Server heat dissipation model correction system and method

Abstract

The invention relates to the technical field of model correction, in particular to a server heat dissipation model correction system and method. The method comprises the following steps: arranging a plurality of temperature sensors on a heat transfer path of a server heat dissipation system, and obtaining a temperature time history curve of each measuring point under power step excitation; performing multi-exponential decay fitting on the temperature time history curve of each measuring point, decomposing to obtain a plurality of time constants and corresponding temperature rise contribution amplitudes, and forming an experimental time constant set; the method comprises the following steps: establishing a computational fluid mechanics simulation model of a server cooling system, setting the same power step input, extracting a simulation temperature time history curve, and performing multi-exponential decay fitting to form a simulation time constant set; structured alignment between a simulation result and an actual measurement result is achieved through a time constant, and the stability and credibility of the model are remarkably improved through proportion type convergence judgment and multi-power working condition verification.

Description

Technical Field

[0001] This invention relates to the field of model calibration technology, and in particular to a server heat dissipation model calibration system and method. Background Technology

[0002] With the continuous development of high-density servers and data centers towards higher computing power and higher power consumption, the heat generation of internal server components is showing a significant upward trend. The design and evaluation of the heat dissipation system has become a key factor affecting the reliability and energy efficiency of the equipment. Server heat dissipation involves three basic heat transfer methods: thermal conduction, which is the process of heat transfer within a solid. In a server, the heat generated by the CPU is first transferred to the heatsink base through thermal grease and then diffuses outward along the heatsink fins; thermal convection, which carries away heat through the flow of fluid (usually air), is how the airflow generated by the server fan carries heat away and exhausts it from the chassis when it passes over the heatsink fins; and thermal radiation, which radiates heat outward in the form of electromagnetic waves. Although it accounts for a smaller proportion of server heat dissipation, it cannot be ignored for high-temperature components.

[0003] Traditional server cooling model correction methods often rely on steady-state temperature data, adjusting convective heat transfer coefficients, thermal resistance parameters, or boundary conditions to make the simulated steady-state results approximate the measured temperatures. However, these methods neglect the thermal inertia and multi-timescale heat transfer characteristics prevalent in server cooling systems, failing to effectively reflect the response differences of different devices and structural levels during transient processes, and easily leading to the problem of "steady-state matching, dynamic distortion." Furthermore, existing correction processes lack unified and quantifiable convergence criteria, often relying on repeated trial and error based on manual experience, which is not only inefficient but also makes it difficult to guarantee consistency and reusability across different projects. Summary of the Invention

[0004] Therefore, it is necessary for the present invention to provide a server heat dissipation model calibration system and method to solve at least one of the above-mentioned technical problems.

[0005] To achieve the above objectives, a server heat dissipation model calibration method includes the following steps: Step S1: Arrange multiple temperature sensors along the heat transfer path of the server cooling system to obtain the temperature time history curves of each measuring point under power step excitation. Step S2: Perform multi-exponential decay fitting on the temperature time history curve of each measuring point to decompose it into several time constants and their corresponding temperature rise contribution amplitudes, forming a set of experimental time constants; Step S3: Establish a computational fluid dynamics simulation model of the server cooling system, set the same power step input, extract the simulation temperature time history curve, perform multi-exponential decay fitting, and form a set of simulation time constants. Step S4: Calculate the relative deviation between the experimental time constant set and the simulation time constant set; determine the corresponding physical heat transfer path based on the numerical range of the time constant, and identify the path to be corrected where there is a time constant deviation; Step S5: Adjust the physical parameters corresponding to the path to be corrected in the computational fluid dynamics simulation model, resimulate and extract the time constants, iterate and adjust until the relative deviation of all time constants is less than the preset threshold, and complete the model correction.

[0006] The present invention also provides a server thermal model calibration system for performing the above-described server thermal model calibration method, the server thermal model calibration system comprising: The data acquisition module is used to deploy multiple temperature sensors along the heat transfer path of the server cooling system to acquire the temperature time history curves of each measuring point under power step excitation. The parameter extraction module is used to perform multi-exponential decay fitting on the temperature time history curve of each measuring point, decompose it to obtain several time constants and their corresponding temperature rise contribution amplitudes, and form a set of experimental time constants. The simulation calculation module is used to establish a CFD simulation model of the server heat dissipation system. It sets the same power step input, extracts the simulation temperature time history curve, performs multi-exponential decay fitting, and forms a set of simulation time constants. The error diagnosis module is used to calculate the relative deviation between the experimental time constant set and the simulation time constant set; determine the corresponding physical heat transfer path based on the numerical range of the time constant, and identify the path to be corrected where there is a time constant deviation. The parameter calibration module is used to adjust the physical parameters corresponding to the path to be calibrated in the computational fluid dynamics simulation model, re-simulate and extract the time constants, and iteratively adjust them until the relative deviation of all time constants is less than a preset threshold, thus completing the model calibration.

[0007] This invention achieves significant engineering value and technological advancements in server heat dissipation modeling and prediction by using closed-loop coupling calibration of experimental measured data with CFD simulation models. First, by introducing power step excitation into a real server heat dissipation system and acquiring temperature time-history curves at multiple measurement points, model calibration no longer relies on a single steady-state condition but fully reflects the dynamic characteristics of the system during transient thermal response, fundamentally enhancing the model's ability to characterize thermal inertia, heat capacity distribution, and differences in heat transfer paths. Second, by performing multi-exponential decay decomposition on the temperature time-history curves and extracting a set of time constants, the complex three-dimensional heat transfer process is mapped into parameters with clear physical meaning. This allows the deviation between simulation and measured results to be quantified, compared, and iteratively corrected, thus avoiding the instability and randomness caused by relying solely on empirical adjustments based on temperature peaks or steady-state errors.

