Performance self-adaptive modeling method for aircraft engine individuals

Abstract

The invention discloses an aero-engine individual-oriented performance self-adaptive modeling method, which is suitable for a large bypass ratio turbofan engine with a multi-rotating component configuration. The method comprises the following steps: identifying a high-sensitivity correction factor based on combined sensitivity analysis of a matrix condition number and TOPSIS; an improved badger algorithm and a nonlinear solver are adopted for serial combination optimization, and modeling is conducted on the correction factors in a staged mode; physical feasibility screening and trend limitation are applied, and solution divergence or distortion is avoided; and external duct leakage and combustion heat release coefficients are introduced, a Bezier curve-based modeling error compensation function is constructed, and continuation from a steady state to dynamic precision is realized. The method gives consideration to algorithm convergence and physical rationality, can efficiently adapt to individual differences of engines, improves modeling precision, stability and robustness, and is suitable for scenes such as individual modeling, model correction and health management.

Description

1. Technical Field

[0001] This invention relates to the field of aero-engine health management technology, and in particular to a performance adaptive modeling method and system for individual aero-engines with multiple rotating components, belonging to the category of component-level model correction and optimization algorithm fusion application. 2. Background Technology

[0002] Currently, individual aero-engines exhibit significant performance differences due to variations in manufacturing processes, assembly deviations, and service conditions. While numerous studies have explored the use of nonlinear solvers (NS) for engine performance adaptation and modeling, NS methods are susceptible to numerical solution failures and physical constraint violations, particularly for engines with large initial model errors and pronounced individual differences. This can lead to solution failures (such as pressure ratios exceeding surge boundaries), limiting their application scope. Furthermore, traditional intelligent optimization algorithms are prone to getting trapped in local optima, exhibiting slow convergence speeds and poor stability when dealing with high-dimensional, multi-objective problems with large initial biases.

[0003] Therefore, there is an urgent need for a performance modeling method that balances numerical solution stability and optimization search capabilities, and is applicable to aero-engines with significant individual differences. 3. Summary of the Invention

[0004] This invention provides an adaptive modeling method and system for aero-engine performance based on an optimized combination strategy. Starting from a general mathematical model, it can quickly complete high-precision modeling of different individual engines by integrating a nonlinear solver and an intelligent optimization algorithm, thus overcoming the shortcomings of existing technologies in terms of modeling stability, applicability, and efficiency.

[0005] Technical solution: To achieve the above objectives, the technical solution adopted by this invention is as follows:

[0006] An adaptive performance modeling method for individual aero-engines includes the following steps:

[0007] Step A), the data preprocessing stage, involves noise filtering and steady-state condition identification of the flight measurement data, and extracting effective steady-state data for modeling.

[0008] Step B), sensitivity analysis of the modified factors: Based on the matrix condition number and entropy weight TOPSIS evaluation method, the characteristic modified factors (AF) are comprehensively ranked to identify high-sensitivity factors and low-sensitivity factors.

[0009] Step C), Performance Correction Stage: A serial combinatorial optimization method is used to correct the performance parameters of the five rotating components of the engine. This optimization method includes: an improved Honey Badger Algorithm (IHBA) with adjustable convergence speed to adapt to different initial deviations; a nonlinear solver (NS) module acting as a filter and compensator to handle highly sensitive correction parameters, improving optimization stability and accuracy; and an iterative constraint mechanism that imposes physical constraints on the pressure ratio, efficiency value, and AF trend to prevent solution divergence.

[0010] Step D), Model Error Compensation Stage: Based on the steady-state optimization solution, the leakage coefficient of the outer bypass duct and the heat release fluctuation coefficient of the combustion chamber are fitted by Bezier curve to achieve rapid error compensation of the mathematical model and improve the accuracy of dynamic simulation.

