Quantitative analysis method for performance dispersivity of mass-produced engine parts considering multi-source information constraint

Abstract

The invention discloses a batch-production engine component performance dispersibility quantitative analysis method considering multi-source information constraint, and relates to the field of aero-engine performance research, and the method specifically comprises the steps: S1, preliminary screening: preliminarily determining the combination and range of component performance correction factors X, and preliminarily selecting a test run data Y combination according to a sensitivity matrix; s2, identifying a single engine, and in the correction factor boundary, sequentially executing a single engine model identification process on a batch of engines by using the primarily selected test run data to obtain correction factor vectors Xi of the engines and collecting the correction factor vectors Xi; s3, statistical characteristic evaluation: evaluating distribution characteristics of the correction factor set, and calculating a mean value mu j and a standard deviation sigma j; s4, judging the reasonability of the mean value; s5, judging the reasonability of the standard deviation; when the mean value and the standard deviation are both in a reasonable range, it is judged that the multi-source information constraint is met, and a correction factor set obtained at the moment is a final performance dispersity identification result.

Description

Technical Field

[0001] The present invention relates to the field of aerospace engine component performance research, and in particular to a quantitative analysis method for the dispersion of performance of mass-produced engine components taking into account multi-source information constraints. Background Art

[0002] Traditionally, quantitative research on aircraft engine performance deviations has relied primarily on zero-dimensional component-level simulation models. By introducing "correction factors" into the mathematical model or employing "parametric regression" methods to adjust component characteristic maps, the simulation results are aligned with test measurement data. Existing methods can be broadly divided into two categories, depending on how the component characteristic maps are corrected: identification methods based on scaling factors and those based on regression analysis. The former, represented by the Kurzke and LiYi-Guang groups, rapidly adjusts characteristic maps through global scaling and translation. This approach is simple to implement and has wide application in engineering, but its accuracy is limited when dealing with local nonlinearities or highly nonstandard operating conditions. The latter, represented by Kong, Tsoutsanis, and Kim, et al., utilize polynomial or nonlinear functions to reconstruct characteristic curves, which can better approximate local variations, but increases computational complexity and data requirements.

[0003] While these methods have made significant progress in theoretical research and single-engine performance simulation, they still face numerous challenges in mass-produced engine applications. The measurement data available for each mass-produced engine is typically limited, with a small measurement cross-section and complex measurement errors. This makes it difficult for traditional model identification methods to effectively filter out measurement errors and accurately identify component performance deviations for each engine based solely on data from a single engine. Furthermore, most existing studies tend to attribute overall engine performance dispersion entirely to performance deviations of rotating components, ignoring the influence of other key factors such as the air system. This can lead to distorted identified performance dispersion of rotating components.

[0004] Therefore, it is necessary to provide a quantitative analysis method for the performance dispersion of mass-produced engine components that considers multi-source information constraints to solve the above problems. Summary of the Invention

[0005] In order to solve the above problems, the present invention provides a quantitative analysis method for the performance dispersion of mass-produced engine components considering multi-source information constraints. From three aspects, namely the condition number of the sensitivity matrix, the rationality constraints of the mean of multiple units, and the dispersion constraints of the identification results of multiple units considering geometric information, the test parameters and correction factors screening method are described. The performance dispersion identification of 100 turbofan engines is completed, and the prediction results are verified using test data.

[0006] To achieve the above objectives, the present invention provides a method for quantitatively analyzing the dispersion of performance of mass-produced engine components taking into account multi-source information constraints, which specifically includes the following steps:

[0007] Step S1: Preliminary screening to determine the combination of correction factors X and the initial boundary range, and test the measurement combination of data Y according to the sensitivity matrix condition number;

[0008] Step S2: Single engine identification: within the given correction factor boundary, use the engine test data to perform a single engine model identification process to obtain the engine correction factor vector X i ;

[0009] Step S3: Calculate the statistical characteristics and calculate the mean μ of all correction factors j and standard deviation σ j Statistics, to assess the distribution of deviations;

[0010] Step S4: Determine the rationality of the mean. If the mean of the correction factor exceeds the reasonable range given by the engineering priori, delete or replace the measurement parameter with a large systematic error associated with it, and return to steps S2 and S3 for identification and statistics.

[0011] Step S5: Determine the rationality of the standard deviation. If the standard deviation of the correction factor significantly exceeds the geometric model prediction interval or the empirically acceptable range, add new correction factor parameters to express component differences; at the same time, adjust the boundaries of each correction factor, and then return to steps S2-S4 for iterative solution. When the mean and standard deviation are both within a reasonable range, it is determined that the multi-source information constraint is met. The correction factor set obtained at this time is the final performance dispersion identification result.

[0012] Preferably, in step S1, an aircraft engine performance model is established to transfer the influence relationship between the performance correction factor of the component and the performance parameters of the entire engine;

[0013] The connection is established by introducing the correction factor X that characterizes the component performance deviation and the state parameter vector C:

[0014] Y=G(X,C)#(1)

[0015] Where G is a nonlinear group function, Y is the test data under different working conditions, Y = [F n ,Wft,Wa0,T1,P1,…,T i ,P i ] represents the performance parameter vector of the whole machine, X=[x fan,m ,x fan,η ,x hpc,m ,x hpc,η ,x hpt,m ,x hpt,η ,x lpt,m ,x lpt,m ,x lpt,η ,…,x c,hpc] represents the flow and efficiency correction factor vectors of each rotating component and the correction quantities of the characteristic parameters of the remaining components. The state parameter vector C includes the engine inlet parameters total pressure P1, total temperature T1 and engine control parameters. The engine control parameters include the low-pressure rotor relative physical speed N1, the high-pressure rotor relative physical speed N2, the low-pressure turbine outlet temperature T6 and the turbine total pressure drop ratio π t ;

[0016] Make a small perturbation ΔX to the correction factor X in formula (1) and observe the change ΔY of the cross-sectional parameters to obtain the local linear approximation:

[0017]

