Intelligent temperature control system and method for submersion liquid-cooled mainframe

Abstract

The application relates to the technical field of host heat dissipation, and discloses an intelligent temperature control system and method of an immersed liquid-cooled host case. The system is integrated in an interface module cabin of the immersed liquid-cooled host case and comprises a sensing layer, a calculation layer, a control layer and a man-machine interaction module; the sensing layer adopts a multi-dimensional sensor combination and a standardized collection process to realize accurate data collection; the calculation layer adopts quantitative algorithms such as real-time heat generation calculation, linear regression trend prediction, segmented adjustment and dynamic correction to output accurate control instructions; and the control layer executes heat dissipation actions through closed-loop adjustment and multiple protection mechanisms. The method comprises the steps of initialization, data collection, quantitative calculation, trend prediction, dynamic adjustment and exception handling. Through the whole-process optimization of 'collection-calculation-control', the application ensures the accuracy, stability and reliability of the temperature control system, and solves the technical problems of uncontrollable heat dissipation, response lag and insufficient stability of traditional liquid-cooled cases.

Description

Technical Field

[0001] This invention relates to the field of host computer heat dissipation technology, specifically to an intelligent temperature control system and method for an immersion liquid-cooled host computer chassis. Background Technology

[0002] As the computing power demands of core computer components such as CPUs and GPUs continue to grow, their power consumption and heat generation have increased significantly, making traditional air cooling insufficient to meet the requirements for efficient heat dissipation. Immersion liquid cooling technology, due to the high thermal conductivity and specific heat capacity of insulating coolants, can quickly transfer heat through direct heat exchange, making it the preferred heat dissipation solution for high-performance computing devices.

[0003] A modular immersion liquid-cooled mainframe chassis already exists in the technology, which improves maintenance convenience by integrating the interface module compartment and cooling system on the top of the rack. However, this technology still has the following key technical defects:

[0004] The heat dissipation control method is simplistic, relying solely on manual switches to control the pump and fan's start and stop. This fails to dynamically adjust the heat dissipation intensity based on the hardware's heat generation status, resulting in a mismatch between heat dissipation efficiency and energy consumption—excessive heat dissipation at low loads leads to energy waste, while insufficient heat dissipation at high loads may cause hardware overheating.

[0005] Lacking scientific quantitative algorithms, relying solely on simple temperature threshold triggering for adjustment, and lacking the ability to predict heat generation trends, it is prone to heat dissipation lag, leading to excessive fluctuations in hardware temperature.

[0006] The data acquisition method is rudimentary, without the use of multi-dimensional sensor combinations, resulting in insufficient data accuracy and comprehensiveness. Furthermore, it lacks a sound closed-loop adjustment and protection mechanism, making it difficult to guarantee system stability and reliability.

[0007] The lack of visual monitoring of temperature and heat dissipation status makes it difficult for users to intuitively grasp the operating status of the equipment, making troubleshooting more difficult.

[0008] Therefore, designing an intelligent temperature control system and method with accurate data acquisition, scientific algorithms, and stable and reliable operation, based on the existing modular immersion liquid-cooled mainframe, has become the key to solving the above-mentioned technical problems. Summary of the Invention

[0009] This invention aims to provide an intelligent temperature control system for an immersion liquid-cooled mainframe chassis. By optimizing data acquisition methods, constructing scientific quantitative algorithms, and designing multiple stabilization mechanisms, it achieves heat generation prediction and precise heat dissipation adjustment. While retaining the original modular maintenance convenience, it improves heat dissipation efficiency and ensures stable and reliable system operation.

[0010] To achieve the above objectives, the present invention adopts the following technical solution.

[0011] An intelligent temperature control system for an immersion liquid-cooled mainframe chassis, the immersion liquid-cooled mainframe chassis including a chassis (1), a frame (2) and a cover plate (3), the top of the chassis (1) being open and containing insulating coolant, the frame (2) being detachably inserted into the chassis (1), the lower mounting bracket (4) supporting the motherboard and power supply, the top being fixed with an interface module compartment (5), and the cover plate (3) being detachably connected to the top of the interface module compartment (5), characterized in that the intelligent temperature control system is integrated within the interface module compartment (5), including:

[0012] The sensing layer employs a multi-dimensional sensor combination and a standardized acquisition process to collect data on the temperature of the core heating element, the temperature of the coolant, the coolant status, and the hardware operating status. The acquisition frequency is 1 time / second, and the detection accuracy error is ≤±0.5℃. The multi-dimensional sensor combination includes a temperature sensor group, a current sensor, an SMBus communication module, and a coolant status monitoring unit.

[0013] The computing layer is electrically connected to the sensing layer via a transfer harness. It uses a real-time heat generation calculation algorithm, a linear regression trend prediction algorithm, a piecewise adjustment algorithm, and a dynamic correction algorithm to calculate the real-time heat generation and predict the future heat generation trend. It outputs speed control commands for the pump (8) and the cooling module (7). In the linear regression trend prediction algorithm, the load scenario correction coefficient λ is set according to the load type. For office scenarios, λ=0.9, for game scenarios, λ=1.1, and for rendering scenarios, λ=1.2. It also supports calibration according to the actual application scenario.

[0014] The control layer is electrically connected to the computing layer, pump (8) and refrigeration module (7) respectively through the adapter harness. It receives control commands and adjusts the coolant circulation speed and fan speed. The control layer is equipped with a closed-loop regulation mechanism, multiple protection mechanisms and power supply switching mechanism to ensure operational stability.

[0015] The human-computer interaction module includes a local interaction unit and a remote interaction unit, which supports visualization of operating status, customization of parameters, and fault warning.

[0016] The computing layer is also electrically connected to the motherboard of the host chassis via a junction box to obtain hardware operating parameters and provide feedback on system status.

[0017] Preferably, the multi-dimensional sensor combination of the perception layer specifically includes:

[0018] (1) The temperature sensor group adopts a waterproof and corrosion-resistant digital temperature sensor (9), including a core component temperature sensor and a coolant temperature sensor; two core component temperature sensors are attached to each of the CPU and GPU core heat dissipation surfaces, and one is attached to each of the motherboard power supply module and hard disk shell; one coolant temperature sensor is fixed in the coolant inlet, outlet and middle area of ​​the enclosure (1), and all are encapsulated in a corrosion-resistant shell; when the difference between the two sensor data of the same core component exceeds 1.5℃, an anomaly mark is activated and the historical average value is used as a replacement.

