A Smart Optimization Method for the Microstructure of a Heat Pipe Swiper

Abstract

This invention provides an intelligent optimization method for the microstructure of a heat pipe wick, relating to the fields of thermal management devices and microstructure optimization technology. The optimization method includes the following steps: S1, establishing a structural parameter-thermal performance index mapping dataset; S2, constructing a surrogate prediction model; S3, establishing a multi-objective optimization model; S4, performing multi-objective optimization based on the dolphin echolocation algorithm; and S5, determining the optimal solution. Based on this, with minimizing temperature difference, minimizing thermal resistance, and maximizing equivalent thermal conductivity as optimization objectives, multi-objective collaborative optimization is performed on numerous design parameters of the wick microstructure. This solves the problems of long design cycles and high costs, difficulty in balancing multiple objectives, difficulty in characterizing nonlinear relationships, difficulty in handling discrete variables, and susceptibility to local optima in existing microstructure design methods.

Description

Technical Field

[0001] This invention relates to the field of thermal management devices and microstructure optimization technology, specifically to a method for intelligent optimization of the microstructure of a heat pipe wick. Background Technology

[0002] Among thermal management devices, ultrathin planar heat pipes have been widely used in thin and light electronic devices such as smartphones, tablets, and laptops due to their advantages such as small size, high thermal conductivity, and excellent temperature uniformity. In recent years, with the rapid development of electronic devices towards high integration and high power consumption, the heat flux density of their internal core chips has increased dramatically, thus placing higher demands on the heat transfer performance of ultrathin planar heat pipes. In ultrathin planar heat pipes, the wick is the core component that determines the heat transfer performance; specifically, the main function of the wick is to use capillary force to drive the condensing working fluid from the condensation section back to the evaporation section, thereby maintaining the continuous gas-liquid circulation. The capillary performance of the wick is closely related to its microstructure, so a reasonable microstructure design can significantly improve capillary force and permeability, reduce the thermal resistance and temperature difference of the heat pipe, and improve the equivalent thermal conductivity. Currently, the wick of copper-based ultrathin planar heat pipes is mostly processed by laser etching, which involves etching multiple parallel channels on a copper substrate as the basic flow channel structure. In order to further enhance capillary performance, existing designs add structures such as micropillars and microribs within the channels.

[0003] However, existing methods for designing microstructures of absorbent cores have the following problems: First, relying on experience and trial and error leads to high time and material costs. Specifically, in the traditional process of capillary wick microstructure design, designers typically rely on experience or single-factor experimental methods to determine microstructure design parameters and optimize them through repeated trial-and-error cycles of "design-processing-testing." For example, Chinese invention patent CN119043057B provides a method for designing capillary wick structures and discloses a method for obtaining parameters through trial and error experiments, namely, "performing performance tests on the obtained capillary wick structure...; if the conditions are met, the designed capillary wick structure meets the standard; if the conditions are not met, the diameter of the support column, the length, width, and height of the unit, or the number of layers in the layer structure are further optimized, and preparation and testing are carried out until the conditions are met." Those skilled in the art understand that the fabrication of wick microstructures involves precision manufacturing processes such as laser etching. Each parameter adjustment requires re-processing of the sample and thermal performance testing. Therefore, for wick microstructures that include numerous design parameters such as channel depth, channel width, micropillar shape (circular, triangular, rhomboid, square, etc.), micropillar position, and micropillar spacing, the aforementioned trial-and-error method leads to long experimental cycles and high material consumption. Moreover, when facing multi-parameter combination optimization problems, it is difficult to achieve global optimization within a limited number of experiments. Therefore, if this problem cannot be addressed, a vicious cycle of "increased number of experiments - extended R&D cycle - increased costs - low design efficiency" will be formed, resulting in product iteration speeds that cannot meet the heat dissipation requirements of the rapidly evolving electronic devices.

[0004] Secondly, it involves conflicting metrics, making it difficult to achieve an optimal balance. Specifically, evaluating the heat transfer performance of heat pipes involves multiple conflicting metrics. For example, reducing thermal resistance and increasing capillary force often impose opposing requirements on the microstructure geometry, while minimizing temperature difference and maximizing equivalent thermal conductivity also have an inverse relationship. Traditional single-factor or empirical design methods struggle to achieve an optimal balance among multiple objectives, easily leading to the optimization of one performance metric at the expense of another. For instance, excessively pursuing low thermal resistance may result in decreased wick permeability, causing the evaporation section to dry out; while excessively pursuing high capillary force may increase flow resistance and reduce heat transfer efficiency. Therefore, if this multi-objective conflict problem cannot be effectively resolved, the overall performance of the heat pipe will be difficult to optimize, thus failing to meet the heat dissipation requirements of high-power electronic devices.

[0005] Third, the complex interactions of parameters make accurate prediction difficult. Specifically, multiple structural parameters of the wick (such as channel depth, channel width, and micropillar shape) exhibit complex nonlinear coupling relationships, resulting in strong nonlinear effects on temperature difference, thermal resistance, and equivalent thermal conductivity. Traditional prediction methods based on empirical formulas or simplified theoretical models struggle to accurately characterize these nonlinear mapping relationships, leading to poor design prediction accuracy and low reliability. For example, changing the channel depth not only affects capillary force but also alters the support effect of the micropillars and the vapor flow space. Furthermore, this effect is coupled with parameters such as micropillar shape and spacing, making a single empirical formula insufficient to comprehensively describe it. This lack of a nonlinear coupling mechanism causes significant discrepancies between simulation results and actual thermal performance, thus failing to accurately reflect the dynamic behavior of the wick.

