A method for identifying the wall pressure parameters of the entire flow channel of a scramjet engine

Abstract

The present invention discloses a method for identifying parameters of the wall surface pressure of a scramjet engine, comprising: obtaining a relevant data set of the wall surface pressure of the whole flow channel; constructing a proxy model of the uncertainty propagation of parameters of the turbulence model of the whole flow channel by using a training set; obtaining multiple identification models and optimal parameters according to the proxy model, Bayesian theory and particle swarm algorithm; obtaining relevant pressure data corresponding to the current identification model by using numerical simulation software; obtaining an error according to the relevant pressure data and pressure data obtained from a real experiment; judging whether the error exceeds a preset threshold, judging whether the current identification model is the last identification model, and performing nesting and fusion processing on all identification models; performing parameter identification on a verification set by using an integrated parameter identification model; analyzing the identification results; optimizing the integrated parameter identification model; and identifying parameters of the wall surface pressure of the whole flow channel by using the optimized integrated parameter identification model.

Description

Technical Field

[0001] The present invention relates to the technical field of engine parameter identification, and in particular to a method for identifying full-flow channel wall pressure parameters of a scramjet engine. Background Art

[0002] Hypersonic vehicles have the advantages of high penetration success rate and strong survivability. They can be used for medium and long-range rapid and precise strikes, time-sensitive target reconnaissance, and round-trip transportation between the earth and the sky. Scramjet engines have many advantages such as light weight, high speed, and high unit thrust. They are the preferred propulsion system for hypersonic vehicles / hypersonic cruise missiles, an indispensable and important part of our military equipment system in the future, and an ideal power unit for near-space hypersonic vehicles. Therefore, the research on scramjet engines has great strategic significance. With the rapid development of computer technology, the ability of computational fluid dynamics to solve engineering problems of scramjet engines has been rapidly improved, playing an important role in theoretical research and engineering applications. In specific engineering experiments, turbulence dominates the entire flow state during the flight of the aircraft, and turbulence simulation has also become a key technology for numerical simulation. In order to improve the numerical simulation accuracy and speed of scramjet engines in complex flows, the parameters of the turbulence model must be identified.

[0003] At present, in engineering applications such as aerodynamic design and optimization, people often only focus on average flow, so RANS simulation has become the mainstream of numerical simulation at this stage. In RANS simulation, turbulence models are needed to solve equations. At present, turbulence model parameters mainly use nominal values ​​or semi-artificial empirical values ​​under simple flows for numerical simulation, which results in large numerical simulation errors for complex flows. Summary of the invention

[0004] The object of the present invention is to provide a method for identifying wall pressure parameters of a scramjet engine full flow channel, so as to improve the accuracy and speed of numerical simulation of the scramjet engine full flow channel in a complex environment.

[0005] The technical solution of the present invention to solve the above technical problems is as follows:

[0006] The present invention provides a method for identifying the wall pressure parameters of a full flow passage of a scramjet engine, the method comprising:

[0007] S101: Obtain relevant data sets of the wall pressure of the entire flow channel of a scramjet engine;

[0008] S102: Divide the relevant data set into a training set and a validation set;

[0009] S103: constructing a proxy model for parameter uncertainty propagation of the full-channel turbulence model using the training set;

[0010] S104: obtaining multiple identification models and optimal parameters of each identification model according to the proxy model, Bayesian theory and particle swarm algorithm;

[0011] S105: according to the optimal parameters of the current identification model, using numerical simulation software, obtaining relevant pressure data corresponding to the current identification model;

[0012] S106: Obtaining an error according to the relevant pressure data and pressure data obtained from actual experiments;

[0013] S107: Determine whether the error exceeds a preset threshold. If so, optimize the current identification model using an error analysis method and return to step S104. Otherwise, proceed to step S108.

[0014] S108: Determine whether the current identification model is the last identification model. If so, proceed to step S109. Otherwise, take the next identification model as the current sub-model and return to step S105.

[0015] S109: performing nesting and fusion processing on all the identification models to obtain an integrated parameter identification model with adaptive turbulence model parameter identification characteristics;

[0016] S110: performing parameter identification on the verification set using the integrated parameter identification model to obtain an identification result;

[0017] S111: Analyze the identification result using the error analysis method to obtain an analysis result;

[0018] S112: Optimizing the integrated parameter identification model using the analysis results.

