A server cabinet fault monitoring system based on big data

Abstract

This invention specifically relates to a server rack fault monitoring system based on big data, belonging to the field of server rack fault monitoring technology. It includes: a data acquisition and synchronization module; a data preprocessing and storage module; a core intelligent analysis and diagnosis module configured to identify sensor faults based on preprocessed data and a big data model; and an alarm and visualization interaction module. In this invention, the correlation analysis between temperature and light intensity, along with a multi-dimensional data cross-validation mechanism, overcomes the limitations of traditional single temperature threshold judgment. On one hand, by combining contact, non-contact, and redundant sensors to collect temperature data, and simultaneously acquiring parameters such as light intensity and attenuation rate through a laser emitter and photodetector, a correlation model between temperature and light intensity is constructed, which can accurately distinguish between actual temperature anomalies and sensor malfunctions. On the other hand, it simultaneously collects correlation data such as sensor power supply voltage and communication signal strength, providing a basis for fault tracing.

Description

Technical Field

[0001] This invention relates to the field of server rack fault monitoring technology, and in particular to a server rack fault monitoring system based on big data. Background Technology

[0002] With the rapid development of the digital economy, the equipment density and computing power requirements of data center server racks continue to rise, making internal temperature stability a core factor in ensuring reliable equipment operation. Current mainstream rack fault monitoring systems primarily rely on temperature sensors to collect data and determine equipment status based on a single temperature threshold; however, this approach has significant limitations:

[0003] On the one hand, it is impossible to effectively distinguish between the actual abnormal temperature of the cabinet and the fault of the temperature sensor itself (such as zero drift, poor line contact, aging failure). False alarms from the sensor often lead to a waste of maintenance resources, or the sensor may fail to report, causing the equipment to overheat and crash. On the other hand, only collecting temperature values ​​while ignoring related parameters such as sensor power supply voltage and communication signal strength makes it difficult to trace the root cause of the fault, increasing the complexity of problem localization.

[0004] At the same time, existing analytical models lack dynamic adaptability and are unable to cope with the complex scenarios of long-term rack operation.

[0005] For example, the rack load fluctuates with business (such as the difference in computing power between weekday peaks and nighttime off-peaks), which causes the temperature baseline to change dynamically. Equipment aging (such as sensor accuracy decay and laser emitter light intensity reduction) will cause the original threshold to gradually become invalid. Traditional fixed models cannot adjust the judgment logic in real time, resulting in a continuous decline in long-term monitoring accuracy.

[0006] In addition, alarm information is mostly presented in the form of single text, lacking a hierarchical mechanism and visualization support. This makes it difficult for operation and maintenance personnel to quickly locate the fault location and assess the scope of impact, further prolonging the fault response time and posing a potential risk to the continuous operation of the data center. Summary of the Invention

[0007] The purpose of this invention is to provide a server rack fault monitoring system based on big data in order to solve the above-mentioned problems.

[0008] In order to achieve the above object, the present invention adopts the following technical solutions:

[0009] A server rack fault monitoring system based on big data includes:

[0010] The data acquisition and synchronization module is configured to capture the status data of the temperature sensors inside the cabinet through multi-dimensional hardware deployment and a precise synchronization mechanism.

[0011] The data preprocessing and storage module is configured to perform deep cleaning and feature engineering on the raw data, build a dataset, and achieve storage and fast retrieval.

[0012] The core intelligent analysis and diagnosis module is configured to identify sensor faults based on preprocessed data and big data models;

[0013] The alarm and visualization interaction module is configured to transform analysis results into alarm information and visualization charts to support operation and maintenance decisions and realize human-computer interaction.

[0014] Preferably, the data acquisition and synchronization module specifically includes:

[0015] Based on the cabinet size, deploy sensors in critical areas, including contact sensors, non-contact infrared sensors, and redundant backup sensors;

[0016] Synchronously acquire sensor operating voltage, communication signal strength, and installation coordinates;

[0017] A red laser and a silicon-based photodetector are mounted in pairs on both sides of the temperature sensor;

[0018] Collect relevant parameters, including original light intensity value, light intensity fluctuation frequency, ambient light background value, and laser emitter operating current;

[0019] Record the initial laser transmittance of each sensor during installation;

[0020] A timestamp is added to each frame of data to ensure time alignment between temperature and light intensity data, and the data is transmitted.

[0021] Data integrity verification is performed on the gateway side, and outliers exceeding the threshold are cached locally and uploaded with priority.

[0022] Preferably, the data preprocessing and storage module specifically includes:

[0023] Data cleaning includes noise suppression, missing value handling, and outlier identification;

[0024] Feature extraction includes temporal feature extraction and spatial feature extraction;

[0025] Store data and preprocessing results within a preset time period; store historical data, segmented by time; store extracted feature vectors for model training and trend analysis;

[0026] A three-level index is built based on rack ID, sensor ID, and timestamp, supporting multi-dimensional retrieval.

[0027] Preferably, the core intelligent analysis and diagnosis module specifically includes:

[0028] By constructing a temperature and light intensity benchmark correlation model using historical normal data, and then using abnormal data to iteratively optimize the model boundary and judgment logic, accurate identification of temperature sensor faults can be achieved when the laser is normal.

