A machine room hardware equipment whole life cycle operation and maintenance management and maintenance prompting method

Abstract

The present application relates to the technical field of monitoring and operation control of machine room equipment, and particularly relates to a machine room hardware equipment full life cycle operation management and maintenance prompting method, which comprises the following steps: obtaining the temperature, electric power load, fan rotating speed and various reference statistics of machine room hardware equipment; fusing the excess heat energy and transient load change amount exceeding the reference to determine a thermoelectric comprehensive stress index, analyzing the cooperative change trend of recent temperature and fan rotating speed sequence and the fluctuation degree of fan rotating speed to determine a heat dissipation response abnormality, adjusting the thermoelectric comprehensive stress index by using the heat dissipation response abnormality to generate a degradation acceleration factor; remodeling the state transition probability determined by the hidden Markov model by using the degradation acceleration factor to evaluate the maintenance urgency, and performing the full life cycle operation management and maintenance prompting. The method improves the accuracy of machine room hardware equipment operation and life cycle management.

Description

Technical Field

[0001] This invention relates to the field of monitoring, operation, and maintenance control technology for data center equipment. Specifically, it relates to a method for the full lifecycle operation and maintenance management and maintenance prompts for data center hardware equipment. Background Technology

[0002] The full lifecycle operation and maintenance management of data center hardware equipment is a core technology to ensure the stable operation of data centers and industrial edge computing nodes. It mainly involves real-time monitoring of the health status of critical equipment such as servers, storage arrays, and network switches, and issuing maintenance prompts when equipment shows signs of failure. Typical components include data acquisition modules, status estimation modules, and decision alarm modules. It is usually applied in high-load scenarios such as Internet data centers, financial settlement centers, and industrial control rooms, aiming to extend equipment life through digital means and reduce the risk of business interruption caused by equipment thermal failure.

[0003] In existing equipment operation and maintenance methods, a typical approach is to use a hidden Markov model. This model infers the hidden performance state level of the equipment by collecting multi-dimensional parameters during equipment operation. The processing steps include operation signal acquisition, transition probability configuration and decoding inference, which abstracts the equipment degradation process into hidden states such as health and wear, thereby realizing equipment operation and maintenance and life cycle management.

[0004] However, in existing equipment operation and maintenance methods based on hidden Markov models, the state transition probability is usually determined based on historical statistical priors or fixed empirical distributions. This is a static, fixed constant matrix. In the actual industrial scenarios of modern high-density data centers, hardware equipment often faces severe load fluctuations caused by tidal computing power tasks, generating strong dynamic thermoelectric stress and sudden local heat island effects. This static state transition probability is difficult to perceive and respond to the accelerated physical degradation caused by excess heat energy and transient load surges. As a result, the model still uses the slow prior degradation rate for evaluation when facing severe physical conditions. This makes the model extremely insensitive to sudden structural degradation signals, and it is very easy to misjudge the performance status of hardware equipment, affecting the accuracy of data center hardware operation and maintenance management.

[0005] Therefore, there is an urgent need for a method that can accurately perceive the actual accelerated degradation process of data center hardware equipment under dynamic thermo-electric stress and abnormal heat dissipation, so as to improve the accuracy of data center hardware equipment operation and maintenance and life cycle management. Summary of the Invention

[0006] To address the problem that existing Hidden Markov Models (HMMs), due to their fixed state transition probabilities, struggle to detect accelerated hardware physical degradation under complex data center conditions, thus affecting the accuracy of data center hardware equipment operation and maintenance and lifecycle management, this invention proposes a method for full lifecycle operation and maintenance management and maintenance prompts for data center hardware equipment, including:

[0007] Obtain the temperature, power load, and fan speed of the computer room hardware at the current moment, as well as the baseline mean, standard deviation, power load, fan speed, and fan speed standard deviation determined during the initial operation phase. The value by which the current temperature exceeds the average temperature reference is taken as excess heat energy, and the difference between the current electrical power load and the previous electrical power load is taken as transient load change. The thermoelectric comprehensive stress index is determined by combining excess heat energy and transient load change. Extract the recent temperature sequence and the recent fan speed sequence at the current moment, analyze the coordinated change trend to determine the degree of heat dissipation response anomaly; determine the degree of fan speed fluctuation based on the fan speed, the fan speed baseline mean, and the fan speed baseline standard deviation; use the degree of fluctuation and the degree of heat dissipation response anomaly to adjust the thermoelectric comprehensive stress index and generate a degradation acceleration factor. The state transition probabilities determined by the Hidden Markov Model are reshaped using a degradation acceleration factor. The maintenance urgency is assessed based on the reshaped state transition probabilities, and full lifecycle operation and maintenance management and maintenance prompts are executed based on the maintenance urgency.

[0008] This technical solution integrates excess heat energy exceeding the benchmark with transient load changes, accurately quantifying the thermoelectric stress index borne by the equipment. Furthermore, by analyzing the coordinated change trend of temperature and fan speed, it keenly captures the phenomenon of sluggish heat dissipation response. Using the degree of heat dissipation response anomaly and fan speed fluctuation, the thermoelectric stress index is nonlinearly adjusted. The resulting degradation acceleration factor can truly reflect the process of physical loss aggravation of the equipment under sudden computing power surges and local heat island effects. Introducing it into the Hidden Markov Model reshapes the state transition probability, enabling the originally static probability matrix to dynamically perceive severe working conditions, thereby improving the accuracy of equipment operation and maintenance and life cycle management.

[0009] Furthermore, the baseline mean temperature, baseline standard deviation temperature, baseline mean electrical power load, baseline mean fan speed, and baseline standard deviation fan speed determined during the initial operation phase are determined as follows: During the initial operation phase when the computer room hardware equipment is put into use and in a healthy state, the initial temperature sequence, initial electrical power load sequence, and initial fan speed sequence are obtained; based on the initial temperature sequence, initial electrical power load sequence, and initial fan speed sequence, mathematical expectation and variance calculations are performed to generate the baseline mean temperature, baseline standard deviation temperature, baseline mean electrical power load, baseline mean fan speed, and baseline standard deviation fan speed.

[0010] Furthermore, the thermoelectric stress index is determined by integrating excess heat energy and transient load changes, including: normalizing the excess heat energy to obtain a normalized thermal stress value; converting the transient load change into a load impact coefficient, and using the load impact coefficient to adjust the normalized thermal stress value to obtain the thermoelectric stress index of the computer room hardware at the current moment.

[0011] This technical solution effectively eliminates the inherent differences in heat dissipation conditions at different physical locations within the computer room by normalizing the allocation of excess heat energy. At the same time, it transforms transient loads into load impact coefficients. These two processes enable the fused thermoelectric stress index to objectively and stably reflect the load borne by the equipment under dynamic computing power allocation.

