A method and system for optimizing design of a Tesla valve type micro-channel heat sink

Abstract

The application discloses a Tesla valve type micro-channel heat sink optimization design method and system, and belongs to the technical field of micro-channel heat dissipation. The non-uniform change of the width of the valve core top and the height of the valve core along the flow direction is introduced as an optimization design variable, the maximization of Nusselt number and the minimization of system pressure drop are taken as optimization objectives, a multi-objective optimization algorithm is used to optimize and solve the optimization design variable, a Pareto optimal solution set is obtained, and an approximate ideal solution sorting method is used to select a scheme with the maximum relative closeness from the Pareto optimal solution set as a comprehensive optimal design scheme. A performance prediction proxy model is constructed by using a Latin hypercube sampling and numerical simulation mode, fast mapping between structure parameters and heat exchange performance is realized, the calculation cost in the optimization process is significantly reduced, the design efficiency is improved, the flow resistance is effectively controlled while the heat exchange capacity is ensured, and the comprehensive heat dissipation performance is further improved.

Description

Technical Field

[0001] This invention relates to the field of microchannel heat dissipation technology, specifically a Tesla valve-type microchannel heat sink optimization design method and system. Background Technology

[0002] With the continuous increase in chip integration and power density, the heat flux density per unit area is constantly increasing. Traditional microchannel cooling methods with simple geometric configurations are no longer sufficient to meet the requirements of enhanced heat transfer and low-resistance operation. Tesla valve-type microchannels, with their asymmetric geometry, can induce periodic diversion, convergence, and recirculation disturbances in the fluid without moving parts, thereby enhancing near-wall heat transfer. Therefore, they are considered a promising new structure for heat dissipation of high heat flux density chips.

[0003] However, most existing Tesla valve-type microchannel heat sink designs employ uniform valve core size configurations, failing to fully consider the possibility of valve core size parameters varying along the flow direction. This makes it difficult to effectively control flow resistance while simultaneously improving heat transfer capacity. Furthermore, there are significant coupling effects among the structural parameters of microchannel heat sinks. Single-factor analysis alone cannot yield the optimal solution that balances heat transfer enhancement and resistance control. While high-precision numerical simulations can directly exhaustively optimize multi-parameter combinations, they suffer from high computational complexity, long optimization cycles, and a tendency to get trapped in local optima. Summary of the Invention

[0004] The purpose of this invention is to provide a Tesla valve-type microchannel heat sink optimization design method and system to solve the above-mentioned problems.

[0005] The technical solution of this invention is:

[0006] A Tesla valve-type microchannel heat sink optimization design method includes the following steps:

[0007] Obtain the optimization design variables and their preset value ranges, including the cooling medium inlet velocity. u in And the valve core top width that varies non-uniformly along the flow direction in the Tesla valve-type microchannel heat sink. W t and valve core height H .

[0008] The optimized design variables and their preset value ranges are input into a pre-constructed performance prediction proxy model. This performance prediction proxy model is a backpropagation neural network model, which has a built-in mapping relationship between input variables and output performance indicators, used to determine the input valve core top width. W t Valve core height H and inlet speed uin Output the corresponding Nusel number. Nu and system voltage drop Δ P .

[0009] The output Nusel number Nu Maximize system voltage drop Δ P With minimization as the optimization objective, a multi-objective optimization algorithm is used to find the optimal solution for the design variables and obtain the Pareto optimal solution set.

[0010] The Pareto optimal solution set is comprehensively evaluated by the approximation ideal solution ranking method. Each solution in the Pareto optimal solution set is taken as a candidate scheme. The Euclidean distance from each candidate scheme to the positive ideal solution and the negative ideal solution is calculated. The relative closeness of each candidate scheme is calculated based on the Euclidean distance. The candidate scheme with the largest relative closeness is selected as the comprehensive optimal design scheme.

[0011] Output the valve core top width corresponding to the comprehensive optimal design scheme. W t Valve core height H and inlet speed u in This is the final design result of the Tesla valve-type microchannel heat sink non-uniform valve core structure.

[0012] Furthermore, the mapping relationship between the input variables and output performance metrics built into the backpropagation neural network model is established in the following way:

[0013] The Latin hypercube sampling method is used to generate sample points within the preset value range of the optimization design variables, and the Nusselt number corresponding to each set of sample points is obtained through numerical simulation. Nu and system voltage drop Δ P A sample database is constructed, and the backpropagation neural network model is trained based on the sample database. The numerical simulation uses a periodic calculation model as the analysis object. The periodic calculation model is obtained from the actual microchannel overall model and is used to reflect the periodic flow and heat transfer characteristics inside the complete microchannel. The number of sample points generated by the Latin hypercube sampling is 60 sets. The sample database is divided into a training set, a validation set, and a test set for training and performance evaluation of the backpropagation neural network model.

