Pump-driven two-phase multi-connected liquid cooling control method and system

Abstract

The invention relates to the technical field of data center thermal management, in particular to a pump-driven two-phase multi-connected liquid cooling control method and system, and the system comprises a liquid cooling module, a plurality of groups of cold plates, a sensor group and a central controller. The liquid cooling module comprises a plate heat exchanger, a liquid storage tank, a parallel refrigerant pump, a control valve group and a tail end flow regulating valve which form a closed circulation loop; the cold plates are arranged in parallel at heat source positions; the sensor group collects parameters such as temperature, pressure and flow; and the central controller executes a self-adaptive fuzzy PID algorithm to dynamically adjust the pump speed, corrects a flow set value through a two-phase flow stability optimization algorithm, starts and stops the pump according to periodic polling, and controls the on-off and the opening degree of a valve. The water-free fluorinated liquid is adopted as a refrigerant, and multi-pump redundancy, double-algorithm cooperation and independent flow regulation design are combined, so that the heat dissipation efficiency and the system stability are improved, the service life of equipment is prolonged, the multi-heat-source differentiation requirements are met, the operation safety is enhanced, and the energy consumption and the maintenance cost are reduced.

Description

Technical Field

[0001] This invention relates to the field of data center thermal management technology, specifically to a pump-driven two-phase multi-phase liquid cooling control method and system. Background Technology

[0002] With the rapid development of digital technology, the computing power density of servers in data center scenarios continues to rise, and the power density of core chips such as CPUs and GPUs is constantly increasing, placing stringent requirements on the efficiency, stability, and adaptability of heat dissipation systems. Traditional heat dissipation methods are gradually becoming insufficient to meet the heat dissipation needs of high power density heat sources, and their limitations are becoming increasingly apparent. Therefore, to address the above problems, a pump-driven two-phase multi-stage liquid cooling control method and system are proposed. Summary of the Invention

[0003] The purpose of this invention is to provide a pump-driven two-phase multi-phase liquid cooling control method and system to solve the problems mentioned in the background art.

[0004] To achieve the above objectives, the present invention provides the following technical solution: A pump-driven two-phase multi-phase liquid cooling control system, comprising: The liquid cooling module consists of a plate heat exchanger, a liquid storage tank, at least two refrigerant pumps connected in parallel, a control valve group, and a terminal flow regulating valve, forming a closed refrigerant circulation loop. Multiple cold plates are connected in parallel between the plate heat exchanger and the refrigerant pump and deployed at the server CPU / GPU heat source location; The sensor group includes a liquid supply temperature sensor, a liquid return temperature sensor, a cold plate outlet temperature sensor, a cold plate outlet pressure sensor, a flow sensor, and a pump inlet and outlet differential pressure sensor. The central controller, electrically connected to the liquid cooling module and sensor group, is configured to execute: Based on the difference between sensor data and a preset threshold, the refrigerant pump speed is dynamically adjusted using an adaptive fuzzy PID algorithm. Control the on / off state and opening degree of the control valve assembly; Each refrigerant pump is started and stopped in rotation according to a preset cycle to achieve balanced wear. The refrigerant flow rate setpoint is corrected in real time using a two-phase flow stability optimization algorithm.

[0005] As a preferred option, the refrigerant pump speed regulation adopts an adaptive fuzzy PID algorithm, and its output speed satisfies: ,in, for The refrigerant pump output speed command at any given time; This represents the real-time parameter deviation value. For deviation values ​​from 0 to Integral at time step; the derivative of the deviation value with respect to time; a proportional coefficient dynamically adjusted by fuzzy rules, an integral coefficient dynamically adjusted by fuzzy rules, a derivative coefficient dynamically adjusted by fuzzy rules, and satisfying: wherein, the dynamically adjusted PID coefficient, respectively represent the proportional p , integral , and derivative ; the PID initial coefficient; the correction amount of the fuzzy inference output.

[0006] As a preferred solution, generated by the following steps: define fuzzy input variables: the deviation of the fuzzy set is {negative large, negative medium, negative small, zero, positive small, positive medium, positive large}; the fuzzy set of the rate of change of the deviation is the same as before; construct a fuzzy rule base, and the rule form is: , , ; solving the fuzzy by the gravity method: , wherein, the correction amount of the fuzzy inference output, respectively represent the proportional p , integral , and derivative ; is the triggering strength of the first rule; is the defuzzification value of the fuzzy subset output by the first , rule; the numerator is the sum of the product of the triggering strength of each rule and the corresponding defuzzification value; and the denominator is the sum of the triggering strength of each rule.

[0007] As a preferred solution, the two-phase flow stability optimization algorithm is: wherein, is the corrected refrigerant flow set value; is the original flow set value; is the stability gain coefficient; is the pressure drop in the gas phase zone in the cold plate; is the density of the liquid refrigerant; is the latent heat of vaporization of the refrigerant; is the dryness at the outlet of the cold plate; The differential of the outlet dryness of the cold plate with respect to time.

[0008] As a preferred solution, the control valve group comprises a bypass valve, and the control logic thereof is: Flow control mode: When the secondary side flow is less than or equal to the opening threshold of the bypass valve, the bypass valve is opened; When the flow is greater than or equal to the closing threshold of the bypass valve, the bypass valve is closed; Linearly adjust the opening degree between the opening threshold and the closing threshold; Pressure difference control mode: When the pump inlet and outlet pressure difference is greater than or equal to the pressure difference opening threshold, the bypass valve is opened; When the pressure difference is less than or equal to the pressure difference closing threshold, the bypass valve is closed.

[0009] As a preferred solution, the opening degree of the bypass valve is controlled by a 0-10V analog signal, 0V corresponds to 0% opening degree, and 10V corresponds to 100% opening degree.

[0010] The central controller also performs: When the refrigerant pump demand speed is less than or equal to 0%, maintain the minimum speed operation; Real-time monitoring of fault signals, in response to refrigerant pump fault alarm immediately close the corresponding pump body.

[0011] As a preferred solution, the refrigerant circulation loop uses anhydrous fluorinated liquid as the refrigerant to realize gas-liquid phase change heat absorption.

[0012] A pump-driven two-phase multi-connection liquid cooling control method is applied to a system, and the control method comprises the following steps: Step S1: Real-time acquisition of liquid supply temperature, return liquid temperature, cold plate outlet dryness, flow and pump pressure difference data by a sensor group; Step S2: The central controller executes a two-phase flow stability optimization algorithm to output a corrected flow set value

[0013] Step S3: Calculate the refrigerant pump demand speed using an adaptive fuzzy PID algorithm; Step S4: Switch the refrigerant pump on and off according to the preset period, and dynamically adjust the opening degree of the bypass valve; Step S5: Linearly adjust the opening degree of the bypass valve based on the flow or pressure difference threshold.

[0014] As a preferred solution, the proportional coefficient of the adaptive fuzzy PID algorithm in step S3 The dynamic adjustment rule includes: When And , wherein, is the absolute value of the real-time parameter deviation value; The absolute value of the rate of change of deviation; This is the dynamically adjusted proportional coefficient; This is the initial proportional coefficient; when and hour, ,in, This represents the absolute value of the real-time parameter deviation. This represents the absolute value of the rate of change of the deviation. This is the dynamically adjusted proportional coefficient; This is the initial proportional coefficient.

