A manifold split cooling method and system based on heat flux perception
Patent Information
- Application Number
- CN202611185995.2
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2026-08-06
- Publication Date
- 2026-09-15
Smart Images
Figure CN122755751A_ABST
Abstract
Description
Technical Field
[0001] This invention relates to the field of thermal management technology, specifically to a manifold split cooling method and system based on heat flux density sensing. Background Technology
[0002] Manifold cooling technology is widely used in heat dissipation of electronic devices, thermal management of power batteries, and industrial heat exchange. Existing manifold cooling systems typically employ a fixed split ratio design, meaning that the flow distribution ratio of each branch pipe is preset during system commissioning and remains unchanged during operation. However, the heat flux density distribution on the surface of a heat source often exhibits significant spatial heterogeneity and temporal dynamics. A fixed split ratio cannot adapt to these variations, leading to overheating in some areas, forming localized hotspots, while other areas are overcooled, resulting in wasted cooling medium.
[0003] Some existing systems attempt to monitor the surface temperature of heat sources using temperature sensors and adjust the flow rate of the cooling medium based on temperature deviations. However, temperature is a result of heat accumulation, not a direct representation of heat generation, and there is a delay in heat conduction from a change in heat flux density to a rise in temperature. This reactive control mode based on temperature feedback inevitably leads to thermal response lag, resulting in temperature overshoot or undershoot during sudden changes in heat load, severely affecting the reliability and service life of the object being cooled.
[0004] Therefore, the main shortcomings of the existing technology are: lack of real-time sensing capability of the spatial distribution of heat flux density on the heat source surface, making it impossible to perform feedforward adjustment at the initial moment of heat load change; the distribution of cooling medium flow can only be adjusted at the macro scale or a single level, making it difficult to achieve precise on-demand distribution from the global to the microchannel level; and the system lacks an adaptive tuning mechanism, making it impossible to dynamically optimize control parameters based on operating data.
[0005] Therefore, there is an urgent need to provide a manifold split cooling method and system that can sense heat flux density distribution in real time, realize multi-level hierarchical feedforward control, and have adaptive optimization capabilities to solve the above-mentioned technical problems. Summary of the Invention
[0006] The purpose of this invention is to provide a manifold shunt cooling method and system based on heat flux density sensing, so as to solve the technical problems existing in the prior art, such as lack of heat flux density distribution sensing, sluggish shunt response, and poor temperature uniformity.
[0007] To achieve the above objectives, according to one aspect of the present invention, a manifold split cooling method based on heat flux density sensing is provided, comprising the following steps: S1: Initialization phase. After the system is powered on, the distributed heat flux density sensor array is calibrated, the regulating valves at each level in the multi-stage manifold diversion structure are reset, and the initial heat flux density reference value is read.
[0008] S2: In the sensing and acquisition phase, the distributed heat flux density sensor array synchronously acquires heat flux density data on the surface of the heat source, constructs a spatial heat flux density distribution map, and calculates the heat flux density gradient.
[0009] S3: The hierarchical decision-making stage is based on the heat flux density distribution and gradient information. It executes hierarchical decisions of the multi-level slave controller, including the first-level global diversion strategy calculation, the second-level fine adjustment of the branch, and the third-level precise compensation.
[0010] S4: Execution control phase. Based on the hierarchical decision results, the first-stage diversion valve, the second-stage regulating valve, and the third-stage compensation valve are driven to achieve dynamic flow distribution in the multi-stage manifold diversion structure.
[0011] S5: Feedback optimization phase, collect feedback data from flow sensor and temperature sensor, evaluate control effect, update heat load prediction model parameters, and achieve adaptive tuning of control parameters.
[0012] Preferably, step S3, the hierarchical decision-making stage, further includes: Level 1 judgment, where the main controller judges whether the heat flux density gradient exceeds a preset threshold; if it does, it triggers Level 1 response and directly adjusts the flow ratio of the Level 1 diversion valve; Level 2 judgment, where the Level 1 slave controller judges whether the heat flux density deviation of each branch exceeds a preset threshold; if it does, it triggers Level 2 fine adjustment and adjusts the opening of the Level 2 regulating valve of the corresponding branch; Level 3 judgment, where the Level 2 slave controller judges whether Level 3 compensation needs to be performed based on the precise compensation target; if so, it triggers Level 3 precise compensation and drives the Level 3 compensation valve to make fine adjustments.
[0013] Preferably, the calculation of the heat flux density gradient in step S2 includes: calculating the first-order partial derivatives of the heat flux density along the X and Y directions of the heat source surface, and calculating the magnitude of the heat flux density gradient and the direction angle of the heat flux density gradient based on the first-order partial derivatives.
[0014] Preferably, in step S4, during the control phase, the response times of each level of control valve meet the preset timing coordination relationship: the response time of the first-level diverter valve is less than the response time of the second-level control valve, and the response time of the second-level control valve is less than the response time of the third-level compensation valve.
[0015] Preferably, in step S5, the feedback optimization stage, based on historical heat flux density data, a time series prediction algorithm is used to predict the heat flux density distribution at the next moment, and the diversion strategy is adjusted in advance accordingly.
[0016] A manifold-based split-flow cooling system based on heat flux density sensing includes: a sensing layer for real-time acquisition of heat flux density data from the surface of a heat source; a main controller for parsing heat flux density distribution information, calculating heat flux density gradient, and generating a global split-flow strategy; a multi-level slave controller, including a first-level slave controller, a second-level slave controller, and a third-level slave controller, for executing hierarchical split-flow control; a multi-level manifold-based split-flow structure, including a first-level manifold, a second-level manifold, and a third-level microchannel, wherein the first-level manifold is equipped with a first-level split-flow valve, the second-level manifold is equipped with a second-level regulating valve, and the third-level microchannel is equipped with a third-level compensating valve; and a cooling medium circulation system for providing cooling medium and driving its circulation within the multi-level manifold-based split-flow structure.
