Design variable value determination method and device of molecular pump and molecular pump

Abstract

The embodiment of the invention provides a design variable value determination method and device of a molecular pump and the molecular pump. The method comprises the steps that blade array geometrical shape parameters are determined according to blade array structure design variables of the molecular pump; inputting the geometric shape parameter of the blade array into a neural network model to obtain a predicted transmission probability parameter output by the neural network model; constructing a fitness function taking a leaf array structure design variable as input and molecular pump performance parameters as output by utilizing the predicted transmission probability parameters; a genetic algorithm is used for designing variables for the leaf column structure to generate a value population; and according to the fitness function value of each individual in the value population, selecting an individual meeting a preset demand condition to obtain a target value of the leaf array structure design variable. And by directly associating the design parameters with the performance indexes, the design process is visualized and intelligent, the tedious step of manually adjusting indirect parameters is reduced, and the design efficiency is remarkably improved.

Description

Technical Field

[0001] The present disclosure relates to the technical field of molecular pumps, and in particular to a method and device for determining the values of design variables of a molecular pump, and a molecular pump. Background Art

[0002] The turbomolecular pump, an advanced device that integrates a pump body, a rotating impeller, a static impeller, and an efficient drive system, drives the directional flow of gas molecules through the high-speed rotation of the impeller, achieving efficient pumping. Its features include fast start-up, strong resistance to atmospheric shock, oil-free operation, and the provision of a clean and ultra-high vacuum environment for high-end fields such as metal smelting, electronics and electrical, aerospace, etc., showing outstanding performance in high-vacuum processing in particular. The overall pumping efficiency of the turbomolecular pump is closely linked to the pumping efficiency of a single blade row, and the latter has a direct and far-reaching impact on the performance of the former. Specifically, the structural design of the blade row, including key geometric parameters such as the blade shape, inclination angle, and pitch-chord ratio, are core factors affecting the pumping capacity of the pump. Fine optimization of these parameters is of guiding significance for the structural design and performance optimization of the pump.

[0003] However, the existing technology has obvious deficiencies in this link. On the one hand, the design process is inefficient. Designers need to manually determine multiple key structural design variables such as tooth root diameter, tooth top diameter, blade height, etc. with the help of changing trends of indirect parameters such as speed ratio, angle, and pitch-chord ratio. Due to the lack of intuitive design methods, this process requires repeated trial and error, which is not only time-consuming and labor-intensive, but also difficult to obtain a global optimal solution, which seriously affects the design efficiency. On the other hand, there are difficulties in optimizing the performance of molecular pumps. The maximum compression ratio and the maximum Ho coefficient are the core parameters for measuring the performance of molecular pumps, and the relationship between the two is intricate. Traditional design methods are difficult to effectively guide how to maximize these two parameters at the same time, and cannot solve this complex dual-objective optimization problem. When determining the values of design variables, designers cannot accurately weigh the relationship between the two, which makes it difficult for molecular pumps to achieve optimal performance.

[0004] These issues in design efficiency and performance optimization severely restrict the quality and efficiency of molecular pump design. Developing new design methods and technologies to more efficiently and accurately determine the values of molecular pump design variables has become a key issue that needs to be addressed in this field. Summary of the Invention

[0005] To address the above-mentioned problems with existing methods for determining the design variables of molecular pumps, the present disclosure provides a method, device, and molecular pump for determining the design variables of a molecular pump, thereby providing a more efficient and accurate solution for determining the design variables of a molecular pump.

[0006] Based on the above problems, in a first aspect, a method for determining the values of design variables of a molecular pump is provided, comprising:

[0007] Determining blade row geometric parameters according to blade row structural design variables of the molecular pump; wherein the blade row geometric parameters are used to determine molecular pump performance, and the blade row geometric parameters are characterized by the blade row structural design variables;

[0008] Inputting the leaf row geometry parameters into a neural network model to obtain a predicted transmission probability parameter output by the neural network model;

[0009] Using the predicted transmission probability parameters, a fitness function is constructed with the blade row structure design variables as input and the molecular pump performance parameters as output;

[0010] Generate a value population for the leaf row structure design variables using a genetic algorithm;

[0011] Individuals that meet preset requirements are selected based on the fitness function value of each individual in the value population to obtain target values of the leaf row structure design variables; wherein the fitness function value is used to characterize the performance parameter value of the molecular pump, and the preset requirement condition characterizes the limitation of the performance parameter values of different molecular pumps.

[0012] In conjunction with the first aspect, in one possible implementation, the neural network model is trained in the following manner:

[0013] Constructing a neural network model with the leaf row geometry parameters as input and the transmission probability parameters as output;

[0014] Extracting a first preset number of sample data within the value range of each leaf row geometric shape parameter to obtain the first preset number of input sample groups;

[0015] For each input sample group, the corresponding transmission probability parameter is solved based on the Monte Carlo simulation method as the output sample group;

[0016] The input sample group is used as input and the output sample group is used as a true value to train the neural network model to obtain the neural network model for predicting transmission probability parameters.

[0017] In combination with the first aspect, in one possible implementation, the neural network model includes an input layer, a hidden layer, and an output layer; the number of nodes in the input layer is the same as the number of types of leaf column geometric shape parameters, and the output layer includes two nodes corresponding to the forward transmission probability and the reverse transmission probability, respectively.

[0018] In conjunction with the first aspect, in one possible implementation, using the predicted transmission probability parameter to construct a fitness function with the blade row structure design variables as input and molecular pump performance parameters as output includes:

[0019] Dividing the blades of the molecular pump corresponding to the values of the blade row structure design variables into a second preset number of segments along the radial direction;

[0020] For each segment, determine the corresponding leaf row geometric parameters according to the segmented leaf row structural design variables;

[0021] Normalizing the leaf row geometric shape parameters corresponding to the second preset value segments by type and inputting the normalized parameters into the neural network model to obtain transmission probability parameters corresponding to the leaf row geometric shape parameters;

[0022] An integration operation is performed on the transmission probability parameter to obtain a corresponding molecular pump performance parameter.

[0023] In conjunction with the first aspect, in one possible implementation, selecting an individual that meets preset requirements based on the fitness function value of each individual in the value population to obtain a target value of the leaf row structure design variable includes:

[0024] Generate a Pareto front based on the blade row structure design variables and molecular pump performance parameters corresponding to each individual in the value population;

[0025] A point that meets a preset requirement condition is selected on the Pareto front, and the value of the leaf row structure design variable corresponding to the point is determined as the target value.

[0026] In conjunction with the first aspect, in one possible implementation, generating a value population for the leaf row structure design variables using a genetic algorithm includes:

[0027] According to the value range of each leaf row structure design variable, the initial population is randomly generated;

[0028] Calculate the molecular pump performance parameters corresponding to each individual in the initial population using the fitness function;

[0029] Arrange the individuals in the initial population in descending order based on non-dominated quick sorting and crowding calculation;

[0030] Perform iterative operations on the current population;

[0031] The iterative operation steps include:

[0032] Selecting a preset number of individuals from the current group in sequence, generating new individuals using a genetic algorithm for the selected individuals, and calculating molecular pump performance parameters corresponding to the new individuals using the fitness function;

[0033] The population set is screened based on non-dominated quick sorting and crowding calculation to generate a new current population with the same number of individuals as the current population, wherein the population set is the union of the current population and the new individuals;

[0034] Determine whether the current population meets the preset convergence conditions;

[0035] If not satisfied, return to the step of performing the iterative operation on the current population;

[0036] If satisfied, the current population is determined as the set of all feasible solutions.

[0037] In combination with the first aspect, in a possible implementation manner, the fitness function value includes: molecular pump performance parameters; the molecular pump performance parameters include: maximum compression ratio and maximum Ho coefficient;

[0038] The step of selecting a point on the Pareto front that meets a preset requirement and determining the value of the leaf row structure design variable corresponding to the point as a target value includes:

[0039] If the maximum compression ratio is the primary requirement, then the point with the highest maximum compression ratio is selected on the Pareto front, and the value of the leaf row structure design variable corresponding to the point is determined as the target value;

[0040] If the maximum Ho coefficient is the primary requirement, then the point with the highest maximum Ho coefficient is selected on the Pareto front, and the value of the leaf row structure design variable corresponding to the point is determined as the target value;

[0041] If the requirements for the maximum compression ratio and the maximum Ho coefficient are balanced, a point on the Pareto front with the smallest difference between the maximum compression ratio and the maximum Ho coefficient is selected, and the value of the leaf row structure design variable corresponding to the point is determined as the target value.