[0011] As a performance adaptive modeling method for individual aero-engines according to the present invention, the specific steps of step A) are as follows:

[0012] Step A1), data noise processing: In this invention, the low-pressure rotor speed n1, the high-pressure rotor speed n2, and the total inlet pressure P of the high-pressure compressor are selected. 25 The total outlet temperature T3 of the high-pressure compressor and the total outlet temperature T5 of the low-pressure turbine are used as measurement parameters. exp The engine's external input parameters u include height H, Mach number Ma, fan inlet total temperature T2, fan inlet total pressure P2, and fuel flow rate W. f And the guide vane angle VSV. Its physical testing and mathematical simulation descriptions are as follows:

[0013] y meas =f exp (u)+y nose

[0014] p = f CLM (AF,u)

[0015] Where p is the CLM output parameter, f exp and f CLM These represent the mappings between the physical object and the mathematical model's input and output, respectively. noise and y meas These are the noise measurement and the test run measurement values, respectively.

[0016] For the raw flight or test data, a moving average filtering method is used to denoise the data, thereby reducing the impact of measurement noise on the performance modeling results.

[0017] Step A2), steady-state point extraction: Identify the steady-state operating point based on the judgment rules of the rate of change of speed, temperature, etc., and extract key input parameters and corresponding measurement output parameters from it for subsequent performance correction.

[0018] As a performance adaptive modeling method for individual aero-engines according to the present invention, the specific steps of step B) are as follows:

[0019] Step B1), Calculation of matrix condition number: Square matrix A∈R n×n The matrix condition number is defined as κ(A) = ||A||||A -1 ||, where ||A|| is the norm of A. For the linear system of equations A·x=b, x and b are n-dimensional vectors. Assuming the error in the observed value b is Δb, then the error between the calculated solution x and the true solution is Δx=A. -1 (b+Δb)-A -1 b = A -1 Δb. From the derivation formula of the matrix condition number, we can finally obtain... This means that if the condition number of matrix A is large, the greater the propagation of observation errors, the more sensitive the system is to these errors, thus affecting the stability of the solution. This invention uses the characteristic correction coefficients as solutions to the linear equation system and the measured parameters as observed values. The larger κ(A) is, the more unstable the solution for the corresponding correction coefficients becomes. For each operating point, a linear approximation system is constructed, and the condition number of the Jacobian matrix corresponding to the correction factor is solved using the measured parameters as the observation vector, evaluating its sensitivity to the output solution.

[0020] Step B2), Entropy-weighted TOPSIS comprehensive analysis: Under multiple operating points, the condition number index is normalized. Since the matrix condition number is a positive index, the evaluation index of m operating points with n characteristic correction coefficients is normalized to obtain matrix z. ij ,

[0021]

[0022] The entropy method is used to allocate the index weights, and the proportion of the ij-th item to the i-th AF is ρ. ij And assign the entropy value q to the j-th index. j The entropy weight Q of the final index is calculated and obtained. j ,

[0023]

[0024] Further construct the centralized standardized decision matrix Z ij =x ij Q j Optimal solution and worst-case scenario make and The Euclidean distance is the characteristic correction factor.

[0025]

[0026] The final comprehensive index obtained by CN-TOPSIS is C i The larger the value, the higher the sensitivity of the corresponding correction coefficient, and the more likely it is to lead to poor numerical solution stability. The TOPSIS method is used to sort all correction factors to quantify the sensitivity of each correction factor.

[0027] Step B3), Correction factor classification: The sensitivity ranking results are used to guide the grouping of subsequent optimization parameters. Local solutions are used for high-sensitivity parameters, and global search optimization is used for low-sensitivity parameters to improve the stability of the solution.

[0028] As a performance adaptive modeling method for individual aero-engines according to the present invention, the specific steps of step C) are as follows:

[0029] Step C1) Construct a modified model. Based on the component-level model, introduce a characteristic correction factor AF. Since the parameters of the component characteristic diagram have strong nonlinearity, they need to be corrected using flow rate, pressure ratio, and efficiency correction factors. The correction factor is the corrected parameters (flow rate, pressure ratio, and efficiency) of the characteristic diagram divided by the original parameters, as shown below.

[0030]

[0031] To improve the granularity of the correction method, AF is expressed using the following formula for different conversion speed lines.

[0032]

[0033] The flow rate, pressure ratio, and efficiency curves of the rotating component are adjusted in a polynomial form.

