Unit body performance tolerance distribution method based on multi-level performance transfer

Abstract

The invention provides a unit body performance tolerance distribution method based on multi-level performance transfer, and belongs to the technical field of aero-engines, and the method comprises the steps: constructing a complete machine forward proxy model comprising a plurality of unit bodies based on geometric feature parameters and complete machine performance parameters; in the model, aiming at each unit body, identifying key geometrical characteristic parameters, and carrying out degree-of-freedom dimension reduction on characteristic distribution of each key geometrical characteristic parameter to obtain a single parameter; inputting a single parameter and a numerical range corresponding to each key geometric feature parameter into the complete machine forward proxy model for updating; and aiming at a complete machine target performance requirement, performing uncertainty reverse propagation based on the updated complete machine forward proxy model to obtain geometric feature parameter distribution of each unit body, and based on the obtained distribution, obtaining a tolerance distribution result by adopting a trade-off space. According to the scheme, the performance consistency and interchangeability of the unit bodies can be guaranteed, and agile research and development, rapid derivation and condition-based maintenance of aviation power are promoted.

Description

Technical Field

[0001] The present application relates to the field of aero-engine technology, and in particular to a method for allocating unit performance tolerance based on multi-level performance transfer. Background Art

[0002] Traditional deterministic design ignores inherent uncertainty, so a large degree of conservatism is required in engine design to reserve sufficient margin to prevent performance from failing to meet standards. In the prior art, there is no method to avoid excessive conservatism and further reduce the margin to optimize the margin requirements in engine design. Summary of the invention

[0003] In view of this, an embodiment of the present application provides a unit body performance tolerance allocation method based on multi-level performance transfer, which at least partially solves the problem of overly conservative margin in engine design existing in the prior art.

[0004] The embodiment of the present application provides a method for allocating unit performance tolerance based on multi-level performance transfer, the method comprising:

[0005] Based on the geometric characteristic parameters and the whole machine performance parameters, a whole machine forward proxy model including multiple units is constructed;

[0006] In the whole machine forward proxy model, key geometric characteristic parameters are identified for each unit body;

[0007] For each unit cell, the characteristic distribution of each key geometric characteristic parameter is reduced in degree of freedom to obtain a single parameter;

[0008] For each unit body, a single parameter corresponding to each key geometric feature parameter and a numerical range corresponding to each key geometric feature parameter are input into the whole machine forward proxy model for updating;

[0009] According to the target performance requirements of the whole machine, uncertainty reverse propagation is performed based on the updated whole machine forward proxy model to obtain the distribution of geometric characteristic parameters of each unit body;

[0010] According to the obtained distribution of geometric characteristic parameters of each unit cell, the tolerance allocation result is obtained using the trade-off space.

[0011] According to a specific implementation of the embodiment of the present application, the whole machine forward proxy model including multiple unit bodies is constructed based on the geometric feature parameters and the whole machine performance parameters, including:

[0012] For each unit body, multiple samples are randomly selected from various geometric feature parameter degradation probabilities, and CFD simulation is performed on all different sample combinations to calculate the component characteristics under the influence of geometric feature parameter degradation;

[0013] The component characteristics obtained by simulating each unit body are input into the zero-dimensional whole machine performance model to obtain the whole machine performance under the joint action of multiple geometric characteristic parameter degradation modes;

[0014] Based on the obtained performance of various whole machines, a whole machine forward surrogate model based on geometric feature parameters and whole machine performance parameters is constructed through the surrogate model method.

[0015] According to a specific implementation of the embodiment of the present application, in the whole machine forward proxy model, for each unit body, identifying key geometric feature parameters includes:

[0016] Performing sensitivity analysis based on the whole machine forward proxy model, and calculating the SHAP value of each geometric characteristic parameter in each unit cell respectively;

[0017] For each unit cell, the geometric feature parameters are sorted according to the SHAP value to identify the key geometric feature parameters.

[0018] According to a specific implementation of the embodiment of the present application, the step of performing degree of freedom dimension reduction on the characteristic distribution of each key geometric characteristic parameter to obtain a single parameter includes:

[0019] The first four statistical moments (μx; σx; Sx; Kx) of the probability density of each key geometric feature parameter are represented by a single parameter to complete the dimensionality reduction of degrees of freedom; among them, μx is the mean, σx is the standard deviation, Sx is the skewness, and Kx is the kurtosis.

[0020] According to a specific implementation of the embodiment of the present application, the step of performing degree of freedom dimension reduction on the characteristic distribution of each key geometric characteristic parameter to obtain a single parameter includes:

[0021] When the characteristic distribution of the geometric characteristic parameters is a triangular distribution, the single parameter dx is defined as:

[0022] dx=x ub -x lb ,

[0023] Among them, x ub is the upper boundary of the triangular distribution, x lb is the lower boundary of the triangular distribution.

[0024] According to a specific implementation of the embodiment of the present application, the single parameter corresponding to each key geometric feature parameter and the numerical range corresponding to each key geometric feature parameter are input into the whole machine forward proxy model for updating, including:

[0025] Constructing a mixed Gaussian distribution based on a single parameter corresponding to each key geometric feature parameter and a numerical range corresponding to each key geometric feature parameter;

[0026] The mixed Gaussian distribution is input into the whole machine forward agent model for updating.

