Physical information machine learning design method for multi-component composite high-thermal-conductivity filler

Abstract

A physical information machine learning design method of a multi-component composite high-thermal-conductivity filler belongs to the technical field of composite high-thermal-conductivity filler design, and comprises the following steps: S1, establishing a material intrinsic attribute data set; s2, establishing a three-dimensional space closest packing model; s3, setting boundary constraint conditions; s4, optimizing a machine learning algorithm; and S5, carrying out proportion conversion. According to the method, an existing machine learning technology is utilized, a comprehensive data set is constructed, a machine learning algorithm is applied, the porosity is reduced and heat conduction paths are increased through accurate material particle size matching, and different three-dimensional space closest packing theoretical models are correspondingly selected according to material intrinsic attributes; therefore, on the basis of optimizing the stacking structure of the optimal filler of the heat-conducting polymer composite material, forming an efficient heat-conducting path and keeping the processing performance of the heat-conducting polymer composite material, the design and manufacturing processes are simplified, the production efficiency is improved, and powerful technical support is provided for effective heat dissipation and reliable and stable work of electronic equipment.

Description

Technical Field

[0001] The present invention relates to the technical field of composite high thermal conductivity filler design, and in particular to a physical information machine learning design method for a multi-component composite high thermal conductivity filler. Background Art

[0002] With the rapid development of electronic technology, the heat dissipation problem of electronic devices has become increasingly prominent. Heat dissipation is directly related to the working performance and service life of electronic devices. In existing heat dissipation solutions for electronic devices, direct contact between microelectronic devices and metal heat sinks is usually used to achieve heat conduction. However, due to the unevenness of the contact surface, the gaps between the two have become the main factor limiting the heat dissipation efficiency. These gaps are usually filled with air with extremely low thermal conductivity, which seriously hinders the effective transfer of heat. In order to overcome this limitation, the existing technology turns to the use of thermally conductive polymer composites as thermal interface materials to improve the efficiency of thermal conduction. However, since the thermal conductivity of pure polymers is generally low, it is necessary to enhance their thermal conductivity by adding fillers with high thermal conductivity.

[0003] The following two strategies have been adopted in the existing research on composite materials with improved thermal conductivity. One method is to use functionalized fillers to improve their dispersibility and compatibility in the polymer matrix, thereby reducing the interfacial thermal resistance between the filler and the matrix. Another method is to increase the thermal conductivity path and further improve the thermal conductivity of the composite material by means of directional arrangement of fillers, construction of separated filler networks, and multi-scale filler composites. The common goal of the above is to form an effective thermal conductive network in the polymer matrix to promote the transfer of phonons (phonons are collective excitations of atomic (or molecular) vibrations in condensed matter, a quasi-particle used to describe the laws of lattice vibrations. Phonon heat transfer refers to the process in which phonons transfer heat in materials as heat carriers) and improve thermal conductivity efficiency. Although existing composite material performance improvement methods have improved thermal conductivity to a certain extent, there is still insufficient research on how to optimize the thermal conductivity path from the perspective of filler stacking. In practical applications, high filling amount is a common method to achieve high thermal conductivity, but this approach often sacrifices the processing performance of composite materials. Therefore, developing a multi-scale filling system and reducing porosity and increasing thermal conductivity paths through precise particle size matching have become the key to balancing thermal conductivity and processing performance. However, there is currently a lack of theoretical guidance for particle size matching of thermal conductive fillers, and it is usually based on parallel control experiments to optimize particle size matching. This method is inefficient and lacks universality. Although the traditional stacking model is feasible in theory, in practical applications, due to the wide variety of alternative materials, finding the best formula often requires exhaustive calculation of all possible combinations, which is not only computationally intensive but also extremely inefficient, thus greatly limiting the application of this advanced theory in practice. Summary of the invention

[0004] Due to technical limitations and the disadvantages of the background technology, the present invention provides a physical information machine learning design method for a multi-component high thermal conductivity filler. Under the joint action of relevant steps, by constructing a comprehensive data set and applying a machine learning algorithm, the stacking structure of the best filler in the thermally conductive polymer composite material is optimized, an efficient thermal conduction path is formed, and the processing performance of the thermally conductive polymer composite material is maintained. At the same time, the design and manufacturing process are simplified, the production efficiency is improved, and a strong technical support is provided for the effective heat dissipation, reliable and stable operation of electronic equipment.