[0008] Building upon this foundation, a convergence determination mechanism based on the proportion of time constants is introduced, transforming the model calibration process from "whether it is close" to "whether it is consistent at the overall thermal dynamic level." This effectively prevents overfitting at individual measurement points or in local areas, ensuring the model's consistency and reliability in global thermal response. Furthermore, after model convergence, verification simulations are conducted using various power step amplitudes to examine the model's temperature prediction deviations under different load disturbances. This ensures that the calibrated CFD model is not only applicable to a single calibration condition but also possesses excellent condition transferability and generalization performance. This multi-level, multi-scale calibration and verification mechanism results in a final heat dissipation simulation model that significantly outperforms traditional methods in terms of accuracy, stability, and applicability. It provides reliable computational basis for server heat dissipation structure optimization, airflow design adjustment, power scheduling strategy formulation, and thermal risk assessment under extreme conditions. This reduces experimental costs and shortens the design cycle while improving the scientific rigor and predictability of heat dissipation system design and operation decisions. Attached Figure Description

[0009] Other features, objects, and advantages of the invention will become more apparent from the following detailed description of non-limiting embodiments with reference to the accompanying drawings: Figure 1 This is a schematic diagram of the steps of a server heat dissipation model calibration method according to the present invention; Figure 2 This is a flowchart of server heat dissipation model calibration based on experimental testing and CFD simulation according to an embodiment of the present invention; Figure 3 This is a schematic diagram of a server heat dissipation model correction system according to the present invention. Detailed Implementation

[0010] The technical method of the present invention will now be clearly and completely described with reference to the accompanying drawings. Obviously, the described embodiments are only some, not all, of the embodiments of the present invention. All other embodiments obtained by those skilled in the art based on the embodiments of the present invention without inventive effort are within the scope of protection of the present invention.

[0011] Furthermore, the accompanying drawings are merely illustrative of the invention and are not necessarily drawn to scale. The same reference numerals in the drawings denote the same or similar parts, and therefore repeated descriptions of them will be omitted. Some block diagrams shown in the drawings are functional entities and do not necessarily correspond to physically or logically independent entities. These functional entities can be implemented in software, in one or more hardware modules or integrated circuits, or in different network and / or processor methods and / or microcontroller methods.

[0012] It should be understood that although the terms "first," "second," etc., may be used herein to describe various units, these units should not be limited by these terms. These terms are used merely to distinguish one unit from another. For example, without departing from the scope of the exemplary embodiments, a first unit may be referred to as a second unit, and similarly, a second unit may be referred to as a first unit. The term "and / or" as used herein includes any and all combinations of one or more of the associated listed items.

[0013] To achieve the above objectives, please refer to Figures 1 to 3 This invention provides a method for calibrating a server heat dissipation model, the method comprising the following steps: Step S1: Arrange multiple temperature sensors along the heat transfer path of the server cooling system to obtain the temperature time history curves of each measuring point under power step excitation. Step S2: Perform multi-exponential decay fitting on the temperature time history curve of each measuring point to decompose it into several time constants and their corresponding temperature rise contribution amplitudes, forming a set of experimental time constants; Step S3: Establish a CFD simulation model of the server cooling system, set the same power step input, extract the simulation temperature time history curve, perform multi-exponential decay fitting, and form a set of simulation time constants. Step S4: Calculate the relative deviation between the experimental time constant set and the simulation time constant set; determine the corresponding physical heat transfer path based on the numerical range of the time constant, and identify the path to be corrected where there is a time constant deviation; Step S5: Adjust the physical parameters corresponding to the path to be corrected in the computational fluid dynamics simulation model, resimulate and extract the time constants, iterate and adjust until the relative deviation of all time constants is less than the preset threshold, and complete the model correction.

[0014] See Figure 2 This diagram illustrates the overall workflow of a server heat dissipation model calibration method. The process consists of a parallel experimental testing system and a CFD simulation system, and iterative calibration and verification of model parameters are achieved through comparative analysis of time constant characteristics. Figure 2As shown, on the experimental testing system side, multiple temperature measuring points are first deployed at key heat transfer path locations in the server cooling system. Experimental tests are conducted under preset power step excitation conditions to obtain temperature time history data corresponding to each measuring point. Based on the obtained temperature time history data, the temperature rise response curves of each measuring point are further processed using multi-exponential fitting to extract a set of experimental time constants reflecting the dynamic characteristics of different heat transfer processes. Simultaneously, on the CFD simulation system side, based on the same boundary conditions and power loading conditions as the experimental test, a simulation model of the server cooling system is constructed and transient thermal simulation calculations are performed to obtain the simulated temperature response results for each corresponding measuring point. The simulated temperature time history data is processed using the same time constant extraction method as the experimental side to obtain a set of simulated time constants, ensuring the comparability of experimental and simulated data at the feature level. Subsequently, as... Figure 2 As shown in the intermediate judgment module, the experimental time constant set is matched with the simulation time constant set, the relative deviation between each time constant is calculated, and it is determined whether the deviation exceeds a preset threshold. When the judgment result is negative, that is, the deviation of each major time constant is within the allowable range, the model directly enters the verification pass state, indicating that the current CFD heat dissipation model can reflect the real server heat dissipation behavior well. When the judgment result is positive, that is, there is a situation where the time constant deviation exceeds the preset threshold, a corresponding parameter adjustment scheme is generated according to the mapping relationship between the time constant and the physical heat transfer path, and the heat capacity parameter, thermal resistance parameter, or thermal conductivity parameter in the CFD simulation model is specifically corrected. The corrected model re-enters the CFD simulation calculation process, forming a closed-loop correction process based on the time constant deviation judgment, until the deviation threshold requirement is met and the verification is passed.

[0015] Furthermore, in step S1, the temperature sensor is arranged on the surface of the CPU chip, the center of the heat sink base, the middle of the heat sink fins, and the heat sink air outlet; the temperature sensor is a fast-response sensor with a response time of less than 0.1 seconds; the data acquisition frequency is 10Hz to 50Hz, and the sampling duration is 300 seconds to 600 seconds.

[0016] In one embodiment, key heat transfer nodes in the server cooling system are analyzed, and temperature sensors are deployed at different physical levels based on the actual heat transfer path and time scale characteristics. Specifically, a first temperature sensor is deployed on the CPU chip surface to directly acquire the transient temperature rise response of the chip junction under power step excitation. This measurement point reflects the rapid thermal response characteristics of the chip and its package. Secondly, a second temperature sensor is deployed at the center of the heat sink base, located in the main heat diffusion region between the chip and the heat sink, to characterize the medium-time scale response during heat conduction from the chip to the heat sink. Thirdly, a third temperature sensor is deployed in the middle of the heat sink fins to acquire the temperature change characteristics during heat diffusion and convection heat transfer within the fin array. Finally, a fourth temperature sensor is deployed at the heat sink outlet to acquire the temperature change of the exhaust air after heat transfer through the fins, reflecting the overall cooling effect at a system-level, slower time scale. These measurement points are distributed sequentially along the main heat transfer path of the server cooling system, ensuring that the acquired temperature time-history data covers the dynamic characteristics corresponding to different physical heat transfer paths.

