AI host system and maintenance management optimization method thereof

Abstract

The invention provides an AI host system and a maintenance management optimization method thereof, and belongs to the technical field of computer system operation and maintenance management and hardware control. The system realizes local centralized control through the main control board and the touch screen, has a fault memory function, and supports automatic parameter recovery after power failure; through a multi-point temperature monitoring and linkage heat dissipation mechanism, accurate temperature control and load prediction heat dissipation are realized. The system operation and maintenance efficiency and the operation stability in a high-temperature environment are improved, and the method is suitable for a high-density AI calculation scene.

Description

Technical Field

[0001] This invention relates to the field of computer system operation and maintenance management and hardware control technology, and in particular to an AI host system with intelligent fault memory and precise linkage heat dissipation functions, and its maintenance and management optimization method. Background Technology

[0002] In scenarios requiring high-density deployment of AI hosts, such as security monitoring and edge computing, system reliability and maintainability are crucial. However, existing systems have significant shortcomings in operation and maintenance management and status control: Firstly, regarding fault recovery, traditional out-of-band management methods (such as BMC) typically only allow for remote switching of hardware, but cannot record and restore the software configuration and operational status at the business level. When a component of the system fails or experiences an unexpected power outage, restoring power often requires maintenance personnel to go to the site in person to manually reconfigure parameters and start the hosts one by one. This is inefficient, error-prone, and leads to extended service interruption time, failing to meet the needs of 24 / 7 continuous operation.

[0003] Secondly, in terms of thermal management, traditional systems often rely on passive responses or coarse control based on a single temperature point. Due to the concentrated heat dissipation of multiple AI hosts in high-density deployments, localized high-temperature areas are easily generated. Traditional solutions cannot accurately detect and respond promptly, potentially leading to equipment damage and shortened lifespan due to overheating. Furthermore, independent temperature control of multiple hosts may result in conflicting thermal strategies, reducing overall heat dissipation efficiency.

[0004] Therefore, there is an urgent need for an AI host system and method that can achieve intelligent operation and maintenance, rapid fault recovery, and precise thermal management. Summary of the Invention

[0005] To address the technical problems existing in the prior art, this invention provides an AI host system with high operation and maintenance efficiency, good system stability, and effective prevention of overheating failures, as well as its maintenance and management optimization method.

[0006] To achieve the above objectives, the present invention adopts the following technical solution: In a first aspect, the present invention provides an AI host system, comprising: Chassis; and integrated within the chassis: Multiple AI hosts are used to perform AI inference tasks; The main control board has non-volatile memory; Localized centralized control interface: Connected to the main control board, it is used to display the operating status of all AI hosts and receive control commands from users for one or more hosts; The fault memory module, functionally implemented by the main control board and its non-volatile memory, is used to capture and store fault parameters when an AI host fault is detected, and automatically restore the system to the state before the fault after power-on. The precise temperature monitoring and coordinated heat dissipation module includes multiple temperature sensors arranged in the AI ​​host area and a heat dissipation unit controlled by the main control board; The main control board is configured to uniformly execute fault memory logic and hierarchical linkage heat dissipation strategy.

[0007] Preferably, the fault memory module is configured to perform the following process: Fault capture: When a hardware error, temperature over-limit, or power abnormality is detected in the AI ​​host, the current task is immediately interrupted, the device identifier, working mode, and operating configuration parameters of the host are captured, and stored as a fault parameter package in the non-volatile memory. State recovery: After the system is powered on again, the fault records in the non-volatile memory are checked first. If a valid fault event is identified, the running configuration parameters stored before the fault are automatically loaded into the corresponding AI host, and the host is triggered to perform a power-on self-test process. After the self-test is passed, normal operation is restored.

[0008] Preferably, in the precise temperature monitoring and coordinated heat dissipation module, the main control board is configured to execute the following strategy: Threshold linkage control: Multiple temperature thresholds are preset. The main control board compares the temperature values ​​collected by each sensor with the current threshold in real time, and outputs a PWM signal to dynamically adjust the operating speed of the heat dissipation unit accordingly. Load prediction and heat dissipation: The main control board analyzes the real-time computing load of each AI host, predicts its short-term heat dissipation trend, and increases the operating intensity of the heat dissipation unit in advance based on the prediction results.

