A method for constructing a reduced-order model of liquid cooling shunt network for arrayed electronic equipment

Abstract

The present invention discloses a method for constructing a reduced-order model of a liquid-cooled shunt network for arrayed electronic devices, and relates to the field of thermal management technology. The present invention sorts out common meta-models in complex liquid-cooled shunt networks, and reduces the meta-models and customized models to one-dimensional black box models based on fluid mechanics empirical formulas, neural network algorithms, etc., thereby building a one-dimensional shunt network for the entire system, solving the data transmission and accurate solution problems of one-dimensional fluid networks, and constructing a heterogeneous integrated thermal management design sub-environment to achieve rapid simulation of complex liquid-cooled shunt networks, thereby improving design efficiency and shortening the design cycle. This method is particularly suitable for application scenarios where rapid iteration of thermal design solutions is required during the product solution design phase.

Description

Technical Field

[0001] The present invention relates to the field of thermal management technology, and in particular to a method for constructing a reduced-order model of a liquid-cooling shunt network for arrayed electronic equipment. Background Art

[0002] The statements in this section merely provide background information related to the present disclosure and may not constitute prior art.

[0003] As electronic devices continue to evolve towards greater integration, high power, and multi-functionality, heat generation continues to increase. Liquid cooling systems based on multi-stage shunt networks are gaining increasing popularity. Liquid cooling shunt networks for array products typically consist of complex, multi-stage liquid cooling structures with a wide range of dimensions. Thermal simulations for large and ultra-large arrays, in particular, can require tens of millions of grid cells, making rapid convergence difficult or even impossible. Therefore, improving the simulation efficiency of complex liquid cooling shunt networks is a critical factor impacting thermal design.

[0004] To meet these requirements, some commercial simulation software (such as Flomaster) has the ability to simulate one-dimensional fluid systems. By simplifying the changes in component inlet and outlet parameters, it can quickly analyze key indicators such as flow, pressure, and temperature, making it suitable for simulating complex systems such as pipeline networks. The liquid cooling distribution network of large-scale arrayed electronic devices has multiple levels and diverse flow channel configurations. The engineering design process requires not only rapid analysis of the system's liquid cooling distribution conditions but also analysis of various specific flow channel configurations. Current commercial simulation software struggles to address these requirements. Patent application number 202310434586.1 proposes a one-dimensional model of a liquid cooling plate and its construction method; patent application number 202411075871.X proposes a method for constructing a digital twin model of a multi-node cold plate parallel liquid cooling pipeline network. This method involves generating a liquid cooling cold plate model, generating a single-row distribution channel submodule for the pipeline network, and generating a supply / return main channel submodule. In summary, the methods used in existing patents do not systematically propose a set of methods and processes for reducing the order of the liquid-cooling shunt network of arrayed electronic devices. They are unable to implement hierarchical calculations of complex multi-level liquid-cooling shunt networks, fail to perform rapid modeling of common units in the liquid-cooling shunt network (hereinafter referred to as metamodels) based on neural network algorithms, fail to provide accurate theoretical solutions for data transmission and accurate solution of one-dimensional fluid networks, and fail to perform secondary packaging of arrayed flow channel units based on metamodels to effectively improve simulation efficiency. Based on this, a method for constructing a reduced-order model of the liquid-cooling shunt network of arrayed products is proposed. Summary of the Invention

[0005] The purpose of the present invention is to provide a method for constructing a reduced-order model of a liquid-cooled shunt network for an arrayed electronic device in response to the problems of low simulation efficiency, difficulty in rapid convergence, or even inability to calculate of complex liquid-cooled shunt networks. This method sorts out common meta-models in complex liquid-cooled shunt networks, and reduces the meta-models and customized models to one-dimensional black box models based on fluid mechanics empirical formulas, neural network algorithms, and the like, thereby building a one-dimensional shunt network for the entire system, solving the data transmission and accurate solution problems of one-dimensional fluid networks, and constructing a heterogeneous integrated thermal management design sub-environment to achieve rapid simulation of complex liquid-cooled shunt networks, thereby improving design efficiency and shortening the design cycle. This method is particularly suitable for application scenarios where rapid iteration of thermal design solutions is required during the product solution design phase.