[0019] (2) The current sensor adopts the ACS712 miniature current sensor (16), which is connected in series on the power supply line from the power supply output to the PCIe expansion card to collect the real-time power supply current of the expansion card;

[0020] (3) The SMBus communication module is integrated into the intelligent control motherboard (10) of the computing layer. It establishes stable communication with the motherboard through the SMBus interface and reads the CPU / GPU operating frequency and load rate data in real time.

[0021] (4) The coolant status monitoring unit includes a liquid level sensor and a dielectric constant sensor, which are used to detect the coolant level height and dielectric constant change, respectively. When the liquid level is lower than the preset threshold or the dielectric constant exceeds the normal range, an alarm signal is triggered.

[0022] Preferably, the closed-loop regulation mechanism of the control layer includes:

[0023] (1) Speed ​​feedback acquisition: The pump drive module (11) integrates Hall speed detection circuit, and the fan adopts 4-wire PWM fan. The actual speed is collected in real time and fed back to the calculation layer. The feedback frequency is ≥1 time / second.

[0024] (2) PWM duty cycle correction: The calculation layer adjusts the output signal through a closed-loop correction formula;

[0025] The corrected formula for the pump is ,in ;

[0026] The correction formula for the fan is: ,in ;

[0027] (3) Adjustment accuracy guarantee: Through continuous feedback correction, ensure that the error between the actual speed of the pump and the target speed is ≤3%.

[0028] Preferably, the multiple protection mechanisms of the control layer include:

[0029] (1) Actuator fault protection: If the pump has no speed feedback for 3 consecutive seconds, the fan will start at the highest speed for cooling; if the fan has no speed feedback for 3 consecutive seconds, the pump will start at the highest speed for cooling; if both fail, an emergency warning will be sent to the operating system.

[0030] (2) Overload protection: The pump and fan speeds, after correction, shall not exceed 100% of the rated speed to avoid hardware overload damage;

[0031] (3) Temperature over-limit protection: When the temperature of the core component is ≥85℃ or the temperature of the coolant is ≥45℃, the highest speed heat dissipation will be started immediately and an alarm will be triggered. If the temperature does not drop after 30 seconds, the user will be prompted to reduce the load.

[0032] (4) Coolant abnormal protection: When the coolant level is lower than the preset threshold or the dielectric constant exceeds the normal range, a graded warning will be activated, and in case of serious abnormality, the power supply to unnecessary hardware will be automatically cut off.

[0033] Preferably, the power supply switching mechanism of the control layer includes: adopting a dual power supply design, with the main power supply taken from the 12V output terminal of the host power supply and the backup power supply taken from the power supply interface of the interface panel. The two are isolated by diodes, and the input terminal is equipped with a filter capacitor (17) and a voltage regulator chip (18). When the main power supply voltage fluctuates beyond 12V±5%, it switches to the backup power supply within ≤10ms. The backup power supply capacity meets the system's full-load operation requirements.

[0034] Preferably, the quantization algorithm used in the computation layer specifically includes:

[0035] (1) Real-time calorific value calculation algorithm:

[0036] CPU / GPU heat generation Where k is the power consumption coefficient, with the CPU's power consumption coefficient ranging from 0.8 to 1.0 and the GPU's power consumption coefficient ranging from 0.9 to 1.1. Rated thermal design power (W), f_actual is the real-time operating frequency (GHz), f_rated is the rated frequency (GHz), and Load is the real-time load rate (0-100%).

[0037] Coolant heat absorption power , where c is the specific heat capacity of the coolant (J / (kg·℃)), ρ is the density of the coolant (kg / m³), V is the volume of the coolant (m³), ΔT is the temperature change in 1 second (℃), and t=1s;

[0038] (2) Linear regression trend prediction algorithm:

[0039] Based on the heat generation data of the most recent 30 minutes, a trend line was fitted using the least squares method. ,in , n=1800 is the sample size;

[0040] Combined with the load scenario correction coefficient λ, we obtain Then convert it to core temperature , where R is the thermal resistance (°C / W) and S is the heat dissipation surface area (m²).

[0041] The rules for determining the value of the load scenario correction factor λ are as follows: λ=0.9 for office scenarios, λ=1.1 for game scenarios, and λ=1.2 for rendering scenarios;

[0042] (3) Segmented adjustment algorithm: Based on the real-time temperature T and the predicted temperature T for the next 5 minutes (300), the target rotational speed is output through a segmented function;

[0043] (4) Dynamic correction algorithm: If the temperature difference between the inlet and outlet of the coolant ,but , The corrected rotational speed shall not exceed 100% of the rated rotational speed; the 5°C threshold is determined based on the physical properties of the coolant and the heat dissipation requirements of the hardware.

[0044] Preferably, the piecewise function satisfies:

[0045] when and At that time, the pump speed is 30% of the rated speed, and the fan speed is 20% of the rated speed;

[0046] when or At that time, the pump speed is 50% of the rated speed, and the fan speed is 40% of the rated speed;

[0047] when or At that time, the pump speed is 80% of the rated speed, and the fan speed is 70% of the rated speed;

[0048] when or At that time, both the pump speed and the fan speed are 100% of their rated speed;

[0049] The execution priority of the dynamic correction algorithm is lower than that of the segmented adjustment algorithm. That is, the initial target speed is determined first by the segmented adjustment algorithm, and then fine-tuned by the dynamic correction algorithm.

[0050] Preferably, it also includes a human-computer interaction module, the human-computer interaction module comprising:

[0051] (1) Local interaction unit, located on the switch panel (502) of the interface module compartment (5), including OLED temperature display window (13), mode switching button (14) and red / yellow dual-color alarm indicator (15); OLED display screen (13) displays various temperature, speed and coolant status data in a cycle; mode switching button (14) supports automatic / manual mode switching; alarm indicator (15) is triggered when the temperature exceeds the threshold, sensor fails or coolant is abnormal, red indicates serious fault and yellow indicates general warning;

[0052] (2) Remote interaction unit, which communicates with the motherboard via the USB / serial port of the intelligent control motherboard (10), supports the installation of monitoring software in the operating system; the monitoring software supports temperature curve display, temperature threshold customization, heat dissipation strategy selection, historical data export (supports data of the last 72 hours), fault log viewing and multi-device network monitoring, and has permission management function, and supports pushing abnormal information to associated terminals.