[0029] Preferably, step one, establishing a basic model based on historical normal data, specifically includes:

[0030] Select stable operating data within a preset time period after sensor installation, and the data must include the complete load cycle;

[0031] Exclude data from the following time periods:

[0032] Abnormal laser intensity or temperature sensor data deviating from the normal range; abnormal cabinet environment; abnormal data acquisition;

[0033] Model components:

[0034] Single-dimensional normal baseline for temperature: Statistical analysis of temperature distribution over various time periods, fitted temperature, and load correlation curves; output of normal temperature range and load response threshold;

[0035] Single-dimensional normal baseline for light intensity: statistically analyzes the long-term attenuation trend of light intensity and calculates the characteristics of light intensity fluctuation; outputs the normal range and fluctuation threshold of light intensity.

[0036] Temperature-light intensity correlation model: The weak correlation was calculated using the Pearson correlation coefficient, and the correlation formula was fitted using multiple linear regression.

[0037] Predicted light intensity = α·temperature + β·initial light intensity + error term;

[0038] Output the range of correlation coefficients and the normal fluctuation range of light intensity with temperature.

[0039] Preferably, step two: optimizing the model with abnormal data, specifically includes:

[0040] The following two types of outlier data were manually labeled to extract fault features that can be used for model optimization:

[0041] Temperature sensor is normal but laser is abnormal: The annotation includes the type of laser failure, the start and end time of the abnormal light intensity, and the trend of change; the key features extracted are abnormal light intensity features and normal temperature features.

[0042] Temperature sensor malfunction but laser function: The annotation includes the type of temperature sensor malfunction and the manifestation of the temperature anomaly; the key features extracted are temperature anomaly features and normal light intensity features.

[0043] Correcting the temperature anomaly detection threshold: Using data from a malfunctioning temperature sensor but a functioning laser, calibrate the temperature anomaly detection threshold.

[0044] Enhance the reliability of determining normal light intensity: use data from a normal temperature sensor but an abnormal laser to clarify the core characteristics of a normal laser;

[0045] Construct a feature template for temperature faults when the laser is normal: Extract typical features from data where the temperature sensor is abnormal but the laser is normal to form a unique judgment template.

[0046] Preferably, step three: determining the temperature sensor malfunction when the laser is functioning normally using the optimized model, specifically includes the following process:

[0047] By using a basic model and optimized feature templates, accurate judgment is achieved through a three-step process: normal light intensity verification, abnormal temperature identification, and fault type matching.

[0048] A1: Verify if the laser is working properly:

[0049] Core verification conditions:

[0050] Light intensity data is within the single-dimensional baseline range, and the attenuation rate is <0.3% / day;

[0051] The light intensity fluctuations are within normal ranges, and the fluctuation period matches the cabinet vibration.

[0052] The relationship between light intensity and temperature conforms to the basic model, and there is no correlation break.

[0053] A2: Identify if the temperature sensor is malfunctioning:

[0054] Anomaly identification conditions:

[0055] The temperature continues to deviate from the optimized baseline, and there is no synchronous shift between adjacent sensors;

[0056] Temperature does not respond to load changes;

[0057] Temperature time-series characteristics are matched with fault templates;

[0058] A3: Matching temperature sensor fault type:

[0059] By comparing the abnormal temperature characteristics with the temperature fault template when the laser is normal, the specific fault type and confidence level are output.

[0060] Preferably, step four: the model continuous optimization mechanism, specifically includes:

[0061] The model is retrained weekly using newly generated normal data and confirmed abnormal data, and the correlation coefficient between temperature and light intensity is updated.

[0062] When a misjudgment is found during manual review, it is immediately added to the negative sample library, and the judgment weight of the corresponding feature is adjusted.

[0063] For different rack types, sub-models are trained separately to avoid misjudgment across scenarios.

[0064] Preferably, the alarm and visualization interaction module specifically includes:

[0065] The analysis results are transformed into intuitive alarm information and visualization charts; based on the abnormal state monitored by sensors, corresponding warning levels are matched and warnings are issued accordingly.

[0066] Maintenance personnel input the processing results of faulty sensors, adjust alarm thresholds, initiate calibration commands, and upload the input information to the cloud. Management personnel can then download the corresponding faulty sensor information remotely, compile historical data from the faulty sensors, and perform relevant analyses.

[0067] Preferably, the system management and external integration module is configured to ensure stable system operation through device resource management and model parameter management, connect to external systems such as environmental monitoring and maintenance work orders, and ensure data security and operational compliance through user permission management.

[0068] In summary, due to the adoption of the above technical solution, the beneficial effects of the present invention are:

[0069] 1. This invention overcomes the limitations of traditional single temperature threshold judgment by using temperature and light intensity correlation analysis and multi-dimensional data cross-validation mechanism. On the one hand, it combines temperature data collected by contact, non-contact and redundant sensors, and simultaneously obtains parameters such as light intensity and attenuation rate through laser emitter and photodetector to construct a correlation model between temperature and light intensity, which can accurately distinguish between real temperature anomalies and sensor malfunctions. On the other hand, it simultaneously collects correlation data such as sensor power supply voltage and communication signal strength to provide a basis for fault tracing.

[0070] 2. This invention dynamically adapts to long-term scenarios such as cabinet load fluctuations and equipment aging through a basic model and an iterative optimization mechanism based on abnormal data: it constructs temperature and light intensity baselines based on historical normal data, and continuously corrects judgment thresholds and feature templates by combining abnormal data to ensure the model's adaptability to load changes and sensor aging; at the same time, through hierarchical alarms and visual interaction, it enables maintenance personnel to quickly locate fault locations and assess the scope of impact. Attached Figure Description

[0071] Further details, features, and advantages of this application are disclosed in the following description of exemplary embodiments in conjunction with the accompanying drawings, in which:

[0072] Figure 1 This is a system structure diagram of the present invention. Detailed Implementation

[0073] Several embodiments of this application will now be described in more detail with reference to the accompanying drawings to enable those skilled in the art to implement this application. This application may be embodied in many different forms and for various purposes and should not be limited to the embodiments set forth herein. These embodiments are provided to make this application thorough and complete, and to fully convey the scope of this application to those skilled in the art. The embodiments described do not limit this application.