[0014] The boundary conditions for the numerical simulation include: a velocity inlet boundary condition is used at the inlet, with the inlet fluid temperature set to 293.15 K; a pressure outlet boundary condition is used at the outlet, with the outlet pressure set to 101.25 kPa; periodic boundary conditions are set on both sides of the calculation model; a constant heat flux density boundary condition with a heat flux density of 100 W / cm² is applied at the bottom of the heat sink; the solid-liquid interface satisfies the temperature continuity condition; a no-slip boundary condition is used for the fluid at the wall; and the remaining outer walls are set as adiabatic boundaries.

[0015] The mapping relationship between input variables and output performance indicators in BPNN is obtained through LHS sampling, numerical simulation, and training with a sample database; while ensuring accuracy, it significantly reduces computational costs and improves optimization efficiency.

[0016] Furthermore, the Tesla valve-type microchannel heat sink's basic structure employs an elliptical valve core arranged in an incremental manner along the flow direction. This serves as the initial configuration for subsequent optimization of the non-uniform valve core structure.

[0017] Furthermore, the multi-objective optimization algorithm is a non-dominated sorting genetic algorithm II.

[0018] The optimization process of the non-dominated sorting genetic algorithm II includes:

[0019] Initialize the population by generating an initial population containing multiple individuals, with each individual corresponding to a set of optimization design variable values.

[0020] Calculate the Nusselt number for each individual in the population. Nu With system voltage drop Δ P , which serves as the objective function value for an individual.

[0021] Perform a fast non-dominated sort on the current population to divide the individuals in the population into different non-dominated levels.

[0022] Calculate the crowding distance between individuals in the same non-dominated hierarchy to maintain population diversity.

[0023] Selection operations are performed based on non-dominant hierarchy and crowding distance to select dominant individuals, and crossover and mutation operations are performed sequentially to generate offspring populations.

[0024] The parent population is merged with the offspring population to form a merged population.

[0025] The merged population is then subjected to another rapid non-dominated sorting and crowding distance calculation, and the best individuals are selected to form a new generation population.

[0026] Determine if the preset maximum number of iterations has been reached. If not, return to calculate the Nusselt number for each individual in the population. Nu With system voltage drop Δ P This step continues iterating; if the desired result is reached, the last generation of the population is output as the Pareto optimal solution set.

[0027] The Pareto optimal solution set refers to a set of non-dominant candidate solutions that cannot further improve one objective without worsening the other. This solution set reflects the optimal trade-off boundary between the Nusselt number and the system voltage drop.

[0028] Furthermore, the width of the valve core top W t The value range is 0.20mm~0.60mm, and the valve core height H The value range is 0.50 mm to 1.10 mm, and the inlet velocity... u in The value range is 0.60 m / s to 1.80 m / s. Under the premise of ensuring the feasibility of microchannel fabrication, the fluid forms appropriate contraction, expansion and backflow disturbance in the valve core area, thereby effectively controlling the flow resistance while enhancing near-wall heat transfer.

[0029] Furthermore, the backpropagation neural network model is a three-layer feedforward neural network structure, with the input layer corresponding to the width of the top of the valve core. W t Valve core height H and inlet speed u in The output layer corresponds to the Nusselt number. Nu and system voltage drop Δ P The hidden layer activation function uses the Tansig function, and the output layer activation function uses the Purelin function. The training algorithm employs Bayesian regularization. The Tansig function introduces nonlinear mapping capabilities, and combined with Bayesian regularization, it effectively suppresses overfitting with limited samples, achieving good control over the Nusselt number Nu and system voltage drop Δ. P High-precision prediction.

[0030] Furthermore, NSGA-II yields a set of Pareto non-dominated solutions, rather than a unique optimal solution. By constructing positive and negative ideal solutions, the distances between each candidate scheme and the ideal and non-ideal points can be calculated, and the final comprehensive optimal non-uniform valve core structure scheme can be selected accordingly. This provides a mathematical basis for subsequent decision-making and point selection in the Pareto solution set, and unifies the processing of benefit-type indicators. Nu (The larger the better) and cost-related indicators (Δ) P The comprehensive evaluation problem between (the smaller the better). The positive ideal solution... A + and negative ideal solution A - satisfy:

[0031] .

[0032] .

