Calibration system and method for engine temperature power attenuation based on differential GPS data

Abstract

The present application relates to the technical field of aviation flight test and data processing, and discloses a turbofan temperature power attenuation calibration system and method based on differential GPS data, the calibration system comprising: a multi-source data acquisition module synchronously acquiring flight parameters, differential GPS and airport meteorological data to construct a multi-source parameter matrix, covering multiple coupled interference variables and making up for the missing defects of test bench working conditions; a parameter inversion module calibrating GPS data and inversely calculating the actual thrust and output power of the turbofan during the take-off running stage; a temperature power correlation coupling module matching power and height data under different temperature gradients and establishing a mapping data set; and a calibration engine carrying a multi-layer perception neural network, relying on the data set to nonlinearly correct the theoretical power attenuation curve and generating a temperature adaptive actual power attenuation curve cluster. The present application forms a closed-loop system of acquisition, inversion, modeling and intelligent calibration, adapts to various interferences in real flight, eliminates the system deviation of the theoretical model, and effectively improves the power measurement accuracy of the aero-engine.

Description

Technical Field

[0001] This invention relates to the field of aviation flight testing and data processing technology, specifically to an engine temperature and power attenuation calibration system and method based on differential GPS data. Background Technology

[0002] The accurate calculation of propeller aircraft flight performance is highly dependent on the actual power output characteristics of the aero-engine. The decay law of engine power with ambient temperature, air pressure, and altitude is a core parameter for evaluating aircraft takeoff performance, verifying obstacle clearance capability, and making takeoff safety decisions. Currently, the industry generally uses the power decay curve provided by the engine manufacturer to characterize the relationship between engine power and ambient temperature, air pressure, and altitude. This theoretical power decay curve is mainly derived from idealized bench test environments combined with theoretical calculation formulas. However, the test process eliminates various interference factors in actual flight scenarios, resulting in significant differences from real flight conditions. For example, during actual aircraft test flights and operation, multiple actual factors such as fluctuations in the aerodynamic efficiency of the engine intake system, differences in fuselage cooling airflow organization, batch differences in fuel quality, individual engine assembly errors, and long-term wear and tear combine to cause a significant systematic deviation between the actual power decay characteristics of the engine and the theoretical decay curve provided by the manufacturer. Moreover, this deviation is further amplified under extreme and complex conditions such as high temperature and high altitude, seriously reducing the accuracy of aero-engine power output calculation. For test pilots and test engineers, inaccurate power decay curves can directly cause deviations in the prediction of aircraft takeoff performance and distortions in the assessment of obstacle clearance performance, greatly affecting the accuracy of flight decisions and flight safety. Summary of the Invention

[0003] In view of this, the purpose of the present invention is to provide an engine temperature power attenuation calibration system and method based on differential GPS data, so as to solve the technical problems mentioned in the prior art.

[0004] An engine temperature power attenuation calibration system based on differential GPS data, the calibration system comprising: The multi-source data acquisition module is used to acquire several types of flight parameters, differential GPS data, and airport meteorological data of the flight target on the same time axis under the set load conditions to construct a multi-source parameter sequence matrix. The parameter inversion calculation module calibrates the differential GPS data based on the flight parameters and airport meteorological data in the multi-source parameter sequence matrix, and calculates the actual output power by inversely solving the actual engine output thrust of the flight target during the takeoff roll phase based on the calibrated multi-source parameter sequence matrix. The temperature and power correlation coupling module is used to match the actual output power and air pressure altitude corresponding to any temperature gradient from the parameter inversion calculation module, and construct a temperature and power mapping relationship dataset. The attenuation characteristic curve calibration engine has a built-in multilayer perceptron deep neural network. The multilayer perceptron deep neural network performs nonlinear calibration on the theoretical power attenuation curve based on the temperature-power mapping relationship dataset, generating a cluster of actual power attenuation characteristic curves that can be adaptively adjusted with temperature.

[0005] Optionally, the multi-source data acquisition module includes a data quality assessment subunit and a time synchronization subunit. The data quality assessment subunit is used to preprocess the acquired flight parameters, differential GPS data, and airport meteorological data. The time synchronization subunit is used to align the preprocessed flight parameters, differential GPS data, and airport meteorological data to the same discrete time node through linear interpolation, constructing a multi-source parameter sequence matrix. The execution logic of this preprocessing includes at least: ① Differential GPS data signal continuity verification: Remove data segments with GPS signal loss, positioning jumps and / or elevation changes; ② Flight parameter steady-state verification: Eliminate abnormal outliers caused by sudden changes in engine speed and / or torque; ③ Airport meteorological data stability judgment verification: Remove sudden interference values ​​of wind speed and / or wind direction during the set monitoring period.