[0027] According to a specific implementation of the embodiment of the present application, the uncertainty back propagation is performed based on the updated whole machine forward proxy model for the whole machine target performance requirement to obtain the geometric characteristic parameter distribution of each unit body, including:

[0028] According to the target performance requirements of the whole machine, the target performance standard deviation σ is calculated according to the confidence interval Ay ;

[0029] In each unit cell, multiple geometric characteristic parameters are extracted from the empirical distribution using Monte Carlo sampling to form the standard deviation of the distribution of multiple geometric characteristic parameters;

[0030] The standard deviations of the distributions of multiple geometric feature parameters are input into the updated whole machine forward proxy model to calculate the standard deviation σ of the simulated distribution of the whole machine performance Ty ;

[0031] When the standard deviation σ of the simulated distribution of the overall performance Ty The standard deviation of the target performance σ Ay When the difference between meets the preset matching requirements, the standard deviation σ of the simulated distribution of the whole machine performance is accepted. Ty ;

[0032] When the standard deviation σ of the simulated distribution of the overall performance Ty The standard deviation of the target performance σ Ay When the difference does not meet the preset matching requirements, the distribution of multiple geometric feature parameters is continuously adjusted for iterative optimization until the preset matching requirements are met;

[0033] A plurality of geometric feature parameter distributions corresponding to satisfying preset matching requirements are obtained.

[0034] According to a specific implementation of the embodiment of the present application, the method of obtaining a tolerance allocation result by using a trade-off space according to the obtained geometric feature parameter distribution of each unit cell includes:

[0035] The obtained geometric characteristic parameter distribution of each unit cell is converted into a trade-off space, and the tolerance allocation result is visualized through a two-dimensional cross-sectional diagram of the trade-off space.

[0036] According to a specific implementation of the embodiment of the present application, the visual display of the tolerance allocation result through the two-dimensional cross-sectional diagram of the trade-off space includes:

[0037] Decomposing the n-dimensional input space of the trade-off space into parameter pairs, wherein one parameter pair is composed of two different geometric characteristic parameters, the n-dimensional input space is composed of the geometric characteristic parameters of all obtained unit cells, and each parameter pair corresponds to a two-dimensional cross section of a hypercube of the trade-off space;

[0038] Contour lines are drawn for each two-dimensional cross section to obtain the m-dimensional output space of the trade-off space, and the tolerance allocation structure is visualized through the m-dimensional output space; wherein, the m-dimensional output space is composed of the performance parameters of the whole machine, and each contour line corresponds to a constraint on the deterministic or probabilistic geometric characteristic parameters of the whole machine performance parameters.

[0039] According to a specific implementation method of an embodiment of the present application, the whole-machine forward agent model is constructed based on a machine learning method.

[0040] Beneficial effects:

[0041] The unit body performance tolerance allocation method based on multi-level performance transfer in the embodiment of the present application constructs a multi-level coupling agent model of parts-unit body-whole machine, and uses discrete event simulation technology to allocate tolerances to the geometric feature parameters of the components that meet the performance requirements of the whole machine. Through reverse tolerance allocation, it is possible to monitor whether the geometric structure parameters or unit body performance change within the range. If it exceeds, the whole machine performance may not meet the operating requirements. Unit bodies that meet the range can be selected for replacement during maintenance to ensure that the performance index requirements of the whole machine can be met. The present invention ensures the consistency of unit body performance through key structural feature tolerance control, thereby ensuring the interchangeability of unit body functional performance, which is of great significance to promoting agile research and development, rapid derivation, and condition-based maintenance of aviation power. BRIEF DESCRIPTION OF THE DRAWINGS

[0042] In order to more clearly illustrate the technical solutions of the embodiments of the present application, the drawings required for use in the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present application. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying creative work.

[0043] Figure 1 A schematic diagram of a flow chart of a method for allocating unit performance tolerance based on multi-level performance transfer according to an embodiment of the present invention;

[0044] Figure 2 A schematic diagram of a forward proxy model establishment process according to an embodiment of the present invention;

[0045] Figure 3 (a) is a schematic diagram of a single parameterization of a triangular distribution according to an embodiment of the present invention, and (b) is a schematic diagram of a single parameterization of a mixed Gaussian distribution according to an embodiment of the present invention;

[0046] Figure 4 A schematic diagram of a forward calculation workflow according to an embodiment of the present invention;

[0047] Figure 5 A schematic diagram of a reverse calculation workflow according to an embodiment of the present invention;

[0048] Figure 6 A schematic diagram of a reverse calculation process based on Monte Carlo according to an embodiment of the present invention;

[0049] Figure 7 FIG. 4 is a schematic diagram of a trade-off space according to an embodiment of the present invention. DETAILED DESCRIPTION

[0050] The embodiments of the present application are described in detail below with reference to the accompanying drawings.