[0009] Preferably, the multi-level temperature thresholds include: The heat dissipation unit operates at low speed when the temperature is below the first threshold. The second threshold is when the heat dissipation unit operates at medium speed if it is higher than the first threshold but lower than the second threshold. The third threshold is when the heat dissipation unit operates at high speed, which is higher than the second threshold but lower than the third threshold. The fourth threshold is set above which the heat dissipation unit operates at ultra-high speed and triggers a system-level alarm.

[0010] Preferably, the main control board periodically acquires the CPU utilization and memory usage of each AI host, and uses a pre-stored heat generation calculation model to predict the internal temperature rise of the chassis in the future. If the predicted temperature will exceed the next level threshold, the fan speed will be increased to the corresponding level in advance.

[0011] In a second aspect, the present invention provides a maintenance and management optimization method for an AI host system, applied to the system described in the first aspect, the method comprising: Fault memory and automatic recovery steps: The main control board monitors the status of the AI ​​host. When a fault is detected, the current fault parameters are captured and stored in non-volatile memory. After the system is powered on again, the fault parameters are automatically read and the corresponding AI host is controlled to restore to the state before the fault. Precise temperature monitoring and coordinated heat dissipation steps: Real-time temperature data inside the chassis is collected by multiple temperature sensors and uploaded to the main control board via I2C bus; The main control board executes a graded coordinated heat dissipation strategy and dynamically controls the operation of the heat dissipation unit based on the temperature data and AI host load prediction.

[0012] Preferably, restoring to the pre-fault state specifically includes: automatically loading the operating configuration parameters stored before the fault, and executing a self-test process to verify the system status. Only after the self-test passes can the system be put into normal operation.

[0013] Preferably, the tiered, interconnected heat dissipation strategy includes: The collected temperature values ​​are compared with preset multi-level temperature thresholds, and a PWM signal is output based on the comparison results to adjust the speed of the heat dissipation unit. Based on the real-time load data of the AI ​​host, its heat generation trend is predicted, and the operating intensity of the heat dissipation unit is adjusted in advance to achieve preventive heat dissipation.

[0014] Preferably, the advance adjustment includes: when it is predicted that the temperature of a local area will exceed the next threshold within the next 30 seconds, the main control board immediately increases the fan speed to the target level, rather than waiting for the actual temperature to reach the threshold before taking action.

[0015] Thirdly, the present invention provides an electronic device including a processor, a memory, and a computer program stored in the memory, wherein the processor executes the computer program to implement the above-described method.

[0016] Fourthly, the present invention provides a computer-readable storage medium having a computer program stored thereon, characterized in that the program, when executed by a processor, implements the above-described method. Compared with the prior art, the beneficial effects of the present invention are as follows: Operation and maintenance efficiency has been greatly improved: the fault memory and automatic recovery functions have reduced the need for manual parameter configuration, and the centralized control via touch screen has simplified the operation process and enabled rapid response.

[0017] Enhanced system availability: Minimizes service interruption time caused by failures and power outages, ensuring the continuity of AI inference tasks.

[0018] Precise and efficient heat dissipation: Through multi-point monitoring, hierarchical control and predictive heat dissipation, local overheating is effectively prevented, and the system's operational stability and equipment lifespan are improved in high-temperature environments.

[0019] Unified control logic: The self-developed main control board enables the centralized execution of all control strategies, eliminating internal conflicts of decentralized control and making the system behavior more consistent and reliable. Attached Figure Description

[0020] The accompanying drawings, which are incorporated in and form part of this specification, illustrate embodiments consistent with the invention and, together with the description, serve to explain the principles of the invention.

[0021] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, for those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0022] Figure 1 This is a schematic diagram of the overall structure of the AI ​​host system of the present invention.

[0023] Figure 2 This is a schematic diagram of the internal structure of the AI ​​host system of the present invention. Figure 1 .

