Multi-node cold plate parallel liquid cooling pipe network digital twinborn model construction method

Through the construction of digital twin models, combined with physical mechanism and machine learning, the design iteration time and simulation accuracy of multi-node cold plate parallel liquid-cooled pipeline system is reduced, real-time monitoring of internal data of liquid-cooled pipeline network is realized, and fault alarm is solved. The problems of low simulation efficiency and unvisible simulation in the prior art.

CN119004718AActive Publication Date: 2024-11-22NANJING RES INST OF ELECTRONICS TECH

Patent Information

Application Number
CN202411075871.X
Authority / Receiving Office
CN · China
Patent Type
Applications(China)
Current Assignee / Owner
Filing Date
2024-08-07
Publication Date
2024-11-22
Estimated Expiration
2044-08-07

AI Technical Summary

Technical Problem

During the design process, the existing multi-node parallel liquid-cooled pipeline system has low simulation efficiency, long design iteration time, and the internal flow characteristics and flow resistance characteristics cannot be intuitively improved, which affects the evaluation of heat dissipation performance.

Method used

The digital twin model construction method of multi-node cold plate parallel liquid-cooled pipe network is adopted. By establishing a hybrid modeling system based on physical mechanism and machine learning, the traditional three-dimensional CFD simulation model is reduced, and the high-fidelity reduction model (ROM) is generated. Through real-time mapping and correction models, multi-case analysis and visual simulation are realized.

Benefits of technology

It greatly shortens the design iteration time and cycle, improves simulation accuracy, realizes real-time monitoring of key data inside the liquid-cooled pipeline network and fault alarms, and reduces electromagnetic interference and false alarms caused by sensor monitoring.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN119004718A_ABST
    Figure CN119004718A_ABST
Patent Text Reader

Abstract

The invention belongs to the technical field of digital twinning, and discloses a method for constructing a digital twinning model of a multi-node cold plate parallel liquid cooling pipe network. The digital twinborn model construction method relates to generation of a liquid-cooling cold plate model, generation of a pipe network single-row sub-runner sub-module and generation of a liquid supply / return main runner sub-module. According to the method, a system model based on hybrid modeling of a physical mechanism and a machine learning method is established, the simulation time is greatly shortened, and the original hour-day simulation time is shortened to the second-minute simulation time; key internal flow, temperature and pressure data are provided for the multi-node cold plate parallel liquid cooling pipe network, and real-time monitoring of a large amount of internal unmeasurable data in operation and maintenance of the liquid cooling pipe network is achieved; according to the invention, necessary pipe network internal data is provided for the operation fault alarm function of the multi-node cold plate parallel liquid cooling pipe network digital twin system, and electromagnetic interference, false alarm and false alarm caused by arrangement of sensors for monitoring are greatly reduced.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] The present invention mainly relates to the field of digital twin technology, and in particular to a method for constructing a digital twin model of a multi-node cold plate parallel liquid cooling pipeline network. Background Art

[0002] Liquid cooling of electronic equipment refers to the process of continuously removing the heat generated by electronic components through the convection heat exchange process between the liquid and the surface of the electronic components in contact with it or the solid surface of the cold plate under the action of circulating liquid, so as to ensure that the temperature of the electronic components is within the normal working range. With its advantages of efficient heat exchange, low noise and compactness, it has been widely used in cooling various electronic equipment, such as data centers, electric vehicles, mobile terminal equipment, IGBT, laser weapons, phased array radars, etc.

[0003] Liquid-cooled cold plates are widely used in the heat dissipation of electronic equipment. The cooling medium with a specific flow rate, temperature and pressure is provided outside the cold plate. The heat generated by the electronic equipment is transferred to the cooling medium through convection heat exchange and heat conduction inside the cold plate, thereby ensuring that the electronic equipment or device works in a suitable temperature range for a long time to ensure the long-term reliability and life of the device. Multi-node parallel liquid cooling pipe network system is widely used in liquid-cooled data centers, liquid-cooled phased array radars and other equipment. Its design includes fluid characteristics analysis, pipe network flow resistance analysis and flow distribution, heat conduction analysis and fluid-solid coupling analysis. The design is mostly carried out through fluid or thermal design software simulation. In the design process, the result response under multiple working conditions is repeatedly iterated to evaluate the heat dissipation performance and finally determine the design parameters. By carrying out thermoelectric joint simulation and software trial and error iteration, the design risk is released, but it is bound to put forward high requirements on simulation accuracy, which leads to a decrease in simulation efficiency. Summary of the invention