[0051] The following describes the implementation methods of the present application through specific examples, and those skilled in the art can easily understand other advantages and effects of the present application from the contents disclosed in this specification. Obviously, the described embodiments are only a part of the embodiments of the present application, rather than all the embodiments. The present application can also be implemented or applied through other different specific implementation methods, and the details in this specification can also be modified or changed in various ways based on different viewpoints and applications without departing from the spirit of the present application. It should be noted that, in the absence of conflict, the following embodiments and the features in the embodiments can be combined with each other. Based on the embodiments in the present application, all other embodiments obtained by ordinary technicians in the field without making creative work belong to the scope of protection of the present application.

[0052] It should be noted that various aspects of the embodiments within the scope of the appended claims are described below. It should be apparent that the aspects described herein can be embodied in a wide variety of forms, and any specific structure and / or function described herein is merely illustrative. Based on the present application, it should be understood by those skilled in the art that an aspect described herein can be implemented independently of any other aspect, and two or more of these aspects can be combined in various ways. For example, any number of aspects described herein can be used to implement the device and / or practice the method. In addition, other structures and / or functionalities other than one or more of the aspects described herein can be used to implement this device and / or practice this method.

[0053] It should also be noted that the illustrations provided in the following embodiments are only schematic illustrations of the basic concept of the present application. The illustrations only show the components related to the present application rather than being drawn according to the number, shape and size of the components in actual implementation. In actual implementation, the type, quantity and proportion of each component may be changed at will, and the component layout may also be more complicated.

[0054] Additionally, in the following description, specific details are provided to facilitate a thorough understanding of the examples. However, it will be understood by those skilled in the art that the aspects described may be practiced without these specific details.

[0055] The present application embodiment provides a unit body performance tolerance allocation method based on multi-level performance transmission, as shown below Figures 1 to 7 A detailed description is given, which specifically includes the following steps:

[0056] Step 1: construct a whole machine forward proxy model including multiple units based on geometric feature parameters and whole machine performance parameters;

[0057] Step 2: In the whole machine forward proxy model, identify key geometric feature parameters for each unit body;

[0058] Step 3: for each unit cell, the characteristic distribution of each key geometric characteristic parameter is subjected to degree of freedom reduction to obtain a single parameter;

[0059] Step 4: for each unit body, input the single parameter corresponding to each key geometric feature parameter and the numerical range corresponding to each key geometric feature parameter into the whole machine forward proxy model for updating;

[0060] Step 5: Based on the target performance requirements of the whole machine, uncertainty reverse propagation is performed based on the updated whole machine forward proxy model to obtain the distribution of geometric characteristic parameters of each unit body;

[0061] Step 6: According to the obtained geometric characteristic parameter distribution of each unit body, the tolerance allocation result is obtained using the trade-off space.

[0062] Inverse uncertainty quantification is an effective means to determine whether the replaced unit body meets the interchangeability of the whole machine performance. It inversely quantifies the unit body performance tolerance range based on the range of the whole machine performance requirements, and further constrains the distribution range of the geometric feature parameters of the parts. It can monitor whether the geometric structure parameters or unit body performance change within the range. If it exceeds, the whole machine performance may not meet the operating requirements. It can also select the unit body that meets the range for replacement during maintenance to ensure that the whole machine performance index requirements can be met. Based on this feature, the embodiment of the present application takes the geometric uncertainty factor as input and the unit body or whole machine performance as output to establish a forward uncertainty propagation agent model; identify key geometric feature parameters through feature extraction and selection methods of sensitivity analysis, simplify the input probability density degrees of freedom and compress the geometric feature space to reduce the calculation dimension; based on the inverse uncertainty quantification method, calculate the selection range of geometric feature parameters under the given performance requirements; finally, in 2D or 3D space, the trade-off space is graphically visualized to intuitively understand the selection interval of geometric feature parameters. This method helps to ensure the consistency of unit performance through tolerance control of key structural features, and further ensure the interchangeability of unit functional performance. It is of great significance to promote agile research and development, rapid derivation, and condition-based maintenance of aviation power.

[0063] In one embodiment, the step of constructing a whole machine forward proxy model including a plurality of unit cells based on geometric feature parameters and whole machine performance parameters includes:

[0064] For each unit body, multiple samples are randomly selected from various geometric feature parameter degradation probabilities, and CFD simulation is performed on all different sample combinations to calculate the component characteristics under the influence of geometric feature parameter degradation;

[0065] The component characteristics obtained by simulating each unit body are input into the zero-dimensional whole machine performance model to obtain the whole machine performance under the joint action of multiple geometric characteristic parameter degradation modes;

[0066] Based on the obtained performance of various whole machines, a whole machine forward surrogate model based on geometric feature parameters and whole machine performance parameters is constructed through the surrogate model method.

[0067] During specific implementation, the uncertainty of the whole machine performance caused by the joint action of degraded geometry is taken into account. By taking multiple samples in a random order from the assumed degradation probabilities of multiple geometric features, CFD simulation is performed on all different sample combinations to calculate the component characteristics under the influence of geometric degradation. The performance of the whole machine under the joint action of multiple geometric degradation modes can be obtained by bringing the simulated component characteristics into the zero-dimensional whole machine performance model. The geometric feature-whole machine performance parameter forward model is constructed through the proxy model method. The proxy model described in this application is to establish a model that can approximately replace the original three-dimensional high-fidelity CFD model and the performance model Y=f(X) for propagation through a certain sample set. Y = f(X) is the mapping from geometric parameters to CFD and then to the whole machine performance, where Y is the whole machine performance and X is the degraded geometric parameters; In order to obtain the mapping of the whole machine performance from the geometric parameters using the proxy model, is the overall performance calculated from the proxy model, X is the degraded geometric parameter, and C is the internal coefficient of the proxy model (integrated in the machine learning model). The ultimate goal is to achieve the Approximately equal to Y. Approximate surrogate model It can save a lot of time, so it is also a necessary method for uncertainty quantification involving CFD. The input of the proxy model is all geometric feature uncertainty parameters, and the output is the performance of the whole machine.

