Processing device and method for optimizing heat dissipation in heat sinks

Abstract

A processing device for optimizing heat dissipation in heat sinks, which includes a storage and a processor. The processor runs a neural network model and executes following actions: generating multiple label temperatures on multiple virtual sample models based on simulation parameters; training the neural network model by using the simulation parameters, the multiple virtual sample models, all the label temperatures, and all labeled regions; and generating multiple predicted temperatures and a predicted region on a virtual test model based on component parameters and the virtual test model, where the multiple predicted temperatures indicate temperatures of multiple positions on a test circuit board, and the predicted region indicates an optimal heat dissipation region of a physical heat sink coated or arranged on the test circuit board.

Description

Technical Field

[0001] This disclosure relates to a heat conduction technology, and more particularly to a processing device and method for optimizing heat dissipation of a heat sink. Prior Technology

[0002] As the performance of electronic devices improves, the design and manufacturing technology of printed circuit boards (PCBs) has become increasingly complex. These PCBs typically contain a large number of electronic components, which generate significant heat during operation. If heat dissipation is not handled effectively, the electronic components may overheat, thereby affecting the stability, reliability, and lifespan of the overall system.

[0003] Heat sinks (such as thermal paste) are a common type of heat dissipation material, widely used for heat transfer between printed circuit boards (PCBs) and other heat sources (i.e., electronic components that may generate heat). However, users often find it difficult to directly know the temperature of different areas on the PCB, and also to determine which area of ​​the PCB to apply the thermal paste to and how much paste to apply for optimal heat dissipation. Therefore, how to quickly predict the temperature of different areas on the PCB and determine the area and amount of thermal paste to apply is a problem that those skilled in the art urgently need to solve. Summary of the Invention

[0004] The purpose of this disclosure is to provide a processing device and method for optimizing heat dissipation of a heat sink, solving the problem in previous technologies where it was difficult to determine which area of ​​the heat sink should be coated or placed, and how much coating / placement should be applied, resulting in the best heat dissipation effect.

[0005] To achieve the above objectives, this disclosure proposes a processing device for optimizing heat dissipation of a heat sink, comprising: A storage device is configured to store multiple instructions, multiple virtual sample models corresponding to multiple sample circuit boards, a simulation parameter of the multiple sample circuit boards, a virtual model under test corresponding to a circuit board under test, and a component parameter of the circuit board under test. Each virtual sample model has a label area with a virtual heat sink, and the label area indicates an optimal heat dissipation area of ​​a physical heat sink coated or disposed on the corresponding sample circuit board. A processor, connected to the storage, is configured to run a neural network model and access the plurality of instructions to perform the following actions: Action a) Generate multiple label temperatures on each virtual sample model based on the simulation parameters; Action b) Using the simulation parameters, the multiple virtual sample models, all the labeled temperatures, and all the labeled regions, train the neural network model of this type; and Action c) Based on the component parameters and the virtual model under test, generate multiple predicted temperatures and a prediction region on the virtual model under test using the trained neural network model, wherein the multiple predicted temperatures indicate the temperature of multiple locations on the circuit board under test, and the prediction region indicates the optimal heat dissipation area of ​​the physical heat sink that should be coated or set on the circuit board under test.

[0006] To achieve the above objectives, this disclosure proposes a method for optimizing heat dissipation of a heat sink, including: Step a) Using a processor, multiple label temperatures are generated on each of a plurality of virtual sample models based on a simulation parameter of a plurality of sample circuit boards, wherein the plurality of virtual sample models correspond to the plurality of sample circuit boards respectively, a label area on each virtual sample model has a virtual heat sink, and the label area indicates an optimal heat dissipation area of ​​one of a physical heat sink coated or disposed on the corresponding sample circuit board; Step b) Using the processor, a neural network model is trained using the simulation parameters, the multiple virtual sample models, all the labeled temperatures, and all the labeled regions; and Step c) Using the processor, based on a component parameter of a circuit board under test and a virtual model under test corresponding to the circuit board under test, multiple predicted temperatures and a predicted region on the virtual model under test are generated using the trained neural network model. The multiple predicted temperatures indicate the temperature of multiple locations on the circuit board under test, and the predicted region indicates the optimal heat dissipation area of ​​the physical heat sink that should be coated or set on the circuit board under test.

