An ai full-dimension chassis monitoring system and method

Abstract

The application discloses an AI full-dimension case monitoring system and method, comprising a multi-channel sensing acquisition module for collecting the global environment of a case main body and multiplexing the original power supply and fan FG interface of the case; an AI intelligent monitoring box built-in with an AI edge operation module and used for pre-processing multi-source data, an LCD visual interaction module, a local storage module, and a case power supply unit, wherein the case power supply unit is electrically connected with the multi-channel sensing acquisition module, the AI edge operation module, the LCD visual interaction module and the local storage module; the AI edge operation module is used for realizing multi-dimensional data trend modeling, load temperature rise prediction and fault root cause positioning, and can identify hidden hazards such as case dust accumulation, fan aging and temperature rise anomaly in advance, so that the lag protection mode of traditional device fault passive alarm is changed into the active protection mechanism of trend prediction and early intervention.

Description

Technical Field

[0001] This invention belongs to the field of intelligent monitoring and protection technology for computer hardware, specifically relating to an AI-powered all-dimensional chassis monitoring system and method. Background Technology

[0002] Desktop PCs and computing server chassis are subject to long-term temperature and humidity fluctuations, dust accumulation, mechanical resonance, power supply ripple, fan aging, and other factors, which can easily lead to PCB electrochemical corrosion, solder joint fatigue cracking, chip thermal aging, power supply attenuation and other faults, thus shortening the service life of the entire machine.

[0003] Existing traditional chassis monitoring equipment mostly has a single monitoring dimension, only collecting local temperature data, which cannot comprehensively identify the causes of hardware damage. Various sensor data are collected independently, lacking the ability to fuse multi-source data and predict time-series trends. Fixed thresholds are not well adapted to different load scenarios such as office, gaming, and full load. Moreover, it only supports passive alarms for exceeding limits, requiring manual troubleshooting after a fault occurs, and cannot automatically determine the cause, resulting in low maintenance efficiency. It cannot identify early silent faults such as voltage spikes and speed drift. In addition, adding a monitoring module requires additional wiring and chassis modification, which can obstruct airflow, affect the overall heat dissipation, and complicate mass production assembly.

[0004] A chassis health monitoring system based on distributed modules is disclosed in patent application publication number CN121636289A. This system deploys independent slave health modules on each board, which are aggregated to the main health module via I2C / IPMB bus and then uploaded to the main CPU of the chassis. Although it achieves distributed acquisition and centralized reporting of temperature and voltage parameters of multiple boards, this detection method is a passive data acquisition method, lacking the ability to fuse multi-source data and predict time-series trends, and is not accurate enough in predicting hardware lifespan. In addition, the distributed slave modules require additional hardware to be installed on each board, which is complicated by wiring, occupies hardware installation space, blocks airflow, and affects heat dissipation. Summary of the Invention

[0005] The purpose of this invention is to provide an AI-powered, multi-dimensional chassis monitoring system and method to address the problems mentioned in the background art, such as the lack of multi-source data fusion and time-series trend prediction capabilities, inaccurate hardware lifespan prediction, and distributed slave module wiring obstructing airflow and affecting heat dissipation.

[0006] To achieve the above objectives, the present invention adopts the following technical solution: An AI-powered all-dimensional chassis monitoring system includes a multi-channel sensor acquisition module for collecting data on the entire chassis environment and reusing the original power supply and fan FG interface of the chassis. The system includes an AI intelligent monitoring box with a built-in AI edge computing module for preprocessing multi-source data, an LCD visualization interaction module, a local storage module, and a chassis power supply unit. The chassis power supply unit is electrically connected to the multi-channel sensing acquisition module, the AI ​​edge computing module, the LCD visualization interaction module, and the local storage module. The AI ​​edge computing module includes a synchronous trigger chip and a trend modeling unit for establishing time-series curves of temperature, humidity, dust, vibration, and temperature; a root cause localization unit for comparing the full-domain time-series data to output the root cause of the fault; and a health scoring unit for calculating the overall health of the machine in six dimensions. The trend modeling unit, the root cause localization unit, and the health scoring unit are all electrically connected to the synchronous trigger chip.

[0007] Preferably, it also includes a partition temperature control drive module for driving the cooling fans of each partition of the chassis. The AI ​​edge computing module is provided with a partition temperature control computing unit for outputting directional speed adjustment commands, and a load prediction unit for predicting the temperature rise trend based on the load. The partition temperature control computing unit is electrically connected to the partition temperature control drive module, and the load prediction unit is placed at the front of the AI ​​computing link and electrically connected to the synchronous trigger chip.

[0008] Preferably, the partition temperature control drive module includes an instruction receiving interface for receiving AI edge computing speed adjustment instructions; Four independent PWM generators corresponding to the CPU area, GPU area, power supply compartment and motherboard; Power drive circuit used to amplify PWM drive signals; Fan signal acquisition loop for interfacing with the chassis's native fan FG interface; A speed feedback circuit is used to provide feedback on the actual speed of each fan group; The command receiving interface, four independent PWM generators, and power drive circuit are electrically connected in sequence. The output of the power drive circuit is connected to each cooling fan in the chassis. The input of the speed feedback circuit is connected to the FG interface of the fan. The output of the speed feedback circuit is transmitted back to the multi-channel sensor acquisition module.

[0009] Preferably, the sensing unit of the multi-channel sensing acquisition module includes a temperature and humidity sensor, a dust concentration sensor, and a vibration sensor. The temperature and humidity sensor, the dust concentration sensor, and the vibration sensor are electrically connected to the multi-channel analog switch, the signal conditioning circuit, and the ADC conversion chip in the multi-channel sensing acquisition module.

[0010] Preferably, the LCD visualization interaction module includes a data receiving interface for receiving AI calculation result data, an LCD liquid crystal display driver chip, a three-color status indicator light driver unit, an image buffer for caching real-time monitoring images, an image buffer unit for storing custom display images, and a log storage subunit for locally storing all abnormal event logs. The data receiving interface, LCD display driver chip, image buffer, and image buffer unit are electrically connected in sequence. The LCD display driver chip drives the LCD display screen and the three-color status indicator light driver unit, respectively. The log storage subunit is electrically connected to the local storage module.

[0011] Preferably, it also includes a remote communication module for synchronizing the whole machine monitoring records to a remote terminal; The remote communication module includes a wired network port circuit, a wireless WiFi communication unit, a data encryption transmission unit, and a standardized report generation unit; The wired network port circuit and the input terminal of the wireless WiFi communication unit are both connected to the data encryption transmission unit. The data encryption transmission unit receives the whole machine monitoring data output by the AI ​​edge computing module, and the standardized report generation unit outputs maintenance reports that can be exported to remote terminals.