[0068] In one embodiment, in the whole machine forward proxy model, identifying key geometric feature parameters for each unit body includes:

[0069] Performing sensitivity analysis based on the whole machine forward proxy model, and calculating the SHAP value of each geometric characteristic parameter in each unit cell respectively;

[0070] For each unit cell, the geometric feature parameters are sorted according to the SHAP value to identify the key geometric feature parameters.

[0071] In the specific implementation, the global sensitivity analysis method is adopted to decompose the original model into the sum of increasing terms of different dimensional parameter combinations based on the system response variance, so that the contribution of a certain uncertainty factor or combination to the total variance of the system response can be quantitatively calculated. Several items with large contributions are selected for the allocation of uncertainty tolerance, and the remaining features are treated as constants in the tolerance allocation.

[0072] Specifically, SHAP (SHapley Additive exPlanations) is a model interpretation method based on game theory. Its core principle comes from the Shapley value in game theory, which can quantify the marginal contribution of each participant (i.e., geometric features) to the overall benefit (i.e., model prediction). The feature contribution value of SHAP can just reflect the importance of geometric features and is suitable for sensitivity analysis. Therefore, in this embodiment, SHAP is combined with the constructed SVR proxy model to perform sensitivity analysis. The overall contribution of geometric features to model prediction can be used as global sensitivity. The larger the average absolute SHAP value, the greater the impact of the geometric features on the prediction results of the model, indicating that the feature is more sensitive. Therefore, the geometric feature parameters are sorted in order from large to small according to the SHAP value, and the geometric feature parameters that are ranked first are the key geometric feature parameters.

[0073] In one embodiment, the step of performing degree of freedom reduction on the characteristic distribution of each key geometric characteristic parameter to obtain a single parameter includes:

[0074] The first four statistical moments (μx; σx; Sx; Kx) of the probability density of each key geometric feature parameter are represented by a single parameter to complete the dimensionality reduction of degrees of freedom; among them, μx is the mean, σx is the standard deviation, S is the skewness, and Kx is the kurtosis.

[0075] Furthermore, the degree of freedom of the characteristic distribution of each key geometric characteristic parameter is reduced to obtain a single parameter, including:

[0076] When the characteristic distribution of the geometric characteristic parameters is a triangular distribution, the single parameter dx is defined as:

[0077] dx=x ub -x lb ,

[0078] Among them, x ub is the upper boundary of the triangular distribution, x lb is the lower boundary of the triangular distribution.

[0079] In practice, the statistical moments of the probability density function are input until the probability that satisfies all constraints is reached. The standard deviation can be reduced and the mean determined, as well as other moments such as skewness (Sx) and kurtosis (Kx). Directly manipulate the statistical moments of all inputs to find multiple possible uncertainty reduction schemes to ensure that the input probability density is achievable. To solve this problem, a parameterized function is proposed to define the shape of the probability density to reduce the input parameters and achieve the purpose of dimensionality reduction. For the triangular distribution, refer to Figure 3 (a), define a single parameter dx as: dx = x ub -x lb , where x ub is the upper boundary of the triangular distribution, x lb is the lower boundary of the triangular distribution. The effect of adjusting a single parameter on the probability density can be determined, for example, by adjusting the value of x relative to the nominal value x. n (the mode in a triangular distribution) reshapes the probability density by scaling the width. For a new width d' x , d' x =x' ub -x l ' b The new parameter of the probability density is calculated from the geometric proportion and the unit area property of the triangle to obtain the corresponding width, that is, x n The distance to the new limit has the same proportional length as shown in the following formula:

[0080]

[0081] The boundary x′ of the new triangular distribution ub , x′ lb It can be obtained from the above equation (1).

[0082] If the triangular distribution is expressed according to the normal probability density, at least three parameters are involved to fix a triangular distribution, such as the maximum value, minimum value and mode of the triangular distribution (the so-called three degrees of freedom). However, if none of these three parameters are fixed according to the Monte Carlo sampling calculation method, the sampling complexity will be increased and the calculation process will be too time-consuming. Therefore, in this embodiment, the three parameters are converted into a parameter called dx (dx in the triangular distribution is equal to the distance from the mode to the minimum value). By fixing some conditions (such as the position of the mode, the ratio of the distance from the maximum and minimum values ​​to the mode), a parameter dx can express a triangular distribution. At this time, the degree of freedom of this triangular distribution is reduced from three dimensions to one dimension, which simplifies the subsequent calculation process and reduces the calculation cost. In the above case, since the degree of freedom of the triangular distribution is reduced from three to one, it will affect all possible distribution spaces of the triangular distribution. In addition, the triangular distribution can also achieve the purpose of parameterization by modifying the upper and lower bounds separately.