[0015] As can be seen from the technical solution provided by the present invention above, the beneficial effects of the pump-driven two-phase multi-phase liquid cooling control method and system provided by the present invention are: I. Significantly improved heat dissipation efficiency and thermal control accuracy: Combining phase change heat absorption with high-efficiency heat exchange: The system uses anhydrous fluorinated liquid as a refrigerant, utilizing its gas-liquid phase change heat absorption characteristics, combined with direct contact heat exchange between the cold plate and the heat source. Compared with traditional air cooling or pure liquid cold plates, the heat dissipation capacity is significantly improved, which can effectively meet the heat dissipation requirements of high power density heat sources. Dynamic flow precise adaptation: The central controller, through the synergy of adaptive fuzzy PID algorithm and two-phase flow stability optimization algorithm, combined with the independent control of the terminal flow regulating valve, can respond to heat source load fluctuations in real time, quickly complete flow and pump speed regulation, effectively control cold plate outlet temperature fluctuations, and solve the problems of lag or local overheating in traditional liquid cooling systems. II. Significantly improved system reliability and equipment lifespan: Multi-pump redundancy and balanced wear design: The liquid cooling module adopts at least two refrigerant pumps connected in parallel, and with the central controller's rotating start-stop mechanism, the operating load of each pump tends to be balanced, extending the overall service life of the pump group; and in the event of a single pump failure, it can quickly switch to the standby pump to ensure uninterrupted refrigerant circulation and improve system redundancy; Dual-mode protection of pressure and flow: The bypass valve of the control valve group can be adjusted in coordination with the flow control mode and the differential pressure control mode. It can quickly divert the refrigerant to balance the state when the circuit flow is abnormal or the pressure is unbalanced, so as to avoid pump overload or pipeline damage and reduce the risk of system failure and shutdown. Rapid fault response and isolation: Relying on the full-link parameter monitoring of the sensor group and the fault diagnosis function of the central controller, anomalies can be located in real time, and faults can be isolated by closing the corresponding branch valves and switching redundant equipment to prevent the fault from spreading and shorten the fault handling response time. Third, it has strong adaptability to various scenarios and is compatible with diverse needs from multiple heat sources: Multiple cold plates connected in parallel with independent adjustment: Multiple cold plates are connected through parallel pipelines. Each cold plate has a terminal flow regulating valve connected in series at its inlet. The refrigerant flow can be independently allocated for heat sources with different power, avoiding the problem of "insufficient heat dissipation for large heat sources and wasted flow for small heat sources". It is compatible with multi-chip heterogeneous hardware architecture. Dynamic adaptation across a wide load range: Under low load, the central controller reduces refrigerant flow by decreasing pump speed and adjusting valve opening to avoid ineffective energy consumption; under high load, multiple pumps can be started simultaneously to increase the total flow and adapt to different scales of heat dissipation needs. IV. Safety and Energy Efficiency Optimization: Waterless coolant and insulation design: Using waterless fluorinated liquid as the coolant, its insulation properties can avoid the risk of leakage and short circuit in traditional water cooling; even if a small amount of leakage occurs, it will not damage the core components, greatly improving the safety of system operation. Algorithm-driven low-energy operation: The adaptive fuzzy PID algorithm dynamically adjusts the pump speed to avoid unnecessary energy consumption; the two-phase flow stability optimization algorithm corrects the flow setpoint to reduce the amount of refrigerant circulating while ensuring heat dissipation, thereby indirectly reducing the power consumption of the pump and reducing the overall system energy consumption. V. Improved ease of operation and maintenance: Precise monitoring and fault location: The sensor group collects various parameters in real time, and the central controller accurately locates the fault point through parameter correlation analysis, reducing the troubleshooting time of maintenance personnel; Operational data traceability and optimization: The central controller automatically records system operation logs, providing data support for maintenance plan formulation and avoiding blind maintenance; at the same time, historical data can be used to optimize algorithm parameters, further improving system adaptability. Attached Figure Description

[0016] Figure 1 This is a schematic diagram of the overall structure of a pump-driven two-phase multi-phase liquid cooling control system according to the present invention. Detailed Implementation

[0017] 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. It should be understood that the specific embodiments described herein are merely illustrative and not intended to limit the invention.

[0018] To better understand the above technical solutions, the following will provide a detailed explanation of the technical solutions in conjunction with the accompanying drawings and specific embodiments.

[0019] like Figure 1 As shown, this embodiment of the invention provides a pump-driven two-phase multi-phase liquid-cooled control system, including a liquid-cooling module, multiple sets of cold plates, a sensor group, and a central controller.

