An engine performance prediction method and system based on flight data

Abstract

The application discloses an engine performance prediction method and system based on flight data, solves the problem that the main contradiction of the traditional engine performance monitoring method is highly dependent on real-time sensor data, can only reflect the current state, and is difficult to comprehensively understand the performance change trend, sets a parameter gradient, comprehensively covers the parameter change range of different flight stages, can better simulate and analyze the performance and behavior of the engine under different flight stages and environmental conditions, the parameter setting mode not only improves the accuracy and reliability of the model, but also increases the diversity and richness of the data, provides strong support for design and optimization, sets an error range, can capture more small changes of environmental parameters that may occur in actual flight, improves the coverage of the data set, and helps the model better understand and predict the engine performance under different environmental conditions, and the error range is reasonably set.

Description

Technical Field

[0001] This invention relates to the field of aero-engine technology, specifically to a method and system for predicting engine performance based on flight data. Background Technology

[0002] Aircraft engine performance prediction plays a crucial role in the modern aviation industry. With the continuous growth of global aviation demand, engine performance directly affects flight safety, operational efficiency, and economic benefits. By accurately predicting engine performance, airlines can take timely measures before problems occur, avoiding potential safety hazards and ensuring flight safety. In addition, performance prediction can help optimize flight plans, reduce unnecessary maintenance and downtime, lower maintenance costs, and improve fuel efficiency, thereby enhancing operational efficiency and economic benefits. Accurate performance prediction provides airlines with a scientific basis for decision-making, helping them make reasonable decisions in special circumstances, such as adjusting flight plans, conducting maintenance, or taking other measures. At the same time, the development and application of engine performance prediction technology promotes innovation and development in aviation technology, providing new ideas and methods for future technological advancements.

[0003] The main problem with traditional engine performance monitoring methods is that they rely heavily on real-time sensor data, which can only reflect the current state and make it difficult to fully understand the trend of performance changes. Especially in extreme weather conditions or special periods, engine performance may be severely affected, thereby threatening flight safety. Therefore, developing a system that can perform performance prediction is of great significance for improving aviation safety and operational efficiency. Summary of the Invention

[0004] To address the shortcomings of existing technologies, this invention provides a method and system for predicting engine performance based on flight data. This solves the main problem of traditional engine performance monitoring methods, which rely heavily on real-time sensor data and can only reflect the current state, making it difficult to fully understand the trend of performance changes.

[0005] To achieve the above objectives, the present invention provides the following technical solution: a method for predicting engine performance based on flight data, comprising the following steps:

[0006] S1. Collect historical normal operating status data of the engine, including engine performance and parameter data and engine operating environment parameter data; preprocess the historical normal operating status data to generate flight data;

[0007] S2. Establish a digital twin model, input flight data, execute regression analysis algorithm, calculate the turbine inlet gas temperature of the engine according to the time series; establish environmental parameter gradients, input them into the digital twin model, obtain the turbine inlet gas temperature corresponding to each set of environmental parameters, and generate a standard dataset.

[0008] S3. Analyze the relationship between the turbine inlet gas temperature and the actual value measured by the sensor in the standard dataset, calculate the residuals, set the residual standard, and analyze the performance of the digital twin model.

[0009] S4. Establish a prediction model, input a standard dataset, execute the Gaussian process regression analysis algorithm, predict the turbine inlet gas temperature for a future period and generate the standard deviation of the predicted values, generate a prediction dataset, analyze the predicted values ​​and the standard deviation of the predicted values, and optimize the prediction model.

[0010] As a further aspect of the present invention, step S1 involves preprocessing historical normal working status data to generate flight data, including the following steps:

[0011] The data is divided into six folders according to the takeoff phase, climb phase, cruise phase, descent phase, approach phase, and landing phase. Each folder is further divided into subfolders according to the flight environment. Flight environment parameters include temperature, atmospheric pressure, and humidity. Each subfolder contains a set of flight environment parameters and the corresponding engine performance parameters, sensor parameters, and material data for that flight environment.

[0012] For the data in each folder, missing value processing, outlier processing, and noise data processing are performed. Data from different sources and formats are standardized to ensure consistency. Multi-source data are then merged to generate flight data.

[0013] As a further aspect of the present invention, step S2, calculating the turbine inlet gas temperature of the engine according to the time sequence, includes the following steps:

[0014] For the six phases—takeoff, climb, cruise, descent, approach, and landing—the subfolders within their respective folders are sorted by recorded time. The data from each subfolder is then input into a digital twin model to calculate the predicted turbine inlet gas temperature. The calculation process is as follows:

[0015] ;

[0016] in, The final turbine inlet gas temperature, Where LHV is the fuel flow rate, and LHV is the lower heating value of the fuel. EF represents combustion efficiency, and EF represents the engine efficiency factor. For airflow, A is the specific heat capacity of air; B is the atmospheric pressure correction factor; C is the temperature correction factor; and D is the humidity correction factor. This refers to the compressor outlet temperature.