[0083] For Gaussian distribution, refer to Figure 3 (b) For distributions with shapes between left-skewed and right-skewed, such as symmetric distributions, we can replace the nominal value x n =d x Defined as input parameter To reshape the Gaussian mixture distribution f gm ,in, and μ′ x Given the upper and lower bounds of the value range, Parameterization involves linear transformation of the constituent functions of the following formula to obtain a new mean and a new standard deviation by modifying the μ′ at the position x to achieve.

[0084]

[0085] in,

[0086]

[0087] Among them, μ1, μ2, σ1, σ2 are the mean and variance of the original mixed Gaussian distribution, μ1', μ'2, σ1', σ'2 are the mean and variance of the reshaped mixed Gaussian distribution, and α1, α2 are the weight coefficients of the reshaped mixed Gaussian distribution.

[0088] Furthermore, the single parameter corresponding to each key geometric feature parameter and the numerical range corresponding to each key geometric feature parameter are input into the whole machine forward proxy model for updating, including:

[0089] Constructing a mixed Gaussian distribution based on a single parameter corresponding to each key geometric feature parameter and a numerical range corresponding to each key geometric feature parameter;

[0090] The mixed Gaussian distribution is input into the whole machine forward agent model for updating.

[0091] In the specific implementation, for example, for the unit body of the high-pressure compressor, the key geometric features identified include the tip clearance increment and the surface roughness, and the degrees of freedom of these two parameters are reduced. It is assumed that both parameters are Gaussian mixture distributions, the range of surface roughness is [10, 50] μm, and the tip clearance increment is [0, 0.8] mm. The parameter dx is used to construct a mixed Gaussian distribution based on a single parameter, and then it is input into the whole machine forward proxy model.

[0092] Furthermore, the uncertainty reverse propagation is performed based on the updated whole machine forward proxy model for the whole machine target performance requirements to obtain the geometric characteristic parameter distribution of each unit body, including:

[0093] According to the target performance requirements of the whole machine, the target performance standard deviation σ is calculated according to the confidence interval Ay ;

[0094] In each unit cell, multiple geometric characteristic parameters are extracted from the empirical distribution using Monte Carlo sampling to form the standard deviation of the distribution of multiple geometric characteristic parameters;

[0095] The standard deviations of the distributions of multiple geometric feature parameters are input into the updated whole machine forward proxy model to calculate the standard deviation σ of the simulated distribution of the whole machine performance Ty ;

[0096] When the standard deviation σ of the simulated distribution of the overall performance Ty The standard deviation of the target performance σ Ay When the difference between meets the preset matching requirements, the standard deviation σ of the simulated distribution of the whole machine performance is accepted. Ty ;

[0097] When the standard deviation σ of the simulated distribution of the overall performance Ty The standard deviation of the target performance σ Ay When the difference does not meet the preset matching requirements, the distribution of multiple geometric feature parameters is continuously adjusted for iterative optimization until the preset matching requirements are met;

[0098] A plurality of geometric feature parameter distributions corresponding to satisfying preset matching requirements are obtained.

[0099] When implementing it, refer to Figure 5 and 6 In each unit cell, Monte Carlo sampling is used to extract possible geometric characteristic parameters from the empirical distribution to form the standard deviation of the distribution Substitute into the forward surrogate model to calculate the standard deviation σ of the simulated distribution of the whole machine performance Ty , and the target performance standard deviation σ Ay Matching is performed, such as satisfying the preset matching requirement Δ=|σ Ty -σ Ay |, then accept the σ Ty If it is not satisfied, continue to adjust σ x Optimization loop. The computational cost is the product of the number of optimization iterations and the number of sampling procedure runs in each loop. For example, each Monte Carlo generates 10,000 samples, the number of iterations is up to 400, and the experimental results are used as the distribution of degraded geometric parameters that meet the performance conditions.

[0100] In one embodiment, the method of obtaining tolerance allocation results by using a trade-off space according to the obtained geometric characteristic parameter distribution of each unit cell includes:

[0101] The obtained geometric characteristic parameter distribution of each unit cell is converted into a trade-off space, and the tolerance allocation result is visualized through a two-dimensional cross-sectional diagram of the trade-off space.

[0102] According to a specific implementation of the embodiment of the present application, the visual display of the tolerance allocation result through the two-dimensional cross-sectional diagram of the trade-off space includes:

[0103] Decomposing the n-dimensional input space of the trade-off space into parameter pairs, wherein one parameter pair is composed of two different geometric characteristic parameters, the n-dimensional input space is composed of the geometric characteristic parameters of all obtained unit cells, and each parameter pair corresponds to a two-dimensional cross section of a hypercube of the trade-off space;

[0104] Contour lines are drawn for each two-dimensional cross section to obtain the m-dimensional output space of the trade-off space, and the tolerance allocation structure is visualized through the m-dimensional output space; wherein, the m-dimensional output space is composed of the performance parameters of the whole machine, and each contour line corresponds to a constraint on the deterministic or probabilistic geometric characteristic parameters of the whole machine performance parameters.