[0020] In this embodiment, the liquid cooling module consists of a plate heat exchanger, a liquid storage tank, at least two refrigerant pumps connected in parallel, a control valve group, and an end flow regulating valve, forming a closed refrigerant circulation loop. Furthermore, the liquid cooling module is the "power core and heat exchange hub" of the pump-driven two-phase multi-phase liquid cooling control system. It achieves efficient refrigerant delivery and heat exchange by constructing a closed refrigerant circulation loop, providing stable heat dissipation support for heat sources such as server CPUs / GPUs. The following will elaborate on this module from an overall perspective to specific details: I. Overall Function Overview: The liquid cooling module is primarily responsible for constructing and maintaining a closed-loop refrigerant circulation. Through the coordinated operation of its internal components, it completes key tasks such as refrigerant storage, pressurization, flow distribution, heat exchange, and pressure stabilization. Its core functions include: providing refrigerant flow power through parallel refrigerant pumps; achieving heat exchange between the refrigerant and the outside environment through plate heat exchangers; ensuring stable refrigerant flow in the loop through a liquid storage tank; and precisely regulating the refrigerant flow and direction using control valve groups and terminal flow regulating valves—ultimately forming a complete closed loop of "power transmission - heat exchange - flow adaptation," ensuring that the refrigerant flows through the cold plates at an appropriate flow rate and state, efficiently absorbing heat from the heat source and providing basic support for system heat dissipation. II. Submodule Composition and Functions: (a) Plate heat exchanger unit: Structure and Connection: The heat exchange core is formed by stacking metal corrugated plates. One side passage (primary side) is connected to the external cooling system (such as the computer room chiller unit), and the other side passage (secondary side) is connected to the refrigerant circulation loop of the liquid cooling module. That is, the inlet is connected to the outlet of the liquid storage tank, and the outlet is connected to multiple cold plate inlets through pipelines. Core function: As a "bridge" for heat exchange, the refrigerant (which may be in a gas-liquid two-phase state) that has absorbed heat from the heat source in the refrigerant circulation loop exchanges heat with the cooling medium on the primary side through the efficient heat conduction of the metal plate. After the refrigerant releases heat, its temperature decreases (the gaseous refrigerant can condense into a liquid state), and it regains its efficient heat absorption capacity, laying the foundation for subsequent heat dissipation by the cold plate. Its corrugated plate structure increases the heat exchange area and can adapt to the large heat exchange requirements in high power density scenarios. (ii) Liquid storage tank unit: Structure and installation location: It adopts a sealed pressure-resistant tank body, with a liquid level sensor (connected to the sensor group) inside. It is connected in series through pipeline between the secondary side outlet of the plate heat exchanger and the inlet of the refrigerant pump to form a "buffer node" of the circulation loop. Core functions: Refrigerant storage and replenishment: Store a certain amount of anhydrous fluorinated liquid refrigerant. When the refrigerant quantity fluctuates due to a small leak or phase change in the circuit, the refrigerant is automatically replenished through the change of liquid level inside the tank to maintain the stability of the total refrigerant quantity in the circuit. Gas-liquid separation and pressure buffering: Since the refrigerant may be in a gas-liquid two-phase state during circulation, the expansion space inside the storage tank can achieve preliminary gas-liquid separation, avoiding the direct entry of gaseous refrigerant into the pump body and causing "cavitation"; at the same time, it buffers the pressure fluctuation of the circuit - when the pump starts or stops or the flow rate is adjusted, the tank can alleviate the sudden increase in pipeline pressure by accommodating some refrigerant, ensuring the stability of the circuit pressure. (III) Parallel refrigerant pump unit: Structure and connection method: It includes at least two refrigerant pumps of the same model (such as positive displacement gear pumps) connected in parallel to the pipeline: the inlet of all pumps is connected to the outlet of the liquid receiver tank through a connecting pipeline, and the outlet is connected to the inlet of the control valve group through a connecting pipeline; each pump is equipped with an independent drive motor and start / stop control module, and is electrically connected to the central controller; Core functions: Provides circulating power: The pump body is driven by a motor to rotate, pressurizing the refrigerant output from the storage tank, so that the refrigerant can obtain sufficient pressure energy to overcome pipeline resistance, cold plate flow resistance, etc., and realize the circulating flow from the pump outlet to the cold plate, plate heat exchanger and back to the storage tank. Multi-pump coordination and rotation operation: The system executes preset cycle start-stop operations according to the central controller's instructions. For example, the system is set to rotate the operating pumps every 6 hours. When pump A has run for 6 hours, the central controller controls pump A to stop and pump B to start, achieving balanced wear of multiple pumps and extending the overall service life of the pump group. At the same time, the parallel operation of multiple pumps can provide redundancy support. If the currently operating pump fails (such as abnormal speed or insufficient pressure), the central controller will respond to the fault alarm and immediately shut down the faulty pump and start the standby pump to ensure that the power of the circuit is not interrupted. Dynamic speed adjustment: The pump's drive motor supports adjustable speed. The central controller calculates the required speed based on sensor data (such as cold plate outlet temperature and return liquid temperature) through an adaptive fuzzy PID algorithm, and then adjusts the motor speed. When the speed increases, the pump's output flow increases, the refrigerant circulation speed increases, and the heat dissipation capacity is enhanced; conversely, the flow decreases, thus achieving dynamic matching between heat dissipation capacity and heat source demand. (iv) Control valve assembly unit: Composition and core components: Includes a bypass valve and several on / off valves (such as solenoid valves). The two ends of the bypass valve are connected to the refrigerant pump outlet main pipeline and the liquid receiver inlet pipeline respectively (forming a bypass branch). The on / off valves are connected in series in the main circuit branch or the pump's independent inlet and outlet pipelines. Core functions: Bypass regulation: Flow control mode and differential pressure control mode are executed according to the central controller's instructions. In flow control mode, when the sensor group detects that the secondary side (i.e., the refrigerant entering the cold plate) flow rate is ≤ the bypass valve opening threshold, the bypass valve opens; when the flow rate is ≥ the closing threshold, the bypass valve closes; if the flow rate is between the two thresholds, the bypass valve adjusts its opening degree linearly according to the 0-10V analog signal (0V corresponds to 0% opening degree, 10V corresponds to 100% opening degree) – diverting part of the refrigerant through the bypass branch to reduce the flow rate entering the main circuit and avoid overload of the cold plate flow rate; In differential pressure control mode, when the differential pressure sensor at the pump inlet and outlet detects a differential pressure greater than or equal to the differential pressure opening threshold, the bypass valve opens, allowing some refrigerant to flow back to the storage tank through the bypass branch, thereby reducing the pump outlet pressure; when the differential pressure is less than or equal to the closing threshold, the bypass valve closes, ensuring that the pump output pressure meets the circuit requirements. Circuit on / off and branch control: On / off valves are turned on and off according to the instructions of the central controller—for example, when a group of cold plates needs maintenance, the on / off valve of the corresponding branch is closed to cut off the refrigerant supply to that cold plate; or when only part of the heat source is running, the branch on / off valve of the idle cold plate is closed to achieve directional distribution of refrigerant flow. (v) Terminal flow regulating valve unit: Installation location and structure: Each cold plate has a terminal flow regulating valve connected in series on its inlet pipe. It adopts an electric regulating ball valve structure and is electrically connected to the central controller. It can receive 0-10V control signals to adjust the opening degree (the opening degree corresponds to the change in the flow coefficient). Core Functions: As a "fine-grained flow distribution node," it adjusts the refrigerant flow of the corresponding cold plate according to the central controller's instructions; based on the cold plate outlet temperature sensor data (e.g., if the outlet temperature of a cold plate is higher than the preset value), the central controller controls the end flow regulating valve at the inlet of that cold plate to increase its