[0017] In one embodiment, to ensure that the temperature time history curve accurately reflects the transient thermal response process under power step excitation, a fast-response temperature sensor with a response time of less than 0.1 seconds is selected, such as a thin-film thermocouple or a miniature platinum resistance sensor. This type of sensor can quickly follow the temperature change after a power surge, avoiding distortion of the initial part of the temperature time history curve due to excessive thermal inertia of the sensor itself.

[0018] For example, during the instantaneous switch of server CPU power from idle to full load, a fast-response sensor can capture the temperature change trend within 0.1 seconds, thus providing a reliable data basis for extracting millisecond to second-level time constants in subsequent multi-exponential decay fitting.

[0019] In one embodiment, the temperature data acquisition parameters are set according to the dynamic characteristics of the server's cooling system. The data acquisition frequency is selected to be between 10Hz and 50Hz to balance data volume and computational complexity while ensuring time resolution. When the acquisition frequency is below 10Hz, it may not be able to fully depict the temperature change details of the rapid thermal response path; when the acquisition frequency is above 50Hz, data redundancy increases, which adversely affects the efficiency of subsequent fitting and processing. The sampling duration is set to 300 to 600 seconds to cover the complete process from the start of power step excitation to the system reaching thermal steady state.

[0020] For example, by setting the sampling frequency to 20 Hz and the sampling duration to 400 seconds, the complete temperature time history curves of the initial rapid heating stage, the intermediate slow diffusion stage, and the final steady-state stage can be obtained simultaneously.

[0021] It should be noted that the upper limit of the sampling duration is not used to introduce additional steady-state information, but to ensure that the slow heating path with the largest time constant in the system can be fully represented, thereby avoiding truncation errors in the subsequent time constant extraction process.

[0022] Furthermore, the multi-exponential decay fitting in step S2 includes: Extract the initial temperature value at time zero and the final stable steady-state temperature value from the temperature time history curve, and calculate the difference between the two as the total temperature rise; In one embodiment, the moment when the power step is applied is defined as the time zero point, and the temperature value corresponding to that moment is determined as the initial temperature value. Subsequently, a time interval with a temperature change rate lower than a preset threshold is selected at the end of the temperature time history curve, and the temperature values ​​within this interval are averaged to obtain the final stable steady-state temperature value. By calculating the difference between the steady-state temperature value and the initial temperature value, the total temperature rise of the measuring point under the power step condition is obtained, which is used to characterize the overall temperature rise level of the measuring point under thermal steady state.

[0023] For example, in the temperature curve of a certain measuring point, the initial temperature is 40 degrees Celsius and the steady-state temperature is 68 degrees Celsius, then the total temperature rise is determined to be 28 degrees Celsius.

[0024] A multi-exponential decay model is established to represent that the temperature rise at any time in the temperature time history curve is equal to the sum of several exponential terms. Each exponential term is in the form of 1 minus a negative exponent with the natural constant as the base and time divided by the time constant, and then multiplied by the temperature rise contribution amplitude. The number of exponential terms is set to 3-4.

[0025] In one embodiment, after obtaining the total temperature rise, a multi-exponential decay model is constructed to describe the time-history behavior of the temperature at the measuring point. This model assumes that the temperature rise process after a power step excitation can be characterized by several superimposed exponential response terms, each of which describes a class of heat transfer processes with similar time scales.

[0026] Specifically, the temperature rise at any given moment is represented as the sum of multiple exponential terms. Each exponential term is mathematically represented as "1 minus the negative exponent of time divided by the corresponding time constant with the natural constant as the base", multiplied by the temperature rise contribution value corresponding to that exponential term. The number of exponential terms is set according to the structural complexity of the server's heat dissipation system, typically selecting 3 to 4 to cover thermal processes at different time scales, such as chip rapid response, structural diffusion, and environmental heat exchange.

[0027] For example, when four exponential terms are selected, the model can simultaneously reflect thermal response characteristics at the millisecond, second, and ten-second levels or more.

[0028] Furthermore, the multi-exponentially decaying pseudo-reducing body in step S2 includes: Using a multi-exponential decay model as the fitting function, the temperature time history curve as the fitting target, and the sum of all amplitude coefficients as the total temperature rise as the constraint, the time constant and temperature rise contribution amplitude of each exponential term are solved. In one embodiment, the established multi-exponential decay model is used as the fitting function, and the temperature time history curve corresponding to the measurement point is used as the fitting target. A constrained nonlinear fitting method is employed to solve for the time constant and temperature rise contribution amplitude of each exponential term. A constraint condition is introduced during the fitting process: the sum of the temperature rise contribution amplitudes of all exponential terms equals the previously determined total temperature rise, thereby ensuring the consistency of the fitting results in an energy sense. By minimizing the error between the model-calculated temperature and the actual measured temperature under this constraint, a set of optimal time constant parameters and corresponding temperature rise contribution amplitudes are obtained.

[0029] For example, when fitting a certain measuring point, the four time constants obtained correspond to the fast response, medium-speed diffusion, slow equilibrium and ambient temperature rise processes, respectively, and the sum of their corresponding temperature rise contribution amplitudes is consistent with the total temperature rise of 28 degrees Celsius.

[0030] All time constants and their corresponding temperature rise contribution values ​​at the same measuring point are combined in sequence to form the time constant feature vector of that measuring point. The time constant feature vectors of multiple measuring points are summarized into the experimental time constant set.

[0031] In one embodiment, after completing the multi-exponential decay fitting for a single measurement point, all time constants obtained for that measurement point and their corresponding temperature rise contribution amplitudes are ordered and combined to form a time constant feature vector for that measurement point. During the combination, the time constants are arranged in ascending order, maintaining a one-to-one correspondence between the time constants and their corresponding temperature rise contribution amplitudes to ensure that the feature vector can stably characterize the multi-timescale thermal response characteristics of that measurement point. Subsequently, the above process is performed on all measurement points in the server cooling system, and the time constant feature vectors obtained from each measurement point are summarized to form an experimental time constant set.

[0032] For example, four measurement points—the CPU surface, the heatsink base, the middle of the fins, and the air outlet—form four time constant feature vectors, which together constitute the experimental time constant set used for subsequent comparison and analysis with simulation results.