[0012] Preferably, the trend modeling unit is integrated inside the main control chip of the AI ​​edge computing module. It performs sliding window caching and multi-dimensional time series curve fitting through the RAM built into the main control chip, and uses exponential moving average (EMA) to smooth and suppress instantaneous load shocks and short-term vibration interference, generating smooth and continuous temperature and humidity curves, dust accumulation curves, vibration fluctuation curves, and zoned temperature time series curves. The generated results are connected to the local storage module via the SPI bus, and trend features are obtained through multi-dimensional feature extraction. The feature extraction results are output to the synchronous trigger chip and synchronously sent to the load prediction unit, the grade judgment unit, the root cause localization unit, and the health scoring unit.

[0013] Preferably, the root cause localization unit is integrated inside the main control chip of the AI ​​edge computing module, and root cause localization is achieved through the following steps: A1. Read the fault feature templates pre-stored in the local storage module and update the cache using the built-in computing resources of the main control chip; A2. Receive the full-dimensional time series data and quantitative features output by the trend modeling unit, and distinguish between instantaneous interference and persistent anomalies through horizontal and vertical bidirectional time series comparison; A3. Determine the root cause of the fault based on logical hierarchical matching; A4. The generated root cause results are encapsulated as standardized fault data packets and transmitted to the synchronous trigger chip and then sent to the LCD visualization interaction module for display. On the other hand, they are written to the local storage module via the SPI bus to complete the archiving and retention of timestamped exception logs.

[0014] Preferably, the health scoring unit is integrated inside the main control chip of the AI ​​edge computing module. The main control chip reads the six-dimensional weight parameters pre-stored in the local storage module via SPI, and receives the mean, slope of change, and fluctuation amplitude of each indicator output by the trend modeling unit, as well as the warning and alarm flags of the grading judgment unit. Each indicator parameter is scored in segments according to the threshold range, and finally the overall health score is obtained by weighting. The six-dimensional weighted parameters include temperature and humidity parameters, dust parameters, vibration parameters, power supply ripple parameters, fan status parameters, and zone temperature parameters.

[0015] An AI-powered, multi-dimensional chassis monitoring method, employing the aforementioned AI-powered, multi-dimensional chassis monitoring system, includes the following steps: S1. The multi-channel sensor acquisition module synchronously acquires raw analog signals of chassis temperature and humidity, dust, vibration, power supply, static electricity, and fan speed. After signal conditioning and ADC digitization, the signals are uploaded in batches to the AI ​​edge computing module. S2 and AI preprocess all digital sensor data to construct continuous time-series curves for temperature, humidity, dust, vibration, and hardware load. S3. Predicts short-term temperature rise trends based on CPU and GPU timing load; S4 and AI divide the CPU, GPU, power supply compartment and motherboard into four temperature control zones, compare the real-time temperature of the zone with the adaptive threshold, and output independent PWM speed adjustment commands to the zone temperature control driver module. S5. Compare real-time sensor data with adaptive thresholds to distinguish between single-item mild warnings, single-item severe alarms, and multi-indicator synchronous linkage anomalies, and identify early silent faults. S6: Mild warning automatically adjusts fan speed to improve heat dissipation; single severe alarm controls the corresponding zone fan to cool at full speed, and the yellow indicator light on the LCD panel flashes continuously; multi-indicator synchronous linkage abnormal start full-domain linkage protection, four fans at full load to enhance heat dissipation, lock the whole machine power supply safety range to suppress resonance, the red indicator light on the panel continuously alarms, and simultaneously generates emergency abnormal data packets. S7. After the alarm is triggered, AI compares the full-dimensional time sequence features and outputs standardized fault root cause records to be stored in the log. S8 uses six-dimensional data weighted calculation to obtain the real-time health score of the whole machine, calls the life decay model to estimate the remaining lifespan of the fan, power supply and motherboard, and pushes maintenance reminders when the threshold is reached. The S9 LCD visual interactive panel displays real-time monitoring data across all dimensions, while three-color indicator lights provide synchronous feedback on the overall machine status. Time-series data and abnormal logs are automatically written to the local storage module for long-term caching. S10, the remote communication module synchronizes the whole machine data and abnormal logs to the remote terminal, and automatically generates hardware operation and maintenance statistical reports for operation and maintenance traceability and analysis.

[0016] The AI-powered, multi-dimensional chassis monitoring system and method proposed in this invention have the following advantages compared with existing technologies: 1. This invention uses an AI edge computing module to achieve multi-dimensional data trend modeling, load temperature rise prediction, and fault root cause location. It can identify hidden deterioration risks such as chassis dust accumulation, fan aging, and abnormal temperature rise in advance, and transform the traditional passive alarm-based protection mode of equipment failure into an active protection mechanism of trend prediction and early intervention. This helps to avoid progressive failures of chassis hardware and improve the stability and service life of chassis operation.

[0017] 2. The system of this invention reuses the original power supply circuit of the chassis and the native FG interface of the fan to complete data acquisition and temperature control drive. There is no need to lay out external lines or modify the chassis structure and fan hardware, eliminating the problem of messy extra wiring. At the same time, it does not block or interfere with the original airflow heat dissipation structure of the chassis. While achieving full-dimensional intelligent monitoring and temperature control, it ensures the original heat dissipation efficiency of the chassis, making it more adaptable and practical. 3. This invention adopts an integrated front panel LCD visual monitoring panel to replace the traditional segmented and scattered operation panel, integrating multiple functions such as data display, status indication, human-machine operation, and log viewing, realizing integrated centralized display and unified control of the entire chassis operation data, which can improve the convenience of equipment operation and maintenance. Attached Figure Description

[0018] Figure 1 This is a schematic diagram of the overall system distribution of the present invention; Figure 2 This is a system block diagram of the present invention.

[0019] In the diagram: 1. Chassis main body; 2. AI intelligent monitoring box; 3. Inlet air temperature and humidity sensor; 4. Internal temperature and humidity sensor; 5. Dust concentration sensor; 6. Vibration sensor; 7. LCD visual monitoring panel; 8. Power supply compartment; 9. Front cooling fan; 10. Dust filter; 11. GPU; 12. Motherboard; 13. CPU cooler; 14. Rear cooling fan; 15. Top cooling fan; 16. Bottom cooling fan. Detailed Implementation

[0020] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. The specific embodiments described herein are merely used to explain the present invention and are not intended to limit the present invention. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0021] This invention provides, for example Figure 1-2 The AI-powered all-dimensional chassis monitoring system shown includes a multi-channel sensor acquisition module for collecting data on the entire environment of the chassis body 1 and reusing the original power supply and fan FG interface of the chassis. By setting up a multi-channel sensor acquisition module to reuse the original power supply and fan FG interface of the chassis, the chassis power supply unit centrally supplies power to all functional modules, simplifying the overall wiring layout, eliminating the need for additional wiring, and avoiding the problem of wiring blocking ventilation ducts and affecting heat dissipation.

