Optimization method of symmetrical sine type wave micro-channel with multiple groups of built-in double ribs

Abstract

The invention discloses a method for optimizing a symmetric sine wave micro-channel with multiple groups of built-in double ribs, and the method comprises the following steps: constructing a physical model containing a micro-channel, and defining the rib spacing, rib offset and rib height as design parameters; building a numerical simulation system, generating design points through Latin hypercube sampling, carrying out flow simulation, and building a back propagation neural network prediction model after data is analyzed and processed through grey relational degree; and finally, by taking maximization of the Nusselt number Nu and minimization of the friction coefficient f as targets, completing multi-target optimization by adopting a second-generation non-dominated sorting genetic algorithm. The problems that in the prior art, multi-parameter optimization efficiency of a composite structure is low, and heat transfer and flow resistance are difficult to balance are solved, the heat transfer and flow comprehensive performance of the optimized structure is remarkably superior to that of a smooth straight wave microchannel, and reliability and effectiveness of the method are verified; and a high-efficiency heat dissipation solution is provided for high-heat-flux scenes such as electronic equipment, aerospace vehicles and the like.

Description

Technical Field

[0001] This invention relates to the field of microchannel heat dissipation technology, specifically to an optimization method for a symmetrical sinusoidal wave microchannel with multiple sets of double ribs. Background Technology

[0002] With the rapid development of miniaturization, integration, and high efficiency in fields such as electronic equipment and aerospace, the power density of equipment has increased dramatically, and the heat flux density has increased significantly. Efficient heat dissipation has become a key technical bottleneck to ensure the stability and service life of equipment. Microchannels, with their advantages of high specific surface area and excellent heat transfer efficiency, have become the core solution to the problem of heat dissipation in high heat flux density. The research and optimization of their heat transfer enhancement technology has become a key direction in the field of power engineering.

[0003] Existing research indicates that different microchannel structures possess corresponding optimal structural parameters. A key challenge in contemporary microchannel design lies in rapidly and accurately determining these optimal design points and their associated parameters. Computational fluid dynamics (CFD) simulations of complex geometries or turbulent flows require fine mesh generation, with single-case calculations often taking hours or even days. When scaling to hundreds of parameter combinations, this computational demand renders traditional methods impractical on the engineering timeline. As engineering challenges evolve towards multiphysics coupling and multi-objective collaborative optimization, the limitations of traditional CFD methods in terms of efficiency, robustness, and applicability become increasingly apparent. Particularly in high-throughput iterative industrial design scenarios, their excessively high computational costs and slow response rates constitute a critical bottleneck hindering technological innovation.

[0004] In summary, existing microchannel structure optimization methods struggle to simultaneously address the optimization requirements of multi-factor coupling and multi-objective balance. Breakthroughs in artificial intelligence technology, through the synergistic integration of CFD numerical simulation and intelligent optimization algorithms, offer a new path for optimizing microchannel heat dissipation systems, thereby achieving more efficient optimization results. Summary of the Invention

[0005] The purpose of this invention is to provide an optimization method for a symmetrical sinusoidal wave microchannel with multiple built-in double ribs. This optimization method effectively solves the problem that existing microchannel structure optimization methods struggle to simultaneously address the optimization requirements of multi-factor coupling and multi-objective balance. To achieve the above objective, this invention adopts the following technical solution.