[0105] In practice, the trade-off space can be visualized by a two-dimensional cross-section of the trade-off space, which can be represented by the following:

[0106] 1) Decompose the n-dimensional input space into parameter pairs, (x1; x2), (x3; x4) ... (xn-1; xn), where x nFor the nth geometry parameter, all n geometry parameters should be visualized in pairs. In case of an odd number of parameters, one parameter can be repeated. Each pair of parameters represents a plane of the hypercube, and these planes intersect at the point of nominal value.

[0107] 2) The m-dimensional output space can be visualized through a two-dimensional cross-section of the space containing contour lines, each of which represents a deterministic or probabilistic constraint on the overall performance parameters, where the n-dimensional input space refers to the geometric parameter space and the m-dimensional output space refers to the overall performance space.

[0108] The degraded geometric characteristic parameters are transformed into a two-dimensional trade-off space, and the area surrounded by it is identified by calculating and drawing contour lines corresponding to specific overall performance constraints (such as temperature T6, fuel consumption SFC and thrust Fn), such as Figure 7 The inner area close to the coordinate axis is the range of values ​​of the degraded geometric parameters that meet the performance conditions of the whole machine. Designers can select values ​​within the area according to actual conditions. The gray area exceeds the performance requirements of the whole machine. The dotted boundary is the optimal performance that meets the conditions of the whole machine, also known as the optimal boundary.

[0109] By using the trade-off space, we can more clearly observe the space where the geometric uncertainty can take values ​​and the space that does not meet the requirements, providing visual support for designers to weigh the tolerance allocation results.

[0110] In one embodiment, the whole-machine forward proxy model is constructed based on a machine learning method.

[0111] The unit performance tolerance allocation method based on multi-level performance transfer of the present application is described in detail below with a specific embodiment. Figures 1 to 7 , specifically including the following steps:

[0112] Taking the high pressure compressor (HPC) unit as an example, in the HPC, due to the influence of the adverse pressure gradient on the blade surface, the geometric uncertainty of the upstream blade will be amplified in the downstream, and finally affect the performance of the whole machine, making the engine performance unable to meet the expected operating standards. This example takes the high pressure compressor of an aircraft engine as a unit and considers the tolerance allocation problem of the three degradation forms of its blades: leading edge erosion, surface roughness, and tip clearance.

[0113] Step 1: Based on the unit body, construct the forward proxy model corresponding to each unit body, integrate the forward proxy models of each unit body, and obtain the forward proxy model of the whole machine.

[0114] When taking samples through LHS, the samples need to be shuffled, and finally 150 different DOE (DESIGN OF EXPERIMENT) combinations are formed. The samples taken need to cover the sample space as much as possible. Therefore, through CFD simulation and zero-dimensional whole machine performance model, the HPC characteristics under the joint action of multiple degradation modes can be obtained, and the component characteristics obtained by simulation are brought into the zero-dimensional whole machine performance model to obtain the whole machine performance under the joint action of multiple degradation modes. Based on SVR (support vector machine regression method), a proxy model from geometric parameters to whole machine performance is constructed. In the SVR proxy model, the absolute percentage error of HPC performance does not exceed 0.4%, the overall error of flow rate is the largest, and the absolute percentage error of exhaust temperature and fuel consumption rate of whole machine performance does not exceed 0.08%.

[0115] LHS stands for Latin hypercube sampling (abbreviated as LHS), which is a method of approximate random sampling from a multivariate parameter distribution. It belongs to the stratified sampling technique and is often used in computer experiments or Monte Carlo integration.

[0116] In this step, multiple samples are randomly selected from the assumed degradation probabilities of multiple geometric features, and CFD simulation is performed on all different sample combinations to calculate the component characteristics under the influence of geometric degradation. The component characteristics obtained by simulation are brought into the zero-dimensional whole machine performance model to obtain the whole machine performance under the joint action of multiple degradation modes. Then, a geometric feature-whole machine performance parameter forward model is constructed through the surrogate model method. The surrogate model input is all geometric feature uncertainty parameters, and the output is the whole machine performance.

[0117] Step 2: Identify key geometric feature parameters

[0118] SHAP is combined with the constructed SVR proxy model to perform sensitivity analysis. The overall contribution of geometric feature parameters to model prediction can be used as global sensitivity. The larger the average absolute SHAP value, the greater the influence of the feature on the prediction results of the model, indicating that the feature is more sensitive. After analysis, the contribution of the tip clearance increment ΔC is the largest, while the contribution of the leading edge erosion proportional factor PF is the smallest, and for the pressure ratio of HPC, the contribution of surface roughness and tip clearance increment is similar.

[0119] In this step, the contribution of degraded geometric parameters to its performance and operational stability is identified through sensitivity analysis, so that they can be focused on and controlled in the surrogate model.

[0120] Step 3: Reduce the degree of freedom for each key geometric feature parameter

[0121] The tip clearance increment and surface roughness are selected as two parameters for DOF reduction. It is assumed that both parameters are Gaussian mixture distributions, the range of surface roughness is [10, 50] μm, and the tip clearance increment is [0, 0.8] mm. The parameter dx is used to construct a mixed Gaussian distribution based on a single parameter, which is then input into the proxy model.