[0006] Optionally, the parameter inversion calculation module includes: The three-dimensional motion vector calculation subunit calculates the ground velocity vector and vector acceleration of the flight target in real time based on differential GPS data at continuous time intervals, and generates differential GPS measured acceleration curves. The wind speed and aerodynamic correction subunit receives wind speed and direction data from flight parameters and / or airport meteorological data, calculates the wind speed vector of the flight target along the flight direction in real time, retrieves the ground speed vector and the wind speed vector to perform a difference operation to obtain the true velocity vector, calculates the instantaneous acceleration of the flight target after aerodynamic correction based on the true velocity vector, and constructs the acceleration change curve. The takeoff and takeoff dynamics modeling subunit establishes the dynamic equations of the flight target during the takeoff and takeoff phase based on the real-time total mass, instantaneous acceleration, dynamic rolling friction coefficient and runway slope of the flight target using Newton's second law, and calculates the instantaneous thrust of the propeller of the flight target in reverse. The shaft output power calculation subunit uses a propeller efficiency map interpolation algorithm to obtain the instantaneous efficiency corresponding to the instantaneous thrust of the propeller of the flight target, and calculates the propeller shaft output power of the flight target based on the instantaneous thrust, true velocity vector and instantaneous efficiency of the propeller of the flight target, as the actual output power.

[0007] Optionally, the parameter inversion calculation module further includes a thrust inversion self-consistency verification subunit, which is configured to: acquire the residual value between the acceleration change curve and the differential GPS measured acceleration curve in real time; and dynamically adjust the dynamic rolling friction coefficient according to the set adjustment unit if the residual value exceeds a preset threshold.

[0008] Optionally, the temperature and power correlation coupling module includes: The taxiing section filtering subunit continuously retrieves the actual output power corresponding to any temperature value from the parameter inversion calculation module within the preset flight taxiing section, and generates a power output data sequence. The power mean extraction subunit performs mean calculation on the power output data sequence to obtain the actual output power corresponding to the current temperature value; The multi-condition sample coupling subunit continuously acquires multiple temperature values ​​and their corresponding actual output power and air pressure altitude under arbitrary temperature gradients, forming one-to-one data pairs to construct a temperature-power mapping relationship dataset.

[0009] Optionally, the temperature and power correlation coupling module further includes a multi-engine aircraft data filtering extension subunit, which is configured to: retrieve real-time flight parameter heading angle and yaw rate data, determine whether the flight target stably maintains the runway centerline without yaw correction during the takeoff roll phase; and select only the smooth roll data segment with yaw rate and lateral control surface correction amount below a preset threshold as the preset flight roll interval.

[0010] Optionally, the multilayer perceptron deep neural network includes: The input layer is provided with a plurality of first neurons, which are used to input sample data. The sample data is at least one of ambient temperature, air pressure altitude, and power value; the power value is either theoretical power or actual output power. The first hidden layer includes multiple second neurons, an activation function U set at the output of each second neuron, and Dropout regularization; The second hidden layer includes multiple third neurons and activation functions set at the output of each third neuron; The output layer has a fourth neuron, which is used to output the actual engine power after calibration by the second hidden layer.

[0011] Optionally, the attenuation characteristic curve calibration engine is provided with a transfer learning fine-tuning subunit, which is configured to: train the multilayer perceptron deep neural network based on the temperature and power mapping relationship dataset corresponding to any one of the multiple sets of flight targets under the same load conditions, and freeze the weights of the input layer and the first hidden layer; for the remaining flight targets, only collect 1 to 2 sets of sample data with different ambient temperatures from the temperature and power mapping relationship dataset to train the multilayer perceptron deep neural network.

[0012] A calibration method for engine temperature and power attenuation based on differential GPS data, applied to the calibration system described above, includes the following steps: S1. Under the set load conditions, acquire several types of flight parameters, differential GPS data and airport meteorological data of the flight target at each temperature point in any temperature gradient on the same time axis to construct a multi-source parameter sequence matrix. S2. The differential GPS data is calibrated based on the flight parameters and airport meteorological data in the multi-source parameter sequence matrix, and the actual output power is calculated by back-solving the actual engine output thrust of the flight target during the takeoff roll phase based on the calibrated multi-source parameter sequence matrix. S3. Construct a dataset of temperature-power mapping relationship based on the actual output power and air pressure altitude corresponding to each temperature point in the temperature gradient, and use it as a training sample set; S4. Input the training sample set into the multilayer perceptron deep neural network, and use the Adam optimizer and MSE loss function to iteratively train until the loss converges, and generate a decay characteristic curve calibration engine. S5. The theoretical power attenuation curve is nonlinearly calibrated using the attenuation characteristic curve calibration engine to generate a cluster of actual power attenuation characteristic curves that can be adaptively adjusted with temperature.

[0013] Optionally, in step S4, the method of training the multilayer perceptron deep neural network using the training sample set specifically includes: S401. Input the ambient temperature, air pressure, altitude, and theoretical power value into the input layer of the multilayer perceptron deep neural network; S402. The input ambient temperature, air pressure, altitude and power values ​​are linearly transformed and calculated by multiple second neurons in the first hidden layer. Negative values ​​are filtered by activation function and nonlinear mapping relationship is introduced to fit the complex nonlinear decay relationship between temperature and power and output to the second hidden layer. At the same time, Dropout regularization is used to randomly block some of the output of the second neurons to suppress model overfitting. S403: The complex nonlinear attenuation relationship between temperature and power output from the first hidden layer is linearly transformed and calculated by multiple third neurons in the second hidden layer. The deep nonlinear mapping features of ambient temperature, air pressure, altitude, and theoretical power value to actual power value are extracted by the activation function and output to the output layer. S404. The output layer performs linear regression calculations on the deep nonlinear mapping characteristics from the ambient temperature, air pressure, altitude, and theoretical power value to the actual power value output by the second hidden layer to obtain the output results.