[0006] This invention provides an optimization method for a symmetrical sinusoidal wave microchannel with multiple built-in double ribs, the optimization method comprising the following steps: Step 1: Establish a physical model of a symmetrical sinusoidal wave microchannel with multiple built-in double ribs, defining the lateral distance between the double ribs as the rib spacing. D The longitudinal spacing is the rib offset. S、 Select rib spacing D Rib offsetS Rib height H r As design parameters, and set the design parameters ( D, S, H r The range of values ​​for ); Step 2: Determine the mathematical model and boundary conditions of the computational fluid domain of the symmetrical sinusoidal wave microchannel with multiple built-in double ribs, and construct a complete numerical simulation system; Step 3: Using the Latin hypercube sampling method, design points are generated within the range of design parameters. Flow simulation is then performed on a symmetrical sinusoidal wave microchannel with multiple built-in double ribs under a set Reynolds number condition to obtain the Nusselt number corresponding to different design points. Nu and coefficient of friction f ; Step 4: Verify the relationship between design parameters and Nusselt number using grey relational analysis. Nu coefficient of friction f The correlation is determined by dividing the design point data into training, testing and validation sets according to a set ratio, and constructing a backpropagation neural network prediction model optimized by a multi-objective genetic algorithm. The backpropagation neural network prediction model includes an input layer, a hidden layer and an output layer, and the number of neurons, activation function and training parameters of each layer are set. Step 5: Maximize the Nusel number Nu and minimizing the coefficient of friction f To achieve the goal, the population size, crossover rate, mutation rate, and maximum number of iterations for multi-objective genetic algorithm optimization are set. The second-generation non-dominated sorting genetic algorithm is used to perform multi-objective optimization on the backpropagation neural network prediction model to obtain the Pareto optimal solution set. Based on the Pareto optimal solution set, the design points corresponding to different decision-making conditions are selected to determine the optimal design parameter combination. The physical model of the symmetrical sinusoidal wave microchannel is then structurally optimized according to the optimal design parameters. Step Six: Simulate and verify the optimized structure of the physical model of the symmetrical sinusoidal wave microchannel. Compare the flow and heat transfer performance of the optimized structure with those of the smooth straight microchannel and the wave microchannel to verify the reliability and effectiveness of the optimization method. In a further preferred embodiment of the above scheme, in step one, the width of the solid computational domain of the theoretical model of the symmetrical sinusoidal wave microchannel with multiple sets of double ribs is... W ,high H and length L The widths are 0.4mm, 0.7mm, and 10mm respectively; the widths of the single-layer microchannels. W c With height H cAll are 0.2 mm; on the bottom surface of the microchannel, 25 sets of composite structures composed of symmetrical sinusoidal sidewall recesses and double ribs are distributed along the central axis of the symmetrical sinusoidal wave microchannel, with the length of a single rib being... L r and width W r The diameters are 0.08 mm and 0.035 mm, respectively.

[0007] In a further preferred embodiment of the above scheme, the symmetrical sinusoidal wave microchannel is a microchannel physical model made of silicon material, and the fluid in the microchannel is pure water; if the fluid is set as a three-dimensional incompressible Newtonian fluid in a continuous laminar flow state, the solid-liquid interface in the microchannel adopts a no-slip boundary condition, the physical properties of silicon remain constant, and the physical properties of pure water are considered constant except for the dynamic viscosity which changes with temperature.

[0008] The mathematical model and boundary conditions of the computational fluid dynamics domain of the symmetrical sinusoidal wave microchannel with multiple sets of built-in double ribs, which are further optimized according to the above scheme, are described as follows: The mathematical model is described by the following equation: Continuity equation: ; Momentum equation: ; Energy equation: ; For the solid domain, the energy equation can be written as: ; It is the Hamiltonian operator, used to describe the spatial rate of change of a physical quantity; U This refers to the fluid velocity vector; ρ This refers to fluid density; UU ) is the velocity tensor; p It refers to the pressure gradient, which describes the spatial distribution and variation of pressure; μ This refers to the dynamic viscosity of the fluid; c p It is the specific heat capacity at constant pressure of the fluid; T f This refers to the fluid temperature; λ f It is the thermal conductivity of the fluid; λ s It is the thermal conductivity of silicon; The boundary conditions are described as follows: The boundary conditions at the microchannel inlet are uniform velocity and uniform temperature. An outlet section is set at the microchannel outlet to control the outlet pressure (P out The pressure (Pa) is set to 0 Pa; a constant and uniform heat flux is applied as a heat source on the bottom surface of the solid; translational periodic boundary conditions are used on both sides of the solid domain to simulate the microchannel group, the contact surface between the fluid domain and the solid domain is set as a thermally coupled interface, and the remaining walls are set as adiabatic boundary conditions.

[0009] In a further preferred embodiment of the above scheme, step three, which uses the Latin hypercube sampling method to generate design points within the range of design parameter values, includes the following steps: Randomly sample 3 design parameters (rib spacing) D Rib offset S rib height H r ) and its parameter value range, and set a total of 56 groups of samples; The range of values ​​for each parameter is divided into 56 equally probable intervals based on the sample size, and one sample point is randomly selected in each interval. By randomly permuting and sorting, the 56 sample values ​​of each of the three design parameters are paired to form 56 unique design points that uniformly cover the three-dimensional parameter space.

[0010] A further preferred embodiment of the above scheme is to select different decisions, including the maximum Nusselt number, based on the Pareto optimal solution set. Nu Minimize the coefficient of friction f and maximizing performance evaluation criteria PEC ; where each satisfies the following expression: in, A w It is the heating wall area of ​​the solid domain of the channel; It is the heat flux density on the heated wall surface; , and These are the inlet temperature, outlet temperature, and heated wall surface temperature, respectively. D h It is the hydraulic diameter of the fluid at the inlet of the smooth microchannel; It is the average thermal conductivity of the fluid domain; It is the area of ​​convective heat transfer between the solid domain and the fluid domain; It is the pressure drop between the inlet and outlet of the microchannel; Nu 0 refers to the Nusselt number of a smooth, straight microchannel; f 0 refers to the coefficient of friction of a smooth, straight microchannel.

