A multi-stage parallel liquid cooling path dynamic allocation system

Abstract

The application discloses a kind of multistage parallel liquid cooling path dynamic allocation system, and it is related to electronic equipment heat dissipation technical field.System includes liquid cooling execution network, temperature sensing network and controller.Liquid cooling execution network has multiple parallel cooling branches with independent electrically controlled valve, and different subareas of cooling object are corresponded respectively.Temperature sensing network collects temperature data.Controller executes following method: data is collected and temperature distribution map is generated;Calculate the average temperature of each subarea, and identify the first type of region of high temperature and the second type of region of low temperature;Cooling demand increment of the first type of region and releasable flow of the second type of region are calculated;With the constraint such as satisfying total flow conservation, and making predicted temperature overall optimal as target, the flow adjustment amount of each branch is solved;Finally, drive valve to execute adjustment.The application dynamically allocates parallel branch flow by software-defined manner, realizes the spatial distribution of cooling capacity according to heat load, solves the problem of uneven heat dissipation, and improves heat dissipation efficiency.

Description

Technical Field

[0001] This invention relates to the field of heat dissipation technology for electronic devices, and in particular to a liquid cooling system and its control method based on electro-digital data processing, specifically a multi-stage parallel liquid cooling path dynamic allocation system. Background Technology

[0002] As the performance of integrated circuits continues to improve, their power consumption and heat flux density are increasing, posing a severe challenge to heat dissipation technology. Liquid cooling, due to its high heat capacity and high heat conduction efficiency, has become the mainstream heat dissipation solution for high-performance computing devices.

[0003] Traditional liquid cooling systems often employ a fixed flow path design, meaning the coolant's flow direction and flow rate ratio are determined during the physical design phase. This design has inherent limitations when facing dynamically changing and spatially unevenly distributed heat loads within the processor chip: cooling resources cannot be dynamically adjusted according to the real-time heat demands of each area. This can easily lead to insufficient heat dissipation in high-load areas, creating localized hotspots and impacting performance and reliability; simultaneously, cooling capacity in low-load areas is wasted, reducing the overall energy efficiency of the cooling system.

[0004] While existing technologies attempt to respond to temperature changes by adjusting pump speed or overall flow rate, this is a coarse-grained, global adjustment that cannot achieve precise, targeted distribution of cooling capacity across different areas of the chip. Therefore, there is an urgent need for a novel liquid cooling system capable of real-time sensing of heat distribution and intelligent allocation of cooling resources to solve the problem of dynamically uneven heat dissipation. Summary of the Invention

[0005] In order to overcome the above-mentioned defects of the prior art, the embodiments of the present invention provide a multi-level parallel liquid cooling path dynamic allocation system to solve the problems mentioned in the background art. The technical problem to be solved by the present invention is that the existing fixed flow path liquid cooling heat dissipation system is difficult to dynamically allocate cooling resources according to the real-time changes and uneven heat load of each area of ​​the equipment, which can easily lead to local overheating or waste of cooling resources.

[0006] To address the aforementioned technical problems, this invention provides a multi-stage parallel liquid cooling path dynamic allocation system, the technical solution of which is as follows: The system includes a liquid-cooled execution network, a temperature sensing network, and a controller.

[0007] The liquid cooling execution network consists of multiple parallel independent cooling branches. Each independent cooling branch is equipped with an electrically controlled flow regulating valve and is used to establish fluid communication with a specific section of the device being cooled.

[0008] The temperature sensing network consists of multiple temperature sensors disposed on the cooled device for collecting temperature signals.

[0009] The controller is connected to the temperature sensing network and all the electrically controlled flow regulating valves. The controller is configured to execute the following control logic: First, the readings of each temperature sensor are collected, and a temperature distribution map of the cooled device is generated based on these readings and the sensor location information.

[0010] Secondly, the average temperature of each zone is calculated based on the temperature distribution map, and according to the ranking of the average temperature, at least one zone with the highest average temperature is identified as a first-class zone that needs enhanced cooling, and at least one zone with the lowest average temperature is identified as a second-class zone that can contribute cooling resources.

[0011] Next, a cooling demand flow rate is calculated for each first-class region, and a release flow rate is calculated for each second-class region.