[0033] Among them, the ideal solution A + This represents the ideal solution formed when all indicators reach their optimal values; the negative ideal solution. A -This represents the ideal solution when each indicator takes its worst value; for benefit-related indicators, such as the average Nusselt number... Nu The maximum value is taken as the positive ideal solution; for cost-related indicators, such as system pressure drop Δ... P The minimum value is taken as the positive ideal solution.

[0034] The relative proximity satisfies: ;in, c i This represents the relative similarity of the i-th candidate solution; c i The larger the value, the closer the candidate solution is to the positive ideal solution and the further away it is from the negative ideal solution, and the better its overall performance.

[0035] Furthermore, the performance of the aforementioned optimal design scheme is compared with that of a Tesla valve-type microchannel heat sink using a uniform valve core size. The comparison indicators include the average Nusselt number. Nu System voltage drop Δ P and comprehensive performance evaluation factors PEC By comparing the performance of the optimized non-uniform valve core structure with that of the uniform valve core structure, the improvement effect of the non-uniform valve core structure in terms of heat transfer capacity, flow resistance and overall performance is quantitatively evaluated.

[0036] A Tesla valve-type microchannel heat sink optimization design system is designed using the above-mentioned optimization design method, including: a parameter input module, a surrogate model storage module, a multi-objective optimization module, a decision evaluation module, and a result output module.

[0037] The parameter input module is used to set the width of the valve core top. W t Valve core height H and inlet speed u in The preset value range.

[0038] The proxy model storage module is communicatively connected to the parameter input module and is used to store pre-built performance prediction proxy models. These performance prediction proxy models are backpropagation neural network models, which have a built-in mapping relationship between input variables and output performance indicators. This is used to determine the performance prediction based on the input valve core top width. W t Valve core height H and inlet speed u in Output the corresponding Nusel number. Nu and system voltage drop Δ P .

[0039] The multi-objective optimization module is communicatively connected to the proxy model storage module. It receives a preset value range output by the parameter input module and calls the performance prediction proxy model in the proxy model storage module, using Nusselt numbers... Nu Maximize system voltage drop Δ P Minimization is the optimization objective. A multi-objective optimization algorithm is used to find the optimal solution for the design variables and obtain the Pareto optimal solution set.

[0040] The decision evaluation module is communicatively connected to the multi-objective optimization module and is used to receive the Pareto optimal solution set output by the multi-objective optimization module. The Pareto optimal solution set is comprehensively evaluated using the approximation ideal solution ranking method, and the scheme with the highest relative closeness is selected as the comprehensive optimal design scheme.

[0041] The result output module is communicatively connected to the decision evaluation module and is used to receive the comprehensive optimal design scheme output by the decision evaluation module, and output the valve core top width corresponding to the comprehensive optimal design scheme. W t Valve core height H and inlet speed u in .

[0042] The proxy model storage module is constructed in the following way: sample points are generated within a preset range of values ​​for the optimization design variables using the Latin hypercube sampling method, and the Nusselt number corresponding to each set of sample points is obtained through numerical simulation. Nu and system voltage drop Δ P A sample database is constructed, and the backpropagation neural network model is trained based on the sample database, so that the backpropagation neural network model has a built-in mapping relationship between input variables and output performance indicators.

[0043] Compared with the prior art, the beneficial effects of the present invention are:

[0044] This invention introduces the non-uniform variation of the valve core top width and valve core height along the flow direction as optimization design variables. This allows for segmental control of the local flow cross-sectional area, valve zone contraction and expansion, and fluid deflection and disturbance intensity within the Tesla valve-type microchannel along the flow direction, thereby effectively controlling flow resistance while enhancing near-wall heat transfer. Based on this, the invention uses maximizing the Nusselt number and minimizing the system pressure drop as optimization objectives. A multi-objective optimization algorithm is employed to find the optimal solution set for the design variables, obtaining the Pareto optimal solution set, thus achieving synergistic optimization of heat transfer capacity and flow resistance. Meanwhile, a performance prediction proxy model with a built-in mapping relationship between input variables and output performance indicators is adopted. Based on the input valve core top width, valve core height and inlet velocity, the corresponding Nusselt number and system pressure drop are quickly output. The Euclidean distance from each candidate scheme to the positive ideal solution and the negative ideal solution is calculated from the Pareto optimal solution set by the approximation ideal solution sorting method. The relative closeness of each candidate scheme is calculated based on the Euclidean distance. The candidate scheme with the largest relative closeness is selected as the comprehensive optimal design scheme. This effectively avoids the defect of traditional methods being prone to getting trapped in local optima, significantly reduces the computational cost in the optimization process, improves design efficiency, and ultimately effectively controls the flow resistance while ensuring heat exchange capacity, thereby achieving a further improvement in comprehensive heat dissipation performance. Attached Figure Description

[0045] Figure 1 This is a flowchart of the design method of the present invention.