[0024] Figure 3 This is a schematic diagram of the internal structure of the AI ​​host system of the present invention. Figure 2 .

[0025] Figure 4 This is a flowchart illustrating the implementation of the fault memory and automatic recovery function of the present invention.

[0026] Figure 5 This is a flowchart of the precise temperature monitoring and linkage heat dissipation control of the present invention. Detailed Implementation

[0027] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only a part of the embodiments of the present invention, and not all of them. Based on the embodiments of the present invention, all other embodiments obtained by those of ordinary skill in the art without creative effort are within the scope of protection of the present invention.

[0028] It should be noted that all directional indications (such as up, down, left, right, front, back, etc.) in the embodiments of the present invention are only used to explain the relative positional relationship and movement of each component in a certain specific posture (as shown in the figure). If the specific posture changes, the directional indication will also change accordingly.

[0029] Furthermore, the use of terms such as "first" and "second" in this invention is for descriptive purposes only and should not be construed as indicating or implying their relative importance or implicitly specifying the number of technical features indicated. Therefore, features defined with "first" and "second" may explicitly or implicitly include at least one of those features. Additionally, the technical solutions of the various embodiments can be combined with each other, but only on the basis of being achievable by those skilled in the art. When the combination of technical solutions is contradictory or impossible to implement, such a combination of technical solutions should be considered non-existent and not within the scope of protection claimed by this invention.

[0030] Example 1 This embodiment describes in detail the specific structure of an AI host system, such as... Figure 1-3 As shown, the AI ​​host system includes a chassis 100, and the following structures integrated into the chassis 100: Multiple AI hosts 103 (e.g., 10 or more) are fixed on a custom bracket 101 to perform inference tasks such as AI image recognition. Each AI host 103 can be independently connected to an external monitoring camera to form a data acquisition unit with a corresponding "host-camera" configuration.

[0031] The single-power-multiple-supply system 104 serves as the sole power source for the system. It is a high-power, high-efficiency power module used to convert AC power into multiple independent and isolated DC power outputs (e.g., 12V, 5V).

[0032] The main control board 106 has non-volatile memory.

[0033] A network switch 111 is integrated into the chassis 100 to enable data exchange between all AI hosts 103 and the external network. The rear panel of the chassis 100 has multiple network cable ports 112 for connecting external cameras and the network.

[0034] The localized centralized control interface 113, preferably an external high-definition capacitive touchscreen in this embodiment, is connected to the main control board 106. It is used to display the status parameters of all AI hosts and receive user control commands for one or more hosts (such as one-click restart or individual on / off switching of a host), achieving localized centralized control with millisecond-level response. The status parameters of the AI ​​hosts include the number of all AI hosts 103, their operating status, system voltage, temperature, and other parameters.

[0035] The main power switch 114 and the switch switch 115 are located on the front panel of the chassis and are used to control the power supply of the entire machine and the network switch, respectively; the filter switch 116 is located on the front panel of the chassis and is used to connect the power filtering function to provide the system with clean and stable mains power input.

[0036] The main unit cover 117 is used to enclose the chassis 100, serving to prevent dust, provide protection, and provide safety isolation.

[0037] The deeply integrated fault memory module is functionally implemented by the main control board and its integrated non-volatile memory (such as EEPROM or FLASH). It is used to capture and store fault parameters when an AI host fault is detected, and automatically restore the system to the state before the fault after the system is powered on again. A unified strategy execution precision temperature monitoring and linkage heat dissipation module includes multiple temperature sensors arranged in the AI ​​host area and a heat dissipation unit (such as a fan) 110 controlled by the main control board; The main control board 106 is configured to uniformly execute fault memory logic and hierarchical linkage heat dissipation strategy, and is connected to the AI ​​host 103 (power and control link), the localized centralized control interface (such as touch screen) 113 (UI interaction link), the temperature sensor (data acquisition link) and the heat dissipation unit (such as fan) 110 (control link) to form a centralized management and control system.

