A micro-channel reactor bolt layout method and system

Abstract

The present application relates to the technical fields of micro-channel type reactor, and discloses a micro-channel type reactor bolt layout method and system, the method comprises the following steps: S1: model construction: the finite element model of micro-channel type reactor is established, and the parametric modeling of micro-channel type reactor is realized by using Python language; S2: target function value determination: the mathematical model of bolt layout problem is determined, the mathematical model of bolt layout problem is simulated, then the single polar plate deformation non-uniformity index is obtained based on the simulation solution, and the single polar plate deformation non-uniformity index is used as the target function value in the bolt layout problem of micro-channel type reactor; S3: bolt layout implementation: taking the structure of micro-channel type reactor as the research object, the non-uniformity of single polar plate deformation is used as the bolt layout target, and the bolt layout of micro-channel type reactor is realized. The problems of poor bolt layout effect of micro-channel type reactor containing mixed variables are solved.

Description

Technical Field

[0001] This invention relates to the field of microchannel reactor technology, specifically a bolt layout method and system for a microchannel reactor. Background Technology

[0002] Microfluidic reactors are widely used in hydrogen energy and carbon-negative reaction fields such as fuel cells, electrolyzers, and CO2 electrolysis reactors. However, due to the non-uniform deformation within and between the microfluidic reactor plates, problems such as uneven current distribution, uneven component degradation, and low gas production can occur, significantly affecting the efficiency and lifespan of the microfluidic reactor.

[0003] Because the overall cost of a fuel cell stack is very high, and multiple rounds of structural modifications may be necessary, optimizing the structure using experimental methods would be too time-consuming and expensive. Therefore, the design phase typically employs a combination of simulation and optimization algorithms. Bayesian optimization algorithms have achieved good results in many practical applications, particularly in parameter tuning, hyperparameter optimization, and machine learning model training. Their advantage lies in their ability to find good solutions with a limited number of function evaluations, making them suitable for the optimization design of complex equipment.

[0004] Currently, most research on Bayesian optimization algorithms is limited to continuous variables, while research on design optimization involving mixed variables—that is, design variables that simultaneously include both continuous and discrete cases. Since mixed variables exist in the optimal design of microfluidic reactor structures, research on Bayesian optimization methods for mixed variables is of practical significance. Summary of the Invention

[0005] To overcome the shortcomings of the prior art, the present invention provides a bolt layout method and system for microchannel reactors, which solves the problems of poor bolt layout effect in microchannel reactors containing mixed variables.

[0006] The technical solution adopted by the present invention to solve the above problems is:

[0007] A bolt layout method for a microchannel reactor includes the following steps:

[0008] S1: Model Building: Establish a finite element model of the microchannel reactor and implement parametric modeling of the microchannel reactor using Python.

[0009] S2: Determination of objective function value: Determine the mathematical model of the bolt layout problem, perform simulation calculations on the mathematical model of the bolt layout problem, and then obtain the non-uniformity index of the single plate deformation based on the simulation solution results. Use the non-uniformity index of the single plate deformation as the objective function value in the bolt layout problem of the microchannel reactor.

[0010] S3: Bolt layout implementation: Taking the microchannel reactor structure as the research object, the non-uniformity of the deformation of the single plate is used as the bolt layout target to realize the bolt layout of the microchannel reactor.

[0011] As a preferred technical solution, step S1 includes the following steps:

[0012] S101: Establish the geometric model of the microchannel reactor and assemble it;

[0013] S102: Establish material properties and assign properties to each component;

[0014] S103: Corresponding analysis steps required when establishing a microchannel reactor simulation;

[0015] S104: Establish contact between the components of the microchannel reactor;

[0016] S105: Establish loads and boundary conditions;

[0017] S106: Configure the grid for each component of the microchannel reactor;

[0018] S107: Create a job to perform simulation calculations.

[0019] As a preferred technical solution, step S2 includes the following steps:

[0020] S201: Determine the design variables and establish a mathematical model for the bolt layout of the microchannel reactor;

[0021] S202: Call the simulation software solver to read the parametric modeling script written in step S1 and perform simulation calculations;

[0022] S203: Extract the vertical displacement of the lower surface of the monopolar plate, output a report file, and obtain the non-uniformity of the monopolar plate deformation by reading the report file and performing data processing, which serves as an indicator of the bolt layout of the microchannel reactor.

[0023] As a preferred technical solution, in step S201, the simulation mathematical model of the microchannel reactor is as follows:

[0024]

[0025] Where find represents the design variable, X represents the set of discrete variables, i and j represent two different design variable numbers, x3,...,x n x1 represents the number of bolts on the longer side, x2 represents the number of bolts on the shorter side, and x represents the distance between adjacent bolts. i Let x represent the design variable numbered i. jLet j represent the discrete variable, RSD(X) represent the relative standard deviation of the displacement on the lower surface of the monopole plate, min:RSD(X) represents finding the minimum value of RSD(X), st represents the constraint condition, KU represents the stiffness matrix multiplied by the overall displacement vector of the reinforcing partition, F represents the total encapsulation force, and l b The length of the long side of the end plate is represented by n, and the total number of discrete variables is represented by D. i The range of values ​​for the discrete variable (number of bolts) is represented by q, and the number of values ​​for the discrete variable is represented by d. iq This represents the values ​​that a discrete variable can take, x. min Indicates the maximum value of the bolt hole spacing, x max This indicates the minimum bolt hole spacing.

[0026] As a preferred technical solution, in step S202, during simulation calculation, a statics solver is used to calculate the bolt layout model, and the nodal displacement matrix U is obtained according to the finite element equation. The calculation formula is as follows:

[0027] KU = F;

[0028] Where K is the overall stiffness matrix and F is the nodal load matrix.

