-dimensional quantitative evaluation report, and outputs it to the adaptive training scheduling module, the visualization review module, and the data management module.

[0028] The operational compliance assessment unit uses rule engine technology to check the compliance of operational procedures and transmits the rule matching results to the multi-dimensional fusion assessment engine. The decision quality assessment unit uses deep learning algorithms to process data and sends the quantitative results output by the algorithm to the multi-dimensional fusion assessment engine. The formula for the deep learning algorithm is as follows: For the special situation evolution time sequence Pilot Operation Sequence and special situation static feature vector F First, a linear transformation is performed on the special situation state and operation, and position encoding is added to capture the temporal position information, that is: ; ; Then, the cross-modal association weights between special situations and operations are learned through a multilayer perceptron (MLP), and bidirectional fusion features are generated, expressed as follows: ; Subsequently, the multi-source information is spliced ​​to obtain the fused features: ; After linear transformation, RLU activation, and batch normalization (BN) optimization, global average pooling and gated recurrent units (GRUs) are used to capture global decision features and temporal decision dynamics, respectively. Finally, the decision quality quantification value is obtained by combining the Sigmoid() and SoftMax functions and multiplying them element by element. Q The calculation formula is: ; This formula integrates temporal coding, cross-modal attention, multi-feature fusion, and recurrent network mechanisms to achieve multi-dimensional quantification of the rationality of decision-making timing and the effectiveness of response plans.

[0029] In the formula, S This is a sequence of special situation evolution states. s i For the special situation evolution time-series state sequence, the first i Special situation state vectors under each time step; T The total time step for the development of the special situation; A For the pilot's operation timing sequence; a i The first in the pilot operation timing sequence i Operation vectors at each time step; F For special static feature vectors; Linear(·) is the linear transformation function; PositionalEncoding( i () represents the time step size i The corresponding position encoding function; s i ' is a special situation' s i The vector obtained after linear transformation and positional encoding; a i ′ is the operation vector a i The vector obtained after linear transformation and positional encoding; a j ′ represents the first ' in the pilot's operation sequence. j Operation vector at each time step a j The feature vector obtained after linear transformation and positional encoding; a k ′ represents the first ' in the pilot's operation sequence. k Operation vector at each time step a k The feature vector obtained after linear transformation and positional encoding; α i,j For the first i Special situation and the first j The cross-modal association weights of the time-series operation; MLP(·) is a multilayer perceptron; S A The operation sequence is a weighted fusion feature of the special situation state based on cross-modal weights; A S This is a weighted fusion feature of the special situation sequence based on cross-modal weights; Concat(·) is the feature concatenation operation; f iThe fused feature vector is obtained by concatenating multi-source information; ReLU(·) is the linear rectified activation function; BN(·) is the batch normalization operation; GlobalAvgPool(·) is the global average pooling operation; GRU(·) is the gated recurrent unit. σ (·) is the Sigmoid activation function; SoftMax(·) is the SoftMax function; Q A quantitative value for decision quality; This is an element-wise multiplication operation.

[0030] Specifically, refer to Figure 4 The intelligent evaluation module assesses performance across four dimensions: operational standardization, decision-making quality, situational awareness, and workload. A comprehensive report is generated through a multi-dimensional fusion evaluation engine. The operational standardization evaluation assesses the completeness of checklist execution, compliance with standard operating procedures, and accuracy of input, transmitting data from each dimension to the multi-dimensional fusion evaluation engine. The decision-making quality evaluation assesses the rationality of decision timing, the effectiveness of response plans, and resource management capabilities, sending corresponding data to the multi-dimensional fusion evaluation engine. The situational awareness evaluation assesses the rationality of attention allocation, environmental situational awareness, and anticipation of special situations, transmitting relevant data to the multi-dimensional fusion evaluation engine. The workload evaluation assesses physiological indicators (heart rate variability / skin conductance), task complexity analysis, and performance-load balance, transmitting data from each dimension to the multi-dimensional fusion evaluation engine. The multi-dimensional fusion evaluation engine integrates the evaluation data from these four dimensions to generate a comprehensive evaluation report. This report is output to both the visualization review module and the adaptive training scheduling module, providing a basis for training review and dynamic adjustment of the training plan.