[0074] Unless otherwise defined, all terms used herein (including technical and scientific terms) shall have the same meaning as commonly understood by one of ordinary skill in the art to which this application pertains. It will be further understood that terms such as those defined in commonly used dictionaries shall be interpreted as having a meaning consistent with their meaning in the relevant field and / or the context of this specification, and shall not be interpreted in an idealized or overly formal sense unless expressly defined herein. Example 1

[0075] Its specific implementation method is combined with the appendix Figure 1 Please provide a detailed explanation.

[0076] Appendix Figure 1 This invention provides a structural block diagram of a server rack fault monitoring system based on big data, which shows the connection relationship between the data acquisition and synchronization module and the alarm and visualization interaction module, and marks the main functional interaction flow of each module.

[0077] In this embodiment, it includes:

[0078] The data acquisition and synchronization module is configured to capture comprehensive data on the status of temperature sensors in the cabinet through multi-dimensional hardware deployment and precise synchronization mechanism, providing original data for subsequent analysis.

[0079] Specifically, it includes:

[0080] Based on the rack size (e.g., a standard 42U rack), deploy sensors in critical areas (e.g., CPU-intensive areas, power module areas, and heat dissipation dead zones), including contact sensors, non-contact infrared sensors, and redundant backup sensors.

[0081] Contact sensors (such as PT100 platinum resistance thermometers): directly attached to the surface of the equipment to collect accurate temperature (accuracy ±0.1℃), with a sampling frequency set to 5Hz (to capture instantaneous temperature fluctuations).

[0082] Non-contact infrared sensor: installed on the cabinet beam, collects the average temperature of the area (coverage radius 30cm), sampling frequency 1Hz (to reduce redundant data);

[0083] Redundant backup sensors: Deploy 1 to 2 additional sensors of the same type next to the core device (such as the main server) for cross-validation.

[0084] In addition to temperature (°C), the sensor's operating voltage (to determine power supply stability), communication signal strength (to determine line connection status), and installation coordinates (accurate to U position + left and right offset mm) are collected simultaneously.

[0085] A 650nm red laser (output power 5mW, to avoid thermal effects) and a silicon-based photodetector (response time <10μs) are mounted in pairs on both sides of the temperature sensor (the angle between the laser path and the normal to the sensor surface is ≤5°).

[0086] Collect relevant parameters, including the original light intensity value (mW / cm²), light intensity fluctuation frequency (Hz, reflecting whether the sensor vibrates), ambient light background value (compensated by both light shield and algorithm), and laser emitter operating current (to determine whether the equipment is aging).

[0087] During installation, record the initial laser transmittance of each sensor (unobstructed, at a standard angle) as the benchmark value for subsequent evaluation (±3% is within the normal deviation range).

[0088] The NTP network time protocol (synchronization accuracy ≤ 1ms) is used to attach a precise timestamp to each frame of data to ensure time alignment between temperature and light intensity data (if the error exceeds 50ms, it is marked as synchronization failure), and the data is then transmitted.

[0089] Close-range cabinets (within the same data center): Direct connection via industrial Ethernet (gigabit bandwidth), using MODBUSTCP protocol to encapsulate data frames (including checksum bits to prevent transmission errors);

[0090] Long-distance server rack clusters (across floors / campus): LoRaWAN wireless transmission is used (transmission distance 2-5km), and data is compressed by the edge gateway (compression rate ≥30%) before being uploaded to reduce bandwidth consumption;

[0091] Data integrity verification is performed on the gateway side (data packets with missing key fields are discarded), and abnormal values ​​exceeding the threshold (such as temperature > 80℃) are locally cached and prioritized for uploading to ensure the real-time performance of emergency data.

[0092] The data preprocessing and storage module is configured to perform deep cleaning and feature engineering on the raw data, build a high-quality dataset, and achieve efficient storage and fast retrieval to support subsequent analysis and modeling.

[0093] Specifically, it includes:

[0094] Data cleaning includes noise suppression, missing value handling, and outlier identification;

[0095] Noise suppression:

[0096] Temperature data: Kalman filtering is used (the filter coefficient is dynamically adjusted in conjunction with the rack fan speed cycle) to eliminate high-frequency fluctuations of ±1℃ caused by airflow disturbances;

[0097] Light intensity data: Impulse noise (such as momentary dust occlusion) is removed using the 3σ criterion, and the original value is retained for continuous low light intensity (>5s) (which may be due to actual occlusion).

[0098] Missing value handling:

[0099] Short-term missing data (<10s): Linear interpolation was used (based on the 10 valid data points before and after the missing data).

[0100] Long-term missing (>10s): Initiate backup sensor data filling and mark "Depends on backup source" with the label.

[0101] Outlier identification: The reasonable range is defined by the IQR (interquartile range) method (temperature: average cabinet temperature ±5℃; light intensity: 70%~130% of the initial value). Values ​​outside the range are marked as "outliers to be verified" and are not directly removed (potential fault clues are retained).

[0102] Feature extraction includes temporal feature extraction and spatial feature extraction;

[0103] Temporal feature extraction:

[0104] Temperature characteristics: maximum / minimum / standard deviation within a 1-minute sliding window, rate of temperature rise (°C / min, exceeding 5°C / min is considered abnormal), and temperature phase shift within a 24-hour period (compared with historical data for the same period).