[0029] As a preferred technical solution, in step S203, the lower surface region of the single-plate is selected, the vertical displacement field in this region is extracted from the simulation results, and the relative standard deviation of the displacement is calculated as the target of the microfluidic reactor simulation. The formula for calculating the relative standard deviation of the displacement is:

[0030]

[0031] Where RSD represents the relative standard deviation of displacement, u i denoted by , where represents the Z-direction displacement of each point on the lower surface of the monopolar plate, and u represents the average value of the Z-direction displacement on the lower surface of the monopolar plate.

[0032] As a preferred technical solution, step S3 includes the following steps:

[0033] S301: Initialize data using Latin hypercube sampling, generate an initial dataset for the non-uniformity of a single plate, and divide the initial dataset into m classes according to the values ​​of discrete variables;

[0034] S302: Train m surrogate models for each type of data;

[0035] S303: Calculate the minimum mean and maximum variance of m surrogate models, and extract the two points with the smallest relative standard deviation on the lower surface of the monopolar plate;

[0036] S304: Add the two points extracted in step S303 to the dataset of the corresponding category in the initial dataset, and retrain the surrogate model corresponding to these two points;

[0037] S305: Calculate the maximum variance and minimum mean of the two newly added surrogate models, and compare them with all the surrogate models obtained in step S302 to obtain the minimum mean and maximum variance among all surrogate models.

[0038] S306: Determine whether the preset number of iterations has been reached. If not, repeat steps S304 and S305. If the preset number of iterations has been reached, output the optimal solution in the dataset.

[0039] As a preferred technical solution, in step S302, the surrogate model is a Gaussian process regression model, and the kernel function of the Gaussian process regression model is:

[0040]

[0041] Where, k(x) i ,x j ) represents the kernel function of the Gaussian process regression model. and l are hyperparameters of the kernel function. Let represent bandwidth and l represent the feature length scale. The process of selecting hyperparameters is the training process of the surrogate model.

[0042] As a preferred technical solution, in step S305, the mean and variance of the displacement prediction values ​​in the surrogate model are respectively:

[0043] m(x) = K * T [K+σ N 2 I] -1 ,

[0044] Cov(x)=K ** -K * T [K+σ N 2 I] -1 K * ;

[0045] Where m(x) represents the mean, Cov(x) represents the variance, and K represents the covariance matrix. * The covariance matrix of sample points and prediction points, K * T K represents * The transpose matrix, K ** σ represents the covariance matrix of the prediction points. N Let I represent the noise variance, and let I represent the identity matrix.

[0046] A bolt layout system for a microchannel reactor, used to implement the aforementioned bolt layout method for a microchannel reactor, comprises the following modules connected in sequence:

[0047] Model building module: used to build a finite element model of a microchannel reactor and to implement parametric modeling of the microchannel reactor using Python language;

[0048] Objective function value determination module: used to determine the mathematical model of the bolt layout problem, perform simulation calculations on the mathematical model of the bolt layout problem, and then obtain the non-uniformity index of the single plate deformation based on the simulation solution results, and use the non-uniformity index of the single plate deformation as the objective function value in the bolt layout problem of the microchannel reactor.

[0049] Bolt layout implementation module: This module is used to implement the bolt layout of a microchannel reactor by taking the non-uniformity of the deformation of the single plate as the target of the bolt layout.

[0050] Compared with the prior art, the present invention has the following advantages:

[0051] (1) For the structure of the microfluidic reactor, a parametric modeling script was written in Python under the simulation software;

[0052] (2) For mixed variable optimization problems where the design variables are both continuous and discrete, a Bayesian optimization algorithm based on classification modeling is proposed. Multiple sub-surrogate models are established according to the different discrete variables. For different prediction points, the corresponding sub-surrogate model is selected for prediction.

[0053] (3) Using a Bayesian optimization algorithm based on classification modeling, the bolt layout and number are used as design variables, and the non-uniformity of the single plate is used as the objective function. The structure with the most uniform deformation of the single plate is selected to achieve the optimized design of the bolt layout of the microfluidic reactor. Attached Figure Description

[0054] Figure 1 This is a flowchart of a bolt layout method for a microchannel reactor according to the present invention;

[0055] Figure 2 This is a simplified geometric model diagram;

[0056] Figure 3 This is a schematic diagram of the equivalent sealing ring;

[0057] Figure 4 This is a schematic diagram of the bolt layout;

[0058] Figure 5 Optimize the basic model diagram for bolt layout;

[0059] Figure 6 A Bayesian optimization graph under the two-point addition criterion;

[0060] Figure 7 Historical convergence curves for Yeats optimization;

[0061] Figure 8 To optimize the bolt distribution and number diagram;

[0062] Figure 9 The optimized diagram shows the bolt distribution and number of bolts.

[0063] Figure 10 To optimize the displacement contour map of the previous single plate;

[0064] Figure 11 The optimized single-plate displacement contour map. Detailed Implementation

[0065] The present invention will be further described in detail below with reference to the embodiments and accompanying drawings, but the embodiments of the present invention are not limited thereto.

[0066] Example 1

[0067] like Figures 1 to 11 As shown, the purpose of this invention is to provide a bolt layout method for microchannel reactors to achieve optimized design of bolt layout in microchannel reactors. This invention aims to address the problem of uneven deformation within and between plates by simulating and optimizing the non-uniformity of single-plate deformation from multiple perspectives, including equipment manufacturing processes, packaging technology, and service degradation.