[0007] Compared to previous technologies, this disclosure first simulates the temperature of the circuit board and then uses a virtual model of the circuit board and simulation parameters to train a neural network model, thereby training a model that can predict the temperature at various locations on the circuit board and the optimal range for coating or setting up a physical heat sink. In this way, this disclosure can quickly predict the temperature of various areas on a printed circuit board and predict the best areas and the best coating / setting amount for coating or setting up a heat sink. Simple Explanation of the Diagram

[0008] Figure 1 is a block diagram illustrating a processing apparatus for optimizing heat dissipation of a heat sink in some embodiments.

[0009] Figure 2A illustrates a top view of a sample circuit board in some embodiments.

[0010] Figure 2B illustrates a slanted view of a virtual sample model in some embodiments.

[0011] Figure 2C illustrates a schematic diagram of a grid coordinate system in some embodiments.

[0012] Figure 3 illustrates a flowchart of a method for optimizing heat dissipation of a heat sink in some embodiments disclosed herein.

[0013] Figure 4A illustrates a schematic diagram of a first initial point cloud in some embodiments.

[0014] Figure 4B illustrates a schematic diagram of a first temperature point cloud map in some embodiments. Implementation

[0015] Referring to FIG1, FIG1 illustrates a block diagram of a processing apparatus 100 for optimizing heat dissipation of a heat sink in some embodiments disclosed herein.

[0016] As shown in Figure 1, in this embodiment, the processing device 100 for optimizing heat dissipation of the heat sink (hereinafter referred to as processing device 100) includes a storage 110 and a processor 120, and the processor 120 is connected to the storage 110.

[0017] In some embodiments, the processing device 100 may be implemented by any data processing device (e.g., a desktop computer, a laptop computer, or a tablet computer) or a server (e.g., a cloud server, a virtual server, or a rack server).

[0018] In this embodiment, the storage device 110 stores multiple computer-executable instructions, multiple virtual sample models corresponding to multiple sample circuit boards, simulation parameters of the multiple sample circuit boards, a virtual model of the circuit board under test corresponding to the circuit board under test, and component parameters of the circuit board under test. Specifically, the processing device 100 disclosed herein uses multiple sample circuit boards to train a neural network model, and uses the trained neural network model to predict relevant information of the circuit board under test.

[0019] In some embodiments, the storage device 110 may be implemented as flash memory, read-only memory, hard disk or any equivalent storage component.

[0020] In some embodiments, each computer-executable instruction can be implemented by any software or firmware. When the processor 120 reads the memory 110 and executes each computer-executable instruction, the heat dissipation optimization method for the heat sink disclosed herein can be implemented (described in detail later).

[0021] The sample circuit board is a physical circuit board used for training. In some embodiments, a physical heat sink is coated or disposed on the optimal heat dissipation area of ​​each sample circuit board. The optimal heat dissipation area is the area with the best heat dissipation and heat conduction effect obtained by the user after conducting a large number of experiments in advance (that is, as long as the physical heat sink is coated or disposed on the optimal heat dissipation area of ​​each sample circuit board, each sample circuit board can have the best heat dissipation and heat conduction effect).

[0022] In some embodiments, each sample circuit board can be implemented by any type of printed circuit board (e.g., rigid printed circuit board or flexible printed circuit board). In some embodiments, the physical heat sink can be implemented by any type of heat sink with heat dissipation and heat conduction capabilities (e.g., thermal adhesive (such as thermal conductive adhesive) or heat sink).

[0023] In this embodiment, each virtual sample model has a labeled area, and the labeled area contains a virtual heat sink. In this embodiment, the labeled area is used to indicate the optimal heat dissipation area of ​​the physical heat sink coated or installed on the corresponding sample circuit board. In other words, the labeled area is a virtualized area on each virtual sample model used to simulate the optimal heat dissipation area on the corresponding sample circuit board.

[0024] In some embodiments, each virtual sample model can be implemented by any type of three-dimensional solid model. In some embodiments, the virtual heat sink can be virtual information (e.g., virtual colloid information or virtual heat sink information) used in three-dimensional virtual space to simulate the heat dissipation and heat conduction state of a physical heat sink.