[0018] Among them, S(X,C) is the sensitivity matrix, and the element S ij Indicates the output y i Correction factor x j Partial derivatives of :

[0019]

[0020] Numerical simulation is performed by finite difference method, and the thrust F is represented by formula (4) n Partial derivative with respect to the fan capacity correction factor:

[0021]

[0022] If the condition number of the matrix cond(S T If S) is too large, use singular value decomposition (SVD):

[0023] S T S=U∑V T #(5)

[0024] where ∑=diag(σ1,σ2,…,σ n ) is a singular value diagonal matrix, σ1≥σ2≥…≥σ n >0, if σ n <<σ1,cond(S T S)=σ1 / σ n The value of indicates that it is very sensitive to the perturbation of certain factors or there is extremely strong coupling;

[0025] To ensure the smooth process of model identification, a subset is first selected from a batch of measurement sections for a specific working condition C0 to construct a reasonable objective function OF;

[0026]

[0027] Among them, a i Represents the weight of the measurement parameter error, y iis the performance parameter of the whole machine, n is the total number of measurement parameters used to construct the objective function, the subscript pre is the model prediction value, and test is the model test value;

[0028] The screening process of the condition number is as follows:

[0029] Step S11: Based on all available engine test parameters Y and correction factor parameters X, construct the initial sensitivity matrix s0, calculate the condition number κ0, and record the measured parameters as the "current best combination" Y best ;

[0030] Step S12: Start from Y best Remove a measurement parameter y i , get a new test parameter set and sub-sensitivity matrix Calculate its condition number Find the combination corresponding to the minimum condition number to become the new Y best ;

[0031] Step S13: Repeat step S12, if Y is found best The condition number has fallen below the specified threshold k max , stop outputting the Y best .

[0032] Preferably, in step S2, the single engine model identification process is as follows:

[0033] Step S21: Initialize the engine performance model Y=G(X,C), determine the correction factor parameter combination X, and the whole engine test parameter combination Y;

[0034] Step S22: Generate an initial correction factor group by particle swarm algorithm, for a given correction factor X i , call the simulation model G to solve the model performance calculation value Y under the boundary condition C corresponding to the actual working point i,pre ;

[0035] Step S23: Y i,pre and measured data Y test Calculate the prediction error err_y i And the objective function OF i ;

[0036] The weight is given according to the measurement standard deviation of the test parameters. The larger the standard deviation, the smaller the weight. At the same time, constraints are added to ensure the calculation error of the measurement parameters.

[0037]

[0038] Step S24: MaxErr_y iConstraints and OF i Update the new generation of correction factor group, repeat step S22-step S23 until the maximum evolution generation is reached. If err_y i If the error constraints are satisfied, the solution is considered to have converged and the optimal correction factor is output.

[0039] Preferably, in step S3, for multiple engines of the same model and the same batch, the correction factor vector X obtained after completing the model identification of each engine is i (i=1,…,N) for statistical analysis, where X i =[x i,1 ,x i,2 ,…,x i,d ] represents the maintenance positive factor vector of the ith engine;

[0040] On this basis, for a specific correction factor x j Calculate the mean and standard deviation on N engines:

[0041]

[0042] Where i, j = 1, 2, ..., d, d represents the dimensional search space, through the above mean μ j and standard deviation σ j , quantify the dispersion and central tendency of the correction factors on multiple engines.

[0043] Preferably, in step S4, the following steps are specifically included:

[0044] S41: Based on the engineering prior knowledge, the correction factor average value constraint range is preset, and the single engine model identification is completed in sequence according to the initial measurement parameter set, and the correction factor average of the batch production engine is calculated. If the mean value of the correction factor is within the preset correction factor range, no screening is required, and Y best =Y0; if the correction factor mean is not within the preset range, execute step S42;

[0045] S42: Record the parameter combination I with the correction factor mean being too large, and best Select the measurement parameter y j , set the y of all engines j The parameters are all in accordance with y j ×(1+δy j ) method, and the corrected test parameters are used to re-identify the engine model and calculate the mean value of the new correction factor Calculate the correction factor for x in set I i The relative change δx i , if satisfied c is the set threshold, indicating the measurement parameter y j The systematic error of has a great influence on the identification result of the mean value of the correction factor with excessive deviation. best Remove; otherwise select a new y j Repeat this step;

[0046] S43: According to the filtered measurement parameter set Y best Complete the model identification of each engine in turn and calculate the average correction factor of the batch production engine If the mean value of the correction factor is within the preset correction factor range, it is considered that Y best is the final screening result; otherwise, return to step S42 to continue screening the test parameters.

[0047] Preferably, in step S5, the following steps are specifically included:

[0048] S51: Bring the collected batch engine key geometry data into the geometry-component association model to determine the prior standard deviation vector σ of the correction factor geo ;

[0049] S52: Utilize the optimal measurement parameter set Y according to the initial correction factor set X0 and the boundary best Complete the model identification of multiple batch-produced engines and calculate the standard deviation vector σ0 of the correction factor. If each dimension of the standard deviation vector σ0 of the correction factor is not greater than the prior standard deviation vector σ geo The corresponding dimension is twice that of the original dimension, so the correction factor set does not need to be changed, x best =Y0, otherwise go to step S53;

[0050] S53: Record the dimension i combination I where the correction factor is too large, select the engine whose correction factor is at the boundary of these dimensions, and select a new correction factor x from the set of candidate correction factors. new Join from X best In the process, the model identification of the selected engine is performed according to the new set of correction factors, and the correction factors {x i ,i∈I} is reasonable. If it is evaluated, it will be added to the best and follow ±6σ geo The principle is X best Redefine the boundaries; if the evaluation fails, select a new x new Repeat this step;

[0051] S54: Complete the model identification of multiple batch engines according to the optimized correction factor set X and the boundary, calculate the standard deviation vector σ of the correction factor, and if each dimension of σ is not greater than the prior standard deviation vector σ geo Corresponding to 2 times of the dimension, it is considered that Xbest is the final optimization result; otherwise, return to step S53 to continue optimization.