[0006] Fourth, the variables are of mixed types and existing algorithms have significant limitations. Specifically, although recent studies have attempted to use neural network models to establish predictive models between process parameters and heat pipe performance, and to use evolutionary algorithms such as genetic algorithms (GA) and particle swarm optimization (PSO) for parameter optimization, the following shortcomings still exist: First, the microstructural parameters of the wick include not only continuous variables such as channel depth and width, but also discrete variables such as micropillar shapes (e.g., circles, triangles, rhombuses, squares, etc.). Traditional gradient optimization methods or optimization algorithms based on continuous variables (such as standard particle swarm optimization and differential evolution algorithms) are difficult to directly handle discrete variables, and often require indirect approximation through binary encoding or penalty function methods, which can lead to information loss, encoding redundancy, and difficulty in achieving efficient collaborative optimization in a mixed space of discrete and continuous variables. Secondly, traditional evolutionary algorithms such as genetic algorithms and particle swarm optimization have adaptive defects in multi-parameter, multi-objective, and mixed-variable optimization of microstructures of wicking cores. That is, in the later stages of iteration, they often get stuck in local optima due to the loss of population diversity, making it difficult to guarantee the discovery of the global optimal solution. For example, genetic algorithms are prone to premature convergence when dealing with multi-peak optimization problems, and particle swarm optimization is prone to problems such as particle convergence and search stagnation in high-dimensional spaces.

[0007] Furthermore, in addition to the aforementioned technical issues, existing microstructure design methods suffer from insufficient coordination among various stages. Specifically, current methods lack an effective closed-loop coordination mechanism between experimental testing, surrogate modeling, optimization, and result verification. This often leads to simulation optimization results deviating from actual processing conditions, resulting in limited engineering guidance value. For example, the optimal parameter combination obtained through optimization may exceed the processing accuracy range of laser etching equipment, or the micropillar shape may be too complex to reliably manufacture. This disconnect between optimization and manufacturing further restricts the application value of liquid-absorbing core microstructure design methods in practical engineering.

[0008] In summary, there is an urgent need for a new method that can overcome the above-mentioned shortcomings and achieve efficient, precise, and intelligent multi-objective optimization of the liquid wick microstructure to meet the growing heat dissipation requirements of high-performance electronic devices. Summary of the Invention

[0009] This invention provides an intelligent optimization method for heat pipe wick microstructures. With minimizing temperature difference, minimizing thermal resistance, and maximizing equivalent thermal conductivity as optimization objectives, it performs multi-objective collaborative optimization on numerous design parameters of the microstructure. This overcomes the problems of long design cycles and high costs, difficulty in balancing multiple objectives, difficulty in characterizing nonlinear relationships, difficulty in handling discrete variables, and easy getting trapped in local optima in existing microstructure design methods, thereby achieving efficient, accurate, and intelligent design of wick microstructures.

[0010] This invention is achieved using the following technical solution: A method for intelligent optimization of the microstructure of a heat pipe wick includes the following steps: S1. Establish a structural parameter-thermal performance index mapping dataset: First, determine which structural parameters are included in the combination of structural parameters of the absorbent core microstructure, so as to clarify the number and type of parameters that can affect the thermal performance considered in this optimization method; then, set up multiple sets of structural parameter combinations with different specific values ​​of each structural parameter, process absorbent core samples according to the multiple sets of structural parameter combinations and conduct thermal performance tests, thereby obtaining multiple sets of thermal performance index combinations corresponding to the multiple sets of structural parameter combinations, so as to form a structural parameter-thermal performance index mapping dataset. S2. Constructing a proxy prediction model: Based on the structural parameter-thermal performance index mapping dataset, train a proxy prediction model; in the proxy prediction model, the combination of structural parameters is the input, and the combination of thermal performance indices is the output; this proxy prediction model can replace real thermal performance testing and is used to quickly predict the corresponding thermal performance index given structural parameters, that is, to realize the rapid mapping from structural parameters to thermal performance indices. S3. Establish a multi-objective optimization model: Using the combination of structural parameters as optimization variables and the objective function composed of the combination of thermal performance indicators as the optimization objective, establish a multi-objective optimization model for the liquid absorption core microstructure; wherein, the combination of thermal performance indicators in the objective function comes from the prediction results of the surrogate prediction model in S2; S4. Multi-objective optimization based on Dolphin Echolocation (DE): The multi-objective optimization model is solved using the Dolphin Echolocation (DE) algorithm to obtain the Pareto optimal solution set; S5. Determine the optimal solution: Select the combination of structural parameters with the best overall performance from the Pareto optimal solution set, and use it as the final design scheme.

[0011] Furthermore, the combination of structural parameters includes channel depth, channel width, micropillar position (the position of the micropillar within the channel), and micropillar spacing (the spacing between micropillars) as continuous variables, and micropillar shape as discrete variables; native encoding mapping is performed on the discrete variables, that is, different discrete variable options are encoded using continuous integers, wherein the discrete variable options (i.e., micropillar shape types) can specifically be circles, triangles, rhombuses, squares, or other shapes known to those skilled in the art.

[0012] Furthermore, the combination of thermal performance indicators includes heat transfer temperature difference, thermal resistance, and equivalent thermal conductivity; the optimization direction of the multi-objective optimization model is to minimize the heat transfer temperature difference, minimize the thermal resistance, and maximize the equivalent thermal conductivity; by taking the reciprocal of the equivalent thermal conductivity, the optimization direction of the multi-objective optimization model is unified to minimizing the objective function. Thus, the conflicting relationships between the objective indicators are resolved, and an optimal balance can be achieved among multiple objectives through a unified optimization direction.