[0005] The technical solution adopted by the present invention to solve its technical problem is: A physical information machine learning design method for a multi-component composite high thermal conductivity filler comprises the following steps: S1: establishing a data set of intrinsic properties of materials, testing the cumulative gradation curve distribution of each material used in the multi-component composite high thermal conductivity filler through a laser particle size analyzer on a PC, and uniformly defining the particle size detection points of the gradation curve. In order to avoid the contingency of a single test result, the cumulative gradation curve distribution of each material should be tested multiple times, and the average value is taken as the final cumulative gradation curve distribution, and the average value is used as the data input, and the data is input into step S4; S2: establishing a three-dimensional space closest packing model, collecting and arranging the modeling formulas of the three-dimensional space closest packing theoretical model on a PC, and compiling the calculation rule function according to the filler particle size distribution range applicable to different modes, which is used for the model call of the subsequent step S4; S3: setting the boundary constraint conditions defined by the physical rules. For a multi-component composite system, the boundary constraint conditions need to be set so that the sum of the total volumes of all systems is equal to 100%, that is, ,in Represents the volume content of different materials, represent different materials; S4: machine learning algorithm optimization, which accelerates the calculation process through machine learning algorithm optimization, and compares it with the optimal formula of the ideal grading curve obtained by the three-dimensional space most dense packing model, and calculates the error between the calculation results and the optimal formula under different material combinations; S5: ratio conversion, because the calculated optimal formula is the volume mixing ratio, it is not convenient to operate in actual operation, so it is necessary to convert the volume mixing ratio of various materials into the mass mixing ratio.

[0006] Furthermore, in step S1, the multi-component high thermal conductivity filler in the material intrinsic property data set includes metal-based thermal conductive fillers and ceramic-based thermal conductive fillers, wherein the metal-based thermal conductive fillers are one or more combinations of silver, copper, aluminum, nickel, and tungsten, and the ceramic-based thermal conductive fillers are one or more combinations of aluminum nitride, silicon carbide, aluminum oxide, magnesium oxide, zinc oxide, boron nitride, etc.

[0007] Furthermore, in S2, the three-dimensional space most dense packing theoretical model is one or more of the continuous grading theory, the discontinuous grading theory and the compressible packing theory. In specific applications, the corresponding appropriate theoretical model calculation formula is selected according to the selection of different materials in practice.

[0008] Furthermore, S4 specifically includes the following processes: (1) defining a hyperparameter grid space; (2) cross-validating hyperparameter combinations; (3) selecting the best hyperparameter combination; (4) training and providing a final model; (5) establishing a machine learning model based on (1) to (4), improving the generalization ability of the model, and finding the hyperparameter configuration that best suits the current task.

[0009] Compared with the prior art, the present invention has the following beneficial effects: the present invention utilizes the existing machine learning technology, and under the joint action of relevant steps, constructs a comprehensive data set, and applies a machine learning algorithm to reduce the porosity and increase the thermal conduction path through precise material particle size matching, and selects different three-dimensional space closest packing theoretical models according to the intrinsic properties of the material, thereby optimizing the stacking structure of the best filler of the thermally conductive polymer composite material, forming an efficient thermal conduction path, maintaining the processing performance of the thermally conductive polymer composite material, and at the same time simplifying the design and manufacturing process, thereby improving production efficiency, and providing strong technical support for the effective heat dissipation and reliable and stable operation of electronic equipment. BRIEF DESCRIPTION OF THE DRAWINGS

[0010] Figure 1 This is a flowchart of a physical information machine learning design method for a multi-component composite high thermal conductivity filler involved in the present invention.

[0011] Figure 2 It is the cumulative particle size distribution gradation curve and statistical results of the intrinsic properties of the material in the present invention.

[0012] Figure 3 The calculation results and time consumed for compounding the ternary powders of the present invention are shown.

[0013] Figure 4 It is part of the calculation results in the present invention, including the combination of different materials, the goodness of fit between the actual cumulative grading curve and the ideal cumulative grading curve, the volumetric dosage of each material and the distribution of the actual cumulative grading curve.

[0014] Figure 5 It is a comparison between the actual cumulative gradation curve of the present invention and the ideal cumulative gradation curve. DETAILED DESCRIPTION