[0052] Therefore, the present invention adopts the above-mentioned quantitative analysis method of the performance dispersion of mass-produced engine components considering multi-source information constraints, which has the following beneficial effects:

[0053] (1) The multi-source information constraint method proposed in this invention integrates the performance data of the whole machine, the prediction results of the geometric component association model and the statistical characteristics of multiple units. The measurement parameter screening method based on the sensitivity matrix condition number can effectively improve the convergence of model identification, from the initial 2% convergence rate to 96%, laying the foundation for the performance dispersion identification of batch engines.

[0054] (2) The present invention introduces component performance rationality constraints and eliminates measurement parameters with systematic errors, making the model identification results more consistent with engineering experience and significantly reducing the mean deviation of the correction factor.

[0055] (3) The present invention combines the prior information provided by the geometry-component association model and introduces the uncertain factors of the air system (the amount of bleed air and leakage in the last stage of the compressor), which effectively avoids the misattribution of component performance dispersion and makes the component performance dispersion closer to the forward prediction result. For example, the standard deviation of the compressor flow capacity is reduced from 2.68% to 0.65%.

[0056] (4) The low-speed airflow method test of the high-pressure turbine guide vane of the present invention shows that the final model identification result is highly consistent with the measured value: the mean error of the high-pressure turbine flow capacity is only 0.35%, which is 7 times higher than the traditional method; the standard deviation is 8%, and the accuracy is increased by 26 times; and the predicted value and the experimental value both conform to the same distribution, which fully proves the accuracy and reliability of the proposed method.

[0057] The technical solution of the present invention is further described in detail below through the accompanying drawings and embodiments. BRIEF DESCRIPTION OF THE DRAWINGS

[0058] Figure 1 This is a dispersion identification flow chart of a method for quantitatively analyzing the dispersion of performance of mass-produced engine components taking into account multi-source information constraints of the present invention;

[0059] Figure 2 is a structural diagram of a turbofan engine in an embodiment of the present invention;

[0060] Figure 3 It is a zero-dimensional model component module of the steady-state performance of the mixed turbofan engine in an embodiment of the present invention;

[0061] Figure 4 This is a diagram showing the effect of correcting the characteristics of a compressor component in an embodiment of the present invention;

[0062] Figure 5 This is a basic flow chart of the single-unit model identification technology in an embodiment of the present invention;

[0063] Figure 6 This is a flow chart of condition number screening based on sensitivity matrix condition number in an embodiment of the present invention;

[0064] Figure 7 Schematic diagram of the components and cross-section of a dual-shaft hybrid turbofan engine in Example 1 of the present invention;

[0065] Figure 8 This is a comparison diagram of model errors of different calculation examples of Example 1 of the present invention;

[0066] Figure 9 This is a comparison diagram of component distribution obtained before and after parameter screening in Example 1 of the present invention;

[0067] Figure 10 A comparison chart of the identified value and the test value of the high-pressure turbine flow capacity in Example 1 of the present invention;

[0068] Figure 11 This is a mean rationality determination diagram in an embodiment of the present invention;

[0069] Figure 12 This is a standard deviation rationality determination diagram in an embodiment of the present invention. DETAILED DESCRIPTION

[0070] The technical solution of the present invention is further described below with reference to the accompanying drawings and embodiments.

[0071] Unless otherwise defined, technical or scientific terms used in the present invention shall have the same meaning as commonly understood by one of ordinary skill in the art to which the present invention belongs.

[0072] The words “include” or “comprising” and similar words used in the present invention mean that the elements before the word include the elements listed after the word, and do not exclude the possibility of also including other elements. The orientation or position relationship indicated by the terms “inside”, “outside”, “upper”, “lower”, etc. is based on the orientation or position relationship shown in the accompanying drawings. It is only for the convenience of describing the present invention and simplifying the description, and does not indicate or imply that the device or element referred to must have a specific orientation, be constructed and operated in a specific orientation. Therefore, it cannot be understood as a limitation of the present invention. When the absolute position of the described object changes, the relative position relationship may also change accordingly. In the present invention, unless otherwise clearly stipulated and limited, the terms such as “attachment” should be understood in a broad sense. For example, it can be a fixed connection, a detachable connection, or an integral whole; it can be directly connected or indirectly connected through an intermediate medium, and it can be the internal connection of two elements or the interaction relationship between two elements. For ordinary technicians in this field, the specific meanings of the above terms in the present invention can be understood according to the specific circumstances.

[0073] Example

[0074] like Figure 1 As shown, a quantitative analysis method for the performance dispersion of mass-produced engine components considering multi-source information constraints specifically includes the following steps:

[0075] Step S1: Preliminary screening to determine the combination of correction factors X and the initial boundary range, and test the measurement combination of data Y according to the sensitivity matrix condition number;

[0076] In step S1, the influence relationship between the performance correction factor of the transmission component and the performance parameters of the whole engine is established in the aircraft engine performance model;

[0077] The aircraft engine performance model is a zero-dimensional simulation model based on thermodynamic properties, component characteristics, and their interoperability. The model is composed of several component modules. For each module, the input parameters are the airflow parameters at the component's inlet. Aerothermodynamic calculations yield the airflow parameters at the component's outlet, including flow rate, pressure, and temperature. Information is transmitted between connected components via airflow parameters, with the outlet parameters of upstream components being transmitted to downstream components. Given engine operating conditions, calculations can be completed for each component from the inlet to the tailpipe, yielding aerothermodynamic parameters for key sections and overall engine performance.

[0078] Taking a hybrid turbofan engine as an example, the framework of the zero-dimensional simulation model is introduced. Its structural diagram is shown as follows: Figure 2 The main components include: air intake, fan, compressor, main combustion chamber, turbine, mixer, afterburner, and tail nozzle. The model can also consider the effects of air bleed and air leakage in the air system. The aerodynamic and mechanical connections of its main components can be abstracted as follows: Figure 3 As shown in the figure, arrows represent the aerodynamic connections between components, while dashed lines represent the mechanical connections. For two aerodynamically connected components, the total temperature, total pressure, and flow rate of the airflow at the upstream component outlet are directly related to the connected component. For mechanically connected components, their physical speed and power are directly related. Figure 3 Each box in the figure represents a component module. Given the airflow parameters at the inlet of the component module and the component performance parameters, the airflow parameters at the outlet of the component module can be obtained through aerodynamic and thermal calculations.