[0068] This invention proposes a structural optimization design method for microchannel reactors based on a classification-based Bayesian algorithm, belonging to the field of structural optimization design. First, a specific microchannel reactor structure is introduced, and a finite element model is established in CAE software. Using Python as the foundation, a script for this invention is written, realizing the parametric modeling of the microchannel reactor structure. Then, a mathematical model for optimizing the bolt layout of the microchannel reactor is established, and a quantitative index for the non-uniformity of the monopole plate is proposed. Finally, a classification-based Bayesian optimization algorithm is proposed to handle mixed optimization design problems involving continuous / discrete variables. Taking the microchannel reactor as the research object, the non-uniformity of the monopole plate deformation is used as the optimization objective. Combining the parametric modeling method and artificial intelligence optimization algorithms, the optimized design of the bolt layout of the microchannel reactor structure is achieved. The discrete variable is the number of bolts.

[0069] A bolt layout method for a microchannel reactor includes the following steps:

[0070] S1: Establish a finite element model of the microchannel reactor and implement parametric modeling in Python.

[0071] S2: Determine the mathematical model of the optimization problem, perform simulation calculations and solutions on the microchannel reactor model, extract relevant results and obtain the non-uniformity index of single plate deformation through data processing, which serves as the objective function value in the optimization design of the microchannel reactor;

[0072] S3: Taking the structure of a microchannel reactor as the research object, the non-uniformity of the deformation of the single plate is used as the optimization objective. Combined with parametric modeling methods, the bolt layout optimization design of the microchannel reactor structure is realized by using Bayesian algorithm.

[0073] Furthermore, the specific implementation method of step S1 is as follows:

[0074] S101: Establish the geometric model of the microchannel reactor and assemble it;

[0075] S102: Establish material properties and assign properties to each component;

[0076] S103: Corresponding analysis steps required when establishing a microchannel reactor simulation;

[0077] S104: Establish contact between the components of the microchannel reactor;

[0078] S105: Establish loads and boundary conditions;

[0079] S106: Configure the grid for each component of the microchannel reactor;

[0080] S107: Create a job to perform simulation calculations.

[0081] Furthermore, the specific implementation method of step S2 is as follows:

[0082] S201: Determine the design variables and establish an optimization mathematical model;

[0083] S202: Call the simulation software solver to read the parametric modeling script written in step S1 and perform simulation calculations;

[0084] S203: Extract the vertical displacement of the lower surface of the monopolar plate, output a report file, and obtain the non-uniformity of the monopolar plate deformation by reading the report file and performing data processing, which serves as an indicator for subsequent optimization design.

[0085] Furthermore, the specific implementation method of step S3 is as follows:

[0086] S301: Initialize data using Latin hypercube sampling, generate an initial dataset for the non-uniformity of a single plate, and divide the initial dataset into m classes according to the values ​​of discrete variables;

[0087] S302: Train m surrogate models for each type of data;

[0088] S303: Extract the two points with the smallest minimum mean and maximum variance from the m proxy model calculators;

[0089] S304: Add the two points from S303 to the dataset of the corresponding category and retrain the corresponding surrogate model;

[0090] S305: Calculate the maximum variance and minimum mean of these two surrogate models and compare them with other surrogate models to obtain the overall minimum mean and maximum variance.

[0091] S306: Determine if the preset number of iterations has been reached. If not, repeat steps S304 and S305. If the preset number of iterations has been reached, output the optimal solution in the dataset.

[0092] A further technical solution is that step S1 is specifically implemented as follows:

[0093] S101: Establish the geometric model of the microchannel reactor and assemble it, and implement it using parameterization;

[0094] The microfluidic reactor consists of an upper end plate, an aluminum middle plate, a single electrode plate, bolts, a membrane electrode, a sealing ring, and a bipolar plate. The two electrodes, the sealing ring, and the membrane electrode form a single cell. The single cell is connected in series and then stacked with the upper end plate, the aluminum middle plate, and the lower end plate to form the microfluidic reactor.

[0095] As can be seen from the structure of the microfluidic reactor, the whole can be equivalent to a symmetrical structure. When establishing the initial model, the following treatment is adopted to simplify the calculation: a quarter of the overall model structure is selected for study; the bolts are modeled using beam elements; when there are multiple repeating structures in the model, such as single cells, array operation is used to generate the model; since the optimization objective examines the state of the single plate after the bolts are tightened, the membrane electrode and the sealing ring are in a compressed state. Since there is a certain gap between the membrane electrode and the plate in the physical model, the simulation requires a lot of time to calculate the process of the plate being compressed to the membrane electrode. Therefore, the thickness of the sealing ring is reduced in this invention so that the membrane electrode is in direct contact with the plate.

[0096] S102: Establish material properties and assign properties to each component, and implement this using parameterization;

[0097] S103: Establish the corresponding analysis steps required for the simulation of microchannel reactors and implement them using parameterization;

[0098] In the simulation calculation of the microchannel reactor, the reactor is subjected to static forces, so a static analysis step is adopted, with a total time of 0.001 and an initial step size of 10. -05 The minimum step size is 10.-08 The maximum step size is 0.001.

[0099] S104: Establish the contact between the components of the microchannel reactor and implement it using parameterization;

[0100] In the simulation calculation of microfluidic reactors, it is necessary to set the contact properties between components. In this invention, binding constraints are used between the various components, and MPC constraints are used between the bolts (which are simplified using beam elements) and the upper and lower end plates.

[0101] Each component is bound together and interacts. The bolt preload is set to a preset value (e.g., 1000N). Considering the influence of gravity, symmetrical boundary conditions are applied to the x and y symmetry planes in the model. The bottom surface of the model is subjected to displacement constraint conditions with zero displacement. A static general solver is used for simulation and solution.