[0017] The formula for calculating the atmospheric pressure correction factor is:

[0018] ;

[0019] in, Given the current atmospheric pressure, Standard atmospheric pressure;

[0020] The formula for calculating the temperature correction factor is:

[0021] ;

[0022] Given the current ambient temperature, Standard temperature;

[0023] The formula for calculating the humidity correction factor is:

[0024] ;

[0025] Given the current ambient humidity, Standard humidity;

[0026] Example: if =0.1 kg / s, LHV = 43 × 10⁶ joules / kg, =0.98, EF=0.95, =10 kg / s =1005 joules / kg·Kelvin =600 Kelvin, =101325 Pascals =101325 Pascals =293 Kelvin, =288 Kelvin, =10 grams per cubic meter =12 grams per cubic meter; therefore, the turbine inlet gas temperature is:

[0027] Kelvin;

[0028] As a further aspect of the present invention, step S2, establishing the environmental parameter gradient, includes the following steps:

[0029] Set the temperature parameter gradient. In the takeoff phase folder, climb phase folder, descent phase folder, approach phase folder, and landing phase folder, the internal temperature parameter gradient is set with -40 degrees Celsius as the temperature origin. The difference between the maximum and minimum values ​​in the temperature gradient is recorded as the range, which is 90 in this case. Set the tolerance t1.

[0030] In the cruise phase folder, the temperature parameter gradient inside starts at -50 degrees Celsius, with a range of 30, and a tolerance of t2 is set.

[0031] Set atmospheric pressure parameter gradients. In the folders for takeoff, climb, descent, approach, and landing, the atmospheric pressure parameter gradients are all set with 100 Pa as the starting point and 100 as the range, with a tolerance of d1.

[0032] In the cruise phase folder, the atmospheric pressure parameter gradient inside is set with 200 Pa as the starting point, 100 as the range, and tolerance d2.

[0033] Set humidity parameter gradients. In the takeoff phase folder, climb phase folder, descent phase folder, approach phase folder, and landing phase folder, the humidity gradients within each folder start at 0% and have a range of 100%. Set the tolerance s1.

[0034] In the cruise phase folder, the humidity parameter gradient inside is set with 0% as the starting point and 10% as the range, with a tolerance of s2.

[0035] As a further aspect of the present invention, a standard dataset is generated based on the gradient of environmental parameters, including the following steps:

[0036] Based on the engine's current stage, from the three sets of parameters—temperature, atmospheric pressure, and humidity—within the environmental parameter gradient, one parameter is extracted from each set in a permutation and combination manner to form an input array. Within the flight data, the temperature, atmospheric pressure, and humidity parameters in each subfolder are defined as a parameter group. In all subfolders, each input array is traversed and searched to find the same parameter group. The engine performance and parameter data of the parameter group are then overwritten with the corresponding engine performance and parameter data of the input array. The updated input array is then input into the digital twin model to obtain the turbine inlet gas temperature, generating a dataset.

[0037] Depending on the engine's current stage, allowable error ranges are set for temperature, atmospheric pressure, and humidity parameters. For example, the allowable error range for temperature parameters is... The allowable error range for atmospheric pressure parameters is 0.5 degrees. 1 Pa, the allowable error range for humidity parameters is 2g / m³; and update each set of input arrays within the environmental parameter gradient, mask the parameter sets contained in a class of datasets, and perform data matching on the updated standard data; in all subfolders, traverse and search each set of updated input arrays to find the parameter sets within the error range of the input arrays, and overwrite the engine performance and parameter data of the corresponding input arrays with the engine performance and parameter data of the parameter sets, input the updated input arrays into the digital twin model to obtain the turbine inlet gas temperature, and generate a second-class dataset;

[0038] Generate standard datasets, including Class I and Class II datasets.

[0039] As a further aspect of the present invention, step S3, which involves calculating the residuals, setting residual standards, and analyzing the performance of the digital twin model, includes the following steps:

[0040] The difference between a dataset in the standard data and the turbine inlet gas temperature corresponding to the same parameter group in the sensor parameters is calculated, and the absolute value of the calculation result is taken as the residual to generate a residual set.

[0041] For a set of residuals, a residual limit c1 is defined. If, within a set of residuals, a certain residual value in the flight data, sorted by engine usage time, has more than 1% of the residuals greater than or equal to the residual limit c1 in the 1000 nearest flight records, then the flight data is optimized, i.e., data augmentation is performed. This includes data enhancement, data expansion, and data synthesis. The data in each stage folder and its subfolders in the flight data are processed, standardized, fused, and enhanced to finally generate processed flight data. The digital twin model building operation is repeated until the turbine inlet gas temperature is less than the residual limit c1.

[0042] The difference between the two types of datasets in the standard data and the turbine inlet gas temperature corresponding to the same parameter group in the sensor parameters is calculated, and the absolute value of the calculation result is taken as the residual to generate a two types of residual set.

[0043] For the type II residual set, a residual limit c2 is defined. If, within the type II residual set, a certain residual value in the flight data, sorted by engine usage time, has more than 1% of the residuals within the nearest 1000 flight records that are greater than or equal to the residual limit c2, then the flight data is optimized, i.e., data augmentation is performed. This includes data enhancement, data expansion, and data synthesis. The data in each stage folder and its subfolders in the flight data are processed, standardized, fused, and enhanced to finally generate processed flight data. The digital twin model building operation is repeated until the turbine inlet gas temperature is less than the residual limit c2.