[0105] Light intensity characteristics: light intensity attenuation rate (daily attenuation / initial value), temperature and light intensity cross-correlation coefficient (under normal installation, it should be weakly correlated, |r|<0.3), number of light intensity abrupt changes (more than 3 times within 1 hour is considered abnormal).

[0106] Spatial feature extraction: Calculate the temperature gradient (°C / m) and light intensity difference between adjacent sensors (compared with the theoretical attenuation model) to identify abnormal clustering patterns in local areas (such as the simultaneous decrease in light intensity of multiple sensors in a certain area).

[0107] It stores data and preprocessing results within a preset time period (supporting millisecond-level queries) to meet real-time monitoring needs; it stores historical data (retained for 1 year), sharded by time (one data block per hour), and supports high compression ratios (compression rate ≥ 10:1); it stores extracted feature vectors (such as daily temperature fluctuation features) for model training and trend analysis.

[0108] A three-level index is built based on rack ID, sensor ID, and timestamp, supporting multi-dimensional retrieval (such as "query all sensors in rack A with a light intensity attenuation rate > 2% / day"), with a query response time of ≤ 1 second.

[0109] The core intelligent analysis and diagnosis module is configured to identify sensor faults based on preprocessed data and big data models;

[0110] Specifically, it includes:

[0111] By constructing a temperature and light intensity benchmark correlation model using historical normal data, and then using abnormal data to iteratively optimize the model boundary and judgment logic, accurate identification of temperature sensor faults can be achieved when the laser is normal.

[0112] Step 1: Establish a basic model based on historical normal data, specifically including:

[0113] Select stable operating data within a preset time period after sensor installation (avoiding the initial installation and commissioning period), and it must include the complete load cycle (such as weekdays / weekends, peak / off-peak periods).

[0114] To avoid contaminating the baseline, data from the following periods should be excluded:

[0115] Abnormal laser intensity (e.g., attenuation > 5%, fluctuation > 1%) or temperature sensor data deviating from the normal range (e.g., exceeding baseline ± 2℃); abnormal cabinet environment (e.g., air conditioner failure, fan stoppage, new equipment obstruction); abnormal data acquisition (e.g., communication interruption, synchronization error > 50ms);

[0116] Sample size requirement: ≥100,000 normal data points for each sensor (calculated at a sampling frequency of 1Hz, approximately 12 days of data) to ensure coverage of various normal scenarios (different loads, temperatures, vibration conditions).

[0117] The basic model consists of three parts: "temperature single-dimensional normal baseline", "light intensity single-dimensional normal baseline" and "temperature-light intensity correlation model", forming a three-dimensional judgment benchmark.

[0118] Model components:

[0119] Temperature single-dimensional normal baseline: statistically analyze the temperature distribution over various time periods (e.g., mean 35℃, standard deviation 2℃ for weekdays 9:00-18:00), fit the temperature and load correlation curves (CPU utilization to predicted temperature); output normal temperature range (e.g., 35℃±3℃), load response threshold (e.g., a 10% increase in utilization should be followed by a 0.5℃ increase in temperature).

[0120] Light intensity single-dimensional normal baseline: statistically analyze the long-term light intensity decay trend (normal ≤0.1% / day), calculate the light intensity fluctuation characteristics (standard deviation within 1 minute ≤0.2mW / cm²); output the normal range of light intensity (initial value ±3%), fluctuation threshold (e.g., fluctuation caused by vibration ≤1%).

[0121] Temperature-light intensity correlation model: A weak correlation was calculated using the Pearson correlation coefficient (light intensity fluctuation is allowed to be ±0.5% for every 10°C increase in temperature), and the correlation formula was fitted using multiple linear regression.

[0122] Predicted light intensity = α·temperature + β·initial light intensity + error term (α and β are coefficients);

[0123] Output the correlation coefficient range (e.g., 0.1±0.05) and the normal fluctuation range of light intensity with temperature;

[0124] Formula explanation:

[0125] α·Temperature: Reflects the influence of temperature on light intensity. α is the temperature influence coefficient, meaning that for every unit change in temperature (e.g., 1℃), the light intensity will change by α units (for example, when the unit of light intensity is mW / cm², it is the change in mW / cm²), reflecting the linear correlation between temperature and light intensity.

[0126] β·Initial Light Intensity: The initial light intensity is the baseline value of light intensity initially acquired by the system when the sensor and laser are functioning normally and installed correctly. β is a weighting coefficient for the initial light intensity, used to retain the basic influence of the initial light intensity on the current predicted light intensity, allowing the prediction to take into account the initial state of the equipment.

[0127] Error term: Represents other factors affecting light intensity that the formula cannot accurately cover, such as slight fluctuations in ambient light and extremely minor instabilities (non-fault-related) in the laser device itself. It is a supplement to the deviation between theoretical calculation and actual situation.

[0128] The method for determining parameters α and β:

[0129] Data Fitting: A large amount of historical data (temperature value, corresponding light intensity value, initial light intensity value) was collected when both the temperature sensor and the laser were functioning normally. A linear regression algorithm was used to make the calculated "α·temperature + β·initial light intensity" as close as possible to the actual light intensity. The optimal values ​​of α and β were determined by minimizing the error. For example, using Python's scikit-learn library, temperature and initial light intensity were input as features, and the actual light intensity as the label; training a linear regression model yielded α and β.