[0012] Then, taking the allocation of release flow from the cooling branch corresponding to the second type of area to the cooling branch corresponding to the first type of area as the basic allocation direction, with the primary constraint of keeping the total flow of the system unchanged, and with the goal of finding the optimal solution that makes the predicted temperature of each zone generally consistent, the precise flow adjustment amount required for each independent cooling branch is calculated.

[0013] Finally, control commands are generated based on the calculated flow adjustment amount, and the corresponding electronically controlled flow regulating valve is driven to perform the action, thereby realizing the dynamic and on-demand distribution of coolant.

[0014] As a further improvement of the present invention, the controller is configured to identify the partitions located in the first N1 positions of the average temperature sort as first-class regions and the partitions located in the last N2 positions as second-class regions, where N1 and N2 are preset positive integers.

[0015] As a further improvement of the present invention, the controller also follows the following additional constraints when calculating the flow adjustment amount of each branch: the flow adjustment amount of the branch flowing to the first type of area must be greater than or equal to zero; the flow adjustment amount of the branch flowing to the second type of area must be less than or equal to zero; and the absolute value of the flow adjustment amount of any branch must not exceed a preset maximum adjustment threshold.

[0016] As a further improvement of the present invention, the controller is also configured to execute a closed-loop iterative optimization process, that is, after the drive valve completes an adjustment, it waits for a preset delay time, and then collects temperature data again and evaluates the cooling effect of the first type of region. If the expected goal is not achieved, the entire process from calculating the requirements to driving the valve is re-executed starting from the current latest state, and the optimal heat dissipation state is approximated through iteration.

[0017] As a further improvement of the present invention, the controller executes a self-learning process when the system is initially started. By sequentially adjusting each valve to different opening degrees and recording the flow and temperature data under steady state, the controller automatically establishes the "opening degree-flow rate" characteristic curve of each valve and the "flow rate-temperature" response model of each zone.

[0018] As a further improvement of the present invention, the target temperature value used by the controller to calculate the cooling demand flow rate for the first type of area is a dynamic parameter that can be adaptively adjusted according to the real-time load rate of the cooled device.

[0019] As a further improvement of the present invention, the system also includes an independent safety monitoring module, which continuously monitors whether the actual operation of each valve is consistent with the command, and whether the pressure of each cooling branch is normal. Once a valve malfunction or abnormal pressure is detected, the module will immediately take over control and force all valves to switch to a preset safe opening state.

[0020] As a further improvement of the present invention, the cooled device is a semiconductor chip, and its partitions correspond to different functional module units on the chip, such as computing core clusters or cache areas.

[0021] As a further improvement of the present invention, the liquid cooling execution network specifically includes a microchannel cold plate, and the multiple independent cooling branches are respectively connected to the independent microchannel group inside the cold plate that is precisely aligned with each chip partition.

[0022] As a further improvement of the present invention, the controller uses linear programming or quadratic programming algorithms to calculate the optimal flow adjustment amount.

[0023] Compared with the prior art, the technical solution provided by the present invention has the following beneficial effects: 1. High heat dissipation precision: By constructing a temperature field in real time and intelligently identifying high and low temperature areas, limited cooling resources can be precisely directed to the hot spots that need them most, effectively suppressing local overheating and improving the temperature uniformity and operational stability of the chip under dynamic load.

[0024] 2. Excellent energy efficiency: By adopting a strategy of redistributing existing resources, the overall heat dissipation efficiency is improved through optimized allocation while keeping the total coolant flow rate of the system basically unchanged. This avoids blindly increasing the total pumping power consumption to deal with local hot spots, thus achieving higher energy utilization efficiency.

[0025] 3. Strong Adaptability: The system has the ability to learn parameters, dynamically adjust the target temperature, and perform closed-loop iterative optimization. It can automatically adapt to different chip characteristics, coolant conditions, and changing workloads, and has good robustness and wide applicability.

[0026] 4. High safety and reliability: The independent hardware safety monitoring mechanism provides fault safety protection for the system, and can quickly intervene when the actuator is abnormal or the pipeline fails, preventing the equipment from being damaged due to heat dissipation failure, and ensuring the long-term reliable operation of the system. Attached Figure Description

[0027] Figure 1 This is a flowchart illustrating the overall dynamic allocation of the multi-stage parallel liquid cooling system of the present invention.