[0053] Based on the above-mentioned intelligent temperature control system, the present invention also provides an intelligent temperature control method for an immersion liquid-cooled mainframe chassis, comprising the following steps:

[0054] S1: Initialization. After the system is powered on, it performs a self-test on the sensing layer sensors, control layer pumps and fans, and coolant status monitoring unit. After the self-test passes, the pumps and fans start at the lowest speed and enter automatic mode. At the same time, the data storage module is initialized and a cyclic overwrite storage strategy is configured (stores the most recent 72 hours of data, and the storage medium capacity is ≥100MB).

[0055] S2: Data acquisition. The sensing layer collects core component temperature, coolant temperature, coolant status, CPU / GPU load and frequency, and PCIe expansion card power supply current data at a frequency of 1 time / second. After range verification (temperature -20℃~120℃, load rate 0%~100%, etc.) and consistency verification (three consecutive data changes exceeding the threshold are marked as abnormal), the data is transmitted to the computing layer.

[0056] S3: Real-time heat generation calculation. The calculation layer calculates the real-time heat generation of the CPU / GPU and the heat absorption power of the coolant through a real-time heat generation calculation algorithm.

[0057] S4: Heat generation trend prediction. The calculation layer fits a trend line based on the data of the last 30 minutes using a linear regression algorithm, and calculates the core temperature for the next 5 minutes by combining the load scenario correction coefficient λ. If there are abnormal data, the corrected valid data is used for trend fitting.

[0058] S5: Target speed is determined. Based on real-time and predicted temperatures, the calculation layer first outputs the initial target speed of the pump and fan through a segmented adjustment algorithm, and then adjusts the target speed through a dynamic correction algorithm in combination with the temperature difference between the inlet and outlet of the coolant.

[0059] S6: Dynamic adjustment. The control layer receives the PWM signal from the calculation layer to drive the pump and fan. Speed ​​accuracy is ensured through speed feedback and closed-loop correction formula. At the same time, the power supply status is monitored in real time. When the main power supply is abnormal, the power supply switching mechanism is triggered.

[0060] S7: Abnormal handling. If the temperature exceeds the threshold, the highest speed cooling will be activated and an alarm will be triggered. If the sensor fails, the emergency cooling mode will be activated and an alarm will be triggered, and the fault log will be recorded. If the load changes abruptly (load change rate > 50% / s) or the temperature prediction error exceeds 3℃, the trend fitting will be skipped, and the target speed will be calculated directly at 1.2 times the current heat generation. If the coolant is abnormal, a graded warning and corresponding emergency measures will be activated. If the temperature does not drop for 30 seconds after exceeding the threshold, the computing layer will send a warning message to the operating system through the motherboard to prompt the user to reduce the hardware load.

[0061] Ideally, this also includes system calibration steps: calibrating the sensor accuracy every 6 months and optimizing the algorithm parameters (including power consumption coefficient, correction coefficient, and load scenario correction coefficient λ) annually to ensure the temperature control accuracy and stability of the system during long-term operation.

[0062] Compared with the prior art, the present invention has the following positive effects:

[0063] (1) The data acquisition is comprehensive and accurate. It adopts a multi-dimensional combination of "temperature sensor group + current sensor + SMBus communication module", with redundant layout and double verification. The temperature detection accuracy error is ≤ ±0.5℃ and the data validity rate is ≥99.5%, providing high-quality data support for algorithm calculation.

[0064] (2) High adjustment accuracy: Through the closed-loop adjustment mechanism, the pump and fan speed error is ≤3%, and the heat dissipation intensity can be dynamically matched according to the hardware heat dissipation status, improving heat dissipation efficiency by more than 30% and reducing energy consumption by 20%~40%;

[0065] (3) Excellent stability and reliability, multiple protection mechanisms cover scenarios such as actuator failure, power failure, and temperature over-limit, the system mean time between failures (MTBF) is ≥10,000 hours, and hardware damage can be avoided through emergency handling in case of failure;

[0066] (4) Interactive and visual: The local display screen is combined with the remote monitoring software to realize real-time visualization of the operating status and fault information, which facilitates user operation and fault diagnosis. Attached Figure Description

[0067] Figure 1 is a schematic diagram of the integrated structure of the intelligent temperature control system and the main unit of the present invention;

[0068] Figure 2 is a schematic diagram of the structure of the present invention exploded;

[0069] Figure 3 is a schematic diagram of the integration of the frame and intelligent temperature control system of the present invention;

[0070] Figure 4 is a schematic diagram of the cover plate structure of the present invention;

[0071] Figure 5 is a schematic diagram of the switch panel (including the human-computer interaction module) of the present invention;

[0072] Figure 6 is a schematic diagram of the interface panel structure of the present invention;

[0073] Figure 7 is a schematic diagram of the connection between the heat exchange module, the refrigeration module, the pump and the control layer of the present invention;

[0074] Figure 8 is a block diagram of the intelligent temperature control system architecture of the present invention;

[0075] Figure 9 is a schematic diagram of the intelligent temperature control method of the present invention;

[0076] Figure 10 is a schematic diagram of the sensor layout of the sensing layer of the present invention;

[0077] Figure 11 is a flowchart of the computational layer algorithm of the present invention;

[0078] Figure 12 is a block diagram illustrating the principle of the closed-loop regulation mechanism of the present invention.

[0079] Explanation of reference numerals in the attached figures:

[0080] 1. Housing; 11. Flange; 2. Frame; 4. Mounting rack; 5. Interface module compartment; 501. Interface opening; 502. Switch panel; 503. Interface panel; 504. Control switch; 505. Adapter; 506. Power supply interface; 507. Handle; 51. Operating section; 52. Horizontal guide rail; 53. Cover plate; 3. Slide rail; 31. Heat exchange module; 6. Refrigeration module; 7. Pump; 8. Circuit switch; 507. Temperature sensor; 9. Intelligent control motherboard; 10. Pump drive module; 11. Fan control module; 12. OLED display screen; 13. Mode switching button; 14. Alarm indicator light; 15. Current sensor; 16. Filter capacitor; 17. Voltage regulator chip; 18. Detailed Implementation

[0081] The present invention will now be further described in conjunction with the accompanying drawings and specific embodiments.