[0102] S105: Establish loads and boundary conditions, and implement them using parameterization;

[0103] In the simulation calculation of the microchannel reactor, the load settings need to consider the bolt preload and gravity after the microchannel reactor is assembled. The bolt preload is set to a preset value (e.g., 1000N), and the gravitational acceleration is 9.8m / s2. In the boundary condition settings, symmetrical boundary conditions need to be set for the quarter model. Since the lower end plate of the microchannel reactor is fixed, displacement boundary conditions need to be set on the lower surface of the lower end plate, and the displacement is 0.

[0104] S106: Mesh settings are configured for each component of the microchannel reactor, and parametric implementation is used;

[0105] The microfluidic reactor structure is discretized into hexahedral solid units, and the mesh size of each component is specified.

[0106] S107: Create a job to perform simulation calculations and implement it using parameterization.

[0107] A further technical solution is that step S2 is specifically implemented as follows:

[0108] S201: Determine the design variables and establish an optimization mathematical model;

[0109] Bolt layout optimization design requires optimizing both the number and position of bolts on the end plate. The overall approach is to set the number of bolts on the long and short sides as discrete variables x1 and x2, respectively, and the distance between bolts as x3 to x... n It is a continuous variable. Figure 6 This is a schematic diagram of the bolt layout. For ease of parametric modeling, [the following is used]... Figure 7Based on the bolt layout, the design is optimized by adding bolts or adjusting their positions. The number of bolts along the short axis is x1∈[3,7], the number of bolts along the long axis is x2∈[3,5], and the distance between adjacent bolts is x... i ∈[0.025,0.05]m, where i=3,4,...,n, and n is the total number of design variables. The constraints are that the bolt hole positions do not exceed the upper end plate, and the total bolt load remains unchanged.

[0110] The optimized model is shown in the following formula:

[0111] find: X = {x1, x2, ..., x} n}

[0112] min:RSD(X)

[0113] stKU=F

[0114]

[0115]

[0116] x i ∈D i D i ={d i1 ,d i2 ,...,d iqi}, i = 1, 2

[0117] x min ≤x i ≤x max i = 1, 2, ..., n

[0118] S202: Call the simulation software solver to read the parametric modeling script written in step S1 and perform simulation calculations;

[0119] The model was calculated using a statics solver, and the nodal displacement matrix U was obtained based on the finite element equations, where K is the overall stiffness matrix and F is the nodal load matrix.

[0120] KU = F

[0121] S203: Extract the vertical displacement of the lower surface of the monopolar plate, output a report file, and obtain the non-uniformity of the monopolar plate deformation by reading the report file and performing data processing, which serves as an indicator for subsequent optimization design.

[0122] When the structure of a microfluidic reactor is poorly designed, excessive deformation of the monopole plate can occur. Since the lower surface of the monopole plate is in close contact with the membrane electrode during operation, excessive deformation can negatively impact the internal structure and performance of the reactor. In practical engineering, it is desirable for the deformation of the monopole plate to be as uniform as possible under the encapsulation load. Therefore, the lower surface region of the monopole plate is selected, and the vertical displacement field within this region is extracted from the simulation results. The relative standard deviation of the displacement is calculated using the following formula and used as the target for optimizing the microfluidic reactor structure.

[0123]

[0124] In the formula, u i The nodal displacements of the selected region for the monopole plate are given, where n is the number of nodes.

[0125] A further technical solution is that step S3 is specifically implemented as follows:

[0126] S301: Initialize the data using Latin hypercube sampling, and divide the initial dataset into m classes based on the values ​​of discrete variables;

[0127] S302: Train m surrogate models for each type of data;

[0128] The surrogate model used in this invention is a Gaussian process regression model, which can be represented by the mean function m(x) and the variance function cov(x), and is defined as follows:

[0129] f(x)~GP(m(x),k(x,x′))

[0130] m(x) = E[f(x)]

[0131] k(x,x′)=E[(f(x)-m(x))(f(x′)-m(x′))]

[0132] In the formula, x and x' are any two input vectors.

[0133] Due to the properties of Gaussian processes, the following joint distribution exists:

[0134]

[0135] In the formula, given the model training data, X = [x1, x2, ..., x...]. n Let y = [y1, y2, ..., y3] be the sample points of the training data. n ] represents the function value of the corresponding sample point, y * To predict the target value, the noise follows an independent and identically distributed Gaussian distribution ε~N(0,σ). N I is the identity matrix, K is the covariance matrix, and K0 is the variance matrix. *For sample point x and prediction point x * The covariance matrix, K ** For the prediction point x * The covariance matrix is ​​given by the following formula:

[0136]

[0137] K * T =[k[x1,x * ],k[x2,x * ],k[x3,x * ],...,k[x n ,x * ]]

[0138] K ** =[k[x * ,x * ]]

[0139] Where k is the kernel function. This invention uses a radial basis function kernel.

[0140] The radial basis function kernel is as follows:

[0141]

[0142] in, and l are hyperparameters of the kernel function, representing bandwidth and feature length scale, respectively.

[0143] As explained above, the kernel function of Gaussian process regression contains several parameters, which are called hyperparameters. With appropriate hyperparameter selection, the surrogate model can fit any nonlinear system; the process of selecting hyperparameters is the model training process.

[0144] The maximum likelihood estimation method is typically used to estimate hyperparameters at a single point by maximizing the likelihood of the Gaussian process regression model. The following equation represents the likelihood estimate of the Gaussian process regression model.

[0145]

[0146] In the formula, X represents a sample point, y represents the function value corresponding to the sample point, and θ represents a hyperparameter. For ease of calculation, the above formula is converted into a negative logarithmic form.