opening—increasing the amount of refrigerant flowing into that cold plate and enhancing its heat absorption capacity; conversely, it reduces the opening to avoid refrigerant waste; through the independent adjustment of each valve, it achieves flow matching between multiple cold plates, ensuring that different heat sources (such as CPU and GPU, or GPUs with different loads) receive matching heat dissipation resources; III. Key Technology Principles: (I) Closed-loop and phase change heat dissipation principle: The core of the liquid cooling module is a "closed refrigerant circulation loop": when the anhydrous fluorinated liquid refrigerant circulates in the loop, it absorbs heat from the CPU / GPU as it flows through the cold plate. Part of the liquid refrigerant undergoes a phase change (gas-liquid phase change absorbs heat), forming a two-phase gas-liquid state. Subsequently, the two-phase refrigerant flows into the secondary side of the plate heat exchanger, exchanges heat with the cooling medium on the primary side, and the gaseous refrigerant condenses into a liquid state, releasing heat. Finally, the refrigerant flows back to the liquid storage tank, is pressurized by the pump, and re-enters the circulation. Through the cycle of "phase change absorbs heat - condensation releases heat," the continuous transfer of heat from the heat source is achieved. The closed structure can prevent refrigerant leakage and prevent external impurities from entering the loop, contaminating the refrigerant, or clogging components. (II) Principle of multi-pump parallel and rotational control: The multi-pump parallel design is based on the concept of "redundancy and balance": a single pump can meet the basic flow requirements when running, while multiple pumps in parallel can increase the total flow by starting simultaneously (under high load), adapting to scenarios with sudden increases in heat sources; and the rotation start-stop mechanism switches the running pumps according to the time period preset by the central controller (such as 8 hours / time), avoiding long-term continuous operation of a single pump - the cumulative running time of each pump tends to be balanced, reducing the risk of failure caused by excessive wear of a single pump. Essentially, it extends the life of the pump group through "load distribution in the time dimension". (III) Dual-mode control principle of bypass valve: The bypass valve's dual-mode control of flow and differential pressure is essentially a "dynamic balance adjustment of the loop": the flow control mode is for "cold plate flow adaptation." When the secondary side flow is too low (possibly due to cold plate blockage), opening the bypass valve can prevent all the refrigerant output by the pump from accumulating in the main loop, resulting in excessive pressure. The differential pressure control mode is for "pump protection." When the pressure difference between the pump inlet and outlet is too large (possibly due to pipeline blockage or insufficient valve opening), the bypass branch can reduce the pump outlet pressure and prevent the pump from being damaged due to overload. The two modes work together to ensure that the loop flow and pressure are within a safe and compatible range. IV. Module Workflow: (a) Initialization phase: After the system starts, the liquid cooling module enters initialization according to the central controller's instructions: the central controller controls all refrigerant pumps to start briefly (10 seconds), and detects whether each pump can establish a basic pressure difference through the pump inlet and outlet differential pressure sensors (to determine whether the pump body is normal); at the same time, it detects whether the initial opening of each valve in the control valve group (bypass valve defaults to 0% opening, on / off valve defaults to closed), and whether the terminal flow regulating valve is in the initial position (such as 50% opening). The liquid level sensor in the storage tank provides feedback on the refrigerant level. If the liquid level is lower than the preset lower limit, the central controller issues a "refrigerant replenishment" prompt; if the liquid level is normal, the module completes initialization and enters standby mode. (II) Normal operation phase: Refrigerant circulation start-up: The central controller starts one refrigerant pump (selecting the initial running pump according to the rotation strategy) based on the initial heat source temperature fed back by the sensor group (such as the initial CPU temperature of 35°C). The pump pressurizes the anhydrous fluorinated liquid in the receiver tank and delivers it to the control valve group. At this time, the bypass valve is closed and the on / off valve is opened according to the preset (such as all cold plate passages are open). The refrigerant flows through the main circuit to the terminal flow regulating valve. Initial flow distribution: The terminal flow regulating valve adjusts its opening according to the initial command of the central controller (e.g., according to the rated power of the heat source, 60% opening is allocated to the CPU cold plate and 70% opening is allocated to the GPU cold plate). The refrigerant flows through each cold plate according to the corresponding flow rate, absorbs heat from the heat source (part of the refrigerant undergoes phase change), and then flows into the secondary side of the plate heat exchanger. Heat exchange: Cooling medium (such as chilled water in the computer room) is introduced into the primary side of the plate heat exchanger and exchanges heat with the refrigerant on the secondary side. The refrigerant releases heat, the gaseous refrigerant condenses into liquid, and after the temperature drops to a preset value (such as 30°C), it flows back to the storage tank through the pipeline to complete one cycle. (III) Dynamic Adjustment Phase: Pump speed regulation: When the heat source load increases (e.g., GPU load increases from 50% to 100%), the cold plate outlet temperature sensor detects that the temperature has risen to 45℃ (above the preset threshold of 38℃), and the central controller calculates the real-time parameter deviation value. (e.g., 7℃), the pump speed is dynamically adjusted using an adaptive fuzzy PID algorithm: if For larger values ​​(e.g., >15℃), increase the scaling factor. Rapidly increase pump speed—increase pump output flow, accelerate refrigerant circulation, enhance the heat absorption capacity of cold plate, until the temperature drops back to within the threshold. Bypass valve adjustment: If the secondary flow sensor detects that the flow rate drops to the bypass valve opening threshold (e.g., 10L / min), the central controller controls the bypass valve to open and linearly adjusts the opening degree (e.g., outputting a 3V signal, corresponding to a 30% opening degree). Some refrigerant flows back to the storage tank through the bypass branch, and the main circuit flow rate is maintained within a safe range. If the pressure difference between the pump inlet and outlet rises to the pressure difference opening threshold (e.g., 0.5MPa), the bypass valve immediately opens (e.g., 50% opening degree), reducing the pump outlet pressure to a safe value (e.g., 0.3MPa) before gradually closing it. Fine-tuning of terminal flow: If the temperature sensor at the outlet of a cold plate detects that the local temperature is too high (e.g., the outlet temperature of a CPU cold plate is 42℃, while others are 38℃), the central controller controls the terminal flow regulating valve corresponding to that cold plate to increase the opening (e.g., from 60% to 75%), thereby increasing the refrigerant flow to that cold plate and specifically reducing its temperature to achieve "on-demand distribution". (iv) Patrol and Fault Response Phase: Multi-pump rotation: When the running pumps have accumulated a preset cycle (e.g., 6 hours), the central controller performs a rotation: first, the standby pump (e.g., pump B) is started; after the speed of pump B is stable and the circuit flow is stable, the original running pump (e.g., pump A) is shut down, and the switching is completed—pump A enters a stop and rest state, and is started again during the next rotation. Pump failure handling: If the operating pump fails (such as a sudden drop in speed or a fault signal being triggered), the central controller will immediately shut down the faulty pump and start the standby pump. The entire switching process will be completed within 5 seconds. If the standby pump is still abnormal after starting (such as no pressure difference), a third pump (if available) will be started to ensure that the power in the circuit is not interrupted. Handling extreme situations: If the central controller calculates that the required speed of the refrigerant pump is ≤0% (such as when the heat source stops and there is no heat dissipation requirement), the pump is controlled to maintain the minimum speed to avoid damage to the pump body due to frequent start-stop; at the same time, the bypass valve is opened to the maximum opening, and the refrigerant circulates through the bypass branch at a small flow rate to maintain stable circuit pressure. (v) Shutdown phase: When the system receives a shutdown command (such as server shutdown), the central controller controls the refrigerant pump speed to gradually decrease (to avoid a sudden drop in pressure), and shuts it off after the speed drops to the minimum speed; at the same time, all on-off valves are closed, and the bypass valve is opened to 50% (to balance the residual pressure in the circuit); finally, the module operation data (such as the running time of each pump and the number of valve adjustments) are recorded to complete the shutdown.