[0044] As a further aspect of the present invention, step S4, generating the prediction dataset, includes the following steps:

[0045] To simulate the environmental conditions during aircraft operation, several sets of environmental parameters are set. Under each set of environmental parameters, the midpoints of the takeoff, climb, cruise, descent, approach, and landing phases are defined as fixed nodes. Using a digital twin model, the aircraft is simulated to operate at a fixed frequency, generating a series of turbine inlet gas temperature values. The standard for the fixed frequency is: a total prediction duration of ten days, five flights per day, and six fixed nodes recorded each time according to the flight phase. Based on the flight phase, a series of turbine inlet gas temperature values ​​are input into the prediction model to generate predicted turbine inlet gas temperatures for several engines and the corresponding standard deviations of the predicted values, forming a prediction dataset.

[0046] As a further aspect of the present invention, analyzing the prediction dataset and optimizing the prediction model includes the following steps:

[0047] Define a limit z for the standard deviation of the predicted value. If the standard deviation of the predicted value is greater than z, it is considered that the corresponding fixed node of the input has fewer than the normal value in the training of the prediction model. Modify the number of daily attendances in the fixed frequency until the standard deviation of the predicted value is less than or equal to z.

[0048] Calculate the mean of the predicted values, and then calculate the confidence interval of the predicted values ​​using the mean and the standard deviation of the predicted values; calculate the range length of the confidence interval, &; define the lower limit of the standard range length as &1 and the upper limit of the standard range length as &2;

[0049] If the range length & is less than the lower limit of the standard range length &1, the allowable error range in the binary dataset is increased by increasing the number of input array groups until the range length & is greater than or equal to the lower limit of the range length &1.

[0050] If the range length & is greater than the lower limit of the standard range length of &2, the allowable error range in the binary dataset is increased by increasing the number of input array groups until the range length & is less than or equal to the lower limit of the range length of &2.

[0051] An engine performance prediction system based on flight data is disclosed. Specifically, the system employs a flight data-based engine performance prediction method to predict engine performance.

[0052] Beneficial effects

[0053] This invention, by setting parameter gradients, comprehensively covers the parameter variation range of different flight stages, enabling better simulation and analysis of engine performance and behavior under different flight stages and environmental conditions. This parameter setting method not only improves the accuracy and reliability of the model but also increases the diversity and richness of the data, thus providing strong support for design and optimization.

[0054] By setting an error range, this invention can capture more subtle changes in environmental parameters that may occur during actual flight, thereby improving the coverage of the dataset. This helps the model better understand and predict engine performance under different environmental conditions. At the same time, setting an error range reasonably can control the number of samples while ensuring data quality, making model training more efficient and effective.

[0055] This invention significantly improves the accuracy and reliability of the model by calculating and analyzing residuals and performing feedback optimization. It defines residual limits c1 and c2 to quantify model performance standards, ensuring that the model's prediction error on different datasets remains within acceptable limits. By analyzing the residual set and identifying the proportion of residuals greater than or equal to the residual limits in the most recent 1000 flight records, it promptly identifies performance deficiencies under specific conditions. To address these deficiencies, it performs data augmentation, including data enhancement, data expansion, and data synthesis, enriching the dataset, improving data diversity and representativeness, and enhancing the model's generalization ability. It repeats the digital twin model building process to ensure that the model's prediction accuracy meets expected standards under various flight conditions. Continuous optimization of model performance improves its robustness and adaptability, ensuring accurate prediction of engine turbine inlet gas temperature in practical applications, providing reliable data support for engine performance evaluation and optimization.

[0056] This invention improves prediction accuracy by defining a standard deviation limit z for predicted values, adjusting the number of daily attendances at a fixed frequency, ensuring sufficient training data for the prediction model, and calculating the mean and confidence interval of the predicted values. It also analyzes the range length & of the confidence interval, defining a lower limit of &1 and an upper limit of &2 for the standard range length, adjusting the allowable error range in the binary dataset, and increasing the number of input arrays to ensure the confidence interval of the predicted values ​​is within a reasonable range. Finally, it analyzes the prediction dataset, optimizes the prediction model, and improves the accuracy and reliability of predictions, providing comprehensive and reliable data support for engine performance evaluation and optimization. Attached Figure Description

[0057] Figure 1 This is a schematic diagram of the method flow of the present invention;

[0058] Figure 2 This is a system block diagram of the present invention. Detailed Implementation

[0059] 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.

[0060] Example 1

[0061] Please see Figure 1 This application provides a method for predicting engine performance based on flight data, including the following steps:

[0062] S1. Collect historical normal operating status data of the engine, including engine performance and parameter data and engine operating environment parameter data; preprocess the historical normal operating status data to generate flight data;

[0063] As a further aspect of the present invention, step S1 involves preprocessing historical normal working status data to generate flight data, including the following steps:

[0064] The data is divided into six folders according to the takeoff phase, climb phase, cruise phase, descent phase, approach phase, and landing phase. Each folder is further divided into subfolders according to the flight environment. Flight environment parameters include temperature, atmospheric pressure, and humidity. Each subfolder contains a set of flight environment parameters and the corresponding engine performance parameters, sensor parameters, and material data for that flight environment.

[0065] For the data in each folder, missing value processing, outlier processing, and noise data processing are performed. Data from different sources and formats are standardized to ensure consistency. Multi-source data are merged to generate flight data.