[0031] In this embodiment, the adaptive training scheduling module adopts a reinforcement learning algorithm, receives a multi-dimensional quantitative evaluation report output by the intelligent evaluation module, identifies the pilot's skill deficiencies, automatically recommends training content, generates a personalized training path planning scheme, and feeds the generated scheme back to the special situation intelligent generation module to adjust the next round of training scenarios, and synchronizes it to the data management module for storage.

[0032] Specifically, the reinforcement learning algorithm used in the adaptive training scheduling module is a mature existing technology in the field. Its core function is to dynamically match the evaluation results, capability shortcomings, and training content based on the multi-dimensional quantitative evaluation report output by the intelligent evaluation module, thereby generating a personalized training path planning scheme. This system does not improve the underlying logic and mathematical framework of this reinforcement learning algorithm, but only adapts it to the scheduling scenario of special situation handling training. The specific application method is as follows: The application logic of this reinforcement learning algorithm revolves around the classic reinforcement learning framework of "state-action-reward": The state space is defined as the multi-dimensional quantitative evaluation results output by the intelligent evaluation module, including key indicators such as operational standardization score, decision quality score, situational awareness score, and workload adaptability (e.g., an operational standardization score below 80 indicates a capability weakness in that dimension); the action space is defined as the set of schedulable training content, covering single special situations (engine failure, hydraulic system failure, etc.) in the basic special situation library, composite special situations generated by the special situation combination unit (e.g., "engine failure + severe weather"), and the suddenness and complexity parameters that can be set by the difficulty adjustment unit (e.g., high suddenness, medium complexity, etc.); the reward function is set based on the pilot's capability improvement after training—if the evaluation score of the corresponding weakness dimension improves by ≥10 points after the next round of training, it is judged as "effective training" and given a positive reward; if the score does not improve or decreases, it is judged as "ineffective training" and given a negative reward. The recommendation strategy for training content is continuously optimized through reward feedback.

[0033] In practical operation, the execution flow of this reinforcement learning algorithm coordinates with the functional steps of the adaptive training scheduling module: First, the algorithm receives a multi-dimensional quantitative evaluation report output by the intelligent evaluation module, analyzes the score distribution in the report, and identifies the pilot's capability weakness dimension (e.g., a decision quality score of 72 points is considered a weakness); Second, based on the current state information corresponding to the "weakness dimension," it selects suitable training content from the action space (e.g., for the decision quality weakness, it selects a composite situation of "hydraulic system failure + landing gear failure" and sets high time pressure parameters); Third, combining the "weakness-training content" matching effect in historical training data (stored in the data management module), the algorithm recommends the optimal training content through iterative learning, generating a personalized training path planning scheme that includes training scenario type, difficulty level, and training duration; Fourth, the algorithm synchronously outputs the generated scheme to the module's feedback unit and storage unit. On the one hand, it feeds back to the intelligent scenario generation module to adjust the generation logic of the next round of training scenarios; on the other hand, it synchronously stores the scheme in the data management module, providing data support for subsequent iterative optimization of training scheduling.

[0034] In this embodiment, the visualization review module supports three-dimensional reproduction of the training process, receives the evaluation report output by the intelligent evaluation module to mark key decision points, and calls the historical data collected by the multimodal data acquisition module stored in the data management module. Combined with the three-dimensional reproduction function, it realizes the analysis function of the disposal process.