[0006] The technical solutions of the present invention are as follows:

[0007] A method for constructing a reduced-order model of a liquid-cooling shunt network for arrayed electronic equipment comprises:

[0008] Step S1: sorting out the meta-model that constitutes the liquid cooling distribution network;

[0009] Step S2: Modeling and packaging the meta-model neural network;

[0010] Step S3: Based on the meta-model modeled and packaged in step S2, the customized model is modeled and packaged;

[0011] Step S4: construct a one-dimensional fluid network consisting of a meta-model and a customized model, and solve the parameters based on the solution algorithm; use the Newton iteration method to solve the equation system including the node mass conservation equation, the flow equation, the global flow equation, and the energy conservation equation;

[0012] Step S5: Encapsulate the meta-model constructed in step S2, the customized model constructed in step S3, and the solution algorithm used in step S4, and add a model interface file for docking with thermal simulation software and tools. After compilation and linking, a model dynamic link library file that can be directly called is generated, which can be called to drive the calculation process.

[0013] Furthermore, step S1 includes:

[0014] Combining the characteristics of the complex liquid cooling shunt network of arrayed electronic equipment, a series of common units that make up the liquid cooling shunt network are sorted out, which is the meta-model.

[0015] Furthermore, the meta-model includes:

[0016] Straight flow channel, gradient flow channel, L-shaped flow channel, T-shaped two-way flow channel, sudden expansion and contraction flow channel, multi-fin straight flow channel and microchannel radiator.

[0017] Furthermore, step S2 includes:

[0018] For simple meta-models, black box models can be established based on relevant empirical formulas of engineering fluid mechanics;

[0019] For complex meta-models, fully parametric modeling is performed one by one, and flow simulation is carried out based on commercial software to obtain flow simulation sample data for each meta-model; the sample data of each meta-model obtained are respectively brought into a neural network with a determined structure for training, and a neural network model representing the behavioral characteristics of each meta-model is obtained; the neural network model is customized and packaged, and then a black box model of each meta-model is constructed.

[0020] Furthermore, the input parameters of the meta-model modeled and packaged in step S2 include physical dimensions and environmental parameters, and the output parameters include pressure and flow.

[0021] Furthermore, the physical dimensions include: flow channel depth, number of fins, and flow channel width; and the environmental parameters include: flow rate and temperature.

[0022] Furthermore, step S3 includes:

[0023] The meta-models modeled and packaged in step S2 are organically combined to form various commonly used flow channel configurations, and each configuration is customized and packaged to construct a black box model of each customized model.

[0024] Furthermore, a one-dimensional fluid network consisting of a meta-model and a customized model is constructed, including:

[0025] The meta-model and customized model are organically arranged to form a one-dimensional fluid network, realizing the reduction of the three-dimensional model to the one-dimensional model.

[0026] Furthermore, in step S4, Newton's iteration method is used to perform an approximate solution.

[0027] Furthermore, the Newton iteration method is used to perform an approximate solution, including:

[0028] Step A: Assign initial values ​​to the pressure of each node and the flow of each component;

[0029] Step B: List the global metamodel equations according to the connection relationship between each metamodel and the known parameters of the nodes and metamodels;

[0030] Step C: Substitute the initial values ​​into the global metamodel equations and calculate the residuals of each equation and the partial derivatives with respect to flow and pressure;

[0031] Step D: Calculate the global residual equation to obtain the correction amount of each variable;

[0032] Step E: Modify the variables according to the obtained correction amount until the iteration reaches the convergence standard.