[0025] In some embodiments, each virtual sample model may have multiple virtual elements, which are virtual electronic elements in a three-dimensional virtual space used to simulate the operation of multiple physical elements on the corresponding sample circuit board.

[0026] In some embodiments, the multiple physical elements can be implemented by elements of any type of physical entity. For example, physical elements can be inductors, capacitors, bridges, housings, shields, or wooden support frames, etc.

[0027] In some embodiments, simulation parameters may include, but are not limited to, mesh size, virtual regions corresponding to each of the multiple physical components on each sample circuit board, the position of the label region on each virtual sample model, the power loss of each physical component, the material properties of each physical component, the temperature limit of each physical component, and the heat transfer equation of each physical component and the physical heat sink.

[0028] In some embodiments, the grid size may indicate the dimensions of each grid in a grid coordinate system in a three-dimensional virtual space (e.g., the length, width, and height of each grid correspond to 1 cm in real space). The grid coordinate system is a commonly used virtual coordinate system in the art and will not be further explained here.

[0029] In some embodiments, the virtual region system corresponding to each physical element on each sample circuit board is used to indicate the three-dimensional virtual region of the virtual element that can be used to simulate each physical element in the above-mentioned grid coordinate system.

[0030] In some embodiments, the processor 120 may pre-use any coordinate system algorithm to convert the three-dimensional real area covered by each entity element in the real coordinate system (e.g., the world coordinate system) into the three-dimensional virtual area covered by the corresponding virtual element in the grid coordinate system, so as to obtain the virtual area corresponding to each entity element.

[0031] In some embodiments, the position of the label area on each virtual sample model is used to indicate the three-dimensional virtual area of ​​the virtual heat sink in the above-mentioned grid coordinate system.

[0032] In some embodiments, the power loss of each physical element is used to indicate the power lost by each physical element in converting electrical energy into useless heat energy.

[0033] In some embodiments, the material properties of each physical element are used to indicate the thermal conductivity or thermal conductivity coefficient of each physical element.

[0034] In some embodiments, the upper temperature limit of each physical component is used to indicate the highest operating temperature that each physical component can withstand (i.e., when the upper temperature limit is exceeded, the physical component cannot operate normally).

[0035] In some embodiments, the heat transfer equations for each physical element are used to indicate how the temperature on each physical element changes over time.

[0036] In some embodiments, the heat transfer equations for a physical heat sink are used to indicate how the temperature on the physical heat sink changes over time.

[0037] In some embodiments, component parameters may include a virtual region corresponding to each of a plurality of other physical components on the circuit board under test, the power loss of all other physical components, the material properties (e.g., thermal conductivity) of each of the other physical components, the upper temperature limit of each of the other physical components, and the heat transfer equation of each of the other physical components. The plurality of other physical components on the circuit board under test may be the same as or different from the plurality of physical components on the sample circuit board.

[0038] It is worth noting that the definitions of virtual regions, power losses, material properties (such as thermal conductivity), upper temperature limits, and heat transfer equations of other physical components included in the component parameters are basically similar to the definitions of virtual regions, power losses, material properties, upper temperature limits, and heat transfer equations of the physical components mentioned above. The only difference is that the former is for the circuit board under test, while the latter is for the sample circuit board. Therefore, it will not be elaborated further here.

[0039] In some embodiments, the processor 120 can use the grid size as a reference to establish a corresponding virtual test model based on the virtual regions corresponding to each physical component on each sample circuit board and the position of the label regions on each virtual sample model, using computational fluid dynamics (CFD) model.

[0040] In some embodiments, the computational fluid dynamics model can be implemented by any type of simulation software with heat conduction simulation capabilities (e.g., ANSYS Fluent software or OpenFOAM software).

[0041] The following practical examples illustrate how to generate a virtual model under test. Referring to Figures 2A-2C, Figure 2A shows a top view of a sample circuit board 210 in some embodiments, Figure 2B shows a side view of a virtual sample model 220 in some embodiments, and Figure 2C shows a schematic diagram of a grid coordinate system {G} in some embodiments.