[0034] Step C2), Improved Honey Badger Algorithm Global Search: The Honey Badger Algorithm (HBA) is a novel metaheuristic optimization algorithm that has advantages over traditional algorithms in solving complex search space optimization problems. Inspired by the foraging behavior of honey badgers, which can use smell to locate food (the digging stage) and rely on honeybees to locate beehives (the bee stage), the specific process is as follows:

[0035] Initialization phase: Initialize the solution set from HL to HU.

[0036] H i =HL i +r1×(HU i -HL i ), i = 1, 2, ..., n

[0037] Where n is the number of parameters to be optimized, H is the population, and r is a random number between 0 and 1, totaling seven.

[0038] Update population:

[0039] During the exploration phase, the movement of potential predator species is influenced by the intensity of predator scent and the distance Hp between the honey badger and the optimal prey. The updated population update expression is as follows:

[0040] H new =H P +F g ×β×I×H p +F g ×r3×α×d i ×|cos(2πr4)×[1-cos(2πr5)]|

[0041] Where β represents the ability to collect food, I represents the odor intensity, and Fg represents the search direction indication.

[0042] During the honey guide phase, when searching for the beehive, the following formula is used to adjust the honey badger's position relative to the honey guide bird.

[0043] H new =H p +F g ×r t ×α×(H p -H i )

[0044] In the two formulas above, the olfactory parameter and the density parameter are defined as follows:

[0045] α=C×exp(-t / t max )

[0046]

[0047] d i =(H p -H i )

[0048] HBA updates particles randomly using two strategies: mining and guiding, until the maximum number of iterations is reached or the objective function is satisfied. The objective function in the model correction is defined as follows.

[0049]

[0050] The objective function is the maximum error across all operating points; the smaller the objective function, the better.

[0051] Based on initial individual differences, engine objects are divided into four categories: maximum difference (ED), large difference (MaD), medium difference (MoD), and small difference (MD). A dynamic adjustment factor FC for convergence speed is set, which is expressed as follows for different initial deviations in the first half of the maximum iteration count:

[0052]

[0053] In the second half, F c for,

[0054] Fc = 2 - log(t) / log(t) max )t>t max / 2

[0055] Where δ is the maximum adjustment range, set based on engineering experience, and r is a random number between 0 and δ. Using a dynamic adjustment factor, personalized convergence paths are achieved for different individuals within IHBA.

[0056] Step C3), Nonlinear Solver NS Screening Mechanism: The principle of the nonlinear solver is to expand the dimension of the equilibrium equations based on the common working equations of the component-level model, using existing experimental data, and then solve for the newly added independent variables. Assuming there are N original common working equations, the number of new equations is determined by the number of measured parameters. Taking the measured parameters T3 and T5 as the selected construction parameters, and AF as the newly added independent variable... η,HPC and AF η,HPT The new equation is as follows:

[0057]

[0058] Among them, f op，N+i For the (N+i)th common working equation, T i,meas For measurement data, T i Calculate data for the model.

[0059] An extended set of equations is established for highly sensitive factors. The residual equations are constructed using the measured values ​​and solved using the Newton-Raphson method. At the same time, solutions that are unsolvable or do not meet the physical conditions (such as pressure ratio less than 1 or efficiency greater than 1) are eliminated.

[0060] Step C4), Serial combination solution and boundary constraints: The NS module is embedded in IHBA. In each optimization iteration, the NS method obtains a set of AFs for the experimental data. For m steady-state measurement points, the correction coefficients are calculated using the following formula.

[0061]

[0062] Use AF iThe characteristic plot is adjusted to screen and reconstruct the optimal solution. 1HBA mainly optimizes the low-sensitivity parameters and provides stable initial values ​​for NS. NS feedback is reasonable to improve the stability and reliability of the solution.

[0063] Step C5), Optimization Process Constraints: The engine mathematical model involves iteration. If the correction coefficients obtained from a certain update of the population are inappropriate, it will lead to internal iteration divergence, resulting in a significant increase in the optimization time. Therefore, the following constraints are added: First, if the pressure ratio is less than 1 and the efficiency is greater than 1 during the numerical iteration of the mathematical model, the population is discarded. Second, ensure that the corrected characteristic curves do not exchange initial positions, i.e., limit AF to be less than the value of the next speed line divided by the value of the current speed line. Physical constraints are added between correction factors to prevent the AF curves from crossing or exhibiting physically meaningless changes, ensuring the interpretability of component performance.