[0042] In conjunction with the first aspect, in one possible implementation, the blade row structure design variables include root diameter, tip diameter, blade height, blade thickness, blade root inclination angle, blade tip inclination angle, and number of blades; the blade row geometric shape parameters include blade inclination angle, speed ratio, and pitch-chord ratio;

[0043] The blade row geometry parameters are determined based on the blade row structure design variables of the molecular pump in the following manner:

[0044] determining the blade inclination angle based on the blade tip inclination angle and the blade root inclination angle;

[0045] determining the speed ratio based on the tooth tip diameter and the tooth root diameter;

[0046] The pitch-chord ratio is determined based on the tooth root diameter, tooth tip diameter, blade height, blade thickness, blade root inclination angle, blade tooth tip inclination angle, and the number of blades.

[0047] In a second aspect, a device for determining the value of a design variable of a molecular pump is provided, comprising:

[0048] a determination module, configured to determine blade row geometric parameters according to blade row structural design variables of the molecular pump; wherein the blade row geometric parameters are used to determine molecular pump performance; and the blade row geometric parameters are characterized by the blade row structural design variables;

[0049] A prediction module, configured to input the leaf row geometry parameters into a neural network model to obtain a predicted transmission probability parameter output by the neural network model;

[0050] A construction module, configured to use the predicted transmission probability parameters to construct a fitness function having the blade row structure design variables as input and the molecular pump performance parameters as output;

[0051] A generation module, configured to generate a value population for the leaf row structure design variables using a genetic algorithm;

[0052] A selection module is used to select individuals that meet preset requirements based on the fitness function value of each individual in the value population to obtain target values of the leaf row structure design variables; wherein the fitness function value is used to characterize the performance parameter value of the molecular pump, and the preset requirement condition is a limitation on the performance parameter values of different molecular pumps.

[0053] In a third aspect, a molecular pump is provided, wherein the blade row structure design variables of the molecular pump are determined by a method for determining the value of a design parameter of a molecular pump as described in the first aspect, or in combination with any possible implementation of the first aspect.

[0054] In a fourth aspect, a computer-readable storage medium is provided, on which a computer program is stored. When the computer program is executed by a processor, the steps of a method for determining the values of design parameters of a molecular pump as described in the first aspect, or in combination with any possible embodiment of the first aspect are executed.

[0055] The beneficial effects of the embodiments of the present disclosure include:

[0056] The disclosed embodiment provides a method and device for determining the values of design variables of a molecular pump, which is applied to the design stage before the manufacture of the molecular pump, and more directly determines the blade row structure design variables of the molecular pump that meet the requirements. First, the blade row geometric shape parameters are determined according to the blade row structure design variables, and a direct mapping between the design variables and the physical structure is established to solve the problem of fuzzy correlation between structural parameters and performance in traditional design; secondly, the geometric shape parameters are input into the neural network model to obtain the predicted transmission probability parameters, and the neural network fitting ability is used to quickly and accurately predict the molecular transmission characteristics, shorten the prediction cycle, and reduce data dependence; then, a fitness function is constructed with design variables as input and performance parameters as output to realize the correlation between performance quantitative evaluation and design variables, providing a search guide for the optimization algorithm; then, a genetic algorithm is used to generate a population of design variable values, and the advantages of the evolutionary algorithm are used to efficiently search for the optimal solution combination in the design variable space, avoiding local optimality; finally, based on the fitness function value, individuals that meet the preset requirements are screened to achieve directional optimization of design variable values and control of performance constraints, avoiding trial and error blindness. In summary, this method for determining the values of molecular pump design variables combines the intelligent modeling capabilities of neural networks and the global optimization capabilities of genetic algorithms. It breaks through the inefficient model of traditional molecular pump design that relies on empirical formulas or a large number of experiments, and achieves efficient and accurate determination of design variable values. It significantly improves the design efficiency and performance optimization level of molecular pumps, and provides reliable technical support for the refined design of high vacuum equipment. BRIEF DESCRIPTION OF THE DRAWINGS

[0057] Figure 1 A flow chart of a method for determining the values of design variables of a molecular pump provided in an embodiment of the present disclosure;

[0058] Figure 2 A comparison chart of the prediction results and the predicted forward transmission probability under the training set provided by the embodiment of the present disclosure;

[0059] Figure 3 A comparison chart of the prediction results and the predicted reverse transmission probability under the training set provided by the embodiment of the present disclosure;

[0060] Figure 4 A comparison chart of the prediction results and the predicted forward transmission probability under the test set provided by the embodiment of the present disclosure;

[0061] Figure 5 A comparison chart of the prediction results and the predicted reverse transmission probability under the test set provided by the embodiment of the present disclosure;

[0062] Figure 6 A Pareto front diagram using blade pitch angle as a design parameter provided in an embodiment of the present disclosure;

[0063] Figure 7 A Pareto front diagram of the marked partial coordinates with the blade pitch angle as the design parameter provided in an embodiment of the present disclosure;

[0064] Figure 8 A Pareto front diagram with design point numbers marked using blade pitch angle as a design parameter provided by an embodiment of the present disclosure;

[0065] Figure 9 A two-dimensional model diagram of a molecular pump provided in an embodiment of the present disclosure;

[0066] Figure 10 A schematic diagram of a device for determining the values of design variables of a molecular pump provided in an embodiment of the present disclosure. DETAILED DESCRIPTION

[0067] The present disclosure provides a method and apparatus for determining the design variable values of a molecular pump, and a molecular pump. Preferred embodiments of the present disclosure are described below in conjunction with the accompanying drawings. It should be understood that the preferred embodiments described herein are intended only to illustrate and explain the present disclosure and are not intended to limit the present disclosure. Furthermore, the embodiments and features of the embodiments herein may be combined with one another unless there is a conflict.

[0068] The embodiment of the present disclosure provides a method for determining the value of a design variable of a molecular pump, such as Figure 1 As shown, it can be implemented as:

[0069] S101. Determining blade row geometry parameters based on blade row structural design variables of the molecular pump; wherein the blade row geometry parameters are used to determine molecular pump performance, and the blade row geometry parameters are characterized by the blade row structural design variables;

[0070] S102, inputting the leaf row geometry parameters into the neural network model to obtain the predicted transmission probability parameters output by the neural network model;

[0071] S103, using the predicted transmission probability parameters to construct a fitness function with the blade row structure design variables as input and the molecular pump performance parameters as output;

[0072] S104, using a genetic algorithm to generate a value population for the leaf row structure design variables;

[0073] S105. Select individuals that meet preset requirements based on the fitness function value of each individual in the value population to obtain target values of the leaf row structure design variables; wherein the fitness function value is used to characterize the performance parameter value of the molecular pump, and the preset requirement condition characterizes the limitation of the performance parameter values of different molecular pumps.

[0074] In the disclosed embodiments, the overall pumping efficiency of the turbomolecular pump is closely related to the pumping efficiency of individual blade rows, and the performance of individual blade rows directly impacts its overall performance. The structural design of the blade row is a key factor influencing the pumping capacity of the turbomolecular pump. By fine-tuning the blade row structural design variables, the turbomolecular pump performance parameters (maximum Ho's coefficient and maximum compression ratio) can be effectively improved, thereby enhancing its overall pumping efficiency.

[0075] The maximum compression ratio and the maximum Ho coefficient are the two core performance parameters of a turbomolecular pump. The maximum compression ratio reflects the turbomolecular pump's ability to reduce gas pressure during the pumping process and is used to measure the maximum pressure ratio the pump can achieve between the inlet and outlet. A high maximum compression ratio means the pump can reduce gas pressure to a lower level in a shorter time. The maximum Ho coefficient, also known as the transmission coefficient, is an important parameter for evaluating the pumping efficiency of a turbomolecular pump. It describes the probability that a gas molecule will be successfully transferred to the next stage or outlet after colliding with the blades in the pump. The maximum Ho coefficient refers to the highest transmission coefficient the pump can achieve under all possible operating conditions. The level of this coefficient directly affects the pumping rate and ultimate pressure of the pump. The higher the maximum Ho coefficient, the more efficient the pump is in transferring gas molecules, enabling it to extract more gas molecules in the same time, thereby achieving a higher vacuum degree. The maximum compression ratio and the maximum Ho coefficient reflect the performance of a turbomolecular pump in terms of pressure reduction capability and pumping efficiency, respectively. Optimizing these two indicators is of great significance for improving the pump's structural design and overall performance and meeting the requirements of high vacuum environments.

[0076] However, the current design process for single-stage molecular pump blade rows has significant limitations. First, designers need to manually determine the blade row structure design variables based on the changing trends of the indirect blade row geometric shape parameters. This process lacks intuitiveness, resulting in low design efficiency. Second, traditional methods cannot clearly guide how to bias and balance the molecular pump performance parameters, namely the maximum compression ratio and the maximum Ho coefficient. When designing a molecular pump, how to determine the appropriate blade row structure design variables for various needs so that the molecular pump performance parameters meet the requirements remains a major challenge in the design field. It essentially constitutes a complex dual-objective optimization problem.

