A flowmeter-free adaptive heat dissipation control method and system based on the linkage between BMC liquid-cooled Tank and HDCV power supply system
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- HANGZHOU JINQUN TECHNOLOGY CO LTD
- Filing Date
- 2026-04-29
- Publication Date
- 2026-06-02
Smart Images
Figure CN122131893A_ABST
Abstract
Description
Technical Field
[0001] This invention relates to the field of thermal management technology for computing devices, and belongs to a flowmeter-free adaptive heat dissipation control method and system based on the linkage of BMC liquid-cooled Tank and HDCV power supply system. Background Technology
[0002] In high-density computing scenarios (referring to applications that achieve extremely high computing performance and energy efficiency within a limited physical space through highly integrated hardware and optimized architecture, with the core being the combination of "space efficiency" and "high performance"), such as AI server scenarios, the combination of liquid-cooled tanks (especially single-phase immersion liquid cooling) and HDCV (high-density DC) power supply systems has become the mainstream solution. However, existing technologies have the following limitations: directly measuring the flow rate, velocity, or volume of fluids such as liquids, gases, or steam through physical sensors to achieve metering, monitoring, or control increases system complexity; the heat dissipation control of liquid-cooled tanks (such as pump speed regulation) mainly relies on temperature feedback, while the power monitoring of HDCV power supply systems operates independently of the liquid cooling system. When the load power suddenly increases, such as when the GPU computing power jumps from 50% to 100%, the temperature change is lagging, causing the liquid cooling system to respond slowly and potentially causing the equipment to overheat temporarily. Existing liquid cooling controls mostly use fixed thresholds (such as increasing the pump speed when the temperature exceeds 40°C), without considering the differences in load types (such as AI training and general computing). Under the same power, the heat distribution of different loads is different (such as concentrated heat generation in the GPU core and uniform heat generation in the CPU). Fixed strategies are prone to overheating or underheating, increasing energy consumption. Summary of the Invention
[0003] To address the aforementioned technical problems, this invention provides a flowmeter-free adaptive heat dissipation control method based on the linkage between BMC liquid-cooled Tank and HDCV power supply system.
[0004] To achieve the above objectives, the present invention provides the following technical solution: A flowmeter-less adaptive heat dissipation control method based on BMC-driven liquid-cooled tank and HDCV power supply system includes the following steps: S1: Parameter Acquisition and Linkage: The HDCV power supply system acquires the input power P of the load in real time and transmits it to the substrate management controller (BMC). The liquid cooling Tank subsystem acquires the surface temperature T1 of the heating device, the liquid inlet temperature Tin, and the liquid outlet temperature Tout and transmits them to the substrate management controller (BMC). S2: Calling the parameter model and preparing control parameters: The baseboard management controller (BMC) calls the linkage parameter library, uses the current load power P and the set inlet and outlet liquid temperature difference ΔT to calculate the speed R of the target liquid cooling pump. In the linkage parameter library, there is a correspondence between the load power P, the inlet and outlet liquid temperature difference ΔT and the speed R of the target liquid cooling pump. S3: Meterless Adaptive Control: After the target liquid cooling pump speed R parameter is prepared, the baseboard management controller (BMC) sends a command to the corresponding liquid cooling pump drive unit, thereby adjusting the speed of the corresponding liquid cooling pump to the speed R of the target liquid cooling pump.
[0005] Furthermore, the establishment of the linkage parameter library is accomplished through the following steps: s21: For different types of load equipment, under various load power P conditions, test the corresponding stable inlet and outlet hydraulic temperature difference ΔT at different liquid cooling pump speeds R, and complete multiple sets of corresponding PR-ΔT data; s22: After obtaining multiple sets of corresponding PR-ΔT data, a functional relationship model between the load power P, the liquid cooling pump speed R and the inlet and outlet liquid temperature difference ΔT is established through data fitting, forming a linkage parameter model; s23: Then, the linkage parameter model is entered through the baseboard management controller (BMC) to form a linkage parameter library.
[0006] Furthermore, the functional relationship model between the load power P, the liquid cooling pump speed R, and the inlet and outlet liquid temperature difference ΔT in step s22 is specifically: R=k∙P+b−c∙ΔT, where k,b,c are fitting coefficients that need to be calibrated based on actual data.
