A method for early warning of hard landing events based on full-motion simulator data

Abstract

This invention discloses a hard landing event early warning method based on full-motion simulator data, comprising the following steps: first, constructing full-motion simulator simulation data and building a multi-physics coupled hard landing dynamics benchmark model; then, using the FLTO falling leaf eddy optimization algorithm to identify unknown parameters of the dynamics benchmark model; performing real-time hard landing early warning based on energy safety margin decay; using the margin zeroing time for prediction and graded early warning; finally, introducing a control delay compensation mechanism, using the predicted aircraft state after delay compensation as input for energy margin and decay rate, and outputting the final hard landing early warning result. This invention can quantify the change in glide energy margin in real time, dynamically determine the hard landing risk level and issue warnings, assist pilots in correcting control and intervening in energy anomalies, reduce fuselage damage and maintenance costs, prevent safety hazards, and provide technical support for proactive safety management during the landing process.

Description

Technical Field

[0001] This invention relates to the field of aviation flight simulation and flight safety early warning technology, and in particular to a method for early warning of hard landing events based on full-motion simulator data. Background Technology

[0002] The landing phase of a civil aircraft is widely recognized as one of the riskiest segments of flight, with a significantly higher probability of accidents and unsafe events compared to other phases. Statistics show that hard landings account for nearly 20% of unsafe landing incidents. In the field of hard landing incident prevention, monitoring parameters related to risk-causing factors is necessary to achieve early risk warning. Accurate diagnosis of hard landings is a crucial step in reducing their harm.

[0003] The current methods used by domestic airlines to diagnose hard landings have the following problems: 1) The decision to make a hard landing is mainly based on the pilot's feedback and the G-value filtering of daily flight quality monitoring.

[0004] 2) Pilots and maintenance personnel generally believe that the current judgment standards and methods are not rigorous and the judgment measures are unreasonable, resulting in problems such as low efficiency and low accuracy, which increases both operating costs and potential risks.

[0005] 3) When reviewing QAR Quick Access Recorder data, only the recorded acceleration values ​​are considered to be within limits, ignoring the analysis of other parameters and environmental factors as causes of hard landing. This approach is overly simplistic and lacks effective trend analysis.

[0006] Currently, support vector machines, Markov methods, LSTM, Bow-Tie models, and machine learning are mainly used to train hard landing risk identification models and optimize parameter weights and hard landing warning criteria. However, due to the limited number of actual unsafe incident cases, there are not enough cases to analyze the risk influencing factors. Summary of the Invention

[0007] The purpose of this invention is to provide a method for early warning of hard landing events based on full-motion simulator data. The full-motion simulator has the same aircraft data package as the real aircraft, and the output system parameters are consistent with those of the real aircraft. By using the full-motion simulator, the impact of various parameters on the risk of unsafe events can be simulated, thereby comprehensively sorting out the causal factors, finding the key factors, and providing key support for risk warning.

[0008] To achieve the above objectives, the present invention provides a method for early warning of re-landing events based on full-motion simulator data, comprising the following steps: S1. Construct simulation data for a full-motion simulator; S2. Establish a multi-physics coupled hard landing dynamics benchmark model; S3. The unknown parameters of the dynamic baseline model in S2 are identified using the FLTO falling leaf eddy optimization algorithm. S4. Real-time early warning of hard landing based on the decay of energy safety margin; S5, Using margin zeroing time Conduct forecasting and tiered early warning; S6. Introduce a control delay compensation mechanism, using the predicted aircraft state after delay compensation as the input for energy margin and decay rate, and output the final hard landing warning result.

[0009] Preferably, S1 includes: S11. Construct random data as the input dataset. The input dataset is the initial flight state dataset, including the control stick, control wheel, rudder, flaps, wind speed, temperature, and barometric altitude. S12. Establish corresponding full-motion simulator databases for different pilots, record training data and test data, including the timing of pilots pulling the stick and releasing the throttle, and perform statistical analysis on them; S13. Import the initial control parameters, environmental parameters and data from the full-motion simulator database into the full-motion simulator, use the full-motion simulator data package to calculate the aircraft state, and output the calculation results. S14. Complete the required parameters for the full-motion simulator. Based on the QTG platform of the Flight Simulator Qualification Test Guide, import the initial control parameters, environmental parameters and aircraft state parameters into the full-motion simulator, run the simulation platform, and output the calculation results. S15. Divide the output data of the full-motion simulator into control data, environmental data and aircraft status data. Control data includes control stick travel and control wheel travel. Environmental data includes ambient temperature. Aircraft status data includes at least flap status, airspeed, pressure altitude, vertical acceleration, rate of descent, angle of attack, bank angle and landing gear strut compression travel.