[0079] The connection is established by introducing the correction factor X that characterizes the component performance deviation and the state parameter vector C:

[0080] Y=G(X,C)#(1)

[0081] Where G is a nonlinear group function, Y is the test data under different working conditions, Y = [F n ,Wft,Wa0,T1,P1,…,T i ,P i ] represents the performance parameter vector of the whole machine, X=[x fan,m ,x fan,η ,x hpc,m ,x hpc,η ,x hpt,m ,x hpt,η ,x lpt,m ,x lpt,m ,x lpt,η ,…,x c,hpc ] represents the flow and efficiency correction factor vectors of each rotating component and the correction quantities of the characteristic parameters of the remaining components. The state parameter vector C includes the engine inlet parameters total pressure P1, total temperature T1 and engine control parameters. The engine control parameters include the low-pressure rotor relative physical speed N1, the high-pressure rotor relative physical speed N2, the low-pressure turbine outlet temperature T6 and the turbine total pressure drop ratio π t ;

[0082] The performance parameters of the entire machine contain information for identifying component performance. However, due to noise and numerical rigidity issues in the solution system, the more measurement parameters the better. Parameters need to be further screened based on information such as convergence and statistical characteristics of multiple units.

[0083] Make a small perturbation ΔX to the correction factor X in formula (1) and observe the change ΔY of the cross-sectional parameters to obtain the local linear approximation:

[0084]

[0085] Among them, S(X,C) is the sensitivity matrix, and the element S ij Indicates the output y i Correction factor xj Partial derivatives of :

[0086]

[0087] Numerical simulation is performed by finite difference method, and the thrust F is represented by formula (4) n Partial derivative with respect to the fan capacity correction factor:

[0088]

[0089] If the condition number of the matrix cond(S T If S) is too large, use singular value decomposition (SVD):

[0090] S T S=U∑V T #(5)

[0091] where ∑=diag(σ1,σ2,…,σ n ) is a singular value diagonal matrix, σ1≥σ2≥…≥σ n >0, if σ n <<σ1,cond(S T S)=σ1 / σ n The value of indicates that it is very sensitive to the perturbation of certain factors or there is extremely strong coupling;

[0092] To ensure the smooth process of model identification, a subset is first selected from a batch of measurement sections for a specific working condition C0, and a reasonable objective function OF is constructed. The weighted root mean square error between the engine model output and the actual engine test data is used as the objective function. The formula is:

[0093]

[0094] Among them, a i Represents the weight of the measurement parameter error, y i is the performance parameter of the whole machine, n is the total number of measurement parameters used to construct the objective function, the subscript pre is the model prediction value, and test is the model test value;

[0095] like Figure 6 As shown in Figure 2, the screening process of the condition number is as follows:

[0096] Step S11: Based on all available engine test parameters Y and correction factor parameters X, construct the initial sensitivity matrix S0, calculate the condition number κ0, and record the measured parameters as the "current best combination" Y best ;

[0097] Step S12: Start from Y best Remove a measurement parameter y i, get a new test parameter set and sub-sensitivity matrix Calculate its condition number Find the combination corresponding to the minimum condition number to become the new Y best ;

[0098] Step S13: Repeat step S12, if Y is found best The condition number is lower than the specified threshold κ max , stop outputting the Y best .

[0099] The parameters and correction methods of the characteristic diagrams are different for different types of components:

[0100] Flow and efficiency correction of compressors and fans: For compression components such as compressors and fans, their characteristic diagrams are constrained by two clusters of curves: flow-efficiency curves and flow-pressure ratio curves, expressed as:

[0101]

[0102] Where: m is the converted flow rate, η is the efficiency, and α is the guide vane angle. cr is the relative conversion speed, π is the pressure ratio. Introduce the correction factor x m and x η Finally, the correction factor is defined as the relative percentage deviation from the standard characteristic, so the corrected flow rate and efficiency are:

[0103]

[0104] For expansion components such as high-pressure turbines and low-pressure turbines, their characteristic diagrams are constrained by two clusters of curves: flow-efficiency curves and flow-pressure ratio curves, which can be expressed as:

[0105]

[0106] The corrected flow rate and efficiency are:

[0107]

[0108] Taking the characteristics of the compressor as an example, take x m =-2%, x η = -2%, showing the corrected component characteristic changes such as Figure 4 shown. Figure 4 In the figure, the solid line represents the design characteristics of the component, and the dotted line represents the actual characteristics of the component after correction.

[0109] Step S2: Single engine identification: within the given correction factor boundary, use the engine test data to perform a single engine model identification process to obtain the engine correction factor vector X i ;

[0110] In step S2, if Figure 5 As shown in the figure, the single engine model identification process is as follows:

[0111] Step S21: Initialize the engine performance model Y=G(X,C), determine the correction factor parameter combination X, and the whole engine test parameter combination Y;

[0112] Step S22: Generate an initial correction factor group by particle swarm algorithm, for a given correction factor X i , call the simulation model G to solve the model performance calculation value Y under the boundary condition C corresponding to the actual working point i,pre ;

[0113] Step S23: Y i,pre and measured data Y test Calculate the prediction error err_y i And the objective function OF i ;

[0114] The weight is given according to the measurement standard deviation of the test parameters. The larger the standard deviation, the smaller the weight. At the same time, constraints are added to ensure the calculation error of the measurement parameters.

[0115]

[0116] Step S24: MaxErr_y i Constraints and OF i Update the new generation of correction factor group, repeat step S22-step S23 until the maximum evolution generation is reached. If err_y i If the error constraints are satisfied, the solution is considered to have converged and the optimal correction factor is output.