[0046] Figure 2 This is a top view of the Tesla valve-type microchannel heat sink foundation structure of the present invention.

[0047] Figure 3 This is a schematic diagram of the computational region for a Tesla valve-type microchannel heat sink.

[0048] Figure 4 This is a schematic diagram of the non-uniform valve core structure parameters of the present invention.

[0049] Figure 5 This is a schematic diagram of the backpropagation neural network of the present invention.

[0050] Figure 6 This is a distribution diagram of the Pareto optimal solution set and the comprehensive optimal design scheme of the present invention.

[0051] Figure 7 This is a bottom temperature comparison map showing the optimal structure and the original, unoptimized structure of this invention. Detailed Implementation

[0052] The following is combined with Figures 1 to 7 The specific embodiments of the present invention will be described in detail below.

[0053] Example

[0054] like Figure 1 As shown, a Tesla valve-type microchannel heat sink optimization design method includes the following steps:

[0055] Obtain the optimization design variables and their preset value ranges. The optimization design variables include the cooling medium inlet velocity. u in And the valve core top width that varies non-uniformly along the flow direction in the Tesla valve-type microchannel heat sink. W t and valve core height H .

[0056] The optimized design variables and their preset value ranges are input into a pre-built performance prediction proxy model. This model is a backpropagation neural network model with a built-in mapping relationship between input variables and output performance indicators, used to determine the input valve core top width. W t Valve core height H and inlet speed u in Output the corresponding Nusel number. Nu and system voltage drop Δ P .

[0057] The output Nusel number Nu Maximize system voltage drop Δ P With minimization as the optimization objective, a multi-objective optimization algorithm is used to find the optimal solution for the design variables and obtain the Pareto optimal solution set.

[0058] The Pareto optimal solution set is comprehensively evaluated by the approximation ideal solution ranking method. Each solution in the Pareto optimal solution set is taken as a candidate scheme. The Euclidean distance from each candidate scheme to the positive ideal solution and the negative ideal solution is calculated. The relative closeness of each candidate scheme is calculated based on the Euclidean distance. The candidate scheme with the largest relative closeness is selected as the comprehensive optimal design scheme.

[0059] Output the valve core top width corresponding to the optimal overall design scheme W t Valve core height H and inlet speed u in This is the final design result of the Tesla valve-type microchannel heat sink non-uniform valve core structure.

[0060] In some embodiments, the mapping relationship between the input variables and output performance metrics built into the backpropagation neural network model is established in the following way:

[0061] The Latin hypercube sampling method is used to generate sample points within a preset range of values ​​for the optimization design variables. Nusselt numbers corresponding to each set of sample points are obtained through numerical simulation. Nu and system voltage drop Δ P Construct a sample database and train a backpropagation neural network model based on the sample database; such as Figure 2 and Figure 3 As shown, to improve numerical computation efficiency and characterize the flow and heat transfer behavior of the overall microchannel heat sink while ensuring computational accuracy, a periodic computational model was used as the analysis object in the numerical simulation. The periodic computational model was obtained from the actual overall microchannel model and was used to reflect the periodic flow and heat transfer characteristics inside the complete microchannel. The number of sample points generated by Latin hypercube sampling was 60 sets. For each set of sample points, numerical simulation was performed based on the above boundary conditions, and the corresponding mean Nusselt number was calculated. Nu With system voltage drop Δ P This allows us to obtain the correspondence between input variables and output performance indicators. The average Nusselt number is used to characterize the convective heat transfer capacity of the microchannel heat sink, and the system pressure drop is used to characterize the resistance loss when the fluid flows through the microchannel.

[0062] The sample database is divided into training, validation, and test sets for training and performance evaluation of backpropagation neural network models.

[0063] In some embodiments, copper is used for the heat sink substrate and cover plate, and silicon is used for the heating element. A thermally conductive silicone grease layer is provided between the heat sink substrate and the heating element interface to reduce interfacial thermal resistance and improve heat transfer efficiency. Deionized water is preferred as the cooling medium. The above material system can meet the processing and heat transfer requirements of microchannel liquid cooling in electronic devices.