[0042] As shown in Figure 2A, the sample circuit board 210 in the real coordinate system {R} has a solid substrate b1, on which multiple solid components r1~r8 are disposed, and a solid heat sink rg is coated or disposed on the solid component r7. In addition, the solid substrate b1 itself can also be regarded as a solid component.

[0043] In other words, the sample circuit board 210 has a physical substrate b1, physical components r1~r8 and physical heat sink rg, each covering a real area in the real coordinate system {R}.

[0044] As shown in Figures 2B-2C, the processor 120 establishes a three-dimensional grid coordinate system {G} based on the grid size, and builds a corresponding virtual sample model 220 for the sample circuit board 210 on the grid coordinate system {G} according to the virtual regions corresponding to the physical substrate b1 and the physical components r1-r8 (i.e., the virtual regions covered by the virtual substrate b2 and the virtual components v1-v8 in the grid coordinate system {G}) and the position of the label region (i.e., the virtual region covered by the virtual heat sink vg in the grid coordinate system {G}).

[0045] In some embodiments, the storage 110 stores a virtual test model corresponding to the circuit board under test. This virtual test model can be generated in a three-dimensional virtual space by simulating the real circuit board under test using the method described above for simulating the sample circuit board. Since the simulation method has been described in detail above, it will not be repeated here.

[0046] In some embodiments, the circuit board under test (PCB) may be a different circuit board from the sample circuit board, and the PCB under test has not yet been coated or fitted with a physical heat sink. In some embodiments, the PCB under test may also be implemented by any type of printed circuit board. Specifically, the PCB under test referred to in this disclosure is a circuit board to which a physical heat sink is coated or fitted onto the area indicated by the inference result after inference is performed using a trained neural network model. This achieves better heat dissipation. In some embodiments, the physical heat sink is a thermal colloid, and the coating amount is the total amount of colloid coated onto the area indicated by the inference result. In some embodiments, the physical heat sink is a heat sink plate, and the fitting amount is the size of the heat sink plate fitted onto the area indicated by the inference result.

[0047] In this embodiment, the processor 120 runs a neural network model 121 and accesses the aforementioned multiple instructions to execute the processing method for heat dissipation optimization of the heat sink as described in the following paragraphs (details to follow).

[0048] In some embodiments, the neural network model 121 can be implemented by any neural network model for image processing (e.g., convolutional neural network model, deep neural network model, YOLO model, or transformer model, etc.).

[0049] In some embodiments, the processor 120 may be implemented by a central processing unit (CPU), a micro control unit (MCU), a programmable logic controller (PLC), a system on chip (SoC), or a field programmable gate array (FPGA).

[0050] Referring also to FIG3, FIG3 illustrates a flowchart of a processing method for optimizing heat dissipation of a heat sink in some embodiments disclosed herein, which is applicable to the processing apparatus 100 shown in FIG1.

[0051] As mentioned above, the storage 110 of the processing device 100 stores multiple virtual sample models and simulation parameters of multiple sample circuit boards, and also stores a virtual model of the circuit board under test and component parameters. As shown in Figure 3, firstly, in step S310, the processor 120 generates multiple label temperatures on each virtual sample model according to the simulation parameters.

[0052] In some embodiments, the processor 120 may use a computational fluid dynamics model to generate the temperatures of each virtual sample model in the multiple grids covered by the grid coordinate system {G}, based on simulation parameters, as label temperatures.

[0053] Specifically, the processor 120 can set the coordinates, power loss, material properties (e.g., thermal conductivity), temperature limit, and heat transfer equation for each virtual sample model within multiple grids covered by the grid coordinate system {G}, based on simulation parameters, to generate multiple simulation result files (e.g., VTK or CSV files) corresponding to each virtual sample model. In some embodiments, the multiple simulation result files respectively contain the power loss, material properties, temperature limit, and heat transfer equation for each point in the corresponding virtual sample model. Furthermore, after the user imports the coordinate system (e.g., the aforementioned grid coordinate system) into the simulation software, the multiple simulation result files can be converted into a first initial point cloud map (with a one-to-one correspondence between the grid and the point cloud). The difference from a pure point cloud distribution map is that the first initial point cloud map includes the simulation parameters, so each point cloud in the point cloud map carries corresponding information such as temperature, pressure, and velocity. Furthermore, the processor 120 then uses a computational fluid dynamics model to generate the temperature of each point cloud in each first initial point cloud map as its respective label temperature.