[0117] Step S3: Calculate the statistical characteristics and calculate the mean μ of all correction factors j and standard deviation σ j Statistics, to assess the distribution of deviations;

[0118] In step S3, for multiple engines of the same model and the same batch, the correction factor vector X obtained after the model identification of each engine is completed is converted into i (i=1,…,N) for statistical analysis, where X i =[x i,1 ,x i,2 ,…,x i,d ] represents the maintenance positive factor vector of the ith engine;

[0119] On this basis, for a specific correction factor x j Calculate the mean and standard deviation on N engines:

[0120]

[0121] Where i, j = 1, 2, ..., d, d represents the dimensional search space, through the above mean μ j and standard deviation σ j , quantifying the dispersion and central tendency of the correction factors on multiple engines, laying the foundation for subsequent constraint analysis based on group statistical characteristics.

[0122] Step S4: Determine the rationality of the mean. If the mean of the correction factor exceeds the reasonable range given by the engineering priori, delete or replace the measurement parameter with a large systematic error associated with it, and return to steps S2 and S3 for identification and statistics.

[0123] like Figure 11 As shown, in step S4, the following steps are specifically included:

[0124] S41: Based on the engineering prior knowledge, the correction factor average value constraint range is preset, and the single engine model identification is completed in sequence according to the initial measurement parameter set, and the correction factor average of the batch production engine is calculated. If the mean value of the correction factor is within the preset correction factor range, no screening is required, and Y best =Y0; if the correction factor mean is not within the preset range, execute step S42;

[0125] S42: Record the parameter combination I with the correction factor mean being too large, and best Select the measurement parameter y j , set the y of all engines j The parameters are all in accordance with y j ×(1+δy j ) method, and the corrected test parameters are used to re-identify the engine model and calculate the mean value of the new correction factor Calculate the correction factor for x in set I i The relative change δx i , if satisfied c is the set threshold, indicating the measurement parameter y j The systematic error of has a great influence on the identification result of the mean value of the correction factor with excessive deviation. best Remove; otherwise select a new y j Repeat this step;

[0126] S43: According to the filtered measurement parameter set Y best Complete the model identification of each engine in turn and calculate the average correction factor of the batch production engine If the mean value of the correction factor is within the preset correction factor range, it is considered that Ybest is the final screening result; otherwise, return to step S42 to continue screening the test parameters.

[0127] Step S5: Determine the rationality of the standard deviation. If the standard deviation of the correction factor significantly exceeds the geometric model prediction interval or the empirically acceptable range, add new correction factor parameters to express component differences; at the same time, adjust the boundaries of each correction factor, and then return to steps S2-S4 for iterative solution. When the mean and standard deviation are both within a reasonable range, it is determined that the multi-source information constraint is met. The correction factor set obtained at this time is the final performance dispersion identification result.

[0128] like Figure 12 As shown, in step S5, the following steps are specifically included:

[0129] S51: Bring the collected batch engine key geometry data into the geometry-component association model to determine the prior standard deviation vector σ of the correction factor geo ;

[0130] S52: Utilize the optimal measurement parameter set Y according to the initial correction factor set X0 and the boundary best Complete the model identification of multiple batch-produced engines and calculate the standard deviation vector σ0 of the correction factor. If each dimension of the standard deviation vector σ0 of the correction factor is not greater than the prior standard deviation vector σ geo The correction factor set does not need to change if the corresponding dimension is doubled. best =Y0, otherwise go to step S53;

[0131] S53: Record the dimension i combination I where the correction factor is too large, select the engine whose correction factor is at the boundary of these dimensions, and select a new correction factor x from the set of candidate correction factors. new Join from X best In the process, the model identification of the selected engine is performed according to the new set of correction factors, and the correction factors {x i ,i∈I} is reasonable. If it is evaluated, it will be added to the best and follow ±6σ geo The principle is X best Redefine the boundaries; if the evaluation fails, select a new x new Repeat this step;

[0132] S54: Complete the model identification of multiple batch engines according to the optimized correction factor set X and the boundary, calculate the standard deviation vector σ of the correction factor, and if each dimension of σ is not greater than the prior standard deviation vector σ geo Corresponding to 2 times of the dimension, it is considered that X best is the final optimization result; otherwise, return to step S53 to continue optimization.

[0133] Through the above method, while ensuring the accuracy of single-unit identification, multi-source information such as multiple units and geometric priors can be used to constrain the group correction factor results, reduce the impact of system errors and measurement noise, and thus improve the robustness and credibility of model identification in engineering applications.

[0134] Example 1

[0135] Taking a certain type of mass-produced dual-shaft mixed-displacement turbofan engine as the object, Figure 7 The main components and cross-sectional diagram of the engine are displayed.

[0136] Data source: Engine performance test data was collected from a standard engine test bench at a certain factory. The measurement parameters are shown in Table 1. The measurement state was the engine's intermediate state. A total of 100 sets of data were collected, and all data have been normalized.

[0137] Table 1

[0138]

[0139] The geometric data comes from the blade profile and clearance data of the rotating components of this batch of engines. It is collected through three-dimensional coordinate measurement combined with blue light scanning.

[0140] At the same time, 35 engines from this batch were sampled and low-speed airflow method tests were conducted using the AF-36 equipment from Fleming & Associates to obtain the deviation distribution of the high-pressure turbine flow capacity.

[0141] The combination of initial test parameters and correction factor parameters, for this type of engine model Y = G (X, C), this embodiment selects the flow capacity and efficiency correction factors of the four rotating parts as the initial correction variables X = [x fan,m ,x fan,η ,…,x lpt,η Table 2 lists the corresponding modified variables and initial bounds. To ensure the convergence of the calculation, the initial bounds are set larger.