[0064] As a performance adaptive modeling method for individual aero-engines according to the present invention, the specific steps of step D) are as follows:

[0065] Step D1), Steady-state benchmark extraction and error analysis: Based on the steady-state solution obtained in step C, construct the skeleton characteristics of the target object. Based on the physical mechanism, the initial modeling error of the engine is mainly attributed to the uncertainty of airflow leakage in the bypass duct and heat release in the combustion chamber.

[0066] Step D2), construct the compensation parameter function: introduce the outer bypass leakage correction factor (CCG) and the combustion chamber heat release correction factor (CCH) as performance compensation factors, and use the constrained interior-point method for optimization solution;

[0067] Step D3), Characteristic curve fitting and compensation: The relationship between the correction factor obtained in step D2 and the rotation speed is fitted by Bezier curve to form a continuously variable modeling error compensation curve, which is used to correct the gas and heat parameters in the original model to achieve high-precision dynamic performance matching.

[0068] Compared with existing technologies, the present invention, employing the above technical solution, forms a performance modeling process that is individual-oriented, structured, and reasonably constrained. This process enables accurate, stable, and efficient modeling in scenarios with limited measurement parameters, large correction amounts, and significant initial deviations. Its advantages are:

[0069] (1) The problem of solution failure of nonlinear solvers under large deviation conditions has been solved;

[0070] (2) It innovatively integrates intelligent optimization algorithms with the NS method to avoid search blind spots and invalid iterations;

[0071] (3) It realizes a seamless transition from steady-state modeling to dynamic simulation, improving the feasibility of practical applications and the credibility of the model. 4. Description of the attached drawings

[0072] Figure 1 This is a structural diagram of a performance adaptive modeling method for individual aero-engines according to the present invention.

[0073] Figure 2 This is a graph showing the results of the sensitivity analysis of the characteristic correction factor;

[0074] Figure 3 This is a schematic diagram of the improved convergence factor variation curve;

[0075] Figure 4 This is a schematic diagram of the optimization process of the serial combination of NS and IHBA;

[0076] Figure 5 This is a schematic diagram illustrating the failure of a nonlinear solver under conditions of large individual deviations.

[0077] Figure 6 This is a complete modeling flowchart considering error compensation;

[0078] Figure 7 This is a comparison chart of the results of compressor characteristic diagram correction using the traditional method and the present invention;

[0079] Figure 8 This is a comparison chart of the output curves of the main performance parameters in the timing simulation and the actual measured values.

[0080] Figure 9 This is a comparison chart showing the correction accuracy between the present invention and traditional algorithms;

[0081] Figure 10 This is a graph showing the cross-validation results under the condition of missing measurement parameters. 5. Detailed Implementation

[0082] The technical solution of the present invention will be further described in detail below with reference to the accompanying drawings.

[0083] The specific implementation steps and verification process of the method of the present invention are described in detail using a certain type of high bypass ratio turbofan engine as an example, under the condition of limited flight data and large individual differences:

[0084] Step A), sensitivity analysis of characteristic correction factors;

[0085] Step B), serial combination optimization correction method;

[0086] Step C), stability enhancement and modeling error compensation;

[0087] Step D), result verification and comparative evaluation.

[0088] The detailed steps of step A) are as follows:

[0089] Step A1), parameter preparation and model building: select 15 characteristic correction factors of five rotating parts of the engine as the analysis objects, and build the input-output mapping relationship based on the component-level model;