[0019] Optionally, before step S101, the method includes:

[0020] According to the working principle and test conditions of scramjet engines, determine the required flow and working conditions for ground wind tunnel tests;

[0021] Use the wall pressure sensor to obtain the original pressure data of the entire flow channel wall under the current incoming flow and working conditions;

[0022] De-noising the original pressure data to obtain a wind tunnel test data set of full-flow channel wall pressure of a scramjet engine;

[0023] Use numerical simulation software to obtain a priori sample data set of the entire flow channel wall pressure;

[0024] The relevant data set is obtained according to the scramjet engine full flow channel wall pressure wind tunnel test data set and the full flow channel wall pressure priori sample data set.

[0025] Optionally, in step S103, the matrix expression of the proxy model is:

[0026]

[0027] in, h Indicates the pressure output value, Hc Represents the proxy model matrix form, represents the number of training samples, represents the coefficient to be solved and after solving it is , express The transpose of Represents the matrix form of the basis function, represents the wall pressure value corresponding to the closure parameter in the training set, represents the number of terms in the polynomial expansion and , represents the order of the polynomial expansion, D represents the number of turbulence model parameters that need to be calibrated, Represents basis functions.

[0028] Optionally, the step S104 includes:

[0029] According to the surrogate model and Bayesian theory, construct a likelihood function;

[0030] Using the likelihood function as the fitness function of the particle swarm algorithm to obtain a target particle swarm algorithm;

[0031] Iterate using the target particle swarm algorithm and save all updated parameters;

[0032] The kernel density estimation method is used to estimate the posterior probability of each parameter and determine the probability density distribution of each parameter within the interval.

[0033] According to the probability density distribution, determining a large posterior probability parameter as an optimal parameter;

[0034] The proxy model, the Bayesian theory, the likelihood function and the target particle swarm optimization algorithm are integrated to obtain the identification model corresponding to each structure of the entire flow channel.

[0035] Alternatively, different likelihood functions construct different errors, assuming:

[0036] M1:

[0037]

[0038] M2:

[0039]

[0040] in, represents the likelihood function in logarithmic form, L represents the number of samples of the true test value, i Indicates the serial number, Indicates i The position of the actual test value corresponding to the position point, Indicates i The actual test value corresponding to the position point, Indicates i The mean of the parameters of the sample points, No. i The variance of the parameter of sample points, represents the measurement error, Represents the approximate output of the surrogate model.

[0041] Optionally, in step S107, the error analysis method includes:

[0042] The average relative error and root mean square error are used to evaluate the accuracy change before and after the turbulence model parameter identification.

[0043] Optionally, the average relative error is:

[0044]

[0045] in, represents the mean relative error, L represents the number of samples of the true test value, i Indicates the serial number, Indicates i The position of the actual test value corresponding to the position point, Indicates i The actual test value corresponding to the position point, Indicates the first i The identified wall pressure value corresponding to each position point.

[0046] Optionally, the root mean square error RMSE for:

[0047]

[0048] in, L represents the number of samples of the true test value, i Indicates the serial number, Indicatesi The actual test value corresponding to the position point, Indicates the first i The identified wall pressure value corresponding to each position point.

[0049] Optionally, in step S109, the adaptive characteristics of the integrated parameter identification model take the detection of the wall pressure inflection point as the main indicator. When different wall pressure inflection points are detected, the relevant functions of the integrated parameter identification model are automatically called, and the parameter identification models at different positions of the entire flow channel are selected for adaptive parameter identification.

[0050] The present invention has the following beneficial effects:

[0051] The present invention can perform block-by-block and integrated identification of the turbulence model parameters of the entire flow path of a scramjet engine in a complex flow environment, and finally form a set of adaptive turbulence model parameter identification methods to automatically identify the turbulence model parameters of the current state, thereby improving the accuracy and speed of numerical simulation, laying a foundation for high-precision calculation work of scramjet engines in the numerical simulation process, and providing technical support for the safe and reliable operation of the engine during flight. BRIEF DESCRIPTION OF THE DRAWINGS

[0052] Figure 1 Flow chart of the method for identifying the full flow channel wall pressure parameters of a scramjet engine according to the present invention

[0053] Figure 2 The figure is a schematic diagram of the structure of the entire flow passage of the scramjet engine of the present invention. DETAILED DESCRIPTION

[0054] The principles and features of the present invention are described below in conjunction with the accompanying drawings. The examples given are only used to explain the present invention and are not used to limit the scope of the present invention.