[0007] Furthermore, S2 also includes a fine-tuning step for calculating the target liquid cooling pump speed R, specifically: the substrate management controller (BMC) corrects the initial target pump speed R based on the difference between the real-time collected inlet and outlet liquid temperature difference ΔT and the target inlet and outlet liquid temperature difference ΔT, to obtain the final target liquid cooling pump speed R.
[0008] Furthermore, S2 also includes predictive control under sudden load changes, specifically: the BMC calculates the power change rate dP / dt of the load power in real time, where t is the time unit s and P is the power unit W. When the power change rate dP / dt > 10% / s, the substrate management controller BMC predicts the load power P at future times based on the current power change trend. Then, the load power P predicted by the substrate management controller BMC is used as input to retrieve the predicted target pump speed R from the linkage parameter library, and the liquid cooling pump speed R is adjusted in advance based on this.
[0009] Furthermore, S2 also includes a model dynamic correction step. During system operation, the substrate management controller (BMC) continuously compares the deviation between the theoretical temperature difference predicted based on the current model and the actual temperature difference ΔT collected in real time. When the deviation continues to exceed a preset correction threshold, the model correction process is triggered to update the parameters of the linkage parameter library using new operating data.
[0010] Furthermore, S2 also includes a fault diagnosis step: when the load power is determined to be in a stable state, the baseboard management controller (BMC) monitors the real-time inlet and outlet liquid temperature difference ΔT. When ΔT is continuously lower than the preset abnormal flow threshold, it determines that the liquid cooling system has insufficient flow or pump failure, and triggers an alarm or protection action.
[0011] Furthermore, the baseboard management controller (BMC) communicates with the HDCV power supply system using the Modbus RTU protocol; the BMC is connected to the pump drive unit via a PWM signal.
[0012] Furthermore, a flowmeter-less adaptive heat dissipation control system based on the linkage between the BMC liquid-cooled tank and the HDCV power supply system is used in the above method. The system includes: a load device, a control signal output interface, a liquid-cooled pump drive unit, a substrate management controller (BMC) module, an HDCV power supply system, and a liquid-cooled tank subsystem. The BMC includes an intelligent control logic unit, a linkage parameter library, and a data acquisition and preprocessing module. The HDCV power supply system includes a function acquisition module, an HDCV control unit, and a DC power distribution module. The function acquisition module communicates with the data acquisition and preprocessing module of the BMC using the Modbus RTU protocol to supply power to the load device and acquire its power data. The liquid-cooled tank subsystem includes a liquid-cooled tank body and a temperature sensor array mounted on the tank body. The liquid-cooled tank subsystem communicates with the BMC. The BMC calls the model in the linkage parameter library and, based on the power data and the coolant temperature data, outputs a speed control signal to the liquid-cooled pump drive unit through the control signal output interface.
[0013] Compared with existing technologies, the present invention provides a design and testing method for managing immersion liquid-cooled computing units, which has the following advantages: 1. Reduce costs and improve reliability: Eliminating physical flow meters reduces hardware costs by 15%-20%, while reducing pipeline resistance and leakage points, improving system reliability by more than 25%.
[0014] 2. Improved heat dissipation response speed: By utilizing the real-time nature of HDCV power data (response ≤10ms), the problem of temperature feedback lag is solved, and the heat dissipation response speed during sudden load changes is improved by 30%-50%, avoiding short-term overheating of the equipment.
[0015] 3. Optimize energy consumption and adaptability: The adaptive strategy based on load type reduces pump energy consumption by 15%-20% (compared to fixed threshold control) and can adapt to the heat characteristics of different computing scenarios (such as AI training and scientific computing).
[0016] 4. Simplified system integration: The liquid cooling and power supply systems are managed in a unified manner through the Baseboard Management Controller (BMC), eliminating the need for additional gateways to convert protocols and reducing system integration complexity by 40%.