[0022] like Figure 1 As shown, an LCD visual monitoring panel 7 is installed on the top of one side of the chassis body. The chassis body contains a power supply compartment 8, a GPU 11, a motherboard 12, and a CPU cooler 13. A dust filter 10 is installed on the side wall of the chassis body. The fans are divided into sections, including a front cooling fan 9, a rear cooling fan 14, and a top cooling fan 15. The LCD visual monitoring panel is integrated into the front panel of the chassis body 1, replacing the traditional fragmented operation panels. This integrated front panel centrally integrates data display, three-color status indicator lights, log viewing, and human-machine interaction functions. The power supply compartment 8 is installed in a separate, enclosed compartment at the lower rear of the chassis body 1, forming an independent temperature control zone. An internal temperature and humidity sensor 4 is located within the compartment, and a rear cooling fan 14 assists in heat dissipation. The GPU 11 is installed in a separate heat dissipation zone in the middle of the chassis, adjacent to the top cooling fan 15, with the airflow directly facing the GPU 11 heatsink fin area. The motherboard 12 is installed at the bottom middle section of the chassis, with a vibration sensor 6 mounted on the back of the motherboard 12, belonging to an independent temperature control zone. A bottom cooling fan 16 assists in blowing air onto the motherboard 12 PCB. The CPU cooler faces the airflow of the front cooling fan 9, and an intake temperature and humidity sensor 3 is located at the front of the cooler's intake airflow.

[0023] The front cooling fan 9 corresponds to the CPU partition, the rear cooling fan 14 corresponds to the power supply compartment partition 8, the top cooling fan 15 corresponds to the GPU partition 11, and the bottom cooling fan 16 corresponds to the motherboard partition 12. All fans' native FG interfaces are uniformly connected to the fan signal acquisition circuit of the partition temperature control driver module, requiring no additional wiring and not obstructing the original airflow.

[0024] The AI ​​intelligent monitoring box 2, which has a built-in AI edge computing module for preprocessing multi-source data, an LCD visualization interaction module, a local storage module, and a chassis power supply unit, are electrically connected to the multi-channel sensing acquisition module, the AI ​​edge computing module, the LCD visualization interaction module, and the local storage module. The AI ​​edge computing module includes a synchronous trigger chip and a trend modeling unit for establishing time-series curves of temperature, humidity, dust, vibration, and temperature; a root cause localization unit for comparing the full-domain time-series data to output the root cause of the fault; and a health scoring unit for calculating the overall health of the machine in six dimensions. The trend modeling unit, the root cause localization unit, and the health scoring unit are all electrically connected to the synchronous trigger chip.

[0025] The AI ​​edge computing module is integrated into the main control chip inside the AI ​​Smart Monitoring Box 2. The main control chip is an STM32F407, which has a built-in hardware FPU floating-point unit and a large-capacity on-chip RAM cache. The module also has a built-in synchronization trigger chip. The synchronization trigger chip is an SN74AHC74, which is electrically connected to the trend modeling unit, load prediction unit, grade determination unit, root cause localization unit, and health scoring unit. It uniformly receives the operation output data of each unit and outputs synchronization timing signals to synchronously drive the zone temperature control drive module, LCD visualization interaction module, and remote communication module.

[0026] The trend modeling unit is integrated into the main control chip of the AI ​​edge computing module. It performs sliding window caching and multi-dimensional time series curve fitting through the RAM built into the main control chip, and uses exponential moving average (EMA) to smooth and suppress instantaneous load shocks and short-term vibration interference, generating smooth and continuous temperature and humidity curves, dust accumulation curves, vibration fluctuation curves, and zoned temperature time series curves. The trend modeling unit builds a ring-shaped FIFO sliding buffer through the built-in RAM of the main control chip and sets a 72-hour time series sliding window. Every 2 seconds, the system receives a set of digital temperature, humidity, dust, vibration and temperature data of four zones uploaded by a multi-channel sensor acquisition module. New data is written to the RAM buffer, and old time series data that exceeds 72 hours is automatically overwritten and discarded, so as to continuously retain continuous and uninterrupted time series samples.

[0027] The trend modeling unit uses multi-dimensional time-series curve fitting algorithms and multi-dimensional feature extraction algorithms to fit temperature, humidity, and dust concentration using second-order polynomial piecewise fitting, and uses the EMA exponential moving average algorithm to smooth partition temperature and vibration acceleration. It filters out instantaneous interference caused by instantaneous CPU full load and external force impact on the chassis, generating smooth and continuous temperature and humidity change curves, dust accumulation curves, vibration fluctuation curves, and partition temperature time-series curves for CPU, GPU11, power supply compartment 8, and motherboard 12.

[0028] After curve fitting is completed, the output is divided into two paths: one path writes the original timing data and curve fitting parameters into the local storage module Flash for long-term storage via the SPI bus; the other path calculates the quantitative trend characteristics of each indicator, such as the mean, slope of change, fluctuation amplitude, and duration of continuous over-limit through the multi-dimensional feature extraction subunit. The feature data packets are uniformly transmitted to the synchronous trigger chip, which then synchronously distributes them to the load prediction unit, the grade judgment unit, the root cause localization unit, and the health scoring unit.

[0029] The root cause localization unit is integrated inside the main control chip of the AI ​​edge computing module, and root cause localization is achieved through the following steps: A1. The fault feature templates pre-stored in the local storage module are read through the built-in computing resources of the main control chip and the cache is updated. The fault feature library retrieval root cause location unit is started and reads the fault features pre-stored in the local storage module through the SPI bus. The fault features include the indicator feature range and judgment logic corresponding to air duct blockage, fan deterioration, power supply abnormality, temperature and humidity exceeding the standard, and sensor failure. The fault features are stored in the unit's internal cache for later use.