[0082] like Figure 1-4 As shown, the intelligent temperature control system of the present invention is integrated into the interface module compartment 5 of the modular immersion liquid-cooled main unit chassis. The main unit chassis is mainly composed of a chassis 1, a frame 2 and a cover plate 3.

[0083] The enclosure 1 is made of transparent acrylic sheet, with a flange 11 on the top. A sealing ring is installed on the flange 11 to ensure sealing performance. The interior of the enclosure 1 is used to contain insulating coolant, forming an immersion heat dissipation environment.

[0084] The rack 2 adopts a detachable design and can be inserted into the enclosure 1 horizontally. The mounting bracket 4 at the bottom is used to support core hardware such as the motherboard, CPU, GPU, hard drive, power supply and PCIe expansion card. The top is fixedly equipped with an interface module compartment 5. The interface module compartment 5 has an interface opening 501 on the side, a switch panel 502 on the front and an interface panel 503 on the back. The core hardware of the heat exchange module 6, cooling module 7, pump 8 and intelligent temperature control system are integrated inside. The top of the interface module compartment 5 is equipped with two symmetrically distributed handles 51 to facilitate the overall removal and maintenance of the rack 2. The side is equipped with an operating part 52 and a horizontal guide rail 53 to ensure precise docking between the rack 2 and the enclosure 1.

[0085] The cover plate 3 is made of a transparent material that matches the box body 1. Its edge is provided with a sliding groove 31, which is detachably connected to the top of the interface module compartment 5 through the sliding groove 31 to achieve a sealed cover of the box body 1.

[0086] like Figure 5-8 As shown in Figure 10, the hardware layout and connection relationship of the intelligent temperature control system are as follows.

[0087] Sensing Layer: Of the eight DS18B20 digital temperature sensors 9, two are attached to the CPU and GPU core heat sinks, one to the motherboard power supply module and one to the hard drive casing, and one is fixed to the coolant inlet, outlet, and central area. All temperature sensors are encapsulated in a corrosion-resistant casing. An ACS712 current sensor 16 is connected in series to the power supply line from the power output to the PCIe expansion card. A liquid level sensor (model: LS-200) is installed on the lower part of the inner side wall of the enclosure 1, and a dielectric constant sensor (model: DK-100) is immersed in the coolant. An SMBus communication module is integrated into the intelligent control motherboard 10 of the computing layer.

[0088] Computing layer: The intelligent control motherboard 10 is fixedly installed on the mounting bracket inside the interface module compartment 5. It is equipped with an STM32F407 microcontroller and a 16MB Flash memory chip. It is electrically connected to various sensors in the perception layer, the actuators in the control layer, and the mainboard of the host chassis through adapter cables.

[0089] Control Layer: Pump drive module 11 and fan control module 12 are integrated on intelligent control motherboard 10. Pump drive module 11 uses DRV8871 chip, is electrically connected to pump 8, and integrates Hall speed detection circuit; fan control module 12 is directly connected to the 4-wire PWM fan of cooling module 7; interface panel 503 is provided with control switch 504, adapter 505, power supply interface 506 and circuit switch 507. The dual power supply of the control layer is taken from the 12V output terminal of host power supply and power supply interface 506 respectively. The input terminal is provided with filter capacitor 17 and voltage regulator chip 18.

[0090] Human-machine interaction module: The local interaction unit is integrated on the switch panel 502, including a 1.3-inch OLED display 13, a mode switching button 14, and a red / yellow dual-color alarm indicator 15; the remote interaction unit is connected to the main board of the host chassis via a USB adapter cable to realize communication with the monitoring software.

[0091] like Figure 7 , 12 As shown, the connection logic of the core components is as follows.

[0092] Signal connection: The temperature sensor 9, current sensor 16, liquid level sensor, and dielectric constant sensor in the sensing layer transmit the collected analog or digital signals to the intelligent control motherboard 10 in the computing layer through the adapter cable bundle; the intelligent control motherboard 10 establishes communication with the host motherboard through the SMBus interface, reads the operating parameters of the CPU / GPU, and feeds back the system status; the PWM control signal output by the intelligent control motherboard 10 is transmitted to the pump 8 and the fan through the pump drive module 11 and the fan control module 12;

[0093] Power supply connection: The main power supply line is led out from the 12V output terminal of the host power supply, filtered by the filter capacitor 17 and regulated by the voltage regulator chip 18 to supply power to the intelligent control motherboard 10 and the actuator; the backup power supply line is led out from the power supply interface 506 of the interface panel 503, and supplies power after passing through the same filtering and voltage regulation circuit. The two power supplies are isolated by diodes to ensure power supply independence.

[0094] Heat dissipation connection: The heat exchange module 6 adopts a metal heat sink and is connected to the pump 8 through a pipeline to form a coolant circulation loop. The coolant flows through the heat exchange module 6 under the drive of the pump 8 and exchanges heat with the fan of the cooling module 7 to achieve rapid heat dissipation.

[0095] The intelligent temperature control system of the present invention is integrated in the interface module compartment 5 of the above-mentioned modular immersion liquid-cooled main unit chassis. The core parameters of the main unit chassis and system are configured as follows.

[0096] Coolant parameters: Mineral oil-based insulating coolant or electronic fluorinated liquid is used, with specific heat capacity c=2200J / (kg·℃), density ρ=880kg / m³, and coolant volume V=0.015m³;

[0097] Core hardware parameters: CPU (TDP=120W, rated frequency f_rated=3.6GHz, thermal resistance R=0.25℃ / W, heat dissipation area S=0.008m²); GPU (TDP=200W, rated frequency f_rated=1.8GHz, thermal resistance R=0.2℃ / W, heat dissipation area S=0.01m²); Cooling module 7 is a 4-wire PWM fan with a rated speed of 2000r / min; Pump 8 is a variable speed DC pump with a rated speed of 3000r / min;

[0098] System performance parameters: The data acquisition frequency of the sensing layer is 1 time / second, and the temperature detection accuracy error is ≤ ±0.5℃; the pump and fan speed error of the control layer is ≤3%, and the main power supply switching response time is ≤10ms; the remote monitoring software of the human-machine interaction module supports Windows 10 and above, and Linux Ubuntu 20.04 and above operating systems.

[0099]

Example 1

[0100] This embodiment is used to explain the application of the present invention in low-load office scenarios.