[0011] In a further preferred embodiment of the above scheme, the population size, crossover rate, mutation rate, and maximum number of iterations for multi-objective genetic algorithm optimization are configured according to set values, based on the average mean squared error of the training and test sets ( MSE ) as the fitness function, using the coefficient of determination ( R 2 The estimation algorithm's fit to the actual values ​​is optimized by using selection, crossover, and mutation operations to improve the initial weights and thresholds of the neural network, satisfying the following: ; ; Among them, the predicted output satisfy: ; It is the first k One predicted value, It is the first k One actual value, N It is the number of prediction points. It is the average of the predicted values; I i express i th Input layer neurons, w ij Indicates will i th Input layer neurons are connected to j th The weights of neurons in the hidden layer, w jk This indicates that j th Hidden layer neurons connect to k th The weights of the output neuron, b j yes j th Threshold of hidden layer neurons b k yes k th Threshold of hidden layer neurons f 1, f 2 are the Tansig function and the Purelin function, respectively.

[0012] In summary, due to the adoption of the above technical solutions, the present invention has the following beneficial technical effects: (2) This invention first establishes a physical model of a symmetrical sinusoidal wave microchannel with multiple sets of double ribs, defines the rib spacing, rib offset, and rib height as design parameters, and constructs a complete numerical simulation system including a fluid domain mathematical model and boundary conditions. Computational fluid dynamics simulations are conducted under a set Reynolds number condition to obtain the heat transfer and flow characteristic parameters corresponding to different design points. Secondly, the design points are generated using the Latin hypercube sampling method, and the correlation between the design parameters and heat transfer and flow objectives is verified through grey relational analysis. Data sets are divided and merged to construct a backpropagation neural network prediction model optimized by a genetic algorithm. Finally, with the goal of maximizing heat transfer performance and minimizing flow resistance, the prediction model is optimized using the NSGA-II algorithm to obtain the Pareto optimal solution set and determine the optimal combination of design parameters. This technical solution, through a combination of numerical simulation and experimental verification, significantly improves the optimization accuracy and efficiency of the composite structure microchannel, providing a theoretically rigorous and engineeringly practical solution for thermal management in high-heat-flux-density scenarios such as high-power electronic devices and aerospace components. (1) The multi-objective optimization method of the present invention systematically solves the problem that existing microchannel structure optimization methods are difficult to balance the optimization requirements of multi-factor coupling and multi-objective balance by integrating computational fluid dynamics simulation, grey relational analysis, genetic algorithm optimization backpropagation neural network prediction model (GA-BP) and non-dominated sorting genetic algorithm (NSGA-Ⅱ). Attached Figure Description

[0013] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0014] Figure 1 This is a flowchart of the steps of the multi-objective optimization method for a symmetrical sinusoidal wave microchannel with built-in multiple sets of double ribs provided by the present invention.

[0015] Figure 2 This is a schematic diagram of the symmetrical sinusoidal wave double-layer microchannel structure with multiple sets of built-in double ribs provided by the present invention.

[0016] Figure 3 This invention provides Nu and f Schematic diagram of the gray relational degree between each parameter Figure 4 This is a schematic diagram of the GA-BP neural network provided by the present invention. Figure 5 This is a schematic diagram illustrating the effect of the number of hidden layer neurons on MSE provided by the present invention. Figure 6 This is a comparative diagram of the training results of the GA-BP neural network provided by the present invention. Figure 7 This is a schematic diagram of the Pareto front curve provided by the present invention.

[0017] Figure 8 This is a comparison chart of the heat transfer and flow performance of the optimal model and the benchmark model under each decision based on the results of multi-objective optimization. Detailed Implementation

[0018] Embodiments of the present invention are described in detail below, examples of which are illustrated in the accompanying drawings, wherein the same or similar reference numerals denote the same or similar elements or elements having the same or similar functions throughout. The embodiments described below with reference to the accompanying drawings are exemplary and intended to explain the present invention, and should not be construed as limiting the present invention.