[0013] Furthermore, in step S4, when updating the individual location based on the dolphin echolocation algorithm, an echolocation mechanism is employed. Specifically, the update result of the discrete variables is processed using discrete mapping rules, namely, rounding and interval restriction operations. The restricted interval is determined by the numerical values ​​of the encoded consecutive integers. Therefore, binary encoding or penalty function approximation is not required when processing discrete variables, achieving unified optimization for both discrete and continuous variables.

[0014] Furthermore, in S4, an encirclement-predation strategy is adopted: the search range is expanded in the early stage of the dolphin echolocation algorithm iteration for global exploration; the search range is narrowed in the later stage of the dolphin echolocation algorithm iteration for local refinement. This allows the algorithm to automatically balance global exploration and local refinement during the iteration process, effectively avoiding premature convergence to a local optimum.

[0015] Furthermore, in step S4, a random proxy mechanism is employed: a subset of individuals is randomly selected with a preset probability, allowing these individuals to perform random walks within the search space without updating their positions as per the echolocation mechanism. This introduces a perturbation, enhancing the algorithm's ability to escape local extremes and further improving global optimization performance.

[0016] Furthermore, in S4, when initializing the algorithm: for the continuous variables, a uniform random initialization method is used; for the discrete variables, a random integer initialization method is used, that is, the variables are selected with equal probability from the continuous integers to be encoded.

[0017] Furthermore, in S1, the specific values ​​of each structural parameter in the combination of structural parameters conform to the range of values; the constraints of the range of values ​​include the processing capability of the microstructure processing equipment for the liquid absorption core, the geometric feasibility of the liquid absorption core, and other factors known to those skilled in the art.

[0018] Furthermore, the process includes step S6, experimental verification of the optimal solution: processing the wicking sample according to the optimal combination of structural parameters and conducting thermal performance tests, comparing the test results with the prediction results of the surrogate prediction model obtained based on the optimal combination of structural parameters. This verifies the effectiveness and accuracy of the optimization method. Step S6, combined with the aforementioned steps S1-S5, forms a closed-loop collaborative mechanism, which helps ensure the matching degree between the prediction results and actual processing conditions.

[0019] Further, step S2 includes the following sub-steps: S21. Data preprocessing: The structural parameter-thermal performance index mapping dataset is normalized and divided into training set and test set according to the ratio. S22. Network Training: The agent prediction model is trained using a numerical optimization algorithm; S23. Model Validation: Use the test set to validate the accuracy of the trained agent prediction model.

[0020] The beneficial effects achieved by this invention are: A method for intelligent optimization of the microstructure of a heat pipe wick, through limiting steps S1-S5, can achieve the following improvements compared to existing technologies: First, it significantly shortens the design cycle and reduces R&D costs: In this invention, a high-precision surrogate prediction model replaces a large number of time-consuming and labor-intensive physical tests, shortening the original "design-processing-testing" trial-and-error cycle, which originally required weeks or even months, to a computer simulation optimization process measured in hours. Furthermore, only a small amount of experimental data is needed to train the surrogate prediction model, and subsequent multi-objective optimization stages are based on this model, eliminating the need for additional physical experiments. Therefore, it can significantly reduce the time required for R&D, material consumption, and testing costs, helping to accelerate product iteration.

[0021] Secondly, it helps achieve an optimal balance among multiple objectives: In this invention, minimizing temperature difference, minimizing thermal resistance, and maximizing equivalent thermal conductivity are the optimization objectives. This unifies the optimization direction of the multi-objective optimization model to the minimization of the objective function. The dolphin echolocation algorithm is used for multi-objective collaborative optimization, automatically finding the Pareto optimal solution set. Then, the combination of structural parameters with the best overall performance can be selected from the Pareto solution set. This resolves the multi-objective conflict problem. Based on the obtained optimal combination of structural parameters, the overall performance of the heat pipe is optimized, thus meeting the high heat dissipation requirements of high-power electronic devices.

[0022] Third, it characterizes the complex relationships between parameters and achieves accurate prediction results: In this invention, driven by experimental data, and through sufficient training and verification, the surrogate prediction model can autonomously learn and accurately capture the complex nonlinear mapping relationship between structural parameters and thermal performance indicators. This effectively improves prediction accuracy and reliability, thus truly reflecting the dynamic behavior of the wicking core.

[0023] Fourth, achieving unified and efficient optimization of discrete and continuous variables: In this invention, the echolocation mechanism of the dolphin echolocation algorithm is used to perform native encoding mapping on discrete variables, and discrete mapping rules are used in the position update process, eliminating the need for binary conversion or penalty function approximation. This achieves seamless and efficient collaborative optimization of mixed variables, avoiding information loss and computational redundancy, and significantly improving the efficiency and accuracy of discrete variable optimization.

[0024] Fifth, it effectively avoids local optima and has strong global optimization capabilities: This invention uses the dolphin echolocation algorithm, which, through its unique encirclement predation strategy and random proxy mechanism, can intelligently balance global exploration and local exploitation, effectively maintain population diversity, and significantly enhance the ability to escape local optima. Therefore, compared with traditional evolutionary algorithms such as genetic algorithms and particle swarm optimization, this method significantly improves convergence accuracy and demonstrates superior stability and robustness in the multi-parameter, multi-objective, and mixed-variable problem of wick microstructure design and optimization.

