Centrifugal fan impeller parameterization optimization method and system based on multiple working conditions
Through the parameterization optimization method of centrifugal fan impeller under multiple operating conditions, the response surface function and proxy model are used, combined with multi-objective optimization and genetic algorithm, the problems of long optimization iteration time and high cost in the existing technology are solved, and the rapid and Reliable impeller optimization effect.
Patent Information
- Application Number
- CN202411201406.6
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2024-08-29
- Publication Date
- 2024-11-15
AI Technical Summary
The existing centrifugal fan impeller optimization method has a long iteration time and high time cost, and the optimization results rely on the experience of engineers, making it difficult to ensure weight loss effect and performance optimization.
The parameterization optimization method of centrifugal fan impeller based on multi-operating conditions is adopted, and the optimal solution is calculated by using parameter definition, operating condition analysis, response surface function establishment and proxy model fitting, combined with multi-objective optimization and constraint control, and the genetic algorithm is used to calculate the optimal solution. Build Pareto cutting-edge, optimize and validate proxy models for efficiency.
It significantly shortens the iteration time, improves optimization efficiency, and reduces time cost. It quickly evaluates the response under different component thickness combinations through the proxy model, avoids a large number of finite element calculations, and ensures the reliability of the optimization results.
Smart Images

Figure CN118965908A_ABST
Abstract
Description
Technical Field
[0001] The present invention relates to the technical field of centrifugal fan optimization, and in particular to a centrifugal fan impeller parameterization optimization method and system based on multiple working conditions. Background Art
[0002] In the prior art, ventilators play an important role in industrial production and daily life. They are widely used in the fields of ventilation, dust removal and cooling in factories, mines, tunnels, cooling towers, vehicles, ships and buildings. For example, in factories, ventilators can effectively remove harmful gases, dust and waste heat, providing workers with a safe and comfortable working environment; in buildings, ventilators can ensure the circulation of indoor air and improve air quality. However, the design of ventilator impellers is generally conservative at this stage, resulting in a large weight. In order to solve this problem, many enterprises have tried corresponding weight reduction methods. However, a considerable part of the weight reduction is only based on trial and error weight reduction based on feeling. Some engineers only rely on experience to reduce the weight of heavy parts such as impeller blades, front discs, and rear discs. This method lacks corresponding theoretical basis and has great safety hazards. In addition, some universities or enterprises use finite element software numerical calculations to strengthen areas with greater stress and thin areas with less stress, and then determine the final optimization solution by re-simulating the optimized solution. This is the optimization method currently used by most people, but the above two solutions are both trial-and-error explorations, with long design iteration times, and the optimization results mainly rely on the experience of engineers, making it difficult to guarantee weight reduction effects.
[0003] The last leading optimization scheme is to first use the optimization software to perform topology optimization to determine the basic configuration of the impeller, then determine the thickness of each component of the impeller through size optimization, and finally perform simulation calculations on the optimized results to determine the final solution. Although the last method uses the software to automatically explore the optimal target response that meets the constraints, such as minimizing the impeller weight under the conditions that the maximum displacement of the impeller is less than 6mm and the first-order frequency is greater than 20Hz, or maximizing the first-order frequency of the impeller under the conditions that the maximum displacement of the impeller is less than 6mm and the impeller weight is reduced by at least 30%, this method has the following problems: First, the optimization iteration time is long and the efficiency is low; second, the relationship between the constraints of the optimization calculation is difficult to control, such as the maximum displacement of less than 6mm and the weight reduction of at least 30% may contradict each other, and it is difficult to obtain good optimization results with such a setting; third, the fan impellers are of various forms and sizes. This method requires topology optimization and size optimization for each impeller, which is extremely time-consuming. At this stage, a parametric optimization method and system for centrifugal fan impellers under multiple working conditions is needed. Summary of the invention
[0004] In order to solve the problems of long iteration time and extremely high time cost in impeller parameter optimization, the present invention provides a centrifugal fan impeller parameter optimization method and system based on multiple working conditions.