[0030] A2. Receives full-dimensional time-series data and quantitative features output by the trend modeling unit, distinguishing between instantaneous interference and persistent anomalies through horizontal and vertical bidirectional time-series comparison; receives full-domain time-series data from the upstream trend modeling unit, including temperature and humidity, dust, vibration, zone temperature, power supply, and fan speed, along with their full-dimensional time-series curves, slopes, and fluctuation amplitude quantitative features, establishing a real-time 72-hour historical time-series dataset. Performs bidirectional comparison of full-domain time-series features, horizontally comparing the current real-time values ​​of each indicator and vertically comparing the 72-hour historical trends; distinguishes between instantaneous pulse interference and long-term persistent indicator deviations, marking single-indicator anomalies and multi-indicator synchronous linkage anomalies as two sets of data.

[0031] A3. Fault root cause determination is completed based on logical hierarchical matching; multi-logical hierarchical fault matching and determination calls the fault template in the cache, using AND / or multi-condition logical hierarchical matching: a: If the temperature and humidity exceed the limits, it is determined that there is a risk of condensation or static electricity in the environment; b: If dust continues to rise and the zone temperature increases, then the dust filter screen 10 in the air duct is determined to be blocked; c: If the vibration amplitude exceeds the standard for a long period of time and the fan speed drifts, it is determined that the fan is loose or the bearing is worn. d: If the power supply voltage deviates or the ripple continuously exceeds the standard, the power supply is considered to be aging; e: If the value of a single sensor remains constant without fluctuation over a long period of time, the sensor is considered to be faulty.

[0032] A4. The generated root cause results are encapsulated as standardized fault data packets and transmitted to the synchronous trigger chip and then sent to the LCD visualization interaction module for display. On the other hand, they are written to the local storage module via the SPI bus to complete the archiving and retention of timestamped exception logs.

[0033] The root cause information is packaged and encapsulated to extract the successfully matched fault type, anomaly monitoring dimension, anomaly start and end time, and indicator change slope, and assembled into a standardized fault root cause text data package.

[0034] Dual-channel synchronous output: Channel 1: Root cause data packets are transmitted to the synchronous trigger chip and simultaneously sent to the LCD visualization interaction module, where the interface intuitively displays the root cause of the fault; Channel 2: Root cause data packets are written to the local storage module, timestamped, and archived to the exception log for later data tracing and maintenance statistics.

[0035] By setting up a trend modeling unit to construct multi-dimensional time-series curves, it is possible to restore the long-term change trends of chassis temperature and humidity, dust, vibration, and zone temperature, and identify the slow-accumulating hidden hardware degradation; by comparing the full-domain time-series data with the root cause localization unit, it can automatically output the root cause of the fault, saving the step of manual troubleshooting; by using the health scoring unit to calculate the overall health of the machine in six dimensions, the overall wear and tear status of the chassis can be quantified.

[0036] It also includes a partition temperature control drive module for driving the cooling fans of each partition of the chassis. The AI ​​edge computing module is equipped with a partition temperature control computing unit for outputting directional speed adjustment commands, and a load prediction unit for predicting the temperature rise trend based on the load. The partition temperature control computing unit is electrically connected to the partition temperature control drive module. The load prediction unit is placed at the front of the AI ​​computing link and electrically connected to the synchronous trigger chip.

[0037] The load prediction unit reuses the RAM and hardware FPU floating-point unit within the AI ​​edge computing main control chip; the input interface connects to the synchronous trigger chip, and the output connects to the zoned temperature control computing unit, with onboard bus transmission, requiring no external wiring; the prediction is specifically achieved through the following steps: B1. The load prediction unit receives 72-hour sliding window time series data output by the trend modeling unit in real time, including CPU usage, GPU11 load, temperature change slope of each partition, and instantaneous peak characteristics. B2. The load prediction unit has a built-in lightweight time-series prediction model that predicts the peak temperature rise and temperature rise rate of CPU and GPU 11 in the next 5-15 minutes based on historical load fluctuation patterns; it can identify temperature rise impacts in advance for sudden load increases such as rendering, AI computing, and gaming. B3. The predicted peak temperature rise, heating rate, and duration of high load are packaged into a prediction data package and synchronously sent to the partition temperature control calculation unit via the synchronous trigger chip. B4. The zone temperature control calculation, combined with the prediction results, dynamically adjusts the corresponding zone heat dissipation threshold and increases the fan speed in advance.

[0038] The partition temperature control drive module includes an instruction receiving interface for receiving AI edge computing speed adjustment instructions; the instruction receiving interface adopts a 3.3V / 12V bidirectional level conversion circuit to receive four partition speed adjustment instructions issued by the partition temperature control computing unit of the AI ​​edge computing module. The system includes four independent PWM generators corresponding to the CPU area, GPU 11 area, power supply compartment 8, and motherboard 12; the instruction receive interface output is electrically connected to each of the four independent PWM generators. The power drive circuit is used to amplify the PWM drive signal; the four PWM generators correspond to the CPU cooling fan, GPU11 cooling fan, power supply 8 fan, and motherboard 12 cooling fan respectively, and independently output PWM speed control signals with different duty cycles and frequencies; the output terminals of the four PWM generators are connected to the power drive circuit, which consists of a MOSFET drive array, a freewheeling protection diode, and an RC snubber circuit to amplify the PWM drive power. The output terminal of the power drive circuit is directly connected to the power supply pins of the four native cooling fans of the chassis to achieve stepless fan speed control; Fan signal acquisition loop for interfacing with the chassis's native fan FG interface; A speed feedback circuit is used to provide feedback on the actual speed of each fan group; The fan signal acquisition loop connects to the native FG speed measurement pins of all fans in the chassis. The speed feedback unit acquires the FG pulse speed signal. The input of the speed feedback unit is connected to the FG interface of each fan. The output of the speed feedback unit is sent back to the multi-channel sensor acquisition module to complete the closed-loop acquisition of fan speed. The system can obtain the real fan speed in real time and use AI calculations to judge fan aging and abnormal speed faults.

[0039] The command receiving interface, four independent PWM generators, and power drive circuit are electrically connected in sequence. The output of the power drive circuit is connected to each cooling fan in the chassis. The input of the speed feedback circuit is connected to the FG interface of the fan. The output of the speed feedback circuit is transmitted back to the multi-channel sensor acquisition module.