[0066] The engine's performance parameters include power, torque, engine efficiency factor, airflow, speed, compression ratio, displacement, thermal efficiency, emissions, combustion efficiency, turbine efficiency, cooling system efficiency, and lubrication system efficiency.

[0067] Sensor parameters, namely various parameters of the engine, are monitored and recorded in real time, including the temperature and pressure of various engine components, the heat released by fuel combustion, the compressor outlet temperature, the gas temperature before the turbine, the fuel flow rate, the speed and rotational acceleration, the torque, the humidity, etc.

[0068] Material data, including the dimensions, density, standard atmospheric pressure, standard temperature, standard humidity, specific heat capacity of air, lower heating value of fuel, elastic modulus, combustion efficiency, yield strength, lower heating value of fuel, hardness, thermal conductivity, and electrical conductivity of various parts of the engine;

[0069] Engine operating environment parameter data, that is, environmental condition parameter data when the engine is working, including parameters such as temperature, atmospheric pressure, and humidity; each set of environmental condition parameter data has a corresponding set of sensor parameters and performance parameters;

[0070] S2. Establish a digital twin model, input flight data, execute regression analysis algorithm, calculate the turbine inlet gas temperature of the engine according to the time series; establish environmental parameter gradients, input them into the digital twin model, obtain the turbine inlet gas temperature corresponding to each set of environmental parameters, and generate a standard dataset.

[0071] As a further aspect of the present invention, step S2, calculating the turbine inlet gas temperature of the engine according to the time sequence, includes the following steps:

[0072] For the six phases—takeoff, climb, cruise, descent, approach, and landing—the subfolders within their respective folders are sorted by recorded time. The data from each subfolder is then input into a digital twin model to calculate the predicted turbine inlet gas temperature. The calculation process is as follows:

[0073] ;

[0074] in, For the final turbine inlet gas temperature, Where LHV is the fuel flow rate, and LHV is the lower heating value of the fuel. EF represents combustion efficiency, and EF represents the engine efficiency factor. For airflow, A is the specific heat capacity of air; B is the atmospheric pressure correction factor; C is the temperature correction factor; and D is the humidity correction factor. This refers to the compressor outlet temperature.

[0075] The formula for calculating the atmospheric pressure correction factor is:

[0076] ;

[0077] in, Given the current atmospheric pressure, Standard atmospheric pressure;

[0078] The formula for calculating the temperature correction factor is:

[0079] ;

[0080] Given the current ambient temperature, Standard temperature;

[0081] The formula for calculating the humidity correction factor is:

[0082] ;

[0083] Given the current ambient humidity, Standard humidity;

[0084] Example: if =0.1 kg / s, LHV = 43 × 10⁶ joules / kg, =0.98, EF=0.95, =10 kg / s =1005 joules / kg·Kelvin =600 Kelvin, =101325 Pascals =101325 Pascals =293 Kelvin, =288 Kelvin, =10 grams per cubic meter =12 grams per cubic meter; therefore, the turbine inlet gas temperature is:

[0085] Kelvin;

[0086] Establishing a digital twin model requires the creation of simulation models, including physical simulation models and thermodynamic models.

[0087] By analyzing flight data and based on the engine's working principle and physical characteristics, detailed models of various engine components (such as the combustion chamber, turbine, and compressor) are created to establish a physical simulation model of the engine; and by simulating heat transfer and energy conversion processes, a thermodynamic model of the engine is established.

[0088] We analyze sensor and performance parameters, extract the most relevant features using PCA, predict turbine inlet gas temperature using linear regression, build a data-driven model using CNN, and improve model accuracy using random forest algorithm.

[0089] Digital twin models can simulate aircraft performance in real time. Combining physical simulation models and data-driven models, they can simulate the performance of aircraft under different environmental conditions and provide more accurate and comprehensive performance predictions.

[0090] Among them, the most relevant features include the temperature and pressure of various engine components, speed and rotational acceleration, torque, humidity, etc.; the dimensions, density, elastic modulus, yield strength, hardness, thermal conductivity and electrical conductivity of various parts of the engine, etc.; and parameters such as air temperature, atmospheric pressure, and humidity.

[0091] The extraction of the most relevant features includes the following steps:

[0092] The data is standardized, and the original feature matrix is ​​calculated using the following formula:

[0093] ;

[0094] It is the mean of the features. It is the standard deviation of the feature;

[0095] Calculate the covariance matrix:

[0096] ;

[0097] Where n is the number of samples;

[0098] The eigenvalues ​​and eigenvectors are calculated using the eigenvalue decomposition formula:

[0099] ;

[0100] Where V is the eigenvector matrix, It is a diagonal matrix, and the matrices on the diagonal are the eigenvalues;

[0101] Arrange the eigenvalues ​​in descending order, select the eigenvectors corresponding to the first k eigenvalues ​​as principal components, and then convert the original feature matrix... Projecting onto the selected principal components yields a new feature matrix, calculated using the following formula:

[0102] ;

[0103] It is a matrix composed of the first k eigenvectors;

[0104] The simulation model and the data-driven model are fused using the Bayesian model averaging method to generate a digital twin model. The simulation results and the data-driven results are then fused using a weighted average to generate the predicted turbine inlet gas temperature of the engine.