[0033] Compared with the existing technology, the beneficial effects of the present invention are:

[0034] The liquid cooling shunt network of large-scale arrayed electronic devices has hundreds or even thousands of shunt branches arranged in an array. If simulation is performed based on traditional commercial software, there are problems such as a huge number of grids, difficulty in convergence, or even inability to calculate. Based on this background, the present invention reduces the order of the complex three-dimensional liquid cooling shunt network to a one-dimensional model, realizes data transfer and accurate solution within the one-dimensional model, and then realizes rapid simulation and evaluation of the complex liquid cooling shunt network. For the liquid cooling shunt network of large-scale arrayed electronic devices, each arrayed shunt branch, complex radiator, etc. can be first reduced to a customized model based on the neural network algorithm, and then a full-system one-dimensional shunt network consisting of a series of meta-models and customized models is constructed to realize hierarchical simulation. The present invention can reduce the development difficulty of complex liquid cooling shunt networks, effectively improve design efficiency, and shorten the simulation time from the original hours to days to seconds to minutes. The present invention is suitable for rapid iteration in the scheme design stage, and has a wide range of application scenarios in fields such as liquid cooling and heat dissipation system design and shunt network fluid dynamics analysis. In particular, it can significantly improve simulation efficiency in the flow heat dissipation simulation of large-scale arrayed electronic devices. BRIEF DESCRIPTION OF THE DRAWINGS

[0035] Figure 1 A flow chart of a method for constructing a reduced-order model for a liquid-cooling shunt network of an arrayed electronic device;

[0036] Figure 2 Flowchart for implementation example simulation;

[0037] Figure 3 This is a schematic diagram of the liquid cooling distribution network for array products;

[0038] Figure 4 This is a schematic diagram of the four-channel RF array;

[0039] Figure 5 Schematic diagram of the three-dimensional structure of the fluid network;

[0040] Figure 6 It is a one-dimensional simulation model of the shunt network;

[0041] Figure 7 for Figure 6 Enlarged schematic diagram of point A in the middle.

[0042] Figure numerals: 1-liquid cooling source, 2-main diversion network, 3-secondary diversion network, 4-radiator, 5-diversion frame, 6-secondary diversion box, 7-microchannel radiator, 8-liquid cooling inlet, 9-liquid cooling outlet, 10-customized model, 11-L-type flow channel, 12-T-type two-way diversion, 13-mass flow inlet, 14-one-dimensional element model. DETAILED DESCRIPTION

[0043] It should be noted that relational terms such as "first" and "second" are used only to distinguish one entity or operation from another entity or operation, and do not necessarily require or imply any actual relationship or order between these entities or operations. Moreover, the terms "comprises," "comprising," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus comprising a series of elements includes not only those elements, but also other elements not explicitly listed, or elements inherent to such process, method, article, or apparatus. In the absence of further limitations, an element defined by the phrase "comprising a ..." does not exclude the presence of additional identical elements in the process, method, article, or apparatus comprising the element.

[0044] The features and performance of the present invention are further described in detail below with reference to the embodiments.

[0045] Example 1

[0046] See also Figure 1 A method for constructing a reduced-order model of a liquid-cooling shunt network for arrayed electronic equipment comprises:

[0047] Step S1: sorting out the meta-model that constitutes the liquid cooling distribution network;

[0048] Step S2: Modeling and packaging the meta-model neural network;

[0049] Step S3: Based on the meta-model modeled and packaged in step S2, the customized model is modeled and packaged;

[0050] Step S4: construct a one-dimensional fluid network consisting of a meta-model and a customized model, and solve the parameters based on the solution algorithm; use the Newton iteration method to solve the equation system including the node mass conservation equation, the flow equation, the global flow equation, and the energy conservation equation;

[0051] Step S5: Encapsulate the meta-model constructed in step S2, the customized model constructed in step S3, and the solution algorithm used in step S4, and add a model interface file for docking with thermal simulation software and tools. After compilation and linking, a model dynamic link library file that can be directly called is generated, which can be called to drive the calculation process.