[0054] In some embodiments, the first initial point cloud is a simulation result file containing coordinate system information, and each of the multiple point clouds in the first initial point cloud has corresponding coordinates, power loss, material properties (e.g., thermal conductivity), upper temperature limit, and heat transfer equation.

[0055] The following description uses a practical example to illustrate the first initial point cloud map. Referring also to Figure 4A, Figure 4A shows a schematic diagram of the first initial point cloud map 410 disclosed in some embodiments.

[0056] As shown in Figure 4A, a virtual sample model has multiple point clouds on multiple grids covered by the grid coordinate system {G}. Each point cloud in the first initial point cloud figure 410 has a corresponding coordinate, power loss, material properties, upper temperature limit and heat transfer equation.

[0057] In some embodiments, the processor 120 may set different colors on each first initial point cloud map for different label temperature values ​​to generate a corresponding first temperature point cloud map. Each of the multiple point clouds in each first temperature point cloud map has a corresponding color, coordinates, power loss, material properties, upper temperature limit, and heat transfer equation.

[0058] The following description uses a practical example to illustrate the first temperature point cloud map. Referring also to Figure 4B, Figure 4B shows a schematic diagram of the first temperature point cloud map 420 disclosed in some embodiments.

[0059] As shown in Figure 4B, a virtual sample model has multiple point clouds on multiple grids covered by the grid coordinate system {G}. Each point cloud in the first temperature point cloud map 420 has a labeled temperature, corresponding color, coordinates, power loss, material properties, upper temperature limit, and heat transfer equation. In this embodiment, when the temperature value of the labeled temperature of the point cloud in the first temperature point cloud map 420 is higher, the color corresponding to the labeled temperature is more biased towards the first color (e.g., red). Conversely, when the temperature value of the labeled temperature of the point cloud in the first temperature point cloud map 420 is lower, the color corresponding to the labeled temperature is more biased towards the second color (e.g., blue).

[0060] In some embodiments, the processor 120 may control a display (not shown) in the processing device 100 or an external display (not shown) to display each of the first initial point cloud maps or the first temperature point cloud maps for the user to view.

[0061] Returning to Figure 3, in step S320, the processor 120 trains a neural network model 121 using simulation parameters, multiple virtual sample models, all label temperatures, and all label regions.

[0062] In some embodiments, the processor 120 may use multiple simulation result files generated by simulation parameters and each virtual sample model as training samples. More specifically, the processor 120 may use the aforementioned data (i.e., coordinates, power loss, material properties, upper temperature limit, and heat transfer equation) of multiple point clouds corresponding to the first initial point cloud map of each virtual sample model, generated according to the simulation parameters, as training samples. Then, the processor 120 may use the label regions and multiple label temperatures on each virtual sample model as multiple training labels corresponding to the multiple training samples. Next, the processor 120 may update the neural network model 121 using the multiple training samples corresponding to each simulation result file (or the first initial point cloud map) and the multiple training labels corresponding to each training sample.

[0063] In some embodiments, the processor 120 may input each training sample into the neural network model 121 to generate corresponding result regions and result temperatures of each point cloud as multiple result labels, and calculate the multiple result labels corresponding to each training sample and the loss values ​​between the multiple training labels (i.e., the loss value between the result temperature and label temperature of the point cloud at the same coordinates and the loss value between the coordinates of the result region and the coordinates of the label region in the grid coordinate system {G}).

[0064] Next, the processor 120 uses the calculated loss value to perform a backpropagation algorithm on the neural network model 121 to update the parameters in the neural network model 121 (i.e., the weight values ​​of each neural network layer in the neural network model 121). In this way, the processor 120 can repeat the same process to complete the update of the neural network model 121 (i.e., complete the training phase) more than a certain number of recursions (i.e., by using this pre-set number of training iterations).