[0090] Step A2) uses the matrix condition number method to evaluate single-point sensitivity and TOPSIS to rank multi-point sensitivity. Under fixed input parameters, the matrix condition number of each correction factor is calculated for each operating point to evaluate its perturbation propagation effect on the output. Thirteen operating points are selected under different flight conditions within the flight envelope of a civil aviation engine. These operating points are then used to evaluate the performance of the finitely measurable parameter y. meas The most influential characteristic correction coefficients are identified and ranked. Taking a single operating point (ground test conditions) as an example, when calculating AF... i When calculating the matrix condition number, the input parameter u is kept constant, while other characteristic correction coefficients are perturbed by a fixed value ΔAF, and ΔAF... i By perturbing by 2%-5% sequentially, AF is obtained. i The matrix condition number under different perturbations; repeating the above work for all characteristic correction coefficients, the parameter sensitivity results at this operating point are obtained, as shown in the figure. Figure 2 As shown in (a). Finally, TOPSIS is used to synthesize the results of multiple operating points to obtain an objective comprehensive index ranking of the characteristic correction coefficients, as follows: Figure 2 As shown in (b), it is difficult to obtain accurate sensitivity analysis results that consider the global situation using only a single point CN. After introducing the TOPSIS method based on entropy weighting, Figure 2 (b) Indices are presented after considering all selected operating points. Under different perturbations, the compressor efficiency correction coefficient and the high-pressure turbine efficiency correction coefficient are significantly affected by the measured parameters more than other parameters, a conclusion consistent with the overall performance study results based on CFM56.

[0091] The detailed steps of step B) are as follows:

[0092] Step B1) Introduces an improved honey badger algorithm, defining different convergence rate adjustment factors FC for different individual difference levels (ED / MaD / MoD / MD) to control the population search size. Figure 3 This demonstrates F under different conditions when r=1. c The change curve of t shows a slow decrease in the early stages of iteration for those with large initial deviations, ensuring a comprehensive search of the solution domain. For those with moderate initial deviations, the decrease gradually accelerates in the early stages of iteration, ensuring both a comprehensive search and improved convergence speed. For those with small initial deviations, the scale continuously decreases in the early stages of iteration, locking onto the region near the target solution. And for deviations greater than t...max / 2, small-scale optimization is used in the vicinity of the objective solution for all three different cases.

[0093] Step B2) Constructs an NS-IHBA serial structure, using IHBA to optimize low-sensitivity parameters and globally explore the solution domain; and using the nonlinear solver (NS) for local solving and accuracy selection of high-sensitivity parameters, such as... Figure 4 As shown, the new random population starts from a local optimum and, through the mining and searching phases, attempts to find all combinations of optimization parameters that satisfy the fitness. Then, NS filters these combinations and retains the solutions that meet the accuracy requirements. The entire optimization process is completed within the constraints, avoiding the algorithm from wasting computational resources in unreasonable local areas.

[0094] Step B3) Establish the optimization objective function. The optimization objective function is defined as the minimum of the maximum relative error between the output and the measured value at multiple operating points, which is used to guide the solution to converge toward the global optimum.

[0095] The detailed steps of step C) are as follows:

[0096] Step C1), nonlinear solver selection mechanism. Although nonlinear solver-based methods have been successfully applied in model modification and model fault diagnosis, research on the applicability of such solvers to aero-engines, which exhibit significant individual differences and deviations from the initial model, remains scarce. Figure 5 As shown, at a speed of 0.95, the pressure ratio corresponding to the surge boundary is less than 6. However, if the pressure ratio at a speed of 0.92 is calculated to be 6 using experimental data from a large-deviation individual object, then the operating point will be completely outside the surge boundary of the initial characteristic diagram. This phenomenon, which contradicts actual physical criteria, can lead to unsolvable cases during model solving, thus causing the nonlinear solver method to fail. Therefore, the NS method is used to construct extended residual equations for sensitive parameters, and solutions with pressure ratio < 1 or efficiency > 1 are discarded.

[0097] Step C2) Correct the application of trend constraints and physical constraints. Apply constraints such as non-crossing and non-abrupt changes to the AF change curve to prevent the generation of non-physical correction curves.

[0098] Step C3) involves modeling an error compensation strategy. Based on the steady-state benchmark solution, the bypass duct leakage coefficient (CCG) and combustion chamber heat release coefficient (CCH) are introduced. A Bezier curve is used to fit their variation with engine speed, forming a dynamic compensation curve. The process of the proposed serial combined optimization scheme is as follows: Figure 6As shown, the steady-state point data extracted from the measurement data are first corrected. In the early stage, the NS-IHBA optimization population expands the solution domain within a set limit according to different individual differences and different convergence rates, reducing the error between measurement parameters and performance parameters. In the later stage, the target accuracy solution is selected under the guidance of the nonlinear solver. Based on the steady-state baseline obtained above, the flow-temperature characteristic curves of individual models are obtained according to the measurement data to compensate for the initial modeling error.