[0038] The deeply integrated fault memory module operates through the following steps: S1. Fault capture: When a hardware error, temperature over-limit, or power abnormality is detected in the AI ​​host, the current task is immediately interrupted, the device identifier, working mode, and operating configuration parameters of the host are captured, and stored as a fault parameter package in the non-volatile memory. S2. State recovery: After the system is powered on again, the fault records in the non-volatile memory are checked first. If a valid fault event is identified, the running configuration parameters stored before the fault are automatically loaded into the corresponding AI host, and the host is triggered to execute the power-on self-test process. After the self-test is passed, normal operation is restored.

[0039] The specific implementation process of steps S1-S2 above is as follows: First, fault detection and capture are performed: During system operation, the main control board monitors each AI host in real time through hardware health status signals, temperature sensor data, and power monitoring circuitry. Once a fault is detected, such as a CPU / memory error, core temperature exceeding a safe threshold, or abnormal power output, the main control board immediately triggers an interrupt. In this interrupt service routine, it accurately captures the unique device identifier (ID) of the currently faulty host, the currently executing working mode (such as inference mode or training mode), and related BIOS / BMC configuration parameters, operating system boot parameters, etc., and packages this information into a fault parameter package.

[0040] Secondly, non-volatile storage is performed: The fault parameter package is immediately written into the non-volatile memory on the main control board, and a timestamp and a valid flag are added. This process ensures that even if the system completely loses power, the critical information at the fault site can be preserved.

[0041] In addition, intelligent recovery is performed: When the system (due to a fault or accident) loses power and then powers on again, the initialization program of the main control board will first check whether there is a valid fault record in the non-volatile memory. If there is, the main control board does not immediately power on all hosts. Instead, it first automatically loads all the running configuration parameters stored before the fault for the faulty host according to the device ID in the record. Subsequently, the main control board triggers the host to execute a strict power-on self-test (POST) process. Only after the self-test passes and it is confirmed that the hardware status has returned to normal, the system allows the host to be put into business operation. This set of processes simulates the operation logic of experienced operation and maintenance personnel, but realizes automated execution at the millisecond level.

[0042] Among them, in the precise temperature monitoring and linkage heat dissipation module, the main control board is configured to execute the following strategies: Threshold linkage control: Multiple levels of temperature thresholds are preset. The main control board continuously compares the temperature values collected by each sensor with the current threshold, and accordingly outputs a PWM (Pulse Width Modulation) signal to dynamically adjust the running speed of the heat dissipation unit; Load prediction heat dissipation: The main control board analyzes the real-time computing load of each AI host, predicts its short-term heat generation trend, and based on the prediction result, increases the running intensity of the heat dissipation unit in advance.

[0043] Specifically, the actual operation of the execution strategy of the precise temperature monitoring and linkage heat dissipation module is as follows: First, deep integration is performed: Multiple high-precision digital temperature sensors are directly embedded near the core heat sink of each AI host and at key nodes of the chassis air duct. All these sensors are directly connected to the self-developed main control board through a unified I2C bus, enabling the main control board to obtain the global temperature field data inside the entire chassis with extremely low latency.

[0044] Secondly, a multi-threshold linkage control strategy is performed: The main control board pre-stores fine multi-level temperature thresholds, including the first threshold T1 - the fourth threshold T4 (for example: T1 = 30 °C, T2 = 40 °C, T3 = 50 °C, T4 = 60 °C, etc.). The main control board continuously compares the readings of each sensor with the multi-level temperature thresholds T1 - T4, and accordingly outputs a PWM signal to dynamically adjust the running speed of the heat dissipation unit: When the temperature of all regions < T1, the heat dissipation unit operates at a low speed (such as a 30% duty cycle) to maintain silence.

[0045] When the temperature of any area is ≥ T1 and < T2, the heat dissipation unit switches to medium speed (e.g., 50% duty cycle) operation.

[0046] When the temperature of any area is ≥ T2 and < T3, the heat dissipation unit switches to high speed (e.g., 80% duty cycle) operation.

[0047] When the temperature of any area is ≥ T3, the heat dissipation unit enters ultra-high speed (100% duty cycle) operation, dissipates heat with all efforts, and issues the highest-level alarm instruction information through the local centralized control interface and network interface.