[0014] The beneficial effects that this invention can produce include: 1. The aero-engine temperature and power attenuation calibration system and method based on differential GPS data provided by this invention filters out interference-free, time-consistent, and steady-state measured flight, meteorological, and positioning data through a multi-source data acquisition module, avoiding invalid data and noise interference. Then, combined with the real flight dynamics modeling logic of the parameter inversion calculation module, it abandons idealized parameters such as fixed friction coefficient, fixed propeller effect, and ideal acceleration, and dynamically adapts to real operating conditions such as runway slope, wind field disturbance, and real-time aircraft mass. At the same time, it converges calculation errors through self-consistent verification closed loop. Finally, through the correlation coupling module, it extracts steady-state and accurate temperature-power mapping sample data, realizing the technical link from theoretical power estimation to measured power inversion. It completely eliminates the systematic deviation between traditional theoretical models and real flight conditions from the data source and the calculation layer, and significantly improves the measurement accuracy of the actual output power of aero-engines.

[0015] 2. This invention relies on a multi-condition sample coupling mechanism to construct a real-world operating condition dataset covering the entire gradient of temperature, pressure, and altitude under extreme operating conditions such as high temperature and high altitude. Then, through the multi-layer nonlinear activation and feature extraction capabilities of a multilayer perceptron deep neural network, it overcomes the limitations of traditional static linear models, accurately fitting the complex nonlinear decay relationship between temperature, pressure, and engine power. Simultaneously, Dropout regularization and iterative optimization training avoid overfitting and insufficient accuracy. Finally, it achieves full-domain nonlinear calibration of the traditional theoretical decay curve, generating a cluster of power decay curves that can adaptively and dynamically adjust with ambient temperature, pressure, and altitude. This effectively solves the pain point of amplified power deviation under extreme and complex operating conditions, significantly improving the matching degree and accuracy of engine power decay characteristics under all operating conditions, especially extreme conditions, fundamentally improving the problem of distorted prediction of aircraft takeoff performance and obstacle clearance performance assessment.

[0016] 3. This invention, through a multi-engine aircraft data filtering subunit, accurately eliminates invalid data caused by yaw corrections and attitude anomalies, achieving standardized filtering of single-engine and multi-engine aircraft data and expanding the range of applicable aircraft types. Simultaneously, it introduces a transfer learning mechanism, relying on full-sample training of the basic aircraft type and freezing the underlying general weights of the model. For other aircraft types, only a small number of samples under different temperature conditions are needed to complete model fine-tuning and adaptation. While ensuring the consistency of power calibration accuracy across multiple aircraft types, it significantly reduces the data acquisition workload and testing costs for multi-aircraft experiments, solving the technical problems of poor model adaptability, high deployment costs, and weak versatility in traditional models. This significantly improves the engineering practicality and scenario adaptability of the entire calibration system, facilitating large-scale deployment and application across multiple aircraft types, and effectively enhancing flight decision-making accuracy and flight safety assurance capabilities. Attached Figure Description

[0017] Figure 1 This is a schematic diagram of the architecture of the engine temperature and power attenuation calibration system based on differential GPS data of the present invention. Figure 2 In this invention Figure 1 The architecture diagram of the multi-source data acquisition module; Figure 3 In this invention Figure 1 The architecture diagram of the parameter inversion calculation module; Figure 4 In this invention Figure 1 The schematic diagram of the temperature and power correlation coupling module; Figure 5 In this invention Figure 1 The architecture diagram of the calibration engine for the attenuation characteristic curve; Figure 6 This is a schematic diagram comparing the actual power inversion mean and theoretical attenuation curve of the calibration system of the present invention under different ambient temperatures; wherein Figure 6 (a) is a schematic diagram showing the variation of power with equivalent airspeed under different ambient temperatures; Figure 6 (b) is a schematic diagram comparing the theoretical power curves and the measured average power under different ambient temperatures; Figure 7 This is a diagram illustrating the training and deployment architecture of the multilayer perceptron deep neural network in this invention. Figure 8 The calibration method of this invention finally generates a measured and calibrated engine temperature-power decay characteristic curve. In the diagram: 1. Multi-source data acquisition module; 11. Data quality assessment subunit; 12. Time synchronization subunit; 2. Parameter inversion calculation module; 21. Three-dimensional motion vector solution subunit; 22. Wind speed and aerodynamic correction subunit; 23. Taxiing dynamics modeling subunit; 24. Shaft output power calculation subunit; 25. Thrust inversion self-consistency verification subunit; 3. Temperature and power correlation coupling module; 31. Taxiing interval screening subunit; 32. Power mean extraction subunit; 33. Multi-condition sample coupling subunit; 34. Multi-engine aircraft data screening and extension subunit; 4. Attenuation characteristic curve calibration engine; 41. Multilayer perceptron deep neural network; 42. Transfer learning fine-tuning subunit. Detailed Implementation