[0019] Combination Figure 1 and Figure 2 As shown, this invention provides an optimization method for a symmetrical sinusoidal wave microchannel with multiple sets of built-in double ribs, characterized in that the optimization method includes the following steps: Step 1: Establish a physical model of a symmetrical sinusoidal wave microchannel with multiple built-in double ribs, defining the lateral distance between the double ribs as the rib spacing. D The longitudinal spacing is the rib offset. S、 Select rib spacing D Rib offset S Rib height H r As design parameters, and set the design parameters ( D, S, H r The range of values ​​for ) Step 2: Determine the mathematical model and boundary conditions of the computational fluid domain of the symmetrical sinusoidal wave microchannel with multiple built-in double ribs, and construct a complete numerical simulation system; Step 3: Using the Latin hypercube sampling method, design points are generated within the range of design parameters. Flow simulation is then performed on a symmetrical sinusoidal wave microchannel with multiple built-in double ribs under a set Reynolds number condition to obtain the Nusselt number corresponding to different design points. Nu and coefficient of friction f ; Step 4: Verify the relationship between design parameters and Nusselt number using grey relational analysis. Nu coefficient of friction f The correlation is determined by dividing the design point data into training, testing and validation sets according to a set ratio, and constructing a backpropagation neural network prediction model optimized by a multi-objective genetic algorithm. The backpropagation neural network prediction model includes an input layer, a hidden layer and an output layer, and the number of neurons, activation function and training parameters of each layer are set. Step 5: Maximize the Nusel number Nu and minimizing the coefficient of friction f To achieve the objective, the population size, crossover rate, mutation rate, and maximum number of iterations for multi-objective genetic algorithm optimization are set. A second-generation non-dominated sorting genetic algorithm is used to optimize the backpropagation neural network prediction model for multiple objectives, obtaining the Pareto optimal solution set. Based on the Pareto optimal solution set, the appropriate decision is selected for different decision states (i.e., maximizing...). Nu , minimize f and maximizing PEC The design point is determined to identify the optimal combination of design parameters, and the physical model of the symmetrical sinusoidal wave microchannel is structurally optimized based on the optimal design parameters. Step Six: The optimized structure of the symmetrical sinusoidal wave microchannel physical model is simulated and verified. The flow and heat transfer performance of the optimized structure are compared with those of the smooth straight microchannel (SMC) and wave microchannel (WMC) to verify the reliability and effectiveness of the optimization method.

[0020] This embodiment systematically solves the problem that existing microchannel structure optimization methods struggle to balance the optimization requirements of multiple factors coupled with multiple objectives by integrating computational fluid dynamics simulation, grey relational analysis, and genetic algorithm optimization into a backpropagation neural network prediction model (GA-BP) and a non-dominated sorting genetic algorithm (NSGA-Ⅱ). First, a physical model of a symmetrical sinusoidal wave microchannel with multiple built-in double ribs is established. Rib spacing, rib offset, and rib height are defined as design parameters. A complete numerical simulation system including a fluid domain mathematical model and boundary conditions is constructed. Computational fluid dynamics simulations are conducted under a set Reynolds number condition to obtain heat transfer and flow characteristic parameters corresponding to different design points. Second, design points are generated using the Latin hypercube sampling method. Grey relational analysis is used to verify the correlation between design parameters and heat transfer and flow objectives. Data sets are divided and merged to construct a backpropagation neural network prediction model optimized by a genetic algorithm. Finally, with the goal of maximizing heat transfer performance and minimizing flow resistance, the prediction model is optimized using the NSGA-II algorithm to obtain the Pareto optimal solution set and determine the optimal combination of design parameters. This technical solution, through a combination of numerical simulation and experimental verification, significantly improves the optimization accuracy and efficiency of composite structure microchannels, providing a solution with both theoretical rigor and engineering practicality for thermal management in high-heat-flux-density scenarios such as high-power electronic devices and aerospace components.

[0021] In the physical model of the symmetrical sinusoidal wave microchannel with built-in multiple sets of double ribs in this invention, the physical model is as follows: Figure 2 As shown, the width of the solid computational domain ( W ),high( H ) and length ( LThe thicknesses are 0.4 mm, 0.7 mm, and 10 mm, respectively; the width of the single-layer microchannel ( W c ) and height ( H c All are 0.2 mm; on the bottom surface of the microchannel, 25 sets of composite structures composed of symmetrical sinusoidal sidewall recesses and double ribs are distributed along the central axis of the symmetrical sinusoidal wave microchannel; the sinusoidal wave profile is composed of y =±Asin(2πx λ w ) describes the length of a single rib ( L r ) and width ( W r The thicknesses are 0.08 mm and 0.035 mm, respectively. The material of the microchannel is silicon, and the fluid is pure water as the coolant. The fluid is assumed to be a three-dimensional incompressible Newtonian fluid in a continuous laminar flow state. The effects of gravity, viscous dissipation and radiative heat dissipation are ignored. The solid-liquid interface adopts a no-slip boundary condition. The physical properties of silicon are kept constant, and the physical properties of water are considered constant except for the dynamic viscosity which changes with temperature. The physical properties of silicon and water are shown in Table 1.