[0052] In this embodiment, specifically, it should be noted that the complex liquid-cooling shunt network has multi-level characteristics, and is generally composed of a main shunt network-a secondary shunt network-a microchannel radiator, etc.; therefore, step S1 combines the characteristics of the complex liquid-cooling shunt network of the arrayed electronic equipment to sort out a series of common units (hereinafter referred to as meta-models) that constitute the liquid-cooling shunt network, such as a direct flow channel, a gradient flow channel, an L-shaped flow channel, a T-shaped two-way shunt, a sudden expansion and contraction flow channel, a multi-fin direct flow channel, a microchannel radiator, etc.

[0053] In this embodiment, specifically, step S2 includes:

[0054] For simple meta-models (such as straight channels and curved pipes), black box models can be established based on relevant empirical formulas of engineering fluid mechanics.

[0055] For each complex metamodel, fully parametric modeling is performed, and flow simulation is performed using commercial software to obtain flow simulation sample data for each metamodel. This sample data for each metamodel is then fed into a neural network with a defined structure for training, resulting in a neural network model that characterizes the behavior of each metamodel. The input parameters for each metamodel are physical dimensions (such as channel depth, number of fins, channel width), and environmental parameters (such as flow rate and temperature), while the output parameters are the corresponding pressure and flow rate. The models are then customized and packaged to construct a black box model for each metamodel.

[0056] In this embodiment, it should be noted that for the commonly used flow channel configurations in the complex liquid cooling network of arrayed electronic devices, secondary packaging can be performed based on the meta-model in step S2, that is, the meta-models are organically combined to form various commonly used flow channel configurations, and each configuration is customized and packaged to construct a black box model of each customized model; the development of customized models can further simplify the multi-level complex liquid cooling shunt network, thereby realizing hierarchical simulation of complex models, effectively improving simulation efficiency;

[0057] In this embodiment, it should be noted that the complex liquid-cooled shunt network can be formed by organically arranging a series of meta-models, thereby realizing the reduction of the three-dimensional model to the one-dimensional model. How to realize the data transmission and accurate solution of the one-dimensional fluid network is the key link of the present invention. The solution parameters of the fluid network include the pressure of each connection node, the flow rate of the meta-model, the flow velocity, etc. The solution of the shunt network mainly adopts the numerical solution method based on fluid mechanics theory, that is, the equation group is composed of the node mass conservation, flow equation, global flow equation, energy conservation equation, etc. The equation group consists of a series of multi-input and multi-output nonlinear equations, which are difficult to solve directly and need to be solved by numerical analysis. The present invention mainly adopts the Newton iteration method for approximate solution. In the solution process, first, the pressure of each node and the flow rate of each component are assigned initial values, and then the global equation group is listed according to the connection relationship of each meta-model and the known parameters of the node and meta-model. The initial values ​​are substituted into the global meta-model equation group to obtain the residuals of each equation and the partial derivatives of the flow rate and pressure. Next, the global residual equation is calculated to obtain the correction amount of each variable. The variables are corrected according to the obtained correction amount until the iteration reaches the convergence standard.

[0058] Therefore, in this embodiment, specifically, constructing a one-dimensional fluid network consisting of a meta-model and a customized model includes:

[0059] The meta-model and customized model are organically arranged to form a one-dimensional fluid network, realizing the reduction of the three-dimensional model to the one-dimensional model.

[0060] In this embodiment, specifically, step S4 uses Newton's iteration method to perform an approximate solution.

[0061] In this embodiment, specifically, the Newton iteration method performs an approximate solution, including:

[0062] Step A: Assign initial values ​​to the pressure of each node and the flow of each component;

[0063] Step B: List the global metamodel equations according to the connection relationship between each metamodel and the known parameters of the nodes and metamodels;

[0064] Step C: Substitute the initial values ​​into the global metamodel equations and calculate the residuals of each equation and the partial derivatives with respect to flow and pressure;

[0065] Step D: Calculate the global residual equation to obtain the correction amount of each variable;

[0066] Step E: Modify the variables according to the obtained correction amount until the iteration reaches the convergence standard.