[0099] The detailed steps of step D) are as follows:

[0100] Step D1) takes a specific performance matching process as an example, compares and analyzes the differences between the traditional method and the proposed method from the perspective of characteristic map adjustment, and verifies the effectiveness of the proposed method through performance parameter simulation. This case uses 150,000 sampling points of measurement data with a sampling interval of 15ms. There are 6 cases where the engine is in a stable operating state. The initial model's maximum error under this measurement data is less than 30%, which is considered MD (Mean Differential). The traditional PSO optimization algorithm and NS-IHBA are used to correct all correction coefficients. The population size of the two algorithms is set to be the same, and they start from the same mathematical model. The external input u of the engine is the same. Measurement data y of the same engine is used. meas Steady-state baseline adaptation is performed. The stopping condition for the optimization algorithm is that the steady-state output p of the model equals the sum of all steady-state measurement parameters y. meas The maximum error is less than 2%. Despite having the same target accuracy, NS-IHBA, compared to traditional algorithms, processes optimization parameters in stages, eliminating unstable and unreasonable factors during the optimization process. The comparison of the characteristic maps of the compressed components before and after correction is shown below. Figure 7 As shown.

[0101] Figure 8 Performance matching results of the combined optimization scheme at different stages are presented. By correcting the steady-state point, NS-IHBA can adjust the output of the initial model to be close to the measured value. Although the steady-state optimization target accuracy is less than 2%, in the time-series calculation, due to the existence of modeling errors, the error with the measured data will be greater than 5%. Therefore, using the NS-IHBA results as the steady-state benchmark, further adjustments are made to the bypass gas flow and combustion chamber heat release fluctuations to generate... Figure 8 The characteristic curve shown in (f) quickly compensates for modeling errors. The combustion chamber heat release characteristic curve on the steady-state baseline is relatively stable, indicating a relatively stable test environment. However, in the low-speed range, the bypass duct flow rate increases with the core engine speed, while in the high-speed range, the bypass duct flow rate tends to saturate. Excessively high speeds lead to a decrease in aerodynamic efficiency, and the bypass duct flow rate gradually decreases. Figure 8As shown in (d) and (e), this paper uses the area between performance parameters and measurement data as the optimization objective function, which is highly applicable to temperature parameters with measurement delays. The simplified compensation method can isolate the delay of external temperature sensors while effectively correcting modeling errors caused by internal heat fluctuations. With compensation based on the characteristic curve, the time-series calculation parameters of the steady-state reference model can quickly and accurately track the actual engine measurement output, providing a more accurate numerical simulation tool for the subsequent design of this individual engine.

[0102] Step D2) verifies accuracy. Simulations ensure consistent population sizes and objective functions for traditional PSO and NS-IHA. The stopping condition for the optimization algorithm is set to the relative error between the current optimal objective function and the global optimal objective function being less than 0.001, indicating that the algorithm has reached a local optimum. Corrections are made for individual engines with an initial difference of MoD. Its experimental data includes six steady-state operating points. All characteristic correction coefficients are adjusted, and all measurement data at all steady-state points are matched. The final accuracy results are as follows: Figure 9 As shown, the maximum and mean values ​​of the measured parameters in the NS-IHBA-corrected individual model at all steady-state points are less than 2%, and both are lower than those of the traditional method. This indicates that NS-IHBA outperforms PSO in terms of exploration capability, resulting in an overall reduction in the error of all parameters. Furthermore, the traditional method gradually struggles with complex optimization problems. Although some parameters may approach the measured values, the objective function prematurely falls into a local optimum when the error exceeds 5%, making further optimization impossible.