[0017] The features and advantages of the present invention will be described in detail through embodiments and in conjunction with the accompanying drawings. Attached Figure Description
[0018] Figure 1 This is an overall architecture diagram of the present invention; Figure 2 This is the parameter correspondence table of the present invention; Figure 3 This is a flowchart of the operation of the present invention. Detailed Implementation
[0019] To make the objectives, technical solutions, and advantages of this invention clearer, the invention will be further described in detail below with reference to the accompanying drawings and embodiments. However, it should be understood that the specific embodiments described herein are merely illustrative and not intended to limit the scope of the invention. Furthermore, descriptions of well-known structures and technologies are omitted in the following description to avoid unnecessarily obscuring the concept of the invention.
[0020] Specific implementation method one: Combining Figures 1 to 3 This embodiment describes the parameter acquisition and linkage process in a flowmeter-free adaptive heat dissipation control method based on BMC and linked with an HDCV power supply system: The HDCV power supply system acquires the input power P of the load in real time, wherein the accuracy of the input power P is ±1%, and transmits the input power P to the substrate management controller BMC. The liquid cooling tank has a multi-point temperature sensor built in it. The liquid cooling tank subsystem acquires the inlet temperature Tin and the outlet temperature Tout and transmits them to the substrate management controller BMC. The difference between the inlet temperature Tin and the outlet temperature Tout is ΔT. In this system, the Baseboard Management Controller (BMC) acts as the master device. The BMC establishes master-slave communication with the HDCV power supply system via an RS-485 bus that follows the Modbus RTU protocol. The BMC periodically reads the real-time collected load voltage and current data from the address specified by the HDCV power supply system after reading the holding register specified by the HDCV power supply system. An example of this process is as follows: the BMC reads the holding register address 0x0000 (voltage) and 0x0001 (current) specified by the HDCV power supply system, and then sends a query frame with function code 03. The HDCV returns the corresponding value. For HDCV's operational status, such as module failure and overload alarms, status monitoring can be achieved through Modbus register mapping and periodic reading by the BMC. When the liquid cooling system requires HDCV to reduce load, the BMC writes a load reduction instruction, such as 50% load, to the HDCV's control register (e.g., 0x0100) via function code 06 (written to a single register).
[0021] Specific Implementation Method Two: Combining Figures 1 to 3 This embodiment describes the parameter model and control parameter preparation process in a flowmeter-less adaptive heat dissipation control method based on BMC-driven liquid-cooled tank and HDCV power supply system linkage: By collecting the current load power P and the set inlet and outlet liquid temperature difference ΔT, the rotational speed R of the target liquid cooling pump is calculated. The load power P, inlet and outlet liquid temperature difference ΔT, and rotational speed R are input to the baseboard management controller (BMC) to establish a linkage parameter library. In the linkage parameter library, the correspondence between load power P, inlet and outlet liquid temperature difference ΔT, and rotational speed R of the target liquid cooling pump is set. The load power P determines the heat generated by the device's GPU or CPU. The higher the load power P, the greater the heat generated by the GPU or CPU. The greater the heat generated by the GPU or CPU, the greater the required flow rate of coolant. This part is an important process that affects the heat dissipation efficiency. Under different load powers P, a specific pump speed R needs to be matched in order to control the temperature difference ΔT within the optimal range. The optimal range is the state that ensures the device does not overheat while avoiding excessive energy consumption of the liquid cooling system. The establishment of the linkage parameter library also involves the following steps: s21: For different types of load equipment, under various load power P conditions, test the corresponding stable inlet and outlet hydraulic temperature difference ΔT at different liquid cooling pump speeds R, and complete multiple sets of corresponding PR-ΔT data; s22: After obtaining multiple sets of corresponding PR-ΔT data, a functional relationship model between the load power P, the liquid cooling pump speed R and the inlet and outlet liquid temperature difference ΔT is established through data fitting, forming a linkage parameter model; s23: Then, the linkage parameter model is entered through the baseboard management controller (BMC) to form a linkage parameter library.