[0101] (a) Scene parameters

[0102] The host computer runs office software such as Word, Excel, and browser, which is a low-load steady-state scenario. The load scenario correction factor λ=0.9. The system initially enters automatic mode, with the pump initial speed at 30% of the rated speed (900r / min) and the fan initial speed at 20% of the rated speed (400r / min).

[0103] (II) Operation Process

[0104] 1. Initialization and Data Acquisition:

[0105] After the system is powered on, the intelligent control motherboard 10 self-tests various sensors in the sensing layer, pump 8, fan, and coolant status monitoring unit in the control layer. After the self-test is passed, the system starts. The sensing layer collects data at a frequency of 1 time / second: CPU load 32%, real-time frequency f_actual = 2.5GHz, GPU load 18%, real-time frequency f_actual = 1.1GHz, CPU dual sensor temperatures are 47.6℃ and 48.1℃ (average 47.8℃), GPU dual sensor temperatures are 51.3℃ and 51.8℃ (average 51.5℃), coolant inlet temperature is 27.5℃, outlet temperature is 29.2℃, middle temperature is 28.3℃, coolant level is 12cm (preset threshold 10cm), dielectric constant is 2.3 (normal range 2.0-2.5), and PCIe expansion card power supply current is 3.2A. The data is valid after range verification (temperature -20℃~120℃, load rate 0%~100%) and consistency verification (data change amplitude ≤5% for 3 consecutive times).

[0106] 2. Real-time heat generation calculation:

[0107] CPU heat output:

[0108] ;

[0109] GPU heat output:

[0110] ;

[0111] Coolant heat absorption power:

[0112] ;

[0113] Because ΔT is small under low load, the heat absorption power is mainly matched to the heat generation of the core components.

[0114] 3. Fever Trend Prediction:

[0115] The computational layer predicts trends by fitting a trend line using the least squares method based on the calorific value data from the most recent 30 minutes (1800 sets). (where a=0.002, b=21.5), combined with λ=0.9, we get Calculate the core temperature in the next 5 minutes (t=300s):

[0116] ;

[0117] .

[0118] 4. Determining the target rotational speed:

[0119] because , , , The segmented adjustment algorithm outputs the initial target speed: 30% of the pump's rated speed (900 r / min) and 20% of the fan's rated speed (400 r / min); the temperature difference between the coolant inlet and outlet, ΔT_liquid, is 29.2 - 27.5 = 1.7℃ < 5℃, so no dynamic correction is required.

[0120] 5. Dynamic adjustment:

[0121] The control layer receives PWM signals to drive the pump and fan to run. The pump drive module 11 acquires the actual speed of 893 r / min through the Hall speed detection circuit, and the fan feeds back the actual speed of 394 r / min through 4-wire PWM. The calculation layer adjusts the speed through a closed-loop correction formula.

[0122] Pump PWM duty cycle:

[0123] Fan PWM duty cycle: ;

[0124] After correction, the actual pump speed is 897 r / min (error 0.33%) and the actual fan speed is 398 r / min (error 0.5%), meeting the requirement of speed error ≤3%; at the same time, the power supply monitoring shows that the main power supply voltage is 12.1V (12V±5%), and there is no need to switch to the backup power supply.

[0125] 6. Steady-state operation performance:

[0126] After running continuously for 2 hours, the system dynamically adjusts once per second, with the CPU temperature stabilizing at 47℃~49℃, the GPU temperature at 51℃~53℃, and the coolant temperature at 27℃~29.5℃, and the fan speed not fluctuating frequently. The local OLED display displays the temperature, fan speed, and coolant status data in a loop, and the remote monitoring software generates a temperature trend curve. No alarm signals are triggered, and the heat dissipation efficiency and energy consumption are well matched.

[0127]

Example 2

[0128] This embodiment is used to explain the application of the present invention in high-load game scenarios.

[0129] (a) Scene parameters

[0130] The host is running a large-scale AAA game, which is a high-load dynamic scene. The load scene correction factor λ=1.1. The system is initially in automatic mode, with the pump speed at 30% of the rated speed (900r / min) and the fan speed at 20% of the rated speed (400r / min).

[0131] (II) Operation Process

[0132] 1. Data Collection:

[0133] Data collected by the perception layer: CPU load 93%, real-time frequency factual = 3.5GHz; GPU load 97%, real-time frequency factual = 1.72GHz; CPU dual sensor temperatures are 71.2℃ and 71.8℃ (average 71.5℃); GPU dual sensor temperatures are 76.3℃ and 76.9℃ (average 76.6℃); coolant inlet temperature 34.5℃, outlet temperature 38.8℃, middle temperature 36.6℃; coolant level 11.8cm, dielectric constant 2.4; PCIe expansion card power supply current 9.8A; data verification is valid.

[0134] 2. Real-time heat generation calculation:

[0135] CPU heat output:

[0136] ;

[0137] GPU heat output:

[0138] ;

[0139] Coolant heat absorption power:

[0140] .

[0141] 3. Fever Trend Prediction:

[0142] A trend line was fitted based on the data from the most recent 30 minutes. Combining λ=1.1, we get Calculate the core temperature for the next 5 minutes:

[0143] ;

[0144] .

[0145] 4. Determining the target rotational speed:

[0146] because , , , The segmented adjustment algorithm outputs the initial target speed: pump 80% rated speed (2400 r / min), fan 70% rated speed (1400 r / min); the coolant inlet and outlet temperature difference ΔT_liquid = 38.8 - 34.5 = 4.3℃ < 5℃, no dynamic correction is required.

[0147] 5. Dynamic adjustment:

[0148] After closed-loop correction, the actual pump speed was 2392 r / min (error 0.33%), and the actual fan speed was 1395 r / min (error 0.36%). After running for 15 minutes, the game scene switch caused the CPU load to surge to 98%, with a load change rate of 60% / s (>50% / s), triggering a fast correction. Skipping trend fitting, the target speed was calculated based on 1.2 times the current heat output: 85% of the pump's rated speed (2550 r / min) and 75% of the fan's rated speed (1500 r / min). After correction, the speed error was ≤2%.

[0149] 6. Steady-state operation performance:

[0150] After 4 hours of continuous operation, the CPU temperature remained stable at 75℃~78℃, the GPU temperature remained stable at 80℃~83℃, and the coolant temperature remained stable at 34℃~39℃. During this period, the main power supply voltage briefly fluctuated to 11.3V (exceeding the 12V±5% range), and the system switched to the backup power supply within 8ms. The heat dissipation was uninterrupted during the power supply switch. The local OLED display 13 updated the speed adjustment data in real time, and the remote monitoring software recorded load changes and power supply switch events without triggering any alarms. The system response was fast and stable.