[0021] In this embodiment, multiple sets of cold plates are connected in parallel between the plate heat exchanger and the refrigerant pump, and deployed at the server CPU / GPU heat source location; Furthermore, the multiple cold plates serve as the "heat exchange terminals" of the pump-driven two-phase multi-stage liquid cooling control system. They are directly attached to core heat sources such as server CPUs / GPUs, and through the flow and phase change of the internal refrigerant, they efficiently transfer the heat generated by the heat source to the refrigerant, making them a key carrier for achieving precise heat dissipation in the system. The following section provides a detailed explanation of the multiple cold plates, from the overall structure to the specifics: I. Overall Function Overview: The core function of multiple cold plates is to act as a heat transfer hub between heat source and refrigerant: through close contact with heat sources such as CPUs / GPUs, the heat generated is quickly transferred into the internally flowing anhydrous fluorinated liquid refrigerant; at the same time, with the help of the internal flow channel design, the refrigerant is guided to form a stable flow (or gas-liquid two-phase flow), allowing the refrigerant to fully absorb heat (some liquid refrigerant will undergo phase change vaporization), and then the heat-absorbing refrigerant is transported back to the plate heat exchanger to complete the heat release, ultimately achieving "directional heat transfer from the heat source"; in addition, the multiple cold plates are set in parallel, which can independently adapt to heat sources of different locations and power (such as CPUs and multiple GPUs in the same server), and with the individual control of the terminal flow regulating valve, the heat dissipation requirements of each heat source can be precisely matched; II. Structural composition and function: (a) Cold plate body structure: Substrate: As the contact carrier between the cold plate and the heat source, it is made of a material with high thermal conductivity (such as copper or aluminum alloy) and is usually 3-5mm thick. The lower surface of the substrate is precision machined (roughness ≤1.6μm) and is tightly attached to the heat sink of the CPU / GPU through thermal grease or phase change pads to ensure that the heat from the heat source is efficiently transferred to the cold plate through thermal conduction. The upper surface is sealed to the flow channel cover to form a closed refrigerant flow channel space. Internal flow channels: The flow channels are located on the upper surface of the substrate or inside the flow channel cover plate, and are in the form of "serpentine", "forked" or "microchannel array" structure (to adapt to different heat source shapes). The cross-sectional dimensions of the flow channels are usually 2-5mm (width) × 1-3mm (depth). Their function is to guide the refrigerant to flow evenly in the cold plate. When flowing directly above the heat source, the refrigerant makes full contact with the substrate to absorb heat. At the same time, the tortuous design of the flow channels can prolong the residence time of the refrigerant in the cold plate, improve the heat absorption efficiency, and avoid "heat exchange blind zones" caused by excessive local flow velocity. Inlet and outlet interfaces: Each cold plate is equipped with an independent refrigerant inlet and outlet interface (such as a quick-connect pagoda connector). The inlet is connected to the main liquid supply line between the plate heat exchanger and the terminal flow regulating valve through a branch pipeline, and the outlet is connected to the main liquid return line between the cold plate and the liquid storage tank through a branch pipeline. Sealing gaskets (such as fluororubber) are provided at the interface to prevent refrigerant leakage. (ii) Parallel connection structure: The parallel connection of multiple cold plates is achieved through pipeline branching: the main liquid supply pipeline extends from the outlet of the plate heat exchanger and is divided into branch liquid supply pipelines matching the number of cold plates through multi-port connectors (such as 3-port and 4-port). Each branch pipeline is connected in series with an end flow regulating valve and then connected to the inlet of the corresponding cold plate. The outlets of each cold plate are collected into the main return liquid pipeline through branch return liquid pipelines and finally flow back to the storage tank. The core of this parallel structure is "independent flow intake for each cold plate" - the flow regulation of a certain cold plate (such as the change of the opening of the end flow regulating valve) will not affect the refrigerant supply of other cold plates, ensuring that each group of cold plates can be individually adapted to the heat dissipation needs of its corresponding heat source. (III) Sensor adapter structure: Sensor installation interfaces are reserved at the cold plate outlet: Temperature sensor installation holes (built-in cold plate outlet temperature sensor) are provided on the pipe wall of the cold plate outlet pipeline to monitor the temperature of the refrigerant when it leaves the cold plate in real time (reflecting the state of the refrigerant after heat absorption); some cold plate outlet pipelines are also equipped with pressure sensor interfaces (built-in cold plate outlet pressure sensor) to monitor the pressure of the refrigerant when it flows out (to help determine whether there is a blockage in the cold plate); these sensors are electrically connected to the central controller through wires to provide data support for flow regulation and fault diagnosis; III. Key Technology Principles: (I) The principle of synergistic effect of contact heat transfer and phase change: The efficient heat dissipation of the cold plate relies on the synergy of "thermal conduction + phase change heat absorption": the heat generated by the heat source (such as the CPU) is first transferred to the cold plate substrate through thermal conduction (the substrate has a thermal conductivity ≥380W / (m・K) and a thermal resistance ≤0.05℃ / W), and then transferred from the substrate to the anhydrous fluorinated liquid refrigerant in the flow channel; when the refrigerant absorbs heat and reaches its boiling point, part of the liquid refrigerant undergoes a phase change. The latent heat of vaporization absorbed during the phase change (which is much greater than the sensible heat of the liquid refrigerant) can quickly reduce the substrate temperature. For example, the latent heat of vaporization of 3M fluorinated liquid is about 120kJ / kg, and the heat absorbed by 1L of refrigerant during phase change is equivalent to more than 10 times the heat absorbed when its liquid temperature rises by 50℃. Therefore, the introduction of phase change improves the heat dissipation capacity of the cold plate by 3-5 times, making it suitable for high power density heat sources (such as a 2000WTDP GPU). (II) Parallel Independent Regulation Principle: The parallel setup of multiple cold plates is based on the concept of "flow distribution on demand": Since the power of the heat source corresponding to each cold plate may be different (e.g., CPU power 200W, GPU power 500W), the refrigerant flow rate entering each cold plate needs to be independently adjusted through the terminal flow regulating valve (connected in series with the cold plate inlet branch pipe). For cold plates with high-power heat sources, the valve opening is increased (e.g., 80%), and the refrigerant flow rate is increased (e.g., 2L / min) to ensure that the refrigerant has enough heat absorption; for cold plates with low-power heat sources, the valve opening is decreased (e.g., 40%), and the flow rate is reduced (e.g., 1L / min) to avoid refrigerant waste. The "flow splitting characteristic" of the parallel pipeline (the pressure of each branch is approximately equal) ensures that the flow regulation of a single cold plate will not interfere with other branches, achieving precise "one plate, one adjustment" adaptation. (III) Flow channel design and the principle of two-phase flow stability: The structural design of the internal flow channels of a cold plate directly affects the stability of the two-phase flow: serpentine or bifurcated flow channels can prevent the refrigerant from "short-circuiting" within the plate (i.e., the refrigerant flows out of the cold plate before fully absorbing heat), while the gradual change in the flow channel cross-section (e.g., a slightly wider inlet section and a slightly narrower heat source section) controls the refrigerant flow velocity (usually maintained at 0.5-1.5 m / s). Too low a flow velocity will cause gaseous refrigerant to accumulate within the plate (forming "gas blockage," hindering the flow of liquid refrigerant); too high a flow velocity will intensify the friction between the gas and liquid phases, leading to pressure fluctuations. A reasonable flow channel design can improve the dryness of the cold plate outlet (…). The dryness is controlled at 0.3-0.7 (i.e., the gas phase ratio is 30%-70%), which avoids "overheating" caused by excessive dryness (the heat absorption capacity decreases after the refrigerant is fully vaporized) and "insufficient heat exchange" caused by excessive dryness, thus providing a basis for the two-phase flow stability optimization algorithm (the central controller corrects the flow rate setpoint based on the dryness of the cold plate outlet). IV. Work Process: (a) Initialization phase: When the system starts, the central controller controls the terminal flow regulating valve to open fully. After the refrigerant pump starts, the refrigerant flows into each cold plate through the main liquid supply line and branch lines. At this time, the air in the cold plate is discharged through the outlet line (the air needs to be discharged in advance when the system is run for the first time) until the cold plate is full of refrigerant (this can be judged by the cold plate outlet temperature sensor: when the temperature tends to be stable, it is considered that the filling is complete). The central controller reads the initial values ​​of the temperature sensors at the outlets of each cold plate (which should be close to the ambient temperature, such as 25°C). If the outlet temperature of a cold plate is abnormal (e.g., much higher than other cold plates, possibly due to pipe blockage), it will issue a "cold plate passage abnormal" warning. (II) Normal operation phase: Heat absorption: After the server starts up, the CPU / GPU begins to generate heat. The heat is transferred to the cold plate substrate through the thermal grease, and the substrate temperature rises (e.g., from 25°C to 40°C). At this time, the refrigerant continuously flows into the cold plate channel under the drive of the pump. After contacting the substrate, it absorbs heat. Under low load (e.g., CPU load 30%), the refrigerant temperature rises but does not reach the boiling point (only sensible heat is absorbed), and it flows out of the cold plate in a liquid state. Under high load (e.g., GPU load 100%), the refrigerant absorbs heat and reaches the boiling point. Part of the liquid state changes to a gaseous state (phase change heat absorption), and it flows out of the cold plate in a gas-liquid two-phase state. Dynamic flow adaptation: The cold plate outlet temperature sensor monitors the outflow temperature in real time (e.g., the outlet temperature of the cold plate corresponding to the GPU rises to 55℃, which is higher than the preset threshold of 45℃). After the data is transmitted to the central controller, the central controller determines that the heat dissipation demand of the heat source corresponding to the cold plate has increased, and controls the opening of the end flow regulating valve at its inlet to increase (e.g., from 50% to 70%). This increases the refrigerant flow rate into the cold plate (e.g., from 1.2L / min to 1.8L / min), enhances the heat absorption capacity of the refrigerant, and the outlet temperature gradually drops back to within the threshold (e.g., 42℃). (III) Abnormal Response Phase: Localized overheating response: If a temperature sensor at a cold plate outlet detects a sudden temperature rise (e.g., from 45°C to 60°C within 10 seconds, possibly due to a sudden increase in the corresponding GPU load), the central controller will not only increase the flow rate to that cold plate but also correct the flow rate setpoint using a two-phase flow stability optimization algorithm. For example, if the dryness of the cold plate outlet ( The growth was too rapid. If the flow rate is increased, the flow rate can be appropriately increased to suppress the flow instability caused by excessive expansion of the gas phase region; Fault isolation response: If the cold plate outlet pressure sensor detects a sudden drop in pressure (e.g., from 0.2MPa to 0.05MPa, possibly due to a leak at the cold plate interface), the central controller immediately closes the end flow regulating valve at the inlet of that cold plate (cutting off the refrigerant supply), and simultaneously isolates the branch pipeline through the on / off valve of the control valve group to prevent a large amount of refrigerant leakage; other cold plates continue to operate normally without affecting overall heat dissipation; (iv) Shutdown phase: After the server shuts down, the heat source stops generating heat, and the central controller controls the refrigerant pump to maintain the lowest speed. The refrigerant continues to circulate in the cold plate for 1-2 minutes (to remove residual heat) until the cold plate outlet temperature drops to near the ambient temperature (e.g., 30°C). Then the pump stops running, and the terminal flow regulating valve resets to its initial opening (e.g., 50%), completing the shutdown process. V. Application Value: (a) Achieving precise heat dissipation from the heat source: Multiple cold plates are directly attached to the core heat sources such as CPU / GPU, resulting in a short heat transfer path (thermal resistance ≤0.1℃ / W). Compared with air cooling (thermal resistance is usually ≥0.5℃ / W), the temperature of the heat source can be reduced by 15-20℃. For example, a 2000W GPU may reach 90℃ under air cooling, while the cold plate heat dissipation can control it below 70℃, avoiding chip throttling or damage caused by high temperature and ensuring stable server operation. (ii) Adapting to diverse needs from multiple heat sources: Because of the parallel independent adjustment design, multiple sets of cold plates can simultaneously meet the heat dissipation needs of different heat sources in the same server: for example, allocating 1L / min of refrigerant flow to a 200W CPU and 2.5L / min of flow to a 500W GPU. Through the individual control of the end flow regulating valve, the problem of "insufficient heat dissipation of large heat sources and waste of flow to small heat sources" is avoided, which is suitable for the "multi-chip heterogeneous" hardware architecture of data center servers. (III) Supporting efficient heat dissipation for two-phase flow: The flow channel design within the cold plate is adapted to phase change heat absorption, allowing the heat dissipation potential of the anhydrous fluorinated liquid to be fully utilized: at the same flow rate, the heat absorption capacity of the refrigerant in the two-phase flow state is 3-5 times that of the pure liquid state. Therefore, it is not necessary to significantly increase the flow rate to cope with high-power heat sources. Compared with the pure liquid cold plate, the refrigerant circulation volume can be reduced by more than 30%, reducing the power consumption of the pump (pump power consumption is positively correlated with flow rate), and indirectly improving the system energy efficiency. (iv) Improve system operational reliability: Contact cooling between the cold plate and the heat source eliminates the need for moving parts such as fans (avoiding dust accumulation and noise issues associated with air cooling), and the waterless fluorinated liquid is non-conductive—even if a small amount of leakage occurs in the cold plate, it will not cause a short circuit in the server, making it more suitable for the high security requirements of data centers compared to water-cooled cold plates; at the same time, a single cold plate failure can be quickly isolated by a valve without affecting the heat dissipation of other heat sources, further improving system redundancy.