[0122] Step 4: Back Propagation of Uncertainty

[0123] In each unit cell, Monte Carlo sampling is used to extract possible geometric parameters from the empirical distribution to form the standard deviation of the distribution. Substitute into the proxy model to calculate the simulated distribution of the whole machine performance σ Ty , and with the target performance range σ Ay Matching is performed, such as satisfying the preset matching requirement Δ=|σ Ty -σ Ay |, then accept the σ Ty If it is not satisfied, continue to adjust σ x Optimization loop. The computational cost is the product of the number of optimization iterations and the number of sampling procedure runs in each loop. Each Monte Carlo generates 10,000 samples and the number of iterations is up to 400 for the test. The experimental results are used as the distribution of degraded geometric parameters that meet the performance conditions.

[0124] In this step, through the reverse calculation workflow, we can start from the known whole machine performance parameters and infer the possible geometric degradation feature distribution that causes these performances, so as to achieve the goal of tolerance allocation.

[0125] Step 5: Geometric feature tolerance assignment

[0126] The trade-off space can be visualized by plotting a two-dimensional cross-section of the trade-off space. Multidimensional trade-off spaces can be visualized by:

[0127] 1) Decompose the n-dimensional input space into pairs of parameters, (x1; x2), (x3; x4) ... (xn-1; xn). All n parameters should be visualized in pairs. In case the number of parameters is odd, one parameter can be repeated. Each pair of parameters represents a plane of the hypercube. These planes intersect at points of nominal value.

[0128] 2) The m-dimensional output space can be visualized through a two-dimensional cross-section of the space containing contour lines, each of which represents a deterministic or probabilistic constraint on the overall machine performance parameters.

[0129] The degenerate geometry d PF With d ΔC Transformed into a two-dimensional trade-off space, the area enclosed by it is identified by calculating and drawing contour lines corresponding to specific overall performance constraints (here considering temperature T6, fuel consumption SFC and thrust Fn), such as Figure 7 The interior of the region is the range of values ​​that the degraded geometric parameters can take to meet the performance conditions of the whole machine. Designers can select values ​​within the region according to actual conditions. The gray part exceeds the performance requirements of the whole machine. The dotted boundary is the optimal performance that meets the conditions of the whole machine, also known as the optimal boundary. Using the trade-off space, we can more clearly observe the space where the geometric uncertainty can take values ​​and the space that does not meet the requirements, providing visual support for designers to weigh the tolerance allocation results.

[0130] The embodiment provided by the present invention is to quantify the range of whole machine performance requirements during the life cycle and inversely quantify the unit body performance tolerance range. It is necessary to first use geometric uncertainty factors as input and unit body or whole machine performance as output to establish a forward uncertainty propagation agent model; identify key geometric features through feature extraction and selection methods using sensitivity analysis, compress the feature space, and retain key feature parameters of the unit body in the application scenario of condition-based maintenance; further simplify the input feature parameters, retain the necessary degrees of freedom to reduce the calculation dimension; based on the inverse uncertainty quantification method, calculate the range of feature parameter selection that meets given performance requirements; finally, in 2D or 3D space, graph the trade-off space to intuitively select the interval.

[0131] Through reverse tolerance allocation, it is possible to monitor whether the geometric structure parameters or unit body performance change within the range. If it exceeds, the performance of the whole machine may not meet the operating requirements. In addition, units that meet the range can be selected for replacement during maintenance to ensure that the performance index requirements of the whole machine can be met. The present invention ensures the consistency of unit body performance through key structural feature tolerance control, thereby ensuring the interchangeability of unit body functional performance, which is of great significance to promoting agile research and development, rapid derivation, and condition-based maintenance of aviation power.

[0132] The above is only a specific implementation of the present application, but the protection scope of the present application is not limited thereto. Any changes or substitutions that can be easily thought of by a person skilled in the art within the technical scope disclosed in the present application should be included in the protection scope of the present application. Therefore, the protection scope of the present application shall be based on the protection scope of the claims.

Claims

1. A unit performance tolerance allocation method based on multi-level performance transfer, characterized in that: The method comprises: Based on the geometric characteristic parameters and the whole machine performance parameters, a whole machine forward proxy model including multiple units is constructed; In the whole machine forward proxy model, key geometric characteristic parameters are identified for each unit body; For each unit cell, the characteristic distribution of each key geometric characteristic parameter is reduced in degree of freedom to obtain a single parameter; For each unit body, a single parameter corresponding to each key geometric feature parameter and a numerical range corresponding to each key geometric feature parameter are input into the whole machine forward proxy model for updating; According to the target performance requirements of the whole machine, uncertainty reverse propagation is performed based on the updated whole machine forward proxy model to obtain the distribution of geometric characteristic parameters of each unit body; According to the obtained distribution of geometric characteristic parameters of each unit cell, the tolerance allocation result is obtained using the trade-off space.