[0077] In view of this, the present invention integrates genetic algorithm and neural network technology to propose a new method for determining the design variable values of molecular pumps. This method aims to establish a direct relationship between the blade row structure design variables and the molecular pump performance parameters, realize the intuitive and intelligent design process, and at the same time explore and solve the bias and balance problems between the maximum compression ratio and the maximum Ho coefficient.

[0078] The method provided by the present invention is mainly used in the design stage of turbomolecular pumps, and aims to achieve precise optimization of blade row structure design variables by combining data-driven and intelligent algorithms, and solve the multivariable coupling optimization problem in the design of core performance parameters of turbomolecular pumps such as pumping rate, ultimate vacuum degree, and energy efficiency. At the same time, it can be extended to other high vacuum equipment manufacturing fields (such as thin film deposition equipment in semiconductor processes, supporting molecular pump design for electron beam lithography equipment), scientific research experimental equipment (such as particle accelerator vacuum systems, molecular pump performance adjustment of high vacuum physics experimental cavities) and precision instruments (such as mass spectrometers, scanning electron microscope vacuum pumping system optimization) and other scenarios. At the same time, it can meet the differentiated quantitative design requirements of molecular pump performance parameters in different application scenarios.

[0079] In the embodiment of the present disclosure, first, the corresponding blade row geometric shape parameters are determined through the blade row structure design variables, and a mathematical mapping relationship between the design variables and the physical structure of the turbomolecular pump is established. The design parameters are converted into quantifiable geometric features, so that the performance influencing factors are made explicit, laying a physical modeling foundation for subsequent performance prediction, and solving the problem of fuzzy correlation between structural parameters and performance in traditional design; then, the geometric shape parameters are input into a pre-trained neural network model, and the neural network's ability to fit complex nonlinear relationships is used to output predicted transmission probability parameters, replacing traditional analytical calculations based on simplified assumptions or time-consuming experimental measurements, and completing the prediction of transmission probability parameters in a multivariable coupling scenario in a short time, thereby improving analysis efficiency and reducing dependence on massive measured data; then, a fitness function is constructed based on the predicted transmission probability parameters, and a blade row structure design variable is established. The direct mapping relationship between the quantity and the performance parameters of the turbomolecular pump provides a quantitative evaluation standard for the optimization of design variables, so that the performance optimization goal is transformed from a qualitative description to a mathematically precise guidance, solving the problem of ambiguous optimization direction in traditional design; then, the global search capability of the genetic algorithm is used to generate a value population containing multiple groups of parameter combinations in the design variable space, and by simulating the selection, crossover, and mutation operations of biological evolution, potential optimal solutions are efficiently explored in complex multidimensional space, avoiding the defect of traditional gradient optimization algorithms that are prone to falling into local optimality, and expanding the search range of design variables; finally, the individuals in the population are screened according to the fitness function value, and the structural design variable combinations that meet the preset performance constraints are retained. The actual application requirements are converted into quantitative screening rules to ensure that the final value meets both performance indicators and engineering feasibility, avoiding the blindness of traditional trial and error methods.

[0080] In summary, this method takes the turbomolecular pump design stage as its core application scenario and constructs a complete process consisting of parameter mapping, intelligent prediction, and quantitative optimization. It integrates the intelligent modeling capabilities of neural networks with the global optimization capabilities of genetic algorithms, breaking through the inefficient model of traditional molecular pump design that relies on empirical formulas or extensive experiments. It has interdisciplinary integration, data-driven characteristics, and engineering practicality. Its effect is manifested in the efficient and accurate determination of design variable values in turbomolecular pump design, improving design efficiency and performance optimization. At the same time, it provides a reusable intelligent optimization methodology for molecular pump design in other high-vacuum scenarios. It is particularly suitable for the design and optimization of complex molecular pump systems with multivariable coupling and significant nonlinear characteristics.

[0081] In another embodiment provided by the present disclosure, the neural network model is trained in the following manner:

[0082] Step 1: Using the leaf row geometry parameters as input and the transmission probability parameters as output, a neural network model is constructed;

[0083] Step 2: extracting a first preset number of sample data within the value range of each leaf row geometric shape parameter to obtain a first preset number of input sample groups;

[0084] Step 3: For each input sample group, the corresponding transmission probability parameter is solved based on the Monte Carlo simulation method as the output sample group;

[0085] Step 4: Take the input sample group as input and the output sample group as true value to train the neural network model to obtain a neural network model for predicting transmission probability parameters.

[0086] In the disclosed embodiments, the neural network model training method is applied to the performance modeling scenario in the turbomolecular pump design phase. The core of the method is to construct a high-precision prediction model based on the mapping relationship between blade row geometry parameters and transmission probability parameters to support the transmission characteristics analysis in turbomolecular pump performance optimization. Specifically, the method can be implemented as follows:

[0087] First, the basic architecture of a neural network model is constructed, using the blade row geometry parameters as input variables and the molecular transmission probability parameters through the blade row structure as output variables. This provides a digital carrier for modeling the transmission characteristics of the molecular pump blade row structure, solving the problem that traditional empirical formulas have difficulty describing the nonlinear relationship between complex geometric structures and molecular motion. In implementation, the above neural network architecture can adopt a multi-layer perceptron (MLP), convolutional neural network (CNN), or graph neural network (GNN). The appropriate network structure is selected based on the spatial characteristics of the blade row geometry parameters. For example, for blade arrays with spatial topological characteristics, GNN can be used to enhance local feature extraction capabilities.

[0088] Next, within the preset value range of each leaf row's geometric shape parameter, a first preset number of sample data is used to form an input sample group containing multiple sets of geometric shape parameter combinations. This systematically covers key areas of the parameter space, ensures the statistical representativeness of the sample data, and avoids insufficient model generalization due to sample bias. During implementation, sample extraction can use methods such as random sampling or stratified sampling. Orthogonal experimental design or Latin Hypercube Sampling (LHS) techniques can also be combined to reduce the number of samples while ensuring sample uniformity. For example, for a range of n parameter values, the LHS method can be used to uniformly select k sample points in each parameter dimension, forming an n×k orthogonal sample matrix, improving the efficiency of parameter space coverage.

[0089] Next, for each input sample set, the corresponding transmission probability parameters are solved using Monte Carlo simulation methods, which serve as the output sample set for model training. The Monte Carlo method overcomes the limitations of traditional analytical methods that rely on simplified assumptions about complex boundary conditions. It can accurately generate true transmission probability values that align with actual physical processes, providing reliable supervision signals for model training.

[0090] Finally, the neural network model was iteratively trained using the input sample set and the corresponding output sample set as the training dataset, employing optimization techniques such as the backpropagation algorithm. By adjusting the model's internal parameters, the error between the model's outputted predicted transmission probability parameters and the Monte Carlo simulation results was gradually reduced, ultimately resulting in a neural network model with accurate prediction capabilities. This step achieves nonlinear mapping learning from geometric shape parameters to transmission probability parameters through a data-driven approach, replacing traditional modeling methods that rely on manual parameter adjustment or trial and error, thereby improving the efficiency and accuracy of complex system modeling.

[0091] In summary, this method combines data-driven development with physical simulation. Monte Carlo simulation generates high-quality training data, ensuring that the model is both consistent with actual physical laws and data-adaptive. A preset number of sample extraction strategies achieves systematic coverage of the range of blade row geometric parameters, improving the model's predictive reliability under all operating conditions. Leveraging the adaptive learning characteristics of neural networks, the method achieves end-to-end modeling from multivariable input to probabilistic parameter output without the need for manually defining complex mathematical expressions. A modeling framework combining simulation data generation with neural network fitting is proposed, breaking through the bottleneck of traditional molecular pump transmission characteristic analysis, which relies on simplified assumptions or expensive experimental measurements. This provides an efficient tool for rapid performance evaluation and optimization during the turbomolecular pump design phase, reducing the number of physical prototyping and experimental verifications, lowering R&D costs, and accelerating product iteration.

[0092] The present disclosure provides a specific implementation method. First, for each of the three leaf row geometric shape parameters, its value range is regarded as a uniform distribution, and M values are extracted from the uniform distribution. Each group of values is used as the input value of the sample, and there are M groups in total. Among them, the size of M is not limited. To ensure the accuracy of the subsequent neural network, during implementation, M can be as small as 50. The larger M is, the more accurate the neural network established in the subsequent steps.

[0093] From the blade inclination angle range [α min ,α max ] randomly select M values from the uniform distribution: α1, α2…α M ; From the range of the chord ratio [s min ,s max ] randomly select M values from the uniform distribution: s1, s2…s M ; From the speed ratio range [C min ,C max ] randomly select M values from the uniform distribution: C1, C2…C M .