[0005] In the first aspect, the present invention provides a parameterized optimization method for a centrifugal fan impeller under multiple working conditions, which adopts the following technical solution: A parameterized optimization method for a centrifugal fan impeller under multiple working conditions, comprising: Parameter definition and working condition analysis are performed according to the thickness of the impeller components, and the thickness parameters are sampled using full factor sampling to obtain response data; DOE analysis is performed based on the response data obtained from parameter definition and operating condition analysis, and the local sensitivity obtained by partial derivative is compared with the average change of the response from DOE analysis; The acquired response data is used to perform PCA principal component analysis to determine the linear correlation between the parameters and the response data; Response surface functions were established using component thickness and response data, and a surrogate model was obtained by fitting the response surface using radial basis functions using data obtained from fractional factorial design and Hammersley sampling; Multi-objective optimization and constraint control based on response surface functions and surrogate models, including using optimization algorithms to calculate the optimal solution and constructing the Pareto frontier from the optimal solution; The proxy model is optimized and verified according to multi-objective optimization and constraint control, and the optimized and verified proxy model is used to perform result evaluation and analysis.
[0006] Furthermore, the parameter definition and operating condition analysis are performed according to the thickness of the impeller components, including defining the thickness of each impeller component as a parameter variable and setting upper and lower limits for the parameter variable, and then defining the centrifugal force condition and the structural natural frequency condition and performing a finite element analysis of the impeller's original size.
[0007] Furthermore, the PCA principal component analysis is performed using the acquired response data, including acquiring the impeller component thickness change data and the response data, calculating the covariance matrix of the original data and performing eigenvalue decomposition to obtain eigenvalues and eigenvectors, constructing a principal component space, projecting the component thickness change and the response data into the principal component space, and performing correlation judgment based on the degree of proximity between the component thickness change and the curve corresponding to the response variable in the principal component space.
[0008] Furthermore, the use of component thickness and response data to establish a response surface function includes selecting a Gaussian function as a radial basis function, collecting data points obtained by finite element calculation that include the thickness value of each component and the corresponding frequency and maximum displacement response value, constructing a response surface function expression with each component thickness combination as input and corresponding frequency and maximum displacement respectively, determining coefficients by solving a linear equation group, and thereby obtaining a response surface function for predicting the frequency and maximum displacement response values under a given component thickness combination.
[0009] Furthermore, the use of radial basis functions to obtain a proxy model by performing response surface fitting on the data obtained by partial factor design and Hammersley sampling includes obtaining data containing the thickness value of each component and the corresponding frequency and maximum displacement response value by partial factor design and Hammersley sampling, selecting the response surface function form according to the characteristics of the data, substituting the data into the response surface function to construct a set of equations to determine the coefficients, and constructing proxy models for frequency and maximum displacement response respectively with component thickness as input and the response surface function after determining the coefficients as the core.
[0010] Furthermore, the method of constructing a Pareto front from the optimal solution includes initializing a population based on an optimization algorithm, calculating the optimal solution by iterative training of the optimization algorithm, performing non-dominated sorting on individuals in the population in each iteration, and collecting non-dominated solutions to construct a Pareto front.
[0011] Furthermore, the optimization and verification of the proxy model based on multi-objective optimization and constraint control includes solving the optimal component thickness value on the proxy model response surface based on multi-objective optimization and constraint control using a genetic algorithm, substituting the thickness value back to generate a new impeller geometry, and verifying the optimization results of the proxy model through finite element model analysis.
[0012] In the second aspect, a centrifugal fan impeller parameter optimization system based on multiple working conditions includes: The data acquisition module is configured to: perform parameter definition and working condition analysis according to the thickness of the impeller components, and sample the thickness parameters using full factor sampling to obtain response data; The preprocessing module is configured to: perform DOE analysis based on the parameter definition and the response data obtained in the operating condition analysis, and compare the local sensitivity obtained by partial derivative with the average change of the response from the DOE analysis; The correlation judgment module is configured to: perform PCA principal component analysis using the acquired response data to judge the linear correlation between the parameter and the response data; A model module, configured to: establish a response surface function using component thickness and response data, and obtain a surrogate model by performing response surface fitting using a radial basis function using data obtained by fractional factorial design and Hammersley sampling; An optimization module is configured to: perform multi-objective optimization and constraint control based on a response surface function and a surrogate model, including calculating an optimal solution using an optimization algorithm and constructing a Pareto frontier from the optimal solution; The output module is configured to: optimize and verify the proxy model according to multi-objective optimization and constraint control, and use the optimized and verified proxy model to perform result evaluation and analysis.
[0013] In a third aspect, the present invention provides a computer-readable storage medium storing a plurality of instructions, wherein the instructions are suitable for being loaded and executed by a processor of a terminal device, for a method for parameterizing and optimizing a centrifugal fan impeller under multiple working conditions.