[0025] In addition, step S6 can be optionally included in this invention, namely, verifying the optimal solution through experiments. This forms a closed-loop design system of "experiment-modeling-optimization-verification," improving the synergistic optimization effect among each stage. The optimal combination of structural parameters obtained through optimization needs to be verified through actual processing and thermal performance testing, and the verification results are fed back to the dataset for iterative updates of the surrogate prediction model. This achieves closed-loop collaboration between design and verification, effectively avoiding the disconnect between optimization and manufacturing, thus making the optimization results more practically valuable in engineering. Attached Figure Description

[0026] Figure 1 This is a schematic diagram of the steps of the optimization method described in the embodiment of the present invention; Figure 2 This is an overall algorithm flowchart of the optimization method described in the embodiments of the present invention; Figure 3 This is a comparison chart of the predicted heat transfer temperature difference and the test temperature difference of the optimization method described in the embodiments of the present invention; Figure 4 This is a comparison diagram of the predicted thermal resistance and the tested thermal resistance of the optimization method described in the embodiments of the present invention; Figure 5This is a comparison chart of the predicted equivalent thermal conductivity and the tested equivalent thermal conductivity of the optimization method described in the embodiments of the present invention. Detailed Implementation

[0027] The technical solutions of the present invention will be clearly and completely described below with reference to the accompanying drawings in the embodiments of the present invention.

[0028] Example 1 This embodiment provides a method for intelligent optimization of the microstructure of a heat pipe wick. Please refer to [link / reference]. Figure 1 and Figure 2 It includes the following steps: S1. Establish a structural parameter-thermal performance index mapping dataset.

[0029] This step aims to obtain high-quality raw data for building the surrogate prediction model through a small number of efficient physical experiments. Specifically, it includes: S11. Determine the combination of structural parameters for the microstructure of the absorbent core: Several key structural parameters of the absorbent core microstructure were selected to form a combination of structural parameters, including the channel depth as a continuous variable. Channel width Microcolumn position (The position of the micropillars within the channel, such as the distance from the center of the micropillar to one side wall of the channel) and the micropillar spacing. (The spacing between the micropillars), and the shape of the micropillars as a discrete variable. In this embodiment, the micro-pillar shapes available as discrete variable options include circles, triangles, rhombuses, and squares.

[0030] S12. Generate the test plan and conduct the test: First, orthogonal experimental design or uniform experimental design methods are used to generate multiple combinations of structural parameters with different values ​​based on the range of values ​​for each structural parameter. The constraints on the value range include the processing capability of the microstructure fabrication equipment (e.g., the processing accuracy of laser etching equipment) and the geometric feasibility of the microstructure.

[0031] Subsequently, multiple wicking core samples were fabricated according to the experimental plan, and an ultra-thin planar heat pipe testing platform (a conventional testing device already in use in this field) was built. The thermal performance parameters of each wicking core sample were measured under the same heating power and cooling conditions. In this embodiment, for each combination of structural parameters, the experiment was repeated three times and the average value was taken to eliminate random errors.

[0032] S13. Form a mapping dataset: This involves integrating various structural parameter combinations with their corresponding thermal performance index combinations. These thermal performance index combinations include the heat transfer temperature difference. Thermal resistance and equivalent thermal conductivity In addition, native encoding mapping is performed on the micro-cylinder shape as a discrete variable, that is, different discrete variable options are encoded using consecutive integers; in this embodiment, circles, triangles, rhombuses, and squares are encoded as 1, 2, 3, and 4, respectively.

[0033] The resulting structural parameter-thermal performance index mapping dataset The expression is as follows: , Where D is the structural parameter-thermal performance index mapping dataset, For the first Combination of structural parameters for each aspiration core sample; They are respectively mapping data and Corresponding heat transfer temperature difference Thermal resistance and equivalent thermal conductivity ; This represents the total number of aspiration core samples.

[0034] S2. Construct a proxy prediction model.

[0035] Because physical experiments are costly and difficult to directly provide a large number of fitness evaluations for optimization algorithms, surrogate prediction models are constructed to replace real experiments, achieving a rapid and accurate mapping from structural parameters to thermal performance indicators. Specifically, this includes: S21. Data Preprocessing: The structural parameter-thermal performance index mapping dataset is normalized and divided into training and test sets proportionally. In this embodiment, the dataset is... The training set was divided into two parts at an 8:2 ratio. and test set The input structural parameters and output thermal performance indicators are normalized to map all data to the [0,1] interval, thus eliminating the influence of different dimensions and accelerating network convergence. The normalization formula is as follows: , in, This is the original data. For the normalized data, and These are the maximum and minimum values ​​of the variable, respectively.

[0036] S22, Network Training; In this embodiment, a three-layer backpropagation neural network (BPNN) is used as the basic architecture of the surrogate prediction model. The input layer contains 5 nodes, corresponding to 5 structural parameters, namely the channel depth. Channel width Microcolumn position Micro-column spacers Micro-column shape The number of nodes in the hidden layer is determined through trial and error or grid search, and the ReLU function is used as the activation function. The output layer contains 3 nodes, corresponding to 3 thermal performance indicators, namely heat transfer temperature difference. Thermal resistance and equivalent thermal conductivity .

[0037] To improve the prediction accuracy and generalization ability of the model, the Levenberg-Marquardt algorithm was used to train a three-layer backpropagation neural network, with mean squared error (MSE) as the loss function. During training, a test set was used. Monitor the generalization performance of the model when the test set If the error no longer decreases after 10 consecutive iterations, training should be terminated early to prevent overfitting.