[0040] The input end of the partition temperature control calculation unit is connected to the synchronous trigger chip to receive the predicted and temperature timing data sent synchronously; the output end is electrically connected to the instruction receiving interface of the partition temperature control drive module. The main control chip's built-in RAM caches the partition predicted temperature rise data output by the prediction unit and the real-time partition temperature and timing change slope sent by the trend modeling unit. At the same time, it reads the adaptive safety temperature thresholds of the CPU area, GPU11 area, power supply compartment 8, and motherboard 12 areas pre-stored in the local storage module in real time through SPI, and updates the temperature comparison parameters of the four areas synchronously with a 2-second calculation cycle. The zone temperature control calculation unit is equipped with a zone temperature difference calculation algorithm and a PWM duty cycle segment conversion algorithm. It independently calculates the real-time temperature difference for the four temperature control zones, corrects the temperature difference value by combining the predicted temperature rise peak, and converts the corresponding PWM frequency and duty cycle parameters according to the temperature difference interval. At the same time, it receives warnings and alarm flags from the graded judgment unit to dynamically correct the speed adjustment parameters. Mild warnings slightly increase the air volume of the corresponding zone, and severe alarms force the zone fan to run at full speed. Multiple indicators are linked to output full-speed heat dissipation parameters for the entire domain in case of abnormality. After parameter conversion, a unified package is generated to produce four independent partition PWM speed control command data packets. Each packet is output to the command receiving interface of the partition temperature control drive module as the control basis for stepless speed regulation of the hardware fan, realizing differentiated pre-heating control in different regions.

[0041] The sensing unit of the multi-channel sensing acquisition module includes a temperature and humidity sensor, a dust concentration sensor 5, and a vibration sensor 6. The temperature and humidity sensor includes, for example, a temperature and humidity sensor 5, a dust concentration sensor 6, and a vibration sensor 7. Figure 1 The air intake temperature and humidity sensor 3 and the internal temperature and humidity sensor 4, the temperature and humidity sensor, the dust concentration sensor 5 and the vibration sensor 6 are electrically connected to the multi-channel analog switch, signal conditioning circuit and ADC conversion chip in the multi-channel sensing and acquisition module.

[0042] In this embodiment, the temperature and humidity sensor used is SHT30, which is placed at two locations: the air inlet of the chassis and the power supply compartment 8. Figure 1 As shown, the system collects simulated temperature and humidity signals inside the chassis in real time; the dust concentration sensor 5 uses a GP2Y1014AU0F to collect simulated dust accumulation concentration in the air duct; the vibration sensor 6 is a piezoelectric vibration detection sensor, mounted on the chassis motherboard 12 and power supply casing to collect the vibration amplitude of the entire machine; the power supply monitoring unit samples the 12V and 5V power supply voltages through a precision resistor voltage divider, and collects the ripple and current of the entire power supply circuit with a sampling resistor; the electrostatic sensing unit collects the electrostatic potential inside the chassis; the fan speed sampling unit directly connects to the chassis's native fan FG speed measurement pin to collect the fan speed pulse signal.

[0043] The outputs of the air inlet temperature and humidity sensor 3, the internal temperature and humidity sensor 4, the dust concentration sensor 5, the vibration sensor 6, the power supply monitoring unit, the electrostatic sensing unit, and the fan speed sampling unit are all connected to a multi-channel analog switch. The multi-channel analog switch uses a CD4051 and switches multiple sensor signals in a time-division manner under the control of the acquisition controller. The output of the multi-channel analog switch is connected to a signal conditioning circuit, which consists of an RC low-pass filter circuit, an instrumentation amplifier circuit, and an anti-aliasing filter circuit to filter out sampling noise and amplify weak sensor signals. The dust concentration sensor 5 is embedded in the air intake duct, located between the dust filter 10 and the air intake fan. The sensing surface is perpendicular to the airflow direction, enabling real-time monitoring of the intake dust concentration. The AI ​​unit learns from the dust change rate and predicts the blockage cycle, generating maintenance reminders.

[0044] Vibration sensor 6 is installed in the middle of the rear panel of the chassis, in the central area of ​​the fan mounting area, with a shock-absorbing pad at the bottom. It is used to collect vibration and resonance signals of the whole machine. The AI ​​unit identifies resonance frequency and abnormal vibration through spectrum analysis.

[0045] Individual alarms will be triggered when relative humidity, dust concentration, and vibration acceleration exceed the normal range. The relative humidity range is set as follows: 40% RH–60% RH is normal; <30% RH is a low humidity warning; >70% RH is a high humidity alarm. Dust concentration (PM2.5) ≤75μg / m³ is normal; 75–150μg / m³ is a warning for mild siltation; >150μg / m³ is a warning for severe blockage.

[0046] Vibration acceleration ≤0.3g is normal; 0.3–0.6g indicates slight resonance warning; >0.6g indicates strong resonance alarm. AI identifies resonance frequencies, potential loosening, and external impacts through spectrum analysis and vibration characteristic modeling; it automatically adjusts fan speed to avoid resonance points when abnormalities occur; strong vibration triggers structural protection prompts to reduce solder joint fatigue and micro-cracks in circuit boards; and it records vibration trends over a long period to predict mechanical aging.

[0047] A fan speed of 0 indicates a stop alarm; a deviation from the rated speed of >±300 rpm indicates a bearing deterioration warning.

[0048] The output of the signal conditioning circuit is connected to an ADC conversion chip, which is an ADS1115, to convert multiple analog signals into high-precision digital signals. The ADC conversion chip is electrically connected to the acquisition controller, which is an STM32F103, to complete channel switching timing control and digital data buffering. The acquisition controller uploads all digital sensor data to the AI ​​edge computing module inside the AI ​​intelligent monitoring box 2 via a serial bus.

[0049] This module directly reuses the original power supply circuit of the chassis, and the fan speed acquisition fully reuses the chassis's native fan FG interface, requiring no additional wiring and without changing the chassis airflow structure.

[0050] The LCD visualization interaction module includes a data receiving interface for receiving AI calculation results, an LCD liquid crystal display driver chip, a three-color status indicator light driver unit, an image buffer for caching real-time monitoring images, an image buffer unit for storing custom display images, and a log storage subunit for locally storing all abnormal event logs. The data receiving interface output is electrically connected to the LCD display driver chip, which is ST7789. The driver chip input is connected to the image buffer and the image buffer unit. The image buffer is AL422B, which temporarily buffers the dynamic curve image monitored in real time. The data receiving interface, LCD display driver chip, image buffer, and image buffer unit are electrically connected in sequence. The LCD display driver chip drives the LCD display screen and the three-color status indicator light driver unit, respectively. The log storage subunit is electrically connected to the local storage module.

[0051] The output of the LCD display driver chip drives the integrated LCD display and the three-color status indicator drive unit. The green indicator corresponds to the overall health of the machine, the yellow corresponds to a mild warning, and the red corresponds to a severe alarm, thus displaying the overall operating status of the machine.