[0055] The present invention provides a method for identifying the wall pressure parameters of the entire flow channel of a scramjet engine, referring to Figure 1 As shown, the method for identifying the full flow channel wall pressure parameters of a scramjet engine includes:

[0056] S101: Obtain relevant data sets of the wall pressure of the entire flow channel of a scramjet engine;

[0057] The relevant data sets here include the full-channel wall pressure wind tunnel test data set and the full-channel wall pressure prior sample data set.

[0058] Prior to this, the present invention determined the incoming flow and working conditions required for the ground wind tunnel test based on the working principle and test conditions of the scramjet engine;

[0059] Use the wall pressure sensor to obtain the original pressure data of the entire flow channel wall under the current incoming flow and working conditions;

[0060] De-noising the original pressure data to obtain a wind tunnel test data set of full-flow channel wall pressure of a scramjet engine;

[0061] In the present invention, reference Figure 2 As shown in FIG. 1 , the full flow path of a scramjet engine includes four parts: an inlet duct, an isolator section, a combustion chamber, and a tail nozzle. Therefore, the full flow path wall pressure wind tunnel test data set of a scramjet engine is also four data sets, including an inlet duct data set, an isolator section data set, a combustion chamber data set, and a tail nozzle data set.

[0062] Use numerical simulation software to obtain a priori sample data set of the entire flow channel wall pressure;

[0063] The numerical simulation software uses AHL3D software independently developed by the China Aerodynamics Research and Development Center. The incoming flow conditions and operating parameters for numerical simulation calculations are consistent with those of ground wind tunnel tests, ensuring that numerical simulations are performed under the same environment. Combined with the independence analysis of the computational grid, the size of the numerical simulation calculation grid is determined, and the nominal value under simple flow is used as one of the results to be compared with the subsequent parameter identification results. The nominal value specifically refers to the middle value of the turbulence model parameter range. At the same time, according to the range of different parameters of the turbulence model selected in the numerical simulation, Latin hypercube is used for sampling to obtain prior samples. In this way, a data set obtained by numerical simulation calculations is obtained, and data sample preprocessing is performed.

[0064] Here, the Latin hypercube method is:

[0065] (1) First, determine the sample size N, that is, the number of samples to be drawn;

[0066] (2) Divide the interval range of each parameter of the turbulence model into N segments;

[0067] (3) Randomly select a value in each of the N segments of each parameter;

[0068] (5) Randomly combine the values ​​extracted from different parameters to obtain the sampling results.

[0069] (6) The sampling results are put into the AHL3D software for recalculation, and the calculation results are analyzed and preprocessed to screen and eliminate completely worthless data to ensure the quality of the basic data set. Completely worthless data includes abnormal data and redundant data, and finally form a priori sample data set of the wall pressure of the entire flow channel.

[0070] It should be noted that the full-flow channel wall pressure prior sample data set also contains four data sets, which correspond to the full-flow channel wall pressure wind tunnel test data set of the scramjet engine.

[0071] The relevant data set is obtained according to the scramjet engine full flow channel wall pressure wind tunnel test data set and the full flow channel wall pressure priori sample data set.

[0072] The wind tunnel test data set of the full flow channel wall pressure of the scramjet engine and the prior sample data set of the full flow channel wall pressure are merged to form a related data set, where the related data set includes the inlet data set, the isolation section data set, the combustion chamber data set and the tail nozzle data set.

[0073] S102: Divide the relevant data set into a training set and a validation set;

[0074] In the present invention, 80% of the data are selected from the relevant data sets as training sets for constructing the proxy model; and 20% of the data are used to verify the accuracy of the proxy model.