[0012] Furthermore, the thermal response anomaly degree is determined as follows: extract the recent temperature sequence and the recent fan speed sequence at the current moment according to a preset time window, calculate the cross-correlation coefficient of the recent temperature sequence and the recent fan speed sequence at different delay step sizes, select the maximum value of the cross-correlation coefficient as the maximum delay cross-correlation coefficient, calculate the complement of the maximum delay cross-correlation coefficient, if the complement is greater than 0, use the complement as the thermal response anomaly degree of the computer room hardware equipment at the current moment, if the complement is not greater than 0, set the thermal response anomaly degree of the computer room hardware equipment at the current moment to 0.

[0013] This technical solution calculates the cross-correlation coefficient between recent temperature sequences and fan speed sequences at different delay steps and extracts the maximum value. It can accurately match the causal physical relationship between temperature rise and fan speed increase in the time domain. It also uses the complement of the maximum delay cross-correlation coefficient to quantify the degree of heat dissipation response anomaly. It directly transforms the complex health status of the heat dissipation system into a mathematical indicator that can be processed by the algorithm, and accurately identifies hidden problems such as mechanical aging or control logic lag inside the heat dissipation system.

[0014] Furthermore, the thermoelectric comprehensive stress index is adjusted using the degree of fluctuation and the degree of heat dissipation response anomaly to generate a degradation acceleration factor. This includes: using the thermoelectric comprehensive stress index of the computer room hardware at the current moment as a base term; if the degree of heat dissipation response anomaly at the current moment is 0, the base term is directly used as the degradation acceleration factor of the computer room hardware at the current moment; if the degree of heat dissipation response anomaly at the current moment is not 0, an amplification multiplier is constructed by combining the product of the degree of fluctuation of the fan speed and the degree of heat dissipation response anomaly with the base constant, and the amplification multiplier is used to adjust the base term to obtain the degradation acceleration factor of the computer room hardware at the current moment.

[0015] This technical solution retains only the influence of basic stress when the heat dissipation function is normal. However, once an abnormal heat dissipation is detected, the product of the fan speed fluctuation and the thermoelectric stress index is used to nonlinearly amplify and adjust the basic terms. This multiplicative amplification mechanism ensures that the model has extremely high sensitivity to high-risk physical conditions such as high temperature accompanied by heat dissipation failure, and can realistically reproduce the accelerated aging process of the equipment.

[0016] Furthermore, the state transition probabilities determined by the Hidden Markov Model are reshaped using a degradation acceleration factor, including: obtaining the performance state level of the data center hardware equipment at the current moment based on the pre-constructed Hidden Markov Model; taking the performance state level at the current moment and all performance state levels above it as the target state level; obtaining the original static transition probabilities from the current performance state level to each target state level; calculating the state span between each target state level and the current performance state level; using the state span and the degradation acceleration factor as mapping variables; substituting them into the algebraic approximation equation based on the second-order Taylor expansion to construct nonlinear amplification weights; and using the amplification weights to weighted normalize the original static transition probabilities from the current performance state level to each target state level to obtain the reshaped state transition probabilities.

[0017] This technical solution amplifies and reshapes the original state transition probability of the Hidden Markov Model by multiplying the degradation acceleration factor and the state level span. It changes the assumption of the traditional model that it relies on uniform aging over time. This reshaping mechanism gives the system the ability to capture potential faults across levels, so that the probability of the equipment transitioning to a higher dangerous state after experiencing a severe instantaneous thermoelectric shock can be immediately amplified, which greatly ensures the real-time performance and accuracy of the prediction results.

[0018] Furthermore, the maintenance urgency is assessed based on the reshaped state transition probabilities, including: selecting target state levels with a higher level than the current performance state level as degraded target state levels; obtaining the reshaped state transition probability from the current performance state level to the degraded target state level, and calculating the state span between the degraded target state level and the current performance state level, mapping the state span to the physical degradation cost weight of the degraded target state level; and weighting and summing the reshaped state transition probabilities from the current performance state level to each degraded target state level based on the degradation cost weight of each degraded target state level to obtain the maintenance urgency of the data center hardware equipment at the current moment.

[0019] This technical solution sets a penalty weight specifically for the deteriorated target state level when assessing maintenance urgency. This nonlinear amplification and weighted summation operation based on the state span emphasizes the high cost of a sudden drop in equipment performance in a physical sense, so that the final maintenance urgency index can better highlight the potential risks that seriously endanger the safe operation of the data center.

[0020] Furthermore, based on maintenance urgency, full lifecycle operation and maintenance management and maintenance prompts are implemented, including: determining whether the data center hardware equipment is currently in the early, middle, or late stage of its lifecycle; when in the early or middle stage, if the current maintenance urgency is greater than a preset maintenance urgency threshold, reducing the power load of the data center hardware equipment; if the current maintenance urgency is not greater than the preset maintenance urgency threshold, maintaining the power load of the data center hardware equipment; when in the late stage, if the current maintenance urgency is greater than the preset maintenance urgency threshold, prompting maintenance personnel to perform hardware equipment repair operations; if the current maintenance urgency is not greater than the preset maintenance urgency threshold, maintaining monitoring of the operational status of the data center hardware equipment.

[0021] Furthermore, the method for determining whether the data center hardware equipment is in the early, middle, or late stage of its life cycle at the current moment is as follows: Based on the Hidden Markov Model, obtain the performance state level of the data center hardware equipment at the current moment, as well as the predefined highest state level representing complete failure; calculate the ratio of the current performance state level to the highest state level; compare the ratio with the preset life cycle division threshold, and determine whether the data center hardware equipment is in the early, middle, or late stage of its life cycle at the current moment based on the comparison result.

[0022] Furthermore, the data center hardware includes server nodes and core switches. When the data center hardware is a server node, the temperature of the processor chip is read as the current temperature; when the data center hardware is a core switch, the temperature of the integrated circuit switching chip is read as the current temperature.

[0023] The present invention has the following effects: This invention assesses the combined thermoelectric stress formed by excess heat energy and dynamic load, and deeply integrates the response hysteresis phenomenon of electromechanical heat dissipation systems. It effectively corrects the inherent defect of slow response of traditional hidden Markov models under complex working conditions, and achieves a fundamental leap from empirical static prediction to physical dynamic prediction. This improves the accuracy of operation and maintenance and life cycle management of data center computing hardware equipment when facing tidal loads and sudden local heat island effects. Attached Figure Description

[0024] Figure 1 This is a schematic diagram of the method flow of the present invention; Figure 2 This is a schematic diagram comparing the accuracy of equipment failure early warning between the present invention and the traditional hidden Markov model method under different load mutation intensities. Figure 3 This is a schematic diagram comparing the monthly unplanned downtime of the present invention with that of the traditional Hidden Markov Model method throughout the entire life cycle of the equipment. Detailed Implementation

[0025] The technical solutions of the present invention will be clearly and completely described below with reference to the accompanying drawings in the embodiments of the present invention.