[0035] Specifically, refer to Figure 5The visual debriefing module enables the review and analysis of the training process through a multi-area collaborative layout: the top global information area displays the training scenario, trainee information, date, and comprehensive score, providing the basic background for the debriefing; the right side multi-dimensional evaluation details area includes an evaluation dimension radar chart, presenting scores for dimensions such as operational standardization, decision quality, situational awareness, and workload, such as 90 points for operational standardization and 85 points for decision quality; a key event list: listing key events such as engine vibration exceeding limits, declaring an emergency, and executing engine shutdown procedures; and expert suggestions and comments: providing improvement suggestions for decision-making decisiveness, process standardization, and system proficiency, analyzing the handling effect from multiple dimensions; the main display area features a 3D flight trajectory reproduction window, which can display the aircraft attitude, flight path, and key event markers, intuitively restoring the spatial dynamic process of training; the bottom main timeline and event marker area are based on time, marking key nodes such as fault triggering, initial decision point, executed actions, and emergency resolution, and supports dragging and zooming the timeline to focus on the handling process at different times. Data linkage across regions not only restores the overall training picture through 3D reconstruction and timeline, but also clarifies improvement directions with the help of multi-dimensional evaluation and expert suggestions, fully supporting the review and analysis of emergency response training.

[0036] Furthermore, in this embodiment, the data stored in the data management module includes operational data, physiological data, and environmental data from the multimodal data acquisition module; multi-dimensional quantitative evaluation reports from the intelligent evaluation module; training scenario data from the special situation scenario intelligent generation module; and training schemes from the adaptive training scheduling module. The data management module supports querying stored data by training batch, pilot identity, and special situation type, and statistically analyzing training effects by evaluation dimension. It can also provide historical special situation data to the special situation scenario intelligent generation module and historical evaluation data to the intelligent evaluation module for reference.

[0037] As a preferred embodiment, a flight training simulation and evaluation method for handling special situations, based on the above system, includes the following steps: Acquire training scenario data output by the special situation scenario intelligent generation module, wherein the training scenario data includes an ordered sequence of special situation evolution nodes generated based on the historical flight accident database and the expert experience database.

[0038] Specifically, the steps for acquiring training scene data include: Retrieve predefined typical special situation scenario data for engine failure and hydraulic system malfunction from the basic special situation library unit; According to the training objective, multiple typical special situation scenario data are combined through the special situation combination unit, and the suddenness and complexity parameters of the special situation are set through the difficulty adjustment unit. Based on the combined special situation data and difficulty parameters, a complete training scheme containing the ordered special situation evolution node sequence is automatically generated by the scenario generator to output the training scenario data.

[0039] Based on the training scenario data, a high-fidelity flight dynamics model is run through the digital twin simulation module to simulate the development process of special situations, thereby generating a real-time simulation data stream containing aircraft state parameters, environmental simulation data, and special situation evolution node trigger information.

[0040] Based on the real-time simulation data stream, the pilot's joystick operation data, eye tracking data, and heart rate variability physiological data are synchronously collected through the multimodal data acquisition module to generate a time-aligned multimodal training dataset.

[0041] Based on the special situation evolution node trigger information and the multimodal training dataset, the multi-dimensional fusion evaluation engine in the intelligent evaluation module is used to quantitatively analyze the pilot's operational standardization, decision quality, situational awareness, and workload before and after each special situation evolution node, so as to generate a multi-dimensional node evaluation report.

[0042] Specifically, the step of performing quantitative analysis based on the special situation evolution node trigger information and the multimodal training dataset includes: Based on the joystick operation data, the operation standardization evaluation unit uses rule engine technology to check its conformity with the standard operating procedure in order to generate an operation standardization score. Based on the special situation evolution node trigger information and pilot operation data, the decision quality assessment unit uses a deep learning algorithm to evaluate the timing of the decision and the rationality of the plan, so as to generate a decision quality score. Based on the eye-tracking data and operational data, the situational awareness assessment unit analyzes the pilot's situational awareness level to generate a situational awareness score. Based on the aforementioned heart rate variability physiological data, the pilot's cognitive workload status is monitored by the workload assessment unit to generate a workload score; Based on the operational standardization score, decision quality score, situational awareness score, and workload score, the multi-dimensional fusion evaluation engine performs weighted fusion to generate the multi-dimensional node evaluation report.