[0130] Scenario adaptation adjustment: The degree to which temperature affects light intensity may vary depending on the server rack (server rack, network rack, etc.) and the installation environment (data center temperature stability, vibration, etc.). For example, in a high-temperature data center, temperature changes may have a more significant impact on light intensity. After the basic fit, α and β can be adjusted within a small range according to the actual scenario to make the formula more suitable for the application scenario.

[0131] The role of the error term is to acknowledge the gap between the model and reality, ensuring that the formula is not absolutely idealized, retaining tolerance for complex and random influencing factors, and avoiding ignoring interference that cannot be fully modeled in reality in pursuit of a precise formula, thus making the light intensity prediction more in line with the real situation. On the one hand, during model training, the error term can help us evaluate the model's fitting effect (such as indicators like mean squared error) and guide the optimization of α and β. On the other hand, in practical applications, when the deviation between the actual light intensity and the predicted value is within a reasonable range of the error term, the equipment can be considered to be in normal condition; only if the deviation is too large should it be judged that there may be a malfunction.

[0132] The output format of the basic model:

[0133] Numerical baselines: such as normal temperature range [30℃, 40℃], normal light intensity range;

[0134] Curved baselines: such as the "load and temperature" fitting curve (used to determine whether the temperature changes normally with the load) and the "temperature and light intensity" correlation curve (used to determine whether the light intensity fluctuates normally with the temperature).

[0135] Feature vector library: Extract time-series features from normal data (such as temperature fluctuation within 10 minutes ≤ 2℃, light intensity attenuation rate ≤ 0.05% / hour) to form normal feature templates.

[0136] Step 2: Optimize the model using outlier data, specifically including:

[0137] Manually label the following two types of abnormal data (marking fault type, abnormal time period, and key features) to extract fault features that can be used for model optimization:

[0138] Temperature sensor is normal but laser is abnormal: The annotation includes the type of laser failure (such as transmitter attenuation, power supply failure), the start and end time of the abnormal light intensity, and the trend of change; the key features extracted are the abnormal light intensity features (such as attenuation rate > 0.5% / day, sudden drop without load correlation) and the normal temperature features (still conforming to the single-dimensional baseline).

[0139] Temperature sensor malfunction but laser function: The annotation includes the type of temperature sensor failure (e.g., drift, circuit failure) and the manifestation of temperature anomaly (e.g., fixed offset, fluctuation); the key features extracted are temperature anomaly features (e.g., deviation from baseline > 3℃, no load response) and normal light intensity features (still conforming to the single-dimensional baseline and correlation model).

[0140] Based on the characteristics of abnormal data, the basic model is modified by boundary correction, weight adjustment, and fault feature library supplementation to improve the ability to identify target scenarios (temperature faults when the laser is normal).

[0141] Correcting the temperature anomaly detection threshold (for temperature sensor failure): Using data from a faulty temperature sensor but a functioning laser, calibrate the temperature anomaly detection threshold.

[0142] For example, in the basic model, a temperature deviation of 3°C from the baseline is considered abnormal, but actual data shows that drift faults usually stabilize only when the temperature deviation is greater than 4°C. Therefore, the judgment threshold is adjusted from 3°C to 4°C (to reduce false positives).

[0143] High-frequency fluctuation characteristics of poor line contact (such as standard deviation > 5℃ within 1 minute) are labeled, and a judgment condition of fluctuation frequency > 2Hz and duration > 10 seconds is added to the model.

[0144] Enhance the reliability of normal light intensity determination (eliminate laser malfunction interference): Use data from a normal temperature sensor but an abnormal laser to identify the core characteristics of a normal laser, avoiding misdiagnosis of laser malfunction as a temperature fault.

[0145] For example, when a laser malfunctions, the light intensity may drop suddenly but the temperature remains normal. This type of data can be used to mark the inverse characteristics of normal light intensity (such as no sudden drop, normal attenuation rate, and fluctuations that conform to vibration laws).

[0146] Add a secondary verification logic to the model to confirm normal light intensity: if the light intensity data meets the single-dimensional baseline and is normally correlated with temperature, then the laser is confirmed to be normal (excluding its interference with temperature determination).

[0147] Construct a feature template for temperature faults when the laser is normal: Extract typical features from data where the temperature sensor is abnormal but the laser is normal to form a unique judgment template;

[0148] Measurement drift (zero drift): Light intensity is within the normal range (e.g., ±3%), combined with temperature deviation from baseline >4℃ and fixed (e.g., continuously higher than 5℃), combined with normal temperature-light intensity correlation (light intensity fluctuates with temperature).

[0149] Poor line contact: light intensity fluctuation <1% combined with high-frequency temperature fluctuation (standard deviation >5℃) and communication signal strength fluctuation >20dBm;

[0150] Hardware failure: unstable light intensity (fluctuation <0.5%) combined with a fixed temperature (e.g., 125℃) and abnormal power supply voltage (<3V).

[0151] Step 3: After optimization, the model determines if the temperature sensor is faulty when the laser is functioning normally. The specific process includes:

[0152] By using a basic model and optimized feature templates, accurate judgment is achieved through a three-step process: normal light intensity verification, abnormal temperature identification, and fault type matching.

[0153] A1: Verify if the laser is working properly (rule out abnormal laser interference):

[0154] Based on the optimized light intensity normal determination logic, it was confirmed that the laser was not abnormal, thus eliminating interference for subsequent temperature fault determination:

[0155] Core verification conditions (must be met simultaneously):

[0156] The light intensity data is within the single-dimensional baseline range (e.g., initial value ±3%), and the attenuation rate is <0.3% / day (no laser attenuation fault).