[0094] According to the above sampling, M input sample groups are formed:

[0095]

[0096] For example, the random sampling function rand in Matlab is used to obtain M samples:

[0097]

[0098] Then, for each input sample group, the corresponding forward transmission probability and reverse transmission probability are solved based on the Monte Carlo simulation method to obtain the output sample group. Specifically, for each of the M input sample groups formed in the above steps, the forward transmission probability and reverse transmission probability are solved using the Monte Carlo simulation method, and the obtained M groups of forward transmission probabilities and reverse transmission probabilities are used as the output sample group of the above neural network model.

[0099] Specifically, for the i-th input sample group [α i , s i , C i ](1≤i≤M), use the Monte Carlo simulation method to solve the corresponding forward transmission probability M12 i and reverse transmission probability M21 i , that is, the i-th output sample group corresponding to the i-th input sample group. For a single-stage blade, the corresponding forward transmission probability M12 is solved according to the blade inclination angle α, the pitch-chord ratio s and the speed ratio C. i and reverse transmission probability M21 i The method is referred to the book "Theory and Practice of Molecular Pump Vacuum Pump" (Bade Chun, Wang Xiaodong, etc., Science Press). Thus, the input and output of the sample form a complete sample:

[0100]

[0101] For example, the sample input [α1 = 31.1080s1 = 1.5572C1 = 1.5368] is substituted into the Monte Carlo simulation method under the turbomolecular pump two-dimensional model to generate the output [M121 = 0.8354M211 = 0.1936], and the calculation takes 0.02 seconds; the sample input [α2 = 24.4652s2 = 0.7356C2 = 1.3645] is substituted into the Monte Carlo simulation method under the turbomolecular pump two-dimensional model to generate the output [M122 = 0.5867M212 = 0.0744], and the calculation takes 0.02 seconds; and so on. M =35.2895s M =1.9963C M =0.8175] is substituted into the Monte Carlo simulation method under the turbomolecular pump two-dimensional model to generate the output [M12 M =0.8133M21 M =0.4175], the calculation time is 0.02 seconds. Thus, the input and output of the sample form a complete sample:

[0102]

[0103] Finally, the obtained input sample group and output sample group are used as true values to train the above neural network model to improve the accuracy of the model.

[0104] In the embodiment of the present disclosure, the model training process, when implemented, further includes the following steps:

[0105] Step a: Divide the input sample group into a training set and a test set, wherein the training set can be implemented as 70%-80% and the test set can be implemented as 20%-30%;

[0106] Step b: setting the loss function (such as mean square error (MSE)) of the neural network model and the optimizer used for neural network training, such as Adam (Adaptive Moment Estimation) and SGD (Stochastic Gradient Descent).

[0107] Step c: Use the above input sample group and output sample group to continuously iterate and optimize the model parameters through the back propagation algorithm to minimize the loss function. After the training is completed, the model is fully tested and the loss function, R 2 and other key evaluation indicators to judge its performance;

[0108] Step d: If the above key evaluation indicators can meet the actual application requirements, the training is terminated; otherwise, if the performance indicators do not meet expectations, adjust the model architecture (such as increasing or decreasing the number of hidden layers, adjusting the number of hidden layer nodes), learning rate, activation function, or expand the size of the sample M, and continue iterative optimization until a neural network model that meets the evaluation criteria is found.

[0109] For example, a three-layer neural network is established, where the hidden layer has 20 nodes. When the number of samples is M = 200, the neural network is trained, and the training set results and test set results are as follows: Figure 2 、 Figure 3 、 Figure 4 and Figure 5 shown.

[0110] Furthermore, to verify the effectiveness of the trained neural network, the sample input [31.10801.55721.5368] was substituted into the neural network to generate M121=0.8431 and M211=0.1925, and the calculation took 0.005 seconds. For the forward and reverse transmission probabilities, the calculation errors compared with the Monte Carlo simulation method were 0.9% and 0.6%.

[0111] Substituting the sample input [24.4652 0.7356 1.3645] into the neural network, we get M122=0.5875 and M212=0.0737, and the calculation time is 0.005s. For the forward and reverse transmission probabilities, the calculation errors compared with the Monte Carlo simulation method are 0.14% and 0.94%.

[0112] Substitute the sample input [35.2895 1.9963 0.8175] into the neural network to obtain M12 M =0.8112 and M21 M =0.4153, the calculation time is 0.005s; for the forward and reverse transmission probabilities, the calculation errors compared with the Monte Carlo simulation method are: 0.26% and 0.53%.

[0113] In the above embodiments, the computing environment of each sample is: processor 11th Gen Intel(R) Core(TM) i5-1135G7@2.40GHz, machine RAM 2.42GHz 16.0GB (15.8GB available).

[0114] In another embodiment provided by the present disclosure, the neural network model includes an input layer, a hidden layer and an output layer; the number of nodes in the input layer is the same as the number of types of leaf column geometric shape parameters, and the output layer includes two nodes corresponding to the forward transmission probability and the reverse transmission probability, respectively.

[0115] In the disclosed embodiments, the initial process of constructing a neural network model can be implemented as follows: First, the network structure is designed, and the number of input layer nodes is determined (3, corresponding to the blade pitch angle α, pitch-chord ratio s, and speed ratio C). This ensures that the full-dimensional geometric shape characteristics of the blade row structure can be fully captured and input into the network, solving the problem of information omission in multi-parameter input scenarios. The number of hidden layers and the number of neurons per layer are set. The number of hidden layers can be implemented as 1-3, and the number of neurons per hidden layer is at least 2. This structure can automatically extract the coupling characteristics between parameters, breaking through the limitations of traditional analytical models that rely on simplified assumptions. During implementation, the hidden layers can adopt the residual network (ResNet) or long short-term memory (LSTM) architecture. The former solves the gradient vanishing problem in deep network training, while the latter is suitable for processing geometric parameters with serial correlation (such as the level parameters of multi-level blade rows). The number of output layer nodes is determined (2, corresponding to the forward transmission probability M12 and the reverse transmission probability M21). By decoupling the probability distribution of the transmission direction, a more refined control basis is provided for molecular pump performance optimization. In addition, an activation function (such as ReLU, Sigmoid, etc.) is selected for the hidden layer. In summary, the present invention proposes a customized neural network architecture for the bidirectional characteristics of the molecular pump transmission process, breaking through the fuzzy characterization of the transmission direction by the traditional single output model.

[0116] In another embodiment provided by the present disclosure, the above step S103, using the predicted transmission probability parameters to construct a fitness function with the blade row structure design variables as input and the molecular pump performance parameters as output, can be implemented as follows:

[0117] Step 1: Divide the blades of the molecular pump corresponding to the values of the blade row structure design variables into a second preset number of segments along the radial direction;

[0118] Step 2: For each segment, determine the corresponding leaf row geometric parameters according to the segmented leaf row structural design variables;

[0119] Step 3: Normalize the leaf row geometric shape parameters corresponding to the second preset value segments by type and input them into the neural network model to obtain transmission probability parameters corresponding to the leaf row geometric shape parameters;

[0120] Step 4: Integrate the transmission probability parameters to obtain the corresponding molecular pump performance parameters.

[0121] In the disclosed embodiment, when constructing a fitness function with the blade row structure design variables as independent variables and the molecular pump performance parameters as dependent variables, it is necessary to follow the numerical relationship between the transmission probability parameters and the molecular pump performance parameters.

[0122] First, the blades corresponding to the blade row structure design variables are radially divided into a second preset number of segments, for example, 10-20 segments based on the blade length or curvature variation characteristics. Discretization is used to convert the continuous blade structure into independently analyzable units, making it easier to capture the differentiated effects of geometric parameters at different radial positions on the transmission characteristics. During implementation, the number of segments can be dynamically adjusted based on the complexity of the blades. A denser segmentation approach, such as locally increasing the number of micro-segments to 50, can be used in blade tip areas where curvature varies dramatically to improve modeling accuracy in key areas.

[0123] Next, for each segment, the corresponding blade row geometry parameters are calculated based on the blade row structural design variables. Parameter mapping is used to decouple the design variables from the global to the local level. For continuously varying design variables, such as radially varying blade pitch angles, cubic spline interpolation or polynomial fitting methods are used to calculate the segmented parameter values. This ensures a smooth transition between geometric features of adjacent segments and avoids physical discontinuities caused by discretization.