[0017] Preferably, the sensing layer includes: a distributed heat flux density sensor array, consisting of multiple heat flux density sensor nodes, arranged at key locations on the surface of the heat source or the contact surface between the heat source and the manifold; a flow sensor, installed in the manifold inlet pipe, for monitoring the total flow rate of the cooling medium; and a temperature sensor, installed at the outlet of each branch of the manifold, for monitoring the outlet temperature of the cooling medium in each branch.
[0018] Preferably, the heat flux density sensor nodes in the distributed heat flux density sensor array are thin-film heat flux density sensors or thermopile heat flux density sensors, and their measurement range is within the preset heat flux density measurement range, and their response time meets the preset response time requirements.
[0019] Preferably, the first-stage diverter valve is a piezoelectric high-speed proportional valve, and its response time meets the preset response time requirement; the second-stage regulating valve is a high-speed proportional solenoid valve, and its response time meets the preset response time requirement; the third-stage compensation valve is a miniature solenoid valve, and its response time meets the preset response time requirement.
[0020] Preferably, the number of branches of the first-stage manifold in the multi-stage manifold diversion structure is within a preset range, the number of microchannels of the second-stage manifold is within a preset range, and the width of the third-stage microchannel is within a preset range.
[0021] Preferably, the main controller communicates with each slave controller using a preset first communication bus method, and the main controller communicates with the distributed heat flux density sensor array using a preset second communication bus method.
[0022] Preferably, the main controller has a built-in heat load prediction model and uses a preset neural network algorithm to predict the time series of heat flux density.
[0023] Preferably, the cooling medium circulation system includes a circulation pump, an expansion tank, and a radiator. The circulation pump is controlled by a variable frequency drive to adjust the flow rate of the cooling medium according to the system's heat load requirements.
[0024] Compared with the prior art, the beneficial effects of the present invention are as follows: This invention uses a distributed heat flux density sensor array to perceive the spatial heat flux density distribution on the surface of a heat source in real time. Combined with heat flux density gradient information, it drives a three-level hierarchical control strategy to achieve precise, on-demand allocation of cooling medium flow from the global level to the microchannel level. Compared to existing cooling methods based on temperature feedback or fixed flow ratios, this invention can actively eliminate local hot spots, control the temperature difference on the heat source surface within a smaller range, and effectively suppress temperature overshoot and undershoot. This invention employs a hierarchical, progressive control architecture with a master controller and multiple slave controllers working in tandem: the first-level response rapidly adjusts the first-level diversion valve based on the direction angle of the heat flux density gradient, achieving macroscopic guidance of the cooling medium flow direction; the second-level response finely adjusts the opening of the second-level regulating valve according to the heat flux density deviation in the branch region; and the third-level response drives the third-level compensation valve via a high-speed PWM signal to achieve millisecond-level local flow compensation. The valves at each level operate in a pre-set sequence, resulting in an overall control response time that is more than an order of magnitude shorter than traditional temperature feedback control. This invention employs an adaptive flow distribution strategy based on heat flux density sensing to match the cooling medium flow rate allocation with the actual spatial distribution of heat load, avoiding overcooling in low heat flux density areas and reducing cooling medium pumping power consumption. Simultaneously, a built-in heat load prediction model forecasts future distribution trends based on historical heat flux density data, enabling feedforward control and further reducing ineffective flow losses, thus effectively improving the overall system energy efficiency ratio. The system of this invention has self-calibration, self-diagnosis, and adaptive optimization capabilities; it automatically completes sensor calibration and valve core repositioning during the initialization phase; it dynamically updates the parameters of the heat load prediction model based on flow and temperature feedback data during operation; and it can adaptively adjust the graded decision threshold according to different operating conditions to achieve online optimization of the control strategy and meet the intelligent thermal management needs under complex dynamic heat load scenarios. Attached Figure Description
[0025] Figure 1 This is a framework diagram of the system in this invention; Figure 2 This is a schematic diagram of the core principle framework for heat flux density gradient calculation and hierarchical decision-making in this invention; Figure 3 This is a flowchart illustrating the hierarchical execution control and feedback optimization process of the multi-level manifold diversion structure in this invention. Detailed Implementation
[0026] To make the objectives, technical solutions, and advantages of this invention clearer and more complete, the invention will be further described in detail below with reference to the accompanying drawings and specific embodiments. It should be understood that the specific embodiments described herein are merely illustrative of the invention and are not intended to limit the invention in any way.
[0027] Example 1: This example focuses on the cooling scenario of power battery modules for new energy vehicles, and details the specific structure and workflow of the manifold shunt cooling method and system based on heat flux density sensing proposed in this invention.
[0028] A manifold split cooling system based on heat flux density sensing includes a sensing layer, a main controller, a multi-stage slave controller, a multi-stage manifold split structure, and a cooling circulation system.
[0029] The sensing layer includes a distributed heat flux density sensor array, a flow sensor group, and a temperature sensor group.
[0030] The distributed heat flux density sensor array consists of 16 heat flux density sensor nodes, evenly arranged in a 4x4 matrix on the lower surface of the power battery module, with a spacing of 100 mm between adjacent nodes. The sensor nodes employ thin-film heat flux density sensors, measuring based on the Seebeck effect, with a measurement range of 0.1 to 500 kW / m², a response time of less than 5 milliseconds, and an accuracy of ±3%. Each node has a built-in temperature compensation circuit and is connected to the main controller via a flexible circuit board.