[0022] Table 1: Physical properties of silicon and water The dynamic viscosity of water is related to temperature as follows: Based on the above conditions, the mathematical model and boundary conditions of the computational fluid dynamics domain of the symmetrical sinusoidal wave microchannel with multiple built-in double ribs are described as follows: The mathematical model is described by the following equation: Continuity equation: ; Momentum equation: ; Energy equation: ; For the solid domain, the energy equation can be written as: ; It is the Hamiltonian operator, used to describe the spatial rate of change of a physical quantity; U This refers to the fluid velocity vector; ρ This refers to fluid density; UU ) is the velocity tensor; p It refers to the pressure gradient, which describes the spatial distribution and variation of pressure; μThis refers to the dynamic viscosity of the fluid; c p It is the specific heat capacity at constant pressure of the fluid; T f This refers to the fluid temperature; λ f λ is the thermal conductivity of the fluid; s It is the thermal conductivity of silicon; In this invention, such as Figure 2 As shown, the symmetrical sinusoidal wave microchannel is modeled using a silicon material. During simulation, the boundary conditions at the microchannel inlet are uniform velocity and uniform temperature. An outlet section is provided at the microchannel outlet to prevent backflow, and the outlet pressure ( P out The pressure (Pa) is set to 0 Pa (relative to atmospheric pressure). A constant and uniform heat flux is applied as a heat source to the bottom surface of the solid. q =1000 kW / m 2 On both sides of the solid domain, translational periodic boundary conditions are used to simulate the microchannel group. The contact surface between the fluid and solid domains is set as a thermally coupled interface, while the remaining walls are set as adiabatic boundary conditions. All walls use no-slip boundary conditions, and the coupling between the boundary conditions and the equations is shown in Table 2. Commercial computational fluid dynamics software is used to solve the governing equations, where a high-resolution scheme is used for spatial discretization, the computational accuracy is set to double precision, and the convergence residuals of all equations are set to 10. -5 .

[0023] Table 2: Detailed boundary conditions for solid and fluid domains Meanwhile, the Nusel number Nu Average coefficient of friction f They are shown below: ; ; in A w It is the heating wall area of ​​the solid domain of the channel; It is the heat flux density on the heated wall surface; , and These are the inlet temperature, outlet temperature, and heated wall surface temperature, respectively. D h It is the hydraulic diameter of the fluid at the inlet of the microchannel; It is the average thermal conductivity of the fluid domain; It is the area of ​​convective heat transfer between the solid domain and the fluid domain; It is the pressure drop between the inlet and outlet of the microchannel.

[0024] Enhanced convective heat transfer in microchannels is often accompanied by increased frictional losses; therefore, a performance evaluation criterion is introduced. PEC The overall performance of microchannels is evaluated using the following criteria. The performance evaluation standard PEC is determined by the following: ; Nu 0 refers to the Nusselt number of a smooth, straight microchannel; f 0 refers to the coefficient of friction of a smooth, straight microchannel.

[0025] In this invention, based on determining the value range of three design parameters and clarifying the boundary conditions, Latin hypercube sampling is used to generate 56 sets of input datasets within the variable study range. CFD numerical simulation calculations are then performed to obtain the corresponding Selby numbers. Nu Average coefficient of friction f As the output dataset, these datasets are randomly divided into training, test, and validation sets, with 70% used as the training set, 15% as the test set, and 15% as the validation set. The design parameters are validated using grey relational analysis. D, S, H r )and Nu , f The correlation.

[0026] In this invention, design parameters ( D, S, H r )and Nu , f Correlation such as Figure 3 As shown, a grey relational degree higher than 0.7 is considered a high correlation; a grey relational degree lower than 0.3 is considered a weak correlation. For Nu In other words, it can be seen that D , S and H r The grey relational strength of these three variables is above 0.7, which means that these three variables have a significant impact on the grey relational relationship between the three variables and the grey relational relationship between the three variables. Nu The effects of these variables are all very significant. S The impact was most significant, followed by D , again H r .for f In other words, it can be seen that D and S The grey relational degree is above 0.7, indicating that these two variables have a high correlation with each other. f It has a strong influence. Compared to the two variables mentioned above, H rAlthough not a major influencing factor, its grey relational degree of 0.6768 still indicates that it can be considered a significant factor. f This has a significant impact. Therefore, predictions are based on these three design variables. Nu and f That's reasonable.