[0067] In this embodiment, step S5 specifically includes developing multi-channel flow network fluid model software based on the aforementioned reduced-order model and thermal-fluid network solution research. To ensure compatibility with other professional simulation software and tools, the models in steps 2 and 3, as well as the solution algorithm in step 4, must be encapsulated. Furthermore, by adding model interface files that interface with professional thermal simulation software and tools, and through compilation and linking, a model dynamic link library file that can be directly called by the software and tools is generated, which can then be used to drive the calculation process.

[0068] Example 2

[0069] The first embodiment realizes the reduced-order modeling of a complex liquid-cooled shunt network by combing the meta-model, encapsulating and modeling the meta-model neural network, modeling and encapsulating the customized model, solving the liquid-cooled shunt network, and encapsulating the shunt network reduced-order model into software. The process is as follows: Figure 1 shown.

[0070] This embodiment is an implementation case of the method proposed in the first embodiment applied to the flow simulation of the shunt network of a four-channel RF array (more channels can be expanded on this basis). The design process is as follows Figure 2 shown.

[0071] The liquid cooling shunt network of the array product mainly consists of the main shunt network 2-secondary shunt network 3-radiator 4, such as Figure 3 As shown, liquid cooling flows from liquid cooling source 1 into the liquid cooling distribution network.

[0072] The four-channel RF array in this embodiment is mainly composed of a three-stage cold plate including a shunt frame 5, a secondary shunt box 6, and a microchannel radiator 7. Figure 4 shown.

[0073] Step S1: Decompose the four-channel array shunt network into a series of combinations of meta-models and customized models, such as Figure 5 As shown, it includes: a customized model 10, an L-shaped flow channel 11, and a T-shaped two-way splitter 12. Liquid cooling flows from the liquid cooling inlet 8 into the four-channel array splitter network and flows out from the liquid cooling outlet 9. In this embodiment, specifically, according to the characteristics of the model, it is specifically disassembled into the following components:

[0074] 2 inlet diverter elements, 4 gradient flow channels (square), narrow inlet and wide outlet elements, 16 corner flow channel (square) elements, 12 straight channel (square) elements, 4 square diverter elements, 8 cross-layer flow channel (vertical inlet, horizontal outlet) elements, 16 sudden expansion flow channel (circular) elements, 8 straight channel (circular) elements, 8 corner flow channel (square) elements, 4 straight channel radiator elements;

[0075] Step S2: Build a one-dimensional thermal fluid network for the RF array by dragging the metamodel and customizing the model icons and connecting them. On the basis of the one-dimensional simulation model of the RF array system, add mass flow inlet elements and pressure outlet elements, and set the inlet flow and outlet ambient pressure values, such as Figure 6 and Figure 7 As shown, it includes: a mass flow inlet 13 and a one-dimensional element model 14;

[0076] Step S3: After the one-dimensional heat flow network model of the RF array is built, the mass flow element and pressure outlet element parameters are input according to the verification conditions and model parameters.

[0077] Step S4: After all component input parameters are entered, set the global solution parameters, mainly including: calculation mode, residual standard, maximum number of iteration steps, and relaxation factor. After the above steps are completed, solve the calculation.

[0078] Step S5: Fluent, a commercial CFD software, was used to perform fluid simulation analysis on the 3D model. The results were compared with the 1D reduced-order network model. The comparison showed that the relative error for indicators such as the system's maximum inlet and outlet pressures and the flow rates at each node were all within 10%, meeting general engineering design requirements.

[0079] The above-described embodiments merely represent specific implementation methods of the present application. While the descriptions are relatively specific and detailed, they should not be construed as limiting the scope of protection of the present application. It should be noted that a person skilled in the art would be able to make numerous variations and improvements without departing from the technical concept of the present application, and all such variations and improvements fall within the scope of protection of the present application.