[0033] Furthermore, the process of solving for the time constants and temperature rise contribution magnitudes of each exponential term also includes the step of determining initial values: The temperature values ​​at each sampling time are extracted from the temperature time history curve, and the difference between the temperature value at each time and the initial temperature value is calculated to form a temperature rise data sequence at each time. In one embodiment, to improve the stability and convergence of the nonlinear fitting of the multi-exponential decay model, the temperature time history curve is numerically transformed to construct a temperature rise data sequence before formally solving for the time constants and temperature rise contribution amplitudes of each exponential term. Specifically, the temperature values ​​corresponding to each sampling time are read one by one from the temperature time history curve, and the initial temperature value at the zero point of time is used as a reference. The initial temperature value is then subtracted from the temperature value at each sampling time to obtain the temperature rise at the corresponding time, thereby forming a temperature rise data sequence arranged in chronological order.

[0034] For example, when the initial temperature is 40 degrees Celsius and the temperature at a certain sampling moment is 52 degrees Celsius, the temperature rise at that moment is determined to be 12 degrees Celsius.

[0035] Take the natural logarithm of each temperature rise value in the temperature rise data sequence to obtain the logarithmic temperature rise data sequence; In one embodiment, after obtaining the temperature rise data sequence, a natural logarithmic transformation is performed on each temperature rise value to obtain a logarithmic temperature rise data sequence. This step is based on the characteristic that the exponential response exhibits an approximately linear relationship in the logarithmic domain, transforming the original exponential growth behavior into a form that facilitates linear analysis through logarithmic transformation. In the logarithmic temperature rise data sequence, each data point corresponds to a sampling time, and its value is the natural logarithm of the temperature rise at that time.

[0036] For example, when the temperature rise at a certain moment is 12 degrees Celsius, the corresponding logarithmic temperature rise value is the natural logarithm of 12.

[0037] Linear fitting was performed on the last segment of the logarithmic temperature rise data sequence where the time was greater than the preset time. The slope and intercept of the fitted line were obtained by using time as the abscissa and the logarithmic temperature rise value as the ordinate. In one embodiment, linear fitting is performed on the last segment of data in the logarithmic temperature rise data sequence where the time is greater than a preset time threshold. This preset time is used to avoid the nonlinear segment caused by the superposition of multiple rapid response processes in the early stage of temperature rise, retaining only the time interval in the later stage of the temperature rise curve where it is approximately dominated by a single exponential term. Using time as the abscissa and the logarithmic temperature rise value as the ordinate, linear fitting is performed on the selected last segment of data to obtain a fitted straight line, and the slope and intercept parameters of this line are extracted.

[0038] For example, data 60 seconds after the power step is selected as the final interval for linear fitting, resulting in a fitted straight line with a negative slope and a constant intercept.

[0039] Taking the reciprocal of the absolute value of the slope yields the initial value of the first time constant; taking the natural exponent of the intercept yields the initial value of the first temperature rise contribution. In one embodiment, the initial parameter value of the first exponential term is calculated based on the slope and intercept obtained from the linear fitting. Specifically, the reciprocal of the absolute value of the slope of the fitted line is used as the initial value of the first time constant; simultaneously, the natural exponent of the intercept of the fitted line is used as the initial value of the first temperature rise contribution amplitude. This time constant and temperature rise contribution amplitude typically correspond to the dominant thermal path in the later stage of the temperature rise process, and their time scale is relatively large.

[0040] For example, when the absolute value of the slope obtained from the linear fitting is 0.02, the initial value of the first time constant is determined to be 50 seconds; when the natural exponent result corresponding to the intercept is 10, the initial value of the first temperature rise contribution amplitude is determined to be 10 degrees Celsius.

[0041] The initial values ​​of the time constant and the initial value of the temperature rise contribution amplitude of the first time constant and the first temperature rise contribution amplitude are calculated based on the first initial value of the time constant and the first initial value of the temperature rise contribution amplitude, which serve as the starting point for the iteration of nonlinear fitting.

[0042] In one embodiment, after obtaining the initial value of the first time constant and the initial value of the first temperature rise contribution amplitude, the temperature rise data sequence is decomposed based on the exponential term to estimate the initial parameters of the remaining exponential terms. Specifically, the initial parameters of the first exponential term are used to calculate the exponential response value corresponding to each sampling time, and this response value is subtracted from the original temperature rise data sequence time by time to obtain the residual temperature rise data sequence. Subsequently, the logarithmic transformation, final linear fitting, and parameter calculation process are repeatedly performed on the residual temperature rise data sequence to obtain the initial values ​​of the time constant and the initial values ​​of the temperature rise contribution amplitude of the remaining exponential terms in sequence.

[0043] For example, after removing the slow heating component corresponding to the first exponential term, linear fitting is performed again on the remaining temperature rise data to obtain the initial value of the time constant corresponding to the medium-timescale thermal diffusion process. All obtained initial values ​​of the time constant and the initial value of the temperature rise contribution magnitude are then used as the iterative starting point for the nonlinear fitting of the multi-exponential decay model.

[0044] Furthermore, the step of calculating the initial values ​​of the time constants and initial values ​​of the temperature rise contribution amplitude for the remaining exponential terms based on the initial value of the first time constant and the initial value of the first temperature rise contribution amplitude, as the iterative starting point for nonlinear fitting, includes: Calculate the value of the first exponential term at each time step based on the initial value of the first time constant and the initial value of the first temperature rise contribution amplitude. Subtract the value of the first exponential term at the corresponding time step from each temperature rise value in the temperature rise data sequence to obtain the residual data sequence. In one embodiment, using the initial value of the first time constant and the initial value of the first temperature rise contribution amplitude, the value of the first exponential term corresponding to each sampling time is calculated one by one according to the pre-established exponential response form. This exponential term value is used to characterize the temperature rise component generated by the slowest or dominant thermal path during the temperature rise process. Subsequently, the temperature rise value at each sampling time in the original temperature rise data sequence is subtracted from the value of the first exponential term at the corresponding time to obtain the residual data sequence reflecting the residual thermal response characteristics.

[0045] For example, when the total temperature rise at a certain moment is 15 degrees Celsius, and the temperature rise component calculated based on the first exponential term is 9 degrees Celsius, the residual temperature rise at that moment is determined to be 6 degrees Celsius.