[0027] In this invention, the genetic algorithm-optimized backpropagation neural network (GA-BP) prediction model comprises an input layer, a hidden layer, and an output layer. The number of neurons, activation functions, and training parameters for each layer are set. In this embodiment, a single-layer hidden layer is used, and the input layer has three neurons with the Purelin activation function. The number of hidden layers is adjusted... Figure 5 shown MSE Error analysis was performed, with the optimal number of neurons being 11 and the Tansig activation function used; the output layer had 2 neurons. GA-BP first optimized the initial weights and thresholds using a genetic algorithm. The initial population size was set to 100, the maximum number of generations to 50, the crossover rate to 0.8, and the mutation rate to 0.1. The Levenverg-Marquardt method was chosen as the training method. The Levenverg-Marquardt method is an optimization algorithm for solving nonlinear least squares problems, combining the advantages of gradient descent and Gauss-Newton methods: by introducing a damping factor to adjust the algorithm's characteristics, when the error between the predicted and actual values ​​is large, the damping factor increases, and the algorithm favors gradient descent to ensure computational stability; when the error is small, the damping factor decreases, and the algorithm favors the faster-converging Gauss-Newton method, thus simultaneously ensuring both training speed and accuracy. The maximum number of iterations was set to 1000, the learning rate to 0.01, and the minimum training error to 0.001. After successfully training the neural network, the output was predicted using the following formula: In the predicted output equation, I i yes i th Input layer neurons, w ij Indicates will i th Input layer neurons are connected to j th The weights of neurons in the hidden layer, w jk This indicates that j th Hidden layer neurons connect to k th The weights of the output neurons. b j yes j thThreshold of hidden layer neurons b k yes k th Threshold of neurons in the hidden layer. f 1, f 2 are the Tansig function and the Purelin function, respectively.

[0028] In this invention, the training results of the GA-BP neural network are as follows: Figure 6 As shown, (a) is a schematic diagram of the fit between the training results and the training set; (b) is a schematic diagram of the fit between the training results and the validation set; (c) is a schematic diagram of the fit between the training results and the validation set; and (d) is a schematic diagram of the fit between the training results and all experimental data. Figure 6 The less the data in (a), (b), (c), and (d) deviates from the predicted line, the better the training result of the GA-BP neural network. Furthermore, the coefficient of determination was calculated based on the actual and predicted values. R 2 This provides a more objective description of the predictive performance of neural networks. R² The core significance is to measure the degree of agreement between the predicted values ​​of the prediction model and the actual values. The value ranges from 0 to 1, and it intuitively reflects the model's ability to explain data patterns. R The closer ² is to 1, the higher the proportion of actual data variation that the model can explain, and the stronger the prediction accuracy. R The closer² is to 0, the worse the model's predictive performance, and the less likely it is to reflect the true patterns in the data. Training results for the neural network show that the GA-BP model performs well on the training, validation, and test sets. R The values ​​were 0.99557, 0.99117, and 0.99374 respectively, totaling... R The value is 0.99447, meaning the model can accurately capture the rib spacing. D Rib offset S rib height H r With Nusel Nu coefficient of friction f The nonlinear relationship between the two data points means that over 99% of the actual data variation can be explained by the model, and the predicted values ​​are almost identical to the actual values ​​simulated by CFD. This metric directly verifies the reliability of the GA-BP model.

[0029] In this invention, the population size, crossover rate, mutation rate, and maximum number of iterations are set for the multi-objective optimization algorithm. NSGA-II is used to perform multi-objective optimization on the prediction model to obtain the Pareto optimal solution set. Based on this solution set, design points corresponding to different optimization objectives are selected to determine the optimal combination of design parameters. The initial population size is 100, the crossover rate is 0.8, the mutation rate is 0.1, and the maximum number of iterations is 100. After 100 iterations, the objective function is as follows, and the optimal Pareto solution set is obtained. Figure 7 As shown.