[0026] This invention provides a method for full lifecycle operation and maintenance management and maintenance prompts for computer room hardware equipment, such as... Figure 1 As shown, it includes: S1: Obtain the baseline statistics of the hardware equipment in the computer room.

[0027] This step aims to acquire the operating data of the equipment in the computer room during the initial operation phase and at the current moment, perform statistical and extraction operations, obtain baseline statistics and the current temperature, power load and fan speed, and perform strict synchronization alignment and standardization preprocessing operations on these raw data.

[0028] Specifically, this step involves the following operations: First, during the initial operation phase when the computer room hardware is in use and in a fully healthy state, the initial temperature sequence, initial power load sequence, and initial fan speed sequence are obtained. Based on the initial temperature sequence, initial power load sequence, and initial fan speed sequence, standard mathematical expectation and variance calculations are performed to generate a series of benchmark statistics, including: the baseline temperature mean. Temperature reference standard deviation Average value of electrical power load Fan speed reference average and fan speed reference standard deviation This represents the baseline level of parameters of the device in a healthy state.

[0029] Simultaneously, based on the specific type of equipment at the current moment, operational parameters are extracted. When the hardware equipment in the data center is a server node, the temperature of the processor chip is directly read as the current temperature by calling the underlying intelligent platform management interface. The power load at the current moment is obtained by reading the product of the current and voltage of the motherboard power supply module. And read the chassis fan feedback signal to obtain the current fan speed. When the core switch is the hardware equipment in the data center, the temperature of the integrated circuit switching chip is directly read as the current temperature using telemetry technology. Read the backplane bus load as the current power load. The fan speed at the current moment is determined by the rotation speed of the fan plug-in module. In addition, the electrical power load of the hardware device at the previous moment is acquired synchronously. .

[0030] S2: Analyze the thermoelectric coupling effect to determine the thermoelectric combined stress index.

[0031] After extracting the underlying baseline parameters, in order to address the issue of nonlinear physical losses in equipment caused by heat accumulation and sudden increases in computing power in high-density data center environments, it is not appropriate to simply consider temperature or load exceeding limits in isolation. In real-world physical scenarios, if the data center air conditioning provides strong cooling without generating high temperatures under high load, its physical losses remain within a controllable range; and when the temperature is slightly high but the load is extremely low, it is unlikely to cause silicon-based lattice fracture.

[0032] Therefore, this step aims to establish a microscopic thermoelectric coupling sensing mechanism for the computer room hardware. Based on baseline statistics and the current temperature and electrical power load, excess heat energy and transient load changes are extracted, and a fusion operation is performed to determine the thermoelectric comprehensive stress index. Through the multiplier effect between the transient electrical stress abrupt change gradient and the physical thermal over-limit, the two are coupled into a thermoelectric comprehensive stress index to reflect the thermoelectric coupling failure condition of the hardware.

[0033] Specifically, this step involves the following operations: The difference between the current temperature and the average temperature baseline is taken as excess heat energy. The difference between the current electrical load and the previous electrical load is taken as transient load change. The combined thermoelectric stress index is determined by integrating the excess heat energy and the transient load change. The excess heat energy is normalized to obtain a normalized thermal stress value. The transient load change is converted into a load impact coefficient, which is used to adjust the normalized thermal stress value to obtain the combined thermoelectric stress index of the computer room hardware at the current moment.

[0034] Specifically, the following relationship is satisfied:

[0035] in, The thermoelectric stress index is a dimensionless comprehensive evaluation index representing the current temperature of the computer room hardware equipment. The temperature of the hardware equipment in the computer room at the current moment. The average temperature is the baseline value. The standard deviation of the temperature reference is... This represents the electrical power load of the computer room hardware at the current moment. This refers to the electrical power load of the computer room hardware equipment at the previous moment. The average value of the electrical power load is the reference value. To find the maximum value function, The preset value is 0.00001, a very small positive number used to prevent the denominator from being zero. It is the absolute value symbol.

[0036] In this relation, For Z-score normalization with single-item filtering function, the overall result is a normalized thermal stress value. It acts as a unidirectional thermal anomaly filter factor, forcibly eliminating situations where the current temperature is below the reference temperature (safety level), extracting only excess heat energy exceeding the absolute health reference, and then dividing it by the reference standard deviation to convert it into a standardized deviation. The larger this term is, the more severe the heat island effect in the rack's internal microenvironment. The second term... This constitutes a load impact factor, used to characterize the dynamic amplification of the normalized thermal stress value by the transient load change. The larger the value, the more significant the amplification effect of the transient load change on the normalized thermal stress value. This partially reflects the relative severity of power load fluctuations, using the absolute difference. The transient surge gradient of the load was quantified, and safe and bounded nonlinear numerical compression and smoothing were performed through a rational algebraic structure of the same order of numerator and denominator. The larger this term is, the greater the relative severity of the fluctuation of electrical power load inside the computer room hardware.

[0037] In this relation, The reason for incrementing by 1 is that in real-world physical scenarios, the sheer volume of heat generated by high temperatures accelerates equipment aging. When the equipment's energy load remains absolutely stable, the transient load change is zero, and the load impact factor is 1. In this case, the health status of the equipment is assessed solely through normalized thermal stress values. When the equipment's energy load fluctuates, the transient load change is greater than zero, and the impact factor is greater than 1. The load impact factor is used to amplify the basic thermal stress, which aligns with the microscopic physical mechanisms within semiconductor devices: under the catalysis of high-temperature noise floor, sudden load surges trigger intense electromigration and localized heat island accumulation effects. This fusion mechanism of a safety constant and a dynamic multiplier reflects the true destructive potential energy under harsh microenvironments.

[0038] If only the power consumption of the device itself increases but the system does not heat up, the thermoelectric stress index S will remain at a very low level because no excess heat is generated. If only the system heats up but the power consumption of the device remains stable, the thermoelectric stress index S only reflects the basic normalized thermal stress value. Only when the power consumption of the device itself surges and the system heats up, the two combine multiplicatively to drive the thermoelectric stress index S to rise significantly, accurately reflecting the working condition that is prone to physical degradation and fracture of components due to the simultaneous surge of electrical stress and thermodynamic stress.

[0039] This relationship is an algebraic reconstruction based on the classic Arrhenius thermal accelerated aging equation and the Coffin-Manson low-cycle fatigue model in semiconductor reliability engineering. Normalized thermal stress is used to characterize the static thermodynamic noise of the Arrhenius effect, and the load mutation term of rational fractional compression is used to accurately map the dynamic electrical fatigue induced by the Coffin-Manson power cycle. Through the product fusion mechanism of the two, the destructive potential energy in the microenvironment is objectively reflected.