[0151]

Example 3

[0152] This embodiment is used to explain the emergency handling of the present invention in the case of abnormal coolant.

[0153] (a) Scene parameters

[0154] The host computer is running video rendering software (load scene correction factor λ=1.2). During operation, the coolant level drops due to a slight leak, and the dielectric constant exceeds the normal range due to the presence of a small amount of impurities.

[0155] (II) Operation Process

[0156] 1. Data Acquisition and Anomaly Detection:

[0157] Data collected by the perception layer: CPU load 91%, real-time frequency 3.4GHz, GPU load 86%, real-time frequency 1.68GHz, CPU temperature 74.2℃, GPU temperature 78.5℃, coolant inlet temperature 36.8℃, outlet temperature 41.2℃, coolant level 9.2cm (below the preset threshold of 10cm), dielectric constant 1.8 (outside the normal range of 2.0-2.5); the calculation layer determines that the coolant is abnormal, triggering the yellow warning indicator 15 to remain lit, and the remote monitoring software pushes a "low coolant level + abnormal dielectric constant" warning message.

[0158] 2. Emergency Response Measures:

[0159] The system initiates a coolant anomaly warning: First, the pump speed is increased to 90% of the rated speed (2700 r / min), and the fan speed is increased to 80% of the rated speed (1600 r / min). After continuous monitoring for 5 minutes, if the coolant level does not rise and the dielectric constant remains at 1.8, a serious anomaly is determined. The system automatically cuts off power to non-essential hardware such as PCIe expansion cards, retaining power only to the CPU and GPU cores, and simultaneously sends an emergency warning to the operating system to "stop the system immediately and check the coolant".

[0160] 3. Treatment effect:

[0161] After emergency handling, the CPU temperature dropped from 74.2℃ to 70.5℃ and the GPU temperature dropped from 78.5℃ to 75.3℃, preventing overheating of core components due to reduced cooling efficiency of the coolant. The local alarm indicator 15 remained solid yellow, and the remote monitoring software recorded the time of the anomaly, the coolant level change curve, and the emergency handling measures to facilitate troubleshooting for the user. After troubleshooting, the user restarted the system using the mode switch button 14, and the system resumed normal operation after passing the self-test.

[0162]

Example 4

[0163] This embodiment is used to explain the fault-tolerant operation of the present invention in sensor failure scenarios.

[0164] (a) Scene parameters

[0165] The host computer is running programming development tools (low-medium load, λ=1.0). One of the CPU's temperature sensors, 9, is not providing data feedback due to poor circuit contact.

[0166] (II) Operation Process

[0167] 1. Fault identification and data correction:

[0168] After system initialization, the perception layer detects that one of the CPU's temperature sensors 9 has no data, determines that the sensor is faulty, triggers the yellow alarm indicator 15 to flash, and records "CPU temperature sensor 1 fault" in the log. The computing layer activates the fault tolerance mechanism, using the real-time data (current temperature 62.3℃) from another normal CPU sensor and combining it with the historical average of the last 10 minutes (61.8℃), correcting it to the valid data of 62.1℃. At the same time, the GPU and other sensor data are normal, and the coolant status is normal.

[0169] 2. Heat dissipation regulation and fault-tolerant operation:

[0170] Based on the corrected temperature data, the segmented adjustment algorithm outputs the pump at 50% of its rated speed (1500 r / min) and the fan at 40% of its rated speed (800 r / min), with a dynamic correction speed error of ≤1.5%. After 2 hours of continuous operation, relying on single sensor data and historical average correction mechanism, the CPU temperature remained stable at 61℃~63℃ and the GPU temperature remained stable at 65℃~67℃, with no abnormalities in heat dissipation adjustment. After the user checked the fault log through remote monitoring software, the sensor circuit was repaired by stopping the system, and the fault was resolved after restarting the system, and alarm indicator 15 went out.

[0171]

Example 5

[0172] This embodiment is used to explain the implementation of the system calibration process of the present invention.

[0173] (a) Calibration scenario

[0174] Sensor accuracy was calibrated after 6 months of system operation, and algorithm parameters were optimized after 1 year of operation.

[0175] (II) Calibration Process

[0176] 1. Sensor accuracy calibration:

[0177] Temperature sensor calibration: Using a standard constant temperature chamber (accuracy ±0.1℃), the CPU, GPU, and coolant temperature sensors 9 were placed in constant temperature environments of 50℃, 70℃, and 90℃ respectively. The deviation between the sensor measurement values ​​and the standard values ​​was recorded. The deviation compensation coefficient was adjusted through the built-in calibration program in the calculation layer. After calibration, the detection accuracy error of all temperature sensors 9 was ≤ ±0.3℃.

[0178] 2. Level and dielectric constant sensor calibration: Inject a standard volume of coolant into the chamber 1, calibrate the level sensor measurement value, and ensure that the level error is ≤ ±0.2cm; use standard dielectric constant samples (2.0, 2.3, 2.5) to calibrate the dielectric constant sensor, and the measurement error after calibration is ≤ ±0.05.

[0179] 3. Algorithm parameter optimization: Based on one year of running data (covering office, gaming, rendering and other scenarios), the power consumption coefficient k was optimized (CPU adjusted to 0.88-0.95, GPU adjusted to 0.98-1.08) and the load scenario correction coefficient λ (new live streaming scenario λ=1.15). The sample data of the linear regression trend prediction algorithm were refitted. After optimization, the error between the predicted temperature and the actual temperature is ≤2℃.

[0180] 4. Calibration Results: After calibration, the system operates in office scenarios, the CPU temperature measurement accuracy is improved to ±0.2℃, the temperature prediction error in gaming scenarios is reduced from 3℃ to 1.8℃, the heat dissipation efficiency is improved by 5%, the energy consumption is reduced by 8%, and the long-term stability of the system is significantly improved.

[0181] The specific embodiments of the present invention, through multi-scenario coverage, verify the feasibility and reliability of the intelligent temperature control system under different loads, abnormal faults, and calibration and maintenance scenarios; the hardware selection, parameter calculation, and process execution in each embodiment strictly follow the limitations of the claims, and the technical effects are consistent with the contents of the invention, fully demonstrating the system's data acquisition accuracy, scientific adjustment, operational stability, and convenient interaction.