[0014] In a fourth aspect, the present invention provides a terminal device comprising a processor and a computer-readable storage medium, wherein the processor is used to implement various instructions; the computer-readable storage medium is used to store multiple instructions, wherein the instructions are suitable for being loaded and executed by the processor, a method for parametric optimization of a centrifugal fan impeller under multiple working conditions.
[0015] In summary, the present invention has the following beneficial technical effects: 1. The present invention establishes a response surface function and a proxy model, uses a radial basis function to obtain a proxy model by fitting the response surface using data obtained by partial factor design and Hammersley sampling, and can quickly evaluate the responses under different component thickness combinations. Compared with directly performing finite element analysis, the calculation speed of the proxy model is greatly improved, thereby significantly shortening the evaluation time in each iteration. In the multi-objective optimization process, a genetic algorithm is used to solve the optimal solution on the proxy model response surface, avoiding time-consuming finite element calculations for a large number of design solutions, and greatly reducing the iteration time.
[0016] 2. The present invention can deeply understand the relationship between parameters and responses before optimization by performing data preprocessing steps such as DOE analysis and PCA principal component analysis, avoiding wasting time and computing resources on irrelevant parameters. The local sensitivity obtained by partial derivative is compared with the average change of the response of DOE analysis, and the linear correlation between parameters and response data is determined by PCA principal component analysis, which helps to determine key parameters, concentrate on optimizing important parameters, and reduce time costs.
[0017] 3. The present invention obtains the thickness change data and response data of the impeller components, calculates the covariance matrix of the original data and performs eigenvalue decomposition to obtain eigenvalues and eigenvectors, constructs the principal component space, and projects the component thickness change and response data into the principal component space for correlation judgment. It can quickly screen out the parameters that have a greater impact on the response, improve optimization efficiency, and reduce time cost. BRIEF DESCRIPTION OF THE DRAWINGS
[0018] Figure 1 It is a schematic diagram of the overall process of a parametric optimization method for a centrifugal fan impeller under multiple working conditions according to an embodiment of the present invention. DETAILED DESCRIPTION
[0019] The present invention is further described in detail below in conjunction with the accompanying drawings.
[0020] Example 1 Reference Figure 1 , a parameterized optimization method for a centrifugal fan impeller under multiple working conditions in this embodiment includes: Parameter definition and working condition analysis are performed according to the thickness of the impeller components, and the thickness parameters are sampled using full factor sampling to obtain response data; DOE analysis is performed based on the response data obtained from parameter definition and operating condition analysis, and the local sensitivity obtained by partial derivative is compared with the average change of the response from DOE analysis; The acquired response data is used to perform PCA principal component analysis to determine the linear correlation between the parameters and the response data; Response surface functions were established using component thickness and response data, and a surrogate model was obtained by fitting the response surface using radial basis functions using data obtained from fractional factorial design and Hammersley sampling; Multi-objective optimization and constraint control based on response surface functions and surrogate models, including using optimization algorithms to calculate the optimal solution and constructing the Pareto frontier from the optimal solution; The proxy model is optimized and verified according to multi-objective optimization and constraint control, and the optimized and verified proxy model is used to perform result evaluation and analysis.
[0021] Specifically, a parameterized optimization method for a centrifugal fan impeller under multiple working conditions includes the following steps: like Figure 1 As shown, S1, parameter definition and working condition analysis are performed according to the thickness of the impeller components, and the thickness parameters are sampled using full factor sampling to obtain response data; According to the variable thickness of structural components, it can be divided into several parts, including front disc, blades, outer rear disc, inner rear disc, arc, cone ring, mouth ring, and outer ring. First, the thickness of each component of the impeller is taken as a parameter variable. For example, the thickness of component A is recorded as T1, the thickness of component B is recorded as T2, and so on, until the thickness of component H is recorded as T8, and the upper and lower limits of the thickness of each component are defined. The centrifugal force condition and the structural natural frequency condition are defined in the finite element software to perform finite element analysis of the original size of the structure. Full factor sampling takes into account the value levels and factor combinations of all factors (i.e., thickness parameters), and can comprehensively explore the parameter space to avoid missing important design schemes. For the eight design variables with upper and lower limits, 256 test schemes can be obtained through full factor sampling, ensuring full coverage of the parameter space. The response data obtained include displacement, stress, frequency and other responses, which can provide comprehensive information for multi-objective optimization. These response data reflect the performance of the impeller under different thickness parameter combinations, which is helpful to find the optimal parameter combination and meet the requirements of multiple objectives at the same time. 256 calculations are performed for each combination in the finite element software. Each calculation simulates the performance of the impeller under centrifugal force conditions and structural natural frequency conditions, and obtains the corresponding displacement, stress, frequency and other response data.