[0046] Take the natural logarithm of each residual value in the residual data sequence, select the last segment of data with a time greater than the preset time for linear fitting, and obtain a new slope and intercept; take the reciprocal of the absolute value of the new slope to obtain the initial value of the second time constant, and take the natural exponent of the new intercept to obtain the initial value of the second temperature rise contribution amplitude. In one embodiment, after obtaining the residual data sequence, a natural logarithmic transformation is performed on each residual value in the residual data sequence to construct a residual logarithmic data sequence. Subsequently, referring to the aforementioned method, the last segment of data in the residual logarithmic data sequence with a time greater than a preset time threshold is selected as the fitting interval. Linear fitting is performed with time as the abscissa and the residual logarithmic value as the ordinate to obtain a new fitted line, and the slope and intercept parameters corresponding to the fitted line are extracted. This fitting process is used to identify the next timescale thermal response characteristics that dominate the residual temperature rise process.

[0047] For example, after removing the first exponential term, linear fitting of the residual data within the interval of 30 to 80 seconds after the power step can yield a linear relationship corresponding to the thermal diffusion process on a medium timescale.

[0048] In one embodiment, the initial parameter values ​​for the second exponential term are calculated based on the new slope and new intercept obtained from linear fitting of the residual data. Specifically, the reciprocal of the absolute value of the new slope is used to obtain the initial value of the second time constant; simultaneously, the natural exponent of the new intercept is used to obtain the initial value of the second temperature rise contribution amplitude. This second exponential term typically corresponds to a faster heat transfer path than the first exponential term, and its time scale is relatively small.

[0049] For example, when the absolute value of the new slope obtained by fitting the residual is 0.1, the initial value of the corresponding second time constant is determined to be 10 seconds; when the result obtained by taking the natural exponent of the new intercept is 5, the initial value of the second temperature rise contribution amplitude is determined to be 5 degrees Celsius.

[0050] All initial values ​​of time constants and initial values ​​of temperature rise contribution magnitude are used as the starting point for nonlinear fitting iterations.

[0051] In one embodiment, the residual data sequence is further processed by separating each term sequentially to obtain the initial values ​​of the time constants and temperature rise contribution amplitudes of the remaining exponential terms, until a preset number of exponential terms is reached. All obtained initial values ​​of time constants and temperature rise contribution amplitudes are then uniformly organized and used as the overall iterative starting point for nonlinear fitting of the multi-exponential decay model.

[0052] For example, when using a four-exponential model, four sets of initial values ​​of time constants and corresponding initial values ​​of temperature rise contribution amplitude can be obtained sequentially, and then input as a whole into the nonlinear fitting algorithm.

[0053] Of particular importance, step S3, establishing the CFD simulation model, includes: establishing a three-dimensional geometric model in CFD software, including the CPU chip, TIM layer, heat sink base, heat sink fins, fan, and PCB board; assigning initial material properties to each component, including thermal conductivity, density, and specific heat capacity; setting the transient power input as a step function; selecting the transient solver with a time step of 0.01 seconds to 0.1 seconds; and setting a virtual temperature probe at the same location as the experiment.

[0054] Furthermore, step S4 includes the following steps: Step S41: Extract the time constant feature vectors of each measurement point from the experimental time constant set and the simulation time constant set respectively, match them according to the magnitude of the time constant values, calculate the relative deviation, and summarize them to form a deviation distribution table; In one embodiment, time constant feature vectors corresponding to the same measurement point are extracted from both the experimental time constant set and the simulation time constant set. For each measurement point, its experimental and simulation time constant feature vectors are first sorted in ascending order of time constant value, and then a one-to-one matching is performed between the two sets of time constants based on the sorted order. After matching, for each pair of experimental and simulation time constants, the relative deviation between them is calculated, and the calculation results, along with information such as the measurement point number and time constant sequence number, are recorded to form a deviation distribution table for subsequent analysis.

[0055] For example, for the CPU surface measurement point, the three time constants obtained from the experiment are 0.3 seconds, 2 seconds and 18 seconds, respectively, and the corresponding time constants obtained from the simulation are 0.4 seconds, 1.6 seconds and 15 seconds. Then, the three sets of relative deviations can be calculated and filled into the deviation distribution table.

[0056] Step S42: Establish a mapping rule between the numerical range of time constant and the physical heat transfer path, wherein the time constant is in the range of 0.1-1 seconds and corresponds to the fast thermal response path of the chip and package, the time constant is in the range of 1-5 seconds and corresponds to the thermal diffusion path of the TIM interface and the heat sink base, the time constant is in the range of 5-30 seconds and corresponds to the thermal balance path of the heat sink fin array, and the time constant is above 30 seconds and corresponds to the slow heating path of the ambient air in the chassis. In one embodiment, thermal response processes with time constants in the range of 0.1 seconds to 1 second are defined as rapid thermal response paths within the chip and its package; thermal response processes with time constants in the range of 1 second to 5 seconds are defined as thermal diffusion paths within the thermal interface material and the heat sink base area; thermal response processes with time constants in the range of 5 seconds to 30 seconds are defined as thermal equilibrium paths within the heat sink fin array; and thermal response processes with time constants greater than 30 seconds are defined as slow temperature rise paths caused by heat exchange between the air and the environment inside the chassis. This mapping rule is used to provide a unified physical interpretation framework for subsequent deviation diagnosis.

[0057] Step S43: For each experimental time constant in the deviation distribution table, determine the corresponding physical heat transfer path type according to the mapping rule to form a path type set corresponding to each measurement point; In one embodiment, for each record in the deviation distribution table, the corresponding experimental time constant value is read, and the corresponding physical heat transfer path type is determined according to the time constant interval to which it belongs. Subsequently, the physical heat transfer path types corresponding to all time constants at the same measurement point are summarized to form a path type set corresponding to that measurement point. This path type set is used to characterize all the main heat transfer paths covered by that measurement point.

[0058] For example, for a measuring point in the middle of a radiator fin, its time constant may fall into both the heat diffusion path interval and the heat equilibrium path interval, thus the path type set at this measuring point contains both of the above-mentioned physical heat transfer paths.

[0059] Step S44: Traverse the deviation distribution table. When the absolute value of the relative deviation of a certain time constant exceeds 15%, search for the physical heat transfer path type corresponding to that time constant from the path type set, and mark the physical heat transfer path as a path to be corrected with parameter errors.

[0060] In one embodiment, when the absolute value of the relative deviation between a certain experimental time constant and its corresponding simulation time constant exceeds a preset threshold of 15%, it is determined that there is a significant inconsistency between the simulated thermal response and the experimental results at that time scale. Subsequently, from the path type set of the measurement point to which the time constant belongs, the physical heat transfer path type corresponding to the time constant interval is searched, and the physical heat transfer path is marked as a path to be corrected with parameter errors.