[0028] Figure 2 This is a detailed flowchart of the temperature field processing and region classification of the present invention.

[0029] Figure 3 This is a flowchart of the flow optimization and allocation calculation branch of the present invention. Detailed Implementation

[0030] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0031] As attached Figures 1 to 3 The present invention discloses a multi-stage parallel liquid cooling path dynamic allocation system. The technical solution of the present invention will be described in detail below with reference to specific embodiments. It should be understood that the embodiments described herein are only for explaining the present invention and are not intended to limit the scope of protection of the present invention.

[0032] Example: A multi-stage parallel liquid cooling path dynamic allocation system and method thereof This embodiment describes a liquid cooling system for electronic devices. The system can dynamically adjust the flow distribution of multiple parallel cooling branches based on the real-time changing heat load of different areas on the device being cooled (taking a processor chip in a computing device as an example).

[0033] 1. System Composition The system mainly consists of three parts: a parallel liquid-cooled execution network, a temperature sensing network, and a control unit.

[0034] 1.1 Parallel Liquid-Cooled Execution Network The network includes a microchannel cold plate and fluid conduits. The microchannel cold plate is in thermal contact with the surface of the processor chip package being cooled via a thermally conductive interface material. The cold plate has internally machined... A group of independent microchannels, in which Integers greater than 1 (in this embodiment) Each microchannel group corresponds to a pre-defined functional area on the chip. The inlet of each microchannel group is connected to the main liquid supply pipe via an independent cooling branch. The cooling branches are connected in parallel to the same supply and return mains. On each cooling branch, an electrically controlled proportional control valve is installed in series. This valve is configured to receive electrical signal commands from the control unit and continuously adjust its opening accordingly, thereby changing the flow resistance and coolant flow rate of its branch.

[0035] 1.2 Temperature Sensing Network This network is used to collect temperature distribution information of the object being cooled. The temperature data comes from two sources: a digital temperature sensor integrated within the processor chip, and auxiliary temperature sensors mounted on the center of each target area on the chip package surface. All sensors are connected to the control unit via a digital communication bus. Let the total number of temperature sensors be... The sensor arrangement should ensure that each pre-divided area... Each functional area has corresponding temperature data available for collection.

[0036] 1.3 Control Unit The control unit is implemented using an embedded microcontroller as its core. It internally stores and runs the control program described in this system. The control unit has corresponding input / output interfaces: the input interface connects to the temperature sensing network to receive temperature data; the output interface connects to the drive circuits of each electronically controlled proportional control valve to send valve control commands. The control unit also includes non-volatile memory for storing calibration parameters and the control model.

[0037] 2. System characteristic parameter calibration method After the system is initially running or after maintenance reset, the control unit performs a parameter calibration process to obtain the key parameters required to establish an accurate control model.

[0038] 2.1 Calibration Preparation The control unit starts the external coolant circulation pump and circulates all the coolant. The opening degree of each electronically controlled proportional control valve is uniformly set to an intermediate value (e.g., 50% opening). The system operates in this state for a sufficient period of time until the flow rate in each branch and the temperature in each area of ​​the chip reach a stable state.

[0039] 2.2 Valve Characteristic Testing and Data Acquisition The control unit independently tests each valve sequentially according to a predetermined order. (The test is repeated for the valve number.) Valve No. (corresponding to valve No.) One cooling branch, For example, the process is as follows: (1) The control unit generates a containing A sequence of instructions with different opening values .For example, The opening values ​​are 30%, 40%, 50%, 60%, and 70%, respectively.

[0040] (2) For each opening instruction in the sequence ( The control unit sends the signal to the valve actuator. After the valve actuates to its designated position, the system enters a waiting state for a duration of [duration missing]. (For example, 20 seconds). The setting must ensure that the system reaches thermal equilibrium. Its specific value can be determined experimentally, but in principle it should be greater than the system's thermal time constant.

[0041] (3) During the steady-state phase, the control unit synchronously collects and records the following data: Valve command opening degree: (unit:%); branch road Real-time volumetric flow rate: (Unit: liters per minute) This can be obtained by measuring a flow meter connected in series on the branch line; all Readings from the temperature sensor: ( ).

[0042] Complete the first After completing all test points for valve number 1, the control unit restores the valve opening to the aforementioned intermediate value, and then repeats the above test process for the next valve until all valves are tested. All valves have been tested.