[0031] The flow sensor group includes an inlet flow sensor and branch flow sensors. The inlet flow sensor is located at the connection between the cooling medium inlet and the primary manifold, with a measurement range of 0 to 100 liters per minute and an accuracy of ±0.5%. The branch flow sensors are located at the inlet of each secondary manifold, with a measurement range of 0 to 20 liters per minute and an accuracy of ±1%.
[0032] The temperature sensor group includes an inlet temperature sensor, an outlet temperature sensor, and a heat source surface temperature sensor. The inlet temperature sensor is a PT1000, with a measurement range of 0 to 150 degrees Celsius and an accuracy of ±0.1 degrees Celsius. The outlet temperature sensor is located at the outlet of each secondary manifold branch and uses an NTC thermistor with an accuracy of ±0.5 degrees Celsius. The heat source surface temperature sensor is a thin-film platinum resistance thermometer with a response time of less than 100 milliseconds and an accuracy of ±0.5 degrees Celsius.
[0033] The main controller uses a 32-bit ARM Cortex-M7 core microcontroller, connects to the aforementioned sensors via an external interface, and communicates with each slave controller via a CAN bus. The main controller is used to construct a spatial heat flux density distribution map, calculate the heat flux density gradient, and execute the first-level global flow distribution strategy calculation.
[0034] The heat flux density distribution map is constructed as follows: based on 16 heat flux density data points, a bilinear interpolation algorithm is used to interpolate the discrete data points into a continuous heat flux density distribution field. This heat flux density distribution map provides an intuitive representation of the global heat flux density distribution for primary decision-making. In specific hierarchical decision-making, primary decision-making is based on the continuous distribution map; secondary and tertiary decision-making directly calls the original data from the sensor nodes corresponding to each branch and microchannel, rather than the interpolated distribution map data, to avoid the additional calculation errors introduced by interpolation affecting the accuracy of fine-tuning.
[0035] The heat flux density gradient is calculated using the finite difference method, calculating the first-order partial derivatives of the heat flux density along the X and Y directions of the heat source surface. The first-order partial derivative along the X direction is equal to the difference in heat flux density between adjacent sensor nodes in the X direction divided by the node spacing; the first-order partial derivative along the Y direction is equal to the difference in heat flux density between adjacent sensor nodes in the Y direction divided by the node spacing. The magnitude of the heat flux density gradient is equal to the square root of the sum of the squares of the two partial derivatives, and the direction angle of the heat flux density gradient is equal to the arctangent of the partial derivatives in the Y and X directions.
[0036] For sensor nodes located at the boundaries of a 4×4 matrix, there are no adjacent nodes in the outer direction, making direct calculation using the central difference method impossible. For boundary nodes, this invention employs a one-sided difference method for supplementary calculation: along the X direction, the first partial derivative of the left boundary node equals the heat flux density of the adjacent node to the right minus the heat flux density of the current node divided by the node spacing (forward difference); for the right boundary node, it equals the heat flux density of the current node minus the heat flux density of the adjacent node to the left divided by the node spacing (backward difference); the same applies to the upper and lower boundary nodes along the Y direction. For the four corner points, the first partial derivatives in both the X and Y directions are calculated using the one-sided difference method. By combining the central difference method (for internal nodes) and the one-sided difference method (for boundary nodes), the gradient field is ensured to be defined at all sensor node locations, thus obtaining a complete heat flux density gradient distribution.
[0037] The main controller is also equipped with a heat load prediction function: it uses a long short-term memory network (LSTM) model to predict the heat flux density distribution at future moments using historical heat flux density data.
[0038] The model's data preprocessing method is as follows: the heat flux density data of 16 nodes in the distributed heat flux density sensor array over the past 10 control cycles (i.e., 100 milliseconds of historical data) are used as the input sequence, with an input dimension of 160 (16 nodes × 10 time steps) and an output dimension of 16 (the predicted heat flux density values of the 16 nodes at the next time step). The input data is scaled to the [0,1] interval using the Min-Max normalization method to eliminate the influence of dimensions and accelerate model convergence.
[0039] The model network structure consists of three LSTM layers, each with 128 hidden units. The activation function within each LSTM unit is tanh, and the output layer is a fully connected layer with the ReLU activation function. To prevent overfitting, a Dropout layer is introduced between the LSTM layers, with a Dropout ratio set to 0.2.
[0040] The model training parameters are set as follows: the training set and validation set are divided in a ratio of 80%:20%; the loss function is mean squared error; the optimizer is Adam, and the learning rate is set to 0.001; the batch size is set to 32; the maximum number of training epochs is set to 100; an early stopping strategy is adopted, and training is terminated early when the validation set loss does not decrease for 10 consecutive training epochs to prevent overfitting.
[0041] Model parameters are stored in the non-volatile memory of the main controller, supporting online incremental updates based on actual operational feedback data. During online operation, the main controller inputs the heat flux density data of the current control cycle into the trained LSTM model, and the model outputs the predicted heat flux density values of each sensor node for the next control cycle. The main controller adjusts the flow distribution strategy in advance based on the prediction results to achieve feedforward control.
[0042] The multi-level slave controller includes one primary slave controller, four secondary slave controllers, and 32 tertiary slave controllers, each located near the corresponding level manifold. Each slave controller receives commands from its superior via a CAN bus and drives the corresponding valve via PWM signals.