[0105] As a further aspect of the present invention, step S2, establishing the environmental parameter gradient, includes the following steps:

[0106] Set the temperature parameter gradient. In the takeoff phase folder, climb phase folder, descent phase folder, approach phase folder, and landing phase folder, the internal temperature parameter gradient is set with -40 degrees Celsius as the temperature origin. The difference between the maximum and minimum values ​​in the temperature gradient is recorded as the range, which is 90 in this case. Set the tolerance t1.

[0107] Example: If the tolerance is 2, then the temperature gradient range is [-40, 50], and there are a total of 45 sets of parameters;

[0108] In the cruise phase folder, the temperature parameter gradient inside starts at -50 degrees Celsius, with a range of 30, and a tolerance of t2 is set.

[0109] Example: If the tolerance is 2, then the temperature gradient range is [-50, -20], and there are a total of 15 sets of parameters;

[0110] Set atmospheric pressure parameter gradients. In the folders for takeoff, climb, descent, approach, and landing, the atmospheric pressure parameter gradients are all set with 100 Pa as the starting point and 100 as the range, with a tolerance of d1.

[0111] Example: If the tolerance is 2, then the atmospheric pressure gradient range is [100, 200], and there are a total of 50 sets of parameters;

[0112] In the cruise phase folder, the atmospheric pressure parameter gradient inside is set with 200 Pa as the starting point, 100 as the range, and tolerance d2.

[0113] Example: If the tolerance is 2, then the atmospheric pressure gradient range is [200, 300], and there are a total of 50 sets of parameters;

[0114] Set humidity parameter gradients. In the takeoff phase folder, climb phase folder, descent phase folder, approach phase folder, and landing phase folder, the humidity gradients within each folder start at 0% and have a range of 100%. Set the tolerance s1.

[0115] Example: If the tolerance is 2%, then the humidity gradient range is [0%, 100%], and there are a total of 50 sets of parameters;

[0116] During the cruise phase, the humidity parameter gradient within the folder starts at 0% and has a range of 10%, with a tolerance of s2.

[0117] Example: If the tolerance is 2%, then the humidity gradient range is [0%, 10%], and there are a total of 5 sets of parameters;

[0118] By setting parameter gradients to fully cover the parameter variation range of different flight stages, the performance and behavior of the engine under different flight stages and environmental conditions can be better simulated and analyzed. This parameter setting method not only improves the accuracy and reliability of the model, but also increases the diversity and richness of the data, thus providing strong support for design and optimization.

[0119] As a further aspect of the present invention, a standard dataset is generated based on the gradient of environmental parameters, including the following steps:

[0120] Based on the engine's current stage, from the three sets of parameters—temperature, atmospheric pressure, and humidity—within the environmental parameter gradient, one parameter is extracted from each set in a permutation and combination manner to form an input array. Within the flight data, the temperature, atmospheric pressure, and humidity parameters in each subfolder are defined as a parameter group. In all subfolders, each input array is traversed and searched to find the same parameter group. The engine performance and parameter data of the parameter group are then overwritten with the corresponding engine performance and parameter data of the input array. The updated input array is then input into the digital twin model to obtain the turbine inlet gas temperature, generating a dataset.

[0121] Depending on the engine's current stage, allowable error ranges are set for temperature, atmospheric pressure, and humidity parameters. For example, the allowable error range for temperature parameters is... The allowable error range for atmospheric pressure parameters is 0.5 degrees. 1 Pa, the allowable error range for humidity parameters is 2g / m³; and update each set of input arrays within the environmental parameter gradient, mask the parameter sets contained in a class of datasets, and perform data matching on the updated standard data; in all subfolders, traverse and search each set of updated input arrays to find the parameter sets within the error range of the input arrays, and overwrite the engine performance and parameter data of the corresponding input arrays with the engine performance and parameter data of the parameter sets, input the updated input arrays into the digital twin model to obtain the turbine inlet gas temperature, and generate a second-class dataset;

[0122] By setting an error range, more subtle changes in environmental parameters that may occur in actual flight can be captured, thereby improving the coverage of the dataset. This helps the model better understand and predict engine performance under different environmental conditions. At the same time, setting an error range appropriately can control the number of samples while ensuring data quality, making model training more efficient and effective.

[0123] Generate standard datasets, including Class I and Class II datasets.

[0124] S3. Analyze the relationship between the turbine inlet gas temperature and the actual value measured by the sensor in the standard dataset, calculate the residuals, set the residual standard, and analyze the performance of the digital twin model.

[0125] As a further aspect of the present invention, step S3, which involves calculating the residuals, setting residual standards, and analyzing the performance of the digital twin model, includes the following steps:

[0126] The difference between a dataset in the standard data and the turbine inlet gas temperature corresponding to the same parameter group in the sensor parameters is calculated, and the absolute value of the calculation result is taken as the residual to generate a residual set.

[0127] For a set of residuals, a residual limit c1 is defined. If, within a set of residuals, a certain residual value in the flight data, sorted by engine usage time, has more than 1% of the residuals greater than or equal to the residual limit c1 in the 1000 nearest flight records, then the flight data is optimized, i.e., data augmentation is performed. This includes data enhancement, data expansion, and data synthesis. The data in each stage folder and its subfolders in the flight data are processed, standardized, fused, and enhanced to finally generate processed flight data. The digital twin model building operation is repeated until the turbine inlet gas temperature is less than the residual limit c1.