[0061] For example, if the relative deviation of a time constant falling within the 5-30 second range is 25%, the thermal balance path corresponding to the radiator fin array can be marked as the path to be corrected.

[0062] Of particular importance, the matching based on the magnitude of the time constant in step S41 includes: The time constants at the same measurement point in the experimental time constant set are sorted in ascending order of value to form an experimental time constant sequence; the time constants at the corresponding measurement points in the simulation time constant set are sorted in the same way to form a simulation time constant sequence. When two sequences have the same number of time constants, they are matched one by one according to their sorting positions. When the number of time constants in two sequences is different, calculate the relative difference between each experimental time constant and all simulated time constants, and select the simulated time constant with the smallest relative difference as the matching object; mark the matched simulated time constants to avoid duplicate matching; For time constants in the experimental time constant sequence that cannot find a matching object, add the modeling details of the corresponding physical path to the simulation model and re-simulate; for time constants in the simulation time constant sequence that are not matched, check whether there is redundant physical process modeling in the simulation model.

[0063] Furthermore, step S5 involves adjusting the physical parameters for the path to be corrected, including: Determine the relationship between the simulated and experimental values ​​of the time constant corresponding to the path to be corrected; In one embodiment, after determining the path to be corrected, the simulated and experimental time constant values ​​corresponding to that path are compared and analyzed to clarify the direction of parameter adjustment. Specifically, the experimental and simulated time constants corresponding to the physical heat transfer path are read from the deviation distribution table to determine whether the simulated time constant is too large or too small relative to the experimental time constant. This determination reflects whether the thermal response speed of the path in the current CFD model is too slow or too fast relative to the real system, thus providing a qualitative basis for subsequent parameter adjustment.

[0064] For example, when the simulation time constant of a certain fast thermal response path is 0.8 seconds, while the experimental time constant is 0.5 seconds, it can be determined that the thermal response speed of the path in the simulation model is too slow.

[0065] When the simulation time constant of a fast thermal response path is greater than the experimental time constant, decrease the thermal capacity parameter or increase the thermal conductivity parameter of the corresponding structural component; when the simulation time constant is less than the experimental time constant, increase the thermal capacity parameter or decrease the thermal conductivity parameter. In one embodiment, when the path to be corrected is determined to be a fast thermal response path and its simulation time constant is greater than the experimental time constant, the thermal capacity or thermal conductivity parameters of the related structural components of that path are adjusted accordingly. Specifically, the equivalent thermal capacity parameter of the structural component in the CFD model is appropriately reduced, or its thermal conductivity parameter is increased while keeping the geometry unchanged, thereby accelerating the thermal response speed in the simulation. Conversely, when the simulation time constant of a fast thermal response path is less than the experimental time constant, the simulated thermal response is slowed down by increasing the thermal capacity parameter or decreasing the thermal conductivity parameter, so that the time constant approaches the experimental value.

[0066] For example, for a fast thermal response path corresponding to a chip and its package, when the simulation time constant is too large, the specific heat capacity parameter of the chip package material can be appropriately reduced to shorten the simulation time constant of the path.

[0067] When the simulation time constant of the heat diffusion path is greater than the experimental time constant, decrease the thermal resistance parameter or decrease the heat capacity parameter of the path; when the simulation time constant is less than the experimental time constant, increase the thermal resistance parameter or increase the heat capacity parameter. In one embodiment, when the path to be corrected is a heat diffusion path, the thermal resistance or thermal capacity parameter of the path is adjusted according to the relationship between the simulation time constant and the experimental time constant. When the simulation time constant of the heat diffusion path is greater than the experimental time constant, it indicates that the heat diffusion along the path is too slow in the simulation. In this case, the heat diffusion efficiency can be improved by reducing the equivalent thermal resistance parameter of the path or simultaneously reducing the thermal capacity parameter of the corresponding structure. When the simulation time constant is less than the experimental time constant, the heat diffusion process in the simulation is slowed down by increasing the thermal resistance parameter or increasing the thermal capacity parameter.

[0068] For example, when the simulation time constant corresponding to the TIM interface is significantly greater than the experimental value, the contact thermal resistance parameter of the interface in the model can be appropriately reduced to enhance the thermal conductivity.

[0069] When the simulation time constant of the thermal equilibrium path or the slow heating path deviates from the experimental value, adjust the thermal capacity parameter or thermal resistance parameter of the structure corresponding to that path. In one embodiment, when the path to be corrected belongs to a thermal equilibrium path or a slow heating path, the simulation time constant is mainly corrected by adjusting the heat capacity or thermal resistance parameters of the structure corresponding to the path. Since such paths usually correspond to large-scale thermal processes such as heat sink fin arrays or air inside the chassis, their time constants are greatly affected by the overall thermal inertia and heat transfer conditions of the system. Therefore, local adjustments to the thermal conductivity parameters are not emphasized; instead, time-scale matching is achieved by changing the equivalent heat capacity or heat transfer resistance.

[0070] For example, when the simulation time constant corresponding to the ambient air of the chassis is significantly smaller than the experimental value, the overall heating process in the simulation can be made slower by increasing the equivalent heat capacity parameter of the air region.

[0071] The parameter adjustment range is set so that the change in the time constant reaches 50%-80% of the absolute value of the relative deviation.

[0072] In one embodiment, when adjusting parameters for various physical heat transfer paths, the adjustment range is uniformly limited to ensure the stability of the calibration process. The goal of each parameter adjustment is set to make the change in the corresponding time constant reach 50% to 80% of the current absolute value of the relative deviation, rather than completely eliminating the deviation all at once. This gradual adjustment method avoids oscillations or new deviations in the model response caused by excessive parameter correction.

[0073] For example, when the relative deviation of a certain time constant is 20%, the parameter adjustment range can be controlled within the range that changes the time constant by about 10% to 16%.

[0074] Furthermore, the convergence determination of the iterative adjustment in step S5 includes: The simulation was re-performed using the adjusted CFD simulation model, a new set of simulation time constants was extracted, and the adjusted relative deviation was calculated. In one embodiment, after adjusting the physical parameters of the path to be corrected, a thermal simulation is re-executed based on the updated CFD simulation model to obtain the corrected simulated temperature time history curve. This simulation process uses the same power step input conditions and measurement point locations as the initial simulation stage to ensure the comparability of the results. Subsequently, the newly obtained simulated temperature time history curve is processed using the same multi-exponential decay fitting method as in step S2 to extract a new set of simulated time constants. This set is then compared one by one with the experimental time constant set, and the relative deviations corresponding to each time constant are recalculated.