[0043] 2.3 Characteristic Parameter Calculation and Model Establishment Based on all the collected test data, the control unit performs offline calculations to establish two key models: Valve characteristic model: For each valve Let its opening command value be... (%), corresponding to the steady-state flow rate is ( Using the data point set obtained from the test. The static characteristic function of the valve is obtained by performing quadratic curve fitting using the least squares method: ; in, , , The coefficients obtained from the fitting process collectively characterize the valve opening under the current system operating pressure. With traffic The mapping relationship between them.

[0044] Regional thermal response model: For the first region on the chip Target areas ( ), and analyze its temperature effect on the branch serving the area (i.e., the first) The response characteristics of flow changes in the branch (one branch). During the calibration process, when the first branch... The change in valve number causes a change in the flow rate of its branch. ( When calculating the average temperature change of all relevant sensor readings in the region, the calculation is performed. ( Define the thermal response coefficient of this region. for: ; The unit is Since increased flow typically leads to a decrease in temperature, It is usually a negative value. This coefficient reflects the sensitivity of regional temperature to changes in flow rate.

[0045] 3. Dynamic Flow Allocation Control Process After parameter calibration is completed, the system enters periodic automatic control mode. Let the control period be... (For example, (seconds). Within each control cycle, the control unit executes the following steps sequentially.

[0046] Step S100: Temperature field information acquisition and digital reconstruction.

[0047] At the start of the control cycle The control unit reads all data via the bus. The current measurement values ​​of each temperature sensor are denoted as set. Based on each sensor Known physical coordinates The control unit performs spatial interpolation calculations.

[0048] As a typical, well-known technique, inverse distance-weighted interpolation can be used. Its inputs are discrete sensor coordinates and temperature values, and its output is a continuous two-dimensional temperature distribution function covering the entire chip package surface. For ease of computation, this function is discretized into a two-dimensional matrix within the controller. This matrix is ​​a complete digital representation of the system's thermal state at the current moment.

[0049] Step S200: Thermal status assessment and classification of cooling areas.

[0050] The control unit is based on a predefined... The geometric boundaries of each functional area, from the matrix Extract the region belonging to each area The temperature values ​​of all pixels (or data points) are calculated, and their arithmetic mean is taken as the average temperature of the region in the current period. .

[0051] As a specific implementation method of this embodiment, the control unit uses a sorting method to dynamically classify the region status: (1) Average temperature of each region ( Sort by value from highest to lowest.

[0052] (2) Set two positive integer parameters and In this embodiment, take , In practical applications, and It can be configured according to heat dissipation requirements and system size; for example, it can be set to the total number of zones. A certain percentage (e.g., 10% to 40%).

[0053] (3) Sort the first... The region with the highest average temperature is identified as the first type of region (i.e., region with high heat dissipation requirements), and its index is denoted as the set of regions. .

[0054] (4) The last part of the sorted sequence The region with the lowest average temperature is classified as a second-class region (i.e., a region that contributes cooling resources), and its index is denoted as the set of regions. .

[0055] (5) The remaining areas are classified as Category III areas.

[0056] Step S300: Quantitative calculation of heat dissipation requirements and available resources.

[0057] Heat dissipation requirement estimation: For each Class I region (Right now Estimate the required increase in coolant flow rate. ( One estimation method based on the linear response assumption is as follows: ; in, The preset temperature target value (e.g., 80) Its settings can be based on the chip's thermal design specifications or the upper limit of the temperature for long-term reliable operation. This represents the thermal response coefficient obtained in this region during the calibration phase.

[0058] Deployable resource assessment: for each Category II area (Right now Assess the redundant coolant flow rate that can be safely redeployed. Query the flow rate currently allocated to this area by the system. ( Set the minimum safe flow rate that must be guaranteed for this area. (For example, 0.3) This is to prevent abnormal temperature increases in the area due to excessive flow reduction. The flow that can be released in this area is... ( )for: ; Step S400: Traffic redistribution decision based on the optimization model.

[0059] Based on the results of steps S200 and S300, the control unit constructs a constrained mathematical optimization problem to solve for the optimal flow adjustment amount of each branch.