[0040] In summary, this step achieves deep coupling quantification of thermal and electric fields by extracting excess heat energy and transient load surge characteristics and performing nonlinear product fusion operations. This results in the ability to isolate single-dimensional false alarms from a physical mechanism perspective and accurately quantify the underlying microscopic destructive potential energy of the equipment.

[0041] S3: Analyze the trend of coordinated change and generate degradation acceleration factors.

[0042] Having determined the microscopic thermoelectric stress, it is important to understand that the overall degradation of data center hardware is determined not only by the local thermoelectric state of the chip but also by the overall physical health of the thermal infrastructure. In a healthy thermal architecture, the fan speed sequence must keep in positive synchronization with the temperature sequence. However, when irreversible structural degradation occurs, such as thermal grease powdering or heat pipe blockage, heat cannot be conducted, inevitably leading to persistently high chip temperatures and ineffective fan operation. At this point, the temperature and fan sequence completely lose their phase-following characteristics, resulting in a thermomechanical control flow decoupling state.

[0043] Therefore, this step aims to explore the health status at the macroscopic system level, determine the anomaly degree of heat dissipation response based on the coordinated change trend of recent temperature series and recent fan speed series, as well as the degree of fluctuation of fan speed, and use the anomaly degree of heat dissipation response to adjust the thermoelectric comprehensive stress index to generate a degradation acceleration factor.

[0044] A time-delay cross-correlation coefficient with tolerance is introduced as the highest-level gating factor for the health status of the heat dissipation system. A nonlinear amplification mechanism is constructed using a sign function to intelligently identify heat dissipation system fault fluctuations at different levels and dynamically adjust the base stress. The aim is to upgrade the apparent environmental data into a degradation acceleration factor that reflects the essence of internal damage to the equipment.

[0045] Specifically, this step involves the following operations: According to the preset time window, the temperature sequence and fan speed sequence within the current moment and a historical time period before that moment are extracted as the recent temperature sequence and recent fan speed sequence for the current moment. Here, the time window length is set to include the current moment and the historical time period of 10 sampling periods before that moment. This is because the heat dissipation system of high-density equipment in the computer room has inherent physical inertia and control system delay in response to chip temperature fluctuations, and short-term peak fluctuations are insufficient to characterize the true state of the system. Selecting a length that includes the current moment and several historical sampling periods is just enough to cover a complete thermodynamic regulation closed loop of temperature rise, fan acceleration, and temperature fall, thereby providing complete and stable sequence support for subsequent accurate evaluation of thermomechanical coupling relationship.

[0046] Calculate the cross-correlation coefficient between the recent temperature sequence and the recent fan speed sequence at different delay steps. Select the maximum cross-correlation coefficient as the maximum delay cross-correlation coefficient. Calculate the complement of the maximum delay cross-correlation coefficient. If the complement is greater than 0, use the complement as the anomaly degree of the heat dissipation response of the data center hardware equipment at the current moment. If the complement is not greater than 0, set the anomaly degree of the heat dissipation response of the data center hardware equipment at the current moment to 0. Specifically, the anomaly degree of the heat dissipation response of the data center hardware equipment at the current moment... ,in, To find the maximum value function, This represents the maximum time-delay cross-correlation coefficient between the recent temperature series and the recent fan speed series, with a value range of [-1, 1]. It is the complement of the maximum time delay cross-correlation coefficient.

[0047] It should be noted that when calculating the cross-correlation coefficient, the classic Pearson correlation coefficient method can be used to assess the phase lag and causal synchronization of two time series. In the current data center scenario, it is used to accurately quantify the hidden physical structural fault of thermo-mechanical control flow decoupling in electromechanical systems. The complement is the value obtained by subtracting the maximum delay cross-correlation coefficient from the constant 1. The maximum delay cross-correlation coefficient reflects the degree to which the fan speed positively follows temperature changes. The closer its value is to 1, the healthier and more responsive the cooling system is. By calculating 1 minus the complement of this coefficient, the indicator representing health is cleverly reversed into a quantitative indicator representing anomaly. When the cross-correlation coefficient is higher, the complement is closer to 0, indicating that the system is more normal. When the cross-correlation coefficient decreases or even becomes negative, it indicates that the fan is out of sync with temperature changes, and the complement increases. This allows for an intuitive and accurate quantification of the degree of decoupling of the cooling architecture's control flow.

[0048] Then, using the thermoelectric stress index of the computer room hardware at the current moment as the basic term, if the heat dissipation response anomaly degree at the current moment is 0, the basic term is directly used as the degradation acceleration factor of the computer room hardware at the current moment; if the heat dissipation response anomaly degree at the current moment is not 0, the thermoelectric stress index is nonlinearly amplified and adjusted by the fluctuation degree of fan speed and the heat dissipation response anomaly degree to obtain the degradation acceleration factor of the computer room hardware at the current moment.

[0049] Specifically, the following relationship is satisfied:

[0050] in, The factors that accelerate the degradation of data center hardware at the current moment. The thermoelectric stress index of the computer room hardware equipment at the current moment. This refers to the degree of thermal response abnormality of the computer room hardware equipment at the current moment. This indicates the degree of fluctuation in fan speed. , The current fan speed. This is the average fan speed as a baseline. The standard deviation of fan speed is the baseline. The preset value is 0.00001, a very small positive number used to prevent the denominator from being zero. It is the absolute value symbol.

[0051] In this relation, This reflects the magnitude of the combined thermoelectric stress experienced by the hardware equipment in the data center at the microscopic level, and is a fundamental factor leading to the physical degradation of the hardware equipment. (square brackets) The internal part further constructs a diagnostic multiplier for the overall abnormality of the heat dissipation system, and uses this multiplier to amplify and adjust the basic items from an environmental perspective.

[0052] square brackets The system integrates macro trends and micro details to make comprehensive anomaly judgments. The anomaly degree of heat dissipation response is derived from cross-correlation analysis over a time window. It reflects the overall macroscopic trend of the heat dissipation system, i.e., whether the fan control flow and temperature evolution have become completely disconnected. The fluctuation of fan speed reflects the microscopic transient details of the fan's mechanical response at the current moment, i.e., the degree of deviation of the fan from the normal operating baseline. By multiplying the macroscopic trend threshold with the microscopic transient fluctuation, a precise judgment mechanism is constructed, as follows: When the cooling system is macroscopically healthier, macroscopic trend indicators... The closer the value is to the minimum of 0, the better the heat dissipation effect. At this point, regardless of the degree of transient fluctuation in the microscopic fan speed, the system accurately determines that this is a normal physical heat exchange adjustment. The strong suppressive effect of the fan speed eliminates the micro-fluctuations, making the square brackets The overall value within the range approaches the fundamental constant 1, thus increasing the degradation acceleration factor. Thermoelectric stress index approaching the microscopic level This ensures, in terms of physical logic, that normal micro-fluctuations in the fan will not be misjudged as a heat dissipation failure, thus avoiding over-adjustment of the basic degradation stress.