[0004] The purpose of the present invention is to provide a method for constructing a digital twin model of a multi-node cold plate parallel liquid cooling pipe network. The digital twin technology has the characteristics of high fidelity, multi-physics, multi-discipline, and multi-scale. By establishing this multi-node cold plate parallel liquid cooling pipe network digital twin model, the traditional three-dimensional CFD simulation model can be reduced to a high-fidelity reduced-order model (ROM). By conducting multi-condition analysis on the digital twin ROM model to evaluate the heat dissipation performance of the cold plate, the design iteration time can be greatly shortened. By establishing a real-time mapping between the physical pipe network and the reduced-order ROM model, the digital twin model is corrected to improve the accuracy and greatly shorten the design cycle. In addition, in the thermal test link, the internal flow characteristics of the parallel liquid-cooled cold plate, the flow resistance characteristics of the pipe network system, and the flow distribution that cannot be directly seen can also be visualized through its twin model simulation.

[0005] To achieve the above object, the present invention provides a method for constructing a digital twin model of a multi-node cold plate parallel liquid cooling pipe network, comprising the following steps:

[0006] Step 1: Establish a simulation model based on a detailed mechanism for the internal flow channel of the liquid-cooled cold plate. According to the working range of the liquid-cooled cold plate, run the simulation model based on the detailed mechanism multiple times to obtain an input-output data set of the liquid-cooled cold plate.

[0007] The input-output data set is used as training data and a machine learning algorithm is used to generate a liquid-cooled cold plate model ROM1;

[0008] Step 2: Building a single-row liquid supply model including a liquid supply manifold and a liquid cooling cold plate around the liquid cooling cold plate model;

[0009] The liquid supply channel supplies liquid to a row of j basic units, each basic unit includes k liquid-cooled cold plates, and the liquid-cooled cold plate model ROM1 generated by step 1 is nested in the single-row liquid supply model to establish an interface model for the liquid supply channel to flow into each liquid-cooled cold plate;

[0010] Step 3: Establish a single-row liquid return model including a liquid return branch channel and a liquid return interface;

[0011] Step 4: Assemble the liquid-cooled cold plate model, the single-row liquid supply model, and the single-row liquid return model to form a decoupled single-row flow channel model;

[0012] Step 5: According to the working range of the liquid cooling network, multiple sets of input-output data sets of the decoupled single-row branch channel model running in the working state are obtained;

[0013] The input-output data set is used as training data, and a machine learning algorithm is used to generate a decoupled single-row runner reduced-order model ROM2;

[0014] Step 6: The pipe network includes a total of i rows of liquid supply branch channels. According to step 5, i decoupled single-row branch channel reduced-order models are generated to form a pipe network single-row branch channel sub-module;

[0015] Step 7: Generation of the main liquid supply / return channel submodule: The main liquid supply channel submodule calculates the flow distribution of each liquid supply branch channel, and performs bidirectional coupling calculation with the single-row branch channel submodule of each row of the pipe network. The main liquid supply channel submodule provides the calculated flow to the single-row branch channel submodule of the pipe network, and the single-row branch channel submodule of the pipe network feeds back the single-row flow resistance data to the main liquid supply channel submodule;

[0016] The liquid return main channel submodule calculates the flow distribution of each liquid return branch channel, and performs bidirectional coupling calculation with the pipe network single-row branch channel submodule of each row. The liquid return main channel submodule provides the calculated flow to the pipe network single-row branch channel submodule, and the pipe network single-row branch channel submodule feeds back the single-row flow resistance data to the liquid return main channel submodule;

[0017] Step 8: Scanning output submodule generation: The scanning output submodule separates the information of the specified liquid cooling cold plate of the specified basic unit according to the predefined liquid cooling cold plate position number and the specified flow information calculated by the supply / return liquid main channel submodule.

[0018] Furthermore, it also includes step 9: combining the supply / return liquid main channel sub-module, the pipe network single-row branch channel sub-module and the scanning output sub-module to establish a liquid cooling pipe network digital twin model based on row attributes and encapsulate it.

[0019] Furthermore, it also includes step 10: liquid cooling pipe network digital twin model SDK generation: input real-time measurement data and output SDK file.

[0020] Furthermore, the interface model of the liquid supply branch channel established in step 2 flowing into each liquid-cooled cold plate is encapsulated.