[0043] The primary controller receives global flow splitting strategy commands from the primary controller. It calculates the primary flow splitting ratio adjustment based on the magnitude and direction of the heat flux density gradient and generates the drive signal for the primary flow splitting valve accordingly.
[0044] The secondary slave controller receives branch adjustment commands from the primary slave controller. Based on the heat flux density deviation of each branch, the secondary slave controller calculates the adjustment amount of the secondary control valve opening and generates the PWM drive signal for the secondary control valve accordingly.
[0045] The third-level slave controller receives compensation commands from the second-level slave controller. Based on the heat flux density deviation in each microchannel region, the third-level slave controller calculates the adjustment amount of the third-level compensation valve opening and generates a high-speed PWM drive signal for the third-level compensation valve accordingly.
[0046] The multi-stage manifold diversion structure includes a primary manifold, a secondary manifold, and a tertiary microchannel, as well as a corresponding primary diversion valve, a secondary regulating valve, and a tertiary compensating valve.
[0047] The primary manifold is a stainless steel main distribution chamber, with its inlet connected to the cooling medium inlet pipe and its outlet branching into four branches that connect to the secondary manifold. The primary distribution valve is installed at the connection between the inlet section and the branch section of the primary manifold. It is a piezoelectric high-speed proportional valve with a response time of less than 10 milliseconds, a control accuracy of ±1%, and a drive signal of 0 to 10 volts.
[0048] There are four secondary manifolds, each with an aluminum alloy flow distribution chamber. Each secondary manifold inlet connects to a branch outlet of the primary manifold, and the outlet is divided into eight microchannels. A secondary regulating valve is installed at the connection between the inlet section and the microchannel outlet section of the secondary manifold. It is a high-speed proportional solenoid valve with a response time of less than 50 milliseconds, driven by a PWM signal with a duty cycle of 0% to 100%.
[0049] The three-stage microchannels comprise 32 units, directly positioned below the heat source surface. Each microchannel has a three-stage compensation valve installed at its inlet, employing a miniature solenoid valve with a response time of less than 100 milliseconds and driven by a high-speed PWM signal.
[0050] The cooling medium circulation system includes a variable frequency magnetic pump, a closed expansion tank, an aluminum plate-fin radiator, and a cooling medium tank. The circulation pump has a rated flow rate of 80 liters per minute and is frequency-controlled; the expansion tank has a volume of 500 ml and a pre-charge pressure of 0.1 MPa; the radiator has a heat exchange area of 2 square meters and is equipped with forced air cooling; the cooling medium is an aqueous solution of ethylene glycol (volume concentration 40%).
[0051] A manifold shunt cooling method based on heat flux density sensing, with a control cycle of 10 milliseconds, includes the following steps: S1: Initialization Phase After the system is powered on, the main controller first executes the initialization program.
[0052] Sensor calibration steps: The main controller controls the distributed heat flux density sensor array to perform calibration. Each sensor node enters self-test mode to check if the sensor circuit is normal; it reads the calibration parameters inside the node and corrects the offset of the measurement zero point; it applies a known heat flux density value through an external standard heat flux source to verify whether the sensor sensitivity is within the allowable range. If calibration fails, the main controller records the faulty node and issues an alarm signal; if calibration passes, the sensor node enters measurement mode.
[0053] Valve spool return procedure: Each slave controller independently controls the return operation. The first-level slave controller sends a return command to the first-level diverter valve, moving the valve spool to the fully open position, then to the fully closed position, and finally to the 50% opening position as the initial operating point. The second-level slave controller sends a return command to each second-level regulating valve, moving the valve spool to the default opening position. The third-level slave controller sends a return command to each third-level compensating valve, moving the valve spool to the default opening position. During the return process, each slave controller monitors the valve position sensor signal to confirm that the valve spool has reached the target position.
[0054] Reference value reading steps: The main controller reads the heat flux density reference value, flow rate reference value, and temperature reference value saved before the last shutdown from the non-volatile memory as the initial reference for this operation. At the same time, the main controller corrects the heat flux density reference value based on the current ambient temperature and the initial temperature of the cooling medium.
[0055] S2: Sensing and Acquisition Phase At the start of each control cycle, the distributed heat flux density sensor array synchronously collects heat flux density data from the surface of the heat source.
[0056] The main controller sends a synchronization acquisition command to the distributed heat flux density sensor array, which includes the acquisition start time and sampling duration parameters. The 16 sensor nodes begin acquisition simultaneously, converting the analog heat flux density signal into a digital signal using an analog-to-digital converter. The sampling frequency is set to 1 kHz, with each node obtaining one heat flux density data point per sample.
[0057] After sampling, each sensor node sends the raw data to the main controller via the integrated circuit bus. Upon receiving the data, the main controller first performs filtering, using a median filter algorithm to remove impulse noise; then it performs denoising, using a Kalman filter algorithm to smooth the signal; finally, it calculates the average heat flux density of each node, resulting in 16 heat flux density data points.
[0058] The main controller constructs a spatial heat flux density distribution map based on 16 heat flux density data points. The distribution map uses a bilinear interpolation algorithm to interpolate the discrete heat flux density data points into a continuous heat flux density distribution field. The main controller stores the distribution map data in memory for subsequent gradient calculations.
[0059] Calculate the first-order partial derivatives of the heat flux density along the X and Y directions of the heat source surface. The first-order partial derivatives are calculated using the central difference method: the X-direction partial derivative equals the heat flux density at the right node minus the heat flux density at the left node divided by twice the node spacing; the Y-direction partial derivative equals the heat flux density at the upper node minus the heat flux density at the lower node divided by twice the node spacing. The magnitude of the heat flux density gradient is equal to the square root of the sum of the squares of the X-direction and Y-direction partial derivatives. The direction angle of the heat flux density gradient is equal to the arctangent of the X-direction and Y-direction partial derivatives.