[0010] Preferably, in S14, the completion operation includes: Considering the sensor size, the angle of attack (AOA) is corrected, which is expressed as: ; in, To correct the angle of attack (AOA); This is the average value from the sensor. , These are class constants determined based on the aircraft's physical properties and different flight conditions; For the sensor distance, take 1 / 2 of the body length; For the aircraft pitch rate, Airspeed; Calculate the vertical component of the downwind direction in a geodetic coordinate system. get: ; in, Airspeed, The pitch angle, To correct the angle of attack, For slope, This represents the vertical velocity.

[0011] Preferably, S2 includes: S21. Calculate air density The formula is: ; in, This refers to atmospheric pressure altitude. The ambient temperature; Calculate the lift coefficient The formula is: ; in, It is the lift coefficient affected by the flaps. It is the lift coefficient affected by the angle of attack; Calculate real-time aerodynamic lift The formula is as follows: ; in, The velocity of an object relative to a fluid; Wing area; The lift coefficient; air density; S22. Construct a nonlinear dynamic model of the landing gear system and define the vertical reaction force of the landing gear on the fuselage. The vertical reaction force is composed of the air spring force and the hydraulic damping force of the shock absorber strut. The expression is: ; in, For the compression stroke of the shock absorber strut; Compression rate; It is a nonlinear variable stiffness function that characterizes the gas compressibility. It is a nonlinear variable damping function that characterizes the damping properties of the oil fluid. When the aircraft's bank angle When the load exceeds a preset threshold, the main landing gear on one side will touch down first and bear the main load. The amplitude is halved; S23. Construct an energy model of the aircraft in the vertical direction, and calculate the total mechanical energy of the aircraft in the vertical plane. The sum of potential energy and kinetic energy is expressed as: ; in, For aircraft quality, The rate of decline; For total mechanical energy The instantaneous rate of change of the aircraft's total energy is obtained by taking the time derivative of the formula. : ; According to Newton's second law Substituting into the above equation, we get: ; S24. Establish a boundary model for the safe power absorption of the landing gear, and define the safe power absorption of the landing gear structure. Characterized by the vertical reaction force currently provided by the landing gear Compression rate The fitting function for the remaining stroke of the support is given by the following formula: ; in, This represents the maximum usable stroke of the shock absorber strut.

[0012] Preferably, in S3, the full-motion simulator outputs a true vertical acceleration. ,make A parameter identification model was established, and the unknown parameters of the dynamic baseline model in S2 were identified using the FLTO falling leaf eddy optimization algorithm, including the following steps: S31. Population initialization: Set the parameter vector of the particles to be identified. Parameter constraint range: ; , These are the upper and lower constraint boundaries of the particle parameter vector; Set population size Maximum number of iterations The population was initialized using SOBOL low-difference sequences; S32. Construct the fitness function to calculate the vertical acceleration using the model. With full-motion simulator real output The root mean square error is used as the fitness function, and the objective is to minimize the fitness value. : ; in, This represents the total number of valid data samples from the simulator. Particle parameter vector The corresponding fitness value, the smaller the value, the higher the parameter identification accuracy; The parameter vector is Time model Calculated vertical acceleration values ​​for the group of samples; For the simulator The true vertical acceleration output of the group of samples; The optimization goal is ; S33, Global Exploration (i.e., the drift phase): Introducing the Levy flight generation perturbation vector: ; The global candidate position update formula is: ; in, Indicates the first The generation The current position of each fallen leaf particle; Indicates the updated candidate locations after global exploration; Indicates the first The global optimal position of the generation population; Let Lévy's flight random perturbation vector satisfy... , This is the global exploration step size coefficient; For element-wise multiplication of vectors; The optimal position for the entire population; S34, Local refinement, i.e., the hovering stage, when particle fitness is continuous The change in the generation is less than the threshold. Initiate a logarithmic shrinkage fine search: ; in, For the first The generation The best historical position of each falling leaf particle; For the first Sub-particles Local search variance of a falling leaf particle; This is a parameter for controlling the logarithmic contraction rate; This represents the current iteration number. This represents the maximum number of iterations. for A 3D random vector whose components follow a uniform distribution ; It is a logarithmic contraction term; This is the local search step size coefficient; For an individual, the best historical position; S35. Escaping local optima, i.e., the flipping phase, when the global optimal solution is continuous. If no significant optimization is found, the flip mechanism is triggered, and three types of complementary perturbation strategies are enabled, as follows: The reflection boundary is flipped, as shown in the following formula: ; In the formula, ; , The upper and lower bounds of the parameter vector; This is the random reflection scaling factor; Gaussian mutagenesis flipping, the formula is as follows: ; In the formula: The coefficient of variation intensity It follows a standard Gaussian distribution; Differential evolution crossover flips, randomly selecting three distinct particles. Generate mutation vectors: ; Where: Differential weighting factor , , , For indexes of distinct population individuals; S36. Introduce a random perturbation mechanism to execute high-frequency low-amplitude perturbations and low-frequency large-amplitude perturbations with preset trigger probabilities; S37, Reaching the maximum number of iterations Or global optimal fitness continuous When the algebraic change is less than the convergence threshold, output the optimal parameter vector: .