[0075] For example, after completing one parameter adjustment, the CPU surface measurement points were re-simulated and fitted, and a new time constant of 0.55 seconds was obtained. Compared with the experimental value of 0.5 seconds, the relative deviation was significantly smaller than the level before adjustment.

[0076] The percentage of time constants whose absolute value of the adjusted relative deviation is less than a preset threshold is used to determine convergence; when the percentage is greater than 90%, convergence is determined. In one embodiment, after obtaining the adjusted relative deviation results, statistical analysis is performed on the deviation distribution of all time constants to determine whether the iterative adjustment process meets the convergence condition. Specifically, the number of time constants whose absolute value of the adjusted relative deviation is less than a preset threshold is counted, and the ratio of this number to the total number of time constants is calculated to obtain the proportion of time constants that meet the deviation requirements. When this proportion is greater than 90%, it is determined that the CFD simulation model in the current iterative state is consistent with the experimental results at the time constant level, and thus the parameter correction process is considered to have converged.

[0077] For example, in an experimental time constant set containing 40 time constants, if the absolute value of the relative deviation of 37 time constants is less than a preset threshold, then the corresponding proportion is 92.5%, which satisfies the convergence judgment condition.

[0078] After convergence, the CFD simulation model is verified by setting different power step amplitude values. The temperature time history curves of each measuring point are extracted. When the temperature deviation of the verification condition is less than 5%, the model correction is confirmed to be complete.

[0079] In one embodiment, after the parameter calibration process is determined to have converged, to verify the generalization effectiveness of the calibrated CFD simulation model under different operating conditions, a power step amplitude different from that in the calibration phase is applied to the model, and a verification simulation is performed. In the verification simulation, the measurement point layout, boundary conditions, and simulation settings remain unchanged; only the magnitude of the power step amplitude is changed to simulate the thermal response of the server under different load levels. Subsequently, the temperature time history curves corresponding to each measurement point are extracted from the verification simulation results and compared with the temperature time history curves obtained under the same power step amplitude under experimental conditions to calculate the temperature deviation between the two.

[0080] For example, when the power step amplitude was adjusted from 100 watts in the original correction stage to 80 watts for verification, it was found that the maximum temperature deviation between the simulated temperature curve and the experimental curve at each measuring point was less than 5%.

[0081] See Figure 3 The present invention also provides a server thermal model calibration system 100 for performing the above-described server thermal model calibration method, wherein the server thermal model calibration system 100 includes: Data acquisition module 101 is used to arrange multiple temperature sensors on the heat transfer path of the server heat dissipation system to obtain the temperature time history curves of each measuring point under power step excitation. The parameter extraction module 102 is used to perform multi-exponential decay fitting on the temperature time history curve of each measuring point, decompose it to obtain several time constants and their corresponding temperature rise contribution amplitudes, and form a set of experimental time constants. The simulation calculation module 103 is used to establish a CFD simulation model of the server heat dissipation system. It sets the same power step input, extracts the simulation temperature time history curve, performs multi-exponential decay fitting, and forms a set of simulation time constants. The error diagnosis module 104 is used to calculate the relative deviation between the experimental time constant set and the simulation time constant set; determine the corresponding physical heat transfer path based on the numerical range of the time constant, and identify the path to be corrected where there is a time constant deviation. The parameter correction module 105 is used to adjust the physical parameters corresponding to the path to be corrected in the computational fluid dynamics simulation model, re-simulate and extract the time constants, and iteratively adjust them until the relative deviation of all time constants is less than a preset threshold, thus completing the model correction.

[0082] Therefore, the embodiments should be considered as exemplary and non-limiting in all respects, and the scope of the invention is defined by the appended claims rather than the foregoing description. Thus, it is intended that all variations falling within the meaning and scope of the equivalents of the application be incorporated into the invention.

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

Claims

1. A method for calibrating a server heat dissipation model, characterized in that, Includes the following steps: Step S1: Arrange multiple temperature sensors along the heat transfer path of the server cooling system to obtain the temperature time history curves of each measuring point under power step excitation. Step S2: Perform multi-exponential decay fitting on the temperature time history curve of each measuring point to decompose it into several time constants and their corresponding temperature rise contribution amplitudes, forming a set of experimental time constants; Step S3: Establish a computational fluid dynamics simulation model of the server cooling system, set the same power step input, extract the simulation temperature time history curve, perform multi-exponential decay fitting, and form a set of simulation time constants. Step S4: Calculate the relative deviation between the experimental time constant set and the simulation time constant set; determine the corresponding physical heat transfer path based on the numerical range of the time constant, and identify the path to be corrected where there is a time constant deviation; Step S5: Adjust the physical parameters corresponding to the path to be corrected in the computational fluid dynamics simulation model, resimulate and extract the time constants, iterate and adjust until the relative deviation of all time constants is less than the preset threshold, and complete the model correction.

2. The server heat dissipation model calibration method according to claim 1, characterized in that, In step S1, temperature sensors are arranged on the surface of the CPU chip, the center of the heat sink base, the middle of the heat sink fins, and the heat sink air outlet; the temperature sensors are fast-response sensors with a response time of less than 0.1 seconds; the data acquisition frequency is 10Hz to 50Hz, and the sampling duration is 300 seconds to 600 seconds.

3. The server heat dissipation model calibration method according to claim 2, characterized in that, Step S2, multi-exponential decay fitting, includes: Extract the initial temperature value at time zero and the final stable steady-state temperature value from the temperature time history curve, and calculate the difference between the two as the total temperature rise; A multi-exponential decay model is established to represent that the temperature rise at any time in the temperature time history curve is equal to the sum of several exponential terms. Each exponential term is in the form of 1 minus a negative exponent with the natural constant as the base and time divided by the time constant, and then multiplied by the temperature rise contribution amplitude. The number of exponential terms is set to 3-4.

4. The server heat dissipation model calibration method according to claim 3, characterized in that, The multi-exponential decay pseudo-reducers in step S2 include: Using a multi-exponential decay model as the fitting function, the temperature time history curve as the fitting target, and the sum of all amplitude coefficients as the total temperature rise as the constraint, the time constant and temperature rise contribution amplitude of each exponential term are solved. All time constants and their corresponding temperature rise contribution values ​​at the same measuring point are combined in sequence to form the time constant feature vector of that measuring point. The time constant feature vectors of multiple measuring points are summarized into the experimental time constant set.