[0021] Furthermore, the liquid supply branch channel corresponding to each group of basic units in step 2 is established by the Modelica model.

[0022] Furthermore, the k liquid-cooled cold plates included in each basic unit in step 2 are connected in parallel.

[0023] Furthermore, the pipe network includes 81 basic units in 9 rows and 9 columns, and each basic unit includes 5 parallel liquid-cooled cold plates.

[0024] Furthermore, in step 10, the TwinDeployer module is used to output the SDK file.

[0025] Beneficial effects:

[0026] The present invention provides a method for constructing a digital twin model of a multi-node cold plate parallel liquid cooling pipe network, and proposes a method for modeling the main flow channel and branch flow channel inside the multi-node cold plate parallel liquid cooling pipe network. The present invention establishes a system model based on hybrid modeling of physical mechanism and machine learning method, which greatly shortens the simulation time from the original hours-days simulation time to seconds-minutes simulation time; the present invention provides key internal flow, temperature, and pressure data for the multi-node cold plate parallel liquid cooling pipe network, and realizes real-time monitoring of a large amount of internal unmeasurable data in the operation and maintenance of the liquid cooling pipe network; the present invention provides the necessary internal pipe network data for the operation fault alarm function of the digital twin system of the multi-node cold plate parallel liquid cooling pipe network, which greatly reduces the electromagnetic interference and false alarms and false alarms caused by the arrangement of sensor monitoring. BRIEF DESCRIPTION OF THE DRAWINGS

[0027] Figure 1 It is a multi-node cold plate parallel liquid cooling pipe network composition and flow channel pipeline diagram involved in an embodiment of the present invention;

[0028] Figure 2It is a basic unit in the liquid cooling network involved in the embodiment of the present invention;

[0029] Figure 3 It is the input / output interface of the digital twin model of the liquid cooling pipe network involved in the embodiment of the present invention;

[0030] Figure 4 It is a flow chart of a method for constructing a digital twin model of a multi-node cold plate parallel liquid cooling pipe network involved in an embodiment of the present invention;

[0031] Figure 5 It is a single-line liquid supply model diagram;

[0032] Figure 6 It is a composition diagram of a single-line liquid return model;

[0033] Figure 7 It is a decoupled single-row runner model diagram;

[0034] Figure 8 This is a comparative analysis diagram of the calculation results of the decoupled single-row branch channel model and the multi-row coupled model;

[0035] Fig. 9 This is a diagram of the numbering method of parallel liquid-cooled cold plates;

[0036] Fig.10 It is the composition and flow channel distribution diagram of the liquid cooling pipe network embodiment;

[0037] Fig.11 It is a flow chart for generating a liquid-cooled cold plate model;

[0038] Fig.12 This is a model diagram of a liquid-cooled cold plate;

[0039] Fig.13 It is a diagram of the single-line liquid supply model;

[0040] Fig.14 It is a diagram of the single-line liquid return model;

[0041] Fig.15 It is a diagram of the decoupled single-row runner model;

[0042] Fig.16 It is the training flow chart of the decoupled single-row split-flow channel reduced-order model;

[0043] Fig.17 It is a submodule diagram of a single-row diversion channel of a pipe network;

[0044] Fig.18 It is the scanned source submodule diagram;

[0045] Fig.19 This is the digital twin model diagram of the liquid cooling network;

[0046] Fig. 20This is the SDK interface diagram for the digital twin model of the liquid cooling network;

[0047] Fig.21 It is the liquid cooling pipe network digital twin model SDK file. DETAILED DESCRIPTION

[0048] The preferred mechanism and implementation method of the present invention are further described below in conjunction with the accompanying drawings and specific implementation methods.

[0049] like Figures 1 to 21 As shown, an embodiment of the present invention discloses a technical solution for a method for constructing a digital twin model of a multi-node cold plate parallel liquid cooling pipe network. The present invention relates to liquid cooling cold plate model generation, single-row liquid supply model generation, single-row liquid return model generation, decoupled single-row branch channel model generation, decoupled single-row branch channel reduced-order model generation, pipe network single-row branch channel sub-module generation, liquid supply / return main channel sub-module generation, scanning output sub-module generation, liquid cooling pipe network digital twin model generation, and liquid cooling pipe network digital twin model SDK generation.