[0128] The difference between the two types of datasets in the standard data and the turbine inlet gas temperature corresponding to the same parameter group in the sensor parameters is calculated, and the absolute value of the calculation result is taken as the residual to generate a two types of residual set.

[0129] For the second type of residual set, a residual limit c2 is defined. If, within the second type of residual set, a certain residual value in the flight data, sorted by engine usage time, has more than 1% of the residuals within the nearest 1000 flight records that are greater than or equal to the residual limit c2, then the flight data is optimized, i.e., data augmentation is performed, including data value-added, data expansion, and data synthesis. The data in each stage folder and its subfolders in the flight data are processed, standardized, fused, and enhanced to finally generate processed flight data. The digital twin model building operation is repeated until the turbine inlet gas temperature is less than the residual limit c2.

[0130] Calculate and analyze residuals to optimize the model through feedback, significantly improving its accuracy and reliability. Define residual limits c1 and c2 to quantify model performance standards and ensure that prediction errors on different datasets remain within acceptable ranges. Analyze the residual set to identify the proportion of residuals greater than or equal to the residual limits in the most recent 1000 flight records, promptly identifying performance deficiencies under specific conditions. Address these deficiencies with data augmentation, including data enhancement, data expansion, and data synthesis, enriching the dataset, increasing its diversity and representativeness, and enhancing the model's generalization ability. Repeat the digital twin model building process to ensure the model's prediction accuracy meets expected standards under various flight conditions. Continuously optimize model performance to improve robustness and adaptability, ensuring accurate prediction of engine turbine inlet gas temperature in practical applications, providing reliable data support for engine performance evaluation and optimization.

[0131] S4. Establish a prediction model, input a standard dataset, execute the Gaussian process regression analysis algorithm, predict the turbine inlet gas temperature for a future period and generate the standard deviation of the predicted values, generate a prediction dataset, analyze the predicted values ​​and the standard deviation of the predicted values, and optimize the prediction model.

[0132] In step S4, generating the prediction dataset includes the following steps:

[0133] To simulate the environmental conditions during aircraft operation, several sets of environmental parameters are set. Under each set of environmental parameters, the midpoints of the takeoff, climb, cruise, descent, approach, and landing phases are defined as fixed nodes. Using a digital twin model, the aircraft is simulated to operate at a fixed frequency, generating a series of turbine inlet gas temperature values. The standard for the fixed frequency is: a total prediction duration of ten days, five flights per day, and six fixed nodes recorded each time according to the flight phase. Based on the flight phase, a series of turbine inlet gas temperature values ​​are input into the prediction model to generate predicted turbine inlet gas temperatures for several engines and the corresponding standard deviations of the predicted values, forming a prediction dataset.

[0134] Analyzing the prediction dataset and optimizing the prediction model includes the following steps:

[0135] Define a limit z for the standard deviation of the predicted value. If the standard deviation of the predicted value is greater than z, it is considered that the corresponding fixed node of the input has fewer than the normal value in the training of the prediction model. Modify the number of daily attendances in the fixed frequency until the standard deviation of the predicted value is less than or equal to z.

[0136] Calculate the mean of the predicted values, and then calculate the confidence interval of the predicted values ​​using the mean and the standard deviation of the predicted values; calculate the range length of the confidence interval, &; define the lower limit of the standard range length as &1 and the upper limit of the standard range length as &2;

[0137] If the range length & is less than the lower limit of the standard range length &1, the allowable error range in the binary dataset is increased by increasing the number of input array groups until the range length & is greater than or equal to the lower limit of the range length &1.

[0138] If the range length & is greater than the lower limit of the standard range length of &2, the allowable error range in the binary dataset is increased by increasing the number of input array groups until the range length & is less than or equal to the lower limit of the range length of &2.

[0139] By defining the standard deviation limit z of the predicted values ​​and adjusting the daily attendance frequency in the fixed frequency range, sufficient training data for the prediction model is ensured, thus improving the accuracy of the predictions. The mean and confidence interval of the predicted values ​​are calculated, and the range length of the confidence interval is analyzed. The lower limit of the standard range length is defined as &1 and the upper limit as &2. The allowable error range in the binary dataset is adjusted, and the number of input array groups is increased to ensure that the confidence interval of the predicted values ​​is within a reasonable range, thereby enhancing the robustness and adaptability of the model. The prediction dataset is analyzed, the prediction model is optimized, and the accuracy and reliability of the predictions are improved, providing comprehensive and reliable data support for engine performance evaluation and optimization.

[0140] Example 2

[0141] Please see Figure 2 A flight data-based engine performance prediction system, specifically employing a flight data-based engine performance prediction method to predict engine performance, includes:

[0142] Data Collection and Preprocessing Unit: Used to collect historical normal operating status data of the engine, including engine performance and parameter data and engine operating environment parameter data. It preprocesses the data to generate flight data.

[0143] Digital Twin Model Building and Calculation Unit: Used to build a digital twin model, input flight data, execute regression analysis algorithms, and calculate the turbine inlet gas temperature of the engine according to the time series. It also establishes environmental parameter gradients, inputs them into the digital twin model, obtains the turbine inlet gas temperature corresponding to each set of environmental parameters, and generates a standard dataset.