[0022] The functional relationship model between the load power P, the liquid cooling pump speed R, and the inlet and outlet liquid temperature difference ΔT in step s22 is specifically: R = k∙P + b−c∙ΔT, where k, b, and c are fitting coefficients that need to be calibrated based on actual data. After encapsulating the fitted mathematical relationship into a model that can be called by the intelligent control logic unit, the calculation logic of "input power P and target temperature difference ΔT → output optimal pump speed R" is clarified. Based on the generated data values, a linear regression is used to fit the mapping relationship. Combining this with the above formula R = k∙P + b−c∙ΔT, the following data can be obtained: Using the first set of data (P=20, R=1500, ΔT=8): 1500=20k+b−8c; Using the second set of data (P=30, R=2000, ΔT=10): 2000=30k+b−10c; Using the third set of data (P=40, R=2500, ΔT=12): 2500=40k+b−12c; By solving the system of equations, the fitting coefficients are obtained: k=50, b=0, c=0 (After simplification, the approximate linear relationship is R=50P−0⋅ΔT, i.e., R=50P). At this point, the mapping model is as follows: For “GPU-X100”, the liquid cooling pump speed R = 50 × P. Based on the above process, multiple sets of correspondences between the liquid cooling pump speed R and the inlet and outlet liquid temperature difference ΔT corresponding to different load power P values can be obtained, and a linkage parameter library including multiple sets of data can be established.
[0023] Specific implementation method three: Combining Figures 1 to 3 This embodiment describes a flow meter-free adaptive heat dissipation control method based on BMC and linked with an HDCV power supply system. The control process is as follows: After the target liquid cooling pump speed R parameter is prepared, the baseboard management controller (BMC) sends a command to the corresponding liquid cooling pump drive unit, thereby adjusting the speed of the corresponding liquid cooling pump to the speed R of the target liquid cooling pump.
[0024] Specific implementation method four: Combination Figures 1 to 3 This embodiment describes an adaptive control process after calculating the target liquid-cooled pump speed R in step S2. Different control processes are implemented based on different load scenarios. 1. Fine-tuning process: The baseboard controller (BMC) queries the database based on the real-time power P of the HDCV, calls the corresponding pump speed R0, and then fine-tunes it through real-time ΔT. If ΔT is higher than the reference value by 1°C, R = R0 + 10%, and the initial target pump speed R is corrected to obtain the final target liquid cooling pump speed R.
[0025] 2. Predictive control process under sudden load: The BMC calculates the power change rate dP / dt of the load power in real time, where t is the time unit (s) and P is the power unit (W). When the power change rate dP / dt > 10% / s, the BMC calls the pump speed R 1-2 seconds in advance according to the predicted power P target value 1 second later. Then, the load power P predicted by the BMC is used as input, and the predicted target pump speed R is retrieved from the linkage parameter library. Based on this, the liquid cooling pump speed R is adjusted in advance.
[0026] Specific Implementation Method Five: Combining Figures 1 to 3 This embodiment describes a system where, due to deviations during long-term operation (such as equipment aging or changes in coolant performance), adjustments are made through a dynamic model correction step. Specifically, the Baseboard Management Controller (BMC) continuously compares the deviation between the theoretical temperature difference predicted by the current model and the actual temperature difference ΔT collected in real-time during system operation. When the deviation continuously exceeds a preset correction threshold (set to ±5% of the design accuracy), the model correction process is triggered. New operating data is used to update the parameters in the linkage parameter library. Specifically, the parameters refer to k, b, and c in the aforementioned formula. After the update, the model is ensured to always match the actual operating conditions. This entire process—establishing a baseline relationship through factory calibration → fitting and modeling to form computational logic → dynamic calling and deviation correction during runtime—achieves precise linkage between the liquid cooling and power supply systems, ensuring heat dissipation efficiency and energy consumption balance in high-density computing scenarios. An example of a real dynamic calling process: When the actual operating power of the GPU is P=35kW: The BMC obtains P=35kW through the power acquisition module of the HDCV power supply system; calls the corresponding mapping model of the GPU R=50×35=1750rpm; sends a command to the liquid cooling pump drive unit through the control signal output interface to adjust the pump speed to 1750rpm; the temperature sensor array of the liquid cooling Tank subsystem collects Tin and Tout, calculates ΔT and feeds it back to the BMC to verify the heat dissipation effect; Actual dynamic correction: If, after one year of operation, the heat generation efficiency changes due to GPU aging, resulting in a deviation such as "the model predicts R=1750rpm, but the actual ΔT=11℃ (exceeding the original design range of 10℃±0.5℃)", the BMC detects the deviation and triggers correction logic to re-collect multiple sets of (P,R,ΔT) data. For example, when P=35kW, R=1800rpm is adjusted, and ΔT=10℃ is measured (meeting design requirements). The new data (35,1800,10) is added to the GPU's linkage parameter library, and the model is refitted. After substituting the new data, the corrected model is obtained as R=51.43P−0.29ΔT. In subsequent calls, the model will dynamically calculate the pump speed based on the corrected coefficients to ensure that ΔT is always within a reasonable range. The adaptive strategy based on the load type reduces pump energy consumption by 15%-20%.