[0053] When the cooling system exhibits more macroscopic anomalies, macroscopic trend indicators... The larger the value, the worse the heat dissipation effect. At this point, microscopic transient fluctuations will be activated directly as a benchmark reflecting the intensity of the anomaly. Under the premise of macroscopic failure of the heat dissipation system, the more violent and uncontrolled the fan speed fluctuations at the microscopic level, the more severe the chaos and oscillation of the underlying temperature control feedback system. This indicates a greater degree of overall anomaly in the heat dissipation system, which will prevent the effective release of the comprehensive thermoelectric stress of the computer room hardware itself at the microscopic level, thus causing a multiplied destructive force on the lifespan of the equipment.

[0054] Therefore, in the case of heat dissipation failure, the greater the amplitude of the micro-uncontrolled fluctuation of the fan, the more severe the overall judgment of the heat dissipation anomaly, and the higher the penalty amplification factor on the basic items. This design not only eliminates the misjudgment that is easily caused by single-dimensional indicators, but also more accurately restores the real acceleration mechanism of equipment degradation caused by harsh heat dissipation environment.

[0055] This relationship is a direct fine-tuning of the classic Cox proportional hazards model in reliability engineering. This type of model accelerates the lifetime prediction of complex mechanical structures by multiplying the base degradation rate by an external covariate state factor. In the current scenario, external cooling system failures are transformed into an acceleration multiplier on the chip's underlying physical degradation rate.

[0056] Finally, this step, through maximum time-delay cross-correlation analysis, achieves logical decoupling between environmental parameters and equipment mechanical health, enabling intelligent identification and progressive amplification of deep structural deterioration pathological features, thus injecting a high-confidence physical driving force into subsequent algorithms.

[0057] S4: Reshape the state transition probabilities determined by the hidden Markov model.

[0058] After extracting the dynamic degradation acceleration factor with extremely high physical confidence for the hardware equipment in the data center, this step uses the degradation acceleration factor to reshape the state transition probability determined by the hidden Markov model, and obtains the reshaped state transition probability. This is a key step to overcome the static limitations of existing algorithms.

[0059] This step aims to reshape the principle of the algorithm without introducing redundant external networks. Instead, it directly injects the degradation acceleration factor extracted from the device microscopic level into the probability inference topology of the Hidden Markov Model. Through the span reward mechanism, it forces the static matrix to dynamically respond to the laws of real physical reality, so that the hidden state inference domain of the model will decisively tilt towards the deep degradation state in harsh environments, so as to accurately reflect the health status of the device.

[0060] Specifically, this step involves the following operations: The performance state level of the data center hardware equipment at the current moment is obtained based on a pre-constructed Hidden Markov Model. The specific process is as follows: offline training is performed using a large amount of multi-dimensional operating data of the same type of hardware equipment throughout its entire life cycle. The Baum-Welch algorithm is used to determine the initial state distribution matrix, observation probability matrix, and historical static state transition probability matrix of the Hidden Markov Model, thus constructing a Hidden Markov Model that represents the equipment degradation process. During online monitoring, the continuous observation sequence of the data center hardware equipment up to the current moment is input into the model, and the Viterbi algorithm is used for decoding and inference to solve for the index of the hidden state with the highest probability at the current moment, which is used as the performance state level at the current moment.

[0061] Based on this, the current performance state level and each higher-order performance state level representing the irreversible degradation direction of the performance state are taken together as the target state level, and the transition probability from the current performance state level to each target state level is obtained.

[0062] In reliability physics, the amplification multiplier of a system's degradation state transition after encountering environmental stress strictly follows the Boltzmann distribution or the Arrhenius exponent law, i.e., it exhibits... The exponential physical evolution of exponential functions. However, directly calculating exponential transcendental functions can significantly increase computational costs.

[0063] This section introduces an algebraic approximation strategy based on second-order Taylor expansion, using the exponential function. The truncation form of the second-order Maclaurin series is: This second-order Taylor expansion has significant engineering implications: it eliminates artificially set empirical constants, with all coefficients derived from rigorous analytic geometry. Furthermore, it reduces complex transcendental functions to a hardware-friendly algebraic structure that relies solely on basic arithmetic operations, and its quadratic terms... It retains the nonlinear characteristics of extreme state transitions during the equipment degradation and evolution process.

[0064] Based on the above principles, this step defines the cross-state damage mapping variable for the hardware device as follows: Through target state hierarchy index Subtract the performance state level index at the current moment Calculate state span Combine it with degradation acceleration factor The product of these factors is used as the cross-state damage mapping variable. Substituting it into the second-order Taylor expansion yields the nonlinear amplification weight. This amplification weight is then used to perform a weighted normalization operation on the original static transition probability to obtain the reshaped state transition probability.

[0065] Specifically, the following relationship is satisfied:

[0066] In this relation, It refers to the performance status level of the redesigned data center hardware at the current moment. Transfer to the target state level The state transition probability, The device's state under absolutely stable operating conditions was obtained through offline training based on massive amounts of historical data. Transition to state The static transition probability, The device's state under absolutely stable operating conditions was obtained through offline training based on massive amounts of historical data. Transition to state The static transition probability; It is the performance status level index determined at the current moment. and It is a target state level index, and guarantees To characterize the direction of irreversible degradation, This is a predefined highest state level representing complete failure. This refers to the degradation acceleration factor of the computer room hardware equipment at the current moment. The performance state level at the current moment With the target state level Cross-state damage mapping variables, , The performance state level at the current moment With the target state level Cross-state damage mapping variables, .

[0067] This formula utilizes the properties of the second-order Taylor expansion: the core lies in This constructs an amplified weight based on the state span distance. When the equipment operating condition is stable or there is no state span decay, When the value approaches 0, the physical lower bound of the amplification weight is adaptively and safely locked at 1, preserving the prior probability determined based on historical steady-state data, and no amplification adjustment is performed; when the equipment faces severe thermoelectric shock... This increases the risk of transitioning to a high-risk state at a distance. The amplification weight increases non-linearly, and when the amplification weight is greater than 1, it amplifies and adjusts the static transition probability.

[0068] This modulation strategy not only keenly captures high-frequency anomalies in the early stages of equipment degradation, but also avoids probability matrix collapse and computational overflow caused by the unbounded divergence of absolute state spans at the underlying algorithmic level. This relationship belongs to the reconstruction paradigm of non-homogeneous hidden Markov models and incorporates the proportional penalty mechanism of relative severity in Markov decision-making processes. In the current scenario, it ensures that even in the face of extreme and severe peak disturbances, the probability amplification factor is forced to converge within a safe physical upper limit, thus improving noise robustness in industrial environments.

[0069] In a specific example, for a monitoring window of a server node in a data center, a degradation acceleration factor characterizing moderate thermoelectric shock was calculated. Assuming a predefined device lifecycle includes four states, At the current moment, it is inferred that the device is in state 1, a healthy state, i.e. The static transition probability matrix obtained offline is: Maintaining health Slight degeneration Severe degradation Facing outage .