[0080] This background section is provided to generally present the context of the invention, and the work of the presently named inventors, the work to the extent described in this background section, and aspects of the description in this section that did not constitute prior art at the time of filing are neither explicitly nor implicitly admitted to be prior art to the present invention.

Claims

1. A method for constructing a reduced-order model of a liquid cooling shunt network for arrayed electronic equipment, characterized in that: include: Step S1: sorting out the meta-model that constitutes the liquid cooling distribution network; Step S2: Modeling and packaging the meta-model neural network; Step S3: Based on the meta-model modeled and packaged in step S2, the customized model is modeled and packaged; Step S4: constructing a one-dimensional fluid network consisting of a meta-model and a customized model, and solving parameters based on a solving algorithm; Step S5: Encapsulate the metamodel constructed in step S2, the customized model constructed in step S3, and the solution algorithm used in step S4, and add a model interface file for connecting with thermal simulation software and tools. After compilation and linking, a model dynamic link library file that can be directly called is generated, which can be called to drive the calculation process; Step S3 includes: The meta-models modeled and packaged in step S2 are organically combined to form various commonly used flow channel configurations, and each configuration is customized and packaged to construct a black box model of each customized model; Build one-dimensional fluid networks consisting of meta-models and custom models, including: The meta-model and customized model are organically arranged to form a one-dimensional fluid network, realizing the reduction of the three-dimensional model to the one-dimensional model.

2. The method for constructing a reduced-order model of a liquid cooling shunt network for an arrayed electronic device according to claim 1, characterized in that: Step S1 includes: Combining the characteristics of the complex liquid cooling shunt network of arrayed electronic equipment, a series of common units that make up the liquid cooling shunt network are sorted out, which is the meta-model.

3. The method for constructing a reduced-order model of a liquid cooling shunt network for an arrayed electronic device according to claim 2, characterized in that: Metamodel, including: Straight flow channel, gradient flow channel, L-shaped flow channel, T-shaped two-way flow channel, sudden expansion and contraction flow channel, multi-fin straight flow channel and microchannel radiator.

4. The method for constructing a reduced-order model of a liquid cooling shunt network for an arrayed electronic device according to claim 1, characterized in that: Step S2 includes: For simple meta-models, black box models can be established based on relevant empirical formulas of engineering fluid mechanics; For complex meta-models, fully parametric modeling is performed one by one, and flow simulation is carried out based on commercial software to obtain flow simulation sample data for each meta-model; the sample data of each meta-model obtained are respectively brought into a neural network with a determined structure for training, and a neural network model representing the behavioral characteristics of each meta-model is obtained; the neural network model is customized and packaged, and then a black box model of each meta-model is constructed.

5. The method for constructing a reduced-order model of a liquid cooling shunt network for an arrayed electronic device according to claim 4, characterized in that: The input parameters of the meta-model completed by modeling and packaging in step S2 include physical dimensions and environmental parameters, and the output parameters include pressure and flow.

6. The method for constructing a reduced-order model of a liquid cooling shunt network for arrayed electronic equipment according to claim 5, characterized in that: Physical dimensions include: channel depth, number of fins, and channel width; environmental parameters include: flow rate and temperature.

7. The method for constructing a reduced-order model of a liquid cooling shunt network for arrayed electronic equipment according to claim 1, characterized in that: In step S4, Newton's iteration method is used to obtain an approximate solution.

8. The method for constructing a reduced-order model of a liquid cooling shunt network for arrayed electronic equipment according to claim 1, characterized in that: Newton iteration method is used for approximate solution, including: Step A: Assign initial values ​​to the pressure of each node and the flow of each component; Step B: List the global metamodel equations according to the connection relationship between each metamodel and the known parameters of the nodes and metamodels; Step C: Substitute the initial values ​​into the global metamodel equations and calculate the residuals of each equation and the partial derivatives with respect to flow and pressure; Step D: Calculate the global residual equation to obtain the correction amount of each variable; Step E: Modify the variables according to the obtained correction amount until the iteration reaches the convergence standard.

Images