[0038] S23. Model Validation: Using the test set The trained surrogate prediction model is validated. In this embodiment, the coefficient of determination between the model's predicted values ​​and the actual experimental values ​​is calculated. and root mean square error RMSE, when Furthermore, if the RMSE is lower than a preset threshold (e.g., 5% of the average thermal performance index), the surrogate prediction model is considered to meet the accuracy requirements and can be used in subsequent optimization processes. Wherein: RMSE= , In the formula, H is the total number of samples participating in the validation. For the h-th sample, This is the model prediction value for the h-th sample; the root mean square error (RMSE) measures the average standard deviation of the error between the predicted value and the true value, and the smaller the value, the higher the prediction accuracy. , In the formula, The average of all true experimental values, i.e. The coefficient of determination It is used to reflect the ability of the surrogate prediction model to explain the variation of real data. The value range is [0,1]. The closer it is to 1, the better the model fits the real value.

[0039] S3. Establish a multi-objective optimization model. The evaluation of heat pipe heat transfer performance involves conflicting indicators; a multi-objective optimization model is used to achieve synergy. Specifically, this includes: A multi-objective optimization model for the wick microstructure was established, using five structural parameters as optimization variables and three thermal performance indices as the objective function. The heat transfer temperature difference was used as the optimization objective. Thermal resistance For a parameter where smaller is better, the equivalent thermal conductivity is... Since a larger value is always better, the optimization direction of the multi-objective optimization model should be to minimize the heat transfer temperature difference, minimize the thermal resistance, and maximize the equivalent thermal conductivity. Based on this, to unify the optimization direction, the equivalent thermal conductivity is taken as its reciprocal, thus transforming the problem into a minimization problem.

[0040] The expression for the multi-objective optimization model is as follows: , The above formula represents the search for a set of optimization variables b such that the three objective functions , , The target vector F is composed of The overall goal is to minimize the value; where T is the transpose symbol; b is the optimization variable vector, corresponding to the combination of structural parameters. ; , , These are three performance metrics to be minimized. , , , , , The surrogate prediction model is based on the combination of structural parameters. The predicted output includes temperature difference, thermal resistance, and equivalent thermal conductivity.

[0041] In addition, based on the processing capabilities of the microstructure fabrication equipment for the liquid suction core and the geometric feasibility of the liquid suction core, the optimization variables must meet the following constraints: Channel depth : , Channel width : , Microcolumn position : , micropillar spacers : , micro-column shape : ∈{1,2,3,4}; In this embodiment, the following example is given: The minimum controllable depth of laser etching equipment is 0.05mm. Excessive depth can lead to a decrease in capillary force or structural collapse. ; Balancing the geometric feasibility of capillary flow space and steam reflux channel, making ; The distance from the center of the microcolumn to one side wall of the channel should be such that it touches the wall or exceeds the channel width. ; Considering that too small a spacing between micropillars can easily cause clogging, while too large a spacing will reduce the capillary enhancement effect, ; ∈{1,2,3,4}, where code 1 corresponds to a circle, code 2 corresponds to a triangle, code 3 corresponds to a rhombus, and code 4 corresponds to a square.

[0042] S4. Multi-target optimization based on the dolphin echolocation algorithm (hereinafter referred to as the DE algorithm).

[0043] The DE algorithm achieves global optimization by simulating the echolocation mechanism, encirclement predation strategy, and random proxy mechanism of dolphins. Its core advantage lies in its ability to directly handle discrete variables and effectively avoid local optima. Specifically, it includes: S41, Algorithm Initialization: First, set the control parameters for the DE algorithm: population size. (Recommended value: 50-100), Maximum number of iterations (Recommended value: 200-500), sound wave attenuation coefficient Encirclement coefficient Random proxy probability (Recommended value: 0.1-0.2).

[0044] Then, the population is randomly initialized within the search space. Among them, for continuous variables (channel depth) Channel width Microcolumn position Micro-column spacers Uniform random initialization is used; for discrete variables (micro-cylinder shape) ), and is initialized with random integers, that is, selected with equal probability from {1,2,3,4}.

[0045] S42. Fitness Evaluation: For population Each individual in ( For the first (Combination of structural parameters of each aspiration core sample), call the surrogate prediction model trained in step S2, and calculate its corresponding three objective function values. That is, predicting the heat transfer temperature difference Predicting thermal resistance and predicted equivalent thermal conductivity .

[0046] The Pareto dominance relation is used to evaluate the merits of individuals. The evaluation logic of this Pareto dominance relation is as follows: if an individual... Non-inferior in all objectives And superior to at least one objective Then it is called Dominate In this fitness evaluation process, non-dominated solutions (i.e., solutions not dominated by any other individual) are obtained based on the above evaluation logic, and these non-dominated solutions are stored in an external archive. middle. Specifically: The Pareto dominance relation is a fundamental criterion used in multi-objective optimization to compare the merits of two solutions (i.e., two sets of microstructure parameter combinations). In this method, each solution corresponds to a set of structural parameter combinations. (i.e., including channel depth) Channel width Microcolumn position Micro-column spacers Micro-column shape Its merits are determined by three objective functions (i.e. (To be decided jointly)

[0047] The specific logic of this dominance relationship is as follows: If the solution... Non-inferior to the solution on all objectives ,Right now , , Simultaneously established; and resolved Superior to the solution at least in one objective That is, it exists If z∈{1,2,3}, then we say that... Dominate ,mean The corresponding microstructure design scheme is superior in overall heat transfer performance. The corresponding microstructure design scheme.