[0070] Begin calculating the amplified weights and numerators of the second-order Taylor expansion for each target state level: When the target state level is 1 Span is 0, mapping variable The amplification weight is: The numerator is: ; When the target state level is 2 Span is 1, mapping variable The amplification weight is: The numerator is: ; When the target state level is 3 The span is 2, and the mapping variable is... The amplification weight is: At this point, the numerator is: ; When the target state level is 4 The span is 3, and the mapping variable is... The amplification weight is: The numerator is: ; Calculate the normalized sum of the denominators: ; Performing a weighted normalization operation yields the reshaped state transition probabilities: the probability of maintaining state 1. The probability of a reasonable slight decrease from the original 98.5% to 93.1%; and the probability of directly jumping to state 4 (outage state) The probability of fatal transition state degradation increased from an extremely low 0.05% to 0.74%. Under the penalty of the Taylor quadratic term, the probability of fatal transition state degradation was nonlinearly amplified by nearly 15 times, activating the hidden Markov model's potential for physical early warning of extremely dangerous and minute precursors.

[0071] S5: Assess the urgency of maintenance and implement full lifecycle operation and maintenance management.

[0072] The preceding steps have constructed a physically driven and environmentally adaptive reinforcement model matrix. This step assesses maintenance urgency based on the reshaped state transition probabilities and executes full lifecycle operation and maintenance management and maintenance prompts based on the maintenance urgency. The aim is to converge the final business logic, deeply explore the micro-anomalies of the probability distribution, and specifically extract the mathematical expectation of the harm of all deteriorating states.

[0073] By introducing a degradation span cost mechanism based on Markov decision processes, the dispersed probability array is collapsed into a single maintenance urgency index, realizing a closed-loop management path from low-level algorithm warning to high-level automatic work order interception, thus eliminating blind spots in human decision-making.

[0074] Specifically, this step involves the following operations: Target state levels with a higher priority than the current performance state level are selected as degradation target state levels. The state transition probability from the current performance state level to the reshaped target state level is obtained, and the state span between the degradation target state level and the current performance state level is calculated. This state span is mapped to the physical degradation cost weight of the degradation target state level. Based on the degradation cost weights of each degradation target state level, the state transition probabilities from the current performance state level to each reshaped target state level are weighted and summed to obtain the maintenance urgency of the data center hardware equipment at the current moment.

[0075] Specifically, the urgency of maintenance satisfies the following relationship:

[0076] in, The urgency of maintaining the data center's hardware at any given moment serves as an indicator to guide business operations. This represents the performance state level at the current moment. This is a predefined highest state level representing complete failure. For the selected deterioration target state level, satisfying conditions, From state Transfer to the deteriorated target state level The probability of state transition after remodeling. Maintenance urgency is essentially the expected level of degradation that the equipment is expected to slide into a deeper failure in the next monitoring cycle.

[0077] This summation formula uses a defined range of summations from... arrive In terms of physical concepts, state maintenance has been eliminated. The probability of inefficiency is calculated by focusing purely on the unidirectional deterioration trend, and a rigorous Markov decision-making expected cost model is constructed. It is the absolute span, equal to the target deterioration state level. With the current state level The difference represents how many performance levels the device dropped at once. This range is directly used as a penalty cost function and multiplied and fused with the probability of remodeling.

[0078] Unlike traditional equal-probability summation models, this design decouples the probability of occurrence from the consequences of degradation: if the model determines that the degradation is only a slight deterioration to the next level and the state span is small, such as... Its contribution to the urgency index is strictly equal to its probability of occurrence. It does not generate excessive interference, but if the model determines that there is a possibility of a direct transition to a high-risk state of complete failure, i.e., it faces the risk of sudden downtime, the state span is large. The penalty cost will increase significantly, and its corresponding weight will rise sharply.

[0079] This weighting mechanism, based on pure mathematical expectation, abandons artificially set empirical constants or safety margins, and objectively amplifies the weight of extremely dangerous and low-probability events that cause leapfrogging. This makes the maintenance urgency index exhibit smooth tracking of slow initial performance degradation, while being sensitive to extreme high-risk leapfrogging precursors.

[0080] This relational expression closely aligns with the classical expected cost function theory in Markov decision processes and is typically used to evaluate the overall expected loss of a strategy in uncertain environments. Its role here is to safely and rigorously collapse a high-dimensional, complex probability distribution matrix into a single operational decision scalar with clear physical meaning that can be directly invoked by an automated work order system.

[0081] This step achieves precise collapse of equipment performance degradation risk assessment to engineering intervention indicators by summing the expected sum of state transition probability and absolute degradation span. Combined with multi-level response strategies, it achieves the effect of preemptive interception of weak and fatal signals before equipment failure and automated closed-loop management throughout the entire life cycle.

[0082] Following this, based on the urgency of maintenance, full lifecycle operation and maintenance management and maintenance prompts will be implemented, specifically including: Based on the service level agreement tolerance of the services carried by the data center and the historical peak distribution of maintenance urgency indicators within a preset time window before previous sudden downtime events, the lower limit of the 95% confidence interval of its lower edge is selected, and then combined with the redundancy of the current business nodes for conversion, so as to ensure that the preset maintenance urgency threshold can both capture fatal precursors in advance and not trigger overly sensitive invalid alarms.

[0083] The method for determining whether any hardware device in a data center is in the early, middle, or late stage of its life cycle at any given moment strictly follows the bathtub curve theory in reliability engineering: obtain the failure rate evolution curve of a large historical sample of such hardware devices, extract the first inflection point where the failure rate stabilizes from high and the second inflection point where the failure rate turns into an exponential increase; calculate the ratio of the average performance state level deduced by the system at the first and second inflection points to the predefined highest state level representing complete failure, and obtain the first and second division thresholds with universal physical significance.

[0084] The performance state level of the data center hardware equipment at the current moment is obtained based on the Hidden Markov Model. The ratio of the current performance state level to the highest state level is calculated. The ratio is compared with a first division threshold and a second division threshold. If the ratio is less than the first division threshold, it is determined to be in the initial stage, i.e., the early failure period. If the ratio is between the two, it is determined to be in the middle stage, i.e., the occasional failure period. If the ratio is greater than the second division threshold, it is determined to be in the final stage, i.e., the wear and tear failure period.

[0085] Implement differentiated management and control for different lifecycle stages: When the equipment is determined to be in the initial or intermediate stage, it indicates that the main hardware structure still has a lifespan. If the current maintenance urgency is greater than the preset maintenance urgency threshold, the system will not trigger scrapping and replacement. Instead, it will trigger online computing power hot migration through the control plane linkage virtualization scheduling platform to reduce the electrical power load of the hardware equipment in the data center and instantly unload local thermoelectric stress. If it is not greater than the preset maintenance urgency threshold, its load will remain unchanged.