[0075] S103: constructing a proxy model for parameter uncertainty propagation of the full-channel turbulence model using the training set;

[0076] The role of the proxy model is to construct a nonlinear mathematical relationship between the input and output of data samples. The present invention uses a chaotic polynomial expansion to propagate the mathematical relationship between the input parameters and the output parameters. The chaotic polynomial expansion is expressed as:

[0077]

[0078] in, represents the fitting result, represents the coefficient to be solved, represents the number of terms in the polynomial expansion, Represents the basis function, the order of the polynomial expansion Number of decision items The size of , here we use the 2nd order expansion, The expression is:

[0079]

[0080] Using regression method to solve the coefficient , the matrix expression of the proxy model is:

[0081]

[0082] in, h Indicates the pressure output value, Hc Represents the proxy model matrix form, represents the number of training samples, represents the coefficient to be solved and after solving it is , express The transpose of Represents the matrix form of the basis function, represents the wall pressure value corresponding to the closure parameter in the training set, represents the number of terms in the polynomial expansion and , represents the order of the polynomial expansion, D represents the number of turbulence model parameters that need to be calibrated, Represents basis functions.

[0083] coefficient It can be solved by the least squares estimation method and can be written as follows:

[0084]

[0085] in, Represents the wall pressure value corresponding to the closure parameter in the training set.

[0086] It should be noted that, since there are four relevant data sets in the present invention, there are also four proxy models determined according to the four relevant data sets.

[0087] S104: obtaining multiple identification models and optimal parameters of each identification model according to the proxy model, Bayesian theory and particle swarm algorithm;

[0088] According to the surrogate model and Bayesian theory, construct a likelihood function;

[0089] Bayesian theory is specifically:

[0090] In the parameter estimation process, the parameter estimation problem is converted into a Bayesian method to solve the inverse problem:

[0091]

[0092] in, For the data set, is the parameter to be identified in the turbulence model, is the posterior distribution of the parameter, is the likelihood function of the parameter, is the prior distribution of the parameter. The prior distribution of the parameter is uniform distribution, that is, we only care about the likelihood function, let:

[0093]

[0094] Bayesian inference is used to construct different likelihood functions, which construct different error hypotheses:

[0095] M1:

[0096]

[0097] M2:

[0098]

[0099] in, represents the likelihood function in logarithmic form, L represents the number of samples of the true test value, i Indicates the serial number, Indicates i The position of the actual test value corresponding to the position point, Indicates i The actual test value corresponding to the position point, Indicates i The mean of the parameters of the sample points, No. i The variance of the parameter at each sample point is represents the measurement error, Represents the approximate output of the surrogate model.

[0100] Using the likelihood function as the fitness function of the particle swarm algorithm to obtain a target particle swarm algorithm;

[0101] The particle swarm optimization algorithm is specifically:

[0102] The particle swarm optimization algorithm relies on the continuous updating of input parameters to find the parameters that make the fitness function optimal. The input parameters are the turbulence model parameters, and the fitness function is the likelihood function. The specific particle swarm optimization process is:

[0103] (1) Initialize particles (parameters), including particle velocity initialization and position initialization;

[0104] (2) Generate new positions for each particle based on the initial positions and velocities;

[0105] (3) Calculate the fitness value of each particle according to the fitness function;

[0106] (4) For each particle, compare the fitness values ​​before and after the update individually and in groups, and retain the optimal fitness value;

[0107] (5) Then update the speed and position, compare the fitness function values, and update the individual and group values ​​respectively;

[0108] (6) During the iteration process, if the constraints are not met, the optimization continues. When the maximum number of iterations is met, the algorithm ends.

[0109] Keep updated parameters during optimization, the position of parameter updates and speed Directly affects the optimization results. The specific expression is:

[0110]

[0111] in, t Indicates t Iterations, i Indicates i parameter iterations, M Indicates the number of turbulence model parameters that need to be calibrated.

[0112] Combining the above-mentioned proxy model, Bayesian theory, likelihood function and optimization algorithm, these methods are integrated to form a fast and efficient turbulence model parameter identification algorithm. For the inlet, isolation section, combustion chamber and tail nozzle of the full flow channel of the scramjet engine, four parameter identification models are finally formed to identify the parameters of the full flow channel turbulence model respectively, that is, in the block identification process, four sets of optimal parameters will be formed.

