Aircraft control simulation platform construction method

Abstract

The invention discloses an aircraft control simulation platform construction method, and the method comprises the steps: constructing a terrain multi-dimensional model, fusing a cross-scale meteorological particle system, and dynamically obtaining an environment disturbance factor; a space obstacle dynamic database is generated in combination with millimeter-level space positioning data, and virtual collision detection is carried out; performing eye movement-gesture collaborative decision, and identifying a control instruction with high confidence; the six-degree-of-freedom kinetic equation is coupled with an unreal engine interface in real time, and a high-fidelity instrument is driven to feed back; generating a training defect thermodynamic diagram, triggering an adaptive optimization strategy and iterating platform parameters; the invention provides a solution integrating multi-modal high-precision data acquisition, dynamic environment-physical rule coupling, intelligent decision and closed-loop optimization so as to realize aircraft control simulation training with high immersion degree, high fidelity and high adaptability.

Description

Technical Field

[0001] The present invention relates to the field of simulation technology, and in particular to a method for constructing an aircraft control simulation platform. Background Art

[0002] Traditional aircraft simulation platforms rely on peripherals such as joysticks and keyboards for their interaction methods, and lack the integrated collection of natural interaction modes such as the operator's body movements and eye movements. In addition, the centimeter-level positioning accuracy cannot meet the quantification requirements of body movements for fine aircraft operation, resulting in insufficient immersion in human-computer interaction and limited command mapping accuracy. Environmental models are mostly based on fixed GIS data or simple meteorological parameters, lacking the ability to generate cross-scale dynamic environmental disturbances. At the same time, the simplified collision model used in the physical rule set has problems such as large detection error and high feedback delay, and cannot truly restore the aerodynamic characteristics and obstacle interaction mechanical characteristics in complex environments.

[0003] In terms of control command decision-making, traditional algorithms independently process eye movement and gesture data, and fail to establish a spatiotemporal collaborative association mechanism. This results in low command recognition in complex scenarios and a high risk of misoperation. In addition, existing platforms lack an adaptive adjustment mechanism based on operational feedback, and training parameters are fixed and preset, unable to be dynamically optimized based on the operator's real-time performance. This results in low training efficiency and an average proficiency time that is longer than actual operation. In response to the aforementioned problems of insufficient interactive realism, rigid environmental simulation, distorted physical feedback, and lack of adaptive capabilities, the present invention provides a solution that integrates multimodal high-precision data acquisition, dynamic environment-physical rule coupling, intelligent decision-making, and closed-loop optimization to achieve highly immersive, highly realistic, and highly adaptable aircraft control simulation training. Summary of the Invention

[0004] The present invention provides a method for constructing an aircraft control simulation platform, which includes: Connect virtual reality equipment, configure a millimeter-level spatial positioning system, build a multi-channel voice array and infrared eye tracking module, and obtain a structured data set; Construct a multi-dimensional terrain model, integrate cross-scale meteorological particle systems, and dynamically obtain environmental disturbance factors; Combined with millimeter-level spatial positioning data to generate a dynamic database of spatial obstacles for virtual collision detection; Use long short-term memory (LSTM) neural networks to make eye movement-gesture collaborative decisions and identify control commands with high confidence; Based on environmental disturbance factors, virtual collision forces, and control commands, the six-degree-of-freedom state vector is coupled with the Unreal Engine interface in real time to drive high-fidelity instrument feedback. Generate a training defect heat map based on instrument feedback data, triggering an adaptive optimization strategy to iterate platform parameters.

[0005] The method for constructing an aircraft control simulation platform as described above, wherein connecting a virtual reality device, configuring a millimeter-level spatial positioning system, building a multi-channel voice array and an infrared eye tracking module, and obtaining a structured data set includes: Equipped with a millimeter-level spatial positioning system, it tracks the operator's position and movements with millimeter-level precision, captures eye movement data in real time at a non-invasive and high-frequency manner, and simultaneously analyzes the coordinates of the gaze point and physiological micro-responses; The positioning, speech, and eye movement data are integrated and the timestamps are aligned with the limb movement acquisition module through the hardware clock synchronization circuit to generate a structured dataset.

[0006] The method for constructing an aircraft control simulation platform as described above, wherein a multi-dimensional terrain model is constructed, a cross-scale meteorological particle system is integrated, an environmental disturbance factor is dynamically acquired, and a dynamic environmental disturbance factor is generated, includes: Access millimeter-level spatial positioning data, integrate geographic information system elevation data, and build a multi-dimensional terrain model; Combined with the cross-scale meteorological particle system, the behavior of meteorological particles from the molecular level to the regional level is simulated, and environmental disturbance factors are dynamically generated.