[0050] The multi-node cold plate parallel liquid cooling pipe network is a unit array composed of several parallel cold plates. Its composition and pipe network pipeline diagram are as follows Figure 1 As shown. The pipe network consists of i×j basic units, such as Figure 2 As shown in the yellow dashed line. Each basic unit contains k parallel liquid-cooled cold plates (5 in the figure), that is, the entire liquid-cooled network contains i×j×K parallel liquid-cooled cold plates. The cold plates are numbered according to their locations in the network, where "cold plate ijk" means the kth liquid-cooled cold plate in the basic unit in the i-th row and j-th column in the network.

[0051] The liquid cooling network is equipped with liquid supply and return channels. There are 1 liquid supply and return main channels. The coolant enters the liquid supply main channel through the network supply port, flows through the i-row liquid supply branch channel to supply liquid to the i-row basic unit, flows into the liquid cooling cold plate for device cooling, enters the i, i+1-row return liquid branch channel to converge into the return liquid main channel, and finally flows out through the network return port to form a complete liquid cooling cycle.

[0052] The input and output interfaces of the digital twin model of the multi-node cold plate parallel liquid cooling network are as follows: Figure 3As shown. The input data of the model are the coolant temperature and flow rate at the inlet of the pipeline network, the coolant pressure at the outlet of the pipeline network, the heating power of cold plates with different positions, and the cold plate position data i, j, and K. The data can be obtained by real-time monitoring of sensors in physical form. The input data drives its digital twin model to carry out real-time calculation and simulation, and outputs the calculation results. It can calculate the outlet pressure, inlet flow rate, inlet enthalpy, heating power, coolant outlet temperature, etc. of the specified "cold plate ijk" in the pipeline network; in the phase change cooling system, it can also be used to calculate the steam mass fraction data of the coolant, which can be used to fully grasp the flow distribution and temperature distribution of each cold plate in the pipeline network, greatly reducing the electromagnetic interference and false alarms caused by the arrangement of sensor monitoring.

[0053] In the input data, the coolant temperature at the inlet of the network, the coolant flow rate at the inlet of the network, and the coolant pressure at the outlet of the network are the flow parameters of the supply and return flow channels of the liquid cooling network, which are used to drive the physical quantities of the flow mechanism in the entire model. The signal source can be a sensor signal or a cooling device connected to the outside of the liquid cooling network. The heating power of the cold plates of the liquid cooling network can be controlled individually, so the model interface is arranged separately for the heating power of each row and column of cold plates. Since the cold plates involved in the digital twin model of the liquid cooling network are all in parallel and numerous, each cold plate needs to calculate a large number of physical quantities, and it is impossible to output all physical quantities in parallel. Therefore, a scanning mechanism is adopted to first number all cold plates, and then the digital twin model of the liquid cooling network outputs all the physical quantities of the corresponding cold plates according to the number. According to the scanning mechanism, the row number, column number, and cold plate number input interface in the basic unit are set for the cold plate number scanning.

[0054] The output interfaces of this model include: the outlet pressure of the specified numbered cold plate, the inlet flow rate of the specified numbered cold plate, the inlet enthalpy of the specified numbered cold plate, the heating power of the specified numbered cold plate, the coolant temperature at the outlet of the specified numbered cold plate, the temperature of the specified numbered cold plate, the mass fraction of the coolant vapor at the outlet of the specified numbered cold plate, etc.

[0055] The overall architecture of the digital twin model construction method for the multi-node cold plate parallel liquid cooling network is as follows: Figure 4 As shown in the figure, the model is generated from three submodules, namely, the "supply / return liquid main channel submodule", the "pipe network single-row branch channel submodule" and the "scanning output submodule". Among the three modules, the supply / return liquid main channel submodule and the single-row branch channel submodule are bidirectionally coupled. The supply / return liquid main channel submodule sends the coolant flow signal of each row to the single-row branch channel submodule, and the single-row branch channel submodule sends the equivalent flow resistance signal of each row to the supply / return liquid main channel submodule, thereby realizing the flow calculation in the entire liquid cooling pipe network. The scanning output module outputs the specific information of the cold plate at the position number according to the single-row flow and supply / return liquid main channel pressure data and the input basic unit position number, including flow, temperature, pressure, calorific value and other data.

[0056] Example 1

[0057] This example takes a liquid cooling network including 9 rows and 9 columns of basic units as an example, wherein each basic unit includes 5 cold plates connected in parallel, and provides an implementation method and a specific implementation process. This example is based on single-phase liquid cooling, but is also applicable to a two-phase liquid cooling system, and this cannot be used to limit the scope of protection of the present invention.