[0060] First, based on the thermal response coefficient obtained from the calibration And assuming that within a single-step adjustment range, the temperature change and flow rate change of the zone are approximately linearly related, then the zone... Performing traffic adjustment Predicted temperature It can be represented as: ; As a specific implementation method of this embodiment, the optimization problem is defined as follows: Decision variables: Defined vector .in, Indicates the plan for the first Flow adjustment made by each cooling branch (unit: ), whose value can be positive or negative.

[0061] Optimization objective: The objective is to ensure that, after flow adjustment, the predicted temperature for all regions is [adjusted / improved]. Compared with reference temperature (usually acceptable) To minimize the overall deviation while prioritizing heat dissipation in the first type of region, the following quadratic objective function is constructed: ; in, These are the weighting coefficients. The weighting coefficients are assigned differentiated values ​​based on the region classification: for Assign higher weight (For example, );for Assign lower weight (For example, For the third type of region, assign a medium weight. (For example, ).

[0062] Constraints: 1. Total flow conservation constraint: This constraint ensures that the flow redistribution process does not change the total input flow of the entire cooling system.

[0063] 2. Flow direction constraint: ; For the third type of area, This constraint clarifies the basic direction for diverting traffic from the second-class area and allocating it to the first-class area.

[0064] 3. Valve limit constraint: ,in This is the maximum permissible adjustment per step (e.g., 0.5). The dynamic adjustment capability of the electronically controlled proportional control valve is determined by its specific characteristics.

[0065] 4. Resource constraints: This constraint ensures that traffic drawn from any Category 2 area does not exceed its currently available limit.

[0066] The control unit invokes its built-in or integrated optimization algorithm solver (e.g., an efficient set method or interior point method solver that can be used to solve quadratic programming problems).

[0067] The input to the solver is the objective function. The coefficient matrix and vector, along with a system of linear equality and inequality equations constituting the constraints, output the optimal flow adjustment vector that minimizes the objective function while satisfying all constraints. .

[0068] Step S500: Generation and execution of control instructions.

[0069] (1) Target flow calculation: Based on the optimal adjustment amount obtained from the optimization solution, calculate the target flow that each branch should achieve in the next control cycle: ; (2) Valve opening calculation: Using the valve characteristic model established during the calibration phase, the target flow rate is calculated. Convert to the corresponding valve opening command .

[0070] Specifically, this can be achieved by solving equations. This can be achieved, or obtained by querying a "flow-opening" correspondence table pre-discretized based on the equation.

[0071] (3) Command issuance and execution: The control unit will issue and execute the calculated opening command set. The signal is sent through its output interface to the actuators of the corresponding electronically controlled proportional control valves. The actuators precisely adjust the valve openings according to the received command signals, thereby physically realizing the dynamic redistribution of coolant flow in each parallel branch.

[0072] Step S600: Closed-loop feedback and iterative optimization.

[0073] As a preferred embodiment of this invention, the system may include a closed-loop feedback and iterative optimization mechanism to further improve the accuracy of a single control cycle and to cope with model mismatch or sudden disturbances.

[0074] (1) System response wait: After the flow adjustment is completed in step S500, the control unit starts a timer and waits for a preset hot response time. (For example, 2.0 seconds). The value of should be greater than or equal to the bulk heat transfer time constant from the chip junction to the coolant of the cooled device, which can be determined by the device's thermal design parameters (thermal capacity, thermal resistance) or by previous experiments.

[0075] (2) Effect evaluation: After the waiting time ends, the control unit re-executes steps S100 and S200 to collect new temperature data and generate a new temperature field matrix. This leads to the updated status of each region.

[0076] (3) Iterative judgment: Calculate the original set of first-class regions Average temperature under the new temperature field Calculate the actual temperature drop. Among them, the expected temperature drop It can be estimated based on the flow increment allocated to the first type of region and its thermal response coefficient, for example... .like Less than a certain proportion (For example, That is, satisfying If the adjustment fails to meet expectations, it is determined that the effect is not as expected and iterative optimization is required.

[0077] (4) Iterative execution: If it is determined that iteration is required, then the latest system state (i.e., the new temperature field) is used. Using the current flow of each branch as a new initial condition, proceed to step S300 to begin a new round of demand quantification, optimization, and instruction execution. To avoid infinite loops due to abnormal situations, a maximum iteration limit can be set (e.g., 3 times).