[0013] Preferably, S4 includes: S41. Real-time calculation of energy safety margin, definition The energy safety margin of the aircraft To ensure safe power absorption by the landing gear With current total power dissipation The difference is expressed as follows: ; in, This indicates that the landing energy is within a safe and absorbable range; This indicates that the current energy dissipation demand has exceeded the landing gear's safe absorption capacity, posing a high risk of a hard landing; S42. Predicting the rate of energy margin decay: Perform real-time analysis of the energy security margin sequence and calculate the rate of decay of the energy security margin at the current moment. This is used to reflect the rate at which the energy danger situation deteriorates: ; in, For time difference.

[0014] Preferably, in S5, the margin zeroing time is defined. The remaining time before the energy safety margin drops to zero, assuming the current decay rate remains constant, is expressed as: ; Using margin zeroing time Forecasting and Tiered Early Warning: Assuming the decay rate remains constant, calculate the remaining time for the energy margin to reach zero. : Only when hour, It has early warning value; Establish hierarchical early warning rules by combining adaptive optimal thresholds: when The situation is currently in a safe state. when At this time, a warning level is triggered, prompting the pilot to correct flight energy; when At that time, a warning level alert was triggered, indicating a high risk of a hard landing, requiring immediate action. The time required to trigger a warning level alert. The time when the attention-level warning is triggered.

[0015] Preferably, in step S6, the delay time is obtained by linear fitting of the full-motion simulator data. and At the moment of early warning decision-making, a control delay compensation mechanism is introduced, using the predicted aircraft state after control delay compensation as the energy margin. and decay rate The input is the final hard landing warning result, which reflects the actual aircraft energy status after the delay takes effect.

[0016] Therefore, the present invention employs the above-mentioned method for early warning of re-landing events based on full-motion simulator data, and the beneficial effects are as follows: (1) This invention uses a full-motion simulator to generate massive simulation data under multiple operating conditions, which makes up for the shortcomings of insufficient real hard landing accident cases and single QAR data dimension; at the same time, it completes key parameters such as angle of attack and vertical wind, gets rid of the traditional single G value judgment mode, comprehensively covers multiple parameters of control, environment and aircraft status, accurately sorts out the causes of hard landing, and consolidates the data foundation of the early warning model.

[0017] (2) This invention has developed two original biomimetic algorithms, FLTO and RAGDO, which respectively complete the identification of unknown parameters of the model and the adaptive optimization of the warning threshold. It effectively overcomes the problems of traditional intelligent algorithms being prone to getting stuck in local optima and poor adaptability of fixed warning thresholds, reduces the false alarm rate and missed alarm rate, and makes the warning standard adaptable to different aircraft types, pilots and meteorological environments, thereby improving the robustness and generalization ability of the system.

[0018] (3) This invention builds an energy margin decay early warning model based on the aircraft energy dissipation mechanism, realizes early graded early warning before hard landing, and changes the traditional post-event review control mode; at the same time, it adds a control delay compensation module to eliminate the early warning error caused by control response lag, which can guide the pilot to correct the operation in time, effectively reduce the risk of hard landing, and reduce aircraft maintenance costs and flight safety hazards.

[0019] The technical solution of the present invention will be further described in detail below with reference to the accompanying drawings and embodiments. Attached Figure Description

[0020] Figure 1 This is an overall flowchart of an embodiment of a hard landing event early warning method based on full-motion simulator data according to the present invention. Detailed Implementation

[0021] The technical solution of the present invention will be further described below with reference to the accompanying drawings and embodiments.