[0052] The data receiving interface is electrically connected to the synchronous trigger chip of the AI ​​edge computing module, synchronously receiving all AI computing data packets, including time-series curves, fault root causes, overall health scores, and zone temperature control status, output by the trend modeling unit, root cause localization unit, health scoring unit, and zone temperature control computing unit. The overall power supply unit provides unified power to the LCD visualization interaction module. The log storage subunit communicates bidirectionally with the local storage module, reading pre-stored custom interface images and writing and retrieving full-cycle time-series data and timestamped abnormal logs in real time, enabling local visualization of monitoring data and querying of historical fault records.

[0053] The local storage module uses a W25Q128 Flash memory chip, which is electrically connected to the AI ​​edge computing module and the LCD visualization interaction module via the SPI bus. It is internally divided into multiple independent storage partitions, which are used to store fault feature templates, six-dimensional health scoring weights, 72-hour time series curve data, long-term historical monitoring data, timestamped abnormal fault logs, and custom interface images. It adopts a cyclic overwrite storage mechanism to automatically clean up overdue redundant data.

[0054] It also includes a remote communication module for synchronizing the whole machine monitoring records to remote terminals; The remote communication module includes a wired network port circuit, a wireless WiFi communication unit, a data encryption transmission unit, and a standardized report generation unit; The wired network port circuit and the input terminal of the wireless WiFi communication unit are both connected to the data encryption transmission unit. The data encryption transmission unit receives the whole machine monitoring data output by the AI ​​edge computing module, and the standardized report generation unit outputs maintenance reports that can be exported to remote terminals.

[0055] The wired network port uses an RJ45 Ethernet circuit, and the wireless WiFi unit uses an ESP8266 module. The input terminals of the two communication units are uniformly connected to the data encryption transmission unit. The data encryption transmission unit receives the whole machine time sequence data, health score, fault log, and hardware life data output by the AI ​​edge computing module, and completes the data packaging and encryption through the AES encryption protocol. The standardized report generation unit has built-in standardized operation and maintenance report templates, automatically summarizes the full life cycle monitoring data, fault records, and maintenance reminders, and generates an exportable Excel operation and maintenance report, which is pushed to the remote operation and maintenance terminal through wired / wireless dual channels to realize remote real-time monitoring and data traceability.

[0056] The health scoring unit is integrated inside the main control chip of the AI ​​edge computing module. The main control chip reads the six-dimensional weight parameters pre-stored in the local storage module via SPI, and receives the mean, slope of change, and fluctuation amplitude of each indicator output by the trend modeling unit, as well as the warning and alarm flags of the grading judgment unit. Each indicator parameter is scored in segments according to the threshold range, and finally the overall health score is obtained by weighting. The health scoring unit is integrated inside the main control chip. After the device is powered on, it reads the six-dimensional weight parameters pre-stored in the local storage module through the SPI bus. The six-dimensional weight parameters include temperature and humidity parameters, dust parameters, vibration parameters, power supply ripple parameters, fan status parameters, and zone temperature parameters, with a total weight of 100.

[0057] The health scoring unit receives the mean, slope, and fluctuation amplitude of each indicator from the trend modeling unit, and simultaneously receives warning and alarm flags from the grading judgment unit. It scores independently in segments according to preset threshold intervals; indicators within the standard range receive full marks, mild warnings result in a slight linear deduction, and severe alarms result in a significant deduction. After completing the sub-item scoring, it performs a weighted sum based on the corresponding weights, outputting a comprehensive health score for the entire machine in the range of 0-100, and classifying it into three health levels: 90-100 points for overall machine health, 70-89 points for mild warning, and below 70 points for severe alarm. Specifically: Temperature and humidity dimension: 40%RH-60%RH gets full marks, deviation from the range results in a linear decrease in points with the amount of deviation; Dust level: ≤75μg / m³ (full marks), the higher the concentration, the lower the score; Vibration dimension: ≤0.3g is full marks, and points will be deducted for increasing vibration amplitude; Power supply dimensions: Voltage and ripple within the standard range receive full marks; deviation and excessive ripple result in deductions. Fan dimension: No speed drift and no vibration are full marks; the greater the speed deviation, the lower the score. Zoned Temperature: All four zones have normal temperatures, which is a perfect score. However, prolonged periods of high temperatures will result in a deduction of points.

[0058] Weighted formula: Overall health score = Temperature and humidity score × weight 1 + Dust score × weight 2 + Vibration score × weight 3 + Power supply score × weight 4 + Fan score × weight 5 + Zone temperature score × weight 6, Calculation result range 0-100.

[0059] 90-100: Healthy (Green Light); 70-89: Mild warning (yellow light); <70: Severe alarm (red light), bound to the graded judgment status.

[0060] The amount of points deducted and the weighting of the scores can be preset and adjusted according to the actual usage environment.

[0061] The health score and level mark are output in two ways: one way is sent to the synchronous trigger chip and sent to the LCD visualization interaction module for display; the other way is written to the local storage module via SPI and saved together with the timing log, which is used to calculate the remaining lifespan of the fan, power supply and motherboard 12 by the life decay model.

[0062] The chassis power supply unit reuses the original 12V power supply circuit of the chassis and is equipped with multiple DC-DC voltage regulator circuits to convert and output stable 3.3V and 5V working voltages respectively, providing unified power supply for the multi-channel sensor acquisition module, AI intelligent monitoring box 2, zone temperature control drive module, LCD visualization interaction module, local storage module, and remote communication module; the entire monitoring system does not require an additional external power adapter and works entirely based on the chassis's native power supply.

[0063] An AI-powered, multi-dimensional chassis monitoring method includes the following steps: S1. Upon power-on, the multi-channel sensor acquisition module starts synchronously. Temperature, humidity, dust, vibration, power supply, and electrostatic sensors simultaneously acquire raw analog signals from inside the chassis. Multiple analog switches select multiple sensor signals in a time-division manner. After being filtered and amplified by the signal conditioning circuit and digitized by the ADC conversion chip, the acquisition controller uploads the digitized sensor data in batches to the AI ​​edge computing module inside the AI ​​intelligent monitoring box 2. At the same time, the speed feedback unit acquires the speed pulse signal from the chassis fan's FG interface and sends it back to the acquisition module for synchronous uploading. After S2 and AI preprocess all digital sensor data, they construct continuous time-series curves for temperature and humidity, dust, vibration, and hardware load. The AI ​​edge computing module receives all digital sensor data and sends it to the trend modeling unit. Relying on the 72-hour circular sliding window to cache time-series data, it generates continuous time-series curves for temperature and humidity, dust, vibration, and CPU / GPU11 / power supply compartment 8 / motherboard 12 partition temperature through EMA smoothing and polynomial fitting algorithms. It extracts the trend characteristics of each indicator and the feature data packet is synchronously sent to the load prediction unit, the grade judgment unit, the root cause localization unit, and the health scoring unit via the synchronous trigger chip. S3, the load prediction unit receives hardware timing load characteristics and predicts short-term temperature rise trends based on CPU and GPU timing load; the partition temperature control calculation unit combines real-time partition temperature, adaptive safety threshold, and temperature rise prediction data to output four directional PWM speed control commands to the partition temperature control drive module.