[0124] The leaf array geometric parameters of each segment are then normalized by type, for example, scaling them to the [0, 1] interval or standardizing them to zero mean and unit variance to eliminate the interference of dimensional differences on the neural network input. These parameters are then input into the model to obtain the transmission probability parameters of each segment. During implementation, the normalization method can be selected based on the characteristics of the parameter distribution. Z-score normalization is used for parameters that follow a normal distribution, while min-max normalization is used for parameters that follow a uniform distribution. Principal component analysis can also be used to reduce the dimensionality of highly correlated geometric parameters, reducing the input dimension and improving model training efficiency.

[0125] Finally, the transmission probability parameters of each segment are integrated to obtain the molecular pump performance parameters. The formula used in the integration operation is given by the prior art and will not be repeated here.

[0126] Overall, the above method constructs a mapping relationship between blade row structure design variables and molecular pump performance parameters through the steps of discrete segmentation, local modeling, and global integration. It features segmented refined modeling, multi-dimensional parameter normalization, and integral performance synthesis. It breaks through the traditional overall modeling's neglect of radial parameter differences, improves the standardization of neural network input and the model's generalization ability, and establishes a quantitative correlation from microstructure to macroscopic properties. It solves the problem of inefficiency in traditional methods for evaluating the performance of complex blade row structures, provides an efficient quantitative tool for the refined design of turbomolecular pumps, and is particularly suitable for multi-objective optimization scenarios of variable parameter blade row structures, which can significantly reduce the number of physical experiments and shorten the R&D cycle.

[0127] In a specific embodiment, the normalized calculation formula can be: X'=(XX min ) / (Xmax -X min ), X is the blade inclination angle α or the pitch-chord ratio s or the speed ratio C.

[0128] Specifically, establishing the fitness function includes the following steps:

[0129] Step a: The independent variables of the fitness function are the blade row structure design variables of the molecular pump: tooth root diameter d, tooth tip diameter D, blade height h, blade thickness t, blade root inclination angle α root , blade tip inclination angle α tip and the number of leaves z;

[0130] Step b: segment the blade along the radial direction into N segments of a second preset value, and the radius of each segment is r1, r2…r N , the corresponding blade inclination angle of each segment is α1, α2…α N .

[0131] Step c: Calculate s for each segment in N segments i and C i : (1≤i≤N), (1≤i≤N).

[0132] Step d: [α i ,s i ,C i ]After normalization, it is substituted into the neural network model to obtain M12 i and M21 i .

[0133] Step e: M12 i and M21 i Use the integral formula to calculate the maximum compression ratio K and the maximum Ho coefficient H of the turbine blade row. The integral formula is referenced from the book "Theory and Practice of Molecular Pumps and Vacuum Pumps" (written by Badchun, Wang Xiaodong, etc., Science Press).

[0134] For example, given d = 90, D = 150, h = 10, t = 1.2, α root =30,α tip =40, z=30, M12 and M21 are divided into 11 discrete points along the radial direction as shown in Table 1:

[0135] Table 1 Calculation results of forward and reverse transmission probabilities at 11 discrete points

[0136]

[0137]

[0138] By integrating M12 and M21 of the above 11 discrete points, we obtain the maximum compression ratio K = 3.24 and the maximum Ho coefficient H = 0.22.

[0139] In another embodiment provided by the present disclosure, the above-mentioned step S105, selecting an individual that meets the preset requirements according to the fitness function value of each individual in the value population to obtain the target value of the leaf row structure design variable, can be implemented as follows:

[0140] Step 1: Generate a Pareto front based on the blade row structure design variables and molecular pump performance parameters corresponding to each individual in the value population;

[0141] Step 2: Select a point on the Pareto front that meets the preset requirements, and determine the value of the leaf row structure design variable corresponding to the point as the target value.

[0142] In the embodiments of the present disclosure, the optimization problem of conflicts among multiple performance parameters of the molecular pump is mainly solved.

[0143] First, a multi-objective optimization algorithm is used to generate the Pareto front based on the blade row design variables and molecular pump performance parameters corresponding to each individual in the value population. The Pareto front encompasses all non-dominated solutions and clearly demonstrates the trade-offs between design variables in the multidimensional performance space. During implementation, mainstream multi-objective optimization techniques such as the non-dominated sorting genetic algorithm and the multi-objective particle swarm optimization algorithm can be used to generate the Pareto front.

[0144] Next, a point that meets the preset requirements is selected on the Pareto front. The preset requirements can be flexibly set in combination with the actual application scenario to meet the design requirements of multiple scenarios.

[0145] In summary, by generating the Pareto front, the potential solution space of design variables in multi-objective optimization is fully revealed, avoiding the traditional single-objective optimization from falling into local optimality due to ignoring the coupling relationship between performance indicators. At the same time, the actual engineering needs are converted into specific screening rules to ensure that the final design variable values ​​not only meet the performance constraints but also meet the priority preferences of actual applications, thereby improving the practicality of the optimization results. The Pareto optimization theory is introduced into the field of molecular pump design, and a design variable screening framework based on a multi-objective evolutionary algorithm is constructed, which solves the inefficiency problem of traditional methods relying on empirical parameter adjustment in multi-index optimization. This method can provide multiple sets of high-performance candidate solutions for turbomolecular pump design, reduce design rework, support customized design, and significantly improve the optimization efficiency and design flexibility of complex systems.

[0146] In another embodiment provided by the present disclosure, the above step S104, using a genetic algorithm to generate a value population for the leaf row structure design variables, can be implemented as follows:

[0147] Step 1: Randomly generate the initial population according to the value range of each leaf row structure design variable;

[0148] Step 2: Use the fitness function to calculate the molecular pump performance parameters corresponding to each individual in the initial population;

[0149] Step 3: Based on non-dominated quick sorting and crowding calculation, arrange the individuals in the initial population in order from best to worst;

[0150] Perform iterative operations on the current population;

[0151] The steps of the iterative operation include:

[0152] Step 4: Select a preset number of individuals from the current group in order, generate new individuals using a genetic algorithm for the selected individuals, and calculate the molecular pump performance parameters corresponding to the new individuals using a fitness function;

[0153] Step 5: Filter the population set based on non-dominated quick sort and crowding calculation to generate a new current population with the same number of individuals as the current population, where the population set is the combination of the current population and the new individuals;

[0154] Step 6: Determine whether the current population meets the preset convergence conditions;

[0155] Step 7: If not satisfied, return to the step of performing iterative operations on the current population;

[0156] Step 8: If satisfied, the current population is determined as the set of all feasible solutions.

[0157] In this disclosed embodiment, an initial population is first randomly generated based on the range of values for each leaf row structure design variable. This operation is intended to provide a diverse starting point for subsequent optimization, enabling the algorithm to search for the optimal solution within a wider search space. During implementation, methods such as Latin hypercube sampling can be used to generate the initial population to improve its representativeness and avoid the uneven distribution issues that may arise from random sampling.

[0158] Subsequently, the fitness function is used to calculate the molecular pump performance parameters corresponding to each individual in the initial population.

[0159] Next, based on non-dominated quick sort and crowding calculation, individuals in the initial population are sorted from best to worst. This allows for rapid screening of relatively high-performing individuals while preventing over-concentration of individuals in a particular area. For example, in a two-dimensional performance space, individuals are sorted based on the maximum compression ratio and the maximum Ho coefficient, placing those with better performance and fewer surrounding individuals at the top.

[0160] An iterative operation is then performed on the current population. A preset number of individuals are sequentially selected from the current population, and new individuals are generated using a genetic algorithm. This process simulates the inheritance and mutation mechanisms in biological evolution, increasing the diversity and evolutionary potential of the population. Mutation operations can also be used to make small changes to specific variables. During the genetic algorithm's generation of new individuals, the crossover and mutation probabilities can be adjusted to balance global and local search capabilities. To expand the search range, the mutation probability can be appropriately increased; to refine the search near an existing optimal solution, the mutation probability can be reduced.

[0161] After generating a new individual, the fitness function is again applied to calculate the corresponding molecular pump performance parameters for subsequent evaluation. The population set (the combination of the current population and the new individual) is then screened based on non-dominated quick sorting and crowding calculation to generate a new current population with the same number of individuals as the current population. This ensures the stability of the population size while continuously updating the quality of the individuals in the population.

[0162] Then determine whether the current population meets the preset convergence conditions. If not, continue iterating. If so, determine the current population as the set of all feasible solutions.

[0163] In summary, the aforementioned steps give the method provided by the present invention strong global search capabilities and the ability to handle multi-objective optimization problems. By applying a genetic algorithm to optimize blade row design variables, the iterative process allows the population to continuously evolve, ultimately finding a more optimal solution. This method can efficiently identify suitable combinations of blade row design variables within a complex design space, improving turbomolecular pump performance while reducing design time and cost, providing an effective solution for the design optimization of related mechanical systems.