[0015] Figure 1As shown in the figure, a physical information machine learning design method for multi-component composite high thermal conductivity filler includes the following processes: Step 1: Establishing a data set of intrinsic properties of materials. For 10 powder raw materials with different average sizes of multi-component composite high thermal conductivity filler, the cumulative gradation curve distribution of each material used in the multi-component composite high thermal conductivity filler is tested by a laser particle size analyzer on the PC side, and the material particle size detection points of the gradation curve are uniformly defined. To avoid test errors, each material is tested 3 times and the average value is taken as the final cumulative particle size distribution value, and the particle size detection points of the gradation curve are unified, and the final results are summarized ( Figure 2 As shown), the material intrinsic property data set is established (this step is used to collect material intrinsic property information). Step 2: Establish a three-dimensional space closest packing theoretical model, collect and organize the modeling formulas of the three-dimensional space closest packing theoretical model on the PC side, and write the calculation rule function according to the material particle size distribution range applicable to different modes, which is used for the model call of the subsequent step S4. Specifically, the present invention adopts the Dinger-Funk model, and its calculation formula is: , where d is the uniform particle size detection point, D max is the maximum particle size in the system, D min is the minimum particle size in the system, n is the distribution modulus, Dis is the best distribution curve in the theoretical state under the system; the three-dimensional space most dense packing theoretical model is one or more of the continuous grading theory, the discontinuous grading theory and the compressible stacking theory. In the application, according to the selection of different materials in practice, the corresponding appropriate theoretical model calculation formula is selected (for the system with a small difference in particle size distribution, the continuous grading theory should be adopted; for the system with a large difference in particle size distribution, the discontinuous grading theory should be adopted; for the system that needs to be extruded / compacted, the compressible stacking theory should be adopted).

[0016] Step 3: Set boundary constraints. For the multi-component compound system, the present invention needs to set boundary constraints so that the sum of the total volume of all systems is equal to 100%. Taking the ternary compound as an example, the system should satisfy , X represents the type of material, 1, 2, and 3 represent the first, second, and third materials, respectively. In addition, all possible combinations of any three of the 10 materials should be established, and the number of combinations generated should be The minimum volume proportion of each material should be 0.01% and the maximum , only material formula combinations that meet the above rules are allowed to proceed to the next step (the result of the calculation without restricting the constraints is meaningless. For example, the total volume fraction of the system is only 100%. If there is no restriction, the sum of the volume fractions of the three materials may exceed 100%, which is not in line with the facts and the calculation is meaningless). Step 4: Use machine learning algorithms to optimize calculations. The present invention uses machine learning algorithms to optimize the calculation speed. While ensuring the accuracy of the calculation, it greatly speeds up the running speed of the program, and compares it with the optimal formula of the ideal grading curve obtained by the most densely packed model in three-dimensional space. The error between the calculation results and the optimal formula under different combination conditions is calculated. When using the traditional exhaustive method for calculation, 3 materials are randomly selected from 10 materials for compounding. It takes more than 20 hours to complete the calculation of all materials that meet the constraints. Using machine learning to first build a model and then use the machine learning model to perform predictive calculations only takes 0.6 seconds ( Figure 3 ); and the results are compared and found that the calculation results of the traditional exhaustive method are completely consistent with the optimization results of the machine learning algorithm. The calculation results give the combination, error, volume proportion and actual cumulative gradation curve distribution of any three different materials ( Figure 4 ); Comparing the ideal cumulative distribution gradation curve with the actual cumulative distribution gradation curve, it can be found that the actual cumulative distribution gradation curve of the calculated formula is highly similar to the ideal cumulative distribution gradation curve ( Figure 5 ); specifically includes the following processes: (1) define the hyperparameter grid space; (2) cross-validate the hyperparameter combination; (3) select the best hyperparameter combination; (4) train and give the final model; (5) establish a machine learning model based on (1) to (4), effectively improve the generalization ability of the model, and find the hyperparameter configuration that best suits the current task (traditional calculations require exhaustive calculations, which takes a lot of time. By adjusting the hyperparameters of the machine learning model in this way, the amount of calculation and calculation time can be greatly reduced, and the accuracy is the same or close to that of the exhaustive method). Step 5: Ratio conversion. Since the calculated optimal formula is the volumetric dosage ratio, it is not easy to operate in actual operation. Therefore, it is necessary to convert the volumetric dosage ratio of various materials into the mass dosage ratio. The ratio conversion formula is , , , In the formula, M1, M2, and M3 represent the mass fractions of the first, second, and third materials respectively; V1, V2, and V3 represent the volume fractions of the first, second, and third materials respectively; ρ1, ρ2, and ρ3 represent the densities of the first, second, and third materials respectively. The purpose of this step is to convert the volume fraction ratio into the mass fraction ratio, which is convenient for the experimenter to sample each material in the experiment.