[0142] Table 2

[0143]

[0144] The high and low pressure rotor speed, inlet temperature and pressure, and guide vane angle are taken as state variables C = [N1, N2, T1, P1, α1, α2]. The rest of the test data y i All of them are selected as variables for constructing the objective function OF, and the weight coefficient a i and the maximum error constraint MaxErr_y i As shown in Table 3;

[0145] Table 3

[0146]

[0147]

[0148] Based on the test data screening method for model convergence, the condition number of the initial sensitivity matrix S0 formed by the correction factors and the measured parameters was calculated to be as high as 1145536336.95, indicating that the correction factors have a strong correlation with the influence vectors of some measured parameters, which can easily lead to non-convergence of the solution. Following the process in steps S11-S13, one measurement variable was deleted in turn and the condition number was recalculated. The results are shown in Table 4:

[0149] Table 4

[0150]

[0151] It can be seen from the data in the table that there is a strong correlation between thrust and inlet flow, exhaust temperature, and exhaust pressure. The existence of random errors will make it difficult to converge the solution. Considering that the thrust error is relatively large during bench testing, thrust is discarded first.

[0152] After completing the model identification for this batch of engines with and without considering thrust, the convergence rate of the solution is greatly improved after removing the thrust, as shown in Table 5.

[0153] Table 5

[0154]

[0155] The calculation example without considering thrust is positioned as Example 1. The correction factors are introduced to recalculate the cross-sectional parameters of the entire batch of engines. The average error of the predicted values of each cross-sectional parameter relative to the measured values is less than 2%, as shown in Table 6. The correction factors of each engine are calculated, and their statistical characteristics are shown in Table 7:

[0156] Table 6

[0157]

[0158] Table 7

[0159]

[0160] Considering the screening of measurement parameters with the rationality constraint of the average deviation of multiple units, the average performance of the correction factor obtained in Example 1 shows that the flow capacity of the intermediate compressor and low-pressure turbine is 9.26% and 8.16% less than the design state respectively. The large deviation does not conform to the engineering experience of the factory performance of new machines. This is probably caused by the systematic error of some measurement parameters used to construct the objective function. According to engineering experience, the radial flow field after the fan is highly non-uniform, and the measured P 23 and T 25It is difficult to accurately represent the average parameters of this cross section.

[0161] By making 100 engines P 23 and T 25 A new example was constructed by varying all parameters by ±1%. The results, shown in Tables 8 and 9, indicate that these two parameters significantly influence the average identification results of the compressor and low-pressure turbine flow capacities for multiple engines, but have little effect on the dispersion of all parameters. If it is impossible to accurately quantify or eliminate their systematic errors, these two parameters can be removed from the objective function.

[0162] Table 8

[0163]

[0164] Table 9

[0165]

[0166] According to the above research, the two parameters P used to construct the model identification objective function are eliminated on the basis of Example 1. 23 and T 25 , and construct Example 2 with all other conditions unchanged. After the calculation, the average error results of the measured cross section are shown in Table 10, except for the eliminated parameter F n 、P 23 and T 25 In addition, the errors of other parameters are reduced from about 1% to 0.33%, indicating that the model identification convergence is higher.

[0167] Table 10

[0168]

[0169] Table 11

[0170]

[0171] According to the geometric data provided by the measurement parameter screening method considering the rationality constraint of the multi-unit mean in step S3, the performance distribution characteristics of the component can be calculated from the front by using the 'geometry-component' association model as shown in Table 12;

[0172] Table 12

[0173]

[0174] Without considering the difference in the corrected mean values, the dispersion of component performance inferred from the whole-machine data is much greater than that predicted by the geometry-component correlation model, especially for the prediction of component flow capacity. This indicates that ignoring non-deterministic factors other than component performance leads to excessive dispersion (standard deviation) in the identification results for this batch of engines, necessitating the addition of new correction factors to improve the accuracy of the standard deviation. The identification results for engines A and B in Example 2 are shown in Table 13.

[0175] Table 13

[0176]

[0177] Considering that the air system itself is non-deterministic and has a relatively large impact on engine performance, among which the compressor last stage bleed air accounts for a large proportion and has the greatest impact on performance, the compressor last stage bleed air correction factor x is increased. cool and the correction factor x for the leakage from the last stage of the compressor to the outside bleed To verify this hypothesis, two engines with compressor flow capacity near the limit were selected. The final stage cold air bleed volume was increased by 5.6% or the final stage bypass air leakage volume was increased by 2%. Model identification was then re-performed. The correction results are shown in Tables 14 and 15. The results show that changes in cold air and leakage volume have a significant impact on the identification results of the eight rotating component correction factors and can be introduced as new correction factors.

[0178] Table 14

[0179]

[0180] Table 15

[0181]

[0182] Finally, we take the component performance standard deviation σ predicted by the geometry-component correlation model and the component performance mean μ predicted in Section 3.33, and calculate x according to the formula i,bound =μ±6σ to calculate the upper and lower boundaries of each variable of the correction factor.

[0183] The compressor's final stage cooling air ratio and leakage to the external duct were added as new uncertainties to the model identification. The cooling air ratio range was ±3%, and the leakage range was 0–3%. The final correction factors and their bounds are shown in Table 16. The test data parameters and maximum constraints used to construct the objective function are shown in Table 17.

[0184] Table 16

[0185]

[0186] Table 17

[0187]

[0188]

[0189] Based on the correction factor combinations and test data combination reference table provided in the batch engine performance dispersion identification process from steps S1 to S5, Example 3 was constructed to complete component performance identification for 100 engines. After the calculations were completed, the average error of each cross-sectional parameter used to construct the target parameters relative to the measured values was approximately 0.3%, as shown in Table 18. The model converged well, and the maximum error of the cross-sectional parameters not used to construct the target function did not exceed 5%, meeting the requirements for engineering applications.

[0190] Table 18

[0191]

[0192] Table 19

[0193]

[0194] Table 19 shows the identification results for the final example. The dispersion of the correction factors for rotating component characteristics is significantly reduced, approximately 1.5 times that of the geometry-component forward prediction results shown in Table 12. Compared to Example 2, the mean is virtually unchanged. Changes in cooling air volume are primarily concentrated within ±1%, with a mean of -0.23%. Leakage is concentrated within 1%, with a mean of 0.68%.