[0007] The method for constructing an aircraft control simulation platform as described above, wherein a dynamic database of spatial obstacles is generated by combining millimeter-level spatial positioning data to perform virtual collision detection, includes: Based on millimeter-level spatial positioning data, a dynamic database of spatial obstacles is constructed and updated in real time to form an interactive physical rule set; Perform millimeter-level virtual collision force detection, calculate force feedback of collision between objects, output virtual collision vector and feedback.

[0008] A method for constructing an aircraft control simulation platform, wherein eye movement-gesture collaborative decision-making is performed using a long short-term memory (LSTM) neural network to identify control instructions with high confidence, includes: Integrate structured datasets as neural network input and use neural network training models to identify the temporal dependencies between eye movements and gestures; According to the trained model, the probability threshold is filtered and instructions with confidence higher than the preset value are output.

[0009] The method for constructing an aircraft control simulation platform, wherein a six-degree-of-freedom state vector is coupled to an Unreal Engine interface in real time based on environmental disturbance factors, virtual collision forces, and control instructions to drive high-fidelity instrument feedback, includes: Receive environmental disturbance factors, virtual collision forces and control instructions, and use Newton-Euler equations to solve the six-degree-of-freedom state vector; Integrate dynamics parameters into Unreal Engine, couple to Unreal Engine interfaces in real time, and drive high-fidelity instrument feedback.

[0010] The method for constructing an aircraft control simulation platform as described above, wherein a training defect heat map is generated based on instrument feedback data, and an adaptive optimization strategy is triggered to iterate platform parameters, including: Collect instrument feedback data, obtain operator performance data, and use data visualization technology to create heat maps to highlight deficient areas in training; Based on heat map analysis, the adaptive optimization strategy is triggered, and the reinforcement learning algorithm is applied to dynamically adjust the training parameters for closed-loop iteration.

[0011] The beneficial effects achieved by the present invention are as follows: Millimeter-level spatial positioning and multi-source data fusion accurately generate dynamic environmental disturbance factors. Combined with an obstacle database and physics rule set, this system faithfully reproduces complex flight environments. Multimodal interaction combined with LSTM decision-making accurately identifies control commands, and the six-degree-of-freedom dynamics equations are coupled with the Unreal Engine in real time, delivering high-fidelity instrument feedback and an enhanced control experience. Heatmaps are generated based on instrument data, triggering adaptive optimization strategies and iterating platform parameters, enabling personalized, efficient training and continuous refinement of aircraft control skills. BRIEF DESCRIPTION OF THE DRAWINGS

[0012] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the following briefly introduces the drawings required for use in the embodiments or the description of the prior art. Obviously, the drawings described below are only some embodiments described in the present invention. For ordinary technicians in this field, other drawings can also be obtained based on these drawings.

[0013] Figure 1 This is a flow chart of a method for constructing an aircraft control simulation platform provided in Example 1 of the present application. DETAILED DESCRIPTION

[0014] The following is a clear and complete description of the technical solutions in the embodiments of the present invention, in conjunction with the accompanying drawings. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of them. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative work are within the scope of protection of the present invention.

[0015] Example 1 like Figure 1 As shown, the first embodiment of the present application provides a method for constructing an aircraft control simulation platform, including: S1: Deploy virtual reality equipment, configure a millimeter-level spatial positioning system, build a multi-channel voice array and infrared eye tracking module, and obtain a structured data set; Establish millimeter-level spatial coordinate mapping, integrate a six-axis attitude sensor, calibrate the infrared eye tracking module, deploy a circular microphone array, and acquire structured data sets in real time. The structured data sets include at least millimeter-level positioning data, operator motion data, and geographic information system elevation data. Deploying virtual reality equipment, configuring a millimeter-level spatial positioning system, building a multi-channel voice array and infrared eye tracking module, and acquiring a structured data set involves the following sub-steps: S11: Equipped with a millimeter-level spatial positioning system, it tracks the operator's position and movements with millimeter-level precision, captures eye movement data in real time at a non-invasive and high-frequency rate, and simultaneously analyzes gaze point coordinates and physiological micro-responses. Load the predefined 3D model of the experimental space, collect spatial data from the markers on the calibration rod, establish a global coordinate system, and complete spatial distortion error compensation mapping. Record the coordinate data returned by the positioning system, compare it to the calibration frame reference value, calculate the static position offset and write it into the compensation matrix, dynamically collect spatial positioning data and IMU data, run the sensor fusion algorithm, and optimize dynamic tracking accuracy to the millimeter level.