[0065] In step S330, the processor 120 generates multiple predicted temperatures and predicted regions on the virtual model under test using a neural network-like model based on the component parameters and the virtual model under test.

[0066] In other words, whenever inference needs to be made about a new circuit board under test, the user generates a new virtual test model for the new circuit board under test using the simulation method described above. Then, the processor 120 can input the new virtual test model and the corresponding component parameters into the trained neural network model 121, so that the processing of the neural network model 121 can convert the virtual test model and the corresponding component parameters into multiple predicted temperatures and predicted regions on the virtual test model (i.e., the usage stage of inference using the trained neural network model 121).

[0067] In this embodiment, multiple predicted temperatures indicate the temperature at multiple locations on the circuit board under test, and the predicted area indicates the optimal heat dissipation area of ​​a physical heat sink coated or disposed on the circuit board under test.

[0068] In some embodiments, the processor 120 can set the coordinates, power loss, material properties, temperature upper limit, and heat transfer equations of each grid covered by the grid coordinate system {G} for the virtual model under test according to the component parameters, so as to generate a second result file (e.g., a VTK file or a CSV file) corresponding to the virtual model under test. After the coordinate system (e.g., the grid coordinate system mentioned above) is imported into the second result file, the second result file can be converted into a second initial point cloud map (similar to the first initial point cloud map 410 in FIG4A). Furthermore, the processor 120 inputs the second result file (or the second initial point cloud map) into the trained neural network model 121, so as to obtain the temperature of each point in the second result file (or each point cloud in the second initial point cloud map) and the predicted region on the second initial point cloud map from the output of the neural network model 121.

[0069] In some embodiments, the multiple point clouds of the second initial point cloud map each have corresponding coordinates, power loss, material properties, upper temperature limit, and heat transfer equation.

[0070] Next, the processor 120 can, according to the positional correspondence between the grid and the point cloud (i.e., find the grid and point cloud with the same coordinates in the grid coordinate system {G}), use the temperature of each point cloud in the second initial point cloud map and the predicted area on the second initial point cloud map as the predicted temperature of each of the multiple grids covered by the virtual model under test in the grid coordinate system {G} and the predicted area of ​​the virtual model under test.

[0071] In some embodiments, the processor 120 may set different colors on each second initial point cloud map for different predicted temperature values ​​to generate corresponding second temperature point cloud maps. Each of the multiple point clouds in each second temperature point cloud map also has a corresponding color, coordinates, power loss, material properties, upper temperature limit and heat transfer equation (similar to the first temperature point cloud map 420 in Figure 4B).

[0072] In some embodiments, the processor 120 may control a display (not shown) in the processing device 100 or an external display (not shown) to display a second initial point cloud map or a second temperature point cloud map for the user to view.

[0073] In some embodiments, the user can process the physical circuit board under test (e.g., coat or install a physical heat sink) based on the predicted temperature and predicted region output by the neural network to obtain the finished circuit board. Furthermore, the processor 120 can calculate the difference between the actual detected temperature and the predicted temperature at each location on the finished circuit board and determine whether the average of these differences is greater than a user-preset difference threshold. When the average of these differences is greater than the difference threshold, it indicates that the inference result of the neural network model 121 is inaccurate. In this case, the processor 120 can retrain the neural network model 121 using other sample circuit boards and corresponding simulation parameters in the same training manner as described above.

[0074] Through the above steps, this disclosure can automatically generate temperatures at various locations on the circuit board under test using a trained neural network model 121, and automatically generate the optimal range for coating or setting up a physical heat sink on the circuit board under test. Based on this range, the coating / setting amount of the physical heat sink can be determined. In this way, this disclosure eliminates the need for extensive and complex testing and inspection work, and can easily obtain the temperatures at various locations on the circuit board under test and the optimal range for coating or setting up a physical heat sink.

[0075] In summary, this disclosure combines pre-temperature simulation of the circuit board with a neural network-like algorithm to train a system capable of predicting the temperature at various locations on the circuit board and the optimal range for coating or placing physical heat sinks. This allows for the rapid inference of the temperature of different areas on the printed circuit board to be processed, as well as the area and amount of heat sink coating or placement. Furthermore, this disclosure employs a method of converting the virtual model with simulated temperatures into a visual point cloud map, enabling users to quickly determine the temperature at each location.