[0164] The specific process of iterative operation of the population in steps 4-8 above can be implemented as follows:

[0165] Step a: Select the top-ranked individual from the current population, i.e., the one with the highest fitness, as the parent generation, perform genetic recombination on the parent generation individual to generate a new offspring individual, and use the fitness function to calculate the molecular pump performance parameters of the new offspring individual;

[0166] Step c, merging the above new offspring individuals with the current population;

[0167] Step d: Sort the individuals in the merged population based on non-dominated quick sort and crowding calculation, determine the fitness of each individual according to the molecular pump performance parameters corresponding to each individual in the current population, and select individuals with higher fitness to form a new population;

[0168] Step e: determine whether the preset convergence condition is currently met. If not, return to step a. If so, determine the current population as the set of all feasible solutions.

[0169] Among them, the above step a, by calling the selection function, selects the individuals with higher ranking, that is, higher fitness, from the current population as the parent; by calling the crossover function and the mutation function, the parent individual is genetically recombined to generate new offspring individuals, thereby increasing the diversity of the population; the preset convergence condition in the above step e can be implemented as reaching the set number of iterations, etc., which is not limited here.

[0170] In another embodiment provided by the present disclosure, the fitness function value includes: molecular pump performance parameters; the molecular pump performance parameters include: maximum compression ratio and maximum Ho coefficient;

[0171] The above step 2, selecting a point on the Pareto front that meets the preset requirements and determining the value of the leaf row structure design variable corresponding to the point as the target value, can be implemented as follows:

[0172] If the maximum compression ratio is the primary requirement, then the point with the highest maximum compression ratio is selected on the Pareto front, and the value of the leaf row structure design variable corresponding to the point is determined as the target value;

[0173] If the maximum Ho coefficient is the primary requirement, then the point with the highest maximum Ho coefficient is selected on the Pareto front, and the value of the leaf row structure design variable corresponding to the point is determined as the target value;

[0174] If the demands for the maximum compression ratio and the maximum Ho coefficient are balanced, the point with the smallest difference between the maximum compression ratio and the maximum Ho coefficient is selected on the Pareto front, and the value of the leaf row structure design variable corresponding to the point is determined as the target value.

[0175] In the disclosed embodiments, a Pareto front has been established, including multiple sets of solutions. During the optimization process, two solutions in each set exhibit a mutually exclusive characteristic: it is impossible to improve the value of one objective function without sacrificing the performance of the other. The Pareto front intuitively demonstrates the balance between the optimization objective and the constraints.

[0176] When selecting a specific solution (i.e., a specific leaf row structural design variable) on the Pareto front, the decision maker's actual needs and preferences can be considered. If the decision maker is highly concerned with the maximum Ho's coefficient, they may tend to choose points on the frontier that are biased to the left (assuming the vertical axis represents the maximum Ho's coefficient) because these points provide a higher maximum Ho's coefficient, even if this may be accompanied by a decrease in the maximum compression ratio. Conversely, if the decision maker places greater emphasis on the maximum compression ratio, they may choose points on the lower right side of the frontier (assuming the horizontal axis represents the maximum compression ratio) because these points optimize the maximum compression ratio. If both the compression ratio and the maximum Ho's coefficient are taken into consideration, then points in the middle of the Pareto front are selected, which ensures that the structural design variable can achieve a good maximum compression ratio and maximum Ho's coefficient at the same time.

[0177] Therefore, the construction and analysis of the Pareto front not only provides decision makers with a comprehensive perspective for understanding the optimization problem, but also enables them to flexibly select the most appropriate design variables for the single-stage leaf row structure according to actual needs.

[0178] For example, given the parameters of tooth tip diameter D = 190 mm, tooth root diameter d = 90 mm, blade thickness t = 1.76 mm, blade height h = 12 mm, number of blades z = 20, and the range of blade tip inclination angle α tip ∈(30°,40°), the range of blade root inclination angle α root Under the condition of ∈(30°,40°), the optimization method of the present invention can be used to obtain the following Figure 6 、 Figure 7 and Figure 8 The Pareto front is shown.

[0179] like Figure 6 As shown, the rightmost design point in the figure corresponds to α root =30°,α tip =30°, at this time the maximum compression ratio K reaches the maximum value of 3.3753, and the maximum Ho coefficient H reaches the minimum value of 0.4002, which is Figure 8 The design point on the far left corresponds to α root =40°,α tip =40°, at this time the maximum Ho coefficient H reaches the maximum value of 0.51889, and the maximum compression ratio K reaches the minimum value of 2.3144, which is Figure 8 Design point 1 in .

[0180] In order to consider the optimal value of the maximum compression ratio and the maximum Ho coefficient at the same time, you can choose Figure 8 Several data points in the middle, such as design points 2, 3, 4, and 5, are included. From a practical application perspective, design point 2 achieves a significantly higher Ho's coefficient than design point 3 at the expense of a slight decrease in the maximum compression ratio, so design point 2 is superior to design point 3. Similarly, design point 5 achieves a higher maximum compression ratio than design point 4 at the expense of a slight decrease in the maximum Ho's coefficient, so design point 5 is superior to design point 4.

[0181] In summary, in order to balance the maximum compression ratio and the maximum Ho's coefficient, it is more reasonable to choose design points 2 and 5. Furthermore, if the designer focuses more on maximizing the maximum Ho's coefficient, then No. 2 can be selected between No. 2 and No. 5; conversely, if the designer focuses more on maximizing the maximum compression ratio, then No. 5 can be selected between No. 2 and No. 5. Among them, the blade inclination angle corresponding to design point 2 is α root =31°,α tip=40°, the blade inclination angle corresponding to design point 5 is α root =30°,α tip =40°.

[0182] In another embodiment provided by the present disclosure, the blade row structure design variables include root diameter, tip diameter, blade height, blade thickness, blade root inclination angle, blade tip inclination angle and number of blades; the blade row geometric shape parameters include blade inclination angle, speed ratio and pitch chord ratio;

[0183] The blade row geometry parameters are determined based on the blade row structure design variables of the molecular pump in the following manner:

[0184] determining a blade inclination angle based on a blade tip inclination angle and a blade root inclination angle;

[0185] Determine the speed ratio based on the tooth tip diameter and tooth root diameter;

[0186] The pitch-chord ratio is determined based on the tooth root diameter, tooth tip diameter, blade height, blade thickness, blade root inclination angle, blade tip inclination angle, and number of blades.

[0187] In the embodiment of the present disclosure, the physical meaning of each blade row structure design variable can be referred to as follows: Figure 9 A two-dimensional model of a molecular pump is shown.

[0188] In the disclosed embodiment, based on the above, in order to facilitate the generation of sample data required for neural network model training, it is necessary to first determine the value ranges of the blade row geometric shape parameters. Before determining the blade row geometric shape parameters and their corresponding value ranges, it is first necessary to obtain the blade row structural design variables and their corresponding value ranges. The value ranges of these variables must comprehensively consider the application environment and economic efficiency of the molecular pump:

[0189] For applications with limited space, the range of tooth root diameter d and tooth top diameter D should be narrowed;

[0190] If the built-in motor of the turbomolecular pump is large, the value range of the tooth root diameter d and the tooth top diameter D needs to be increased accordingly;

[0191] In order to effectively control manufacturing costs, the condition α can be set root =α tip , that is, the inclination angle of the blade from the root to the top of the tooth remains unchanged.

[0192] In summary, the specific values of the design variables need to be flexibly adjusted to adapt to different molecular pump usage needs and cost control requirements.

[0193] For example, the value range of D can be set to [100mm, 150mm], the value range of d can be set to [80mm, 100mm], the range of t can be set to [1mm, 1.2mm], the range of h can be set to [8mm, 10mm], the range of z can be set to [15, 40], and α root The range can be set to [10°, 40°], α tip The range can be set to [20°,40°].

[0194] In turbomolecular pump theory, for single-stage blades, the blade geometry is often characterized by the blade inclination angle α, the pitch-chord ratio s, and the dimensionless velocity ratio C. The blade row geometry determines the pumping performance of the blade row, which in turn affects the performance parameters of the turbomolecular pump. The blade row geometry is determined by the blade row structural design variables. Therefore, the blade row geometry and its corresponding value range are determined based on these blade row structural design variables and their corresponding value ranges.

[0195] The range of blade inclination angle α is defined as the range from blade root inclination angle α root The minimum value of the blade tip inclination angle α tip The maximum value of α min =α root The minimum value of α max =α tip The maximum value of .