[0016] Execute the tracking delay test program, record the physical displacement instructions and virtual scene response timestamps at high speed, calculate the end-to-end motion-to-photon delay, adjust the rendering pipeline and transmission protocol delay to ≤20ms, monitor the continuity of positioning data, and confirm that there is no frame loss or jump.

[0017] Start the eye tracker's underlying driver, load the pupil-corneal reflection algorithm kernel, perform automatic optimization of the camera's initial focal length and exposure parameters, guide the operator to gaze at a preset nine-point grid sequence through the calibration interface, dynamically collect the geometric relationship between the corneal reflection points and the pupil contour, and establish a user-specific three-dimensional eye movement model.

[0018] A hardware-level timestamp generator aligns eye movement data streams, physiological signal streams, and the simulation master clock, establishing a multimodal synchronous transmission channel. Dynamic tracking verification is performed, pupil center displacement vectors are analyzed in real time, and the three-dimensional coordinates of the gaze point in the simulation visual coordinate system are output. Blink frequency, pupil diameter change, and iris microtwitches are simultaneously marked. This also triggers the PPG signal peak detection algorithm to extract heart rate variability features.

[0019] S12: Fusion of positioning, speech, and eye movement data, and alignment of timestamps with the limb motion acquisition module through a hardware clock synchronization circuit to generate a structured dataset.

[0020] The hardware-level precision clock synchronization circuit is activated to distribute nanosecond-precision time synchronization pulses to the spatial positioning base station array, eye tracking module, bone conduction microphone array, and inertial motion capture suit. A unified reference clock tag is embedded in real time into each sensor's raw data stream. Data streams from each sensor are received in parallel via a dedicated interface. A kernel-level real-time thread is invoked to perform hardware interrupt-triggered preprocessing on four types of data using a ring buffer structure: applying Kalman filtering to smooth jitter on spatial coordinates, performing speckle noise suppression on eye movement signals, initiating beamforming noise reduction on voice streams, and implementing gravity compensation on body movement data.

[0021] The timestamp alignment engine is activated, using the iris reflection event from the eye movement module as the benchmark trigger point to dynamically calibrate the offset of other sensor data frames within the synchronization period. A cross-modal fusion algorithm is then run to bind the pose data to the head skeleton nodes, map the eye gaze coordinates to the helmet display spatial coordinate system, associate the voice command time window with the joint motion trajectory, and synchronously integrate the limb motion vectors into the whole-body motion tree to generate a structured dataset.

[0022] S2: Construct a multi-dimensional terrain model, integrate cross-scale meteorological particle systems, and dynamically obtain environmental disturbance factors; The system loads geographic information system elevation raster data and registers it to the spatial positioning coordinate system through UTM projection conversion; it simultaneously accesses the meteorological mesoscale numerical analysis results, discretizes and deploys cross-scale meteorological particle systems in the simulation airspace, simulates the behavior of meteorological particles from the molecular level to the regional level, and dynamically generates environmental disturbance factors.

[0023] Based on the spatial positioning system, the GIS elevation data and the cross-scale meteorological particle system are integrated to generate dynamic environmental disturbance factors, including the following sub-steps: S21: Access millimeter-level spatial positioning data, integrate geographic information system elevation data, and build a multi-dimensional terrain model; The operator's three-dimensional spatial coordinates and attitude angle data are obtained in real time through the millimeter-level spatial positioning module, and the geographic information system data interface is triggered synchronously. According to the latitude and longitude information of the current spatial coordinates, the elevation data of the corresponding area is called from the preloaded terrain database.

[0024] The local coordinate system of the spatial positioning data is converted into the global geodetic coordinate system of the geographic information system through the spatial transformation matrix, so that the spatial reference of the positioning point and the elevation data are kept consistent; for the elevation grid points around the positioning point, the inverse distance weighted interpolation algorithm is used to dynamically generate a continuous terrain surface, and the interpolated elevation data is superimposed with the spatial positioning data to generate a dynamic terrain height matrix centered on the operator's current position. The terrain undulation characteristics are updated in real time through bilinear sampling to generate a multi-dimensional terrain model.

[0025] S22: Combine cross-scale meteorological particle systems to simulate the behavior of meteorological particles from the molecular level to the regional level, and dynamically generate environmental disturbance factors.