[0195] Combined with Table 6, Table 10, and Table 18, the average error results of the cross-sectional parameters of Example 1, Example 2, and Example 3 are shown as follows: Figure 8 After the model identification is completed for the three examples, the solution errors are effectively reduced, proving the effectiveness of the model identification method itself. The parameters P obtained by the screening method of measurement parameters 31 , P6, T6, Wft, and Wa1 have much smaller errors than those of the unselected parameters, indicating that the system tends to adopt the information of these parameters in the solution process, which meets the purpose of parameter screening.

[0196] The distribution diagrams of the eight correction factors of rotating parts obtained from the three examples are shown in the figure below: Figure 9 As more factors are considered, the mean of the component performance correction factor obtained becomes more reasonable, and the standard deviation is close to the result of the forward prediction of the geometry-component correlation model, which proves the effectiveness of the performance dispersion identification method for mass-produced engines considering multi-source information constraints.

[0197] In order to verify the accuracy of the calculation results of the final example, the parameters obtained from the three examples are compared with the high-pressure turbine flow capacity distribution results obtained by low-speed airflow test measurement of 35 engines sampled from this batch of engines, such as Figure 10 shown.

[0198] Based on the definition of the correction factor, the error of the mean and standard deviation of the example relative to the test value is calculated according to the formula. Where μ represents the mean, σ represents the standard deviation, the subscript pre represents the predicted value of the example, and test represents the test value. The calculation results are shown in Table 20:

[0199]

[0200] In terms of mean, the predicted values in Example 3 differed only by 0.34% from the experimental values, a sevenfold improvement compared to Example 1, which represents a traditional method. Although the mean error in Example 2 was smaller, it resulted in a significant miscalculation of the dispersion. This was due to the neglect of the air system's impact on performance. During the reverse identification of the model, the dispersion of the air system was attributed to the rotating components, resulting in a much greater miscalculation of the component system dispersion than was actually expected. Regarding dispersion, the standard deviation of the predicted values in the final example differed by only 8%, representing a 26-fold improvement in accuracy compared to the traditional method. A Kolmogorov-Smirnov normality test was performed on both the final example and the experimental values. The null hypothesis was not rejected at a significance level of α = 0.05, indicating that both conformed to a normal distribution. In summary, the results of the final example closely matched the experimental values, demonstrating a significant improvement in accuracy, demonstrating the accuracy of this identification method.

[0201] Table 20

[0202]

[0203] Therefore, the present invention adopts the above-mentioned quantitative analysis method of the performance dispersion of mass-produced engine components considering multi-source information constraints. The proposed multi-source information constraint method integrates the whole machine performance data, the prediction results of the geometric component association model and the statistical characteristics of multiple units, thereby overcoming the defects of relying solely on the whole machine test data in the performance identification of mass-produced engine components, which often leads to unstable solutions due to insufficient information and difficult evaluation of component performance dispersion.

[0204] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention rather than to limit the same. Although the present invention has been described in detail with reference to the preferred embodiments, those skilled in the art should understand that they can still modify or replace the technical solutions of the present invention with equivalents, and these modifications or equivalent replacements cannot cause the modified technical solutions to deviate from the spirit and scope of the technical solutions of the present invention.

Claims

1. A quantitative analysis method for the performance dispersion of mass-produced engine components considering multi-source information constraints, characterized by: The specific steps include: Step S1: Preliminary screening to determine the combination of correction factors X and the initial boundary range, and test the measurement combination of data Y according to the sensitivity matrix condition number; Step S2: Single engine identification: within the given correction factor boundary, use the engine test data to perform a single engine model identification process to obtain the engine correction factor vector X i ; Step S3: Calculate the statistical characteristics and calculate the mean μ of all engine correction factors j and standard deviation σ j Statistics, to assess the distribution of deviations; Step S4: Determine the rationality of the mean. If the mean of the correction factor exceeds the reasonable range given by the engineering priori, delete or replace the measurement parameter with a large systematic error associated with it, and return to steps S2 and S3 for identification and statistics. Step S5: Determine the rationality of the standard deviation. If the standard deviation of the correction factor significantly exceeds the geometric model prediction interval or the empirically acceptable range, add new correction factor parameters to express component differences; at the same time, adjust the boundaries of each correction factor, and then return to steps S2-S4 for iterative solution. When the mean and standard deviation are both within a reasonable range, it is determined that the multi-source information constraint is met. The correction factor set obtained at this time is the final performance dispersion identification result.

2. A quantitative analysis method for the performance dispersion of mass-produced engine components considering multi-source information constraints according to claim 1, characterized in that: In step S1, the influence relationship between the performance correction factor of the transmission component and the performance parameters of the whole engine is established in the aircraft engine performance model; The connection is established by introducing the correction factor X that characterizes the component performance deviation and the state parameter vector C: Y=G(X,C)#(1) Where G is a nonlinear group function, Y is the test data under different working conditions, Y = [F n ,Wft,Wa0,T1,P1,…,T i ,P i ] represents the performance parameter vector of the whole machine, X=[x fan,m ,x fan,η ,x hpc,m ,x hpc,η ,x hpt,m ,x hpt,η ,x lpt,m ,x lpt,m ,x lpt,η ,…,x c,hpc ] represents the flow and efficiency correction factor vectors of each rotating component and the correction quantities of the characteristic parameters of the remaining components. The state parameter vector C includes the engine inlet parameters total pressure [1, total temperature T1 and engine control parameters. The engine control parameters include the low-pressure rotor relative physical speed N1, the high-pressure rotor relative physical speed N2, the low-pressure turbine outlet temperature T6 and the turbine total pressure drop ratio π t ; Make a small perturbation ΔX to the correction factor X in formula (1) and observe the change ΔY of the cross-sectional parameters to obtain the local linear approximation: Among them, S(X,C) is the sensitivity matrix, and the element S ij Indicates the output y i Correction factor x j Partial derivatives of : Numerical simulation is performed by finite difference method, and the thrust F is represented by formula (4) n Partial derivative with respect to the fan capacity correction factor: If the condition number of the matrix cond(S T If S) is too large, use singular value decomposition (SVD): S T S=U∑V T #(5) where ∑=diag(σ1,σ2,…,σ n ) is a singular value diagonal matrix, σ1≥σ2≥…≥σ n >0, if σ n <<σ1,cond(S T S)=σ1 / σ n The value of indicates that it is very sensitive to the perturbation of certain factors or there is extremely strong coupling; To ensure the smooth process of model identification, a subset is first selected from a batch of measured sections for a specific working condition C0 to construct a reasonable objective function OF; Among them, a i Represents the weight of the measurement parameter error, y i is the performance parameter of the whole machine, n is the total number of measurement parameters used to construct the objective function, the subscript pre is the model prediction value, and test is the model test value; The screening process of the condition number is as follows: Step S11: Based on all available engine test parameters Y and correction factor parameters X, construct the initial sensitivity matrix S0, calculate the condition number κ0, and record the measured parameters as the "current best combination" Y best ; Step S12: Start from Y best Remove a measurement parameter y i , get a new test parameter set and sub-sensitivity matrix Calculate its condition number Find the combination corresponding to the minimum condition number to become the new Y best ; Step S13: Repeat step S12, if Y is found best The condition number has fallen below the specified threshold k max , stop outputting the Y best .