[0048] During the execution of the multi-threshold linkage control strategy in this embodiment, the main control board precisely controls the fan speed through PWM signals to achieve the best balance between noise and heat dissipation efficiency.

[0049] In addition, a preventive heat dissipation strategy based on load prediction is carried out: the main control board periodically (e.g., every 5 seconds) obtains the CPU utilization rate, memory occupancy rate, and power consumption data of each AI host through the operating system interface or monitoring chip. A heat generation calculation model (which can be a linear or lightweight non-linear model trained based on historical data) is preset in the system. The main control board uses this model to predict the temperature rise trend of each area in the short term (e.g., in the next 30 seconds) based on the current load data. If it is predicted that the temperature of a certain area will exceed the threshold corresponding to the current fan speed gear of the heat dissipation unit within the next 30 seconds (for example, currently operating at medium speed and the predicted temperature will exceed T2), the main control board will immediately increase the fan speed of the heat dissipation unit to the target gear (e.g., increase to high speed) in advance.

[0050] It should be noted that in addition to the above-mentioned linear model example of the heat generation calculation model, the model can also be in the form of a look-up table, and the corresponding relationship between different load combinations and temperature rise is established in advance based on experimental data. The heat generation calculation model can be a well-known thermodynamic model based on power consumption and heat dissipation coefficient in the art, or can be obtained by simple linear regression fitting based on historical operation data. Its specific form does not affect the implementation of the core innovative solution of the present invention. The implementation method of this model is a model well-known to those skilled in the art, and the present invention does not limit it here.

[0051] Through the preventive heat dissipation strategy based on load prediction in this embodiment, it effectively avoids the sudden temperature rise caused by the system's thermal inertia and the possible overheat protection shutdown, ensuring the smooth and continuous calculation tasks.

[0052] Embodiment 2 As Figure 4-5 shown, this embodiment elaborates in detail an optimization method for the maintenance management of an AI host system. The method is applied to the system described in Embodiment 1, and the method includes: Fault memory and automatic recovery steps: The main control board monitors the status of the AI ​​host. When a fault is detected, the current fault parameters are captured and stored in non-volatile memory. After the system is powered on again, the fault parameters are automatically read and the corresponding AI host is controlled to restore to the state before the fault. "Restore to the state before the fault" specifically includes: automatically loading the running configuration parameters stored before the fault and executing a self-test process to verify the system status. The system can only be put into normal operation after the self-test is passed.

[0053] Precise temperature monitoring and coordinated heat dissipation steps: Multiple temperature sensors collect real-time internal temperature data from the chassis and upload it to the main control board via the I2C bus. Based on the temperature data and AI host load prediction, the main control board executes a tiered coordinated heat dissipation strategy to dynamically control the operation of the heat dissipation units. The tiered coordinated heat dissipation strategy includes: The collected temperature values ​​are compared with preset multi-level temperature thresholds, and a PWM signal is output based on the comparison results to adjust the speed of the heat dissipation unit. Based on the real-time load data of the AI ​​host, its heat generation trend is predicted, and the operating intensity of the heat dissipation unit is adjusted in advance to achieve preventive heat dissipation.

[0054] The "advance adjustment" includes: when it is predicted that the temperature in a local area will exceed the next threshold within the next 30 seconds, the main control board immediately increases the fan speed to the target level, instead of waiting for the actual temperature to reach the threshold before taking action.

[0055] Specifically, the above method includes the following steps: S101: Fault memory and automatic recovery process, such as Figure 4 As shown, it includes the following steps: S11: During system operation, the main control board continuously monitors the status of each AI host. Specifically, the main control board collects data such as CPU / memory error status, core temperature, and power output fluctuations of each AI host in real time through hardware health monitoring interfaces (such as BMC / IPMI), temperature sensors, power management chips, etc.

[0056] Example: When the CPU temperature of an AI host exceeds 90℃ or the power output is lower than 4V, the main control board determines that the AI ​​host is in an "abnormal state".