[0196] The value range of speed ratio C [C min ,C max ] is obtained by the following first and second formulas or first and third formulas:

[0197] By the first formula Calculate the velocity ratio C, where C represents the velocity ratio, v represents the blade row linear velocity, and u represents the molecular thermal motion velocity; and the molecular thermal motion velocity u is calculated using the second or third formula;

[0198] If the Monte Carlo simulation method under the turbomolecular pump two-dimensional model is used to generate samples, then according to the second formula Calculate the molecular thermal motion velocity u;

[0199] If the Monte Carlo simulation method under the turbomolecular pump three-dimensional model is used to generate samples, then according to the third formula: Calculate the molecular thermal motion velocity u;

[0200] Among them, u represents the molecular thermal motion speed, R represents the gas constant, T represents the temperature, and M represents the gas molecular modulus.

[0201] For a specific gas and working conditions, T and M remain constant, so the speed ratio C depends mainly on the linear velocity v of the blade row. Since v is proportional to the angular velocity w (the constant rotation speed of the molecular pump) and the radius r (v = w r), the maximum value of C occurs at the maximum radius D max C max =w·D max ; The minimum value occurs at the minimum radius d min C min =w·D min Among them, D max is the maximum value of the tooth tip diameter D of the structure blade row design variable within the set value range, d min It is the minimum value of the tooth root diameter d within the set value range.

[0202] Therefore, when r = d min When The minimum value of C is obtained; when r=D max When The maximum value of C is obtained.

[0203] The pitch-chord ratio s is calculated using the following fourth formula:

[0204]

[0205] Among them, s represents the pitch-chord ratio, a represents the pitch length, b represents the chord length, r represents the blade radius, z represents the number of blades, t represents the blade thickness, α represents the blade inclination angle, and h represents the blade height.

[0206] To minimize s = a / b, a must be minimized and b must be maximized. This requires minimizing r, maximizing z, and maximizing t, while also minimizing α to minimize the numerator. At the same time, h must be maximized and α again minimized to maximize the denominator. If the minimum value of s is less than 0, the minimum value of s starts at 0. Conversely, to maximize s, a must be maximized and b must be minimized. This requires maximizing r, minimizing z, and minimizing t, while also maximizing α to maximize the numerator. At the same time, h must be minimized and α again maximized to minimize the denominator.

[0207] Therefore, when r = d min 、z=z max , t=t max 、α=α min , h=h max When s reaches its minimum value s min When r=D max 、z=z min , t=t min 、α=α max , h=h minWhen s reaches its maximum value s max .

[0208] For example, if D∈[100mm,150mm],d∈[80mm,100mm],t∈[1mm,1.2mm],h∈[8mm,10mm],z∈[15,40], α root ∈[10°,30°],α tip ∈[20°,40°]. Then, according to the above method:

[0209] First, the blade inclination angle α ranges from [10°, 40°];

[0210] Secondly, further, given the operating temperature T = 293.15K, the speed w is 60000r / min = 6283.2 rad / s, and the gas molecular weight M = 28g / mol, then if the Monte Carlo simulation method under the turbomolecular pump two-dimensional model is used to generate samples, then Furthermore, when r = d min =40, the minimum value of C min =0.04*6283.2 / 295=0.85, the maximum value of C max =0.075*6283.2 / 295=1.3.

[0211] Therefore, the speed ratio C ranges from [0.85, 1.6].

[0212] Again, when r = 40, z = 40, t = 1.2, α = 10, h = 10, s min =-0.0111;

[0213] When r=75,z=15,t=1,α=40,h=8,s max =3.2.

[0214] Therefore, the range of the chord ratio s is (0,3.2], because s min =-0.0111 is less than 0, so the minimum value of s starts from 0 and cannot be equal to 0.

[0215] Based on the same disclosed concept, the embodiments of the present disclosure also provide a device for determining the design variable values of a molecular pump. Since the principles of the problems solved by these devices are similar to those of the aforementioned methods, the implementation of the device can refer to the implementation of the aforementioned methods, and the repeated parts will not be repeated.

[0216] With the above Figure 1 Corresponding to the method shown, the embodiment of the present disclosure also provides a device for determining the value of the design variable of a molecular pump, such as Figure 10 As shown, including:

[0217] A determination module 1001 is configured to determine blade row geometry parameters based on blade row structural design variables of the molecular pump; wherein the blade row geometry parameters are used to determine molecular pump performance; and the blade row geometry parameters are characterized by the blade row structural design variables;

[0218] Prediction module 1002, for inputting leaf row geometry parameters into a neural network model to obtain predicted transmission probability parameters output by the neural network model;

[0219] A construction module 1003 is used to construct a fitness function using the predicted transmission probability parameters, with the blade row structure design variables as input and the molecular pump performance parameters as output;

[0220] A generation module 1004 is used to generate a value population for the leaf row structure design variables using a genetic algorithm;

[0221] The selection module 1005 is used to select individuals that meet the preset requirements based on the fitness function value of each individual in the value population to obtain the target value of the leaf row structure design variable; wherein the fitness function value is used to characterize the performance parameter value of the molecular pump, and the preset requirement condition characterizes the limitation of the performance parameter value of different molecular pumps.

[0222] In another embodiment provided by the present disclosure, the neural network model in the prediction module 1002 is trained in the following manner:

[0223] A neural network model is constructed with leaf row geometry parameters as input and transmission probability parameters as output;

[0224] Sampling a first preset number of sample data within the value range of each leaf row geometric shape parameter to obtain a first preset number of input sample groups;

[0225] For each input sample group, the corresponding transmission probability parameter is solved based on the Monte Carlo simulation method as the output sample group;

[0226] The input sample group is taken as input and the output sample group is taken as the true value to train the neural network model, thereby obtaining a neural network model for predicting transmission probability parameters.

[0227] In another embodiment provided by the present disclosure, the above-mentioned neural network model includes an input layer, a hidden layer and an output layer; the number of nodes in the input layer is the same as the number of types of leaf column geometric shape parameters, and the output layer includes two nodes corresponding to the forward transmission probability and the reverse transmission probability, respectively.

[0228] In another embodiment provided by the present disclosure, the above-mentioned construction module 1003 is used to radially divide the blades of the molecular pump corresponding to the values of the blade row structure design variables into a second preset value of segments; for each segment, the corresponding blade row geometric shape parameters are determined according to the blade row structure design variables of the segment; the blade row geometric shape parameters corresponding to the second preset value of segments are normalized by type and then input into the neural network model to obtain transmission probability parameters corresponding to the blade row geometric shape parameters; the transmission probability parameters are integrated to obtain corresponding molecular pump performance parameters.

[0229] In another embodiment provided by the present disclosure, the above-mentioned selection module 1005 is used to generate a Pareto front based on the leaf row structure design variables and molecular pump performance parameters corresponding to each individual in the value population; select a point on the Pareto front that meets the preset demand conditions, and determine the value of the leaf row structure design variable corresponding to the point as the target value.

[0230] In another embodiment provided by the present disclosure, the generation module 1004 is configured to randomly generate an initial population based on a value range of each leaf row structure design variable; calculate a molecular pump performance parameter corresponding to each individual in the initial population using the fitness function; and sort the individuals in the initial population in descending order based on non-dominated quick sorting and crowding calculation;

[0231] Perform iterative operations on the current population;

[0232] The iterative operation steps include:

[0233] Selecting a preset number of individuals from the current group in sequence, generating new individuals using a genetic algorithm for the selected individuals, and calculating molecular pump performance parameters corresponding to the new individuals using the fitness function;

[0234] The population set is screened based on non-dominated quick sorting and crowding calculation to generate a new current population with the same number of individuals as the current population, wherein the population set is the union of the current population and the new individuals;

[0235] Determine whether the current population meets the preset convergence conditions;

[0236] If not satisfied, return to the step of performing the iterative operation on the current population;

[0237] If satisfied, the current population is determined as the set of all feasible solutions.

[0238] In another embodiment provided by the present disclosure, the fitness function value includes: molecular pump performance parameters; molecular pump performance parameters include: maximum compression ratio and maximum Ho coefficient; the selection module 1005 is used to

[0239] If the maximum compression ratio is the primary requirement, then the point with the highest maximum compression ratio is selected on the Pareto front, and the value of the leaf row structure design variable corresponding to the point is determined as the target value;

[0240] If the maximum Ho coefficient is the primary requirement, then the point with the highest maximum Ho coefficient is selected on the Pareto front, and the value of the leaf row structure design variable corresponding to the point is determined as the target value;

[0241] If the requirements for the maximum compression ratio and the maximum Ho coefficient are balanced, a point on the Pareto front with the smallest difference between the maximum compression ratio and the maximum Ho coefficient is selected, and the value of the leaf row structure design variable corresponding to the point is determined as the target value.