[0022] In this embodiment, the sensor group includes a liquid supply temperature sensor, a liquid return temperature sensor, a cold plate outlet temperature sensor, a cold plate outlet pressure sensor, a flow sensor, and a pump inlet and outlet differential pressure sensor. Furthermore, the sensor array serves as the "sensory nerve center" of the pump-driven two-phase multi-stage liquid cooling control system. By collecting core parameters such as temperature, pressure, and flow rate at various key nodes of the system in real time, it provides accurate and real-time data support for the dynamic adjustment, fault diagnosis, and optimized control of the central controller. It is the foundation for the system to achieve a closed loop of "sensing-decision-execution." The following section elaborates on the sensor array from an overall perspective to specific details: I. Overall Function Overview: The core function of the sensor array is to achieve "real-time sensing and data feedback" of the entire system's status: by deploying various sensors at key nodes in the refrigerant circulation loop (such as the liquid supply line, liquid return line, cold plate outlet, pump inlet and outlet, etc.), it continuously monitors parameters such as refrigerant temperature, pressure, flow rate, and pump operating pressure differential; the collected raw data is preprocessed and transmitted to the central controller to provide input for core algorithms such as adaptive fuzzy PID control, two-phase flow stability optimization, and fault diagnosis; at the same time, by judging abnormal parameter thresholds, it triggers early warning signals in real time to ensure that the system can be promptly detected and dealt with when parameters deviate from the normal range; II. Submodule Composition and Functions: The sensor group consists of various types of sensors, which can be divided into temperature, pressure and flow sensors according to the monitored objects. Each sensor is deployed in key locations of the system according to functional requirements to form a fully covered sensing network. (a) Temperature sensors: Liquid supply temperature sensor: Installation location: On the pipeline between the secondary side outlet of the plate heat exchanger and the main liquid supply line of the cold plate (i.e., the main pipeline section before the refrigerant enters the cold plate). Technical specifications: Employs a Pt1000 platinum resistance sensor; measurement range -20℃ to 100℃; accuracy ±0.1℃; response time ≤1s. Function: Real-time monitoring of the refrigerant temperature (supply liquid temperature) before it enters the cold plate, reflecting the heat exchange effect of the plate heat exchanger. If the supply liquid temperature is too high (e.g., exceeding 35°C), it indicates that the heat exchanger is not dissipating heat sufficiently, and the primary side cooling medium flow rate needs to be adjusted accordingly. At the same time, this temperature is the basic parameter for calculating the heat absorption temperature difference of the cold plate (the difference between the supply liquid temperature and the cold plate outlet temperature). Return liquid temperature sensor: Installation location: On the pipeline between the main return line of the cold plate and the inlet of the liquid storage tank (i.e., the main pipeline section after the refrigerant leaves the cold plate). Technical specifications: Same as the liquid supply temperature sensor, measuring range -20℃-100℃, accuracy ±0.1℃; Function: Monitors the overall temperature of the refrigerant after absorbing heat from the heat source (return liquid temperature), reflecting the overall heat dissipation load of the system—the difference between the return liquid temperature and the supply liquid temperature (i.e., the total temperature difference of the system). Combined with flow data, it can calculate the real-time heat dissipation of the system (heat dissipation = flow rate × refrigerant specific heat capacity × temperature difference), providing a macroscopic basis for pump speed adjustment. Cold plate outlet temperature sensor: Installation location: On the outlet branch pipe of each group of cold plates (adjacent to the outlet interface of the cold plate); Technical specifications: Employs NTC thermistors (miniaturized design, adaptable to branch pipeline space), measurement range 0℃-80℃, accuracy ±0.2℃, response time ≤0.5s; Function: Individually monitor the refrigerant temperature flowing out of each group of cold plates, directly reflecting the heat dissipation effect of the corresponding heat source. For example, if the outlet temperature of a GPU cold plate is higher than the preset threshold (such as 50°C), it indicates that the heat source is not dissipating heat properly, and the central controller will increase the flow rate of the cold plate accordingly. At the same time, this temperature is also the core indicator for judging whether the cold plate is experiencing "local overheating". (ii) Pressure sensors: Cold plate outlet pressure sensor: Installation location: downstream of the temperature sensor on each cold plate outlet branch pipe (to form a "temperature and pressure integrated monitoring point" in conjunction with the temperature sensor); Technical specifications: Employs a miniature diffused silicon pressure sensor, measuring range 0-1MPa, accuracy ±0.5%FS (full scale), output 4-20mA analog signal; Function: Monitors the pressure of refrigerant flowing out of the cold plate to help determine the internal condition of the cold plate. If the outlet pressure of a cold plate drops suddenly (e.g., below 0.1MPa), it may be due to leakage at the cold plate interface or blockage of the flow channel. If the pressure continues to rise, it may be due to sticking of the terminal flow regulating valve, providing data support for fault location. Pump inlet and outlet differential pressure sensors: Installation location: Connect the inlet manifold and outlet manifold of the refrigerant pump respectively through two pressure taps (bridging the pump body inlet and outlet). Technical specifications: Differential pressure sensor, measuring range 0-0.5MPa, accuracy ±0.2%FS, response time ≤10ms; Function: Real-time monitoring of the pressure difference between the pump inlet and outlet (i.e., the pump head) to reflect the pump's operating status. During normal operation, the pressure difference is positively correlated with the pump speed (the higher the speed, the greater the pressure difference). If the speed remains unchanged but the pressure difference drops suddenly, it may be due to pump cavitation or internal wear. If the pressure difference exceeds the safety threshold, it will trigger the bypass valve to open, protecting the pump from overload damage. (iii) Flow sensor: Installation location: On the main cold plate liquid supply line (the main pipeline after all branch pipelines are combined), located upstream of the end flow regulating valve; Technical specifications: Employs an electromagnetic flowmeter (compatible with conductive or non-conductive refrigerants), measuring range 0-100L / min, accuracy ±0.5%FS, and supports bidirectional measurement (to prevent misjudgment due to refrigerant backflow). Function: Monitors the total refrigerant flow into all cold plates, which is the core basis for the flow control mode. When the total flow is lower than the bypass valve opening threshold, the bypass valve is triggered to adjust. At the same time, the total flow, combined with the return liquid temperature and the supply liquid temperature, can calculate the total heat dissipation of the system, providing a basis for decision-making for multi-pump coordinated operation (such as starting the standby pump under high load). III. Key Technology Principles: (a) The principle of data acquisition synchronization: The sensor group adopts a "timestamp synchronous acquisition" mechanism: the central controller triggers all sensors to sample simultaneously through a unified clock signal (1ms accuracy), ensuring that parameters such as supply temperature, return temperature, flow rate, and pressure at the same moment can be correlated and analyzed; for example, if the cold plate outlet temperature rises at a certain moment, and the flow sensor shows a decrease in total flow, it can be quickly determined that the heat dissipation problem is caused by insufficient flow, rather than a sudden increase in heat source power, thus avoiding erroneous adjustment by the central controller; (II) High-precision measurement and anti-interference principle: To address the electromagnetic environment of the liquid cooling system (where server rooms experience high-frequency electromagnetic interference), the sensor array employs multiple anti-interference designs: Hardware-wise: The temperature sensor uses a three-wire connection (to eliminate wire resistance error), and the output signals of the pressure and flow sensors are transmitted through shielded cables (the shield is grounded at one end). At the software level: the raw data is filtered by a sliding window (window size of 5-10 sampling points) to remove instantaneous pulse interference, and then cross-validated by temperature and pressure (if there is a corresponding relationship between refrigerant saturation temperature and pressure, the data is marked as abnormal if the deviation exceeds the threshold) to ensure that the data output to the central controller is accurate and reliable. (III) Principles of Parameter Correlation Analysis: The parameters collected by the sensor array are not isolated, but are interconnected through physical laws to form a "parameter chain": for example, the relationship between the cold plate outlet temperature and the flow rate follows the principle that "the higher the flow rate, the lower the outlet temperature" (when the heat source power is stable); the relationship between the pump inlet and outlet pressure difference and the flow rate follows the principle that "the pressure difference increases with the square of the flow rate" (pipeline resistance characteristics); the central controller can use these correlations to verify abnormal data from individual sensors—if the outlet temperature of a cold plate shows an abnormally high increase, but the corresponding flow sensor shows a normal flow rate and the temperatures of other cold plates are stable, it may be a fault in that temperature sensor rather than an actual heat dissipation problem, thus improving the system's fault tolerance; IV. Work Process: (a) Initialization phase: After the system starts up, the sensor group performs a self-test along with the central controller: each sensor outputs an initial signal (such as the temperature sensor outputting the ambient temperature and the pressure sensor outputting atmospheric pressure). The central controller checks whether the signal is within the normal range (such as whether the initial value of the temperature sensor is within -20℃ to 100℃). If a sensor has no signal or the signal is out of range, it is marked as a "faulty sensor" and an alarm is triggered. After completing the self-test, the sensor group enters the preheating state (lasts for 30 seconds). After the temperature sensor resistance stabilizes and the zero-point drift of the pressure sensor is eliminated, it enters the normal acquisition mode. (II) Real-time data acquisition phase: The central controller sends acquisition commands to all sensors according to a preset sampling period (usually 100ms / time), and each sensor synchronously acquires the current parameters: the temperature sensor converts the temperature signal into a resistance value (Pt1000) or a voltage value (NTC), and the pressure and flow sensors convert the physical quantity into a 4-20mA or 0-10V analog signal; The sensor transmits the raw signal to the signal conditioning module of the central controller, which converts it into a digital quantity through A / D conversion (16-bit precision). After filtering and verification, the digital quantity is stored in the real-time database for the control algorithm to call. (III) Abnormal Early Warning Stage: The central controller compares sensor data with preset thresholds in real time (such as the liquid supply temperature threshold of 30℃±5℃, and the pump inlet / outlet pressure difference threshold of 0.3MPa±0.1MPa). If the data is within the threshold range, it continues to be collected normally; if the data exceeds the threshold (e.g., the cold plate outlet temperature reaches 55℃, exceeding the upper limit of 50℃), a first-level warning is triggered, and the central controller starts the adjustment strategy (e.g., increasing the corresponding cold plate flow). If the data exceeds the safety limit (e.g., the pressure difference between the pump inlet and outlet reaches 0.6MPa, exceeding the limit of 0.5MPa), a level two warning is triggered, and protective measures are immediately implemented (e.g., opening the bypass valve and reducing the pump speed). For abnormal data collected three times consecutively (excluding transient interference), mark them as "fault parameters" and use multi-sensor correlation analysis to locate the fault point (e.g., "high outlet temperature of a cold plate + normal flow rate" → poor heat exchange of the cold plate; "normal temperature + sudden drop in pressure" → pipeline leakage). (iv) Shutdown phase: When the system shuts down, the sensor group continues to collect data for 30 seconds, recording the final state before shutdown (such as the outlet temperature of each cold plate, total flow rate, and pump pressure difference) as a system operation log archive; then the data acquisition process is completed when the central controller is powered off. V. Application Value: (a) Providing a data foundation for precise control: The real-time parameters collected by the sensor array serve as the "input source" for the central controller's adjustment strategy: for example, the adaptive fuzzy PID algorithm needs to be based on the deviation between the supply temperature and the target value ( ) Calculate pump speed; the two-phase flow stability optimization algorithm needs to be based on the dryness of the cold plate outlet ( Correcting the flow rate setpoint—without high-precision sensor data, the control algorithm will lose its basis and will be unable to achieve "heat dissipation on demand"; (ii) Achieving early fault detection and early fault location: Through real-time monitoring and correlation analysis of multiple parameters, the sensor group can issue early warnings before the fault escalates: for example, a slow drop in the outlet pressure of the cold plate may indicate a loose interface, and abnormal fluctuations in the pressure difference between the pump inlet and outlet may indicate pump wear. These early signals can help maintenance personnel intervene in advance and reduce downtime by more than 60%. (III) Supporting system energy efficiency optimization: Based on the flow rate and temperature difference data collected by sensors, the system can calculate the real-time heat dissipation and energy consumption (such as pump power consumption), and then optimize the operation strategy: for example, at low load (small heat dissipation), by reducing the pump speed and flow rate, the energy consumption per unit heat dissipation is reduced by 20%-30%, realizing "heat dissipation on demand and dynamic optimization of energy consumption". (iv) Improve system redundancy and reliability: The sensor group adopts a "dual backup" design for key parameters (such as temperature sensors installed in the main liquid supply line and important cold plate branches). When a sensor fails, a replacement value can be calculated from the data of other related sensors (such as calculating the average liquid supply temperature using the return liquid temperature and the total flow rate), ensuring uninterrupted system control and improving overall redundancy.