[0022] S2. Perform DOE analysis based on the parameter definition and response data obtained in the operating condition analysis, and compare the local sensitivity obtained by partial derivative with the average change of the response from the DOE analysis; After the finite element analysis in step S1, various response data such as displacement, stress, frequency, etc. can be obtained. These response data reflect the performance of the impeller under different component thicknesses and working conditions. The main effect of the results is obtained through DOE analysis, that is, the average change of each response when the size thickness changes. First, the components related to thickness in the impeller are determined as factors, such as blades, front discs, rear discs, etc., and different levels are set for each factor. For example, the blade thickness can be set to three levels of thin, medium, and thick. Full factorial design or partial factorial design is performed according to the selected factors and levels. The full factorial design will consider all level combinations of all factors, while the partial factorial design will select a part of the combination for experiment to reduce the number of experiments. Different combinations of component thicknesses are simulated and calculated in the finite element software to obtain the corresponding response data, such as displacement, stress, frequency, etc. Ensure that sufficiently accurate data is collected for subsequent analysis, and perform statistical analysis on the collected response data. Specifically, for each factor, the average value of the response at different levels of the factor is calculated. Then, compare the differences between these averages to determine the main effect of the factor on the response. By analyzing the differences in the average values of the responses at each factor level, the main effect relationship between the change in dimensional thickness and the average change in each response can be determined. For example, if it is found that when the blade thickness increases from thin to medium, the average value of the first-order frequency increases significantly, and when it increases from medium to thick, the average value of the first-order frequency does not change much, then it can be concluded that the increase in blade thickness within a certain range has a positive main effect on the first-order frequency. In addition, by taking the partial derivative of the structure near the initial value, the local sensitivity of the response can be obtained, which is compared with the average change in the response obtained by DOE (equivalent to the average sensitivity). If the difference is small and within an acceptable range, the main effect relationship can be considered reliable; if the difference is large, it is necessary to further check the accuracy of the analysis method and data, or adjust the experimental design and analysis parameters.
[0023] S3, using the acquired response data to perform PCA principal component analysis to determine the linear correlation between the parameters and the response data; Collect the thickness variation data and response data of the impeller components, and assume that the thickness variation data of the components is a The moment X of , where n is the number of samples and p is the number of component thickness parameters. The response data is a Calculate the covariance matrix of the component thickness variation data matrix X and the response data matrix Y, where q is the number of response variables. and , for the covariance matrix and Perform eigenvalue decomposition respectively. For matrix C, the formula for eigenvalue decomposition is: ,in, is a matrix consisting of eigenvectors, It is a diagonal matrix, and the elements on its diagonal are eigenvalues. The principal components are selected according to the size of the eigenvalues. The first k eigenvectors with larger eigenvalues are selected as the principal components, where k satisfies the cumulative variance contribution rate to reach a certain proportion (such as 80% or 90%). The calculation formula for the cumulative variance contribution rate is: , Among them, k represents the number of selected principal components, and p represents the dimension of the original data, that is, the number of features. Represents the i-th eigenvalue. The eigenvalue is obtained by eigenvalue decomposition of the covariance matrix, which reflects the variance of the data in the direction of the corresponding eigenvector. Then, the component thickness change data and the response data are projected into the principal component space respectively. For the data matrix X, the projected matrix is ,in is a matrix composed of the first k principal component eigenvectors. For the data matrix Y, the projected matrix is , in the principal component space, observe the projected data matrix and , qualitative judgment is made by observing the closeness of the curves corresponding to the component thickness change and the response variable in the principal component space. If the curves are close, it means that the correlation is strong; if the curves are far apart, it means that the correlation is weak. If the two lines are reversed, it means that there is a negative correlation between them. Negative correlation means that when one response variable increases, the other response variable will decrease. In the principal component space, the reverse curves reflect this opposite trend of change. For example, for the maximum displacement response and the first-order frequency response, if their principal component curves are reversed, it means that increasing the maximum displacement of the impeller may lead to a decrease in the first-order frequency. This is because in this case, there is a mutually restrictive relationship between the two response variables.