[0242] In another embodiment provided by the present disclosure, the blade row structure design variables include root diameter, tip diameter, blade height, blade thickness, blade root inclination angle, blade tip inclination angle, and number of blades; the blade row geometric shape parameters include blade inclination angle, speed ratio, and pitch-chord ratio;

[0243] The determination module 1001 is configured to determine the blade row geometric parameters based on the blade row structural design variables of the molecular pump in the following manner:

[0244] determining a blade inclination angle based on a blade tip inclination angle and a blade root inclination angle;

[0245] Determine the speed ratio based on the tooth tip diameter and tooth root diameter;

[0246] The pitch-chord ratio is determined based on the tooth root diameter, tooth tip diameter, blade height, blade thickness, blade root inclination angle, blade tip inclination angle, and number of blades.

[0247] An embodiment of the present disclosure provides a molecular pump, wherein the blade row structure design variables of the molecular pump are determined by a method for determining the design variable values of a molecular pump provided by any embodiment of the present disclosure.

[0248] An embodiment of the present disclosure provides a computer-readable storage medium having a computer program stored thereon. When executed by a processor, the computer program executes the steps of a method for determining design variable values for a molecular pump provided in any embodiment of the present disclosure. The storage medium may be a volatile or non-volatile computer-readable storage medium.

[0249] Through the description of the above embodiments, those skilled in the art can clearly understand that the embodiments of the present disclosure can be implemented through hardware or by means of software plus a necessary general hardware platform. Based on this understanding, the technical solutions of the embodiments of the present disclosure can be embodied in the form of a software product, which can be stored in a non-volatile storage medium (which can be a CD-ROM, a USB flash drive, a mobile hard disk, etc.), including a number of instructions for enabling a computer device (which can be a personal computer, a server, or a network device, etc.) to execute the methods described in the various embodiments of the present disclosure.

[0250] Those skilled in the art will understand that the accompanying drawings are merely schematic diagrams of a preferred embodiment, and the modules or processes in the accompanying drawings are not necessarily required for implementing the present disclosure.

[0251] Those skilled in the art will appreciate that the modules in the devices of the embodiments may be distributed in the devices of the embodiments as described in the embodiments, or may be located in one or more devices different from the embodiments with corresponding changes. The modules of the above embodiments may be combined into one module or further split into multiple submodules.

[0252] The serial numbers of the above-mentioned embodiments of the present disclosure are for description only and do not represent the advantages or disadvantages of the embodiments.

[0253] Obviously, those skilled in the art may make various changes and modifications to the present disclosure without departing from the spirit and scope of the present disclosure. Thus, if these modifications and variations of the present disclosure fall within the scope of the claims of the present disclosure and their equivalents, the present disclosure is intended to include these modifications and variations.

Claims

1. A method for determining the value of a molecular pump design variable, characterized in that: include: Determining blade row geometric parameters according to blade row structural design variables of the molecular pump; wherein the blade row geometric parameters are used to determine molecular pump performance, and the blade row geometric parameters are characterized by the blade row structural design variables; Inputting the leaf row geometry parameters into a neural network model to obtain a predicted transmission probability parameter output by the neural network model; Using the predicted transmission probability parameters, a fitness function is constructed with the blade row structure design variables as input and the molecular pump performance parameters as output; Generate a value population for the leaf row structure design variables using a genetic algorithm; An individual that meets the preset requirement conditions is selected based on the fitness function value of each individual in the value population to obtain the target value of the leaf row structure design variable; wherein the fitness function value is used to characterize the performance parameter value of the molecular pump, and the preset requirement condition is a limitation on the performance parameter values of different molecular pumps.

2. The method according to claim 1, wherein The neural network model is trained in the following way: Constructing a neural network model with the leaf row geometry parameters as input and the transmission probability parameters as output; Extracting a first preset number of sample data within the value range of each leaf row geometric shape parameter to obtain the first preset number of input sample groups; For each input sample group, the corresponding transmission probability parameter is solved based on the Monte Carlo simulation method as the output sample group; The input sample group is used as input and the output sample group is used as a true value to train the neural network model to obtain the neural network model for predicting transmission probability parameters.

3. The method according to claim 2, wherein The neural network model includes an input layer, a hidden layer and an output layer; the number of nodes in the input layer is the same as the number of types of leaf column geometric shape parameters, and the output layer includes two nodes corresponding to the forward transmission probability and the reverse transmission probability respectively.

4. The method according to claim 1, wherein The predicted transmission probability parameters are used to construct a fitness function with the blade row structure design variables as input and the molecular pump performance parameters as output, including: Dividing the blades of the molecular pump corresponding to the values of the blade row structure design variables into a second preset number of segments along the radial direction; For each segment, determine the corresponding leaf row geometric parameters according to the segmented leaf row structural design variables; Normalizing the leaf row geometric shape parameters corresponding to the second preset value segments by type and inputting the normalized parameters into the neural network model to obtain transmission probability parameters corresponding to the leaf row geometric shape parameters; An integration operation is performed on the transmission probability parameter to obtain a corresponding molecular pump performance parameter.

5. The method according to claim 1, wherein The selecting of an individual that meets preset requirements based on the fitness function value of each individual in the value population to obtain a target value of the leaf row structure design variable includes: Generate a Pareto front based on the blade row structure design variables and molecular pump performance parameters corresponding to each individual in the value population; A point that meets a preset requirement condition is selected on the Pareto front, and the value of the leaf row structure design variable corresponding to the point is determined as the target value.

6. The method according to claim 1, wherein The method of using a genetic algorithm to generate a value population for the leaf row structure design variables includes: According to the value range of each leaf row structure design variable, the initial population is randomly generated; Calculate the molecular pump performance parameters corresponding to each individual in the initial population using the fitness function; Arrange the individuals in the initial population in descending order based on non-dominated quick sorting and crowding calculation; Perform iterative operations on the current population; The iterative operation steps include: Selecting a preset number of individuals from the current group in sequence, generating new individuals using a genetic algorithm for the selected individuals, and calculating molecular pump performance parameters corresponding to the new individuals using the fitness function; The population set is screened based on non-dominated quick sorting and crowding calculation to generate a new current population with the same number of individuals as the current population, wherein the population set is the union of the current population and the new individuals; Determine whether the current population meets the preset convergence conditions; If not satisfied, return to the step of performing the iterative operation on the current population; If satisfied, the current population is determined as the set of all feasible solutions.

7. The method according to claim 5, wherein The fitness function value includes: molecular pump performance parameters; the molecular pump performance parameters include: maximum compression ratio and maximum Ho coefficient; The step of selecting a point on the Pareto front that meets a preset requirement and determining the value of the leaf row structure design variable corresponding to the point as a target value includes: If the maximum compression ratio is the primary requirement, then the point with the highest maximum compression ratio is selected on the Pareto front, and the value of the leaf row structure design variable corresponding to the point is determined as the target value; If the maximum Ho coefficient is the primary requirement, then the point with the highest maximum Ho coefficient is selected on the Pareto front, and the value of the leaf row structure design variable corresponding to the point is determined as the target value; If the requirements for the maximum compression ratio and the maximum Ho coefficient are balanced, a point on the Pareto front with the smallest difference between the maximum compression ratio and the maximum Ho coefficient is selected, and the value of the leaf row structure design variable corresponding to the point is determined as the target value.

8. The method according to claim 1, wherein The blade row structure design variables include root diameter, tip diameter, blade height, blade thickness, blade root inclination angle, blade tip inclination angle and number of blades; the blade row geometric shape parameters include blade inclination angle, speed ratio and pitch chord ratio; The blade row geometry parameters are determined based on the blade row structure design variables of the molecular pump in the following manner: determining the blade inclination angle based on the blade tip inclination angle and the blade root inclination angle; determining the speed ratio based on the tooth tip diameter and the tooth root diameter; The pitch-chord ratio is determined based on the tooth root diameter, tooth tip diameter, blade height, blade thickness, blade root inclination angle, blade tooth tip inclination angle, and the number of blades.

9. A device for determining the value of a design variable of a molecular pump, characterized in that: include: a determination module, configured to determine blade row geometric parameters according to blade row structural design variables of the molecular pump; wherein the blade row geometric parameters are used to determine molecular pump performance; and the blade row geometric parameters are characterized by the blade row structural design variables; A prediction module, configured to input the leaf row geometry parameters into a neural network model to obtain a predicted transmission probability parameter output by the neural network model; A construction module, configured to use the predicted transmission probability parameters to construct a fitness function having the blade row structure design variables as input and the molecular pump performance parameters as output; A generation module, configured to generate a value population for the leaf row structure design variables using a genetic algorithm; A selection module is used to select individuals that meet preset requirements based on the fitness function value of each individual in the value population to obtain target values of the leaf row structure design variables; wherein the fitness function value is used to characterize the performance parameter value of the molecular pump, and the preset requirement condition is a limitation on the performance parameter values of different molecular pumps.

10. A molecular pump, characterized in that: The blade row structure design variables of the molecular pump are determined by the method for determining the design variable values of a molecular pump as described in claims 1 to 8.

Images