[0043] Based on the multi-dimensional node evaluation report, the adaptive training scheduling module identifies the pilot's skill gaps and dynamically adjusts the training strategy based on reinforcement learning algorithms to generate personalized subsequent training scenario configuration schemes.

[0044] Based on the personalized subsequent training scenario configuration scheme and the historical training data stored in the data management module, the training process is three-dimensionally reproduced and key decision points are marked through the visualization review module to generate a visual interactive review analysis report.

[0045] The visualized, interactive debriefing analysis report is fed back to the special scenario intelligent generation module and the intelligent evaluation module to update the expert experience base and optimize the evaluation model parameters, forming an adaptive closed-loop training process.

[0046] Furthermore, in a typical practical application scenario, for basic emergency handling training of newly recruited pilots, the emergency scenario intelligent generation module retrieves a single emergency scenario, "single engine failure," from the basic emergency database. The difficulty adjustment unit is set to "with warning, medium complexity, and moderate time pressure." After integration by the scenario generator, a training plan is generated, verified, and output to the digital twin simulation module. The digital twin simulation module constructs a corresponding high-fidelity aircraft model and environmental model to simulate the chain effects of thrust changes and aerodynamic imbalances after engine failure, synchronizing status data in real time. The multimodal data acquisition module records the pilot's control stick corrections, throttle adjustments, and other operational data, as well as heart rate and eye... Physiological data such as movement; the intelligent evaluation module checks the compliance of procedures through the operational standardization evaluation unit, and the decision quality evaluation unit analyzes the rationality of decisions in conjunction with the special situation evolution nodes, and generates an evaluation report after multi-dimensional fusion; the adaptive training scheduling module identifies the shortcoming of "delayed decision timing" based on the report, calls the reinforcement learning algorithm to recommend composite special situation training of "single engine failure + small-scale weather change", feeds back to the special situation scenario intelligent generation module to adjust the next round of scenarios, and stores it in the data management module; the visualization debriefing module reproduces the process of aircraft yaw correction, power compensation, etc. in three dimensions during training, marks the key decision point of "delay of 5 seconds for the first thrust adjustment", and assists the pilot in debriefing.

[0047] In another application scenario, for the assessment of senior pilots' handling of complex emergencies, the emergency scenario intelligent generation module generates a composite emergency situation of "hydraulic system failure + partial electrical system failure + strong crosswind weather" through the emergency combination unit. The difficulty is adjusted to "no warning, high complexity, short-term pressure". The scenario generator outputs the scenario after completing logical verification. The digital twin simulation module accurately simulates the aircraft response under the cross-linking of multiple system failures and severe weather conditions, such as the surge in control force after hydraulic power assist failure and the partial failure of instruments due to electrical failure. The multimodal data acquisition module captures the pilot's operation data such as high force on the control stick and frequent switching of backup instruments, as well as the high-intensity cognitive load reflected by EEG signals. The intelligent assessment module evaluates operations from multiple dimensions, including operational standardization (such as backup instrument call procedures), decision quality (such as the priority of handling hydraulic or electrical faults), situational awareness (perception of the situation under multiple faults), and workload (peak load reflected by physiological data). The adaptive training scheduling module recommends "special training on optimizing the allocation of resources for multiple faults" if it determines that "multi-system fault resource management is insufficient" based on the assessment results. The visual debriefing module fully restores the handling process and, combined with the expert advice area's commentary on "fault priority judgment," helps senior pilots further improve their ability to handle complex scenarios. All process data is stored and subsequently retrieved and analyzed by the data management module.

[0048] Therefore, the above-mentioned method and system for flight training simulation and evaluation oriented towards emergency response solves the problems of lack of systematic generation of emergency scenarios, strong subjectivity in the evaluation process, and imprecise training feedback in traditional flight training. It improves the scientific nature and efficiency of emergency response training, can accurately enhance pilots' emergency response capabilities, and provides a guarantee for aviation safety.