[0024] S4, establishing a response surface function using component thickness and response data, and using radial basis functions to obtain a surrogate model by fitting the response surface using data obtained by partial factorial design and Hammersley sampling; Finite element calculations are performed on the impeller component thickness combinations determined by the partial factor design to obtain the corresponding frequency and maximum displacement response values. These response data reflect the performance of the impeller under the specific thickness combinations covered by the partial factor design. Hammersley sampling is used to perform more uniform sampling in the parameter space. The frequency and maximum displacement response values under different thickness combinations are also obtained through finite element calculations. After that, the Gaussian function is used as the radial basis function. The radial basis function expression is: , in, Represents the input vector x (component thickness combination) and the known data points The distance between them, c represents a constant, and a series of data points are obtained by finite element calculation , among which, among which, is the number of data points, n is the number of data points, and each data point Contains three component thickness values and their corresponding frequencies and maximum displacement response value , for the frequency response y, the response surface function can be expressed as: , in, is the coefficient of the frequency response to be determined. Each data point has a corresponding coefficient. is the combination of component thicknesses for which the frequency response is to be predicted, is the radial basis function, n is the number of known data points, and for the maximum displacement response z, the response surface function is: , in, is the coefficient of displacement response to be determined. In order to determine the coefficient and , substituting the known data points into the response surface function, we get a linear equation system: ,and , By solving this linear equation system, we get and Based on the determined coefficients, the response surface function is used to quickly predict the frequency and maximum displacement response values for a given component thickness combination.
[0025] For the simulation of functions with more complex relationships between variables and responses, the radial basis function method is used to obtain the proxy model by fitting the response surface using the data obtained by partial factor design and Hammersley sampling. This can make the fitting residual very small, but it may cause the effect of overfitting. In order to determine whether overfitting will occur, the data obtained by Latin hypercube sampling is needed as a verification scheme. The finally obtained proxy model can replace the finite element calculation. The construction of the proxy model is specifically to substitute the collected data into the response surface function, construct a linear equation group, solve the equation group by the above-mentioned response surface function solution method, determine the coefficients in the response surface function, take the component thickness as input, and use the response surface function after the coefficients are determined as the core of the proxy model. For the two response variables of frequency and maximum displacement, respectively, the proxy model is constructed. For example, for the frequency response, the proxy model can be expressed as: , in, is the thickness of the component to be predicted, is the response surface function of frequency. Similarly, the proxy model for the maximum displacement response can be obtained. The proxy model is verified using data that does not participate in the fitting, such as data obtained by Latin hypercube sampling.
[0026] S5. Multi-objective optimization and constraint control based on response surface function and surrogate model, including using optimization algorithm to calculate the optimal solution and constructing the Pareto frontier from the optimal solution; The response function is a mathematical expression obtained by analyzing and fitting component thickness and response data. It takes component thickness as input and response as output. The response function can quickly calculate the response value under a given component thickness, providing a calculation method for the objective function for multi-objective optimization. The proxy model is a simplified representation of the actual physical system, which is established through methods such as partial factor design and Hammersley sampling, and is obtained using response surface fitting. The proxy model quickly evaluates the performance of different component thickness combinations during the multi-objective optimization process, and obtains the optimal solution through an optimization algorithm. The optimization algorithm uses a genetic algorithm. Taking the genetic algorithm as an example, a set of initial component thickness combinations is first randomly generated as a population. Then, the objective function value (i.e., each response value) of each individual is calculated through the response function and the agent model. According to the objective function value, a new population is generated using genetic operations. This process is repeated until the stopping condition is met (such as reaching the maximum number of iterations or the objective function value is no longer significantly improved). Then, the Pareto frontier is constructed. In the multi-objective optimization process, as the optimization algorithm iterates, a series of non-dominated solutions will be obtained. These non-dominated solutions constitute the Pareto frontier. The Pareto frontier shows the trade-off relationship between different objectives. For example, in the parametric optimization of the centrifugal fan impeller, the Pareto frontier includes a series of different component thickness combinations, each of which corresponds to different frequencies and maximum displacement response values. According to actual needs, the most suitable optimization scheme is selected from the Pareto frontier. In addition, in multi-objective optimization, in addition to multiple objective functions, there are also some constraints. For example, the maximum stress of the impeller cannot exceed a certain value, or the component thickness must be within a certain range. According to actual needs, these constraints are implemented by adding a constraint processing mechanism in the optimization algorithm.