[0064] S4 and AI divide the CPU, GPU, power supply compartment, and motherboard into four temperature control zones. They compare the real-time temperature of each zone with the adaptive threshold and output independent PWM speed control commands to the zone temperature control driver module. The zone temperature control driver module receives the speed control commands through the command receiving interface. The four independent PWM generators generate corresponding duty cycle speed control signals, which are amplified by the power drive circuit and drive the corresponding zone cooling fans to steplessly adjust their speed. The fan FG interface continuously feeds back the actual speed to the speed feedback unit, which then transmits the data to the multi-channel sensor acquisition module to form a closed-loop temperature control adjustment. S5. Compare real-time sensor data with adaptive thresholds to distinguish between single-item mild warnings, single-item severe alarms, and multi-indicator synchronous linkage anomalies, identify early silent faults, and classify and judge the real-time sensor indicators with adaptive safety thresholds to distinguish between single-item mild warnings, single-item severe alarms, and multi-indicator synchronous linkage anomalies, mark early silent faults in the chassis, and push the anomaly mark synchronously to the root cause location unit and health scoring unit. S6. The system implements protection strategies based on the level of anomalies: for mild warnings, the fan speed is automatically adjusted to improve heat dissipation; for severe alarms, the corresponding zone fan is controlled to run at full speed to dissipate heat, and the yellow indicator light on the LCD panel flashes continuously; for multiple indicators, the system initiates full-domain linkage protection when an anomaly occurs, with four fans running at full load to enhance heat dissipation, locking the power supply safety zone of the whole machine to suppress resonance, and the red indicator light on the panel continuously alarming, while simultaneously generating an emergency anomaly data packet. S7. After triggering a warning / alarm, the root cause location unit retrieves the local storage fault feature template, compares the full-dimensional time-series features horizontally and vertically to complete the fault root cause matching, encapsulates standardized fault records, pushes them synchronously to the LCD panel to display the fault text, and writes them to the local storage module to archive the timestamped abnormal log. The S8 system uses a six-dimensional data weighted calculation to determine the real-time health score of the entire machine. It calls a lifespan decay model to estimate the remaining lifespan of the fan, power supply, and motherboard (12). When the lifespan reaches a threshold, it pushes a maintenance reminder. The health score unit reads the six-dimensional weighted parameters stored locally, combines the real-time values, trends, and graded abnormality markers of each sensor indicator, and scores them separately to calculate a weighted real-time health score of 0-100 for the entire machine. At the same time, it calls the built-in lifespan decay model to estimate the remaining lifespan of the fan, power supply, and motherboard (12) based on long-term dust, vibration, and temperature time-series data. When the lifespan is close to the threshold, it pushes panel maintenance reminders and remote maintenance prompts. The S9 features a real-time rotating display of comprehensive monitoring data on an LCD visual interactive panel. Three-color indicator lights synchronously reflect the overall system status. Time-series data and anomaly logs are automatically written to the local storage module for long-term caching. The integrated LCD visual interactive module synchronously receives all AI calculation results through a data receiving interface. The LCD display rotates six-dimensional real-time monitoring curves, zone temperatures, overall system health score, fault root causes, and remaining hardware lifespan information. The three-color indicator lights synchronously match the overall system health, warning, and alarm status. All time-series data and anomaly logs are continuously written to the local storage module for long-term caching, and the panel supports local retrieval of historical maintenance records. S10: The remote communication module synchronizes the whole machine data and abnormal logs to the remote terminal, and automatically generates hardware operation and maintenance statistical reports for operation and maintenance traceability and analysis. The remote communication module receives AI edge computing whole machine monitoring data, fault logs and health scores in real time. The standardized report generation unit automatically summarizes the periodic operation and maintenance data to generate exportable reports, which are synchronously pushed to the remote operation and maintenance terminal through wired network port or wireless WiFi, realizing remote real-time monitoring, historical fault tracing and equipment periodic operation and maintenance statistical analysis.

[0065] Finally, it should be noted that the above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art can still modify the technical solutions described in the foregoing embodiments or make equivalent substitutions for some of the technical features. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the protection scope of the present invention.

Claims

1. An AI-powered, multi-dimensional chassis monitoring system, characterized in that: This includes a multi-channel sensor acquisition module for collecting data on the entire environment of the chassis and reusing the original power supply and fan FG interface of the chassis; The system includes an AI intelligent monitoring box with a built-in AI edge computing module for preprocessing multi-source data, an LCD visualization interaction module, a local storage module, and a chassis power supply unit. The chassis power supply unit is electrically connected to the multi-channel sensing acquisition module, the AI ​​edge computing module, the LCD visualization interaction module, and the local storage module. The AI ​​edge computing module includes a synchronous trigger chip and a trend modeling unit for establishing time-series curves of temperature, humidity, dust, vibration, and temperature; a root cause localization unit for comparing the full-domain time-series data to output the root cause of the fault; and a health scoring unit for calculating the overall health of the machine in six dimensions. The trend modeling unit, the root cause localization unit, and the health scoring unit are all electrically connected to the synchronous trigger chip.

2. The AI-powered all-dimensional chassis monitoring system according to claim 1, characterized in that: It also includes a partition temperature control drive module for driving the cooling fans of each partition of the chassis. The AI ​​edge computing module is equipped with a partition temperature control computing unit for outputting directional speed adjustment commands, and a load prediction unit for predicting the temperature rise trend based on the load. The partition temperature control computing unit is electrically connected to the partition temperature control drive module. The load prediction unit is placed at the front of the AI ​​computing link and electrically connected to the synchronous trigger chip.