[0144] Standard Dataset Analysis Unit: Used to analyze the relationship between turbine inlet gas temperature and actual values ​​measured by sensors within a standard dataset, calculate residuals, set residual standards, and analyze the performance of the digital twin model;

[0145] Prediction Model Building and Optimization Unit: Used to build prediction models, input standard datasets, execute Gaussian process regression analysis algorithms, predict turbine inlet gas temperature for a future period and generate prediction standard deviations, generate prediction datasets, analyze prediction values ​​and prediction standard deviations, and optimize prediction models;

[0146] Data augmentation and model optimization unit: used to optimize flight data based on residual analysis results.

[0147] Example 3

[0148] In its specific implementation, this embodiment includes all the implementation processes of the two sets of embodiments described above.

[0149] Some of the data in the above formulas are numerical calculations with dimensions removed, and the contents not described in detail in this specification are all prior art known to those skilled in the art.

[0150] The above embodiments are only used to illustrate the technical methods of the present invention and are not intended to limit it. 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 be made to the technical methods of the present invention without departing from the spirit and scope of the technical methods of the present invention.

Claims

1. A method for predicting engine performance based on flight data, characterized in that, Includes the following steps: S1. Collect historical normal operating status data of the engine, including engine performance and parameter data and engine operating environment parameter data; preprocess the historical normal operating status data to generate flight data; S2. Establish a digital twin model, input flight data, execute regression analysis algorithm, calculate the turbine inlet gas temperature of the engine according to the time series; establish environmental parameter gradients covering different flight stages, input them into the digital twin model, obtain the turbine inlet gas temperature corresponding to each set of environmental parameters, and generate a standard dataset, which includes a first-class dataset and a second-class dataset. S3. Analyze the relationship between the turbine inlet gas temperature and the actual value measured by the sensor in the standard dataset, calculate the residuals, set the residual standard, and analyze the performance of the digital twin model. S4. Establish a prediction model, input a standard dataset, execute the Gaussian process regression analysis algorithm, predict the turbine inlet gas temperature for a future period and generate the standard deviation of the predicted values, generate a prediction dataset, analyze the predicted values ​​and the standard deviation of the predicted values, and optimize the prediction model.

2. The method for predicting engine performance based on flight data according to claim 1, characterized in that, In step S1, the historical normal working status data is preprocessed to generate flight data, including the following steps: The data is divided into six folders according to the takeoff phase, climb phase, cruise phase, descent phase, approach phase, and landing phase. Each folder is further divided into subfolders according to the flight environment. Flight environment parameters include temperature, atmospheric pressure, and humidity. Each subfolder contains a set of flight environment parameters and the corresponding engine performance parameters, sensor parameters, and material data for that flight environment. For the data in each folder, missing value processing, outlier processing, and noise data processing are performed. Data from different sources and formats are standardized to ensure consistency. Multi-source data are then merged to generate flight data.

3. The method for predicting engine performance based on flight data according to claim 1, characterized in that, In step S2, the turbine inlet gas temperature of the engine is calculated according to the time series, including the following steps: For the six phases—takeoff, climb, cruise, descent, approach, and landing—the subfolders within their respective folders are sorted by recorded time. The data from each subfolder is then input into a digital twin model to calculate the predicted turbine inlet gas temperature. The calculation process is as follows: Among them, T TET For the final turbine inlet gas temperature, m fuel Where LHV is the fuel flow rate, η is the lower heating value of the fuel, and η is the fuel flow rate. combustion For combustion efficiency, EF is the engine efficiency factor, m air For airflow, c p A is the specific heat capacity of air; B is the atmospheric pressure correction factor; C is the temperature correction factor; and T is the humidity correction factor. compressor This refers to the compressor outlet temperature. The formula for calculating the atmospheric pressure correction factor is: Among them, P ambient Given the current atmospheric pressure, P standard Standard atmospheric pressure; The formula for calculating the temperature correction factor is: T ambient Given the current ambient temperature, T standard Standard temperature; The formula for calculating the humidity correction factor is: H ambient H represents the current ambient humidity. standard This is the standard humidity.

4. The method for predicting engine performance based on flight data according to claim 1, characterized in that, In step S2, establishing the environmental parameter gradient includes the following steps: Set the temperature parameter gradient. In the takeoff phase folder, climb phase folder, descent phase folder, approach phase folder, and landing phase folder, the internal temperature parameter gradient is set with -40 degrees Celsius as the temperature origin. The difference between the maximum and minimum values ​​in the temperature gradient is recorded as the range, which is 90 in this case. Set the tolerance t1. In the cruise phase folder, the temperature parameter gradient inside starts at -50 degrees Celsius, with a range of 30, and a tolerance of t2 is set. Set atmospheric pressure parameter gradients. In the folders for takeoff, climb, descent, approach, and landing, the atmospheric pressure parameter gradients are all set with 100 Pa as the starting point and 100 as the range, with a tolerance of d1. In the cruise phase folder, the atmospheric pressure parameter gradient inside is set with 200 Pa as the starting point, 100 as the range, and tolerance d2. Set humidity parameter gradients. In the takeoff phase folder, climb phase folder, descent phase folder, approach phase folder, and landing phase folder, the humidity gradients within each folder start at 0% and have a range of 100%. Set the tolerance s1. In the cruise phase folder, the humidity parameter gradient inside is set with a starting point of 0% and a range of 10%, with a tolerance of s2.