[0060] The main controller simultaneously receives data from both the flow sensor group and the temperature sensor group. The inlet flow sensor and branch flow sensors transmit the cooling medium flow data to the main controller after analog-to-digital conversion. The inlet temperature sensor and each outlet temperature sensor transmit the cooling medium temperature data to the main controller. The main controller stores the flow and temperature data, along with the heat flux density data, in its memory to form complete system status data.
[0061] S3: Hierarchical Decision-Making Stage Based on heat flux density distribution and gradient information, the master controller and multiple slave controllers perform hierarchical decisions. The first-level judgment is based on the continuous heat flux density distribution map: the master controller extracts the integral heat flux and overall gradient direction of each region from the distribution map and determines whether it is necessary to trigger global flow guidance.
[0062] Level 1 Decision: The main controller determines whether the maximum value of the heat flux density gradient exceeds a first preset threshold. The first preset threshold is set based on the rated heat load of the heat source and the cooling system capacity; in this embodiment, it is set to 50 kW per cubic meter. If the maximum value of the heat flux density gradient exceeds the first preset threshold, it indicates significant uneven heat flow distribution on the heat source surface, triggering a Level 1 response. The main controller calculates the target flow rate ratio of each branch based on the direction angle of the heat flux density gradient and generates a Level 1 flow splitting strategy command. The Level 1 flow splitting strategy command includes the target opening value of each branch and is sent to the Level 1 slave controller via the CAN bus. If the maximum value of the heat flux density gradient does not exceed the first preset threshold, the main controller sends the gradient information to the Level 1 slave controller via the CAN bus, proceeding to Level 2 Decision.
[0063] Secondary Judgment: After receiving the gradient information sent by the master controller, the primary slave controller calculates the average heat flux density of each branch of the primary manifold, i.e., the average heat flux density data points in the corresponding area of each branch. The primary slave controller determines whether the deviation of the average heat flux density of each branch from the target value exceeds a second preset threshold. The second preset threshold is set according to the temperature uniformity requirements, and in this embodiment, it is set to 10%. If the average heat flux density deviation of a certain branch exceeds the second preset threshold, it indicates that the heat source area corresponding to that branch is overheated or undercooled, triggering secondary fine adjustment. The primary slave controller calculates the target opening adjustment amount of each secondary regulating valve based on the branch heat flux density deviation and generates a secondary adjustment command. The secondary adjustment command includes the branch number to be adjusted and the adjustment amount, and is sent to the corresponding secondary slave controller via the CAN bus. If the deviation of each branch is within the allowable range, the primary slave controller sends the heat flux density distribution details of each branch to the corresponding secondary slave controller, proceeding to the tertiary judgment.
[0064] Level 3 Judgment: After receiving the heat flux density distribution details sent by the Level 1 slave controller, each Level 2 slave controller calculates the heat flux density value of each microchannel region in the Level 2 manifold, i.e., the heat flux density data of the sensor node corresponding to each microchannel. The Level 2 slave controller determines whether the deviation of the heat flux density value of each microchannel region from the target value exceeds a third preset threshold. The third preset threshold is set according to the microchannel flow regulation accuracy requirements, and is set to 5% in this embodiment. If the heat flux density deviation of a certain microchannel region exceeds the third preset threshold, it indicates that there is a temperature deviation at the heat source point corresponding to that microchannel, triggering Level 3 precise compensation. The Level 2 slave controller calculates the target opening adjustment amount of each Level 3 compensation valve based on the microchannel heat flux density deviation and generates a Level 3 compensation command. The Level 3 compensation command includes the microchannel number to be compensated and the adjustment amount, and is sent to the corresponding Level 3 slave controller via the CAN bus. After receiving the compensation command, each Level 3 slave controller determines whether to perform Level 3 compensation based on the precise compensation target. If the compensation command contains adjustment amount information, precise compensation is required; if the compensation command does not contain adjustment amount, the current opening of the Level 3 compensation valve remains unchanged.
[0065] S4: Execution Control Phase Based on the hierarchical decision-making results, each controller drives the corresponding regulating valve to achieve dynamic flow distribution in the multi-stage manifold diversion structure.
[0066] After receiving the primary flow splitting strategy command from the master controller, the primary slave controller parses the target opening values of each branch in the command. The primary slave controller sends a drive signal to the primary flow splitting valve, controlling the valve spool to move to the target position. The drive signal for the primary flow splitting valve is a voltage signal between 0 and 10 volts, with the voltage value proportional to the valve opening. When the target opening value is 50%, the corresponding drive voltage is 5 volts. Simultaneously, the primary slave controller monitors the valve position feedback signal, and completes the primary flow splitting control after confirming that the valve spool has reached the target position. The response time of the primary flow splitting valve is less than 10 milliseconds.
[0067] After receiving the secondary regulation command from the primary slave controller, each secondary slave controller parses the branch number and adjustment amount in the command. The secondary slave controller calculates the target opening of the secondary control valve based on the adjustment amount: a positive adjustment increases the target opening, and a negative adjustment decreases it. The secondary slave controller sends a PWM drive signal to the corresponding secondary control valve, controlling the valve spool to move to the target position. The drive signal for the secondary control valve is a PWM pulse width modulation signal with a frequency of 1 kHz, and the duty cycle is proportional to the valve opening. A target opening of 50% corresponds to a 50% duty cycle. Simultaneously, the secondary slave controller monitors the valve position feedback signal and completes the secondary branch regulation after confirming that the valve spool has reached the target position. The response time of the secondary control valve is less than 50 milliseconds.