[0049] This document uses specific examples to illustrate the principles and implementation methods of the present invention. The descriptions of the above embodiments are only for the purpose of helping to understand the method and core ideas of the present invention. Furthermore, those skilled in the art will recognize that, based on the ideas of the present invention, there will be changes in the specific implementation methods and application scope. Therefore, the content of this specification should not be construed as a limitation of the present invention.

Claims

1. A method for flight training simulation and evaluation for special situation handling, characterized in that, The method comprises the following steps: obtaining training scene data output by a special situation scenario intelligent generation module, wherein the training scene data contains an ordered special situation evolution node sequence generated based on a historical flight accident library and an expert experience library; running a high-fidelity flight dynamics model through a digital twin simulation module to simulate the development process of a special situation according to the training scene data, to generate a real-time simulation data stream containing aircraft state parameters, environmental simulation data and special situation evolution node trigger information; synchronously collecting pilot joystick operation data, eye tracking data and heart rate variability physiological data through a multi-modal data acquisition module according to the real-time simulation data stream, to generate a time-aligned multi-modal training data set; quantitatively analyzing the operation standardization, decision quality, situational awareness and workload of the pilot before and after each special situation evolution node through a multi-dimensional fusion evaluation engine in an intelligent evaluation module according to the special situation evolution node trigger information and the multi-modal training data set, to generate a multi-dimensional node evaluation report; identifying the pilot's capability short board through an adaptive training scheduling module according to the multi-dimensional node evaluation report, and dynamically adjusting the training strategy based on a reinforcement learning algorithm, to generate a personalized subsequent training scene configuration scheme; reproducing the training process in three dimensions and marking key decision points through a visualized review module according to the personalized subsequent training scene configuration scheme and the historical training data stored in the data management module, to generate a visualized interactive review analysis report; feeding back the visualized interactive review analysis report to the special situation scenario intelligent generation module and the intelligent evaluation module to update the expert experience library and optimize the evaluation model parameters, forming an adaptive closed-loop training process.

2. The method of claim 1, wherein, The training scene data acquisition step comprises: calling pre-defined engine failure and hydraulic system fault typical special situation scene data from a basic special situation library unit; combining multiple typical special situation scene data through a special situation combination unit according to the training target, and setting the suddenness parameter and complexity parameter of the special situation occurrence through a difficulty adjustment unit; automatically generating a complete training scheme containing the ordered special situation evolution node sequence through a scenario generator according to the combined special situation data and difficulty parameters, to output the training scene data.

3. The method of claim 1, wherein, The step of quantitatively analyzing according to the special situation evolution node trigger information and the multi-modal training data set comprises: checking the compliance of the joystick operation data with the standard operation procedure through an operation standardization evaluation unit using rule engine technology, to generate an operation standardization score; evaluating the decision timing and scheme rationality through a decision quality evaluation unit using a deep learning algorithm according to the special situation evolution node trigger information and the pilot operation data, to generate a decision quality score; analyzing the situational awareness level of the pilot through a situational awareness evaluation unit according to the eye tracking data and operation data, to generate a situational awareness score; monitoring the cognitive load state of the pilot through a workload evaluation unit according to the heart rate variability physiological data, to generate a workload score; According to the operation normative score, the decision quality score, the situational awareness score and the workload score, weighted fusion is performed by the multi-dimensional fusion evaluation engine to generate the multi-dimensional node evaluation report.

4. A flight training simulation and evaluation system for performing the method of any one of claims 1 to 3, characterized by Comprise: An intelligent special situation scenario generation module configured to generate training scenario data containing an ordered special situation evolution node sequence; A digital twin simulation module configured to run a high-fidelity simulation and output a real-time simulation data stream; A multi-modal data acquisition module configured to synchronously acquire data and generate a multi-modal training data set; An intelligent evaluation module configured to perform quantitative analysis and generate a multi-dimensional node evaluation report; An adaptive training scheduling module configured to generate a personalized training scenario configuration scheme according to the evaluation report; A visualized review module configured to generate an interactive review analysis report; A data management module configured to store all scenario, acquisition, evaluation and review data, and support historical data query and feedback.

Images