[0058] Example Pipeline network distribution Fig.10 As shown, the network contains 81 basic units and 405 cold plates. The location number information is shown in Fig.10 shown.

[0059] The method for constructing a digital twin model of a multi-node cold plate parallel liquid cooling pipe network in this embodiment includes the following steps:

[0060] Step 1: Establish a simulation model of the internal flow channel of the liquid-cooled cold plate based on a detailed mechanism, generate a certain number of input boundary conditions according to the working range of the liquid-cooled cold plate, run the simulation model based on the detailed mechanism multiple times, obtain corresponding simulation results, and thus obtain the input-output data set of the liquid-cooled cold plate;

[0061] The input-output data set is used as training data and a machine learning algorithm is used to generate a liquid-cooled cold plate model ROM1.

[0062] Step 2: Around the liquid-cooled cold plate model, a single-row liquid supply model including a liquid supply manifold and a liquid-cooled cold plate is established;

[0063] Single-line liquid supply model Figure 5 The yellow framed liquid manifold is a row of j basic units supplying liquid, and each basic unit contains k liquid-cooled cold plates (k is 5 in the figure). As shown in each group, the liquid supply manifold corresponding to the basic unit is established by the Modelica model. The liquid-cooled cold plate model (ROM1) is generated by step 1 and nested in the single-row liquid supply model for use. Finally, the interface model of the liquid supply manifold flowing into each cold plate is established;

[0064] The specific single-line liquid supply model is as follows: Fig.13 As shown in the figure. The blue module is the pipeline between two basic units in the liquid supply channel, the rose red module is the liquid cooling cold plate model. The pink module is the basic unit interface model with two interfaces and three interfaces.

[0065] Step 3: Establish a single-row liquid return model including a liquid return branch channel and a liquid return interface;

[0066] Single-line return model Figure 6 As shown in the yellow box, it consists of a return liquid shunt channel and a return liquid interface between the return liquid shunt channel and the basic unit. The model is as follows Fig.14The orange module is the pipeline between two units in the return flow channel.

[0067] Step 4: To characterize the flow characteristics of multiple parallel pipes in the pipe network, this model construction method first decouples the complex multiple parallel flow channels in the pipe network. The decoupling method is as follows: Figure 7 As shown in the figure, the specific analysis method is: taking the liquid cooling cold plate model (ROM1) as the core, and coordinating with its flow mechanism, a complex flow network model of the liquid cooling network is constructed. Single-row decoupled and multi-row coupled network models are established respectively, and their key output parameters are investigated under the same boundary conditions. The comparison results of the output parameters are shown in Figure 8 As shown in the figure, the error between the decoupled single-row result and the 3-row / 5-row coupled result is about 0.0002%. Therefore, the decoupling of the complex flow network in the official website is acceptable, and the impact on the accuracy of multi-row parallel calculation is negligible.

[0068] Assemble the liquid-cooled cold plate model, the single-row liquid supply model and the single-row liquid return model to form a decoupled single-row flow channel model;

[0069] The model is composed of Fig.15 As shown, the capacitor element groups on the upper and lower sides are used to construct the decoupling split surface.

[0070] Step 5: According to the working range of the liquid cooling network, multiple sets of input-output data sets of the decoupled single-row branch channel model running in the working state are obtained;

[0071] The input-output data set is used as training data, and a machine learning algorithm is used to generate a decoupled single-row runner reduced-order model ROM2;

[0072] The specific workflow is as follows: Fig.16 shown.

[0073] Step 6: The pipe network includes a total of i rows of liquid supply branch channels. According to step 5, i decoupled single-row branch channel reduced-order models are generated to form a pipe network single-row branch channel sub-module;

[0074] Since the pipe network contains 9 rows in total, it is necessary to generate 9 single-row branch channel reduced-order models to form a single-row branch channel sub-module of the pipe network, such as Fig.17 shown.