[0113] Iterate using the target particle swarm algorithm and save all updated parameters;

[0114] The kernel density estimation method is used to estimate the posterior probability of each parameter and determine the probability density distribution of each parameter within the interval.

[0115] According to the probability density distribution, determining a large posterior probability parameter as an optimal parameter;

[0116] The proxy model, the Bayesian theory, the likelihood function and the target particle swarm optimization algorithm are integrated to obtain the identification model corresponding to each structure of the entire flow channel.

[0117] S105: according to the optimal parameters of the current identification model, using numerical simulation software, obtaining relevant pressure data corresponding to the current identification model;

[0118] That is: the four sets of optimal parameters are recalculated in combination with the AHL3D numerical simulation software to obtain the identified inlet duct wall pressure, isolation section wall pressure, combustion chamber wall pressure and tail nozzle wall pressure as well as the flow field data corresponding to each part.

[0119] S106: Obtain the error based on the relevant pressure data and the pressure data obtained from the actual test; compare the identified inlet duct wall pressure, isolation section wall pressure, combustion chamber wall pressure and tail nozzle wall pressure results with the actual test values, and analyze the degree of agreement between the numerical simulation calculation results and the actual test values. At the same time, compare the numerical simulation calculation results of the nominal values ​​of the turbulence model parameters (that is, the nominal values ​​of the turbulence model parameters specifically refer to the middle values ​​within the range of the turbulence model parameter) with the calculation results after identification, and analyze the errors before and after the turbulence model parameter identification.

[0120] S107: Determine whether the error exceeds a preset threshold. If so, optimize the current identification model using an error analysis method and return to step S104. Otherwise, proceed to step S108.

[0121] Here, the error analysis method includes:

[0122] The average relative error and root mean square error are used to evaluate the accuracy change before and after the turbulence model parameter identification.

[0123] Optionally, the average relative error is:

[0124]

[0125] in, represents the mean relative error, L represents the number of samples of the true test value, i Indicates the serial number, Indicates i The position of the actual test value corresponding to the position point, Indicates i The actual test value corresponding to the position point, Indicates the first i The identified wall pressure value corresponding to each position point.

[0126] Optionally, the root mean square error RMSE for:

[0127]

[0128] in, L represents the number of samples of the true test value, i Indicates the serial number, Indicates i The actual test value corresponding to the position point, Indicates the first i The identified wall pressure value corresponding to each position point.

[0129] S108: Determine whether the current identification model is the last identification model. If so, proceed to step S109. Otherwise, take the next identification model as the current sub-model and return to step S105.

[0130] Since the present invention has four identification models, the current operation is to perform block identification on each identification model.

[0131] S109: performing nesting and fusion processing on all the identification models to obtain an integrated parameter identification model with adaptive turbulence model parameter identification characteristics;

[0132] The adaptive characteristic of the integrated parameter identification model of the present invention takes the detection of the wall pressure inflection point as the main indicator. When different wall pressure inflection points are detected, the relevant functions of the integrated parameter identification model are automatically called, and the parameter identification models at different positions of the entire flow channel are selected for adaptive parameter identification.

[0133] S110: performing parameter identification on the verification set using the integrated parameter identification model to obtain an identification result;

[0134] S111: Analyze the identification result using the error analysis method to obtain an analysis result;

[0135] The error analysis method here is the same as the above-mentioned error analysis method. Since the timing of causing the error is uncertain, it is necessary to analyze the errors caused at each timing.

[0136] S112: Optimizing the integrated parameter identification model using the analysis results.

[0137] The above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, improvements, etc. made within the spirit and principle of the present invention should be included in the protection scope of the present invention.