[0147]

[0148] The gray wolf optimization algorithm is used to minimize the above equation to obtain reasonable hyperparameter values, and the training of the surrogate model is completed.

[0149] S303: Extract the two points with the smallest target value from the minimum mean and maximum variance of the m proxy model calculators;

[0150] S304: Add the two points from S303 to the dataset of the corresponding category and retrain the corresponding surrogate model;

[0151] S305: Calculate the maximum variance and maximum mean of these two surrogate models and compare them with other surrogate models to obtain the overall maximum mean and maximum variance.

[0152] The mean and variance of the predicted values ​​are as follows:

[0153] m(x) = K * T [K+σ N 2 I] -1

[0154] Cov(x)=K ** -K * T [K+σ N 2 I] -1 K *

[0155] In the formula, K is the covariance matrix, σ N Let I be the noise variance, and I be the identity matrix.

[0156] S306: Determine if the preset number of iterations has been reached. If not, repeat steps S304 and S305. If the preset number of iterations has been reached, output the optimal solution in the dataset.

[0157] Example 2

[0158] like Figures 1 to 11 As shown, as a further optimization of Embodiment 1, this embodiment also includes the following technical features based on Embodiment 1:

[0159] like Figure 1 As shown, the present invention provides a bolt layout method for a microchannel reactor, comprising the following steps:

[0160] S1: Establish a finite element model of the microchannel reactor and implement parametric modeling in Python.

[0161] A further technical solution is that step S1 is specifically implemented as follows:

[0162] S101: Establish the geometric model of the microchannel reactor and assemble it, and implement it using parameterization;

[0163] The microfluidic reactor consists of an upper end plate, an aluminum middle plate, a single electrode plate, bolts, a membrane electrode, a sealing ring, and a bipolar plate. The two electrodes, the sealing ring, and the membrane electrode form a single cell. The single cell is connected in series and then stacked with the upper end plate, the aluminum middle plate, and the lower end plate to form the microfluidic reactor.

[0164] As can be seen from the structure of the microfluidic reactor, the whole can be equivalent to a symmetrical structure. To simplify calculations when establishing the initial model, the following treatment was adopted: a quarter of the overall model was selected for study; bolts were modeled using beam elements; when multiple repeating structures, such as single cells, existed in the model, array operations were used to generate the model; since the optimization objective examines the state of a single electrode plate after the bolts are tightened, both the membrane electrode and the sealing ring are in a compressed state. Because there is a certain gap between the membrane electrode and the electrode plate in the physical model, the simulation requires a significant amount of time to calculate the process of the electrode plate compressing to the membrane electrode. Therefore, the thickness of the sealing ring was reduced in this invention, allowing the membrane electrode to directly contact the electrode plate; the final simplified geometric model is as follows: Figure 2 As shown in the figure, and part of the program implementation is shown in Table 1.

[0165] Table 1. Main Program Codes for Geometric Modeling and Assembly Implementation

[0166]

[0167] S102: Establish material properties and assign properties to each component, and implement this using parameterization;

[0168] In the simulation calculation of the microchannel reactor, it is necessary to assign values ​​to the material properties of the components. The actual material parameters of each component in the microchannel reactor are shown in Table 2. The materials of each component are considered to be isotropic materials, and the equivalent elastic modulus of the sealing ring is calculated using the equivalent modulus calculation method.

[0169] Table 2 Mechanical Properties of Materials Used in Components

[0170]

[0171] The equivalent elastic modulus of the sealing ring is calculated. The equivalent elastic modulus of the sealing region at the current stage is obtained from the equivalent stiffness model, and the relationship is as follows:

[0172]

[0173] Where k is the equivalent stiffness, E is the equivalent elastic modulus, and l is the component thickness.

[0174] The relationship between the sealing ring thickness and the membrane electrode thickness is as follows: Figure 3As shown, there is a gap between the electrode plate and the membrane electrode at this point. Experiments show that the sealing ring of the microfluidic reactor, after being bolted together, experiences a force of 20 MPa. At this point, there is no gap between the electrode plate and the membrane electrode. Based on the force balance and deformation coordination relationship, the sealing area can be equivalently represented. After the equivalent representation, the membrane electrode and the lower surface of the electrode plate are just in contact with each other without contact pressure. The elastic modulus E of the equivalent sealing ring of the current model can be obtained by simultaneously applying the following formulas. s' = 95.29 MPa.

[0175] k s' x+k r x = F

[0176] k s' x = p s As

[0177]

[0178]

[0179] F = F bolt n bolt

[0180] In the formula, k is... s' The equivalent rubber pad stiffness, k r For the equivalent stiffness of the diffusion layer, A s l is the pressure-bearing area of ​​the rubber pad. s' E represents the equivalent thickness of the rubber pad. r Let A be the diffusion layer modulus. r l is the pressure-bearing area of ​​the diffusion layer. r E represents the thickness of the diffusion layer. s' F represents the equivalent elastic modulus of the rubber pad, and F represents the total sealing force. bolt n is the bolt preload. bolt This represents the number of bolts.

[0181] The material setting procedure is shown in Table 3.

[0182] Table 3. Main Program Codes for Material Setting Procedure

[0183]

[0184] S103: Establish the corresponding analysis steps required for the simulation of microchannel reactors and implement them using parameterization;

[0185] In the simulation calculation of the microchannel reactor, the reactor is subjected to static forces, so a static analysis step is adopted, with a total time of 0.001 and an initial step size of 10. -05 The minimum step size is 10. -08 The maximum step size is 0.001.

[0186] The program implementation is shown in Table 4.