[0022] Unless otherwise defined, the technical or scientific terms used in this invention shall have the ordinary meaning understood by one of ordinary skill in the art to which this invention pertains. The terms "first," "second," and similar terms used in this invention do not indicate any order, quantity, or importance, but are merely used to distinguish different components. Terms such as "comprising" or "including" mean that the element or object preceding the word encompasses the elements or objects listed following the word and their equivalents, without excluding other elements or objects.

[0023] like Figure 1 As shown, a method for early warning of hard landing events based on full-motion simulator data includes the following steps: S1. Construct full-motion simulator simulation data, including: S11. Construct random data as the input dataset. The input dataset is the initial flight state dataset, including the control stick, control wheel, rudder, flaps, wind speed, temperature, and barometric altitude. Each parameter generates a dataset at fixed intervals, such as 0, 5, and 10 knots of wind.

[0024] S12. Establish corresponding full-motion simulator databases for different pilots, record training data and test data, including the timing of pilots pulling the stick and releasing the throttle, and perform statistical analysis on them.

[0025] S13. Import the initial control parameters, environmental parameters and data from the full-motion simulator database into the full-motion simulator, use the full-motion simulator data package to calculate the aircraft state, and output the calculation results. S14. Complete the required parameters (such as horizontal wind) for the full-motion simulator. Based on the QTG platform of the Flight Simulator Qualification Test Guide, import the initial control parameters, environmental parameters and aircraft state parameters into the full-motion simulator, run the simulation platform, and output the calculation results.

[0026] This step of completing the task includes: 1) Correct AOA calculation: ; Considering the sensor size, the corrected angle of attack (AOA) can be expressed as follows: ; in, To correct the angle of attack (AOA); This is the average value from the sensor. , These are class constants determined based on the aircraft's physical properties and different flight conditions; For the sensor distance, take 1 / 2 of the body length; For the aircraft pitch rate, This is airspeed.

[0027] 2) Vertical wind calculation: Since the recorder did not record vertical wind, it is necessary to calculate the vertical component of the wind in the geodetic coordinate system. ,get: ; in, Airspeed, The pitch angle, To correct the angle of attack, For slope, This represents the vertical velocity.

[0028] S15. Data Classification: The output data of the full-motion simulator is divided into control data, environmental data, and aircraft status data. Control data includes control stick travel and control wheel travel. Environmental data includes ambient temperature, etc. Aircraft status data includes at least flap status, airspeed, pressure altitude, vertical acceleration, rate of descent, angle of attack, bank angle, and landing gear strut compression travel.

[0029] S2. Establish a multi-physics coupled re-landing dynamics baseline model, including: S21. Calculate air density The formula is: ; in, This refers to atmospheric pressure altitude. The ambient temperature; Calculate the lift coefficient The formula is: ; in, It is the lift coefficient affected by the flaps. It is the lift coefficient affected by the angle of attack; Calculate real-time aerodynamic lift The formula is as follows: ; in, The velocity of an object relative to a fluid; Wing area; For lift coefficient, The density of air can be calculated using the aforementioned formula.

[0030] S22. Construct a nonlinear dynamic model of the landing gear system and define the vertical reaction force of the landing gear on the fuselage. The vertical reaction force is composed of the air spring force and the hydraulic damping force of the shock absorber strut. The expression is: ; in, For the compression stroke of the shock absorber strut; The compression rate can be directly derived from the rate of descent. Conversion; It is a nonlinear variable stiffness function that characterizes the compressibility of gas, and typically varies with... Increase and increase; It is a nonlinear variable damping function that characterizes the damping properties of the oil and is related to both compression speed and stroke.

[0031] When the aircraft's bank angle When the angle exceeds a preset threshold (e.g., 5°), the main landing gear on one side will touch down first and bear the main load. The amplitude is halved.

[0032] S23. Construct an energy model of the aircraft in the vertical direction, and calculate the total mechanical energy of the aircraft in the vertical plane. The sum of potential energy and kinetic energy is expressed as: ; in, For aircraft quality, The rate of decline.

[0033] For the above total mechanical energy The instantaneous rate of change of the aircraft's total energy is obtained by taking the time derivative of the formula. : .

[0034] According to Newton's second law Substituting into the above equation, we get: .