[0076] 100: A treatment device for optimizing heat dissipation of heat sinks. 110: Storage 120: Processor 121: Neural Network Model 210: Sample Circuit Board {R}: Real coordinate system r1~r8: Solid components b1: Solid substrate rg: solid heat sink 220: Virtual Sample Model v1~v8: Virtual Components b2: Virtual substrate vg: Virtual heat sink {G}: Grid coordinate system S310~S330: Steps 410: First Initial Point Cloud Map 420: First Temperature Point Cloud Map

Claims

1. A processing device for optimizing heat dissipation of a heat sink, comprising: A storage device configured to store multiple instructions, multiple virtual sample models corresponding to multiple sample circuit boards, a simulation parameter of the multiple sample circuit boards, a virtual model under test corresponding to a circuit board under test, and a component parameter of the circuit board under test, wherein a label area on each of the virtual sample models has a virtual heat sink, the label area indicating an optimal heat dissipation area of ​​one of the physical heat sinks coated or disposed on the corresponding sample circuit board; and a processor connected to the storage device, configured to run a neural network model, and access the multiple instructions to perform the following actions: Action a) generating multiple first initial point cloud maps corresponding to each of the virtual sample models based on the simulation parameters, and calculating the temperature of multiple point clouds on each of the first initial point cloud maps as multiple label temperatures, wherein each of the multiple point clouds in each of the first initial point cloud maps has at least one of a corresponding power loss, material properties, temperature upper limit, and heat transfer equation and coordinates; Action b) training the neural network model using the simulation parameters, the multiple virtual sample models, all the label temperatures, and all the label areas; Action c) Based on the component parameters and the virtual test model, generate a second initial point cloud map of the virtual test model using the trained neural network model, calculate the temperature of multiple point clouds on the second initial point cloud map as multiple predicted temperatures, and calculate a prediction region, wherein the multiple predicted temperatures indicate the temperature of multiple locations on the test circuit board, and the prediction region indicates the optimal heat dissipation area of ​​the physical heat sink that should be coated or set on the test circuit board, wherein each of the multiple point clouds in the second initial point cloud map has at least one of the corresponding power loss, material properties, temperature upper limit, and heat transfer equation and coordinates; and Action d) Calculate multiple differences between multiple actual detected temperatures at multiple locations on a processed test circuit board and the multiple predicted temperatures, and when the average of one of the multiple differences is greater than a difference threshold, retrain the neural network model using the simulation parameters corresponding to other sample circuit boards, wherein the prediction region of the processed test circuit board has been coated or set with the physical heat sink based on the predicted temperature.

2. The processing apparatus for heat dissipation optimization of a heat sink as described in claim 1, wherein the simulation parameters include the grid size in a grid coordinate system in a three-dimensional virtual space, the virtual region corresponding to each of the plurality of physical components on each of the sample circuit boards, the position of the label region on each of the virtual sample models, the power loss of all the physical components, the material properties of each of the physical components, the upper temperature limit of each of the physical components, and the heat transfer equations of each of the physical components and the physical heat sink; wherein the component parameters include the virtual region corresponding to each of the plurality of other physical components on the circuit board under test, the power loss of all other physical components, the material properties of each of the other physical components, the upper temperature limit of each of the other physical components, and the heat transfer equations of each of the other physical components.

3. The processing apparatus for heat dissipation optimization of a heat sink as described in claim 2, wherein the processor is further configured to perform the following actions: based on the grid size, and according to the position of the virtual region corresponding to each of the physical components on each of the sample circuit boards and the label region on each of the virtual sample models, a corresponding virtual model to be tested is established using a computational fluid dynamics model.

4. The processing apparatus for heat dissipation optimization of a heat sink as described in claim 1, wherein in action a), the processor is configured to perform the following action: generating, based on the simulation parameters, the temperatures of each of the virtual sample models within a plurality of grids in a grid coordinate system as the label temperatures of the point cloud of the first initial point cloud map using a computational fluid dynamics model.