5. The method for predicting engine performance based on flight data according to claim 4, characterized in that, Based on the gradient of environmental parameters, a standard dataset is generated, including the following steps: Based on the engine's current stage, from the three sets of parameters—temperature, atmospheric pressure, and humidity—within the environmental parameter gradient, one parameter is extracted from each set in a permutation and combination manner to form an input array. Within the flight data, the temperature, atmospheric pressure, and humidity parameters in each subfolder are defined as a parameter group. In all subfolders, each input array is traversed and searched to find the same parameter group. The engine performance and parameter data of the parameter group are then overwritten with the corresponding engine performance and parameter data of the input array. The updated input array is then input into the digital twin model to obtain the turbine inlet gas temperature, generating a dataset. Based on the engine's current stage, allowable error ranges are set for temperature, atmospheric pressure, and humidity parameters. Each input array within the environmental parameter gradient is updated, and parameter sets contained in one type of dataset are masked. Data matching is performed on the updated standard data. In all subfolders, each updated input array is traversed and searched to find parameter sets within the input array's error range. The engine performance and parameter data of the parameter sets are then overwritten with the corresponding engine performance and parameter data of the input array. The updated input array is then input into the digital twin model to obtain the turbine inlet gas temperature, generating a second-class dataset.

6. The method for predicting engine performance based on flight data according to claim 1, characterized in that, In step S3, the residuals are calculated, residual standards are set, and the performance of the digital twin model is analyzed, including the following steps: The difference between a dataset in the standard data and the turbine inlet gas temperature corresponding to the same parameter group in the sensor parameters is calculated, and the absolute value of the calculation result is taken as the residual to generate a residual set. For a set of residuals, a residual limit c1 is defined. If, within a set of residuals, a certain residual value exists in the flight data, sorted by engine usage time, and within the 1000 nearest flight records, more than 1% of the residuals are greater than or equal to the residual limit c1, then the flight data is optimized, i.e., data augmentation is performed. This includes data enhancement, data expansion, and data synthesis. The data in each stage folder and its subfolders in the flight data are processed, standardized, fused, and enhanced to finally generate processed flight data. The digital twin model building operation is repeated until the turbine inlet gas temperature is less than the residual limit c1. The difference between the two types of datasets in the standard data and the turbine inlet gas temperature corresponding to the same parameter group in the sensor parameters is calculated, and the absolute value of the calculation result is taken as the residual to generate a two types of residual set. For the second type of residual set, a residual limit c2 is defined. If, within the second type of residual set, a certain residual value in the flight data, sorted by engine usage time, has more than 1% of the residuals within the nearest 1000 flight records that are greater than or equal to the residual limit c2, then the flight data is optimized, i.e., data augmentation is performed. This includes data enhancement, data expansion, and data synthesis. The data in each stage folder and its subfolders in the flight data are processed, standardized, fused, and enhanced to finally generate processed flight data. The digital twin model building operation is repeated until the turbine inlet gas temperature is less than the residual limit c2.

7. The method for predicting engine performance based on flight data according to claim 1, characterized in that, In step S4, generating the prediction dataset includes the following steps: To simulate the environmental conditions during aircraft operation, several sets of environmental parameters are set. Under each set of environmental parameters, the midpoints of the takeoff, climb, cruise, descent, approach, and landing phases are defined as fixed nodes. Using a digital twin model, the aircraft is simulated to operate at a fixed frequency, generating a series of turbine inlet gas temperature values. The standard for the fixed frequency is: a total prediction duration of ten days, five flights per day, and six fixed nodes recorded each time according to the flight phase. Based on the flight phase, a series of turbine inlet gas temperature values ​​are input into the prediction model to generate predicted turbine inlet gas temperatures for several engines and the corresponding standard deviations of the predicted values, forming a prediction dataset.

8. The method for predicting engine performance based on flight data according to claim 7, characterized in that, Analyzing the prediction dataset and optimizing the prediction model includes the following steps: Define a limit z for the standard deviation of the predicted value. If the standard deviation of the predicted value is greater than z, it is considered that the corresponding fixed node of the input has fewer than the normal value in the training of the prediction model. Modify the number of daily attendances in the fixed frequency until the standard deviation of the predicted value is less than or equal to z. Calculate the mean of the predicted values, and then calculate the confidence interval of the predicted values ​​using the mean and the standard deviation of the predicted values; calculate the range length of the confidence interval, &; define the lower limit of the standard range length as &1 and the upper limit of the standard range length as &2; If the range length & is less than the lower limit of the standard range length &1, the allowable error range in the binary dataset is increased by increasing the number of input array groups until the range length & is greater than or equal to the lower limit of the range length &1. If the range length & is greater than the lower limit of the standard range length of &2, the allowable error range in the binary dataset is increased by increasing the number of input array groups until the range length & is less than or equal to the lower limit of the range length of &2.

9. An engine performance prediction system based on flight data, characterized in that, The system specifically employs the flight data engine performance prediction method as described in any one of claims 1-8 to predict engine performance.

Images