[0157] The light intensity fluctuations are within normal range (standard deviation <1% within 1 minute), and the fluctuation period matches the cabinet vibration (such as fan speed) (excluding laser loosening or power supply failure).

[0158] The relationship between light intensity and temperature conforms to the basic model (e.g., light intensity fluctuation within ±0.5% for every 10°C increase in temperature), with no breaks in the correlation (excluding faults such as laser wavelength shift).

[0159] A2: Identify whether the temperature sensor is malfunctioning (based on optimized temperature determination logic):

[0160] After confirming that the laser is functioning normally, anomalies are identified by comparing temperature data with the optimized baseline.

[0161] Anomaly detection conditions (any one of them must be met):

[0162] The temperature continues to deviate from the optimized baseline (e.g., > ±4℃ for 30 consecutive minutes), and there is no synchronous shift between adjacent sensors (excluding environmental anomalies).

[0163] Temperature does not respond to load changes (e.g., when CPU utilization increases from 20% to 80%, the temperature change is <1℃, which is below the normal response threshold).

[0164] Temperature time-series characteristics are matched with fault templates (such as high-frequency fluctuations, fixed value locking);

[0165] A3: Matching temperature sensor fault type (accurate location problem):

[0166] By comparing the abnormal temperature characteristics with the temperature fault template when the laser is normal, the specific fault type and confidence level are output:

[0167] For example: A sensor displays a temperature of 39℃ (baseline 34℃, deviation 5℃), the laser light intensity is normal (10mW / cm²±0.2), and the temperature and light intensity correlation is normal (the light intensity increases slightly by 0.3% with increasing temperature) to match the measurement drift (zero drift) template with a confidence level of 98%.

[0168] If the feature matching degree is less than 60% (such as a new type of fault), a manual review process will be triggered, and the case will be added to the feature library for subsequent model iterations.

[0169] Step 4: Continuous model optimization mechanism, specifically including:

[0170] The model is retrained weekly using newly generated normal data and confirmed abnormal data (after labeling), and the correlation coefficient between temperature and light intensity is updated (to adapt to slight sensor aging).

[0171] When a misjudgment is found during manual review (such as classifying normal temperature fluctuations as a fault), the case is immediately added to the negative sample library, and the judgment weight of the corresponding feature is adjusted (such as reducing the abnormal weight of short-term fluctuations).

[0172] For different rack types (such as high-density server racks and network racks), sub-models are trained separately (due to differences in load and vibration environment, the normal baseline may vary) to avoid misjudgment across scenarios.

[0173] The alarm and visualization interaction module is configured to transform analysis results into intuitive alarm information and visualization charts to support operation and maintenance decisions and achieve efficient human-computer interaction.

[0174] The analysis results are transformed into intuitive alarm information and visual charts to support operation and maintenance decisions and achieve efficient human-computer interaction.

[0175] Emergency Alarm (P1 Level): Sensor failure (such as hardware damage) results in the inability to monitor the temperature of critical areas, or loose installation may cause the equipment to overheat. Triggering conditions: Immediately send an alarm via SMS, telephone, and audible and visual alarm in the computer room, and respond within 15 minutes.

[0176] Important alarm (P2 level): Sensor drift / minor obstruction (light intensity attenuation of 5%~10%), which does not affect basic monitoring but needs to be processed and pushed to the operation and maintenance APP within 30 minutes and displayed at the top of the system homepage;

[0177] Alarm (P3 level): Slight angular deviation (light intensity attenuation <5%) is recorded in the maintenance list and a summary report is generated weekly.

[0178] Alarm details include: faulty sensor ID, rack location (with 3D rack diagram for location), fault type (with data waveform diagram), impact assessment (e.g., "may cause a 2°C deviation in temperature monitoring in this area"), and handling suggestions (e.g., "recalibrate the sensor and adjust the angle using a laser level").

[0179] Rack heat map: The temperature of each area is displayed in color (blue < 25℃, red > 40℃), with anomaly light intensity markers (yellow flashing dots) superimposed.

[0180] Sensor status list: sorted by fault level, displaying key parameters such as temperature value, light intensity attenuation rate, and last calibration time;

[0181] Trend Comparison Chart: The current temperature / light intensity curve of the same sensor is compared with the historical baseline for the same period, visually demonstrating the deviation.

[0182] Historical analysis function:

[0183] Fault statistics report: Statistics on fault type distribution, processing time, and repeat faulty sensors by time period (day / week / month);

[0184] Installation status trend: Displays the curve of sensor light intensity attenuation rate over time, and predicts the maintenance time required (e.g., "The light intensity attenuation is expected to exceed 10% after 15 days").

[0185] Model accuracy analysis: Displays the recognition accuracy of various faults and details of misjudged cases (supports manual labeling and correction).

[0186] Interactive operation: Supports maintenance personnel to manually enter processing results (such as "sensor replaced"), adjust alarm thresholds (for special load cabinets), and initiate calibration commands (triggering the re-acquisition of laser initial values). The entered information is uploaded to the cloud, and administrators can download the information of the corresponding faulty sensors remotely and analyze the historical data of the faulty sensors.

[0187] The system management and external integration module is configured to ensure stable system operation through device resource management and model parameter management, connect to external systems such as environmental monitoring and maintenance work orders, and ensure data security and operational compliance through user permission management.

[0188] Real-time monitoring of the operating status of temperature sensors, laser emitters, and transmission gateways (e.g., automatic alarm when laser operating temperature > 60℃), and recording of equipment lifespan (e.g., prompting when a sensor has been in operation for 18 months and is approaching its 24-month replacement cycle).