[0023] In this embodiment, the central controller is electrically connected to the liquid cooling module and the sensor group, and is configured to execute: Based on the difference between sensor data and a preset threshold, the refrigerant pump speed is dynamically adjusted using an adaptive fuzzy PID algorithm. Control the on / off state and opening degree of the control valve assembly; Each refrigerant pump is started and stopped in rotation according to a preset cycle to achieve balanced wear. The refrigerant flow setpoint is corrected in real time using a two-phase flow stability optimization algorithm; The refrigerant pump speed regulation adopts an adaptive fuzzy PID algorithm, and its output speed satisfies: ,in, for The refrigerant pump output speed command at any given time; This represents the real-time parameter deviation value. For deviation values ​​from 0 to Integral at time step; This is the derivative of the deviation value with respect to time; The proportional coefficient is dynamically adjusted using fuzzy rules. The integral coefficients are dynamically adjusted using fuzzy rules. Let be the differential coefficients that are dynamically adjusted through fuzzy rules, and satisfy: ,in, The PID coefficients are dynamically adjusted. They represent the proportion p respectively ,integral ,differential ; These are the initial coefficients for the PID controller; This is the correction amount for the output of fuzzy inference; Generate by following these steps: Define fuzzy input variables: The fuzzy set is {negative large, negative medium, negative small, zero, positive small, positive medium, positive large}; Deviation change rate The fuzzy set is the same as before; Construct a fuzzy rule base, with the rule format as follows: , , ; Defuzzy resolution using the centroid method: , ,in, This is the correction amount for the fuzzy inference output. They represent the proportion p respectively ,integral ,differential ; For the first The trigger strength of the rule; For the first , Each rule outputs a sharpening value for a fuzzy subset; the numerator is the sum of the products of the trigger strength of each rule and its corresponding sharpening value; the denominator is the sum of the trigger strengths of each rule. The two-phase flow stability optimization algorithm is as follows: ,in, The corrected refrigerant flow rate setpoint; Set the value for the raw flow rate; This is the stability gain coefficient; This refers to the pressure drop in the vapor phase region within the cold plate. The density of the liquid refrigerant; The latent heat of vaporization of the refrigerant; Dryness of cold-rolled plate outlet; This represents the derivative of the cold plate outlet dryness with respect to time. Furthermore, the central controller is the "nerve center and decision-making core" of the pump-driven two-phase multi-phase liquid cooling control system. Through real-time interaction with the liquid cooling module and sensor group, it integrates collected data, executes core algorithms, and outputs control commands to achieve dynamic regulation and stable operation management of the entire system. It is a key carrier connecting the entire "sensing-analysis-execution" link. The following section elaborates on the central controller from overall structure to details: I. Overall Function Overview: The core function of the central controller is to achieve "intelligent decision-making and precise control" of the system: on the one hand, through electrical connection with the sensor group, it receives data such as supply liquid temperature, return liquid temperature, cold plate outlet parameters (temperature, pressure, dryness), flow rate and pump differential pressure in real time, and uses the pre-processed data as input for the control algorithm; on the other hand, based on this data, it dynamically calculates control quantities such as refrigerant pump speed and control valve group opening through built-in adaptive fuzzy PID algorithm and two-phase flow stability optimization algorithm, and then sends instructions to the liquid cooling module (pump, valve, etc.) to complete operations such as flow regulation, pump rotation, and fault handling; at the same time, it is also responsible for recording system operation data, executing pump wear equalization strategy, and ultimately ensuring dynamic adaptation of refrigerant circulation and heat dissipation demand, and maintaining stable heat source temperature; II. Submodule Composition and Functions: The central controller is composed of a hardware core unit and a software functional unit working together. The hardware provides the operating foundation, and the software implements the algorithm and logic control. Each sub-module has a clear division of labor and is closely linked. (a) Data Receiving and Preprocessing Unit: Hardware foundation: Equipped with a 16-bit A / D conversion module (sampling rate 1kHz) and multiple signal interfaces (supporting 0-10V analog signals, 4-20mA current signals and digital signals), which are respectively connected to the status feedback terminals of various sensors in the sensor group and the liquid cooling module; Core functions: Real-time data reception: Synchronously receive sensor group data (such as voltage signal from liquid supply temperature sensor and current signal from flow sensor) and liquid cooling module status signals (such as pump operation feedback and valve opening feedback) at a cycle of 100ms / time. Preprocessing operations: Filtering the raw data (using a 5-point moving average filter to remove instantaneous noise), scaling conversion (converting voltage / current signals into physical quantities, such as converting a 0-10V signal into a temperature value of -20℃-100℃), and validity verification (if a cold plate outlet temperature data exceeds the reasonable range of 0℃-80℃, it is marked as "invalid data" and replaced with the average of the first 3 values). Data caching: Preprocessed data is temporarily stored in a 2MB cache for real-time access by the control algorithm unit, and simultaneously written to local storage (16GB capacity) at a frequency of 1 second / time to retain historical data; (ii) Control Algorithm Unit: As the "computing core" of the central controller, it incorporates an adaptive fuzzy PID algorithm and a two-phase flow stability optimization algorithm. It achieves efficient computation through a 32-bit MCU (180MHz main frequency) and is the core carrier of control decision-making. Adaptive Fuzzy PID Algorithm Module: Input: Real-time parameter deviation values ​​collected by the sensor array (e.g., the difference between the cold plate outlet temperature and the target temperature) and the rate of change of deviation ( Differential with respect to time); Calculation process: Dynamically adjusting PID coefficients based on fuzzy rules: Definition and The fuzzy set (negative large, negative medium, negative small, zero, positive small, positive medium, positive large) is used through a rule base (such as "if..."). For the upright and If it is positive and small, then The coefficient correction amount is derived from the fact that the value is positive and small. Press again Obtain dynamic coefficients ; Pump speed calculation command: Substitute into the formula Output ( (Pump speed command at all times). Output: Speed ​​control signal for refrigerant pump (0-10V analog signal, corresponding to 0%-100% speed); Two-phase flow stability optimization algorithm module: Input: Raw flow rate setpoint collected by the sensor array Pressure drop in the gas phase region inside the cold plate Liquid refrigerant density Latent heat of vaporization of refrigerant and the dryness of the cold plate outlet (Calculated using temperature-pressure correlation); Calculation process: According to the formula Calculate the corrected flow rate setpoint (in, (This is the stability gain coefficient, with a default value of 0.8); Output: The corrected flow rate setpoint, used as the target reference value for the adaptive fuzzy PID algorithm (e.g., based on...). (Adjusting the target deviation of the pump speed); (iii) Execution control unit: Hardware foundation: Equipped with a multi-channel D / A conversion module (12-bit output accuracy) and a relay output interface, which are respectively connected to the refrigerant pump drive module, the actuator of the control valve group, and the control terminal of the end flow regulating valve; Core functions: Pump control: The speed command output by the adaptive fuzzy PID algorithm ( The signal is converted into a 0-10V analog signal and sent to the frequency converter drive module of the refrigerant pump to adjust the pump speed. At the same time, a cycle command is generated according to a preset cycle (e.g., 8 hours / time), and the start and stop switching of the pump is controlled by a relay (e.g., after the current pump A has been running for 8 hours, pump B is started first, and pump A is turned off after its speed stabilizes). Valve control: Receives flow regulation requests from the algorithm unit and outputs a 0-10V opening signal to the terminal flow regulating valve (0V corresponds to 0% opening, 10V corresponds to 100% opening); for bypass valves, outputs commands according to flow / differential pressure control mode (e.g., when the secondary flow rate is ≤5L / min, outputs a 3V signal corresponding to 30% opening). Boundary protection: Execute special control logic - when the pump speed required by the algorithm is ≤0%, force the output of the minimum speed signal (e.g., corresponding to 2V voltage, maintain 20% speed); when a pump fault alarm signal is received (e.g., pump current over-limit feedback), immediately output a shutdown command to cut off the power supply to the corresponding pump; (iv) Fault Diagnosis and Recording Unit: Fault monitoring: Real-time comparison of sensor data with preset safety thresholds (e.g., pump inlet and outlet pressure difference ≥ 0.5MPa is considered over-limit, cold plate outlet pressure ≤ 0.1MPa is considered abnormal), and simultaneously receiving fault feedback signals from the liquid cooling module (e.g., pump stall alarm, valve jamming feedback). Fault Handling: If a fault is detected, the corresponding strategy is triggered immediately—such as starting the standby pump when the pump fails, or closing the end flow regulating valve of the corresponding branch when the cold plate outlet pressure is abnormal; at the same time, a fault code is generated (such as "E01" representing a pump A fault, and "E02" representing a bypass valve jam). Data logging: Write the time of failure, type of failure, and sensor data at the time of failure (such as flow rate and temperature at the time of failure) to local storage, and generate daily operation logs (including the cumulative pump runtime, number of valve adjustments, frequency of failures, etc.). III. Key Technology Principles: (I) Closed-loop control principle: The core control logic of the central controller is based on "closed-loop feedback": real-time parameters collected by the sensor group (such as the cold plate outlet temperature) are used as "feedback quantities" and compared with preset target values ​​(such as 45℃) to obtain the deviation. Algorithm unit based on The control quantity is calculated (such as increasing the pump speed). After the control unit adjusts the actuator (pump), the feedback quantity changes accordingly (temperature decreases) until the deviation is reduced to the allowable range (such as ±2℃). For example, when the GPU load suddenly increases and the cold plate outlet temperature rises to 50℃ (deviation +5℃), the central controller increases the flow rate by increasing the pump speed, so that the temperature drops back to 45℃, forming a closed loop of "deviation-adjustment-deviation elimination". (II) Principle of Dual Algorithm Collaboration: The adaptive fuzzy PID algorithm and the two-phase flow stability optimization algorithm do not operate independently, but rather have a collaborative relationship of "feedforward" + "feedback": Two-phase flow stability optimization algorithm based on cold plate outlet dryness rate of change ( Correct the flow rate setting in advance. (Feedforward regulation) avoids parameter fluctuations caused by two-phase flow oscillations (such as preventing local overheating caused by excessive dryness); the adaptive fuzzy PID algorithm is based on the corrected... The pump speed is dynamically adjusted (feedback regulation) to precisely match the corrected flow demand based on the deviation from the actual flow rate. The combination of these two methods not only suppresses potential system fluctuations but also improves regulation accuracy—temperature fluctuations can be reduced by 40% compared to single PID control. (III) Multi-task scheduling principle: The central controller needs to handle multiple tasks simultaneously, including data acquisition, algorithm calculation, control output, and fault monitoring. It achieves efficient operation through a "priority scheduling mechanism": core tasks (such as data reception and algorithm calculation) are set to the highest priority, occupying 60% of the MCU's computing resources and being forcibly executed at 100ms intervals; secondary tasks (such as log generation and polling timers) are set to low priority and executed when core tasks are idle (such as during algorithm calculation intervals); fault handling tasks are set to emergency priority, immediately interrupting the current task (such as pausing polling timers) upon triggering, and prioritizing fault response (such as starting a backup pump). This scheduling mechanism ensures that the delay of critical operations (such as real-time adjustment and fault response) is ≤10ms, avoiding control lag caused by task congestion.