[0103] Step D3) verifies robustness. If the algorithm can still function normally even when measurement data is missing or incomplete, and can handle or compensate for missing data, the system is considered to have high robustness. This paper uses the leave-one-out method to cross-compare the robustness of different correction methods. It iterates through all cases where one measurement parameter is missing, keeping other algorithm conditions consistent, with the initial error at MD, and corrects three steady-state points. The parameters to be corrected are selected as n1, n2, and P. 25 For parameters P25, T3, and T5, one parameter is discarded sequentially, and the maximum error of the remaining parameters is corrected to within 2%. The discard amounts for each of the three cases are verified. Here, the NS equations are constructed using the two remaining parameters from P25, T3, and T5 to build a nonlinear equation system, and AF is solved. HPCeff and AF HPTeff The cross-validation results are as follows: Figure 10 As shown. The three steady-state data points selected represent the throttle pushing process, during which both pressure and temperature parameters continuously increase. The HBA method exhibits the worst robustness; under condition 1, although its correction results ensure that the maximum error during the rising phase of other parameters is less than 2%, P25 shows a decreasing trend, therefore its R... 2The value is less than 0, therefore the method fails when parameters are missing. Traditional PSO and NS-IHBA methods can accurately predict parameter trends even with missing parameters. However, although... Figure 10 This indicates that the PSO method has a shorter average time consumption than NS-IHBA in the case of MD, but its MAE and RSME for missing parameter prediction are both higher than NS-IHBA. Besides using reconstructed equations to establish a screening mechanism, NS-IHBA also guides the solution to converge in a reasonable direction in stages, making it more closely aligned with the physical object. Therefore, in terms of accuracy, its RMSE and MAE are an order of magnitude smaller than PSO's, and its RSME is also significantly lower. 2 Largest, highest accuracy

[0104] In summary, the present invention combines Figures 2 to 10 This paper systematically demonstrates a novel adaptive modeling method for individual aero-engine performance. This method possesses the following significant advantages: it balances global optimization and local selection, avoiding numerical divergence and local optima; it flexibly adapts to varying degrees of individual engine differences, enhancing universality; it effectively improves model reliability by combining physical constraints, dynamic convergence, and error compensation; and it achieves high-precision performance modeling consistent under both steady-state and dynamic conditions.

Claims

1. A performance adaptive modeling method for individual aero-engines, characterized in that, Includes the following steps: Step A), the data preprocessing step, includes denoising the flight or test measurement data collected by the engine, identifying and screening steady-state points, and extracting key steady-state measurement points for performance modeling; Step B), the sensitivity analysis step of the correction factor, includes a comprehensive evaluation method based on matrix condition number and entropy weight TOPSIS, which quantifies the sensitivity of the characteristic correction factor of each rotating component, and classifies the correction factor into high-sensitivity and low-sensitivity categories according to the sensitivity ranking results. Step C), the performance adaptation modeling step, includes an optimization method consisting of a series combination of the Improved Honey Badger Algorithm (IHBA) and a nonlinear solver (NS). For low-sensitivity factors, a global optimization search is used; for high-sensitivity factors, an extended set of equations is constructed and solved locally by NS. Constraints are applied to and solutions to non-physical or divergent solutions, thereby achieving gradual correction of the component-level characteristics of the multi-rotating parts. Step D), the modeling error compensation step, further fine-tunes the engine bypass flow rate and combustion chamber heat release parameters based on the output of the corrected steady-state benchmark model, and constructs a Bezier curve fitting function of the bypass leakage coefficient (CCG) and heat release fluctuation coefficient (CCH) as a function of speed to compensate for individual modeling errors and improve the accuracy of the model in dynamic simulation.

2. The performance adaptive modeling method for individual aero-engines according to claim 1, characterized in that, The convergence factor in IHBA is automatically adjusted based on individual initial differences, including four levels of bias: Minor, Moderate, Major, and Extreme.

3. The performance adaptive modeling method for individual aero-engines according to claim 1, characterized in that, The NS module is used to eliminate unsolvable particles during the optimization iteration process and to solve for highly sensitive correction parameters.

4. The method according to claim 1, characterized in that, The model error compensation part uses two coefficients, CCG and CCH, to adjust the bypass flow and combustion chamber heat release, and uses the steady-state reference solution as the fitting basis.

5. A performance adaptive modeling method for individual aero-engines based on any one of claims 1-4, characterized in that, include: Data processing module; Sensitivity analysis module; Serial combination optimization module; Model error compensation module; and component-level mathematical model.

Images