[0026] The cross-scale meteorological particle engine is launched, and based on the terrain height matrix and preset meteorological parameters, the multi-scale particle simulation domain is initialized: the molecular-level particle layer simulates Brownian motion through the Langevin dynamic equation to calculate the impact of air molecule thermal disturbances on microscale airflow; the mesoscopic particle layer uses the lattice Boltzmann method to solve the particle distribution function with a discrete velocity model; the regional-level particle layer uses the vorticity constraint form of the Navier-Stokes equation to track the advection and diffusion of turbulent vortices through the vortex particle method. The three-scale particle behaviors are integrated in the particle aggregation module, and the environmental disturbance factor is calculated according to spatial position and time. The environmental disturbance factor calculation formula is:

[0027] It represents the molecular-scale coupling coefficient, which is used to adjust the contribution weight of molecular-scale particle motion to macroscopic perturbations; Indicates the number of molecular-level particle sampling, which is the number of molecular particles counted and used for Monte Carlo integration. The larger the value, the more accurate the thermal noise statistics. represents the momentum vector of the kth air molecule, reflecting the microscopic momentum increase caused by the Brownian motion of the air molecule; , represents the mass of the kth air molecule, The instantaneous velocity vector of the kth air molecule is given by the Langevin equation Solve, represents the damping coefficient; represents the Gaussian white noise term, simulating the randomness of molecular collisions.

[0028] represents the vortex scale coupling coefficient, which is used to adjust the strength of the vortex force term. Indicates location The vorticity vector at the point; r represents the distance from the target point X to the vortex source point The relative position vector of ; t represents the current moment; The differential form of Biot-Savart law is used to describe the velocity field induced by the vortex at point x, characterizing the non-uniform effect of the vortex inertial force on the spacecraft. represents the terrain-scale coupling coefficient, which is used to adjust the gain of the terrain gradient force; represents the gravitational acceleration vector; It represents the gradient vector of the terrain height matrix Z, which represents the projection of gravity along the terrain slope and reflects the steady disturbances dominated by the terrain, such as slope effects and valley winds.

[0029] S3: Generate a dynamic database of spatial obstacles based on millimeter-level spatial positioning data for virtual collision detection; The system receives real-time millimeter-level spatial positioning data streams and discretizes the operator's surroundings into millimeter-level voxel units using a dynamic voxel meshing algorithm. It then integrates geographic information data with pre-set obstacle models to build and update a dynamic spatial obstacle database. The system uses a separating axis theorem collision detection engine to calculate the penetration depth and contact normal vectors between key points on the aircraft surface and obstacle pixels, and generates a six-degree-of-freedom virtual collision force vector using a spring-damper force model.

[0030] Combined with millimeter-level spatial positioning data, a dynamic database of spatial obstacles is generated to form an interactive physical rule set, and millimeter-level virtual collision force detection is performed, including the following sub-steps: S31: Based on millimeter-level spatial positioning data, a dynamic database of spatial obstacles is constructed and updated in real time to form an interactive physical rule set; The system accesses real-time 3D coordinate data from a millimeter-level spatial positioning system, fits discrete positioning points into continuous motion trajectories, and constructs a real-time pose matrix for dynamic objects. It then activates a multi-threaded data parsing module to perform semantic segmentation on the spatial positioning data. It uses a geometric feature matching algorithm to identify the basic geometric units of obstacles and generates a 3D surface model with normal vector information based on triangulated meshing technology. It also simultaneously loads a pre-set physical property library to form structured data units containing both geometric features and physical parameters.

[0031] When new frame spatial positioning data is input, an incremental update mechanism is triggered: coordinate mutations are detected using the Euclidean distance threshold method. Objects whose displacement exceeds the threshold are marked as "moving," and their bounding box spatial orientation is recalculated. Objects for which valid coordinates are not detected are marked as "invalid entities" and removed from the database after confirmation by a delayed deletion queue. The system establishes interaction constraints between objects based on Newtonian mechanics, defining momentum transfer formulas and energy loss models for collision responses. It also uses a spatial hash table to grid-partition obstacles and establish a "grid index-object ID" mapping relationship, forming an interactive physical rule set that includes a dynamic entity list, geometric constraints, and physical interaction rules.

[0032] S32: Perform millimeter-level virtual collision force detection, calculate the force feedback of the collision between objects, output the virtual collision vector and feedback; Based on the 3D surface model, bounding box parameters (OBB directional bounding box), and physical properties in the interactive physics rule set, a spatial hash table is used to quickly locate pairs of objects within the same grid partition in the current frame, forming a set of objects to be detected. Coarse collision detection is performed at the bounding box level: the separating axis theorem is used to determine whether the bounding boxes of two objects intersect. If the minimum translation vector of the separating axis is detected to be less than the millimeter-level accuracy threshold, precise geometric collision detection is triggered. For convex objects, the GJK algorithm is used to calculate the minimum penetration depth and contact normal. For non-convex objects, ray tracing and facet intersection testing are performed based on the triangular mesh to extract the 3D coordinates and local surface normals of all contact points.