2. The unit performance tolerance allocation method based on multi-level performance transfer according to claim 1 is characterized in that: The whole machine forward proxy model including multiple unit bodies is constructed based on the geometric feature parameters and the whole machine performance parameters, including: For each unit body, multiple samples are randomly selected from various geometric feature parameter degradation probabilities, and CFD simulation is performed on all different sample combinations to calculate the component characteristics under the influence of geometric feature parameter degradation; The component characteristics obtained by simulating each unit body are input into the zero-dimensional whole machine performance model to obtain the whole machine performance under the joint action of multiple geometric characteristic parameter degradation modes; Based on the obtained performance of various whole machines, a whole machine forward surrogate model based on geometric feature parameters and whole machine performance parameters is constructed through the surrogate model method.

3. The unit performance tolerance allocation method based on multi-level performance transfer according to claim 1 is characterized in that: In the whole machine forward proxy model, for each unit body, key geometric feature parameters are identified, including: Performing sensitivity analysis based on the whole machine forward proxy model, and calculating the SHAP value of each geometric characteristic parameter in each unit cell respectively; For each unit cell, the geometric feature parameters are sorted according to the SHAP value to identify the key geometric feature parameters.

4. The unit performance tolerance allocation method based on multi-level performance transfer according to claim 1 is characterized in that: The method of performing a degree of freedom dimension reduction on the characteristic distribution of each key geometric characteristic parameter to obtain a single parameter includes: The first four statistical moments (μx; σx; Sx; Kx) of the probability density of each key geometric feature parameter are represented by a single parameter to complete the dimensionality reduction of degrees of freedom; among them, μx is the mean, σx is the standard deviation, Sx is the skewness, and Kx is the kurtosis.

5. The unit performance tolerance allocation method based on multi-level performance transfer according to claim 1 is characterized in that: The method of performing a degree of freedom dimension reduction on the characteristic distribution of each key geometric characteristic parameter to obtain a single parameter includes: When the characteristic distribution of the geometric characteristic parameters is a triangular distribution, the single parameter dx is defined as: dx=x ub -x lb , Among them, x ub is the upper boundary of the triangular distribution, x lb is the lower boundary of the triangular distribution.

6. The unit performance tolerance allocation method based on multi-level performance transfer according to claim 1 is characterized in that: The step of inputting the single parameter corresponding to each key geometric feature parameter and the numerical range corresponding to each key geometric feature parameter into the whole machine forward proxy model for updating includes: Constructing a mixed Gaussian distribution based on a single parameter corresponding to each key geometric feature parameter and a numerical range corresponding to each key geometric feature parameter; The mixed Gaussian distribution is input into the whole machine forward agent model for updating.

7. The unit performance tolerance allocation method based on multi-level performance transfer according to claim 6 is characterized in that: According to the target performance requirements of the whole machine, uncertainty reverse propagation is performed based on the updated whole machine forward proxy model to obtain the distribution of geometric characteristic parameters of each unit body, including: According to the target performance requirements of the whole machine, the target performance standard deviation σ is calculated according to the confidence interval Ay ; In each unit cell, multiple geometric characteristic parameters are extracted from the empirical distribution using Monte Carlo sampling to form the standard deviation of the distribution of multiple geometric characteristic parameters; The standard deviations of the distributions of multiple geometric feature parameters are input into the updated whole machine forward proxy model to calculate the standard deviation σ of the simulated distribution of the whole machine performance Ty ; When the standard deviation σ of the simulated distribution of the overall performance Ty The standard deviation of the target performance σ Ay When the difference between meets the preset matching requirements, the standard deviation σ of the simulated distribution of the whole machine performance is accepted. Ty ; When the standard deviation σ of the simulated distribution of the overall performance Ty The standard deviation of the target performance σ Ay When the difference does not meet the preset matching requirements, the distribution of multiple geometric feature parameters is continuously adjusted for iterative optimization until the preset matching requirements are met; A plurality of geometric feature parameter distributions corresponding to satisfying preset matching requirements are obtained.

8. The unit performance tolerance allocation method based on multi-level performance transfer according to claim 1 is characterized in that: The method of obtaining tolerance allocation results by using a trade-off space according to the obtained geometric characteristic parameter distribution of each unit body includes: The obtained geometric characteristic parameter distribution of each unit cell is converted into a trade-off space, and the tolerance allocation result is visualized through a two-dimensional cross-sectional diagram of the trade-off space.

9. The unit performance tolerance allocation method based on multi-level performance transfer according to claim 8, characterized in that: The method of visually displaying the tolerance allocation result through a two-dimensional cross-sectional diagram of the trade-off space includes: Decomposing the n-dimensional input space of the trade-off space into parameter pairs, wherein one parameter pair is composed of two different geometric characteristic parameters, the n-dimensional input space is composed of the geometric characteristic parameters of all obtained unit cells, and each parameter pair corresponds to a two-dimensional cross section of a hypercube of the trade-off space; Contour lines are drawn for each two-dimensional cross section to obtain the m-dimensional output space of the trade-off space, and the tolerance allocation structure is visualized through the m-dimensional output space; wherein, the m-dimensional output space is composed of the performance parameters of the whole machine, and each contour line corresponds to a constraint on the deterministic or probabilistic geometric characteristic parameters of the whole machine performance parameters.

10. The unit performance tolerance allocation method based on multi-level performance transfer according to any one of claims 1 to 9, characterized in that: The whole machine forward agent model is constructed based on a machine learning method.

Images