5. The server heat dissipation model calibration method according to claim 4, characterized in that, The process of solving for the time constant and temperature rise contribution of each exponential term also includes the step of determining initial values: The temperature values ​​at each sampling time are extracted from the temperature time history curve, and the difference between the temperature value at each time and the initial temperature value is calculated to form a temperature rise data sequence at each time. Take the natural logarithm of each temperature rise value in the temperature rise data sequence to obtain the logarithmic temperature rise data sequence; Linear fitting was performed on the last segment of the logarithmic temperature rise data sequence where the time was longer than the preset time. The slope and intercept of the fitted line were obtained by using time as the abscissa and the logarithmic temperature rise value as the ordinate. Taking the reciprocal of the absolute value of the slope yields the initial value of the first time constant; taking the natural exponent of the intercept yields the initial value of the first temperature rise contribution. The initial values ​​of the time constant and the initial value of the temperature rise contribution amplitude of the first time constant and the first temperature rise contribution amplitude are calculated based on the first initial value of the time constant and the first initial value of the temperature rise contribution amplitude, which serve as the starting point for the iteration of nonlinear fitting.

6. The server heat dissipation model calibration method according to claim 5, characterized in that, The step of calculating the initial values ​​of the time constants and initial values ​​of the temperature rise contribution amplitude for the remaining exponential terms based on the initial value of the first time constant and the initial value of the first temperature rise contribution amplitude, as the iterative starting point for nonlinear fitting, includes: Calculate the value of the first exponential term at each time step based on the initial value of the first time constant and the initial value of the first temperature rise contribution amplitude. Subtract the value of the first exponential term at the corresponding time step from each temperature rise value in the temperature rise data sequence to obtain the residual data sequence. Take the natural logarithm of each residual value in the residual data sequence, select the last segment of data with a time greater than the preset time for linear fitting, and obtain a new slope and intercept; take the reciprocal of the absolute value of the new slope to obtain the initial value of the second time constant, and take the natural exponent of the new intercept to obtain the initial value of the second temperature rise contribution amplitude. All initial values ​​of time constants and initial values ​​of temperature rise contribution magnitude are used as the starting point for nonlinear fitting iterations.

7. The server heat dissipation model calibration method according to claim 6, characterized in that, Step S4 includes the following steps: Step S41: Extract the time constant feature vectors of each measurement point from the experimental time constant set and the simulation time constant set respectively, match them according to the magnitude of the time constant values, calculate the relative deviation, and summarize them to form a deviation distribution table; Step S42: Establish a mapping rule between the numerical range of time constant and the physical heat transfer path, wherein the time constant is in the range of 0.1-1 seconds and corresponds to the fast thermal response path of the chip and package, the time constant is in the range of 1-5 seconds and corresponds to the thermal diffusion path of the heat conduction interface and the heat sink base, the time constant is in the range of 5-30 seconds and corresponds to the thermal balance path of the heat sink fin array, and the time constant is above 30 seconds and corresponds to the slow heating path of the ambient air in the chassis. Step S43: For each experimental time constant in the deviation distribution table, determine the corresponding physical heat transfer path type according to the mapping rule to form a path type set corresponding to each measurement point; Step S44: Traverse the deviation distribution table. When the absolute value of the relative deviation of a certain time constant exceeds 15%, search for the physical heat transfer path type corresponding to that time constant from the path type set, and mark the physical heat transfer path as a path to be corrected with parameter errors.

8. The server heat dissipation model calibration method according to claim 7, characterized in that, Step S5 involves adjusting the physical parameters for the path to be corrected, including: Determine the relationship between the simulated and experimental values ​​of the time constant corresponding to the path to be corrected; When the simulation time constant of a fast thermal response path is greater than the experimental time constant, decrease the thermal capacity parameter or increase the thermal conductivity parameter of the corresponding structural component; when the simulation time constant is less than the experimental time constant, increase the thermal capacity parameter or decrease the thermal conductivity parameter. When the simulation time constant of the heat diffusion path is greater than the experimental time constant, decrease the thermal resistance parameter or decrease the heat capacity parameter of the path; when the simulation time constant is less than the experimental time constant, increase the thermal resistance parameter or increase the heat capacity parameter. When the simulation time constant of the thermal equilibrium path or the slow heating path deviates from the experimental value, adjust the thermal capacity parameter or thermal resistance parameter of the structure corresponding to that path. The parameter adjustment range is set so that the change in the time constant reaches 50%-80% of the absolute value of the relative deviation.

9. The server heat dissipation model calibration method according to claim 8, characterized in that, The convergence criteria for iterative adjustment in step S5 include: The computational fluid dynamics simulation model was re-performed using the adjusted model. A new set of simulation time constants was extracted, and the adjusted relative deviation was calculated. The percentage of time constants whose absolute value of the adjusted relative deviation is less than a preset threshold is used to determine convergence; when the percentage is greater than 90%, convergence is determined. After convergence, the CFD simulation model is verified by setting different power step amplitude values. The temperature time history curves of each measuring point are extracted. When the temperature deviation of the verification condition is less than 5%, the model correction is confirmed to be complete.

10. A server heat dissipation model calibration system, characterized in that, For performing the server thermal model calibration method as described in claim 1, the server thermal model calibration system comprises: The data acquisition module is used to deploy multiple temperature sensors along the heat transfer path of the server cooling system to acquire the temperature time history curves of each measuring point under power step excitation. The parameter extraction module is used to perform multi-exponential decay fitting on the temperature time history curve of each measuring point, decompose it to obtain several time constants and their corresponding temperature rise contribution amplitudes, and form a set of experimental time constants. The simulation calculation module is used to establish a computational fluid dynamics simulation model of the server heat dissipation system. It sets the same power step input, extracts the simulation temperature time history curve, performs multi-exponential decay fitting, and forms a set of simulation time constants. The error diagnosis module is used to calculate the relative deviation between the experimental time constant set and the simulation time constant set; determine the corresponding physical heat transfer path based on the numerical range of the time constant, and identify the path to be corrected where there is a time constant deviation. The parameter calibration module is used to adjust the physical parameters corresponding to the path to be calibrated in the computational fluid dynamics simulation model, re-simulate and extract the time constants, and iteratively adjust them until the relative deviation of all time constants is less than a preset threshold, thus completing the model calibration.

Images