[0033] After determining the collision pair and contact information, the system calculates the virtual collision force vector feedback based on the relative motion state at the moment of collision. The calculation formula is:

[0034] Where k represents the virtual elastic coefficient; Indicates the penetration depth, which is the relative displacement of the collision bodies penetrating each other in the virtual space; represents the unit normal vector of the collision contact surface; c represents the virtual damping coefficient, which quantifies the energy dissipation of the virtual collision; Represents the relative velocity vector of the collision point; represents the normal component of the relative velocity; represents the vector force that converts the scalar projection into the normal direction; represents the virtual friction coefficient, which quantifies the degree of sliding friction of the virtual collision; represents the scalar magnitude of the normal force; It represents the sign indicator function. When the value in the bracket is not 0, it means there is relative motion in the tangential direction. represents the tangential component of the relative velocity; Represents the unit tangent vector of the contact surface; Convert the tangential velocity component into a tangential vector force; The force feedback vectors of each pair of colliding objects are transformed into the world coordinate system, and a timestamp is added to each force vector using the hardware synchronized clock.

[0035] S4: Eye movement-gesture collaborative decision-making is performed through a long short-term memory (LSTM) neural network to identify control commands with high confidence; The system integrates eye movement and gesture data in real time, pre-processes it for noise reduction, and then feeds it into an LSTM neural network. It then uses a gating mechanism to capture temporal dependencies. Model inference generates a probability distribution for commands, sets a probability threshold to filter out low-confidence outputs, and ultimately outputs control commands with confidence levels exceeding a preset value.

[0036] The long short-term memory (LSTM) neural network is used to make eye movement-gesture coordinated decisions and identify high-confidence control commands, including the following sub-steps: S41: Integrate structured datasets as neural network input and use neural network training models to identify the temporal dependencies between eye movements and gestures; Feature engineering was performed on the structured dataset: temporal features such as pupil diameter change rate, gaze point displacement velocity, and scan frequency were extracted from the eye movement data; spatial-temporal features such as joint angle change, limb movement trajectory curvature, and movement duration were extracted from the gesture data to form a multidimensional feature vector sequence. A three-layer LSTM neural network architecture was constructed: the input layer receives the feature vector (the dimension is the number of eye movement features + the number of gesture features), the middle layer contains 128 memory cells (with sigmoid activation functions set for the forget gate, input gate, and output gate), and the output layer uses the Softmax activation function to generate the probability distribution of the command category.

[0037] During the training phase, the system batches the labeled dataset using a sliding time window to minimize the cross-entropy loss function. Backpropagation is used to update network weights, with an early stopping mechanism to prevent overfitting. The resulting model parameters (weight matrix, bias vector) are then stored as an executable file. The real-time feature sequence is fed into the trained network, and forward propagation is used to calculate the probability distribution of eye-gesture coordination patterns, completing the mapping from motion data to control intent.

[0038] S42: Based on the trained model, filtering is performed through a probability threshold, and outputting instructions with a confidence level higher than a preset value; The probability distribution vector of the command category is received to suppress random noise interference. The threshold decision module is activated: the smoothed probability value of each command category is compared with the preset confidence threshold. If the highest probability value exceeds the threshold and the difference with the second highest probability value is greater than 0.2, the corresponding command is marked as a high-confidence command. If the conditions are not met, multimodal consistency verification is triggered: the gaze duration and gesture trajectory completion score in the eye movement data are retrieved, and auxiliary evidence such as the similarity with the standard action template is calculated based on the dynamic time warping algorithm. The posterior probability is calculated using the Bayesian fusion formula. If the posterior probability rises above the threshold, the command is confirmed to be valid.

[0039] For confirmed high-confidence instructions, the system adds a timestamp and converts it into a unified format for output; for instructions that fail verification, an invalid instruction identifier is generated and an exception log is recorded, while the current frame data is stored in a buffer for subsequent model optimization.

[0040] S5: Based on environmental disturbance factors, virtual collision forces, and control commands, the six-degree-of-freedom state vector is coupled with the Unreal Engine interface in real time to drive high-fidelity instrument feedback; The system receives environmental disturbance factors, collision force vectors, and control commands, constructs a six-degree-of-freedom dynamics solver, uses Newton-Euler equations to describe vehicle motion, and uses quaternions to represent attitude to avoid gimbal lock. The calculated results are converted into Force / Torque inputs for Actor components via the Unreal Engine physics interface, driving the high-fidelity instrument model to respond in real time and simultaneously updating the visual feedback in the rendering engine.