[0027] S6. Optimize and verify the proxy model based on multi-objective optimization and constraint control, and use the optimized and verified proxy model to perform result evaluation and analysis; Combined with the Excel table, the proxy model is regarded as a mathematical function. Different component thickness values are entered in the Excel table. The input is the component thickness and other parameters. The formula in Excel will automatically call the calculation formula of the proxy model. The output is the response value (such as frequency, maximum displacement, etc.), calculate the value of the radial basis function, and combine it with the determined coefficients and , calculate the corresponding frequency and maximum displacement response values, the genetic algorithm searches on the response surface of the proxy model, generates new individuals (i.e., component thickness combinations) through continuous iteration, and calculates their fitness values. The fitness values are usually determined based on multiple objective functions and constraints. After multiple iterations, the genetic algorithm gradually converges to the optimal solution, i.e., a set of component thickness values under multi-objective optimization and constraint control. The optimal component thickness values are substituted back into the impeller's geometric model to generate a new impeller geometric structure. This new geometric structure represents the optimal design solution under multi-objective optimization and constraint control. The newly generated geometric structure is subjected to finite element model analysis to verify the optimization results of the proxy model. Example 2 The difference between this embodiment and embodiment 1 is that this embodiment provides a parameterized optimization system for centrifugal fan impellers under multiple working conditions; The data acquisition module is configured to: perform parameter definition and working condition analysis according to the thickness of the impeller components, and sample the thickness parameters using full factor sampling to obtain response data; The preprocessing module is configured to: perform DOE analysis based on the parameter definition and the response data obtained in the operating condition analysis, and compare the local sensitivity obtained by partial derivative with the average change of the response from the DOE analysis; The correlation judgment module is configured to: perform PCA principal component analysis using the acquired response data to judge the linear correlation between the parameter and the response data; A model module, configured to: establish a response surface function using component thickness and response data, and obtain a surrogate model by performing response surface fitting using a radial basis function using data obtained by fractional factorial design and Hammersley sampling; An optimization module, configured to: perform multi-objective optimization and constraint control based on a response surface function and a surrogate model, including calculating an optimal solution using an optimization algorithm, and constructing a Pareto frontier from the optimal solution; The output module is configured to: optimize and verify the proxy model according to multi-objective optimization and constraint control, and use the optimized and verified proxy model to perform result evaluation and analysis.
[0028] A computer-readable storage medium stores a plurality of instructions, wherein the instructions are suitable for being loaded and executed by a processor of a terminal device for a parameterized optimization method of a centrifugal fan impeller under multiple working conditions.
[0029] A terminal device includes a processor and a computer-readable storage medium, wherein the processor is used to implement various instructions; the computer-readable storage medium is used to store multiple instructions, wherein the instructions are suitable for being loaded and executed by the processor to implement a method for parameterizing and optimizing a centrifugal fan impeller under multiple working conditions.
[0030] The above are all preferred embodiments of the present invention, and are not intended to limit the protection scope of the present invention. Therefore, any equivalent changes made based on the structure, shape, and principle of the present invention should be included in the protection scope of the present invention.
Claims
1. A parameterized optimization method for centrifugal fan impeller under multiple working conditions, characterized in that: include: Parameter definition and working condition analysis are performed based on the thickness of the impeller components, and the thickness parameters are sampled using full factor sampling to obtain response data; DOE analysis is performed based on the response data obtained from parameter definition and operating condition analysis, and the local sensitivity obtained by partial derivative is compared with the average change of the response from DOE analysis; The acquired response data is used to perform PCA principal component analysis to determine the linear correlation between the parameters and the response data; Response surface functions were established using component thickness and response data, and a surrogate model was obtained by fitting the response surface using radial basis functions using data obtained from fractional factorial design and Hammersley sampling; Multi-objective optimization and constraint control based on response surface functions and surrogate models, including using optimization algorithms to calculate the optimal solution and constructing the Pareto frontier from the optimal solution; The proxy model is optimized and verified according to multi-objective optimization and constraint control, and the optimized and verified proxy model is used to perform result evaluation and analysis.
2. The method for parameterized optimization of a centrifugal fan impeller under multiple working conditions according to claim 1, characterized in that: The parameter definition and operating condition analysis based on the thickness of the impeller components include defining the thickness of each impeller component as a parameter variable and setting upper and lower limits for the parameter variable, then defining the centrifugal force condition and the structural natural frequency condition and performing a finite element analysis of the impeller's original size.