[0189] Dynamically allocate computing resources (such as automatically increasing the CPU / memory ratio of the analysis module during peak fault periods) and provide storage capacity warnings (automatically expand storage when remaining space is <20%).

[0190] Critical nodes (such as data transmission gateways) adopt a primary / backup switching mechanism (switching time < 10 seconds) to avoid system interruption due to single point of failure.

[0191] Save historical model versions (at least 5), support rollback (when the accuracy of the new model drops), and record model training logs (training data volume, number of iterations, accuracy).

[0192] Parameter configuration: Allows administrators to adjust core parameters (such as temperature drift judgment threshold and light intensity attenuation alarm threshold), and supports custom configuration templates according to rack type (such as network rack / server rack).

[0193] Sensor fault information is pushed to the computer room air conditioning system via API interface to trigger temporary control (such as increasing the air conditioning fan speed in the fault area).

[0194] Operation and maintenance work order system: Automatically creates fault work orders (including processing priority and required tool list), and synchronizes work order status (such as dispatched, completed) to the monitoring system;

[0195] Log auditing system: Records all operations (such as parameter modification, manual marking) and system events (such as alarm triggering, model update), with log retention time ≥1 year, and supports compliance auditing.

[0196] The access roles are divided into system administrator (configuration parameters, management users), operation and maintenance engineer (handling alarms, performing maintenance), and read-only user (viewing reports, monitoring status);

[0197] Record all key user actions (such as modifying alarm thresholds and marking faults as handled), supporting traceability and auditing;

[0198] Sensitive data (such as rack location coordinates) is stored in encrypted form and transmitted using the HTTPS protocol to prevent information leakage.

[0199] The above formulas are all dimensionless calculations. The formulas are derived from software simulations based on a large amount of collected data to obtain the most recent real-world results. The preset parameters in the formulas are set by those skilled in the art according to the actual situation.

[0200] The foregoing has only described certain exemplary embodiments of the present invention by way of illustration. Undoubtedly, those skilled in the art can modify the described embodiments in various ways without departing from the spirit and scope of the present invention. Therefore, the foregoing drawings and descriptions are illustrative in nature and should not be construed as limiting the scope of protection of the claims of the present invention.

[0201] It should be noted that, in this document, the use of relational terms such as "first" and "second" is merely for distinguishing one entity or operation from another, and does not necessarily require or imply any such actual relationship or order between these entities or operations. Furthermore, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such a process, method, article, or apparatus. Without further limitations, an element defined by the phrase "comprising one..." does not exclude the presence of other identical elements in the process, method, article, or apparatus that includes the element.

[0202] It should be understood that in the various embodiments of this application, the order of the above-mentioned processes does not imply the order of execution. The execution order of each process should be determined by its function and internal logic, and should not constitute any limitation on the implementation process of the embodiments of this application.

[0203] Those skilled in the art will recognize that the units and algorithm steps of the various examples described in conjunction with the embodiments disclosed herein can be implemented in electronic hardware, or a combination of computer software and electronic hardware. Whether these functions are implemented in hardware or software depends on the specific application and design constraints of the technical solution. Those skilled in the art can use different methods to implement the described functions for each specific application, but such implementation should not be considered beyond the scope of this application.

[0204] Those skilled in the art will understand that, for the sake of convenience and brevity, the specific working processes of the systems, devices, and units described above can be referred to the corresponding processes in the foregoing method embodiments, and will not be repeated here.

[0205] The units described as separate components may or may not be physically separate. The components shown as units may or may not be physical units; that is, they may be located in one place or distributed across multiple network units. Some or all of the units can be selected to achieve the purpose of this embodiment according to actual needs.

[0206] In addition, the functional units in the various embodiments of this application can be integrated into one processing unit, or each unit can exist physically separately, or two or more units can be integrated into one unit.

[0207] The above description is merely a specific embodiment of this application, but the scope of protection of this application is not limited thereto. Any variations or substitutions that can be easily conceived by those skilled in the art within the scope of the technology disclosed in this application should be included within the scope of protection of this application. Therefore, the scope of protection of this application should be determined by the scope of the claims.

[0208] The foregoing has only described certain exemplary embodiments of the present invention by way of illustration. Undoubtedly, those skilled in the art can modify the described embodiments in various ways without departing from the spirit and scope of the present invention. Therefore, the foregoing drawings and descriptions are illustrative in nature and should not be construed as limiting the scope of protection of the claims of the present invention.