[0018] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0019] Please see Figure 1As shown, this invention provides an engine temperature and power attenuation calibration system based on differential GPS data. The calibration system includes a multi-source data acquisition module 1, a parameter inversion calculation module 2, a temperature and power correlation coupling module 3, and an attenuation characteristic curve calibration engine 4. The multi-source data acquisition module 1 is used to acquire several types of flight parameters, differential GPS data, and airport meteorological data of the flight target on the same time axis under a set load condition to construct a multi-source parameter sequence matrix. This matrix covers core interference variables such as temperature, air pressure, motion state, and equipment conditions during actual flight, compensating for the deficiency of bench tests in not considering the coupling interference of multiple factors in actual flight, and ensuring the comprehensiveness of data for subsequent power calibration. The parameter inversion calculation module 2 is based on the multi-source parameter sequence... The flight parameters and airport meteorological data in the matrix are used to calibrate the differential GPS data, and the actual output power is calculated by back-calculating the actual engine thrust of the flight target during the takeoff roll phase based on the calibrated multi-source parameter sequence matrix. The temperature and power correlation coupling module 3 is used to match the actual output power and air pressure altitude corresponding to any temperature gradient from the parameter inversion calculation module 2 to construct a temperature and power mapping relationship dataset. The attenuation characteristic curve calibration engine 4 has a built-in multilayer perceptron deep neural network 41. The multilayer perceptron deep neural network 41 performs nonlinear calibration on the theoretical power attenuation curve based on the temperature and power mapping relationship dataset, generating a cluster of actual power attenuation characteristic curves that can adaptively adjust with temperature. This calibration system solves the core problem of the disconnect between theoretical models and actual operating conditions from the bottom up by constructing a fully closed-loop autonomous calibration system of data acquisition, parameter inversion, correlation modeling and intelligent calibration. It adapts to interference factors in the real flight scenario, eliminates the systematic deviation of the theoretical power attenuation curve, and significantly improves the overall accuracy of aero-engine power measurement.

[0020] In some embodiments, taking a typical single-engine turboprop aircraft test flight scenario as an example, the calibration system is deployed in the flight test data processing center of the test flight organization. In a typical test flight, the data sources involved include: a flight parameter recording system (FDR / DAR) that records parameters such as engine torque, propeller speed, fuel flow, indicated airspeed, barometric altitude, and total atmospheric temperature at a sampling rate of not less than 16Hz; a differential GPS system (DGPS) that records the aircraft's geographical location (longitude, latitude, geodetic altitude), eastward speed, northward speed, vertical speed, and precise GPS time at a sampling rate of 20Hz; and an automatic weather observation system (AWOS) that provides minute-level updates of data such as field pressure, dry-bulb temperature, dew point temperature, average wind speed, average wind direction, and gust wind speed.

[0021] In other embodiments, the trained attenuation characteristic curve calibration engine 4 is embedded in an embedded computer. Before takeoff, the system obtains the latest airport temperature and air pressure via data link, and, combined with the aircraft weight and takeoff configuration, directly outputs the actual available power under the current conditions, enabling the flight control system to dynamically calculate the calibrated flight speed and obstacle clearance gradient.

[0022] Furthermore, such as Figure 2 As shown, the multi-source data acquisition module 1 has a built-in data quality assessment subunit 11 and a time synchronization subunit 12. The data quality assessment subunit 11 is used to preprocess the acquired flight parameters, differential GPS data, and airport meteorological data. The time synchronization subunit 12 is used to align the preprocessed flight parameters, differential GPS data, and airport meteorological data of the data quality assessment subunit 11 to the same discrete time node through linear interpolation, construct a multi-source parameter sequence matrix, realize the standardization and time-series unified processing of multi-dimensional measured data, and avoid the interference of invalid data on the calibration results. The execution logic of this preprocessing includes at least the following: ① Differential GPS data signal continuity verification: Remove data segments with GPS signal loss, positioning jumps and / or elevation changes; avoid deviations in the calculation of flight parameters such as acceleration or velocity caused by positioning failure data, and ensure the authenticity and stability of flight motion status data; ② Flight parameter steady-state verification: Eliminate abnormal outliers in engine speed and / or torque; exclude non-steady-state interference factors of the engine to ensure that the subsequent power inversion data are all from effective operating conditions, and improve the accuracy of actual power solution; ③ Airport meteorological data stability judgment verification: Remove sudden interference values ​​of wind speed and / or wind direction during the set monitoring period; ensure the matching of meteorological parameters with flight conditions, avoid distortion of temperature-power correlation model caused by instantaneous meteorological anomalies, and ensure the universality of calibration curves.