[0068] After receiving the level 3 compensation command from the level 2 slave controller, each level 3 slave controller parses the microchannel number and adjustment amount in the command. The level 3 slave controller calculates the target opening of the level 3 compensation valve based on the adjustment amount. The level 3 slave controller sends a high-speed PWM drive signal to the corresponding level 3 compensation valve, controlling the valve core to move to the target position. The drive signal for the level 3 compensation valve is a high-speed PWM pulse width modulation signal with a frequency of 10 kHz, and its duty cycle is proportional to the valve opening. Simultaneously, the level 3 slave controller monitors the valve position feedback signal, and completes the level 3 precise compensation after confirming that the valve core has reached the target position. The response time of the level 3 compensation valve is less than 100 milliseconds.
[0069] After the control operation is completed, each slave controller feeds back the results to the master controller via the CAN bus. The master controller records the actual opening value and response time of each valve for subsequent evaluation of the control effect.
[0070] It should be noted that in this embodiment, the response times of the control valves at each stage meet a preset timing coordination relationship: the first-stage diversion valve has the shortest response time (less than 10 milliseconds in this embodiment), the second-stage control valve has the next shortest response time (less than 50 milliseconds in this embodiment), and the third-stage compensation valve has the longest response time (less than 100 milliseconds in this embodiment). The design principle of this timing coordination relationship is as follows: the first-stage diversion valve is responsible for macroscopic flow guidance based on the direction angle of the heat flux density gradient, and needs to respond quickly when the heat flux density gradient changes abruptly, guiding the mainstream of the cooling medium to the high heat flux density region, thus requiring the highest response speed; the second-stage control valve is responsible for fine adjustment at the branch level, with a relatively small action range, thus requiring the next fastest response speed; the third-stage compensation valve is responsible for precise compensation at the microchannel level, with the smallest adjustment range, and needs to be fine-tuned after the second-stage adjustment has stabilized, thus requiring a relatively low response speed. Through this 'fast macroscopic, slow microscopic' timing coordination, it is ensured that the actions of each valve do not interfere with each other, forming a stable cascaded control response chain, avoiding flow oscillations caused by the simultaneous action of each valve.
[0071] S5: Feedback Optimization Phase At the end of each control cycle, the main controller collects feedback data from the flow sensor and temperature sensor, evaluates the control effect, updates the parameters of the heat load prediction model, and achieves adaptive optimization of the control parameters.
[0072] The main controller collects data from the inlet flow sensor to obtain the current total flow rate of the cooling medium; it also collects data from the flow sensors of each branch to obtain the current flow rate of the cooling medium in each branch; it calculates the deviation between the actual flow rate and the target flow rate of each branch, and if the deviation exceeds the allowable range, it records the deviation value for subsequent valve adjustment optimization.
[0073] The main controller collects data from the inlet temperature sensor to obtain the current inlet temperature of the cooling medium; it also collects data from each outlet temperature sensor to obtain the current outlet temperature of the cooling medium in each branch; it calculates the deviation between the outlet temperature of each branch and the target temperature, and if the deviation exceeds the allowable range, it records the deviation value for subsequent control strategy adjustments.
[0074] The main controller evaluates the control effect: based on heat flux density distribution data and temperature distribution data, it calculates the temperature uniformity index of the heat source surface, defined as the difference between the highest and lowest temperatures on the heat source surface. The control effect evaluation results are used to determine whether control parameters need to be adjusted.
[0075] The main controller updates the parameters of the heat load prediction model: the heat flux density data of the current control cycle is added to the historical data sequence, and the data of the past 10 control cycles is maintained by using a sliding window method; the historical data sequence is input into the long short-term memory network model to predict the heat flux density distribution of the next control cycle; the flow diversion strategy of the next control cycle is adjusted according to the prediction results to achieve feedforward control.
[0076] The main controller performs adaptive tuning of control parameters: based on historical control results, an adaptive control algorithm is used to adjust the thresholds and adjustment coefficients at each level. The first, second, and third preset thresholds are dynamically adjusted according to the actual control effect, enabling the system to adapt to different thermal load conditions.
[0077] The main controller determines whether the system needs to enter a special operating mode: if the heat flux density gradient changes drastically and exceeds the alarm threshold, the system automatically switches to gradient warning mode, executes a rapid response strategy, and prioritizes system stability; if the local temperature exceeds the alarm threshold, the system automatically switches to hot spot alarm mode, and prioritizes cooling of the hot spot area; if the heat load remains below the threshold, the system automatically switches to energy-saving mode, reduces the speed of the circulating pump to save energy. In energy-saving mode, the main controller adjusts the opening setpoint of each level of valve proportionally according to the change in the speed of the circulating pump, keeping the flow distribution ratio of each branch unchanged.
[0078] The above steps are repeated in each control cycle to achieve continuous dynamic distribution and adaptive optimization of cooling medium flow.
[0079] Example 2: The main difference between this example and Example 1 lies in the arrangement of the distributed heat flux density sensor array and the type of sensor.
[0080] In this embodiment, the heat source surface is the CPU chip inside the data center server, with dimensions of 25 mm by 25 mm. The distributed heat flux density sensor array uses 9 sensor nodes arranged in a 3x3 matrix, with an 8 mm spacing between adjacent nodes. The sensor nodes are thermopile-type heat flux density sensors with a response time of 2 milliseconds and a measurement range of 0.01 to 1000 kW / m². The sensor nodes are attached to the contact surface between the CPU chip surface and the heat sink via a flexible circuit board.