[0182] Although embodiments of the invention have been shown and described, those skilled in the art will recognize that various changes, modifications, substitutions and alterations can be made to these embodiments without departing from the principles and spirit of the invention, the scope of which is defined by the appended claims and their equivalents.

Claims

1. An intelligent temperature control system for an immersion liquid-cooled mainframe chassis, the immersion liquid-cooled mainframe chassis comprising a chassis (1), a frame (2), and a cover plate (3), wherein the top of the chassis (1) is open and contains insulating coolant, the frame (2) is detachably inserted into the chassis (1), the lower mounting bracket (4) carries the motherboard and power supply, an interface module compartment (5) is fixed on the top, and the cover plate (3) is detachably connected to the top of the interface module compartment (5), characterized in that, The intelligent temperature control system is integrated into the interface module compartment (5), including: The sensing layer employs a multi-dimensional sensor combination and a standardized acquisition process to collect data on the temperature of the core heating element, the temperature of the coolant, the coolant status, and the hardware operating status. The acquisition frequency is 1 time / second, and the detection accuracy error is ≤±0.5℃. The multi-dimensional sensor combination includes a temperature sensor group, a current sensor, an SMBus communication module, and a coolant status monitoring unit. The computing layer is electrically connected to the sensing layer via a transfer harness. It uses a real-time heat generation calculation algorithm, a linear regression trend prediction algorithm, a piecewise adjustment algorithm, and a dynamic correction algorithm to calculate the real-time heat generation and predict the future heat generation trend. It outputs speed control commands for the pump (8) and the cooling module (7). In the linear regression trend prediction algorithm, the load scenario correction coefficient λ is set according to the load type. For office scenarios, λ=0.9, for game scenarios, λ=1.1, and for rendering scenarios, λ=1.

2. It also supports calibration according to the actual application scenario. The control layer is electrically connected to the computing layer, pump (8) and refrigeration module (7) respectively through the adapter harness. It receives control commands and adjusts the coolant circulation speed and fan speed. The control layer is equipped with a closed-loop regulation mechanism, multiple protection mechanisms and power supply switching mechanism to ensure operational stability. The human-computer interaction module includes a local interaction unit and a remote interaction unit, which supports visualization of operating status, customization of parameters, and fault warning. The computing layer is also electrically connected to the motherboard of the host chassis via a junction box to obtain hardware operating parameters and provide feedback on system status.

2. The intelligent temperature control system for the immersion liquid-cooled mainframe chassis according to claim 1, characterized in that, The multi-dimensional sensor combination of the perception layer specifically includes: (1) The temperature sensor group adopts a waterproof and corrosion-resistant digital temperature sensor (9), including a core component temperature sensor and a coolant temperature sensor; two core component temperature sensors are attached to each of the CPU and GPU core heat dissipation surfaces, and one is attached to each of the motherboard power supply module and hard disk shell; one coolant temperature sensor is fixed in the coolant inlet, outlet and middle area of ​​the enclosure (1), and all are encapsulated in a corrosion-resistant shell; when the difference between the two sensor data of the same core component exceeds 1.5℃, an anomaly mark is activated and the historical average value is used as a replacement. (2) The current sensor adopts the ACS712 miniature current sensor (16), which is connected in series on the power supply line from the power supply output to the PCIe expansion card to collect the real-time power supply current of the expansion card; (3) The SMBus communication module is integrated into the intelligent control motherboard (10) of the computing layer. It establishes stable communication with the motherboard through the SMBus interface and reads the CPU / GPU operating frequency and load rate data in real time. (4) The coolant status monitoring unit includes a liquid level sensor and a dielectric constant sensor, which are used to detect the coolant level height and dielectric constant change, respectively. When the liquid level is lower than the preset threshold or the dielectric constant exceeds the normal range, an alarm signal is triggered.

3. The intelligent temperature control system for the immersion liquid-cooled mainframe chassis according to claim 1, characterized in that, The closed-loop regulation mechanism of the control layer includes: (1) Speed ​​feedback acquisition: The pump drive module (11) integrates Hall speed detection circuit, and the fan adopts 4-wire PWM fan. The actual speed is collected in real time and fed back to the calculation layer. The feedback frequency is ≥1 time / second. (2) PWM duty cycle correction: The calculation layer adjusts the output signal through a closed-loop correction formula; The corrected formula for the pump is ,in ; The correction formula for the fan is: ,in ; (3) Adjustment accuracy guarantee: Through continuous feedback correction, ensure that the error between the actual speed of the pump and the target speed is ≤3%.

4. The intelligent temperature control system for the immersion liquid-cooled mainframe chassis according to claim 1, characterized in that, The multiple protection mechanisms of the control layer include: (1) Actuator fault protection: If the pump has no speed feedback for 3 consecutive seconds, the fan will start at the highest speed for cooling; if the fan has no speed feedback for 3 consecutive seconds, the pump will start at the highest speed for cooling; if both fail, an emergency warning will be sent to the operating system. (2) Overload protection: The pump and fan speeds, after correction, shall not exceed 100% of the rated speed to avoid hardware overload damage; (3) Temperature over-limit protection: When the temperature of the core component is ≥85℃ or the temperature of the coolant is ≥45℃, the highest speed heat dissipation will be started immediately and an alarm will be triggered. If the temperature does not drop after 30 seconds, the user will be prompted to reduce the load. (4) Coolant abnormal protection: When the coolant level is lower than the preset threshold or the dielectric constant exceeds the normal range, a graded warning will be activated, and in case of serious abnormality, the power supply to unnecessary hardware will be automatically cut off.

5. The intelligent temperature control system for the immersion liquid-cooled mainframe chassis according to claim 1, characterized in that, The power supply switching mechanism of the control layer includes: adopting a dual power supply design, with the main power supply taken from the 12V output terminal of the host power supply and the backup power supply taken from the power supply interface of the interface panel. The two are isolated by diodes, and the input terminal is equipped with a filter capacitor (17) and a voltage regulator chip (18). When the main power supply voltage fluctuates beyond 12V±5%, it switches to the backup power supply within ≤10ms. The backup power supply capacity meets the system's full-load operation requirements.