[0017] Figure 1 As shown in , the present invention has the following advantages: (1) Based on the requirement of thermal conductive filler particle size matching, the present invention reduces porosity and increases thermal conductivity path by accurately matching material particle sizes, selects different three-dimensional space closest packing theoretical models according to the intrinsic properties of the material, and provides a strong theoretical basis for the design of multi-element powder compounding: (2) The present invention allows the boundary constraint conditions to be customized, because in real situations, the filling rate of powder in three-dimensional space cannot reach 100%, and there must be certain gaps. The present invention can use boundary constraints to provide a guarantee for subsequent more reliable models and improved prediction quality. (3) In the present invention, after establishing the material intrinsic property data set, the three-dimensional space closest packing model and boundary constraints, the hyperparameter space is established and cross-validated through the machine learning optimization algorithm, and the optimal prediction model is obtained according to the optimal hyperparameter combination, which accelerates the calculation speed very quickly and reduces the lowest prediction cost. (4) The present invention converts the volumetric doping ratio of different materials into the mass ratio of different materials by converting the volumetric doping ratio of the calculation results, which can make it more convenient for experimental personnel to verify the results and verify the model results. In addition, the present invention can also allow unlimited expansion of the size of the material intrinsic property data set, and different three-dimensional space closest packing theoretical models can be selected when selecting different materials; when the data set operation complexity is greater, the effect of the method of the present invention is more significant, in addition to the optimization strategy in the invention, machine learning algorithms such as Bayesian optimization can also be flexibly used to achieve the same purpose. The present invention not only provides a strong theoretical support for multi-powder compounding, but also can greatly save computing time and computing costs.

[0018] The above shows and describes the basic principles and main features of the present invention and the advantages of the present invention. It is obvious to those skilled in the art that the present invention is limited to the details of the above exemplary embodiments, and that the present invention can be implemented in other specific forms without departing from the spirit or basic features of the present invention. Therefore, no matter from which point of view, the embodiments should be regarded as exemplary and non-restrictive. The scope of the present invention is defined by the appended claims rather than the above description, and it is intended that all changes that fall within the meaning and scope of the equivalent elements of the claims are included in the present invention.

[0019] In addition, it should be understood that although the present specification is described according to the implementation mode, the implementation mode does not only include an independent technical solution. This narrative method of the specification is only for the sake of clarity. Those skilled in the art should regard the specification as a whole. The technical solutions in the embodiments can also be appropriately combined to form other implementation modes that can be understood by those skilled in the art.

Claims

1. A physical information machine learning design method for multi-component composite high thermal conductivity filler, characterized in that: The following steps are involved: S1: Establishment of material intrinsic property data set. Test the cumulative gradation curve distribution of each material used in the multi-component composite high thermal conductivity filler by laser particle size analyzer on PC, and uniformly define the particle size detection points of the gradation curve. In order to avoid the contingency of a single test result, the cumulative gradation curve distribution of each material should be tested multiple times, and the average value is taken as the final cumulative gradation curve distribution, which is used as data input, and the data is input to step S4; S2: Establishment of the closest packing model in three-dimensional space. Collect and organize the modeling formulas of the closest packing theoretical model in three-dimensional space on the PC side, and compile the calculation rule function according to the filler particle size distribution range applicable to different modes, which is used for the model call in the subsequent step S4; S3: Setting of boundary constraints. For multi-component complex systems, boundary constraints need to be set so that the sum of the total volume of all systems is equal to 100%, that is, ,in Represents the volume content of different materials, represent Different materials; S4: Machine learning algorithm optimization, which accelerates the calculation process through machine learning algorithm optimization, and compares it with the optimal formula of the ideal gradation curve obtained by the three-dimensional space most dense packing model to find the error between the calculation results and the optimal formula under different material combination conditions; S5: Ratio conversion. Since the calculated optimal formula is the volumetric dosage ratio, it is not convenient to operate in actual operation. Therefore, it is necessary to convert the volumetric dosage ratio of various materials into the mass dosage ratio.

2. The physical information machine learning design method of a multi-component composite high thermal conductive filler according to claim 1 is characterized in that: In step S1, the multi-component high thermal conductivity filler in the material intrinsic property data set includes metal-based thermal conductive fillers and ceramic-based thermal conductive fillers, wherein the metal-based thermal conductive fillers are one or more combinations of silver, copper, aluminum, nickel, and tungsten, and the ceramic-based thermal conductive fillers are one or more combinations of aluminum nitride, silicon carbide, aluminum oxide, magnesium oxide, zinc oxide, boron nitride, etc.

3. The physical information machine learning design method of a multi-component composite high thermal conductive filler according to claim 1 is characterized in that: In S2, the three-dimensional space most dense packing theoretical model is one or more of the continuous grading theory, the discontinuous grading theory and the compressible packing theory. In specific applications, the corresponding appropriate theoretical model calculation formula is selected according to the selection of different materials in practice.

4. The physical information machine learning design method of a multi-component composite high thermal conductive filler according to claim 1 is characterized in that: S4 specifically includes the following processes: (1) defining the hyperparameter grid space; (2) cross-validating the hyperparameter combinations; (3) selecting the best hyperparameter combination; (4) training and giving the final model; (5) building a machine learning model based on (1) to (4), improving the generalization ability of the model, and finding the hyperparameter configuration that best suits the current task.

Images