[0023] Furthermore, such as Figure 3As shown, the parameter inversion calculation module 2 includes a three-dimensional motion vector calculation subunit 21, a wind speed and aerodynamic correction subunit 22, a taxiing dynamics modeling subunit 23, a shaft output power calculation subunit 24, and a thrust inversion self-consistency verification subunit 25. The three-dimensional motion vector calculation subunit 21 calculates the ground speed vector and vector acceleration of the flight target in real time based on differential GPS data at continuous time intervals, and generates a differential GPS measured acceleration curve. It relies on high-precision differential GPS measured data to restore the true motion state of the aircraft, replacing the traditional theoretical motion parameter estimation method and realizing the true acquisition of flight dynamic parameters. The wind speed and aerodynamic correction subunit 22 receives wind speed and direction data from flight parameters and / or airport meteorological data, calculates the wind speed vector of the flight target along the flight direction in real time, and retrieves the ground speed vector and wind speed vector to perform difference calculation to obtain the true velocity vector. Based on the true velocity vector, it calculates the instantaneous acceleration of the flight target after aerodynamic correction, constructs the acceleration change curve, and makes up for the deficiency of traditional models that ignore the influence of actual aerodynamic wind disturbance. The system recreates the aircraft's true airspeed and force state, matching real flight aerodynamic conditions. The taxiing dynamics modeling subunit 23, based on the flight target's real-time total mass, instantaneous acceleration, dynamic rolling friction coefficient, and runway slope, establishes the dynamic equations of the flight target during the takeoff taxiing phase using Newton's second law. It then reverse-calculates the instantaneous propeller thrust of the flight target, accurately recreating the aircraft's true ground taxiing force scenario. This effectively corrects the thrust calculation deviation caused by the traditional theoretical model's neglect of runway, friction, and real-time mass, achieving high-precision inversion of the engine's true thrust. The shaft output power calculation subunit 24 uses a propeller efficiency graph interpolation algorithm to obtain the instantaneous efficiency corresponding to the flight target's instantaneous propeller thrust. Based on the flight target's instantaneous propeller thrust, true velocity vector, and instantaneous efficiency, it calculates the flight target's propeller shaft output power as the actual output power. This abandons fixed theoretical propeller efficiency parameters and uses graph interpolation to adapt to the changes in propeller efficiency under different operating conditions, accurately matching the true power output conversion relationship. The thrust inversion self-consistency verification subunit 25 is configured to: acquire the residual value between the acceleration change curve and the differential GPS measured acceleration curve in real time; if the residual value exceeds the preset threshold, dynamically adjust the dynamic rolling friction coefficient according to the set adjustment unit, so as to continuously converge the thrust and power inversion calculation error, ensure the stability and accuracy of power solution under full taxiing conditions, and avoid systematic deviation caused by fixed values ​​of a single parameter.

[0024] In the above, the propeller shaft output power P of the flying target actual for: ; Among them, F T V is the instantaneous thrust of the propeller of the flying target. true The true velocity vector of the flying target. The instantaneous efficiency of the flight target is given; this instantaneous efficiency is specifically obtained through two-dimensional interpolation, indexed by the advance ratio J and the power coefficient Cp.

[0025] In the above, the dynamic equations of the flight target during the takeoff roll phase are: ; Among them, F f For the dynamic rolling friction of the runway: F f =mgf, where f is the dynamic rolling friction coefficient (the system can adaptively correct according to the acceleration residual, unlike the traditional fixed friction coefficient); g is the gravitational acceleration; F g Runway slope resistance: F g =mgsinθ, where θ is the actual slope of the runway; F wind Let m be the aerodynamic drag, m be the real-time total mass of the flying target, and a be the instantaneous acceleration.

[0026] Furthermore, such as Figure 4 As shown, the temperature and power correlation coupling module 3 includes a taxiing interval screening subunit 31, a power mean extraction subunit 32, a multi-condition sample coupling subunit 33, and a multi-engine aircraft data screening extension subunit 34. The taxiing interval screening subunit 31 continuously retrieves the actual output power corresponding to any temperature value from the parameter inversion calculation module 2 within a preset flight taxiing interval, generating a power output data sequence. This ensures that the data samples accurately match the core application scenarios of aircraft takeoff performance verification and obstacle clearance capability assessment, improving the adaptability of the final calibration curve to actual flight decisions. The power mean extraction subunit 34... 2. The power output data sequence is averaged to obtain the actual output power corresponding to the current temperature value, eliminating the instantaneous power fluctuation error at a single point in time and refining the engine steady-state power output characteristics under various temperature conditions; the multi-condition sample coupling subunit 33 continuously acquires multiple temperature values ​​and their corresponding actual output power and air pressure altitude under any temperature gradient, forming a one-to-one data pair to construct a temperature and power mapping relationship dataset. By covering the full-condition scenario with multiple temperature and altitude gradients, it focuses on supplementing the high-temperature and high-altitude extreme condition samples to make up for the defects of distortion of traditional theoretical curves under extreme conditions. The multi-engine aircraft data filtering extension subunit 34 is configured to: retrieve real-time data of flight parameter heading angle and yaw rate to determine whether the flight target is stably maintaining the runway centerline without yaw correction during the takeoff roll phase; select only the stable roll data segment with yaw rate and lateral control surface correction amount below the preset threshold as the preset flight roll interval, thereby eliminating the power calculation interference caused by abnormal roll attitude of multi-engine aircraft, ensuring the validity of multi-engine aircraft sample data, achieving full coverage and accurate calibration of single-engine and multi-engine aircraft, and improving the system's versatility.