[0086] When the equipment is determined to be in the final stage, it indicates that the hardware has suffered irreversible physical wear and tear. If the current maintenance urgency is greater than the preset maintenance urgency threshold, the system directly generates a high-priority work order, prompting the maintenance personnel to perform the hardware equipment decommissioning and physical replacement operations. If it is not greater than the threshold, the sampling frequency of the operation status monitoring is increased.

[0087] Another method to determine whether the hardware equipment in the data center is currently in the early, middle, or late stage of its life cycle is: The performance state level of the data center hardware device at the current moment is obtained based on the Hidden Markov Model, as well as the predefined highest state level representing complete failure; the ratio of the current performance state level to the highest state level is calculated; the ratio is compared with the preset life cycle division threshold, and the data center hardware device is determined to be in the early, middle or late stage of its life cycle at the current moment based on the comparison result.

[0088] Here, the preset life cycle division thresholds are preferably set to 0.3 and 0.7. The reason for setting these two thresholds is that, according to the bathtub curve failure theory widely followed by semiconductor devices, the life cycle of the equipment can be naturally divided into the early failure period, the accidental failure period, and the wear-out failure period. The threshold of 0.3 corresponds to the point when the equipment passes the early break-in period and enters stable operation, while the threshold of 0.7 is the inflection point where a large amount of electromigration and thermal fatigue damage accumulates, causing the failure rate to start to rise exponentially. Using these two thresholds can accurately divide the equipment into the early stage, middle stage, or late stage of the life cycle.

[0089] For example, if a server in the data center calculates its current performance status level... With the highest state level ratio The ratio was compared with the preset lifecycle segmentation threshold, and it was found that... Based on the comparison results, it was determined that the hardware device was in the early stage of its life cycle. Since it was in the early stage and the urgency of maintenance at the current moment was greater than the preset maintenance urgency threshold, the system did not generate a scrap replacement work order. Instead, it immediately linked the virtualization scheduling platform to trigger online computing power hot migration, reduce the power load of the computer room hardware device, and instantly unload the thermal and electrical pressure.

[0090] To further and intuitively verify the application effect and engineering value of the method of this invention in actual industrial scenarios, a 60-month comparative test was conducted on the AI ​​large-scale model training cluster nodes in the data center of an Internet enterprise. The method of this invention and the traditional Hidden Markov Model method were used respectively to perform parallel full lifecycle operation and maintenance management on 200 rack-mounted server nodes of the same batch, hardware configuration, and business load type. During the test, equipment operation data and operation and maintenance event statistics were collected synchronously, resulting in the following comparison chart: Figure 2This paper presents a comparison of the hardware degradation failure prediction accuracy between the proposed method and the traditional HMM method under different load mutation intensities. The horizontal axis represents the load mutation intensity, defined as the increase ratio of the transient peak power load to the baseline rated load. The test scope covers all business scenarios from stable rated operation to extreme peak load, including typical high-load conditions in data centers such as the startup of large-scale model batch training tasks, peak energy consumption and traffic, and sudden increases in computing power. The vertical axis represents the equipment failure prediction accuracy, calculated as the ratio of the number of irreversible hardware degradation failure events accurately predicted by the method 72 hours in advance within the test period to the total number of actual hardware degradation failures within the test period. Under low-fluctuation operating conditions where the load mutation intensity is below 40%, both methods maintain high early warning accuracy. The accuracy of the method in this invention is above 98.8%, while the accuracy of the traditional HMM method is above 95%. The performance difference between the two is small, which is within the effective application range of static prior models under stable operating conditions. However, as the load mutation intensity continues to increase, the traditional HMM method, due to its use of a static state transition probability matrix based on historical statistical priors, cannot detect in real time the accelerated physical degradation of equipment caused by the coupling of excess heat energy accumulation and transient load surges. Its ability to capture sudden structural degradation signals is significantly reduced. The accuracy of early warnings showed a significant downward trend, reaching only 38.7% under extreme peak load conditions. In contrast, this invention quantifies the microscopic physical losses of equipment by real-time calculation of the thermoelectric stress index and generates a dynamic degradation acceleration factor by combining the anomaly degree of heat dissipation response and the degree of fan speed fluctuation. This reshapes the state transition probability of the hidden Markov model, enabling the model to dynamically perceive the degradation acceleration process under severe conditions. The early warning accuracy can still be maintained at over 95.2%, improving the accuracy of operation and maintenance management of data center hardware under tidal dynamic loads.

[0091] Figure 3 The results show a comparison of the average monthly unplanned downtime of a single device managed by the method of this invention and the traditional HMM method over the entire 60-month service life of the device. The horizontal axis represents the device operating time in months, fully covering the entire life cycle of the device from commissioning to the end of its service life. The vertical axis represents the average monthly unplanned downtime of a single device, which is defined as the number of service interruption downtime events caused by hardware performance degradation without effective early warning, excluding downtime caused by non-hardware degradation factors such as planned maintenance shutdowns and network failures.

[0092] Meanwhile, this test strictly follows the bathtub curve failure theory widely used in semiconductor devices and electronic equipment, dividing the entire equipment lifecycle into three stages: the initial stage (early failure period, 0-18 months), the mid-stage (accidental failure period, 18-42 months), and the final stage (wear-out failure period, 42-60 months). As shown in the figure, in the initial stage of the lifecycle, the number of equipment downtimes managed by both methods is at a low level. This is because the equipment has just completed its initial break-in period, and the main hardware structure and cooling system are in good condition, resulting in a low probability of hardware degradation failure events. In this stage, the method of this invention captures early response anomalies in the cooling system through cross-correlation analysis of temperature and fan speed sequences, identifying and intervening in early hidden dangers of the equipment, and controlling the monthly unplanned downtime of a single device to a very low value, lower than the traditional HMM method. In the mid-lifecycle stage, as equipment enters a period of stable service with occasional failures, the number of downtimes using traditional HMM methods begins to significantly exceed that of the method described in this invention. This is because the method described in this invention assesses the urgency of equipment maintenance in real time and employs load regulation strategies such as online computing power thermal migration and dynamic power limits to address sudden thermoelectric coupling shocks during this stage. This reduces the continuous physical wear and tear on the equipment and minimizes downtime events caused by sudden load shocks. In the late-lifecycle stage, as equipment enters its wear-out and failure period, irreversible physical aging and structural wear occur. Traditional HMM methods cannot predict the risk of sudden failures due to equipment aging and can only adopt a passive maintenance model of repairing after a failure. The number of unplanned downtimes per month increases exponentially, reaching a maximum of 2.5 times per month. In contrast, the method described in this invention, through a differentiated maintenance strategy throughout the entire lifecycle, assesses the failure risk of equipment in advance during its late stages and promptly prompts maintenance personnel to perform hardware replacement when the maintenance urgency exceeds a threshold. This reduces the risk of business interruption throughout the equipment's lifecycle and ensures the continuous and stable operation of data center services.