[0057] S12: Determine if a fault is detected (such as hardware error, excessive temperature, abnormal power supply, etc.). If so, execute S13. The fault diagnosis logic built into the main control board includes: Hardware errors: such as ECC memory errors, PCIe link disconnection; Temperature exceeding limits: such as core temperature > 90℃; Power supply abnormalities: such as voltage drops or overcurrent protection triggering.

[0058] S13: Immediately capture fault context: The main control board triggers the interrupt service routine to interrupt the current task, reads and records the faulty host's ID, working mode, configuration parameters (such as BIOS settings, OS boot parameters, inference task configuration), etc., to form a fault parameter package.

[0059] Example: The fault parameter packet format is: {Host ID: 03, Fault type: "Temperature overlimit", Configuration parameters: "boot_mode=UEFI, power_limit=65W", Timestamp: "2025-07-11 14:05:32"}.

[0060] S14: Persistent storage: Write the fault parameter package to the non-volatile memory (such as EEPROM or Flash) on the main control board and mark it as "to be recovered" to ensure that the fault scene information is not lost even if the system is completely powered off.

[0061] S15: The system can perform protective shutdown or issue alarms: The main control board can decide whether to trigger a protective shutdown of the entire machine or only shut down the faulty host according to the severity of the fault, and notify the operation and maintenance personnel through the touch screen / network alarm.

[0062] S16: When the system is powered on again, the main control board first reads the fault records: During the startup initialization phase, the main control board prioritizes scanning the fault records in the non-volatile memory.

[0063] S17: Determine if there are any unrecovered valid fault records. If there are records marked as "to be recovered", proceed to recovery process S18; otherwise, start normally.

[0064] S18: Automatic parameter reset: The main control board loads the operating parameters stored before the fault to the faulty host. Specifically, the main control board writes the operating parameters stored before the fault to the host ID in the fault record, including: BIOS configuration, operating system boot parameters and inference task configuration file.

[0065] Example: Before the failure, host 03 was running in "high-performance mode". Upon recovery, the CPU power limit was automatically set to 65W. S19: Perform self-test and verification: The main control board triggers the faulty host to perform a power-on self-test (POST), checking the hardware status of the CPU, memory, storage, network, etc. If the self-test passes, the fault record is cleared, and the host is put into operation; if the self-test fails, a new fault is recorded and an alarm is triggered. This process greatly improves operation and maintenance efficiency.

[0066] In this embodiment, step S101 uses non-volatile storage and interrupt capture mechanisms to save a "snapshot" of the fault scene, reducing manual intervention; through automatic parameter resetting and self-test verification, it simulates the manual recovery process, achieves millisecond-level automated execution, improves system availability, and is suitable for AI inference scenarios that run 24 / 7.

[0067] S102: Precise temperature monitoring and coordinated heat dissipation process, such as Figure 5 As shown, it includes the following steps: S21: Multiple temperature sensors continuously collect the temperature of key points inside the chassis. Digital temperature sensors are deployed near the CPU heatsink, memory modules, and air inlets and outlets of each AI host and networked via I2C bus.

[0068] S22: Temperature data is uploaded to the main control board in real time via the I2C bus. In actual operation, the main control board reads data from each sensor in a polling manner, with a sampling frequency of 1Hz.

[0069] S23: The main control board compares the temperature with multiple thresholds in real time. In actual operation, four temperature thresholds are preset from the first threshold T1 to the fourth threshold T4 (e.g., T1=30℃, T2=40℃, T3=50℃, T4=60℃). The main control board compares the readings of each sensor with the thresholds to determine the current heat dissipation level.

[0070] S24: Parallel Load Prediction: The main control board acquires the real-time load of the AI ​​host and inputs it into the prediction model to calculate the short-term temperature rise. In actual operation, the main control board periodically (e.g., every 5 seconds) acquires data such as CPU utilization, memory usage, and power consumption of each AI host through the operating system interface, inputs it into the preset heat generation calculation model (e.g., temperature rise = a × CPU utilization + b × memory usage + c), and predicts the temperature change trend within the next 30 seconds.

[0071] S25: Comprehensive Decision: The main control board generates control commands for the heat dissipation unit by combining the current measured temperature and the predicted temperature rise.