[0041] Based on environmental disturbance factors, virtual collision forces, and control commands, the six-degree-of-freedom state vector is coupled to the Unreal Engine interface in real time to drive high-fidelity instrument feedback. This involves the following steps: S51: Receives environmental disturbance factors, virtual collision forces, and control instructions, and uses the Newton-Euler equation to solve the six-degree-of-freedom state vector; The environmental disturbance factor, virtual collision force, and control instructions are received synchronously through a high-speed data pipeline. The external force is synthesized in the body coordinate system to calculate the total force and total torque:

[0042] in represents the density of the aircraft; g represents the acceleration of gravity, the direction of which points to the center of the earth; V represents the volume domain of the aircraft; S represents the surface area of ​​the aircraft; P represents the surface static pressure; n represents the unit external normal vector of the surface element; the viscous stress tensor, which describes the distribution of viscous forces, such as air friction; represents the environmental disturbance factor; represents the virtual collision force; t represents the time t; represents the instantaneous change of the virtual collision force at time t; r represents the position vector from the surface element to the center of mass, pointing from the center of mass to the surface element; The continuous dynamics equations are converted into state space differential form using the Newton-Euler equations:

[0043] It represents the time rate of change of linear velocity in the body coordinate system b; is the linear velocity vector in the body coordinate system; m represents the inertia scalar, describing the translational inertia; is the total force acting on the aircraft; represents the angular velocity vector in the body coordinate system b; Momentum vector in the body coordinate system; Coriolis inertial force is the virtual inertial force of a moving object in a rotating system, reflecting the coupling effect of the reference system rotation; It represents the time rate of change of angular velocity in the body coordinate system b; I represents the inertia tensor; represents the inverse matrix of the inertia tensor; is the total torque acting on the aircraft; is the angular momentum vector in the body coordinate system; It represents the gyroscopic torque, which is the inertial torque generated by rotational coupling. It is formed by the cross product of angular momentum and angular velocity, reflecting the coupling effect of the inertia tensor. The fourth-order Runge–Kutta method (RK4) is used for numerical integration to iteratively calculate the slope and update the state with a fixed step size.

[0044] After completing the status update, the system performs coordinate system conversion and output processing. Through the direction cosine matrix Convert to Earth's inertial frame:

[0045] in represents the linear velocity of the inertial frame; Represents the direction cosine matrix of the inertial coordinate system-body coordinate system; Represents the inverse transformation matrix from the inertial coordinate system to the body coordinate system; is the linear velocity vector in the body coordinate system.

[0046] Angular velocity Convert to Euler angles by quaternion integration , to avoid singularities:

[0047] in, Indicates the rate of change of the quaternion q over time; q represents the unit quaternion, describing the attitude of the aircraft; 0 represents the scalar part, because angular velocity is a vector, corresponding to the 'pure vector form' of the quaternion; represents the angular velocity vector in the body coordinate system b; Represents the transpose operator, which converts row vectors into column vectors to facilitate matrix operations.

[0048] The final output is the six-degree-of-freedom state vector:

[0049] Represents the position vector in the inertial coordinate system, describing the absolute position of the aircraft in the inertial space; Represents the linear velocity vector in the inertial coordinate system, describing the translational velocity of the aircraft in the inertial space; An angle vector representing the body's posture; Represents the angular velocity vector in the body coordinate system b.

[0050] S52: Integrate dynamic parameters into Unreal Engine, couple Unreal Engine interfaces in real time, and drive high-fidelity instrument feedback; The 6DOF state vector is received through the API and passed to Unreal Engine's physics engine module. Data conversion and format adaptation are performed through the plugin interface. Using Unreal Engine's Chaos physics engine, linear and angular velocities are converted into the simulated object's motion state. Position and Euler angles are used to update the object's position and orientation in the scene.

[0051] Unreal Engine's material and lighting systems are used to update the aircraft model's surface reflections and ambient occlusion effects in real time based on the aircraft's current motion state and ambient lighting conditions, enhancing visual realism. In rendering the cockpit instrument panel, dynamic state parameters are mapped to the virtual instrument pointer angles and digital readouts, achieving high-fidelity instrument display through Unreal Engine's UI rendering system.

[0052] S6: Generate a training defect heat map based on instrument feedback data, triggering the adaptive optimization strategy to iterate platform parameters; Collect instrument feedback data and operator performance data, and use data visualization techniques to create heat maps to highlight areas of training deficiency. Based on heat map analysis, trigger an adaptive optimization strategy, apply reinforcement learning algorithms to dynamically adjust training parameters, and conduct closed-loop iterations.