[0064] The boundary conditions for the numerical simulation include: a velocity inlet boundary condition is used at the inlet, with the inlet fluid temperature set to 293.15 K; a pressure outlet boundary condition is used at the outlet, with the outlet pressure set to 101.25 kPa; periodic boundary conditions are set on both sides of the computational model; a constant heat flux density boundary condition with a heat flux density of 100 W / cm² is applied at the bottom of the heat sink; the solid-liquid interface satisfies the temperature continuity condition; no-slip boundary conditions are used for the fluid at the wall; and the remaining outer walls are set as adiabatic boundaries.

[0065] The mapping relationship between input variables and output performance indicators in BPNN is obtained through LHS sampling, numerical simulation, and training with a sample database; while ensuring accuracy, it significantly reduces computational costs and improves optimization efficiency.

[0066] The Tesla valve-type microchannel heat sink's basic structure employs an elliptical valve core arranged in an incremental manner along the flow direction. This serves as the initial configuration for subsequent optimization of the non-uniform valve core structure.

[0067] In some embodiments, the multi-objective optimization algorithm is a non-dominated sorting genetic algorithm II.

[0068] The optimization process of the non-dominated sorting genetic algorithm II includes:

[0069] S1 initializes the population, generating an initial population containing multiple individuals, with each individual corresponding to a set of optimization design variable values.

[0070] S2 calculates the Nusselt number for each individual in the population. Nu With system voltage drop Δ P , which serves as the objective function value for an individual.

[0071] S3 performs a fast non-dominated sort on the current population, dividing the individuals in the population into different non-dominated levels.

[0072] S4 calculates the crowding distance between individuals in the same non-dominated hierarchy to maintain population diversity.

[0073] S5 performs selection operations based on non-dominant hierarchy and crowding distance to select dominant individuals, and then performs crossover and mutation operations in sequence to generate offspring populations.

[0074] S6 merges the parent population with the offspring population to form a merged population.

[0075] S7 performs another rapid non-dominated sorting and crowding distance calculation on the merged population, and selects the best individuals to form a new generation population.

[0076] S8 determines whether the preset maximum number of iterations has been reached. If not, it returns to S2 to continue iterating; if it has been reached, it outputs the last generation of the population as the Pareto optimal solution set.

[0077] The Pareto optimal solution set refers to a set of non-dominant candidate solutions that cannot further improve one objective without worsening the other. This solution set reflects the optimal trade-off boundary between the Nusselt number and the system voltage drop.

[0078] like Figure 4 As shown, the non-uniform valve core structure in this embodiment is mainly characterized by the width of the valve core top. W t and valve core height H This is achieved through changes, by altering... W t and HThe value of adjusts the local flow cross-sectional area near the valve core, the degree of contraction and expansion in the valve region, and the intensity of fluid deflection and disturbance, thereby affecting near-wall fluid renewal, the degree of thermal boundary layer disruption, and overall flow resistance. A non-uniform valve core structure refers to a valve core structure where the dimensional parameters are not constant along the flow direction but are designed to vary according to optimization requirements to achieve a synergistic match between enhanced heat transfer and pressure drop control. In some embodiments, the top width of the valve core... W t The value range is 0.20mm~0.60mm, valve core height H The value range is 0.50 mm to 1.10 mm, and the inlet velocity is... u in The value range is 0.60 m / s to 1.80 m / s. While ensuring the feasibility of microchannel fabrication, the fluid in the valve core region is made to generate appropriate contraction, expansion, and backflow disturbances, thereby effectively controlling flow resistance while enhancing near-wall heat transfer.

[0079] In some embodiments, such as Figure 5 As shown, the backpropagation neural network model is a three-layer feedforward neural network structure, with the input layer corresponding to the width of the top of the valve core. W t Valve core height H and inlet speed u in The output layer corresponds to the Nusselt number. Nu and system voltage drop Δ P The hidden layer activation function uses the Tansig function, and the output layer activation function uses the Purelin function. The training algorithm employs Bayesian regularization. The Tansig function introduces nonlinear mapping capabilities, and combined with Bayesian regularization, it effectively suppresses overfitting with limited samples, achieving good control over the Nusselt number Nu and system voltage drop Δ. P High-precision prediction.

[0080] like Figure 6 As shown, since NSGA-II obtains a set of Pareto non-dominated solutions, rather than a unique optimal solution, by constructing positive and negative ideal solutions, the distances between each candidate scheme and the ideal and non-ideal points can be further calculated. Based on this, the final comprehensive optimal non-uniform valve core structure scheme can be selected, providing a mathematical basis for subsequent decision-making and point selection of the Pareto solution set, and uniformly handling benefit-type indicators. Nu (The larger the better) and cost-related indicators (Δ) P A comprehensive evaluation problem involving the optimal balance between (the smaller the better). Positive ideal solution. A + and negative ideal solution A - satisfy:

[0081] .