[0035] S24. Establish a boundary model for the safe power absorption of the landing gear, and define the safe power absorption of the landing gear structure. Characterized by the vertical reaction force currently provided by the landing gear Compression rate The fitting function for the remaining stroke of the support is given by the following formula: ; in, For the maximum available travel of the shock absorber strut; when When approaching the limit, the compression speed is too high, or the stroke is about to be exhausted. The value will decrease dramatically; this functional relationship can be pre-fitted from test data of a simulator under different operating conditions.

[0036] S3. The unknown parameters of the dynamic baseline model in S2 are identified using the FLTO falling leaf eddy optimization algorithm.

[0037] In this invention, , , , For unknown parameters; the full-motion simulator outputs the true vertical acceleration. ,make By establishing a parameter identification model, this invention proposes a novel original biomimetic intelligent algorithm—Fall-en-LeavesTurbulenceOptimizer (FLTO). This algorithm is based on the multimodal motion of fallen leaves in complex turbulence as a biological prototype.

[0038] (I) Biological biomimetic principles The Falling Leaf Eddy Optimizer (FLTO) is based on the motion patterns of fallen leaves in atmospheric turbulence, and is divided into four typical motion modes, each corresponding to a different search logic in the algorithm: Horizontal drift phase: Fallen leaves drift a long distance under the influence of airflow, corresponding to the global exploration of the algorithm.

[0039] Circling phase: Fallen leaves enter a local vortex with small circular motion, corresponding to a fine local search in the algorithm.

[0040] Flipping stage: When the airflow changes abruptly or the leaves are close to the ground, they become unstable and flip over, and the corresponding algorithm escapes the local optimal disturbance.

[0041] Random perturbation: Intermittent atmospheric turbulence causes slight shaking of fallen leaves, corresponding to global random perturbation in the algorithm.

[0042] FLTO integrates multimodal motion features, possessing strong global search capabilities, high local convergence accuracy, and resistance to local optima.

[0043] This invention employs the FLTO (Falling Leaf Whirl) optimization algorithm to identify unknown parameters in the dynamic baseline model of S2, including the following steps: S31. Population initialization: Set the parameter vector of the particles to be identified. Parameter constraint range: ; , These are the upper and lower constraint boundaries of the particle parameter vector; Set population size Maximum number of iterations The population was initialized using SOBOL low-difference sequences to ensure uniform coverage of the search space.

[0044] S32. Construct the fitness function to calculate the vertical acceleration using the model. With full-motion simulator real output The root mean square error is used as the fitness function, and the objective is to minimize the fitness value. : ; in, This represents the total number of valid data samples from the simulator. Particle parameter vector The corresponding fitness value, the smaller the value, the higher the parameter identification accuracy; The parameter vector is Time model Calculated vertical acceleration values ​​for the group of samples; For the simulator The true vertical acceleration output of the group of samples; The optimization goal is .

[0045] S33, Global Exploration (i.e., the drift phase): Introducing the Levy flight generation perturbation vector: .

[0046] The global candidate position update formula is: ; in, Indicates the first The generation The current position of each fallen leaf particle; Indicates the updated candidate locations after global exploration; Indicates the first The global optimal position of the generation population; Let Lévy's flight random perturbation vector satisfy... , This is the global exploration step size coefficient; For element-wise multiplication of vectors; This is the globally optimal position for the population.

[0047] S34, Local refinement, i.e., the hovering stage, when particle fitness is continuous The change in the generation is less than the threshold. Initiate a logarithmic shrinkage fine search: ; in, For the first The generation The best historical position of each falling leaf particle; For the first Sub-particles Local search variance of a falling leaf particle; This is a parameter for controlling the logarithmic contraction rate; This represents the current iteration number. This represents the maximum number of iterations. for A 3D random vector whose components follow a uniform distribution ; This is a logarithmic shrinkage term that gradually narrows the local search range as iterations progress, thereby improving convergence accuracy. This is the local search step size coefficient; This represents the optimal position in an individual's history.

[0048] S35. Escaping local optima, i.e., the flipping phase, when the global optimal solution is continuous. If no significant optimization is found, the flip mechanism is triggered, and three types of complementary perturbation strategies are enabled, as follows: (1) The reflection boundary is flipped, and the formula is as follows: ; In the formula, ; , The upper and lower bounds of the parameter vector; This is the random reflection scaling factor. When a particle goes out of bounds, it is remapped to the feasible region of parameters through mirror reflection, achieving a sudden positional change and escaping the local optimum.