[0048] During algorithm execution, a non-dominated solution is a solution in the current population that is not dominated by any other individual; that is, no other solution is inferior to this non-dominated solution on all objectives and superior to it on at least one objective. Non-dominated solutions represent a set of Pareto-optimal microstructure designs found so far, and these solutions will be stored in an external archive. This serves as the basis for the final selection of the optimal combination of structural parameters.

[0049] In this embodiment, the following example is given: Consider the following two sets of structural parameter combinations: : =0.12mm =0.25mm =0.12mm =0.30mm =1 (circle); : =0.15mm =0.30mm =0.15mm =0.40mm =2 (triangles); If predicted by the proxy prediction model, The corresponding predicted heat transfer temperature difference and thermal resistance are both lower than The corresponding predicted heat transfer temperature difference and thermal resistance, and The corresponding predicted equivalent thermal conductivity reciprocal (i.e. It is also lower than (meaning the equivalent thermal conductivity is higher). The reciprocal of the corresponding predicted equivalent thermal conductivity is called... Dominate .

[0050] S43. Individual location update based on echolocation and random agent mechanisms: One of the core mechanisms of the DE algorithm is echolocation, which simulates the behavior of dolphins locating prey by emitting sound waves and receiving echoes. In this method, this mechanism can be used for individual location updates, especially for the direct processing of discrete variables. This approach eliminates the need for binary encoding or penalty function approximation, achieving unified optimization for both discrete and continuous variables. Specifically, the echolocation mechanism: For population Each individual in its new location Updated according to the following formula: , in, For individuals updated based on echolocation mechanism new location, , , A random number within the interval [0,1]; For the current population The optimal individual (non-dominated solution) in the solution; Individuals selected randomly; For the population Individual at the center; sound wave attenuation coefficient and encirclement coefficient It serves as a control coefficient, used to adjust the search step size.

[0051] Furthermore, regarding the micro-cylinder shape as a discrete variable... Furthermore, the arithmetic operations in the above update formula need to be replaced with discrete mapping rules, that is, rounding and interval restriction operations need to be performed; specifically, the calculated update result is rounded to the nearest integer and restricted to the interval [1,4], which can be expressed as: , in, micro-column shape The new location is updated based on echolocation and discrete mapping rules. It is calculated using the aforementioned position update formula.

[0052] Furthermore, to prevent premature convergence and getting trapped in local optima, the DE algorithm introduces a random proxy mechanism. This mechanism ensures that a certain proportion of individuals in the population are always exploring, effectively maintaining population diversity and enhancing the algorithm's ability to escape local extremes. Specifically, this random proxy mechanism: With random proxy probability (Recommended value: 0.1-0.2) To randomly select a subset of individuals, instead of performing echolocation-based location updates on these individuals, a random walk is performed within the search space. This can be expressed as: , in, For individuals updated based on a random agent mechanism new location, This is the random walk strength coefficient (recommended value: 0.05-0.1). These are the upper and lower bounds of the search space, respectively. It is a random vector uniformly distributed in the interval [0,1].

[0053] S44. Algorithm iteration process based on the encirclement predation strategy: The DE algorithm can simulate the cooperative behavior of dolphins surrounding prey using an encirclement strategy. This strategy allows the algorithm to automatically balance global exploration and local exploitation during iteration, avoiding the premature convergence of traditional algorithms. Specifically: In the early stages of algorithm iteration, increase the sound wave attenuation coefficient. To expand the search scope and conduct a global exploration; in the later stages of algorithm iteration, reduce To narrow down the encirclement and conduct detailed local development; The decay law is as follows: , in, This represents the current iteration number. This is the initial attenuation coefficient (recommended value is 0.9). This represents the maximum number of iterations (recommended value is 200-500).

[0054] S45. External archive update and iteration termination: After each iteration, the fitness of the individuals after the position update is evaluated again to obtain a new non-dominated solution, and then the current population is... The non-dominated solutions in the archive are merged with the existing non-dominated solutions in the external archive A, and a new set of non-dominated solutions is selected from the merged set to update the external archive A. If the size of the external archive A exceeds the preset limit (recommended value is 100), the existing crowding distance sorting strategy is used to retain the solution with the most uniform distribution.

[0055] When the maximum number of iterations is reached or continuous The algorithm terminates when the external archive A is no longer updated after the second iteration, and outputs all non-dominated solutions in the external archive A, which is the Pareto optimal solution set.

[0056] S5. Determine the optimal solution. This step aims to select the combination of structural parameters with the best overall performance from the Pareto optimal solution set, which will be used as the final design scheme. Specifically, this includes: The Pareto optimal solution set output in step S4 contains multiple sets of mutually non-dominated optimal structural parameter combinations; to select the final design scheme from these, an entropy weighting method is used for objective weighting decision-making, including: S51. Data Standardization: For each solution (combination of structural parameters) in the Pareto optimal solution set, its corresponding three objective function values ​​( The standardization process is performed; for indicators where smaller is better, the standardization formula is: , in, Let be the standardized result of the j-th optimization objective corresponding to the m-th Pareto optimal structural parameter combination; m is the m-th structural parameter combination to be evaluated in the Pareto optimal solution set; j is the index of the optimization objective. ∈{1,2,3}, the three index values ​​correspond to the heat transfer temperature difference respectively. Thermal resistance Reciprocal of equivalent thermal conductivity ; Let j be the set of objective function values ​​corresponding to all combinations of structural parameters within the Pareto optimal solution set for the j-th optimization objective; For the m-th Pareto optimal combination of structural parameters, the specific objective function value of the j-th optimization objective is obtained after substituting it into the calculation. For the j-th optimization objective, find the maximum value within the Pareto optimal solution set; For the j-th optimization objective, find the minimum value within the Pareto optimal solution set.