[0072] If a temperature rise is predicted to cause a step-up, the speed-up command will be executed in advance. For example, if it is predicted that the temperature in a certain area will exceed the next level threshold within 30 seconds, the speed of the heat dissipation unit will be increased in advance.

[0073] If not, then execute the corresponding speed command based on the threshold range of the current measured temperature.

[0074] Example: The current temperature is 38℃ (medium speed), and it is predicted that it will rise to 43℃ in 30 seconds (high speed should be switched), then the main control board will immediately increase the fan speed to high speed.

[0075] S26: Execution Control: Based on instructions, the speed of the heat dissipation unit is dynamically adjusted via PWM signals, supporting multi-level speed regulation (such as 20%, 40%, 60%, 80%, 100% duty cycle), and avoiding policy conflicts among multiple heat dissipation units through centralized control. This process realizes the transformation of heat dissipation from "passive response" to "active prevention".

[0076] In this embodiment, step S102 involves multi-point monitoring: global perception of the temperature field to avoid local blind spots; predictive control: adjusting heat dissipation in advance based on load trends to overcome system thermal inertia; and centralized coordination: the main control board uniformly schedules all heat dissipation units to avoid competition and conflict.

[0077] This method has the following beneficial effects: Fault recovery and heat dissipation control are fully automated, reducing manual intervention and improving operation and maintenance efficiency; Rapid recovery and precise temperature control ensure business continuity and hardware security, and enhance system stability; The main control board centrally executes all strategies, ensuring consistent and reliable system behavior and unified control logic.

[0078] This method achieves automated operation and maintenance, recoverable status, and intelligent heat dissipation of AI host systems through two core processes: fault memory and automatic recovery, and precise temperature monitoring and coordinated heat dissipation. It is particularly suitable for high-density, high-availability AI inference scenarios.

[0079] The specific implementation methods of the above methods and steps are the same as the specific implementation methods of the functions of each module in Embodiment 1, and will not be repeated here.

[0080] Example 3 The present invention also provides an electronic device, including: a processor, a transmitting device, an input device, an output device, and a memory. The processor may be implemented using a general-purpose CPU (Central Processing Unit), a microprocessor, an application-specific integrated circuit, or one or more integrated circuits, and is used to execute relevant programs to implement the technical solutions provided in the embodiments of this application. The memory may be implemented using a read-only memory (ROM), a static storage device, a dynamic storage device, or a random access memory (RAM), and is used to store computer program code. The computer program code includes computer instructions. When the processor executes the computer instructions, the electronic device executes a method as described in any of the above possible implementation methods.

[0081] Example 4 The present invention also provides a computer-readable storage medium storing a computer program, the computer program including program instructions, which, when executed by a processor of an electronic device, cause the processor to perform a method as described in any of the above possible implementations.

[0082] In the description of this specification, the references to terms such as "an embodiment," "example," "specific example," etc., indicate that a specific feature, structure, material, or characteristic described in connection with that embodiment or example is included in at least one embodiment or example of the present invention. In this specification, the illustrative expressions of the above terms do not necessarily refer to the same embodiment or example. Furthermore, the specific features, structures, materials, or characteristics described may be combined in any suitable manner in one or more embodiments or examples.

[0083] The above description is merely a specific embodiment of the present invention, enabling those skilled in the art to understand or implement the invention. Various modifications to these embodiments will be readily apparent to those skilled in the art, and the general principles defined herein may be implemented in other embodiments without departing from the spirit or scope of the invention. Therefore, the present invention is not to be limited to the embodiments shown herein, but is to be accorded the widest scope consistent with the principles and novel features claimed herein.

Claims

1. An AI host system, characterized in that, include: Chassis; And those integrated into the chassis: Multiple AI hosts are used to perform AI inference tasks; The main control board has non-volatile memory; Localized centralized control interface: Connected to the main control board, it is used to display the operating status of all AI hosts and receive control commands from users for one or more hosts; The fault memory module, functionally implemented by the main control board and its non-volatile memory, is used to capture and store fault parameters when an AI host fault is detected, and automatically restore the system to the state before the fault after power-on. The precise temperature monitoring and coordinated heat dissipation module includes multiple temperature sensors arranged in the AI ​​host area and a heat dissipation unit controlled by the main control board; The main control board is configured to uniformly execute fault memory logic and hierarchical linkage heat dissipation strategy.