[0187] Table 4. Main Program Code for Implementation of Analysis Step Settings

[0188]

[0189] S104: Establish the contact between the components of the microchannel reactor and implement it using parameterization;

[0190] In the simulation calculation of microfluidic reactors, it is necessary to set the contact properties between components. In this invention, binding constraints are used between the various components, and MPC constraints are used between the bolts (which are simplified using beam elements) and the upper and lower end plates.

[0191] Each component is bound together and interacts with it. The bolt preload is set, the influence of gravity is considered, the x and y symmetry planes in the model are symmetrically bounded, the bottom surface of the model is subjected to a displacement constraint condition with zero displacement, and a static universal solver is used for simulation and solution.

[0192] The program implementation is shown in Table 5.

[0193] Table 5. Main Program Code for Implementation of Analysis Steps

[0194]

[0195] S105: Establish loads and boundary conditions, and implement them using parameterization;

[0196] In the simulation calculation of the microchannel reactor, the load settings need to consider the bolt preload and gravity after the microchannel reactor is assembled. The bolt preload is set and the gravitational acceleration is 9.8 m / s2. In the boundary condition settings, symmetrical boundary conditions need to be set for the quarter model. Since the lower end plate of the microchannel reactor is fixed, a displacement boundary condition needs to be set on the lower surface of the lower end plate, with a displacement of 0.

[0197] The program implementation is shown in Table 6.

[0198] Table 6. Main Program Code for Setting Load Boundary Conditions

[0199]

[0200] S106: Mesh settings are configured for each component of the microchannel reactor, and parametric implementation is used;

[0201] The microfluidic reactor structure is discretized into hexahedral solid units, and the mesh size of each component is specified.

[0202] The program implementation is shown in Table 7.

[0203] Table 7. Main Program Code for Mesh Generation

[0204]

[0205] S107: Create a job to perform simulation calculations and implement it using parameterization.

[0206] The program implementation is shown in Table 8.

[0207] Table 8. Main program code for mesh generation.

[0208]

[0209] S2: Simulate and solve the microchannel reactor model, extract relevant results, and output a report file RPT as initial data;

[0210] A further technical solution is that step S2 is specifically implemented as follows:

[0211] S201: Determine the design variables and establish an optimization mathematical model;

[0212] Bolt layout optimization design requires optimizing both the number and position of bolts on the end plate. The overall approach is to set the number of bolts on the long and short sides as discrete variables x1 and x2, respectively, and the distance between bolts as x3 to x... n It is a continuous variable. Figure 4 This is a schematic diagram of the bolt layout. For ease of parametric modeling, [the following is used]... Figure 5 Based on the bolt layout, the design is optimized by adding bolts or adjusting their positions. The number of bolts along the short axis is x1∈[3,7], the number of bolts along the long axis is x2∈[3,5], and the distance between adjacent bolts is x... i ∈[0.025,0.05]m, where i=3,4,...,n, and n is the total number of design variables. The constraints are that the bolt hole positions do not exceed the upper end plate, and the total bolt load remains unchanged.

[0213] The optimized model is shown in the following formula:

[0214] find: X = {x1, x2, ..., x} n}

[0215] min:RSD(X)

[0216] stKU=F

[0217]

[0218]

[0219] x i ∈D i Di ={d i1 ,d i2 ,...,d iqi}, i = 1, 2

[0220] x min ≤x i ≤x max i = 1, 2, ..., n

[0221] S202: Simulation calculations are performed using a Python script that utilizes the parametric modeling established in S1;

[0222] S203: Extract the vertical displacement of the lower surface of the monopole plate and output the report file RPT as initial data.

[0223] Read the calculation results of S201 and extract the vertical displacement of the lower surface of the monopole plate. Part of the program is shown in Table 9.

[0224] Table 9. Main Program Code for Result Extraction

[0225]

[0226] When the structure of a microfluidic reactor is poorly designed, excessive deformation of the monopole plate can occur. Since the lower surface of the monopole plate is in close contact with the membrane electrode during operation, excessive deformation can negatively impact the internal structure and performance of the reactor. In practical engineering, it is desirable for the deformation of the monopole plate to be as uniform as possible under the encapsulation load. Therefore, the lower surface region of the monopole plate is selected, and the vertical displacement field within this region is extracted from the simulation results. The relative standard deviation of the displacement is calculated using the following formula and used as the target for optimizing the microfluidic reactor structure.

[0227]

[0228] In the formula, u i The nodal displacements of the selected region for the monopole plate are given, where n is the number of nodes.

[0229] S3: Taking the structure of a microchannel reactor as the research object, the non-uniformity of the deformation of the single plate is used as the optimization objective. Combined with parametric modeling methods, the Bayesian algorithm is used to realize the optimized design of the microchannel reactor structure.

[0230] A further technical solution is that step S3 is specifically implemented as follows:

[0231] S301: Initialize the data using Latin hypercube sampling, and divide the initial dataset into m classes based on the values ​​of discrete variables;

[0232] First, 75 initial sample points are generated using Latin hypercube sampling. The objective function values ​​corresponding to these sample points are then calculated using finite element simulation, and an initial surrogate model is established. The optimization iteration count is set to 50.

[0233] The gray wolf algorithm was used to optimize the point addition criterion. The population size of the gray wolf algorithm was set to 100 and the number of iterations was set to 500.

[0234] S302: Train m surrogate models for each type of data;

[0235] The surrogate model used in this invention is a Gaussian process regression model, which can be represented by the mean function m(x) and the variance function cov(x), and is defined as follows:

[0236] f(x)~GP(m(x),k(x,x′))

[0237] m(x) = E[f(x)]

[0238] k(x,x′)=E[(f(x)-m(x))(f(x′)-m(x′))]

[0239] In the formula, x and x' are any two input vectors.