[0053] Based on the instrument feedback data, a training defect heat map is generated, the adaptive optimization strategy is triggered, and the platform parameters are iterated, including the following sub-steps: S61: Collect instrument feedback data, obtain operator performance data, and use data visualization technology to create heat maps to highlight areas of deficiency in training; By capturing instrument feedback and operator interaction data, the data streams are synchronized by timestamp and stored in a circular buffer. Preprocessing is performed on a continuous 5-second data window: IMU and virtual instrument data are first fused through Kalman filtering to compensate for sensor noise, and then dynamic features are extracted.

[0054] Based on the preprocessing results, a multidimensional feature space is constructed: time series data is mapped into a three-dimensional tensor of [operation instructions, state responses, error indicators]. A density clustering algorithm is used to discover abnormal clusters in the feature space, the clustering quality is evaluated using the silhouette coefficient, and defect types are defined in combination with domain knowledge.

[0055] Finally, the three-dimensional feature space is projected onto a two-dimensional operating plane, and kernel density estimation is used to calculate the defect probability density distribution. The cumulative defect score is calculated over time, and time weights are added. Finally, the heat map data is rendered to the training interface via the Canvas API, and markers and quantitative indicators are overlaid on the defect areas, forming a multi-dimensional, interactive visualization of training defects.

[0056] S62: Based on heat map analysis, trigger the adaptive optimization strategy, apply the reinforcement learning algorithm to dynamically adjust the training parameters, and perform closed-loop iteration.

[0057] The areas and links where operators frequently make mistakes are extracted from the heat map, and the data is used as input to calculate the operator's performance score in different scenarios through the reward function in the reinforcement learning algorithm.

[0058] Utilizing a reinforcement learning model, the system uses operator performance as feedback to update training parameters and optimize the difficulty and complexity of training scenarios, ensuring the adaptability and effectiveness of the training process. The optimized training parameters are applied to the training scenarios, adjusting environmental disturbance factors, virtual collision forces, and control command generation logic in real time to improve training effectiveness.

[0059] Corresponding to the above embodiment, an embodiment of the present invention provides a computer storage medium, comprising: at least one memory and at least one processor; The memory is used to store one or more program instructions; The processor is used to run one or more program instructions to execute a method for constructing an aircraft control simulation platform.

[0060] Corresponding to the above embodiment, an embodiment of the present invention provides a computer-readable storage medium, which contains one or more program instructions, and the one or more program instructions are used by a processor to execute a method for constructing an aircraft control simulation platform.

[0061] The embodiments disclosed in the present invention provide a computer-readable storage medium, in which computer program instructions are stored. When the computer program instructions are executed on a computer, the computer executes the above-mentioned method for constructing an aircraft control simulation platform.

[0062] In the embodiments of the present invention, the processor may be an integrated circuit chip having signal processing capabilities. The processor may be a general-purpose processor, a digital signal processor (DSP), an application-specific integrated circuit (ASIC), a field programmable gate array (FPGA), or other programmable logic devices, discrete gate or transistor logic devices, or discrete hardware components.

[0063] The methods, steps, and logic diagrams disclosed in the embodiments of the present invention can be implemented or executed. A general-purpose processor can be a microprocessor or any conventional processor. The steps of the methods disclosed in the embodiments of the present invention can be directly implemented and executed by a hardware decoding processor, or by a combination of hardware and software modules within the decoding processor. The software modules can be located in a storage medium well-established in the art, such as random access memory, flash memory, read-only memory, programmable read-only memory, electrically erasable programmable memory, or registers. The processor reads the information from the storage medium and, in conjunction with its hardware, completes the steps of the aforementioned methods.

[0064] The storage medium may be a memory and may be, for example, a volatile memory or a nonvolatile memory, or may include both volatile and nonvolatile memory.

[0065] Among them, the non-volatile memory can be a read-only memory (ROM), a programmable read-only memory (PROM), an erasable programmable read-only memory (EPROM), an electrically erasable programmable read-only memory (EEPROM), or a flash memory.

[0066] Volatile memory may be random access memory (RAM), which is used as an external cache memory. By way of example and not limitation, many forms of RAM are available, such as static random access memory (SRAM), dynamic random access memory (DRAM), synchronous dynamic random access memory (SDRAM), double data rate synchronous dynamic random access memory (DDRSDRAM), enhanced synchronous dynamic random access memory (ESDRAM), synchronous link dynamic random access memory (SLDRAM), and direct RAM bus random access memory (DRRAM).

[0067] The storage media described in the embodiments of the present invention are intended to include, but are not limited to, these and any other suitable types of memory.