[0082] .

[0083] Among them, the ideal solution A + This represents the ideal solution formed when all indicators reach their optimal values; the negative ideal solution. A - This represents the ideal solution when each indicator takes its worst value; for benefit-related indicators, such as the average Nusselt number... Nu The maximum value is taken as the positive ideal solution; for cost-related indicators, such as system pressure drop Δ... P The minimum value is taken as the positive ideal solution.

[0084] The degree of closeness is satisfied: ;in, c i This represents the relative similarity of the i-th candidate solution; c i The larger the value, the closer the candidate solution is to the positive ideal solution and the further away it is from the negative ideal solution, and the better its overall performance.

[0085] The sorting method for approximating the ideal solution includes the following steps:

[0086] Suppose there are m candidate solutions in the Pareto optimal solution set, and there are n evaluation indicators. Construct a decision matrix.

[0087] .

[0088] in, x ij This represents the value of the i-th scheme on the j-th index.

[0089] The evaluation indicators in the decision matrix are normalized to obtain the normalized matrix:

[0090] .

[0091] in, t ij Let be the normalized value of the i-th scheme on the j-th indicator. Normalization is used to eliminate the influence of differences in the dimensions and orders of magnitude of different indicators on the comprehensive evaluation results.

[0092] Based on the weight of each indicator ω j Calculate the weighted normalized matrix:

[0093] .

[0094] in, ω j This represents the weight of the j-th evaluation indicator;a ij The weighted index value is shown.

[0095] Constructing the ideal solution A + and negative ideal solution A - :

[0096] .

[0097] .

[0098] Here, a positive ideal solution represents the ideal solution formed when each indicator takes its optimal value, while a negative ideal solution represents the ideal solution formed when each indicator takes its worst value. For benefit-related indicators, such as the average Nusselt number... Nu The maximum value is taken as the positive ideal solution; for cost-related indicators, such as system pressure drop Δ... P The minimum value is taken as the positive ideal solution.

[0099] Calculate the Euclidean distances from each candidate solution to the positive and negative ideal solutions:

[0100] .

[0101] .

[0102] Where, d i + d represents the distance between the i-th candidate solution and the positive ideal solution; i - This represents the distance between the i-th candidate solution and the negative ideal solution.

[0103] Calculate the relative similarity of each candidate solution:

[0104] .

[0105] in, c i This represents the relative similarity of the i-th candidate solution; c i The larger the value, the closer the scheme is to the positive ideal solution and the further away it is from the negative ideal solution, and the better its overall performance.

[0106] Based on the relative similarity of each candidate solution Sort the solutions and select the one with the highest relative similarity as the overall optimal solution:

[0107] .

[0108] in, A finalThis represents the final, optimal overall design scheme.

[0109] This embodiment obtains a comprehensive optimal solution through non-uniform valve core parameter design, BPNN surrogate modeling, NSGA-II optimization, and TOPSIS point selection. The comprehensive optimal design solution is compared with a Tesla valve-type microchannel heat sink using a uniform valve core size. The optimization results, compared with the original baseline structure, show improvements in average Nusselt number, system voltage drop, and comprehensive performance evaluation factor. PEC The improvements are evident in various metrics, including: comparing the performance of the optimal design with that of a Tesla-type microchannel heat sink using a uniform valve core size; the comparison metrics include the average Nusselt number. Nu System voltage drop Δ P and comprehensive performance evaluation factors PEC By comparing the performance of the optimized non-uniform valve core structure with that of the uniform valve core structure, the improvement effect of the non-uniform valve core structure in terms of heat transfer capacity, flow resistance and overall performance is quantitatively evaluated.

[0110] The original reference structure is based on a Tesla-type microchannel heat sink with uniform valve core size.

[0111] By comparison, it can be found that the non-uniform valve core structure obtained by the method of this embodiment can effectively control the flow resistance while ensuring high heat exchange capacity, thus exhibiting better overall heat dissipation performance.

[0112] As shown in Table 1, the optimal Tesla valve-type microchannel obtained using the method in this embodiment has higher average Nusselt number and system voltage drop than the original baseline structure, but its comprehensive performance evaluation factor is lower. PEC The increase to 1.50 indicates that the optimization scheme significantly improves the overall flow heat transfer performance while incurring a certain pressure reduction cost.

[0113] Table 1 Comparison of heat transfer performance parameters of microchannels with different structures

[0114]

[0115] like Figure 7 As shown, compared with the original structure, the optimized non-uniform valve core structure obtained by the method in this embodiment can significantly reduce the range of the high-temperature region at the bottom and improve the uniformity of temperature distribution. This indicates that the multi-objective optimization design method proposed in this embodiment can not only obtain the optimal compromise solution in a mathematical sense, but also demonstrate better thermal management performance in actual heat sink heat dissipation.