[0081] In this embodiment, due to the small area of the heat source, the gradient calculation uses the five-point difference method instead of the central difference method to improve the calculation accuracy of the boundary region. Simultaneously, the heat load prediction model uses gated recurrent units instead of long short-term memory networks to reduce the computational load. The gated recurrent unit model contains two gated recurrent unit layers, each containing 64 hidden units. Other steps are the same as in Embodiment 1.
[0082] Example 3: The main difference between this example and Example 1 lies in the threshold triggering logic and control parameters of the hierarchical decision-making stage.
[0083] This embodiment is designed for cooling applications of high-power laser arrays, where the dynamic range and rate of change of heat flux density are large. The first preset threshold is set to 200 kW / m³, the second to 15%, and the third to 8%. The first-stage shunt valve employs a piezoelectric high-speed proportional valve with a response time requirement of less than 10 milliseconds. The driving signal is a 0-10 volt voltage signal, and its response speed is higher than that of the piezoelectric high-speed proportional valve configuration in Embodiment 1, to accommodate the rapid changes in heat flux density of high-power lasers. The second-stage regulating valve employs a high-speed proportional solenoid valve with a response time requirement of less than 50 milliseconds. The third-stage compensation valve employs a miniature solenoid valve with a response time requirement of less than 100 milliseconds.
[0084] In this embodiment, the timing coordination of the valves at each stage is the same as in Embodiment 1, still satisfying the principle of "fast macroscopic response and slow microscopic response," where the first-stage diverter valve responds the fastest and the third-stage compensation valve responds the slowest. The first-stage diverter valve uses a piezoelectric high-speed proportional valve to achieve a rapid response within 10 milliseconds, ensuring that the mainstream cooling medium is quickly directed to the high heat flux density region when the heat flux density gradient changes abruptly. The second-stage regulating valve has a response time of less than 50 milliseconds, performing fine adjustment of the branch level after the macroscopic guidance is completed. The third-stage compensation valve has a response time of less than 100 milliseconds, performing localized and precise compensation of the microchannel level after the adjustment of each branch has stabilized. The valves at each stage operate sequentially to avoid flow oscillation.
[0085] In this embodiment, after the first-level judgment is triggered, when the main controller calculates the flow distribution direction based on the heat flux density gradient direction angle, an additional inertial filtering stage is introduced to avoid frequent operation of the first-level diversion valve caused by high-frequency heat flux fluctuations. The time constant of the inertial filter is set to 5 control cycles.
[0086] In this embodiment, the time series prediction algorithm in the feedback optimization stage uses an autoregressive moving average model with exogenous variables instead of a long short-term memory network to adapt to embedded environments with limited computing resources. Other steps are the same as in Embodiment 1.
[0087] Example 4: The main difference between this example and Example 1 is the number of levels in the multi-stage manifold split structure.
[0088] In this embodiment, the heat source is a large battery module containing 32 series-connected battery cells, with dimensions of 1200 mm x 800 mm. The multi-stage manifold distribution structure is configured as a four-stage distribution: the primary manifold distributes the cooling medium to four secondary manifolds, each secondary manifold distributes the cooling medium to four tertiary manifolds, and each tertiary manifold distributes the cooling medium to four quaternary microchannels. The corresponding slave controllers are configured at four levels: primary slave controller, secondary slave controller, tertiary slave controller, and quaternary slave controller.
[0089] The response times of valves at each stage still follow the principle of "fast macroscopic, slow microscopic": the first-stage diverter valve has the fastest response time, set to less than 10 milliseconds, and uses a piezoelectric high-speed proportional valve to guide the global flow macroscopically based on the direction angle of the heat flux density gradient; the second-stage regulating valve has the next fastest response time, set to less than 50 milliseconds, and is responsible for fine adjustment at the second-stage manifold branch level; the third-stage regulating valve has the next slowest response time, set to less than 100 milliseconds, and is responsible for further adjustment at the third-stage manifold branch level; the fourth-stage compensation valve has the slowest response time, set to less than 200 milliseconds, and is responsible for precise compensation at the fourth-stage microchannel level. The fourth-stage compensation valve has the smallest adjustment range and requires fine-tuning after the first three stages of adjustment have stabilized, thus requiring the lowest response speed.
[0090] The hierarchical decision-making stage is expanded to four levels of judgment: Level 1 judgment handles the global heat flux density gradient. If the maximum value of the heat flux density gradient exceeds the first preset threshold (set to 80 kW / m³ in this embodiment), a Level 1 response is triggered, adjusting the Level 1 diverter valve; otherwise, it proceeds to Level 2 judgment. Level 2 judgment handles the deviation of the Level 2 manifold branches. If the average deviation of the heat flux density of a certain branch exceeds the second preset threshold (set to 15% in this embodiment), a Level 2 response is triggered, adjusting the corresponding Level 2 regulating valve; otherwise, it proceeds to Level 3 judgment. Level 3 judgment handles the deviation of the Level 3 manifold branches. If the deviation exceeds the third preset threshold (set to 10% in this embodiment), a Level 3 response is triggered, adjusting the corresponding Level 3 regulating valve; otherwise, it proceeds to Level 4 judgment. Level 4 judgment handles the precise compensation of the Level 4 microchannels. If the deviation exceeds the fourth preset threshold (set to 5% in this embodiment), a Level 4 response is triggered, adjusting the corresponding Level 4 compensation valve.