5. The processing apparatus for heat dissipation optimization of a heat sink as described in claim 1, wherein in action b), the processor is configured to perform the following actions: generating a plurality of simulation result files corresponding to the plurality of virtual sample models respectively based on the simulation parameters; using the plurality of simulation result files as a plurality of training samples; using the label region on each of the virtual sample models and the plurality of label temperatures as a plurality of training labels corresponding to the plurality of training samples; and updating the neural network model using the plurality of training samples and the plurality of training labels corresponding to the plurality of training samples.

6. A method for optimizing heat dissipation of a heat sink, comprising: Step a) Using a processor, generate multiple first initial point cloud maps corresponding to each of the multiple virtual sample models based on a simulation parameter of multiple sample circuit boards, and calculate the temperature of multiple point clouds on each of the first initial point cloud maps as multiple label temperatures, wherein the multiple virtual sample models correspond to the multiple sample circuit boards respectively, a label area on each virtual sample model has a virtual heat sink, and the label area indicates the optimal heat dissipation area of ​​one of the physical heat sinks coated or set on the corresponding sample circuit board, and each of the multiple point clouds of the first initial point cloud map has at least one of the corresponding power loss, material properties, temperature upper limit and heat transfer equation and coordinates; Step b) Using the processor, train a neural network model using the simulation parameter, the multiple virtual sample models, all the label temperatures and all the label areas; Step c) Using the processor, based on a component parameter of a circuit board under test and a virtual test model corresponding to the circuit board under test, a second initial point cloud map on the virtual test model is generated using the trained neural network model. The temperatures of multiple point clouds on the second initial point cloud map are calculated as multiple predicted temperatures, and a prediction region is calculated. The multiple predicted temperatures indicate the temperatures of multiple locations on the circuit board under test, and the prediction region indicates the optimal heat dissipation area of ​​the physical heat sink that should be coated or set on the circuit board under test. Each of the multiple point clouds in the second initial point cloud map has at least one of the corresponding power loss, material properties, temperature upper limit, and heat transfer equation and coordinates. Step d) Multiple differences between multiple actual detected temperatures at multiple locations on a processed circuit board under test and the multiple predicted temperatures are calculated. When the average of one of the multiple differences is greater than a difference threshold, the neural network model is retrained using simulation parameters corresponding to other sample circuit boards. The prediction region of the processed circuit board under test has been coated or set with the physical heat sink based on the predicted temperature.

7. The heat dissipation optimization method for a heat sink as described in claim 6, wherein the simulation parameters include the mesh size in a network coordinate system in a three-dimensional virtual space, the virtual region corresponding to each of the plurality of physical components on each of the sample circuit boards, the position of the label region on each of the virtual sample models, the power loss of all the physical components, the material properties of each of the physical components, the upper temperature limit of each of the physical components, and the heat transfer equations of each of the physical components and the physical heat sink, wherein the component parameters include the virtual region corresponding to each of the plurality of other physical components on the circuit board under test, the power loss of all other physical components, the material properties of all other physical components, the upper temperature limit of each of the other physical components, and the heat transfer equations of each of the other physical components.

8. The method for optimizing heat dissipation of a heat sink as described in claim 7 further includes: Using the processor, based on the grid size, and according to the virtual regions corresponding to the physical components on each sample circuit board and the positions of the label regions on each virtual sample model, a computational fluid dynamics model is used to establish the corresponding virtual test model.

9. The method for optimizing heat dissipation of a heat sink as described in claim 6, wherein step a) comprises: Using the processor, based on the simulation parameters, a computational fluid dynamics model is used to generate the temperature of each of the multiple grids covered by the virtual sample model in a grid coordinate system as the label temperature of the point cloud of the first initial point cloud map.

10. The method for optimizing heat dissipation of a heat sink as described in claim 6, wherein step b) comprises: The processor generates multiple simulation result files corresponding to the multiple virtual sample models based on the simulation parameters. The processor uses the multiple simulation result files as multiple training samples; the processor uses the label regions and multiple label temperatures on each virtual sample model as multiple training labels corresponding to the multiple training samples; and the processor uses the multiple training samples and the multiple training labels corresponding to the multiple training samples to update the neural network model.

Images