[0240] Due to the properties of Gaussian processes, the following joint distribution exists:

[0241]

[0242] In the formula, given the model training data, X = [x1, x2, ..., x...]. n Let y = [y1, y2, ..., y3] be the sample points of the training data. n ] represents the function value of the corresponding sample point, y * To predict the target value, the noise follows an independent and identically distributed Gaussian distribution ε~N(0,σ). N I is the identity matrix, K is the covariance matrix, and K0 is the variance matrix. * For sample point x and prediction point x * The covariance matrix, K ** For the prediction point x * The covariance matrix is ​​given by the following formula:

[0243]

[0244] K * T =[k[x1,x * ],k[x2,x * ],k[x3,x * ],...,k[x n ,x* ]]

[0245] K ** =[k[x * ,x * ]]

[0246] Where k is the kernel function. This invention uses a radial basis function kernel.

[0247] The radial basis function kernel is as follows:

[0248]

[0249] in, and l are hyperparameters of the kernel function, representing bandwidth and feature length scale, respectively.

[0250] As explained above, the kernel function of Gaussian process regression contains several parameters, which are called hyperparameters. With appropriate hyperparameter selection, the surrogate model can fit any nonlinear system; the process of selecting hyperparameters is the model training process.

[0251] The maximum likelihood estimation method is typically used to estimate hyperparameters at a single point by maximizing the likelihood of the Gaussian process regression model. The following equation represents the likelihood estimate of the Gaussian process regression model.

[0252]

[0253] In the formula, X represents a sample point, y represents the function value corresponding to the sample point, and θ represents a hyperparameter. For ease of calculation, the formula is converted into a negative logarithmic form.

[0254]

[0255] The gray wolf optimization algorithm is used to minimize the above equation to obtain reasonable hyperparameter values, and the training of the surrogate model is completed.

[0256] S303: Extract the two points with the largest maximum values ​​of the mean and variance of the m proxy model calculators;

[0257] The addition criterion, also known as the search function, is used after constructing the surrogate model with the initial dataset. It is then constructed using the model's posterior distribution, and subsequently optimized using the Grey Wolf Optimization Algorithm to obtain new points. The construction of the addition criterion needs to balance development and exploration. Development refers to considering the location where the optimal point is most likely to exist, while exploration refers to considering the location where the model's uncertainty is highest, avoiding convergence to local extrema. Commonly used addition criteria include the Expectation Boosting (EI) criterion, the Probability Boosting (PI) criterion, and the Lower Confidence Bound (LCB) criterion.

[0258] The invention employs a two-point addition criterion, the expression of which is shown in the following formula:

[0259]

[0260] In the formula, κ is the control coefficient. When optimizing the maximum value, κ is 1, and when optimizing the minimum value, κ is -1.

[0261] By using this point-addition criterion, each iteration in the optimization process generates two points: one representing the point closest to the extreme point under the current surrogate model, and the other representing the point with the highest uncertainty under the current surrogate model. This satisfies the optimization algorithm's balance between development and search capabilities. However, during the optimization calculation, different iterations may obtain the same point using the point-addition criterion. Calculating the same sample point not only increases computational cost but also does not improve the accuracy of the surrogate model. Therefore, the Bayesian algorithm under the two-point-addition criterion needs to remove these duplicate points to avoid repeated calculations and addition to the dataset. However, as can be seen from the following formula, when an unknown point coincides with a point in the dataset (i.e., the objective function value of that point is known), the variance of its predicted value is 0. Therefore, during the optimization of I2(x), no data points will coincide with the original dataset. Thus, it is only necessary to judge the newly added points when optimizing I1(x) to avoid overlap between the newly added points and the original dataset. The Bayesian optimization framework under the two-point-addition criterion is as follows: Figure 6 .

[0262] S304: Add the two points from S303 to the dataset of the corresponding category and retrain the corresponding surrogate model;

[0263] S305: Calculate the maximum variance and maximum mean of these two surrogate models and compare them with other surrogate models to obtain the overall maximum mean and maximum variance.

[0264] The mean and variance of the predicted values ​​are as follows:

[0265] m(x) = K * T [K+σ N 2 I] -1

[0266] Cov(x)=K ** -K * T [K+σ N 2 I] -1 K *

[0267] In the formula, K is the covariance matrix, σ N Let I be the noise variance, and I be the identity matrix.

[0268] S306: Determine if the preset number of iterations has been reached. If not, repeat steps S304 and S305. If the preset number of iterations has been reached, output the optimal solution in the dataset.

[0269] Figure 7 The historical convergence curves of the Bayesian optimization are presented. The horizontal axis represents the number of iterations, and the vertical axis represents the optimal objective function value at the current iteration number. The graph shows that the optimization algorithm converges after 10 iterations.

[0270] Through optimization calculations, an optimal set of bolt numbers and relative positions was finally obtained, as follows: Figure 9 As shown. The optimized bolt layout is substituted into the initial equivalent model for calculation. Simultaneously, the non-uniformity index of the unipolar plate deformation in the equivalent model under the original bolt layout is calculated. Figure 8 To optimize the comparison of the structures before and after, the calculation results were compared, and the following results were obtained: Figure 9 The stress distribution cloud diagrams under different conditions are shown in Table 10. As shown in Table 10, the non-uniformity of the single-pole plate deformation corresponding to the optimized structure is 0.6%, which is a reduction of 17.8% compared to the optimization results of the reinforced partition in Section 4.3.2.