[0091] This embodiment demonstrates that the hierarchical control architecture of the present invention can be flexibly expanded to four or more levels according to the scale of the heat source and cooling requirements. The response time of each level always maintains a "fast macroscopic, slow microscopic" timing relationship: the higher the level (closer to the global response), the faster the response; the lower the level (closer to the microchannel response), the slower the response. This progressively slower timing ensures that the valve actions at each level do not interfere with each other, forming a stable cascaded control response chain. Other steps are similar to those in Embodiment 1 and will not be repeated.
[0092] The above are merely preferred embodiments of the present invention and are not intended to limit the present invention in any other way. Any person skilled in the art may make changes or modifications to the above-disclosed technical content to create equivalent embodiments that can be applied to other fields. However, any simple modifications, equivalent changes, and modifications made to the above embodiments based on the technical essence of the present invention without departing from the scope of the present invention shall still fall within the protection scope of the present invention.
Claims
1. A manifold split cooling method based on heat flux perception, characterized in that, Includes the following steps: S1: Initialization phase. After the system is powered on, the distributed heat flux density sensor array is calibrated, the regulating valves at each level in the multi-stage manifold diversion structure are reset, and the initial heat flux density reference value is read. S2: Sensing and acquisition phase, distributed heat flux density sensor array synchronously acquires heat flux density data on the surface of heat source, constructs spatial heat flux density distribution map, and calculates heat flux density gradient; S3: The hierarchical decision-making stage, based on heat flux density distribution and gradient information, executes hierarchical decisions of multi-level slave controllers, including first-level global flow splitting strategy calculation, second-level fine-tuning of branches, and third-level precise compensation; S4: Execution control stage, based on the hierarchical decision results, drives the first-stage diversion valve, the second-stage regulating valve and the third-stage compensation valve to achieve dynamic flow distribution of the multi-stage manifold diversion structure; S5: Feedback optimization phase, collect feedback data from flow sensor and temperature sensor, evaluate control effect, update heat load prediction model parameters, and achieve adaptive tuning of control parameters.
2. The heat flux aware manifold split cooling method of claim 1, wherein, Step S3, the hierarchical decision-making stage, includes: S31: First-level judgment: The main controller judges whether the heat flux density gradient exceeds the preset threshold. If it does, it triggers the first-level response and directly adjusts the flow ratio of the first-level flow divider valve. If it does not exceed the threshold, it proceeds to step S32. S32: Secondary judgment. The primary judgment is based on whether the heat flux density deviation of each branch exceeds the preset threshold. If it exceeds the threshold, the secondary fine adjustment is triggered to adjust the opening of the secondary regulating valve of the corresponding branch. If it does not exceed the threshold, proceed to step S33. S33: Level 3 judgment. The Level 2 controller determines whether Level 3 compensation needs to be performed based on the precision compensation target. If so, Level 3 precision compensation is triggered, driving the Level 3 compensation valve to make fine adjustments.
3. The heat flux aware manifold split cooling method of claim 2, wherein, In step S31, the first-level response calculates the flow distribution direction that needs to be adjusted based on the direction angle of the heat flux density gradient, and directs the cooling medium to the high heat flux density region.
4. The manifold split cooling method based on heat flux density sensing according to claim 2, characterized in that, In step S32, the secondary fine adjustment calculates the adjustment amount of the secondary regulating valve opening of the corresponding branch based on the direction and magnitude of the heat flux density deviation, and drives the secondary regulating valve to act through the PWM signal.
5. The manifold split cooling method based on heat flux density sensing according to claim 2, characterized in that, In step S33, the three-stage precise compensation is achieved by driving the three-stage compensation valve through a high-speed PWM signal, thereby realizing millisecond-level flow compensation control.
6. The manifold split cooling method based on heat flux density sensing according to claim 1, characterized in that, Step S2, which involves calculating the heat flux density gradient, includes calculating the first-order partial derivatives of the heat flux density along the X and Y directions of the heat source surface, and calculating the magnitude and direction angle of the heat flux density gradient based on the first-order partial derivatives.
7. The manifold split cooling method based on heat flux density sensing according to claim 1, characterized in that, In the execution control stage described in step S4, the response times of each level of regulating valve meet the preset timing coordination relationship, with the first-level diversion valve responding the fastest, followed by the second-level regulating valve, and the third-level compensation valve responding the slowest.
8. A manifold split cooling system based on heat flux density sensing, characterized in that, include: A distributed heat flux density sensor array is used to synchronously acquire heat flux density data of the heat source surface; The main controller communicates with the distributed heat flux density sensor array to construct a spatial heat flux density distribution map and calculate the heat flux density gradient, and executes a first-level global flow splitting strategy calculation. Multi-level slave controllers, including level 1, level 2, and level 3 slave controllers, are used to perform level 2 branch fine adjustment and level 3 precise compensation in hierarchical decision-making. The multi-stage manifold distribution structure includes a primary manifold, a secondary manifold, and a tertiary microchannel, as well as a primary distribution valve installed in the primary manifold, a secondary regulating valve installed in the secondary manifold, and a tertiary compensation valve installed in the tertiary microchannel, used to achieve dynamic distribution of cooling medium flow. Flow sensors and temperature sensors are used to collect feedback data.
9. The manifold split cooling system based on heat flux density sensing according to claim 8, characterized in that, The distributed heat flux density sensor array consists of several heat flux density sensor nodes, each of which is arranged at a key position on the surface of the heat source.
10. The manifold split cooling system based on heat flux density sensing according to claim 8, characterized in that, The main controller integrates a heat load prediction model, which is used to predict the heat flux density distribution at future times based on historical heat flux density data.