[0116] In summary, this embodiment establishes a periodic calculation model for the heat sink of a Tesla valve-type microchannel, and combines it with non-uniform valve core parameter design, sample database construction, backpropagation neural network surrogate modeling, non-dominated sorting genetic algorithm II multi-objective optimization, and TOPSIS comprehensive decision-making to form a complete multi-objective optimization design method for non-uniform valve core structures. This method can significantly improve design efficiency, reduce the cost of repetitive numerical simulation calculations, and achieve an effective balance between the average Nusselt number and the system pressure drop, showing good prospects for engineering applications.

[0117] A Tesla valve-type microchannel heat sink optimization design system is designed using the above-mentioned optimization design method, including: a parameter input module, a surrogate model storage module, a multi-objective optimization module, a decision evaluation module, and a result output module.

[0118] The parameter input module is used to set the width of the valve core top. W t Valve core height H and inlet speed u in The preset value range.

[0119] The proxy model storage module communicates with the parameter input module to store pre-built performance prediction proxy models. These performance prediction proxy models are backpropagation neural network models, which have a built-in mapping relationship between input variables and output performance indicators. This mapping is used to determine the performance based on the input valve core top width. W t Valve core height H and inlet speed u in Output the corresponding Nusel number. Nu and system voltage drop Δ P .

[0120] The multi-objective optimization module communicates with the surrogate model storage module to receive the preset value range output by the parameter input module and to call the performance prediction surrogate model in the surrogate model storage module, using Nusselt numbers. Nu Maximize system voltage drop Δ P Minimization is the optimization objective. A multi-objective optimization algorithm is used to find the optimal solution for the design variables and obtain the Pareto optimal solution set.

[0121] The decision evaluation module is connected to the multi-objective optimization module and is used to receive the Pareto optimal solution set output by the multi-objective optimization module. The Pareto optimal solution set is comprehensively evaluated by the approximation ideal solution ranking method, and the scheme with the highest relative closeness is selected as the comprehensive optimal design scheme.

[0122] The result output module is communicatively connected to the decision evaluation module. It receives the comprehensive optimal design scheme output by the decision evaluation module and outputs the valve core top width corresponding to the comprehensive optimal design scheme. W t Valve core height H and inlet speed u in .

[0123] The proxy model storage module is constructed as follows: sample points are generated within a preset range of values ​​for the optimization design variables using the Latin hypercube sampling method, and the Nusselt number corresponding to each set of sample points is obtained through numerical simulation. Nu and system voltage drop Δ P A sample database is constructed, and a backpropagation neural network model is trained based on the sample database, so that the backpropagation neural network model has a built-in mapping relationship between input variables and output performance indicators.

[0124] The above-disclosed embodiments are merely preferred embodiments of the present invention. However, the embodiments of the present invention are not limited thereto, and any variations that can be conceived by those skilled in the art should fall within the protection scope of the present invention.

Claims

1. A Tesla valve-type microchannel heat sink optimization design method, characterized in that, Includes the following steps: Obtain the optimization design variables and their preset value ranges, including the cooling medium inlet velocity. u in And the valve core top width that varies non-uniformly along the flow direction in the Tesla valve-type microchannel heat sink. W t and valve core height H ; The optimized design variables and their preset value ranges are input into a pre-constructed performance prediction proxy model. This performance prediction proxy model is a backpropagation neural network model, which has a built-in mapping relationship between input variables and output performance indicators, used to determine the input valve core top width. W t Valve core height H and inlet speed u in Output the corresponding Nusel number. Nu and system voltage drop Δ P ; The output Nusel number Nu Maximize system voltage drop Δ P With minimization as the optimization objective, a multi-objective optimization algorithm is used to find the optimal solution set for the design variables and obtain the Pareto optimal solution set. The Pareto optimal solution set is comprehensively evaluated by the approximation ideal solution sorting method. Each solution in the Pareto optimal solution set is taken as a candidate scheme. The Euclidean distance from each candidate scheme to the positive ideal solution and the negative ideal solution is calculated. The relative closeness of each candidate scheme is calculated based on the Euclidean distance. The candidate scheme with the largest relative closeness is selected as the comprehensive optimal design scheme. Output the valve core top width corresponding to the comprehensive optimal design scheme. W t Valve core height H and inlet speed u in This is the final design result of the Tesla valve-type microchannel heat sink non-uniform valve core structure.