[0024] In this embodiment, the control valve group includes a bypass valve, and its control logic is as follows: Flow control mode: When the secondary flow rate is less than or equal to the bypass valve opening threshold, the bypass valve is opened. When the flow rate is greater than or equal to the bypass valve closing threshold, the bypass valve is closed. The opening degree is linearly adjusted between the opening threshold and the closing threshold. Differential pressure control mode: When the pressure difference between the pump inlet and outlet is greater than or equal to the pressure difference opening threshold, the bypass valve is opened; When the differential pressure is less than or equal to the differential pressure shut-off threshold, the bypass valve is closed.

[0025] In this embodiment, the opening degree of the bypass valve is controlled by a 0-10V analog signal, where 0V corresponds to 0% opening and 10V corresponds to 100% opening.

[0026] The central controller also performs: When the refrigerant pump requires a speed of ≤0%, maintain the lowest speed operation. Real-time monitoring of fault signals; immediate shutdown of the corresponding pump body in response to refrigerant pump fault alarm.

[0027] In this embodiment, anhydrous fluorinated liquid is used as the refrigerant in the refrigerant circulation loop to achieve gas-liquid phase change heat absorption.

[0028] A pump-driven two-phase multi-phase liquid cooling control method is applied to the system. The control method includes the following steps: Step S1: Real-time data collection of supply liquid temperature, return liquid temperature, cold plate outlet dryness, flow rate, and pump differential pressure using a sensor array; Step S2: The central controller executes the two-phase flow stability optimization algorithm and outputs the corrected flow setpoint. Step S3: Calculate the required speed of the refrigerant pump using an adaptive fuzzy PID algorithm; Step S4: Switch the start and stop of the refrigerant pump according to the preset cycle, and dynamically adjust the opening of the bypass valve; Step S5: Linearly adjust the bypass valve opening based on the flow rate or differential pressure threshold.

[0029] In this embodiment, the proportional coefficient of the adaptive fuzzy PID algorithm in step S3 The dynamic adjustment rules include: when and hour, ,in, This represents the absolute value of the real-time parameter deviation. The absolute value of the rate of change of deviation; This is the dynamically adjusted proportional coefficient; This is the initial proportional coefficient; when and hour, ,in, This represents the absolute value of the real-time parameter deviation. This represents the absolute value of the rate of change of the deviation. This is the dynamically adjusted proportional coefficient; This is the initial proportional coefficient.

[0030] Although embodiments of the invention have been shown and described, it will be understood by those skilled in the art that various changes, modifications, substitutions and alterations can be made to these embodiments without departing from the principles and spirit of the invention, the scope of which is defined by the appended claims and their equivalents.

Claims

1. A pump-driven two-phase multi-phase liquid-cooled control system, characterized in that: include: The liquid cooling module consists of a plate heat exchanger, a liquid storage tank, at least two refrigerant pumps connected in parallel, a control valve group, and a terminal flow regulating valve, forming a closed refrigerant circulation loop. Multiple cold plates are connected in parallel between the plate heat exchanger and the refrigerant pump, and deployed at the server CPU / GPU heat source location; The sensor group includes a liquid supply temperature sensor, a liquid return temperature sensor, a cold plate outlet temperature sensor, a cold plate outlet pressure sensor, a flow sensor, and a pump inlet and outlet differential pressure sensor. The central controller, electrically connected to the liquid cooling module and sensor group, is configured to execute: Based on the difference between sensor data and a preset threshold, the refrigerant pump speed is dynamically adjusted using an adaptive fuzzy PID algorithm. Control the on / off state and opening degree of the control valve assembly; Each refrigerant pump is started and stopped in rotation according to a preset cycle to achieve balanced wear. The refrigerant flow rate setpoint is corrected in real time using a two-phase flow stability optimization algorithm.

2. The pump-driven two-phase multi-phase liquid-cooled control system according to claim 1, characterized in that: The speed regulation of the refrigerant pump adopts an adaptive fuzzy PID algorithm, and its output speed satisfies: ,in, for The refrigerant pump output speed command at any given time; This represents the real-time parameter deviation value. For deviation values ​​from 0 to Integral at time step; This is the derivative of the deviation value with respect to time; The proportional coefficient is dynamically adjusted using fuzzy rules. The integral coefficients are dynamically adjusted using fuzzy rules. Let be the differential coefficients that are dynamically adjusted through fuzzy rules, and satisfy: ,in, The PID coefficients are dynamically adjusted. They represent the proportion p respectively ,integral ,differential ; These are the initial coefficients for the PID controller; This is the correction amount for the output of fuzzy inference.

3. The pump-driven two-phase multi-phase liquid-cooled control system according to claim 2, characterized in that: The Generate through the following steps: Define fuzzy input variables: deviation The fuzzy set is {negative large, negative medium, negative small, zero, positive small, positive medium, positive large}; Deviation change rate The fuzzy set is the same as before; Construct a fuzzy rule base, with the rule format as follows: , , ; Defuzzy resolution using the centroid method: , ,in, This is the correction amount for the fuzzy inference output. They represent the proportion p respectively ,integral ,differential ; For the first The trigger strength of the rule; For the first , Each rule outputs a sharpening value for a fuzzy subset; the numerator is the sum of the products of the trigger strength of each rule and its corresponding sharpening value; the denominator is the sum of the trigger strengths of each rule.

4. The pump-driven two-phase multi-phase liquid-cooled control system according to claim 1, characterized in that: The two-phase flow stability optimization algorithm is as follows: ,in, The corrected refrigerant flow rate setpoint; Set the value for the raw flow rate; This is the stability gain coefficient; This refers to the pressure drop in the vapor phase region within the cold plate. The density of the liquid refrigerant; The latent heat of vaporization of the refrigerant; Dryness of cold-rolled plate outlet; This is the derivative of the dryness of the cold plate outlet with respect to time.

5. The pump-driven two-phase multi-phase liquid-cooled control system according to claim 1, characterized in that: The control valve assembly includes a bypass valve, and its control logic is as follows: Flow control mode: When the secondary flow rate is less than or equal to the bypass valve opening threshold, the bypass valve is opened; When the flow rate is greater than or equal to the bypass valve closing threshold, the bypass valve is closed. The opening degree is linearly adjusted between the opening threshold and the closing threshold. Differential pressure control mode: When the pressure difference between the pump inlet and outlet is greater than or equal to the pressure difference opening threshold, the bypass valve is opened; When the differential pressure is less than or equal to the differential pressure shut-off threshold, the bypass valve is closed.

6. A pump-driven two-phase multi-phase liquid-cooled control system according to claim 5, characterized in that: The opening degree of the bypass valve is controlled by a 0-10V analog signal, where 0V corresponds to 0% opening and 10V corresponds to 100% opening.

7. The pump-driven two-phase multi-phase liquid-cooled control system according to claim 1, characterized in that: The central controller also performs: When the refrigerant pump requires a speed of ≤0%, maintain the lowest speed operation. Real-time monitoring of fault signals; immediate shutdown of the corresponding pump body in response to refrigerant pump fault alarm.

8. The pump-driven two-phase multi-phase liquid-cooled control system according to claim 1, characterized in that: The refrigerant circulation loop uses anhydrous fluorinated liquid as the refrigerant to achieve gas-liquid phase change heat absorption.

9. The pump-driven two-phase multi-phase liquid cooling control method according to claim 1, characterized in that: The control method is applied to the system according to any one of claims 1-8, and the control method includes the following steps: Step S1: Real-time data collection of supply liquid temperature, return liquid temperature, cold plate outlet dryness, flow rate, and pump differential pressure using a sensor array; Step S2: The central controller executes the two-phase flow stability optimization algorithm and outputs the corrected flow setpoint. Step S3: Calculate the required speed of the refrigerant pump using an adaptive fuzzy PID algorithm; Step S4: Switch the start and stop of the refrigerant pump according to the preset cycle, and dynamically adjust the opening of the bypass valve; Step S5: Linearly adjust the bypass valve opening based on the flow rate or differential pressure threshold.

10. A pump-driven two-phase multi-phase liquid cooling control method according to claim 9, characterized in that: The proportional coefficient of the adaptive fuzzy PID algorithm in step S3 The dynamic adjustment rules include: when and hour, ,in, This represents the absolute value of the real-time parameter deviation. The absolute value of the rate of change of deviation; This is the dynamically adjusted proportional coefficient; This is the initial proportional coefficient; when and hour, ,in, This represents the absolute value of the real-time parameter deviation. This represents the absolute value of the rate of change of the deviation. This is the dynamically adjusted proportional coefficient; This is the initial proportional coefficient.

Images