Claims

1. A method for full lifecycle operation and maintenance management and maintenance prompts for computer room hardware equipment, characterized in that, include: The system acquires the current temperature, power load, and fan speed of the computer room hardware, along with the baseline mean and standard deviation of temperature, power load, fan speed, and fan speed determined during the initial operation phase. The value by which the current temperature exceeds the baseline mean is considered excess heat energy. The difference between the current power load and the previous power load is considered transient load change. The system integrates excess heat energy and transient load change to determine the thermoelectric stress index, including: ;in, The thermoelectric stress index is a dimensionless comprehensive evaluation index representing the current temperature of the computer room hardware equipment. The temperature of the hardware equipment in the computer room at the current moment. The average temperature is the baseline value. The standard deviation of the temperature reference is... This represents the electrical power load of the computer room hardware at the current moment. This refers to the electrical power load of the computer room hardware equipment at the previous moment. The average value of the electrical power load is the reference value. To find the maximum value function, The preset value is 0.00001, a very small positive number used to prevent the denominator from being zero. The absolute value symbol is used; the recent temperature sequence and recent fan speed sequence at the current moment are extracted, and the co-variation trend is analyzed to determine the degree of heat dissipation response anomaly; based on the fan speed, the baseline mean of fan speed, and the baseline standard deviation of fan speed, the degree of fan speed fluctuation is determined, and the degree of fluctuation and the degree of heat dissipation response anomaly are used to adjust the thermoelectric comprehensive stress index to generate a degradation acceleration factor, including: ;in, The factors that accelerate the degradation of data center hardware at the current moment. This refers to the degree of thermal response abnormality of the computer room hardware equipment at the current moment. This indicates the degree of fluctuation in fan speed. , The current fan speed. This is the average fan speed. The standard deviation of fan speed is the baseline. The preset value is 0.00001, a very small positive number used to prevent the denominator from being zero. The absolute value symbol is used. The state transition probabilities determined by the Hidden Markov Model are reshaped using a degradation acceleration factor, including: obtaining the performance state level of the data center hardware equipment at the current moment based on the pre-built Hidden Markov Model; taking the performance state level at the current moment and all performance state levels above it as the target state level; obtaining the original static transition probabilities from the current performance state level to each target state level; calculating the state span between each target state level and the current performance state level; using the state span and degradation acceleration factor as mapping variables; substituting them into the algebraic approximation equation based on the second-order Taylor expansion to construct nonlinear amplification weights; using the amplification weights to weight and normalize the original static transition probabilities from the current performance state level to each target state level to obtain the reshaped state transition probabilities; assessing the maintenance urgency based on the reshaped state transition probabilities; and performing full lifecycle operation and maintenance management and maintenance prompts based on the maintenance urgency.

2. The method for full lifecycle operation and maintenance management and maintenance prompts of computer room hardware equipment according to claim 1, characterized in that, The baseline mean temperature, baseline standard deviation temperature, baseline mean electrical power load, baseline mean fan speed, and baseline standard deviation fan speed determined during the initial operation phase are determined as follows: During the initial operation phase when the computer room hardware equipment is put into use and in a healthy state, the initial temperature sequence, initial electrical power load sequence, and initial fan speed sequence are obtained; based on the initial temperature sequence, initial electrical power load sequence, and initial fan speed sequence, mathematical expectation and variance calculations are performed to generate the baseline mean temperature, baseline standard deviation temperature, baseline mean electrical power load, baseline mean fan speed, and baseline standard deviation fan speed.

3. The method for full lifecycle operation and maintenance management and maintenance prompts of computer room hardware equipment according to claim 1, characterized in that, The thermal response anomaly is determined as follows: the recent temperature sequence and the recent fan speed sequence are extracted according to a preset time window. The cross-correlation coefficients of the recent temperature sequence and the recent fan speed sequence under different delay steps are calculated. The maximum value of the cross-correlation coefficient is selected as the maximum delay cross-correlation coefficient. The complement of the maximum delay cross-correlation coefficient is calculated. If the complement is greater than 0, the complement is used as the thermal response anomaly of the computer room hardware at the current moment. If the complement is not greater than 0, the thermal response anomaly of the computer room hardware at the current moment is set to 0.

4. The method for full lifecycle operation and maintenance management and maintenance prompts of computer room hardware equipment according to claim 1, characterized in that, The maintenance urgency assessment based on the reshaped state transition probability includes: selecting target state levels with a higher level than the current performance state level as degraded target state levels; obtaining the reshaped state transition probability from the current performance state level to the degraded target state level, and calculating the state span between the degraded target state level and the current performance state level, mapping this state span to the physical degradation cost weight of the degraded target state level; and weighting and summing the reshaped state transition probabilities from the current performance state level to each degraded target state level based on the degradation cost weight of each degraded target state level to obtain the maintenance urgency of the data center hardware equipment at the current moment.

5. The method for full lifecycle operation and maintenance management and maintenance prompts of computer room hardware equipment according to claim 1, characterized in that, Based on maintenance urgency, full lifecycle operation and maintenance management and maintenance prompts are implemented, including: determining whether the data center hardware equipment is currently in the early, middle, or late stage of its lifecycle; when in the early or middle stage, if the current maintenance urgency is greater than a preset maintenance urgency threshold, reducing the power load of the data center hardware equipment; if the current maintenance urgency is not greater than the preset maintenance urgency threshold, maintaining the power load of the data center hardware equipment; when in the late stage, if the current maintenance urgency is greater than the preset maintenance urgency threshold, prompting maintenance personnel to perform hardware equipment repair operations; if the current maintenance urgency is not greater than the preset maintenance urgency threshold, maintaining monitoring of the operational status of the data center hardware equipment.

6. The method for full lifecycle operation and maintenance management and maintenance prompts of computer room hardware equipment according to claim 5, characterized in that, The method to determine whether the hardware equipment in the data center is in the early, middle or late stage of its life cycle at the current moment is as follows: obtain the performance state level of the hardware equipment in the data center at the current moment based on the Hidden Markov Model, as well as the highest state level that represents complete failure as predefined. Calculate the ratio of the current performance state level to the highest state level; The ratio is compared with the preset lifecycle segmentation threshold, and the comparison result determines whether the computer room hardware is in the early, middle or late stage of its lifecycle at the current moment.

7. The method for full lifecycle operation and maintenance management and maintenance prompts of computer room hardware equipment according to claim 1, characterized in that, The data center hardware includes server nodes and core switches. When the data center hardware consists of server nodes, the temperature of the processor chip is read as the current temperature. When the data center hardware consists of core switches, the temperature of the integrated circuit switching chip is read as the current temperature.

Images