[0027] In this embodiment, as Figure 6 As shown in (a), the shaft power data sequence before the set taxiing speed V1 is extracted from multiple full-power static takeoff test flight data performed under different ambient temperatures. Figure 6 As shown in (b), the mean power is extracted from the extracted shaft power data sequence in the interval from 0.7V1 to V1, forming power-temperature data pairs. A high-confidence mapping dataset is constructed by accumulating data from different temperature days. Figure 6 (a) It can be seen that all three curves exhibit sawtooth oscillations, indicating that at the same temperature, power fluctuates slightly instantaneously with increasing equivalent airspeed, but the average power within the range does not show a clear monotonic trend. For example, across the entire speed range from 40kt to 140kt, the average power of each temperature curve remains relatively stable, indicating that airspeed is not the dominant factor in power attenuation under this condition; temperature is the decisive variable. Meanwhile, the light blue shaded area (approximately 95~120kt) represents the commonly used airspeed range for cruise. The fluctuation amplitude of the three curves in the cruise range is consistent with that in the low-speed range, indicating a stable power difference caused by temperature throughout the cruise flight. At the high-speed end of 140kt, the high-temperature red curve shows a significant drop, proving that power attenuation intensifies under the coupling of high temperature and high speed. Figure 6 (b) It can be seen that under the low temperature condition of 15℃, the theoretical value is about 995 horsepower, the measured average is about 980 horsepower, and the measurement deviation is 15 horsepower. The error range is narrow, the experimental data dispersion is small, and the theoretical and measured values ​​are in good agreement, indicating that the model prediction accuracy is excellent at low temperatures. Under the medium temperature condition of 28℃, the theoretical value is about 950 horsepower, the measured average is about 900 horsepower, and the deviation increases to 50 horsepower. The error length increases significantly, and the power fluctuations in multiple tests at the same temperature become larger, indicating that the model underestimates the power loss caused by high temperature. Under the high temperature condition of 42℃, the theoretical value is about 905 horsepower, the measured average is only 815 horsepower, and the deviation reaches 95 horsepower, which is more than 6 times the deviation at low temperature. The error span is the largest, indicating that the dispersion of experimental data is significantly improved under high temperature conditions, and the theoretical model error deteriorates significantly. It is evident that existing theoretical models only consider the assumptions of an ideal atmosphere and no additional losses, neglecting the additional losses under high-temperature conditions, such as nonlinear losses caused by high temperatures, such as increased lubrication resistance of lubricating oil, increased turbine cooling bleed air loss, amplified total pressure loss in the intake manifold, and decreased combustion efficiency in the combustion chamber. These losses increase nonlinearly with increasing temperature, so the higher the temperature, the greater the difference between theoretical calculations and actual experiments.

[0028] In this embodiment, as Figure 7As shown, the multilayer perceptron deep neural network 41 includes an input layer, a first hidden layer, a second hidden layer, and an output layer. The input layer has three first neurons, which correspond one-to-one with the input ambient temperature, air pressure altitude, and power value, respectively; the power value is the theoretical power or the actual output power. The first hidden layer includes 64 second neurons, activation functions at the output of each second neuron, and Dropout regularization. The second hidden layer includes 32 third neurons and activation functions at the output of each third neuron. The output layer has one fourth neuron, which outputs the actual engine power calibrated by the second hidden layer. The activation function can be a ReLU function, a Sigmoid function, or a Tanh function, but to avoid the gradient vanishing problem during gradient backpropagation in the deep neural network, the ReLU function is preferred.

[0029] Furthermore, such as Figure 5 As shown, the attenuation characteristic curve calibration engine 4 is equipped with a transfer learning fine-tuning subunit 42. The transfer learning fine-tuning subunit 42 is configured to: train a multilayer perceptron deep neural network 41 based on the temperature and power mapping relationship dataset corresponding to any one of the multiple sets of flight targets under the same load conditions, and freeze the weights of the input layer and the first hidden layer; for the remaining flight targets, only 1 to 2 sets of sample data with different ambient temperatures from the temperature and power mapping relationship dataset are collected to train the multilayer perceptron deep neural network 41, which is used to significantly reduce the data collection workload of multi-aircraft calibration, improve the model iteration and adaptation efficiency, and at the same time ensure the consistency and accuracy of power attenuation calibration of multiple aircraft.

[0030] This invention also provides a method for calibrating engine temperature and power attenuation based on differential GPS data, applied to the aforementioned calibration system. The calibration method includes the following steps: Step S1: Under the set load conditions, acquire several types of flight parameters, differential GPS data, and airport meteorological data of the flight target at each temperature point on the same time axis in any temperature gradient to construct a multi-source parameter sequence matrix. Step S2: Based on the flight parameters and airport meteorological data in the multi-source parameter sequence matrix, the differential GPS data is calibrated, and the actual output power is calculated by inversely solving the actual engine output thrust of the flight target during the takeoff roll phase based on the calibrated multi-source parameter sequence matrix. Step S3: Construct a dataset of temperature-power mapping relationship based on the actual output power and air pressure altitude corresponding to each temperature point in the temperature gradient, and use it as a training sample set; Step S4: Input the training sample set into the multilayer perceptron deep neural network 41, and use the Adam optimizer and MSE loss function to iteratively train until the loss converges, and generate the decay characteristic curve calibration engine 4. Step S5: Perform nonlinear calibration on the theoretical power attenuation curve using the attenuation characteristic curve calibration engine 4 to generate a cluster of actual power attenuation characteristic curves that can be adaptively adjusted with temperature.

[0031] In this embodiment, as Figure 8 As shown, the confidence interval is narrow in the low-temperature range, indicating low statistical uncertainty. The confidence interval is significantly wider in the high-temperature range, demonstrating that random disturbances and operational fluctuations in engine power output are significantly enhanced under high-temperature conditions, expanding the range of statistical prediction confidence and requiring higher fault tolerance for high-temperature power prediction. The power curve calibrated by the aforementioned calibration system remains entirely within the high confidence interval, indicating high reliability and applicability to complex high-temperature conditions. Furthermore, the manufacturer's theoretical curve, set at the factory, gradually deviates from the confidence interval as ambient temperature increases. Therefore, the calibration system and method of this invention effectively solve the specific technical problem of nonlinear amplification of the theoretical-actual power deviation due to temperature increases in existing technologies by constructing a complete closed-loop system from GPS measured data inversion to correlation coupling to MPL nonlinear calibration. Both the theoretical and actual power curves show a continuous downward trend with increasing temperature; the higher the ambient temperature, the more linearly the maximum usable power of the aero-engine decreases. Low temperatures result in denser, colder air and higher intake mass flow rates, leading to stronger combustion and work efficiency. In high-temperature environments, the thinner air reduces intake flow, while increased losses in turbine cooling and lubrication systems cause a continuous decrease in output power. Furthermore, across the entire temperature range, the actual calibration curve consistently lies below the manufacturer's theoretical curve, indicating that the original theoretical model overestimates the engine's actual output power throughout the entire process, and the actual power performance is lower than the manufacturer's stated theoretical value.