2. The system according to claim 1, characterized in that, The fault memory module is configured to perform the following process: Fault capture: When a hardware error, temperature over-limit, or power abnormality is detected in the AI ​​host, the current task is immediately interrupted, the device identifier, working mode, and operating configuration parameters of the host are captured, and stored as a fault parameter package in the non-volatile memory. State recovery: After the system is powered on again, the fault records in the non-volatile memory are checked first. If a valid fault event is identified, the running configuration parameters stored before the fault are automatically loaded into the corresponding AI host, and the host is triggered to perform a power-on self-test process. After the self-test is passed, normal operation is restored.

3. The system according to claim 1, characterized in that, In the precise temperature monitoring and coordinated heat dissipation module, the main control board is configured to execute the following strategy: Threshold linkage control: Multiple temperature thresholds are preset. The main control board compares the temperature values ​​collected by each sensor with the current threshold in real time, and outputs a PWM signal to dynamically adjust the operating speed of the heat dissipation unit accordingly. Load prediction and heat dissipation: The main control board analyzes the real-time computing load of each AI host, predicts its short-term heat dissipation trend, and increases the operating intensity of the heat dissipation unit in advance based on the prediction results.

4. The system according to claim 3, characterized in that, The multi-level temperature thresholds include: The heat dissipation unit operates at low speed when the temperature is below the first threshold. The second threshold is when the heat dissipation unit operates at medium speed if it is higher than the first threshold but lower than the second threshold. The third threshold is when the heat dissipation unit operates at high speed, which is higher than the second threshold but lower than the third threshold. The fourth threshold is set above which the heat dissipation unit operates at ultra-high speed and triggers a system-level alarm.

5. The system according to claim 1, characterized in that, The specific implementation of the load prediction heat dissipation includes: the main control board periodically obtains the CPU utilization and memory usage of each AI host, and predicts the internal temperature rise of the chassis in the future period of time through the pre-stored heat generation calculation model. If the predicted temperature will exceed the next level threshold, the fan speed will be increased to the corresponding level in advance.

6. A method for optimizing the maintenance and management of an AI host system, characterized in that, Applied to the system as described in any one of claims 1-5, the method comprises: Fault memory and automatic recovery steps: The main control board monitors the status of the AI ​​host. When a fault is detected, the current fault parameters are captured and stored in non-volatile memory. After the system is powered on again, the fault parameters are automatically read and the corresponding AI host is controlled to restore to the state before the fault. Precise temperature monitoring and coordinated heat dissipation steps: Real-time temperature data inside the chassis is collected by multiple temperature sensors and uploaded to the main control board via I2C bus; The main control board executes a graded coordinated heat dissipation strategy and dynamically controls the operation of the heat dissipation unit based on the temperature data and AI host load prediction.

7. The method according to claim 6, characterized in that, The process of restoring the system to its pre-fault state includes: automatically loading the pre-fault stored operating configuration parameters and performing a self-test process to verify the system status. The system can only be put into normal operation after the self-test is passed.

8. The method according to claim 6, characterized in that, The hierarchical and coordinated heat dissipation strategy includes: The collected temperature values ​​are compared with preset multi-level temperature thresholds, and a PWM signal is output based on the comparison results to adjust the speed of the heat dissipation unit. Based on the real-time load data of the AI ​​host, its heat generation trend is predicted, and the operating intensity of the heat dissipation unit is adjusted in advance to achieve preventive heat dissipation.

9. The method according to claim 8, characterized in that, The advance adjustment includes: when it is predicted that the temperature in a local area will exceed the next threshold within the next 30 seconds, the main control board immediately increases the fan speed to the target level, instead of waiting for the actual temperature to reach the threshold before taking action.

10. A computer-readable storage medium having a computer program stored thereon, characterized in that, When the program is executed by the processor, it implements the method as described in any one of claims 6 to 9.

Images