[0078] 4. Adaptive strategies and security monitoring 4.1 Adaptive Parameter Adjustment To enable the system to adapt to different workloads, the control unit monitors the real-time load rate of the processor chip and dynamically adjusts key control parameters accordingly. Specifically, the control unit periodically (e.g., every 10 seconds) calculates the processor's average load rate within the most recent time window. .

[0079] like (For example, Then the target temperature in step S300 will be... From default value (For example, 80) Dynamically downgraded to (For example, 75) This allows the system to adopt a more aggressive cooling strategy under heavy loads, prioritizing performance and safety.

[0080] like (For example, ), then Upgraded to (For example, 82) This allows the system to relax temperature control targets under light loads, which helps reduce regulation frequency and energy consumption, thus achieving energy efficiency optimization.

[0081] like Between and In between, maintain .

[0082] 4.2 Independent security monitoring To ensure system hardware reliability, a high-priority security monitoring task, independent of the main control loop, runs continuously in the background, with an execution frequency far exceeding that of the main control loop. (For example, running at 100Hz).

[0083] Valve status monitoring: This task reads the valve position feedback signals of each electronically controlled proportional control valve in real time. and the opening degree relative to the most recently issued command from the control unit. Compare. If any valve is detected... satisfy (in To allow for tolerances, for example If the valve remains stuck (at full scale) for more than a set time (e.g., 1 second), it is determined that the valve is stuck, loose, or has a drive failure.

[0084] Pipeline pressure monitoring: This task involves real-time reading of pressure sensor data installed on each cooling branch (typically at the inlet or outlet of the electronically controlled proportional control valve). If any satisfy or (in For example, a preset safe operating pressure range. , If this state persists, it is determined that there may be pipe blockage, coolant leakage, or main circulation pump malfunction in the system.

[0085] Once the safety monitoring task detects any of the above-mentioned abnormal states, it will immediately implement fault-safe protection operations: First, it will suspend the optimized control process of the main control unit; then, it will forcibly send a unified, preset safety opening command to the actuators of all electronically controlled proportional control valves. (For example, 60% opening) to bring the system into a known and stable basic cooling state; at the same time, alarm information containing fault codes and details is reported to the host computer or network management unit through the system management interface (such as IPMI, USB, etc.).

[0086] This system continuously and cyclically executes the aforementioned dynamic flow allocation control process. Through online temperature field construction and intelligent regional classification, it achieves precise perception and problem localization of heat load space. By constructing an optimization problem model that includes flow conservation, directional constraints, and capacity limitations, it achieves globally optimal decision-making for cooling resource allocation schemes. Through closed-loop iteration and parameter adaptation, it enhances the system's robustness and scenario adaptability. Ultimately, without significantly increasing the system's total pumping power consumption, it efficiently allocates cooling capacity from areas with lower temperatures and lighter loads to areas with higher temperatures and more pressing loads.

[0087] Compared to traditional fixed-path liquid cooling solutions, this system can more effectively suppress localized hot spots on the chip, reduce the maximum junction temperature, and improve the uniformity of chip surface temperature. This directly translates to improved operational stability, reliability, and long-term lifespan of high-performance computing devices when dealing with complex, dynamic, and uneven workloads. Furthermore, by improving the utilization efficiency of cooling resources, a higher energy efficiency ratio is also achieved at the system level.

[0088] Finally, the following points should be noted: First, in the description of this application, it should be noted that, unless otherwise specified and limited, the terms "installation", "connection", and "linkage" should be interpreted broadly, and can be mechanical or electrical connections, or internal connections between two components, or direct connections. "Up", "down", "left", "right", etc. are only used to indicate relative positional relationships. When the absolute position of the described object changes, the relative positional relationship may change. Secondly: The accompanying drawings of the embodiments disclosed in this invention only involve the structures involved in the embodiments disclosed in this invention. Other structures can refer to the general design. In the absence of conflict, the same embodiment and different embodiments of this invention can be combined with each other. In conclusion, the above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the protection scope of the present invention.