[0057] S52, Entropy Weight Calculation: Based on the above standardization results, the objective weights of each optimization objective are calculated using the entropy weight method, as shown in the following formula: ; in, The information entropy of the j-th optimization objective is used to measure the degree of dispersion of the objective in the Pareto optimal solution set; , is the proportion of the standardized result of the m-th Pareto optimal solution under the j-th optimization objective to the sum of the standardized results of all solutions to that objective; is the objective weight of the j-th optimization objective calculated based on information entropy. The larger the weight, the higher the discrimination and information content of the objective in multi-objective evaluation; k is the normalization coefficient, used to normalize the value of information entropy to the interval [0,1]; M is the number of solutions in the Pareto optimal solution set. The sum of the information entropy redundancy of all optimization objectives is used to normalize the weights of each objective to the [0,1] interval.

[0058] S53. Calculation of comprehensive evaluation value: Based on the entropy weights of each optimization objective, a comprehensive score is calculated for each solution in the Pareto optimal solution set, and its comprehensive evaluation value is determined by the following formula: , in, The value is the comprehensive evaluation value of the m-th Pareto optimal solution; n is the total number of optimization objectives, in this embodiment n=3, corresponding to the heat transfer temperature difference. Thermal resistance Reciprocal of equivalent thermal conductivity These are the three types of optimization objectives.

[0059] Then, select The largest solution is taken as the optimal combination of structural parameters.

[0060] S6. Experimental Verification of the Optimal Solution. To achieve closed-loop collaboration between design and verification and effectively avoid the disconnect between optimization and manufacturing, step S6 is performed. Specifically, it includes: The liquid-absorbing core sample was fabricated according to the optimal combination of structural parameters, an ultrathin planar heat pipe was assembled, and its thermal performance was tested. The test results were compared with the prediction results of the surrogate prediction model based on the optimal combination of structural parameters to verify the effectiveness of the optimization method provided in this embodiment. Please refer to... Figure 3 , Figure 4 and Figure 5The figures show comparisons between predicted and tested temperature differences, predicted and tested thermal resistance, and predicted and tested equivalent thermal conductivity. As can be seen from the figures, the errors between the actual test results and the corresponding model prediction results are small, thus proving the effectiveness of the optimization method.

[0061] In summary, the optimization method provided in this embodiment effectively solves the problems of long design cycle, high cost, difficulty in balancing multiple objectives, difficulty in characterizing nonlinearity, difficulty in handling discrete variables, and easy getting trapped in local optima in the design of wick microstructures by constructing a surrogate prediction model, establishing a multi-objective optimization framework, and introducing the DE algorithm and its unique echo localization, encirclement predation, and random surrogate mechanisms. It realizes efficient, accurate, and intelligent multi-objective optimization design of wick microstructures, thus providing solid technical support for the performance improvement and rapid iteration of ultrathin planar heat pipes.

[0062] Specifically, to verify the effectiveness of this method, optimization was performed according to steps S1-S5 above to obtain a set of optimal structural parameter combinations. Ten absorbent core samples were fabricated according to this optimal structural parameter combination and their thermal performance was tested. The actual test results (real experimental values) were compared with the prediction results (model prediction values) of the surrogate prediction model. Specifically: Regarding the heat transfer temperature difference ΔT: the measured temperature difference range for the 10 wicking samples was 26.0℃-49.0℃, and the predicted temperature difference range was 26.0℃-50.0℃. The absolute error range of the prediction results was 0.5℃-3.0℃, with an average absolute error of 1.65℃; the relative error range was 1.79%-7.35%, with an average relative error of 4.56%. Among them, the relative error of 7 samples was less than 5%, and the relative error of 3 samples (sample 2, sample 5, and sample 7) was between 6% and 8%.

[0063] Regarding thermal resistance R: the measured thermal resistance of the 10 wicking samples ranged from 12.0℃ / W to 28.0℃ / W, while the predicted thermal resistance ranged from 11.0℃ / W to 31.0℃ / W. The absolute error of the prediction results ranged from 0.5℃ to 3.0℃, with an average absolute error of 1.35℃ / W; the relative error ranged from 2.94% to 11.11%, with an average relative error of 7.02%. Among them, the relative error of 6 samples was less than 8%, and the relative error of 4 samples (sample 1, sample 6, sample 8, and sample 9) was between 8% and 12%.

[0064] In equivalent thermal conductivity In terms of performance: the measured equivalent thermal conductivity of the 10 wicking samples ranged from 820 W / (m·K) to 1630 W / (m·K), while the predicted equivalent thermal conductivity ranged from 840 W / (m·K) to 1650 W / (m·K). The absolute error of the prediction results ranged from 10 W / (m·K) to 45 W / (m·K), with an average absolute error of 29 W / (m·K); the relative error ranged from 0.62% to 3.70%, with an average relative error of 2.25%. The relative errors of all 10 samples were less than 4%, with 8 samples less than 3%.

[0065] Based on the above data, the coefficients of determination (R²) of the surrogate prediction model constructed by this method for heat transfer temperature difference, thermal resistance and equivalent thermal conductivity are 0.961, 0.947 and 0.973, respectively, all greater than 0.95; the root mean square error (RMSE) of the prediction of the three indicators are 1.82℃, 1.51℃ / W and 32.6 W / (m·K), respectively, all lower than the corresponding preset thresholds.