3. The method for parameterized optimization of a centrifugal fan impeller under multiple working conditions according to claim 1, characterized in that: The PCA principal component analysis is performed using the acquired response data, including acquiring the impeller component thickness change data and the response data, calculating the covariance matrix of the original data and performing eigenvalue decomposition to obtain eigenvalues and eigenvectors, constructing a principal component space, projecting the component thickness change and the response data into the principal component space, and performing correlation judgment based on the degree of proximity of the curves corresponding to the component thickness change and the response variable in the principal component space.
4. The method for parameterized optimization of a centrifugal fan impeller under multiple working conditions according to claim 1, characterized in that: The method of establishing a response surface function using component thickness and response data includes selecting a Gaussian function as a radial basis function, collecting data points obtained by finite element calculation that include thickness values of each component and corresponding frequencies and maximum displacement response values, constructing a response surface function expression that takes each component thickness combination as input and corresponds to frequencies and maximum displacements, respectively, and determining coefficients by solving a linear equation group, thereby obtaining a response surface function for predicting frequency and maximum displacement response values under a given component thickness combination.
5. The method for parameterized optimization of a centrifugal fan impeller under multiple working conditions according to claim 1, characterized in that: The method uses radial basis functions to obtain a proxy model by performing response surface fitting on the data obtained by partial factor design and Hammersley sampling, including obtaining data including the thickness value of each component and the corresponding frequency and maximum displacement response value by partial factor design and Hammersley sampling, selecting the response surface function form according to the characteristics of the data, substituting the data into the response surface function to construct an equation group to determine the coefficients, and using the component thickness as input and the response surface function after determining the coefficients as the core to respectively construct proxy models for frequency and maximum displacement response.
6. The method for parameterized optimization of centrifugal fan impellers under multiple working conditions according to claim 1, characterized in that: The method of constructing the Pareto frontier from the optimal solution includes initializing the population based on the optimization algorithm, calculating the optimal solution through iterative training of the optimization algorithm, performing non-dominated sorting on individuals in the population in each iteration, and collecting non-dominated solutions to construct the Pareto frontier.
7. The method for parameterized optimization of centrifugal fan impellers under multiple working conditions according to claim 1, characterized in that: The optimization and verification of the proxy model according to multi-objective optimization and constraint control includes solving the optimal component thickness value on the proxy model response surface based on multi-objective optimization and constraint control using a genetic algorithm, substituting the thickness value back to generate a new impeller geometry structure, and verifying the optimization result of the proxy model through finite element model analysis.
8. A centrifugal fan impeller parameter optimization system based on multiple working conditions, executing the method according to any one of claims 1 to 7, characterized in that: include: The data acquisition module is configured to: perform parameter definition and working condition analysis according to the thickness of the impeller components, and sample the thickness parameters using full factor sampling to obtain response data; The preprocessing module is configured to: perform DOE analysis based on the parameter definition and the response data obtained in the operating condition analysis, and compare the local sensitivity obtained by partial derivative with the average change of the response from the DOE analysis; The correlation judgment module is configured to: perform PCA principal component analysis using the acquired response data to judge the linear correlation between the parameter and the response data; A model module, configured to: establish a response surface function using component thickness and response data, and obtain a surrogate model by performing response surface fitting using a radial basis function using data obtained by fractional factorial design and Hammersley sampling; An optimization module is configured to: perform multi-objective optimization and constraint control based on a response surface function and a surrogate model, including calculating an optimal solution using an optimization algorithm and constructing a Pareto frontier from the optimal solution; The output module is configured to: optimize and verify the proxy model according to multi-objective optimization and constraint control, and use the optimized and verified proxy model to perform result evaluation and analysis.
9. A computer-readable storage medium storing a plurality of instructions, characterized in that: The instructions are suitable for being loaded by a processor of a terminal device and executing a parameterized optimization method for a centrifugal fan impeller under multiple working conditions as described in any one of claims 1 to 7.
10. A terminal device, comprising a processor and a computer-readable storage medium, wherein the processor is used to implement each instruction; and the computer-readable storage medium is used to store multiple instructions, characterized in that: The instructions are suitable for being loaded by a processor and executing a parameterized optimization method for a centrifugal fan impeller under multiple working conditions as described in any one of claims 1 to 7.