[0075] Step 7: Generation of the main liquid supply / return channel submodule: The main liquid supply channel submodule calculates the flow distribution of each liquid supply branch channel, and performs bidirectional coupling calculation with the single-row branch channel submodule of each row of the pipe network. The main liquid supply channel submodule provides the calculated flow to the single-row branch channel submodule of the pipe network, and the single-row branch channel submodule of the pipe network feeds back the single-row flow resistance data to the main liquid supply channel submodule;

[0076] The liquid return main channel submodule calculates the flow distribution of each liquid return branch channel, and performs bidirectional coupling calculation with the pipe network single-row branch channel submodule of each row. The liquid return main channel submodule provides the calculated flow to the pipe network single-row branch channel submodule, and the pipe network single-row branch channel submodule feeds back the single-row flow resistance data to the liquid return main channel submodule;

[0077] Step 8: Scan output submodule generation: The scan output submodule separates the information of the specified liquid cooling cold plate of the specified basic unit according to the pre-defined liquid cooling cold plate position number and the specified flow information calculated by the supply / return liquid main channel submodule; the specific numbering method of the scan output module is as follows Fig. 9 As shown in the figure, i represents the row number, j represents the column number, and k represents the cold plate number in each basic unit. By inputting i, j, and k, each cold plate can be accurately located. The packaged scanning output submodule is shown in the figure. Fig.18 shown.

[0078] Step 9: Combine the main flow channel submodule for supply / return liquid, the single-row branch channel submodule for pipe network and the scanning output submodule to establish a digital twin model of the liquid cooling pipe network based on row attributes and encapsulate it; the encapsulated configuration is as follows: Fig.19 shown.

[0079] Step 10: Liquid cooling network digital twin model SDK generation: Use TwinDeployer module to output SDK file, which can realize the rapid calculation and output of liquid cooling network digital twin model calculation results. The specific process is as follows: Use TwinDeployer 2021R1 module to output SDK file. Import the twin file generated by Twin Bulider into the TwinDeployer environment and define the input data. The interface is as follows: Fig. 20 As shown; after completing the settings, you can directly output the SDK folder as follows Fig.21 shown.

[0080] The present invention provides a method for constructing a digital twin model of a multi-node cold plate parallel liquid cooling pipe network, and establishes a digital twin model of a cold plate parallel liquid cooling pipe network through one-dimensional Modelica. A liquid-cooled cold plate model is constructed, and its reduced-order model (ROM1) is formed through working condition training, and a "decoupled single-row shunt channel model" is constructed and its reduced-order model (ROM2) is formed through working condition training. A ROM model nesting technology and a parallel cold plate scanning output traversal algorithm are proposed. The method is used in the single-row shunt channel submodule and the scanning output submodule of this model. In the single-row shunt channel submodule, the above-mentioned single-row decoupling method is used to perform model training and reduction on the "decoupled single-row shunt channel model". In the scanning output module, each parallel cold plate model (ROM1) in the above-mentioned decoupled single-row shunt channel reduced-order model (ROM2) is nested, thereby realizing the function of outputting the cold plate calculation result of the corresponding position number according to the input cold plate position number.

[0081] The above method can realize the interconnection between the liquid cooling network and all parallel cold plates, build a digital twin model of multi-node cold plate parallel liquid cooling network, and finally realize the rapid calculation of operation monitoring and simulation of key data such as temperature, flow, pressure, etc. in the liquid cooling network and parallel cold plates. At the same time, the input boundary of the digital twin model calculation can be measured by a limited number of sensors, and the model can be driven by the measured data to realize the real-time simulation and visualization of liquid cooling heat dissipation performance driven by the measured data. Compared with the deployment of a large number of temperature, flow, and pressure sensors for monitoring, the design cost can be greatly reduced, and it has engineering significance for the heat dissipation evaluation, performance improvement, and reliability improvement of electronic equipment in the cold plate.

[0082] Finally, it should be noted that the above are only preferred embodiments of the present invention and are not intended to limit the present invention. Although the present invention has been described in detail with reference to the embodiments, those skilled in the art can still modify the technical solutions described in the aforementioned embodiments or make equivalent substitutions for some of the technical features therein. However, any modifications, equivalent substitutions, improvements, etc. made within the spirit and principles of the present invention should be included in the protection scope of the present invention.