[0027] Specific Implementation Method Six: Combination Figures 1 to 3 This embodiment describes a flow meter-free adaptive heat dissipation control method based on a BMC-managed liquid-cooled tank and an HDCV power supply system, which further includes a fault diagnosis step: When the load power P is determined to be stable, the baseboard management controller (BMC) monitors the real-time inlet and outlet liquid temperature difference ΔT. When ΔT is continuously lower than a preset abnormal flow threshold, such as <2℃, it determines that the liquid cooling system flow is insufficient or the pump is faulty, completes the function of replacing the flow meter, and triggers a warning or replaces the backup pump if a backup pump is available to protect the overall cooling chain.
[0028] Specific Implementation Method Six: Combination Figures 1 to 3 This embodiment describes a flowmeter-less adaptive heat dissipation control system based on the linkage between a BMC liquid-cooled tank and an HDCV power supply system, used in the aforementioned method. The system includes: a load device, a control signal output interface, a liquid-cooled pump drive unit, a substrate management controller (BMC) module, an HDCV power supply system, and a liquid-cooled tank subsystem. The BMC includes an intelligent control logic unit, a linkage parameter library, and a data acquisition and preprocessing module. The HDCV power supply system includes a function acquisition module, an HDCV control unit, and a DC power distribution module. The function acquisition module communicates with the data acquisition and preprocessing module of the BMC using the Modbus RTU protocol to supply power to the load device and acquire its power data. The liquid-cooled tank subsystem includes a liquid-cooled tank body and a temperature sensor array mounted on the tank body. The liquid-cooled tank subsystem communicates with the BMC. The BMC calls a model from the linkage parameter library and, based on the power data and the coolant temperature data, outputs a speed control signal to the liquid-cooled pump drive unit through the control signal output interface.
[0029] The above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions or improvements made within the spirit and principles of the present invention should be included within the protection scope of the present invention.
Claims
1. A flowmeter-less adaptive heat dissipation control method based on the linkage of BMC liquid-cooled tank and HDCV power supply system, characterized in that, Includes the following steps: S1: Parameter acquisition: The HDCV power supply system acquires the input power P of the load in real time and transmits it to the substrate management controller (BMC). The liquid cooling Tank subsystem acquires the surface temperature T1 of the heating device, the liquid inlet temperature Tin and the liquid outlet temperature Tout and transmits them to the substrate management controller (BMC). The difference between the liquid inlet temperature Tin and the liquid outlet temperature Tout is ΔT. S2: Call and control parameter preparation: Calculate the target liquid cooling pump speed R by collecting the current load power P and the set inlet and outlet liquid temperature difference ΔT. Establish a linkage parameter library by inputting the load power P, inlet and outlet liquid temperature difference ΔT, and speed R into the baseboard management controller (BMC). In the linkage parameter library, the corresponding relationship between load power P, inlet and outlet liquid temperature difference ΔT, and target liquid cooling pump speed R is set. S3: Execution Control: After the target liquid cooling pump speed R parameter is prepared, the baseboard management controller (BMC) sends a command to the corresponding liquid cooling pump driver, thereby adjusting the speed of the corresponding liquid cooling pump to the speed R of the target liquid cooling pump.
2. The flowmeter-less adaptive heat dissipation control method based on the linkage of BMC liquid-cooled Tank and HDCV power supply system as described in claim 1, characterized in that: The establishment of the linkage parameter library is completed through the following steps: s21: For different types of load equipment, under various load power P conditions, test the corresponding stable inlet and outlet hydraulic temperature difference ΔT at different liquid cooling pump speeds R, and complete multiple sets of corresponding PR-ΔT data; s22: After obtaining multiple sets of corresponding PR-ΔT data, a functional relationship model between the load power P, the liquid cooling pump speed R and the inlet and outlet liquid temperature difference ΔT is established through data fitting, forming a linkage parameter model; s23: Then, the linkage parameter model is entered through the baseboard management controller (BMC) to form a linkage parameter library.