[0271] As described above, the present invention can be implemented well.

[0272] All features disclosed in all embodiments of this specification, or steps in all methods or processes implied in the disclosure, may be combined and / or extended or replaced in any way, except for mutually exclusive features and / or steps.

[0273] The above description is merely a preferred embodiment of the present invention and is not intended to limit the present invention in any way. Based on the technical essence of the present invention, any simple modifications, equivalent substitutions, and improvements made to the above embodiments within the spirit and principles of the present invention shall still fall within the protection scope of the present invention.

Claims

1. A bolt layout method for a microchannel reactor, characterized in that, Includes the following steps: S1: Model Building: Establish a finite element model of the microchannel reactor and implement parametric modeling of the microchannel reactor using Python. S2: Determination of objective function value: Determine the mathematical model of the bolt layout problem, perform simulation calculations on the mathematical model of the bolt layout problem, and then obtain the non-uniformity index of the single plate deformation based on the simulation solution results. Use the non-uniformity index of the single plate deformation as the objective function value in the bolt layout problem of the microchannel reactor. S3: Bolt layout implementation: Taking the microchannel reactor structure as the research object, the non-uniformity of the deformation of the single plate is used as the bolt layout target to realize the bolt layout of the microchannel reactor. Step S3 includes the following steps: S301: Initialize data by Latin hypercube sampling, generate an initial dataset for the non-uniformity of a single plate, and divide the initial dataset into m classes according to the values ​​of discrete variables, where the discrete variable is the number of bolts; S302: Train m surrogate models for each type of data; S303: Calculate the minimum mean and maximum variance of m surrogate models, and extract the two points with the smallest relative standard deviation on the lower surface of the monopolar plate; S304: Add the two points extracted in step S303 to the dataset of the corresponding category in the initial dataset, and retrain the corresponding proxy model for these two points; S305: Calculate the maximum variance and minimum mean of the two newly added surrogate models, and compare them with all the surrogate models obtained in step S302 to obtain the minimum mean and maximum variance among all surrogate models. S306: Determine whether the preset number of iterations has been reached. If not, repeat steps S304 and S305. If the preset number of iterations has been reached, output the optimal solution in the dataset.

2. The bolt layout method for a microchannel reactor according to claim 1, characterized in that, Step S1 includes the following steps: S101: Establish the geometric model of the microchannel reactor and assemble it; S102: Establish material properties and assign properties to each component; S103: Corresponding analysis steps required when establishing a microchannel reactor simulation; S104: Establish contact between the components of the microchannel reactor; S105: Establish loads and boundary conditions; S106: Configure the grid for each component of the microchannel reactor; S107: Create a job to perform simulation calculations.

3. The bolt layout method for a microchannel reactor according to claim 1, characterized in that, Step S2 includes the following steps: S201: Determine the design variables and establish a mathematical model for the bolt layout of the microchannel reactor; S202: Call the simulation software solver to read the parametric modeling script written in step S1 and perform simulation calculations; S203: Extract the vertical displacement of the lower surface of the monopolar plate, output a report file, and obtain the non-uniformity of the monopolar plate deformation by reading the report file and performing data processing, which serves as an indicator of the bolt layout of the microchannel reactor.

4. The bolt layout method for a microchannel reactor according to claim 1, characterized in that, In step S202, during the simulation calculation, a statics solver is used to calculate the bolt layout model. The nodal displacement matrix U is obtained based on the finite element equations. The calculation formula is as follows: ； Where K is the overall stiffness matrix and F is the nodal load matrix.

5. The bolt layout method for a microchannel reactor according to claim 4, characterized in that, In step S203, the lower surface region of the single plate is selected, the vertical displacement field in this region is extracted from the simulation results, and the relative standard deviation of the displacement is calculated as the target of the microfluidic reactor simulation. The formula for calculating the relative standard deviation of the displacement is: ； in, Indicates the relative standard deviation of displacement. This represents the Z-direction displacement of each point on the lower surface of the monopole plate. This represents the average displacement in the Z direction of the lower surface of the monopole.

6. The bolt layout method for a microchannel reactor according to claim 5, characterized in that, In step S302, the surrogate model is a Gaussian process regression model, and the kernel function of the Gaussian process regression model is: ； in, The kernel function represents the Gaussian process regression model. and These are hyperparameters of the kernel function. Indicates bandwidth. The process of selecting hyperparameters, which represents the feature length scale, is the training process of the surrogate model.

7. The bolt layout method for a microchannel reactor according to claim 6, characterized in that, In step S305, the mean and variance of the displacement predictions in the surrogate model are as follows: ； in, This represents the mean. Represents variance. K'' Represents the covariance matrix. The covariance matrix between sample points and prediction points, express The transpose of the matrix, This represents the covariance matrix of the predicted points. σ N Indicates the noise variance. I Let represent the identity matrix, and y represent the objective function value corresponding to the training sample.

8. A bolt layout system for a microchannel reactor, characterized in that, A bolt layout method for implementing a microchannel reactor as described in any one of claims 1 to 7 comprises the following modules connected in sequence: Model building module: used to build a finite element model of a microchannel reactor and to implement parametric modeling of the microchannel reactor using Python language; Objective function value determination module: used to determine the mathematical model of the bolt layout problem, perform simulation calculations on the mathematical model of the bolt layout problem, and then obtain the non-uniformity index of the single plate deformation based on the simulation solution results, and use the non-uniformity index of the single plate deformation as the objective function value in the bolt layout problem of the microchannel reactor. Bolt layout implementation module: This module is used to implement the bolt layout of a microchannel reactor by taking the non-uniformity of the deformation of the single plate as the target of the bolt layout.

Images