[0049] (2) Gaussian mutation inversion, the formula is as follows: ; In the formula: The coefficient of variation intensity It follows a standard Gaussian distribution.

[0050] (3) Differential evolution crossover flip, randomly select three distinct particles Generate mutation vectors: ; Where: Differential weighting factor , , , For distinct population individuals; compared with the current particle No repetition; after generating the mutation vector, it is crossed with the original particle position to complete the differential flip perturbation and enrich the population diversity.

[0051] S36. Introduce a random perturbation mechanism to execute high-frequency low-amplitude perturbations and low-frequency large-amplitude perturbations with preset trigger probabilities: High-frequency low-amplitude disturbance: trigger probability .

[0052] Low-frequency large-amplitude disturbances: each Execution by proxy, trigger probability Effect on population proportion Individuals.

[0053] S37. Convergence Judgment and Output: Termination condition: Reaching the maximum number of iterations Or global optimal fitness continuous The iteration ends when the algebraic change is less than the convergence threshold, and the optimal parameter vector is output. .

[0054] FLTO uses the turbulent motion of falling leaves as its sole biomimetic prototype and integrates four unique mechanisms: global exploration of Levy's flight, logarithmic contraction of local search, multi-strategy flip escape, and dual random perturbation. Compared with least squares and traditional swarm intelligence algorithms, it can effectively avoid local optima and improve the accuracy and robustness of nonlinear parameter identification.

[0055] S4. Real-time early warning for hard landing based on energy safety margin decay, including: S41. Real-time calculation of energy safety margin, definition The energy safety margin of the aircraft To ensure safe power absorption by the landing gear With current total power dissipation The difference is expressed as follows: ; in, This indicates that the landing energy is within a safe and absorbable range; This indicates that the current energy dissipation demand has exceeded the landing gear's safe absorption capacity, posing a significant risk of a hard landing.

[0056] S42. Predicting the rate of energy margin decay: Perform real-time analysis of the energy security margin sequence and calculate the rate of decay of the energy security margin at the current moment. This rate is used to reflect the rate at which the energy danger situation deteriorates: ; in, For time difference.

[0057] S5, Define margin zeroing time The remaining time before the energy safety margin drops to zero, assuming the current decay rate remains constant, is expressed as: ; Note that here... Calculation only when A value less than 0 (meaning the margin is decreasing) has warning significance.

[0058] Further use of margin zeroing time Forecasting and Tiered Early Warning: Assuming the decay rate remains constant, calculate the remaining time for the energy margin to reach zero. : Only when hour, It has early warning value.

[0059] Establish hierarchical early warning rules by combining adaptive optimal thresholds: when The situation is currently in a safe state.

[0060] when At that time, a warning level alert is triggered, prompting the pilot to correct flight energy.

[0061] when At that time, a warning level alert was triggered, indicating an extremely high risk of a hard landing, requiring immediate action. The time required to trigger a warning level alert. The time when the attention-level warning is triggered.

[0062] S6. Introduce a control delay compensation mechanism, using the predicted aircraft state after delay compensation as the input for energy margin and decay rate, and output the final hard landing warning result.

[0063] Given the time delay between input from the control stick / wheel and the resulting change in angle of attack / gradient, this delay time needs to be obtained through linear fitting of the full-motion simulator data. and At the moment of early warning decision-making, a control delay compensation mechanism is introduced, using the predicted aircraft state after control delay compensation as the energy margin. and decay rate The input is the final hard landing warning result, which reflects the actual aircraft energy status after the delay takes effect.

[0064] Therefore, the present invention adopts the above-mentioned method for early warning of hard landing events based on full-motion simulator data, which can realize early warning of hard landing in advance, promote the transformation of landing safety from post-event review to pre-event proactive prevention and control, effectively correct the early warning deviation caused by control lag, reduce the occurrence rate of hard landing, save maintenance expenses, and improve the safety of approach and landing operations.

[0065] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and not to limit them. Although the present invention has been described in detail with reference to preferred embodiments, those skilled in the art should understand that modifications or equivalent substitutions can still be made to the technical solutions of the present invention, and these modifications or equivalent substitutions cannot cause the modified technical solutions to deviate from the spirit and scope of the technical solutions of the present invention.