3. The AI-powered all-dimensional chassis monitoring system according to claim 2, characterized in that: The partition temperature control drive module includes an instruction receiving interface for receiving AI edge computing speed adjustment instructions; Four independent PWM generators corresponding to the CPU area, GPU area, power supply compartment and motherboard; Power drive circuit used to amplify PWM drive signals; Fan signal acquisition loop for interfacing with the chassis's native fan FG interface; A speed feedback circuit is used to provide feedback on the actual speed of each fan group; The command receiving interface, four independent PWM generators, and power drive circuit are electrically connected in sequence. The output of the power drive circuit is connected to each cooling fan in the chassis. The input of the speed feedback circuit is connected to the FG interface of the fan. The output of the speed feedback circuit is transmitted back to the multi-channel sensor acquisition module.

4. The AI-powered all-dimensional chassis monitoring system according to claim 3, characterized in that: The sensing and acquisition unit of the multi-channel sensing and acquisition module includes a temperature and humidity sensor, a dust concentration sensor, and a vibration sensor. The temperature and humidity sensor, the dust concentration sensor, and the vibration sensor are electrically connected to the multi-channel analog switch, the signal conditioning circuit, and the ADC conversion chip in the multi-channel sensing and acquisition module.

5. The AI-powered all-dimensional chassis monitoring system according to claim 4, characterized in that: The LCD visualization interaction module includes a data receiving interface for receiving AI calculation results, an LCD liquid crystal display driver chip, a three-color status indicator light driver unit, an image buffer for caching real-time monitoring images, an image buffer unit for storing custom display images, and a log storage subunit for locally storing all abnormal event logs. The data receiving interface, LCD display driver chip, image buffer, and image buffer unit are electrically connected in sequence. The LCD display driver chip drives the LCD display screen and the three-color status indicator light driver unit, respectively. The log storage subunit is electrically connected to the local storage module.

6. The AI-powered all-dimensional chassis monitoring system according to claim 5, characterized in that: It also includes a remote communication module for synchronizing the whole machine monitoring records to remote terminals; The remote communication module includes a wired network port circuit, a wireless WiFi communication unit, a data encryption transmission unit, and a standardized report generation unit; The wired network port circuit and the input terminal of the wireless WiFi communication unit are both connected to the data encryption transmission unit. The data encryption transmission unit receives the whole machine monitoring data output by the AI ​​edge computing module, and the standardized report generation unit outputs maintenance reports that can be exported to remote terminals.

7. The AI-powered all-dimensional chassis monitoring system according to claim 6, characterized in that: The trend modeling unit is integrated into the main control chip of the AI ​​edge computing module. It performs sliding window caching and multi-dimensional time series curve fitting through the RAM built into the main control chip, and uses exponential moving average (EMA) to smooth and suppress instantaneous load shocks and short-term vibration interference, generating smooth and continuous temperature and humidity curves, dust accumulation curves, vibration fluctuation curves, and zoned temperature time series curves. The generated results are connected to the local storage module via the SPI bus, and trend features are obtained through multi-dimensional feature extraction. The feature extraction results are output to the synchronous trigger chip and synchronously sent to the load prediction unit, the grade judgment unit, the root cause localization unit, and the health scoring unit.

8. The AI-powered all-dimensional chassis monitoring system according to claim 7, characterized in that: The root cause localization unit is integrated inside the main control chip of the AI ​​edge computing module, and root cause localization is achieved through the following steps: A1. Read the fault feature templates pre-stored in the local storage module and update the cache using the built-in computing resources of the main control chip; A2. Receive the full-dimensional time series data and quantitative features output by the trend modeling unit, and distinguish between instantaneous interference and persistent anomalies through horizontal and vertical bidirectional time series comparison; A3. Determine the root cause of the fault based on logical hierarchical matching; A4. The generated root cause results are encapsulated as standardized fault data packets and transmitted to the synchronous trigger chip and then sent to the LCD visualization interaction module for display. On the other hand, they are written to the local storage module via the SPI bus to complete the archiving and retention of timestamped exception logs.

9. The AI-powered all-dimensional chassis monitoring system according to claim 8, characterized in that: The health scoring unit is integrated inside the main control chip of the AI ​​edge computing module. The main control chip reads the six-dimensional weight parameters pre-stored in the local storage module via SPI, and receives the mean, slope of change, and fluctuation amplitude of each indicator output by the trend modeling unit, as well as the warning and alarm flags of the grading judgment unit. Each indicator parameter is scored in segments according to the threshold range, and finally the overall health score is obtained by weighting. The six-dimensional weighted parameters include temperature and humidity parameters, dust parameters, vibration parameters, power supply ripple parameters, fan status parameters, and zone temperature parameters.

10. An AI-powered full-dimensional chassis monitoring method, employing the AI-powered full-dimensional chassis monitoring system according to any one of claims 1-9, characterized in that, Includes the following steps: S1. The multi-channel sensor acquisition module synchronously acquires raw analog signals of chassis temperature and humidity, dust, vibration, power supply, static electricity, and fan speed. After signal conditioning and ADC digitization, the signals are uploaded in batches to the AI ​​edge computing module. S2 and AI preprocess all digital sensor data to construct continuous time-series curves for temperature, humidity, dust, vibration, and hardware load. S3. Predicts short-term temperature rise trends based on CPU and GPU timing load; S4 and AI divide the CPU, GPU, power supply compartment and motherboard into four temperature control zones, compare the real-time temperature of the zone with the adaptive threshold, and output independent PWM speed adjustment commands to the zone temperature control driver module. S5. Compare real-time sensor data with adaptive thresholds to distinguish between single-item mild warnings, single-item severe alarms, and multi-indicator synchronous linkage anomalies, and identify early silent faults. S6: Mild warning automatically adjusts fan speed to improve heat dissipation; single severe alarm controls the corresponding zone fan to cool at full speed, and the yellow indicator light on the LCD panel flashes continuously; multi-indicator synchronous linkage abnormal start full-domain linkage protection, four fans at full load to enhance heat dissipation, lock the whole machine power supply safety range to suppress resonance, the red indicator light on the panel continuously alarms, and simultaneously generates emergency abnormal data packets. S7. After the alarm is triggered, AI compares the full-dimensional time sequence features and outputs standardized fault root cause records to be stored in the log. S8 uses six-dimensional data weighted calculation to obtain the real-time health score of the whole machine, calls the life decay model to estimate the remaining lifespan of the fan, power supply and motherboard, and pushes maintenance reminders when the threshold is reached. The S9 LCD visual interactive panel displays real-time, multi-dimensional monitoring data in a rotating manner, and the three-color indicator lights synchronously reflect the overall status of the machine. Time-series data and abnormal logs are automatically written to the local storage module for long-term caching. S10, the remote communication module synchronizes the whole machine data and abnormal logs to the remote terminal, and automatically generates hardware operation and maintenance statistical reports for operation and maintenance traceability and analysis.

Images