Claims

1. A multi-stage parallel liquid cooling path dynamic allocation system, characterized in that, include: A liquid cooling execution network consists of multiple independent cooling branches connected in parallel. Each of the independent cooling branches is equipped with an electrically controlled flow regulating valve and is used to fluidly communicate with a section of the device being cooled. A temperature sensing network consists of multiple temperature sensors disposed on the cooled device; A controller, connected to the temperature sensing network and all the electrically controlled flow regulating valves, is configured to: Collect readings from each temperature sensor and generate a temperature distribution map of the cooled device based on the readings and the sensor positions; The average temperature of each zone is calculated based on the temperature distribution map. Based on the ranking of the average temperatures, at least one zone with the highest average temperature is identified as a first-class zone, and at least one zone with the lowest average temperature is identified as a second-class zone. Calculate a cooling demand flow for each partition identified as the first type of region, and calculate a release flow for each partition identified as the second type of region; The allocation direction is to transfer the releasable flow from the cooling branch corresponding to the second type of region to the cooling branch corresponding to the first type of region. The first constraint is to keep the total flow of the system unchanged. The goal is to find the optimal solution that makes the predicted temperature of each zone tend to be consistent. The required flow adjustment amount for each independent cooling branch is calculated. The corresponding electronically controlled flow regulating valve is activated based on the flow adjustment amount.

2. The multi-stage parallel liquid cooling path dynamic allocation system according to claim 1, characterized in that, The controller is configured to: identify the partitions located in the first N1 positions of the average temperature sort as first-class regions, and identify the partitions located in the last N2 positions as second-class regions, where N1 and N2 are preset positive integers.

3. The multi-stage parallel liquid cooling path dynamic allocation system according to claim 1, characterized in that, When calculating the required flow adjustment for each branch, the controller also follows the second and third constraints; The second constraint is: the flow adjustment of branches flowing into the first type of area is greater than or equal to zero, and the flow adjustment of branches flowing into the second type of area is less than or equal to zero. The third constraint is that the absolute value of the flow adjustment of any branch is not greater than a preset maximum adjustment threshold.

4. The multi-stage parallel liquid cooling path dynamic allocation system according to claim 1, characterized in that, The controller is also configured to execute an iterative optimization process: After the valve is activated, a preset delay time is waited, and then the temperature sensor readings are collected again to generate a new temperature distribution map. Based on the new temperature distribution map, determine whether the temperature drop in the first type of region has reached the expected level; If the expected result is not achieved, the current state will be used as the initial condition to re-execute the steps of calculating the cooling demand flow rate, calculating the release flow rate, calculating the flow rate adjustment amount, and driving the valve.

5. The multi-stage parallel liquid cooling path dynamic allocation system according to claim 1, characterized in that, The controller executes a learning process upon system startup, the learning process including: Adjust each electrically controlled flow regulating valve to a different opening degree in sequence, and record the flow value of the branch and the temperature value of the affected zone after each opening degree stabilizes; Based on the recorded data, the correspondence between the opening degree and flow rate of each electrically controlled flow regulating valve was established, as well as the response relationship between flow rate changes and temperature changes in each zone.

6. The multi-stage parallel liquid cooling path dynamic allocation system according to claim 1, characterized in that, The target temperature value used to calculate the cooling demand flow rate for the first type of regional partition is a dynamic value, and the controller adjusts the magnitude of the target temperature value according to the real-time load rate of the cooled device.

7. The multi-stage parallel liquid cooling path dynamic allocation system according to claim 1, characterized in that, It also includes a security monitoring module, which is configured as follows: Monitor whether the actual opening degree of each electrically controlled flow regulating valve is consistent with the commanded opening degree; Monitor whether the fluid pressure of each independent cooling branch is within the preset normal range; When an abnormality is detected in the valve or pressure, control all electrically controlled flow regulating valves to switch to a preset safe opening degree.

8. The multi-stage parallel liquid cooling path dynamic allocation system according to claim 1, characterized in that, The cooled device is a semiconductor chip, and the partitions correspond to different functional module units on the semiconductor chip.

9. A multi-stage parallel liquid cooling path dynamic allocation system according to claim 8, characterized in that, The liquid cooling execution network includes a microchannel cold plate, and the multiple independent cooling branches are respectively connected to the independent microchannel groups corresponding to each partition in the microchannel cold plate.

10. A multi-stage parallel liquid cooling path dynamic allocation system according to claim 1, characterized in that, The controller uses linear programming or quadratic programming algorithms to find the optimal solution.

Images