[0066] Compared to traditional empirical formulas, this method improves prediction accuracy by approximately 40%-60%. Using this method, the design cycle for wicking microstructures can be shortened from 4-6 weeks using traditional trial-and-error methods to 1-2 weeks, a reduction of approximately 60%-80%, thus significantly reducing R&D costs and accelerating product iteration. The wicking microstructure optimized using this method exhibits approximately 15%-25% lower thermal resistance and approximately 20%-35% higher equivalent thermal conductivity compared to traditional empirical designs.

[0067] Furthermore, under the same number of iterations (Tmax=300), the DE algorithm of this method was compared with the traditional genetic algorithm (GA) and particle swarm optimization (PSO). The results showed that the convergence accuracy of the DE algorithm was improved by about 28% and 32% compared with GA and PSO, respectively. Moreover, the variance of the convergence result of the DE algorithm after 10 independent runs was 42% smaller than that of GA and 35% smaller than that of PSO. Therefore, the DE algorithm of this method showed better stability and robustness.

[0068] It should be noted that the parts not described in detail or in elaboration in the above solutions are all prior art and do not constitute improvements made by this invention to existing technology, nor are they within the protection scope of this invention's technical solutions. Therefore, they will not be elaborated upon further in this document. Of course, the above content is merely a preferred embodiment of this invention and should not be considered as limiting the scope of the embodiments of this invention. This invention is also not limited to the above examples; equivalent changes and improvements made by those skilled in the art within the substantial scope of this invention should all fall within the patent coverage of this invention.

Claims

1. A method for intelligent optimization of the microstructure of a heat pipe wick, characterized in that, Includes the following steps: S1. Establish a structural parameter-thermal performance index mapping dataset: The structural parameter combination of the liquid absorption core microstructure is determined, liquid absorption core samples are processed according to multiple sets of the structural parameter combination and thermal performance is tested, and multiple sets of thermal performance index combinations corresponding to the multiple sets of the structural parameter combination are obtained to form a structural parameter-thermal performance index mapping dataset. The combination of structural parameters includes channel depth, channel width, micropillar position, and micropillar spacing as continuous variables, and micropillar shape as discrete variables; native encoding mapping is performed on the discrete variables, that is, different discrete variable options are encoded using consecutive integers; The combination of thermal performance indicators includes heat transfer temperature difference, thermal resistance, and equivalent thermal conductivity. S2. Constructing a proxy prediction model: Based on the structural parameter-thermal performance index mapping dataset, a surrogate prediction model is trained; in the surrogate prediction model, the combination of structural parameters is the input, and the combination of thermal performance indices is the output. S3. Establish a multi-objective optimization model: A multi-objective optimization model for the liquid absorption core microstructure is established, using the combination of structural parameters as optimization variables and the objective function formed by the combination of thermal performance indicators as the optimization objective. The optimization direction of the multi-objective optimization model is to minimize the heat transfer temperature difference, minimize the thermal resistance, and maximize the equivalent thermal conductivity. By taking the reciprocal of the equivalent thermal conductivity, the optimization direction of the multi-objective optimization model is unified to minimize the objective function. S4. Multi-target optimization based on dolphin echolocation algorithm: The dolphin echolocation algorithm is used to solve the multi-objective optimization model to obtain the Pareto optimal solution set; In step S4, when updating the individual location based on the dolphin echolocation algorithm, an echolocation mechanism is used; wherein, the update result of the discrete variable is subjected to a discrete mapping rule, that is, rounding and interval restriction operation are performed; the restricted interval is determined by the numerical value of the consecutive integers being encoded. In step S4, an encirclement predation strategy is adopted: the search range is expanded in the early stage of the dolphin echolocation algorithm iteration, and the search range is narrowed in the later stage of the dolphin echolocation algorithm iteration. S5. Determine the optimal solution: The optimal combination of structural parameters with the best overall performance is selected from the Pareto optimal solution set and used as the microstructure optimization scheme for the heat pipe wick.

2. The intelligent optimization method for the microstructure of a heat pipe wick according to claim 1, characterized in that: In S4, a random proxy mechanism is also adopted: a portion of individuals are randomly selected with a preset probability, and these individuals are allowed to wander randomly in the search space without performing individual position updates based on the echolocation mechanism.

3. The intelligent optimization method for the microstructure of a heat pipe wick according to claim 1, characterized in that: In step S4, during algorithm initialization: for the continuous variables, a uniform random initialization method is used; for the discrete variables, a random integer initialization method is used, that is, the variables are selected with equal probability from the continuous integers to be encoded.

4. The intelligent optimization method for the microstructure of a heat pipe wick according to claim 1, characterized in that: In S1, the specific values ​​of each structural parameter in the combination of structural parameters conform to the value range. The constraints on the range of values ​​include the processing capability of the microstructure processing equipment for the liquid absorption core and the geometric feasibility of the liquid absorption core.

5. The intelligent optimization method for the microstructure of a heat pipe wick according to claim 1, characterized in that: It also includes step S6, experimental verification of the optimal solution: The liquid-absorbing core sample was processed according to the optimal combination of structural parameters and its thermal performance was tested. The test results were compared with the prediction results of the surrogate prediction model based on the optimal combination of structural parameters.

6. The intelligent optimization method for the microstructure of a heat pipe wick according to claim 1, characterized in that, S2 includes the following sub-steps: S21. Data Preprocessing: The structural parameter-thermal performance index mapping dataset is normalized and divided into training and testing sets according to a certain ratio. S22, Network Training: The agent prediction model is trained using a numerical optimization algorithm; S23. Model Validation: The accuracy of the trained agent prediction model was validated using a test set.

Images