3. The method for quantitatively analyzing the dispersion of performance of mass-produced engine components considering multi-source information constraints according to claim 2, characterized in that: In step S2, the single engine model identification process is as follows: Step S21: Initialize the engine performance model Y=G(X,C), determine the correction factor parameter combination X, and the whole engine test parameter combination Y; Step S22: Generate an initial correction factor group by particle swarm algorithm, for a given correction factor X i , call the simulation model G to solve the model performance calculation value Y under the boundary condition C corresponding to the actual working point i,pre ; Step S23: Y i,pre and measured data Y test Calculate the prediction error err_y i And the objective function OF i ; The weight is given according to the measurement standard deviation of the test parameters. The larger the standard deviation, the smaller the weight. At the same time, constraints are added to ensure the calculation error of the measurement parameters. Step S24: MaxErr_y i Constraints and OF i Update the new generation of correction factor group, repeat step S22-step S23 until the maximum evolution generation is reached. If err_y i If the error constraints are satisfied, the solution is considered to have converged and the optimal correction factor is output.

4. The method for quantitatively analyzing the dispersion of performance of mass-produced engine components considering multi-source information constraints according to claim 3, characterized in that: In step S3, for multiple engines of the same model and the same batch, the correction factor vector X obtained after the model identification of each engine is completed is converted into i (i=1,…,N) for statistical analysis, where X i =[x i,1 ,x i,2 ,…,x i,d ] represents the maintenance positive factor vector of the ith engine; On this basis, for a specific correction factor x j Calculate the mean and standard deviation on N engines: Where i, j = 1, 2, ..., d, d represents the dimensional search space, through the above mean μ j and standard deviation σ j , quantify the dispersion and central tendency of the correction factors on multiple engines.

5. The method for quantitatively analyzing the dispersion of performance of mass-produced engine components considering multi-source information constraints according to claim 4, characterized in that: In step S4, the following steps are specifically included: S41: Based on the engineering prior knowledge, the correction factor average value constraint range is preset, and the single engine model identification is completed in sequence according to the initial measurement parameter set, and the correction factor average of the batch production engine is calculated. If the mean value of the correction factor is within the preset correction factor range, no screening is required, and Y best =Y0; if the correction factor mean is not within the preset range, execute step S42; S42: Record the parameter combination I with the correction factor mean being too large, and best Select the measurement parameter y j , set the y of all engines j The parameters are all in accordance with y j ×(1+δy j ) method, and the corrected test parameters are used to re-identify the engine model and calculate the mean value of the new correction factor Calculate the correction factor for x in set I i The relative change δx i , if satisfied c is the set threshold, indicating the measurement parameter y j The systematic error of has a great influence on the identification result of the mean value of the correction factor with excessive deviation. best Remove; otherwise select a new y j Repeat this step; S43: According to the filtered measurement parameter set Y best Complete the model identification of each engine in turn and calculate the average correction factor of the batch production engine If the mean value of the correction factor is within the preset correction factor range, it is considered that Y best is the final screening result; otherwise, return to step S42 to continue screening the test parameters.

6. The method for quantitatively analyzing the dispersion of performance of mass-produced engine components considering multi-source information constraints according to claim 4, characterized in that: In step S5, the following steps are specifically included: S51: Bring the collected batch engine key geometry data into the geometry-component association model to determine the prior standard deviation vector σ of the correction factor geo ; S52: Utilize the optimal measurement parameter set Y according to the initial correction factor set X0 and the boundary best Complete the model identification of multiple batch-produced engines and calculate the standard deviation vector σ0 of the correction factor. If each dimension of the standard deviation vector σ0 of the correction factor is not greater than the prior standard deviation vector σ geo The correction factor set does not need to change if the corresponding dimension is doubled. best =Y0, otherwise go to step S53; S53: Record the dimension i combination I where the correction factor is too large, select the engine whose correction factor is at the boundary of these dimensions, and select a new correction factor x from the set of candidate correction factors. new Join from X best In the process, the model identification of the selected engine is performed according to the new set of correction factors, and the correction factors {x i ,i∈I} is reasonable. If it is evaluated, it will be added to the best and follow ±6σ geo The principle is X best Redefine the boundaries; if the evaluation fails, select a new x new Repeat this step; S54: Complete the model identification of multiple batch engines according to the optimized correction factor set X and the boundary, calculate the standard deviation vector σ of the correction factor, and if each dimension of σ is not greater than the prior standard deviation vector σ geo Corresponding to 2 times of the dimension, it is considered that X best is the final optimization result; otherwise, return to step S53 to continue optimization.

Images