[0068] Those skilled in the art will appreciate that in one or more of the above examples, the functions described herein can be implemented using a combination of hardware and software. When software is used, the corresponding functions can be stored in a computer-readable medium or transmitted as one or more instructions or codes on a computer-readable medium. Computer-readable media include computer storage media and communication media, wherein communication media includes any medium that facilitates the transmission of computer programs from one place to another. The storage medium can be any available medium that can be accessed by a general-purpose or special-purpose computer.

[0069] The specific implementation methods described above further illustrate the objectives, technical solutions and beneficial effects of the present invention in detail. It should be understood that the above description is only a specific implementation method of the present invention and is not intended to limit the scope of protection of the present invention. Any modifications, equivalent replacements, improvements, etc. made on the basis of the technical solution of the present invention should be included in the scope of protection of the present invention.

Claims

1. A method for constructing an aircraft control simulation platform, characterized in that: include: Connect virtual reality equipment, configure a millimeter-level spatial positioning system, build a multi-channel voice array and infrared eye tracking module, and obtain a structured data set; Construct a multi-dimensional terrain model, integrate cross-scale meteorological particle systems, and dynamically obtain environmental disturbance factors; Combined with millimeter-level spatial positioning data to generate a dynamic database of spatial obstacles for virtual collision detection; Use long short-term memory (LSTM) neural networks to make eye movement-gesture collaborative decisions and identify control commands with high confidence; Based on environmental disturbance factors, virtual collision forces, and control commands, the six-degree-of-freedom state vector is coupled with the Unreal Engine interface in real time to drive high-fidelity instrument feedback. Generate a training defect heat map based on instrument feedback data, triggering an adaptive optimization strategy to iterate platform parameters.

2. The method for constructing an aircraft control simulation platform according to claim 1, characterized in that: Connect virtual reality equipment, configure a millimeter-level spatial positioning system, build a multi-channel voice array and infrared eye tracking module, and obtain a structured data set, including: Equipped with a millimeter-level spatial positioning system, it tracks the operator's position and movements with millimeter-level precision, captures eye movement data in real time at a non-invasive and high-frequency manner, and simultaneously analyzes the coordinates of the gaze point and physiological micro-responses; The positioning, speech, and eye movement data are integrated and the timestamps are aligned with the limb movement acquisition module through the hardware clock synchronization circuit to generate a structured dataset.

3. The method for constructing an aircraft control simulation platform according to claim 1, wherein: Construct a multi-dimensional terrain model, integrate cross-scale meteorological particle systems, dynamically obtain environmental disturbance factors, and generate dynamic environmental disturbance factors, including: Access millimeter-level spatial positioning data, integrate geographic information system elevation data, and build a multi-dimensional terrain model; Combined with the cross-scale meteorological particle system, the behavior of meteorological particles from the molecular level to the regional level is simulated, and environmental disturbance factors are dynamically generated.

4. The method for constructing an aircraft control simulation platform according to claim 1, wherein: Combined with millimeter-level spatial positioning data, a dynamic database of spatial obstacles is generated to perform virtual collision detection, including: Based on millimeter-level spatial positioning data, a dynamic database of spatial obstacles is constructed and updated in real time to form an interactive physical rule set; Perform millimeter-level virtual collision force detection, calculate force feedback of collision between objects, output virtual collision vector and feedback.

5. The method for constructing an aircraft control simulation platform according to claim 1, wherein: Through the long short-term memory (LSTM) neural network, eye movement and gesture collaborative decision-making is carried out to identify high-confidence control commands, including: Integrate structured datasets as neural network input and use neural network training models to identify the temporal dependencies between eye movements and gestures; According to the trained model, the probability threshold is filtered and instructions with confidence higher than the preset value are output.

6. The method for constructing an aircraft control simulation platform according to claim 1, wherein: Based on environmental disturbance factors, virtual collision forces, and control commands, the six-degree-of-freedom state vector is coupled with the Unreal Engine interface in real time to drive high-fidelity instrument feedback, including: Receive environmental disturbance factors, virtual collision forces and control instructions, and use Newton-Euler equations to solve the six-degree-of-freedom state vector; Integrate dynamics parameters into Unreal Engine, couple to Unreal Engine interfaces in real time, and drive high-fidelity instrument feedback.

7. The method for constructing an aircraft control simulation platform according to claim 1, characterized in that: Generate a training defect heat map based on instrument feedback data, triggering an adaptive optimization strategy to iterate platform parameters, including: Collect instrument feedback data, obtain operator performance data, and use data visualization technology to create heat maps to highlight deficient areas in training; Based on heat map analysis, the adaptive optimization strategy is triggered, and the reinforcement learning algorithm is applied to dynamically adjust the training parameters for closed-loop iteration.

Images