Claims

1. A method for identifying the full-flow channel wall pressure parameters of a scramjet engine, characterized in that: The method for identifying the full flow channel wall pressure parameters of a scramjet engine comprises: S101: Obtain relevant data sets of the wall pressure of the entire flow channel of a scramjet engine; S102: Divide the relevant data set into a training set and a validation set; S103: constructing a proxy model for parameter uncertainty propagation of the full-channel turbulence model using the training set; S104: According to the proxy model, Bayesian theory and particle swarm algorithm, multiple identification models and optimal parameters of each identification model are obtained, specifically including: According to the surrogate model and Bayesian theory, a likelihood function is constructed; different likelihood functions construct different errors. Assume that: M1： M2： in, represents the likelihood function in logarithmic form, L represents the number of samples of the true test value, i Indicates the serial number, Indicates i The position of the actual test value corresponding to the position point, Indicates i The actual test value corresponding to the position point, Indicates i The mean of the parameters of the sample points, No. i The variance of the parameter at each sample point is represents the measurement error, represents the approximate output of the surrogate model; Using the likelihood function as the fitness function of the particle swarm algorithm to obtain a target particle swarm algorithm; Iterate using the target particle swarm algorithm and save all updated parameters; The kernel density estimation method is used to estimate the posterior probability of each parameter and determine the probability density distribution of each parameter within the interval. According to the probability density distribution, determining a large a posteriori probability parameter as an optimal parameter; The proxy model, the Bayesian theory, the likelihood function and the target particle swarm optimization algorithm are integrated to obtain the identification model corresponding to each structure of the entire flow channel; S105: according to the optimal parameters of the current identification model, using numerical simulation software, obtaining relevant pressure data corresponding to the current identification model; S106: Obtaining an error according to the relevant pressure data and pressure data obtained from actual experiments; S107: Determine whether the error exceeds a preset threshold. If so, optimize the current identification model using an error analysis method and return to step S104. Otherwise, proceed to step S108. The error analysis method includes: The average relative error and root mean square error are used to judge the change in accuracy before and after the turbulence model parameter identification; The average relative error is: in, represents the mean relative error, Indicates the first i The identified wall pressure value corresponding to each position point; The root mean square error RMSE for: ； S108: Determine whether the current identification model is the last identification model. If so, proceed to step S109. Otherwise, take the next identification model as the current identification model and return to step S105. S109: performing nesting and fusion processing on all the identification models to obtain an integrated parameter identification model with adaptive turbulence model parameter identification characteristics; S110: performing parameter identification on the verification set using the integrated parameter identification model to obtain an identification result; S111: Analyze the identification result using the error analysis method to obtain an analysis result; S112: optimizing the integrated parameter identification model using the analysis results to obtain an optimized integrated parameter identification model; S113: Using the optimized integrated parameter identification model, the full flow channel wall pressure parameters of the scramjet engine are identified.

2. The method for identifying the full flow passage wall pressure parameters of a scramjet engine according to claim 1, characterized in that: Before step S101, the method includes: According to the working principle and test conditions of scramjet engines, determine the required flow and working conditions for ground wind tunnel tests; Use the wall pressure sensor to obtain the original pressure data of the entire flow channel wall under the current incoming flow and working conditions; De-noising the original pressure data to obtain a wind tunnel test data set of full-flow channel wall pressure of a scramjet engine; Use numerical simulation software to obtain a priori sample data set of the entire flow channel wall pressure; The relevant data set is obtained according to the scramjet engine full flow channel wall pressure wind tunnel test data set and the full flow channel wall pressure priori sample data set.

3. The method for identifying the full flow passage wall pressure parameters of a scramjet engine according to claim 1, characterized in that: In step S103, the matrix expression of the proxy model is: in, h Indicates the pressure output value, Hc Represents the proxy model matrix form, represents the number of training samples, represents the coefficient to be solved and after solving it is , express The transpose of Represents the matrix form of the basis function, represents the wall pressure value corresponding to the closure parameter in the training set, represents the number of terms in the polynomial expansion and , represents the order of the polynomial expansion, D represents the number of turbulence model parameters that need to be calibrated, Represents basis functions.

4. The method for identifying the full flow passage wall pressure parameters of a scramjet engine according to any one of claims 1 to 3, characterized in that: In step S109, the adaptive characteristics of the integrated parameter identification model take the detection of the wall pressure inflection point as the main indicator. When different wall pressure inflection points are detected, the relevant functions of the integrated parameter identification model are automatically called, and the parameter identification models at different positions of the entire flow channel are selected for adaptive parameter identification.

Images