Claims

1. A method for constructing a digital twin model of a multi-node cold plate parallel liquid cooling network. It is characterized in that The following steps are involved: Step 1: Establish a simulation model based on a detailed mechanism for the internal flow channel of the liquid-cooled cold plate. According to the working range of the liquid-cooled cold plate, run the simulation model based on the detailed mechanism multiple times to obtain an input-output data set of the liquid-cooled cold plate. The input-output data set is used as training data and a machine learning algorithm is used to generate a liquid-cooled cold plate model ROM1; Step 2: Building a single-row liquid supply model including a liquid supply manifold and a liquid cooling cold plate around the liquid cooling cold plate model; The liquid supply channel supplies liquid to a row of j basic units, each basic unit includes k liquid-cooled cold plates, and the liquid-cooled cold plate model ROM1 generated by step 1 is nested in the single-row liquid supply model to establish an interface model for the liquid supply channel to flow into each liquid-cooled cold plate; Step 3: Establish a single-row liquid return model including a liquid return branch channel and a liquid return interface; Step 4: Assemble the liquid-cooled cold plate model, the single-row liquid supply model, and the single-row liquid return model to form a decoupled single-row flow channel model; Step 5: According to the working range of the liquid cooling network, multiple sets of input-output data sets of the decoupled single-row branch channel model running in the working state are obtained; The input-output data set is used as training data, and a machine learning algorithm is used to generate a decoupled single-row runner reduced-order model ROM2; Step 6: The pipe network includes a total of i rows of liquid supply branch channels. According to step 5, i decoupled single-row branch channel reduced-order models are generated to form a pipe network single-row branch channel sub-module; Step 7: Generation of the main liquid supply / return channel submodule: The main liquid supply channel submodule calculates the flow distribution of each liquid supply branch channel, and performs bidirectional coupling calculation with the single-row branch channel submodule of each row of the pipe network. The main liquid supply channel submodule provides the calculated flow to the single-row branch channel submodule of the pipe network, and the single-row branch channel submodule of the pipe network feeds back the single-row flow resistance data to the main liquid supply channel submodule; The liquid return main channel submodule calculates the flow distribution of each liquid return branch channel, and performs bidirectional coupling calculation with the pipe network single-row branch channel submodule of each row. The liquid return main channel submodule provides the calculated flow to the pipe network single-row branch channel submodule, and the pipe network single-row branch channel submodule feeds back the single-row flow resistance data to the liquid return main channel submodule; Step 8: Scanning output submodule generation: The scanning output submodule separates the information of the specified liquid cooling cold plate of the specified basic unit according to the predefined liquid cooling cold plate position number and the specified flow information calculated by the supply / return liquid main channel submodule.

2. According to the method for constructing a digital twin model of a multi-node cold plate parallel liquid cooling pipe network as shown in claim 1, It is characterized in that It also includes step 9: combining the supply / return liquid main channel sub-module, the pipe network single-row branch channel sub-module and the scanning output sub-module to establish a liquid cooling pipe network digital twin model based on row attributes and encapsulate it.

3. According to the method for constructing a digital twin model of a multi-node cold plate parallel liquid cooling pipe network as shown in claim 1, It is characterized in that It also includes step 10: liquid cooling pipe network digital twin model SDK generation: input real-time measurement data and output SDK file.

4. According to the method for constructing a digital twin model of a multi-node cold plate parallel liquid cooling pipe network as shown in claim 1, It is characterized in that The interface model of the liquid supply branch channel established in step 2 flowing into each liquid-cooled cold plate is encapsulated.

5. According to the method for constructing a digital twin model of a multi-node cold plate parallel liquid cooling pipe network as shown in claim 1, Features: The liquid supply branch channel corresponding to each group of basic units in step 2 is established by the Modelica model.

6. According to the method for constructing a digital twin model of a multi-node cold plate parallel liquid cooling pipe network as shown in claim 1, Features: The k liquid-cooled cold plates included in each basic unit in step 2 are connected in parallel.

7. According to the method for constructing a digital twin model of a multi-node cold plate parallel liquid cooling pipe network as shown in claim 1, Features: The pipe network includes 81 basic units in 9 rows and 9 columns, and each basic unit includes 5 parallel liquid-cooled cold plates.

8. According to the method for constructing a digital twin model of a multi-node cold plate parallel liquid cooling pipe network as shown in claim 3, Features: In step 10, the TwinDeployer module is used to output the SDK file.

Citation Information

Patent Citations

  • Debugging method and system of immersed liquid cooling server based on digital twinning

    CN113961410A

  • Liquid-cooling cold plate model construction method based on digital twinning

    CN115048743A

  • Temperature field-structure field online simulation method, system and equipment for heat treatment

    CN117763788A

Cited By

  • Server heat dissipation adjusting method and device, electronic equipment and storage medium

    CN121099589A

  • A server heat dissipation adjusting method and device, electronic equipment and storage medium

    CN121099589B