[0030] ; ; =[ D , S , H r ]; ; Each point on the front surface of the Pareto optimal solution set is a Pareto optimal solution, the core meaning of which is to "maximize the Nusselt number". Nu "and minimizing the coefficient of friction" f "Between these two conflicting objectives, an unimprovable equilibrium is reached—that is, the performance of one objective cannot be improved without sacrificing the performance of the other. Each point corresponds to a set of optimal design parameter combinations (rib spacing)." D Rib offset S rib height H r Based on the obtained Pareto solution set, in the case of constant flow, if only considering... Nu Value (minimize 1 / Nu If so, then first select endpoint A (MC- Nu max If only considering f Value (Minimize) f If so, then first select endpoint B (MC- f min Point C (MC-) PEC max This represents a compromise between the two. Furthermore, simulations were conducted to verify the optimized structure, and the flow field was compared and analyzed with two benchmark models: the Smooth Straight Microchannel (SMC) and the Wave Microchannel (WMC). Figure 8 As shown, this embodiment includes SMC, WMC, and MC- Nu max MC- f min and MC- PECmax Nusel number Nu Friction resistance coefficient f and comprehensive performance evaluation PEC The comparison chart shows that the optimized MC-maximized Nusselt number model... Nu max of Nu Its value is 3.60 times that of SMC; the best overall performance model is MC- PEC max of Nu The value is MC- Nu max 88.7%, while f The value is only 51.6%. PEC The value reached 1.77. This indicates that the optimized MC- PEC max The model not only improves heat transfer efficiency but also achieves a balance between flow field characteristics and flow resistance, thereby verifying the reliability and effectiveness of the multi-objective optimization method of this invention.

[0031] The above description discloses only one preferred embodiment of the present invention, and should not be construed as limiting the scope of the present invention. Those skilled in the art will understand that all or part of the processes of the above embodiments can be implemented, and equivalent changes made in accordance with the claims of the present invention are still within the scope of the invention.

Claims

1. A method for optimizing a symmetrical sinusoidal wave microchannel with multiple sets of built-in double ribs, characterized in that, The optimization method includes the following steps: Step 1: Establish a physical model of a symmetrical sinusoidal wave microchannel with multiple built-in double ribs, defining the lateral distance between the double ribs as the rib spacing. D The longitudinal spacing is the rib offset. S、 Select rib spacing D Rib offset S Rib height H r As design parameters, and set the design parameters ( D, S, H r The range of values ​​for ); Step 2: Determine the mathematical model and boundary conditions of the computational fluid domain of the symmetrical sinusoidal wave microchannel with multiple built-in double ribs, and construct a complete numerical simulation system; Step 3: Using the Latin hypercube sampling method, design points are generated within the range of design parameters. Under the set Reynolds number condition, the flow and heat transfer of a symmetrical sinusoidal wave microchannel with multiple sets of double ribs are simulated to obtain the Nusselt number corresponding to different design points. Nu and coefficient of friction f ; Step 4: Verify the relationship between design parameters and Nusselt number using grey relational analysis. Nu coefficient of friction f The correlation is determined by dividing the design point data into training, testing and validation sets according to a set ratio, and constructing a backpropagation neural network prediction model optimized by a multi-objective genetic algorithm. The backpropagation neural network prediction model includes an input layer, a hidden layer and an output layer, and the number of neurons, activation function and training parameters of each layer are set. Step 5: Maximize the Nusel number Nu and minimizing the coefficient of friction f To achieve the goal, the population size, crossover rate, mutation rate, and maximum number of iterations for multi-objective genetic algorithm optimization are set. The second-generation non-dominated sorting genetic algorithm is used to perform multi-objective optimization on the backpropagation neural network prediction model to obtain the Pareto optimal solution set. Based on the Pareto optimal solution set, the design points corresponding to different decision-making conditions are selected to determine the optimal design parameter combination. The physical model of the symmetrical sinusoidal wave microchannel is then structurally optimized according to the optimal design parameters. Step Six: Simulate and verify the optimized structure of the physical model of the symmetrical sinusoidal wave microchannel. Compare the flow and heat transfer performance of the optimized structure with those of the smooth straight microchannel and the wave microchannel to verify the reliability and effectiveness of the optimization method.

2. The optimization method for a symmetrical sinusoidal wave microchannel with multiple sets of built-in double ribs according to claim 1, characterized in that, In step one, the width of the solid computational domain of the theoretical model of the symmetrical sinusoidal wave microchannel with multiple sets of double ribs is calculated. W ,high H and length L The widths are 0.4mm, 0.7mm, and 10mm respectively; the widths of the single-layer microchannels. W c With height H c All are 0.2 mm; on the bottom surface of the microchannel, 25 sets of composite structures composed of symmetrical sinusoidal sidewall recesses and double ribs are distributed along the central axis of the symmetrical sinusoidal wave microchannel, with the length of a single rib being... L r and width W r The diameters are 0.08 mm and 0.035 mm, respectively.