Claims

1. A method for early warning of heavy landing events based on full-motion simulator data, characterized in that, Includes the following steps: S1. Construct simulation data for a full-motion simulator; S2. Establish a multi-physics coupled hard landing dynamics benchmark model; S3. The unknown parameters of the dynamic baseline model in S2 are identified using the FLTO falling leaf eddy optimization algorithm. S4. Real-time early warning of hard landing based on the decay of energy safety margin; S5, Use margin zeroing time Conduct forecasting and tiered early warning; S6. Introduce a control delay compensation mechanism, using the predicted aircraft state after delay compensation as the input for energy margin and decay rate, and output the final hard landing warning result.

2. The method for early warning of re-landing events based on full-motion simulator data according to claim 1, characterized in that, S1 includes: S11. Construct random data as the input dataset. The input dataset is the initial flight state dataset, including the control stick, control wheel, rudder, flaps, wind speed, temperature, and barometric altitude. S12. Establish corresponding full-motion simulator databases for different pilots, record training data and test data, including the timing of pilots pulling the stick and releasing the throttle, and perform statistical analysis on them; S13. Import the initial control parameters, environmental parameters and data from the full-motion simulator database into the full-motion simulator, use the full-motion simulator data package to calculate the aircraft state, and output the calculation results. S14. Complete the required parameters for the full-motion simulator. Based on the QTG platform of the Flight Simulator Qualification Test Guide, import the initial control parameters, environmental parameters and aircraft state parameters into the full-motion simulator, run the simulation platform, and output the calculation results. S15. Divide the output data of the full-motion simulator into control data, environmental data and aircraft status data. Control data includes control stick travel and control wheel travel. Environmental data includes ambient temperature. Aircraft status data includes at least flap status, airspeed, pressure altitude, vertical acceleration, rate of descent, angle of attack, bank angle and landing gear strut compression travel.

3. The method for early warning of re-landing events based on full-motion simulator data according to claim 2, characterized in that, In S14, the completion operation includes: Considering the sensor size, the angle of attack (AOA) is corrected, which is expressed as: ； in, To correct the angle of attack (AOA); This is the average value from the sensor. , These are class constants determined based on the aircraft's physical properties and different flight conditions; For the sensor distance, take 1 / 2 of the body length; For the aircraft pitch rate, Airspeed; Calculate the vertical component of the downwind direction in a geodetic coordinate system. get: ； in, Airspeed, The pitch angle, To correct the angle of attack, For slope, This represents the vertical velocity.

4. The method for early warning of re-landing events based on full-motion simulator data according to claim 3, characterized in that, S2 include: S21. Calculate air density The formula is: ； in, This refers to atmospheric pressure altitude. Ambient temperature; Calculate the lift coefficient The formula is: ； in, It is the lift coefficient affected by the flaps. It is the lift coefficient affected by the angle of attack; Calculate real-time aerodynamic lift The formula is as follows: ； in, The velocity of an object relative to a fluid; Wing area; The lift coefficient; air density; S22. Construct a nonlinear dynamic model of the landing gear system and define the vertical reaction force of the landing gear on the fuselage. The vertical reaction force is composed of the air spring force and the hydraulic damping force of the shock absorber strut. The expression is: ； in, For the compression stroke of the shock absorber strut; Compression rate; It is a nonlinear variable stiffness function that characterizes the gas compressibility. It is a nonlinear variable damping function that characterizes the damping properties of the oil fluid. When the aircraft's bank angle When the load exceeds a preset threshold, the main landing gear on one side will touch down first and bear the main load. The amplitude is halved; S23. Construct an energy model of the aircraft in the vertical direction, and calculate the total mechanical energy of the aircraft in the vertical plane. The sum of potential energy and kinetic energy is expressed as: ； in, For aircraft quality, The rate of decline; For total mechanical energy The instantaneous rate of change of the aircraft's total energy is obtained by taking the time derivative of the formula. : ； According to Newton's second law Substituting into the above equation, we get: ； S24. Establish a boundary model for the safe power absorption of the landing gear, and define the safe power absorption of the landing gear structure. Characterized by the vertical reaction force currently provided by the landing gear Compression rate The fitting function for the remaining stroke of the support is given by the following formula: ； in, This represents the maximum usable stroke of the shock absorber strut.