[0032] Furthermore, in step S4, the method for training the multilayer perceptron deep neural network 41 using the training sample set specifically includes: Step S401: Input the ambient temperature, air pressure, altitude, and theoretical power value into the input layer of the multilayer perceptron deep neural network 41; Step S402: The 64 second neurons of the first hidden layer perform linear transformation calculations on the input ambient temperature, air pressure, altitude, and power values; and the ReLU activation function is used to filter negative values, and a nonlinear mapping relationship is introduced to fit the complex nonlinear decay relationship between temperature and power and output to the second hidden layer. At the same time, Dropout regularization is used to randomly block the output of some neurons to suppress model overfitting. Step S403: The complex nonlinear attenuation relationship between temperature and power output from the first hidden layer is calculated by linear transformation through the 32 third neurons of the second hidden layer, and the deep nonlinear mapping features of ambient temperature, air pressure altitude, and theoretical power value to actual power value are extracted by the activation function ReLU and output to the output layer. Step S404: The output layer performs linear regression calculations on the ambient temperature, air pressure, altitude, and deep nonlinear mapping characteristics from theoretical power value to actual power value output by the second hidden layer to obtain the output results.

Claims

1. An engine temperature and power attenuation calibration system based on differential GPS data, characterized in that, The calibration system includes: The multi-source data acquisition module (1) is used to acquire several types of flight parameters, differential GPS data and airport meteorological data of the flight target on the same time axis under the set load conditions to construct a multi-source parameter sequence matrix; The parameter inversion calculation module (2) calibrates the differential GPS data based on the flight parameters and airport meteorological data in the multi-source parameter sequence matrix, and calculates the actual output power by inversely solving the actual engine output thrust of the flight target during the takeoff roll phase based on the calibrated multi-source parameter sequence matrix. The temperature and power correlation coupling module (3) is used to match the actual output power and air pressure altitude corresponding to any temperature gradient from the parameter inversion calculation module (2) and construct a temperature and power mapping relationship dataset; The attenuation characteristic curve calibration engine (4) has a built-in multilayer perceptron deep neural network (41). The multilayer perceptron deep neural network (41) performs nonlinear calibration on the theoretical power attenuation curve based on the temperature and power mapping relationship dataset, and generates a cluster of actual power attenuation characteristic curves that can be adaptively adjusted with temperature.

2. The engine temperature and power attenuation calibration system based on differential GPS data according to claim 1, characterized in that, The multi-source data acquisition module (1) has a built-in data quality assessment subunit (11) and a time synchronization subunit (12). The data quality assessment subunit (11) is used to preprocess the acquired flight parameters, differential GPS data, and airport meteorological data. The time synchronization subunit (12) is used to align the preprocessed flight parameters, differential GPS data, and airport meteorological data of the data quality assessment subunit (11) to the same discrete time node through linear interpolation to construct a multi-source parameter sequence matrix. The execution logic of this preprocessing includes at least the following: ① Differential GPS data signal continuity verification: Remove data segments with GPS signal loss, positioning jumps and / or elevation changes; ② Flight parameter steady-state verification: Eliminate abnormal outliers caused by sudden changes in engine speed and / or torque; ③ Airport meteorological data stability judgment verification: Remove sudden interference values ​​of wind speed and / or wind direction during the set monitoring period.

3. The engine temperature and power attenuation calibration system based on differential GPS data according to claim 1, characterized in that, The parameter inversion calculation module (2) includes: The three-dimensional motion vector calculation subunit (21) calculates the ground velocity vector and vector acceleration of the flight target in real time based on differential GPS data at continuous time intervals, and generates differential GPS measured acceleration curves; The wind speed and aerodynamic correction subunit (22) receives wind speed and wind direction data from flight parameters and / or airport meteorological data, calculates the wind speed vector of the flight target along the flight direction in real time, retrieves the ground speed vector and the wind speed vector to perform a difference operation to obtain the true speed vector, calculates the instantaneous acceleration of the flight target after aerodynamic correction based on the true speed vector, and constructs the acceleration change curve. The takeoff and takeoff dynamics modeling subunit (23) establishes the dynamic equation of the takeoff and takeoff target based on the real-time total mass, instantaneous acceleration, dynamic rolling friction coefficient and runway slope of the flight target through Newton's second law, and calculates the instantaneous thrust of the propeller of the flight target in reverse. The shaft output power calculation subunit (24) uses the propeller efficiency map interpolation algorithm to obtain the instantaneous efficiency corresponding to the instantaneous thrust of the propeller of the flight target, and calculates the propeller shaft output power of the flight target based on the instantaneous thrust, true velocity vector and instantaneous efficiency of the propeller of the flight target, as the actual output power.