3. The optimization method for a symmetrical sinusoidal wave microchannel with multiple sets of built-in double ribs according to claim 2, characterized in that, The symmetrical sinusoidal wave microchannel uses a microchannel physical model made of silicon material, and the fluid inside the microchannel is pure water. If the fluid is set as a three-dimensional incompressible Newtonian fluid in a continuous laminar flow state, the solid-liquid interface inside the microchannel adopts a no-slip boundary condition, the physical properties of silicon remain constant, and the physical properties of pure water are considered constant except for the dynamic viscosity which changes with temperature.

4. The optimization method for a symmetrical sinusoidal wave microchannel with multiple sets of built-in double ribs according to claim 2, characterized in that, The mathematical model and boundary conditions of the computational fluid dynamics domain for a symmetrical sinusoidal wave microchannel with multiple built-in double ribs are described below: The mathematical model is described by the following equation: Continuity equation: ； Momentum equation: ； Energy equation: ； For the solid domain, the energy equation can be written as: ； It is the Hamiltonian operator, used to describe the spatial rate of change of a physical quantity; U This refers to the fluid velocity vector; ρ This refers to fluid density; UU ) is the velocity tensor; p It refers to the pressure gradient, which describes the spatial distribution and variation of pressure; μ This refers to the dynamic viscosity of the fluid; c p It is the specific heat capacity at constant pressure of the fluid; T f This refers to the fluid temperature; λ f It is the thermal conductivity of the fluid; λ s It is the thermal conductivity of silicon; The boundary conditions are described as follows: The boundary conditions at the microchannel inlet are uniform velocity and uniform temperature. An outlet section is set at the microchannel outlet to control the outlet pressure ( P out The pressure (Pa) is set to 0 Pa; a constant and uniform heat flux is applied as a heat source on the bottom surface of the solid; translational periodic boundary conditions are used on both sides of the solid domain to simulate the microchannel group, the contact surface between the fluid domain and the solid domain is set as a thermally coupled interface, and the remaining walls are set as adiabatic boundary conditions.

5. The optimization method for a symmetrical sinusoidal wave microchannel with multiple sets of built-in double ribs according to claim 2, characterized in that, Step three, which uses the Latin hypercube sampling method to generate design points within the range of design parameter values, includes the following steps: Randomly sample 3 design parameters (rib spacing) D Rib offset S rib height H r ) and its parameter value range, and set a total of 56 groups of samples; The range of values ​​for each parameter is divided into 56 equally probable intervals based on the sample size, and one sample point is randomly selected in each interval. By randomly permuting and sorting, the 56 sample values ​​of each of the three design parameters are paired to form 56 unique design points that uniformly cover the three-dimensional parameter space.

6. The optimization method for a symmetrical sinusoidal wave microchannel with multiple sets of built-in double ribs according to claim 1, characterized in that, Based on the Pareto optimal solution set, different decisions are selected, including the maximum Nusselt number. Nu Minimize the coefficient of friction f and maximizing performance evaluation criteria PEC ; where each satisfies the following expression: in, A w It is the heating wall area of ​​the solid domain of the channel; It is the heat flux density on the heated wall surface; , and These are the inlet temperature, outlet temperature, and heated wall surface temperature, respectively. D h It is the hydraulic diameter of the fluid at the inlet of the smooth microchannel; It is the average thermal conductivity of the fluid domain; It is the area of ​​convective heat transfer between the solid domain and the fluid domain; It is the pressure drop between the inlet and outlet of the microchannel; Nu 0 refers to the Nusselt number of a smooth, straight microchannel; f 0 refers to the coefficient of friction of a smooth, straight microchannel.

7. The optimization method for a symmetrical sinusoidal wave microchannel with multiple sets of built-in double ribs according to claim 1, characterized in that, The population size, crossover rate, mutation rate, and maximum number of iterations for multi-objective genetic algorithm optimization are configured according to set values, using the average mean squared error of the training and test sets (...). MSE ) as the fitness function, using the coefficient of determination ( R 2 The estimation algorithm's fit to the actual values ​​is optimized by using selection, crossover, and mutation operations to improve the initial weights and thresholds of the neural network, satisfying the following: ; ; Among them, the predicted output satisfy: ； It is the first k One predicted value, It is the first k One actual value, N It is the number of prediction points. It is the average of the predicted values; I i express i th Input layer neurons, w ij Indicates will i th Input layer neurons are connected to j th The weights of neurons in the hidden layer, w jk This indicates that j th Hidden layer neurons connect to k th The weights of the output neuron, b j yes j th Threshold of hidden layer neurons b k yes k th Threshold of hidden layer neurons f 1, f 2 are the Tansig function and the Purelin function, respectively.

Images