Claims

1. A server rack fault monitoring system based on big data, characterized in that, include: The data acquisition and synchronization module is configured to capture the status data of the temperature sensors inside the cabinet through multi-dimensional hardware deployment and a precise synchronization mechanism. The data preprocessing and storage module is configured to perform deep cleaning and feature engineering on the raw data, build a dataset, and achieve storage and fast retrieval. The core intelligent analysis and diagnosis module is configured to identify sensor faults based on preprocessed data and big data models; Step 1: Establish a basic model based on historical normal data; Step 2: Optimize the model using outlier data, specifically including: The following two types of outlier data were manually labeled to extract fault features that can be used for model optimization: Temperature sensor is normal but laser is abnormal: The annotation includes the type of laser failure, the start and end time of the abnormal light intensity, and the trend of change; the key features extracted are abnormal light intensity features and normal temperature features. Temperature sensor malfunction but laser function: The annotation includes the type of temperature sensor malfunction and the manifestation of the temperature anomaly; the key features extracted are temperature anomaly features and normal light intensity features. Correcting the temperature anomaly detection threshold: Using data from a malfunctioning temperature sensor but a functioning laser, calibrate the temperature anomaly detection threshold. Enhance the reliability of determining normal light intensity: use data from a normal temperature sensor but an abnormal laser to clarify the core characteristics of a normal laser; Construct a feature template for temperature faults when the laser is normal: Extract typical features from data where the temperature sensor is abnormal but the laser is normal to form a unique judgment template; Step 3: After optimization, the model determines if the temperature sensor is faulty when the laser is functioning normally. The specific process includes: By using a basic model and optimized feature templates, accurate judgment is achieved through a three-step process: normal light intensity verification, abnormal temperature identification, and fault type matching. A1: Verify if the laser is working properly: Core verification conditions: Light intensity data is within the single-dimensional baseline range, and the attenuation rate is <0.3% / day; The light intensity fluctuations are within normal ranges, and the fluctuation period matches the cabinet vibration. The relationship between light intensity and temperature conforms to the basic model, and there is no correlation break. A2: Identify if the temperature sensor is malfunctioning: Anomaly identification conditions: The temperature continues to deviate from the optimized baseline, and there is no synchronous shift between adjacent sensors; Temperature does not respond to load changes; Temperature time-series characteristics are matched with fault templates; A3: Matching temperature sensor fault type: By comparing the abnormal temperature characteristics with the temperature fault template when the laser is normal, the specific fault type and confidence level are output. The alarm and visualization interaction module is configured to transform analysis results into alarm information and visualization charts to support operation and maintenance decisions and realize human-computer interaction.

2. The server rack fault monitoring system based on big data according to claim 1, characterized in that, The data acquisition and synchronization module specifically includes: Based on the cabinet size, deploy sensors in critical areas, including contact sensors, non-contact infrared sensors, and redundant backup sensors; Synchronously acquire sensor operating voltage, communication signal strength, and installation coordinates; A red laser and a silicon-based photodetector are mounted in pairs on both sides of the temperature sensor; Collect relevant parameters, including original light intensity value, light intensity fluctuation frequency, ambient light background value, and laser emitter operating current; Record the initial laser transmittance of each sensor during installation; A timestamp is added to each frame of data to ensure time alignment between temperature and light intensity data, and the data is transmitted. Data integrity verification is performed on the gateway side, and outliers exceeding the threshold are cached locally and uploaded with priority.

3. The server rack fault monitoring system based on big data according to claim 1, characterized in that, The data preprocessing and storage module specifically includes: Data cleaning includes noise suppression, missing value handling, and outlier identification; Feature extraction includes temporal feature extraction and spatial feature extraction; Store data and preprocessing results within a preset time period; store historical data, segmented by time; store extracted feature vectors for model training and trend analysis; A three-level index is built based on rack ID, sensor ID, and timestamp, supporting multi-dimensional retrieval.

4. The server rack fault monitoring system based on big data according to claim 1, characterized in that, The core intelligent analysis and diagnosis module specifically includes: By constructing a temperature and light intensity benchmark correlation model using historical normal data, and then using abnormal data to iteratively optimize the model boundary and judgment logic, accurate identification of temperature sensor faults can be achieved when the laser is normal.

5. A server rack fault monitoring system based on big data according to claim 4, characterized in that, Step 1: Establish a basic model based on historical normal data, specifically including: Select stable operating data within a preset time period after sensor installation, and the data must include the complete load cycle; Exclude data from the following time periods: Abnormal laser intensity or temperature sensor data deviating from the normal range; abnormal cabinet environment; abnormal data acquisition; Model components: Single-dimensional normal baseline for temperature: Statistical analysis of temperature distribution over various time periods, fitted temperature, and load correlation curves; output of normal temperature range and load response threshold; Single-dimensional normal baseline for light intensity: statistically analyzes the long-term attenuation trend of light intensity and calculates the characteristics of light intensity fluctuation; outputs the normal range and fluctuation threshold of light intensity. Temperature-light intensity correlation model: The weak correlation was calculated using the Pearson correlation coefficient, and the correlation formula was fitted using multiple linear regression. Predicted light intensity = α·temperature + β·initial light intensity + error term; Output the range of correlation coefficients and the normal fluctuation range of light intensity with temperature.

6. The server rack fault monitoring system based on big data according to claim 1, characterized in that, Step 4: Continuous model optimization mechanism, specifically including: The model is retrained weekly using newly generated normal data and confirmed abnormal data, and the correlation coefficient between temperature and light intensity is updated. When a misjudgment is found during manual review, it is immediately added to the negative sample library, and the judgment weight of the corresponding feature is adjusted. For different rack types, sub-models are trained separately to avoid misjudgment across scenarios.

7. A server rack fault monitoring system based on big data according to claim 1, characterized in that, The alarm and visualization interaction module specifically includes: The analysis results are transformed into intuitive alarm information and visualization charts; based on the abnormal state monitored by sensors, corresponding warning levels are matched and warnings are issued accordingly. Maintenance personnel input the processing results of faulty sensors, adjust alarm thresholds, initiate calibration commands, and upload the input information to the cloud. Management personnel can then download the corresponding faulty sensor information remotely, compile historical data from the faulty sensors, and perform relevant analyses.

8. A server rack fault monitoring system based on big data according to claim 1, characterized in that, The system management and external integration module is configured to ensure stable system operation through device resource management and model parameter management, connect to external systems such as environmental monitoring and maintenance work orders, and ensure data security and operational compliance through user permission management.

Images