5. The method for early warning of re-landing events based on full-motion simulator data according to claim 4, characterized in that, In S3, the full-motion simulator outputs realistic vertical acceleration. ,make A parameter identification model was established, and the unknown parameters of the dynamic baseline model in S2 were identified using the FLTO falling leaf eddy optimization algorithm. The steps included: S31. Population initialization: Set the parameter vector of the particles to be identified. Parameter constraint range: ; , These are the upper and lower constraint boundaries of the particle parameter vector; Set population size Maximum number of iterations The population was initialized using SOBOL low-difference sequences; S32. Construct a fitness function to calculate vertical acceleration using the model. With the real output of the full-motion simulator The root mean square error is used as the fitness function, and the objective is to minimize the fitness value. : ； in, This represents the total number of valid data samples from the simulator. Particle parameter vector The corresponding fitness value, the smaller the value, the higher the parameter identification accuracy; The parameter vector is Time model Calculated vertical acceleration values ​​for the group of samples; For the simulator The true vertical acceleration output of the group of samples; The optimization goal is ; S33, Global Exploration (i.e., the drift phase): Introducing the Levy flight generation perturbation vector: ; The global candidate position update formula is: ； in, Indicates the first The generation The current position of each fallen leaf particle; Indicates the updated candidate locations after global exploration; Indicates the first The globally optimal position of the generation population; Let Lévy's flight random perturbation vector satisfy... , This is the global exploration step size coefficient; For element-wise multiplication of vectors; The optimal position for the entire population; S34, Local refinement, i.e., the hovering stage, when particle fitness is continuous The change in the generation is less than the threshold. Initiate a logarithmic shrinkage fine search: ； in, For the first The generation The best historical position of each falling leaf particle; For the first Sub-particles Local search variance of a falling leaf particle; This is a parameter for controlling the logarithmic contraction rate; This represents the current iteration number. This represents the maximum number of iterations. for A 3D random vector whose components follow a uniform distribution ; It is a logarithmic contraction term; This is the local search step size coefficient; For an individual, the best historical position; S35. Escaping local optima, i.e., the flipping phase, when the global optimal solution is continuous. If no significant optimization is found, the flip mechanism is triggered, and three types of complementary perturbation strategies are enabled, as follows: The reflection boundary is flipped, as shown in the following formula: ； In the formula, ; , The upper and lower bounds of the parameter vector; This is the random reflection scaling factor; Gaussian mutagenesis flipping, the formula is as follows: ； In the formula: The coefficient of variation intensity It follows a standard Gaussian distribution; Differential evolution crossover flips, randomly selecting three distinct particles. Generate mutation vectors: ； Where: Differential weighting factor , , , For indexes of distinct population individuals; S36. Introduce a random perturbation mechanism to execute high-frequency low-amplitude perturbations and low-frequency large-amplitude perturbations with preset trigger probabilities; S37, Reaching the maximum number of iterations Or global optimal fitness continuous When the algebraic change is less than the convergence threshold, output the optimal parameter vector: 。 6. The method for early warning of re-landing events based on full-motion simulator data according to claim 5, characterized in that, S4 include: S41. Real-time calculation of energy safety margin, definition The energy safety margin of the aircraft To ensure safe power absorption by the landing gear With current total power dissipation The difference is expressed as follows: ； in, This indicates that the landing energy is within a safe and absorbable range; This indicates that the current energy dissipation demand has exceeded the landing gear's safe absorption capacity, posing a high risk of a hard landing; S42. Predicting the rate of energy margin decay: Perform real-time analysis of the energy security margin sequence and calculate the rate of decay of the energy security margin at the current moment. This is used to reflect the rate at which the energy danger situation deteriorates: ； in, For time difference.

7. The method for early warning of re-landing events based on full-motion simulator data according to claim 6, characterized in that, In S5, the margin zeroing time is defined. The remaining time before the energy safety margin drops to zero, assuming the current decay rate remains constant, is expressed as: ； Using margin zeroing time Forecasting and Tiered Early Warning: Assuming the decay rate remains constant, calculate the remaining time for the energy margin to reach zero. : Only when hour, It has early warning value; Establish hierarchical early warning rules by combining adaptive optimal thresholds: when The situation is currently in a safe state. when At this time, a warning level is triggered, prompting the pilot to correct flight energy; when At that time, a warning level alert was triggered, indicating a high risk of a hard landing, requiring immediate action. The time required to trigger a warning level alert. The time when the attention-level warning is triggered.

8. The method for early warning of re-landing events based on full-motion simulator data according to claim 7, characterized in that, In S6, the delay time is obtained by linear fitting of the full-motion simulator data. and At the moment of early warning decision-making, a control delay compensation mechanism is introduced, using the predicted aircraft state after control delay compensation as the energy margin. and decay rate The input is the final hard landing warning result, which reflects the actual aircraft energy status after the delay takes effect.

Images