6. The intelligent temperature control system for the immersion liquid-cooled mainframe chassis according to claim 1, characterized in that, The quantization algorithm used in the computation layer specifically includes: (1) Real-time calorific value calculation algorithm: CPU / GPU heat generation Where k is the power consumption coefficient, with the CPU's power consumption coefficient ranging from 0.8 to 1.0 and the GPU's power consumption coefficient ranging from 0.9 to 1.

1. Rated thermal design power (W), f_actual is the real-time operating frequency (GHz), f_rated is the rated frequency (GHz), and Load is the real-time load rate (0-100%). Coolant heat absorption power , where c is the specific heat capacity of the coolant (J / (kg・℃)), ρ is the density of the coolant (kg / m³), V is the volume of the coolant (m³), ΔT is the temperature change in 1 second (℃), and t=1s; (2) Linear regression trend prediction algorithm: Based on the heat generation data of the most recent 30 minutes, a trend line was fitted using the least squares method. ,in , n=1800 is the sample size; Combined with the load scenario correction coefficient λ, we obtain Then convert it to core temperature , where R is the thermal resistance (°C / W) and S is the heat dissipation surface area (m²). The rules for determining the value of the load scenario correction factor λ are as follows: λ=0.9 for office scenarios, λ=1.1 for game scenarios, and λ=1.2 for rendering scenarios; (3) Segmented adjustment algorithm: Based on the real-time temperature T and the predicted temperature T for the next 5 minutes (300), the target rotational speed is output through a segmented function; (4) Dynamic correction algorithm: If the temperature difference between the inlet and outlet of the coolant ,but , The corrected rotational speed shall not exceed 100% of the rated rotational speed; the 5°C threshold is determined based on the physical properties of the coolant and the heat dissipation requirements of the hardware.

7. The intelligent temperature control system for the immersion liquid-cooled mainframe chassis according to claim 5, characterized in that, The piecewise function satisfies: when and At that time, the pump speed is 30% of the rated speed, and the fan speed is 20% of the rated speed; when or At that time, the pump speed is 50% of the rated speed, and the fan speed is 40% of the rated speed; when or At that time, the pump speed is 80% of the rated speed, and the fan speed is 70% of the rated speed; when or At that time, both the pump speed and the fan speed are 100% of their rated speed; The execution priority of the dynamic correction algorithm is lower than that of the segmented adjustment algorithm. That is, the initial target speed is determined first by the segmented adjustment algorithm, and then fine-tuned by the dynamic correction algorithm.

8. The intelligent temperature control system for the immersion liquid-cooled mainframe chassis according to claim 1, characterized in that, It also includes a human-computer interaction module, which includes: (1) Local interaction unit, located on the switch panel (502) of the interface module compartment (5), including OLED temperature display window (13), mode switching button (14) and red / yellow dual-color alarm indicator (15); OLED display screen (13) displays various temperature, speed and coolant status data in a cycle; mode switching button (14) supports automatic / manual mode switching; alarm indicator (15) is triggered when the temperature exceeds the threshold, sensor fails or coolant is abnormal, red indicates serious fault and yellow indicates general warning; (2) Remote interaction unit, which communicates with the motherboard via the USB / serial port of the intelligent control motherboard (10), supports the installation of monitoring software in the operating system; the monitoring software supports temperature curve display, temperature threshold customization, heat dissipation strategy selection, historical data export (supports data of the last 72 hours), fault log viewing and multi-device network monitoring, and has permission management function, and supports pushing abnormal information to associated terminals.

9. A method for intelligent temperature control of an immersion liquid-cooled mainframe chassis, implemented based on the intelligent temperature control system described in any one of claims 1-8, characterized in that, Includes the following steps: S1: Initialization. After the system is powered on, it performs a self-test on the sensing layer sensors, control layer pumps and fans, and coolant status monitoring unit. After the self-test passes, the pumps and fans start at the lowest speed and enter automatic mode. At the same time, the data storage module is initialized and a cyclic overwrite storage strategy is configured (stores the most recent 72 hours of data, and the storage medium capacity is ≥100MB). S2: Data acquisition. The sensing layer collects core component temperature, coolant temperature, coolant status, CPU / GPU load and frequency, and PCIe expansion card power supply current data at a frequency of 1 time / second. After range verification (temperature -20℃~120℃, load rate 0%~100%, etc.) and consistency verification (three consecutive data changes exceeding the threshold are marked as abnormal), the data is transmitted to the computing layer. S3: Real-time heat generation calculation. The calculation layer calculates the real-time heat generation of the CPU / GPU and the heat absorption power of the coolant through a real-time heat generation calculation algorithm. S4: Heat generation trend prediction. The calculation layer fits a trend line based on the data of the last 30 minutes using a linear regression algorithm, and calculates the core temperature for the next 5 minutes by combining the load scenario correction coefficient λ. If there are abnormal data, the corrected valid data is used for trend fitting. S5: Target speed is determined. Based on real-time and predicted temperatures, the calculation layer first outputs the initial target speed of the pump and fan through a segmented adjustment algorithm, and then adjusts the target speed through a dynamic correction algorithm in combination with the temperature difference between the inlet and outlet of the coolant. S6: Dynamic adjustment. The control layer receives the PWM signal from the calculation layer to drive the pump and fan. Speed ​​accuracy is ensured through speed feedback and closed-loop correction formula. At the same time, the power supply status is monitored in real time. When the main power supply is abnormal, the power supply switching mechanism is triggered. S7: Abnormal handling. If the temperature exceeds the threshold, the highest speed cooling will be activated and an alarm will be triggered. If the sensor fails, the emergency cooling mode will be activated and an alarm will be triggered, and the fault log will be recorded. If the load changes abruptly (load change rate > 50% / s) or the temperature prediction error exceeds 3℃, the trend fitting will be skipped, and the target speed will be calculated directly at 1.2 times the current heat generation. If the coolant is abnormal, a graded warning and corresponding emergency measures will be activated. If the temperature does not drop for 30 seconds after exceeding the threshold, the computing layer will send a warning message to the operating system through the motherboard to prompt the user to reduce the hardware load.

10. The intelligent temperature control method for an immersion liquid-cooled mainframe chassis according to claim 9, characterized in that, It also includes system calibration steps: the sensor accuracy is calibrated every 6 months, and the algorithm parameters, including power consumption coefficient, correction coefficient, and load scenario correction coefficient λ, are optimized annually to ensure the temperature control accuracy and stability of the system in long-term operation.

Images