4. The engine temperature and power attenuation calibration system based on differential GPS data according to claim 3, characterized in that, The parameter inversion calculation module (2) further includes a thrust inversion self-consistency verification subunit (25), which is configured to: acquire the residual value between the acceleration change curve and the differential GPS measured acceleration curve in real time, and dynamically adjust the dynamic rolling friction coefficient according to the set adjustment unit if the residual value exceeds the preset threshold.

5. The engine temperature and power attenuation calibration system based on differential GPS data according to claim 1, characterized in that, The temperature and power correlation coupling module (3) includes: The taxiing section screening subunit (31) continuously retrieves the actual output power corresponding to any temperature value from the parameter inversion calculation module (2) within the preset flight taxiing section, and generates a power output data sequence. The power mean extraction subunit (32) performs mean calculation on the power output data sequence to obtain the actual output power corresponding to the current temperature value; The multi-condition sample coupling subunit (33) continuously acquires multiple temperature values ​​and their corresponding actual output power and air pressure altitude under any temperature gradient, forming a one-to-one data pair to construct a temperature and power mapping relationship dataset.

6. The engine temperature and power attenuation calibration system based on differential GPS data according to claim 5, characterized in that, The temperature and power correlation coupling module (3) further includes a multi-engine aircraft data filtering extension subunit (34), which is configured to: retrieve real-time data of flight parameter heading angle and yaw rate, determine whether the flight target is stably maintaining the runway centerline without yaw correction during the takeoff roll phase; and select only the smooth roll data segment with yaw rate and lateral control surface correction amount below the preset threshold as the preset flight roll interval.

7. The engine temperature and power attenuation calibration system based on differential GPS data according to claim 1, characterized in that, The multilayer perceptron deep neural network (41) includes: The input layer is provided with a plurality of first neurons, which are used to input sample data. The sample data is at least one of ambient temperature, air pressure altitude, and power value; the power value is either theoretical power or actual output power. The first hidden layer includes multiple second neurons, activation functions set at the output of each second neuron, and Dropout regularization; The second hidden layer includes multiple third neurons and activation functions set at the output of each third neuron; The output layer has a fourth neuron, which is used to output the actual engine power after calibration by the second hidden layer.

8. The engine temperature and power attenuation calibration system based on differential GPS data according to claim 7, characterized in that, The attenuation characteristic curve calibration engine (4) is equipped with a transfer learning fine-tuning subunit (42). The transfer learning fine-tuning subunit (42) is configured to: train the multilayer perceptron deep neural network (41) based on the temperature and power mapping relationship dataset corresponding to any one of the multiple sets of flight targets under the same load conditions, and freeze the weights of the input layer and the first hidden layer; for the remaining flight targets, only 1 to 2 sets of sample data with different ambient temperatures in the temperature and power mapping relationship dataset are collected to train the multilayer perceptron deep neural network (41).

9. A method for calibrating engine temperature and power attenuation based on differential GPS data, characterized in that, Applied to the calibration system according to any one of claims 1-8, the calibration method comprises the following steps: S1. Under the set load conditions, acquire several types of flight parameters, differential GPS data and airport meteorological data of the flight target at each temperature point in any temperature gradient on the same time axis to construct a multi-source parameter sequence matrix. S2. The differential GPS data is calibrated based on the flight parameters and airport meteorological data in the multi-source parameter sequence matrix, and the actual output power is calculated by back-solving the actual engine output thrust of the flight target during the takeoff roll phase based on the calibrated multi-source parameter sequence matrix. S3. Construct a dataset of temperature-power mapping relationship based on the actual output power and air pressure altitude corresponding to each temperature point in the temperature gradient, and use it as a training sample set; S4. Input the training sample set into the multilayer perceptron deep neural network (41), and use the Adam optimizer and MSE loss function to iteratively train until the loss converges, and generate the decay characteristic curve calibration engine (4). S5. The theoretical power attenuation curve is nonlinearly calibrated by the attenuation characteristic curve calibration engine (4) to generate a cluster of actual power attenuation characteristic curves that can be adaptively adjusted with temperature.

10. The method for calibrating engine temperature and power attenuation based on differential GPS data according to claim 9, characterized in that, In step S4, the method for training the multilayer perceptron deep neural network (41) using the training sample set specifically includes: S401, Input the ambient temperature, air pressure altitude and theoretical power value into the input layer of the multilayer perceptron deep neural network (41); S402. The input ambient temperature, air pressure, altitude and power values ​​are linearly transformed and calculated by multiple second neurons in the first hidden layer. Negative values ​​are filtered by activation function and nonlinear mapping relationship is introduced to fit the complex nonlinear decay relationship between temperature and power and output to the second hidden layer. At the same time, Dropout regularization is used to randomly block some of the output of the second neurons to suppress model overfitting. S403. The complex nonlinear attenuation relationship between temperature and power output from the first hidden layer is linearly transformed and calculated by multiple third neurons in the second hidden layer. The deep nonlinear mapping features of ambient temperature, air pressure, altitude, and theoretical power value to actual power value are extracted by the activation function and output to the output layer. S404. The output layer performs linear regression calculations on the deep nonlinear mapping characteristics from the ambient temperature, air pressure, altitude, and theoretical power value to the actual power value output by the second hidden layer to obtain the output results.

Images