2. The Tesla valve-type microchannel heat sink optimization design method according to claim 1, characterized in that, The mapping relationship between the input variables and output performance metrics built into the backpropagation neural network model is established in the following way: The Latin hypercube sampling method is used to generate sample points within the preset value range of the optimization design variables, and the Nusselt number corresponding to each set of sample points is obtained through numerical simulation. Nu and system voltage drop Δ P A sample database is constructed, and the backpropagation neural network model is trained based on the sample database. The numerical simulation uses a periodic calculation model as the analysis object. The periodic calculation model is obtained from the actual microchannel overall model and is used to reflect the periodic flow and heat transfer characteristics inside the complete microchannel. The Latin hypercube sampling generates 60 sets of sample points; the sample database is divided into a training set, a validation set, and a test set for training and performance evaluation of the backpropagation neural network model.

3. The Tesla valve-type microchannel heat sink optimization design method according to claim 1, characterized in that, The Tesla valve-type microchannel heat sink's basic structure uses an elliptical valve core arranged in an incremental manner along the flow direction.

4. The Tesla valve-type microchannel heat sink optimization design method according to claim 1, characterized in that, The multi-objective optimization algorithm is the non-dominated sorting genetic algorithm II.

5. The Tesla valve-type microchannel heat sink optimization design method according to claim 1, characterized in that, The top width of the valve core W t The value range is 0.20mm~0.60mm, and the valve core height H The value range is 0.50mm~1.10mm, and the inlet velocity... u in The value range is 0.60 m / s to 1.80 m / s.

6. The Tesla valve-type microchannel heat sink optimization design method according to claim 1, characterized in that, The backpropagation neural network model is a three-layer feedforward neural network structure, with the input layer corresponding to the width of the top of the valve core. W t Valve core height H and inlet speed u in The output layer corresponds to the Nusselt number. Nu and system voltage drop Δ P The hidden layer activation function uses the Tansig function, the output layer activation function uses the Purelin function, and the training algorithm uses the Bayesian regularization algorithm.

7. The Tesla valve-type microchannel heat sink optimization design method according to claim 1, characterized in that, Positive Ideal Solution A + and negative ideal solution A - satisfy: ； 。 8. The Tesla valve-type microchannel heat sink optimization design method according to claim 1, characterized in that, Also includes: The performance of the aforementioned optimal design scheme was compared with that of a Tesla valve-type microchannel heat sink using a uniform valve core size. The comparison metrics included the average Nusselt number. Nu System voltage drop Δ P and comprehensive performance evaluation factors PEC .

9. A Tesla valve-type microchannel heat sink optimization design system, characterized in that, Designing using the optimization design method according to any one of claims 1 to 8, comprising: The parameter input module is used to set the width of the valve core top. W t Valve core height H and inlet speed u in The preset value range; The proxy model storage module, communicatively connected to the parameter input module, is used to store pre-built performance prediction proxy models. These performance prediction proxy models are backpropagation neural network models, which have a built-in mapping relationship between input variables and output performance indicators. This is used to store the pre-built performance prediction proxy models based on the input valve core top width. W t Valve core height H and inlet speed u in Output the corresponding Nusel number. Nu and system voltage drop Δ P ; A multi-objective optimization module, communicatively connected to the proxy model storage module, receives a preset value range output by the parameter input module and calls the performance prediction proxy model in the proxy model storage module, using Nusselt numbers. Nu Maximize system voltage drop Δ P With minimization as the optimization objective, a multi-objective optimization algorithm is used to find the optimal solution for the design variables and obtain the Pareto optimal solution set. The decision evaluation module is communicatively connected to the multi-objective optimization module. It is used to receive the Pareto optimal solution set output by the multi-objective optimization module, and to comprehensively evaluate the Pareto optimal solution set by the approximation ideal solution sorting method, and select the scheme with the highest relative closeness as the comprehensive optimal design scheme. The result output module, communicatively connected to the decision evaluation module, receives the comprehensive optimal design scheme output by the decision evaluation module and outputs the valve core top width corresponding to the comprehensive optimal design scheme. W t Valve core height H and inlet speed u in .

10. The Tesla valve-type microchannel heat sink optimization design system according to claim 9, characterized in that, The proxy model storage module is constructed in the following way: sample points are generated within a preset range of values ​​for the optimization design variables using the Latin hypercube sampling method, and the Nusselt number corresponding to each set of sample points is obtained through numerical simulation. Nu and system voltage drop Δ P A sample database is constructed, and the backpropagation neural network model is trained based on the sample database.

Images