3. The flowmeter-less adaptive heat dissipation control method based on the linkage of BMC liquid-cooled tank and HDCV power supply system as described in claim 2, characterized in that: The functional relationship model between the load power P, the liquid cooling pump speed R, and the inlet and outlet liquid temperature difference ΔT in step s22 is as follows: , where k,b,c are the fitting coefficients.
4. The flowmeter-less adaptive heat dissipation control method based on the linkage of BMC liquid-cooled Tank and HDCV power supply system as described in claim 1, characterized in that: The S2 further includes a fine-tuning step for calculating the target liquid cooling pump speed R when the load is stable. Specifically, the substrate management controller (BMC) corrects the initial target pump speed R based on the difference between the real-time collected inlet and outlet liquid temperature difference ΔT and the target inlet and outlet liquid temperature difference ΔT, to obtain the final target liquid cooling pump speed R.
5. The flowmeter-less adaptive heat dissipation control method based on the linkage of BMC liquid-cooled Tank and HDCV power supply system as described in claim 1, characterized in that: The S2 also includes predictive control during load abrupt changes, specifically: the BMC calculates the power change rate of the load power in real time. , The unit of time is seconds (s). The power unit is W, and the rate of change of power is... If the load power P is greater than the preset value, the Baseboard Management Controller (BMC) predicts the load power P at future times based on the current power change trend. Then, using the load power P predicted by the Baseboard Management Controller (BMC) as input, the BMC retrieves the predicted target pump speed R from the linkage parameter library and adjusts the liquid cooling pump speed R in advance based on this.
6. The flowmeter-less adaptive heat dissipation control method based on the linkage of BMC liquid-cooled Tank and HDCV power supply system as described in claim 1, characterized in that: The S2 also includes a model dynamic correction step. During system operation, the substrate management controller (BMC) continuously compares the deviation between the theoretical temperature difference predicted based on the current model and the actual temperature difference ΔT collected in real time. When the deviation continues to exceed the preset correction threshold, the model correction process is triggered to update the parameters of the linkage parameter library using new operating data.
7. The flowmeter-less adaptive heat dissipation control method based on the linkage of BMC liquid-cooled Tank and HDCV power supply system as described in claim 1, characterized in that: The S2 also includes a fault diagnosis step: when the load power is determined to be stable, the baseboard management controller (BMC) monitors the real-time inlet and outlet liquid temperature difference ΔT. When ΔT is continuously lower than the preset abnormal flow threshold, it determines that the liquid cooling system has insufficient flow or pump failure, and triggers an alarm or protection action.
8. The flowmeter-less adaptive heat dissipation control method based on the linkage of BMC liquid-cooled Tank and HDCV power supply system as described in claim 1, characterized in that: The baseboard management controller (BMC) communicates with the HDCV power supply system using the Modbus RTU protocol; the BMC is connected to the pump drive unit via a PWM signal.
9. A flowmeter-less adaptive heat dissipation control system based on the linkage of BMC liquid-cooled tank and HDCV power supply system, characterized in that, The system for implementing the method of any one of claims 1 to 8 comprises: a load device, a control signal output interface, a liquid-cooled pump drive unit, a substrate management controller (BMC) module, an HDCV power supply system, and a liquid-cooled tank subsystem. The BMC includes an intelligent control logic unit, a linkage parameter library, and a data acquisition and preprocessing module. The HDCV power supply system includes a function acquisition module, an HDCV control unit, and a DC power distribution module. The function acquisition module communicates with the data acquisition and preprocessing module of the BMC using the Modbus RTU protocol to supply power to the load device and acquire its power data. The liquid-cooled tank subsystem includes a liquid-cooled tank body and a temperature sensor array disposed on the tank body. The liquid-cooled tank subsystem communicates with the BMC. The BMC calls the model in the linkage parameter library and, based on the power data and the coolant temperature data, outputs a speed control signal to the liquid-cooled pump drive unit through the control signal output interface.