A heat dissipation unit control method and electronic device

Abstract

The application discloses a heat dissipation unit control method and electronic equipment, relates to the technical field of heat dissipation unit control, and comprises the following steps: acquiring acoustic signals and vibration signals of a heat dissipation unit in a server under multiple operating states, determining baseline pressure mark characteristics and resonance frequency spectrum of the heat dissipation unit, and establishing a heat dissipation acoustic archive; acquiring and analyzing real-time acoustic signals and real-time vibration signals of the heat dissipation unit to obtain real-time pressure mark characteristics and real-time frequency spectrum, and judging whether the operating state of the heat dissipation unit is abnormal; in response to abnormal operation of the heat dissipation unit, judging whether the heat dissipation unit has resonance risk; and in response to the existence of resonance risk, generating a fusion adjustment instruction based on a feature comparison result and a frequency spectrum comparison result. The application can realize real-time sensing of the operating state of the heat dissipation unit and pre-control, avoid resonance risk and abnormal state, and reduce the generation of noise, so as to solve the problems of noise control lag, dependence on hardware structure modification and inability to avoid resonance risk in advance in the related art.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of heat dissipation unit control, and particularly relates to a heat dissipation unit control method and an electronic device. BACKGROUND

[0002] With the continuous improvement of server computing performance, the running power and quantity of heat dissipation units (such as fans, pumps, etc.) in the server continue to increase, resulting in increasingly prominent problems of aerodynamic noise and structural resonance in the case; in particular, the noise not only affects the working environment of the data center, but also may mask the early failure characteristics of the equipment, reducing the system reliability; therefore, it is necessary to effectively suppress the noise of the heat dissipation unit.

[0003] In the prior art, on the one hand, a speed regulation mode of temperature feedback is usually used to control the heat dissipation unit to compensate and adjust after the noise is generated, but it belongs to after-the-fact adjustment and cannot identify and avoid the noise risk caused by resonance or abnormal state in advance; on the other hand, there are also schemes for reducing noise through hardware design and material improvement, but they are too dependent on hardware and are difficult to adapt to complex and variable operating conditions. SUMMARY

[0004] The present application provides a heat dissipation unit control method capable of identifying the running state of the heat dissipation unit in real time based on acoustic signals and vibration signals and pre-controlling to avoid resonance risks and abnormal states, thereby reducing the generation of noise, to at least solve the problems of noise control lag, dependence on hardware structure modification and inability to identify and avoid noise risks caused by resonance and abnormal states in advance in the related art.

[0005] The present application provides a heat dissipation unit control method, comprising:

[0006] Acoustic signals and vibration signals of the heat dissipation unit in the server in multiple running states are obtained, baseline pressure pattern features and resonance frequency spectrum of the heat dissipation unit are determined, and a heat dissipation acoustic archive is established;

[0007] In response to the operation of the server, real-time acoustic signals and real-time vibration signals of the heat dissipation unit are obtained and analyzed, real-time pressure pattern features and real-time frequency spectrum are obtained;

[0008] According to the feature comparison result of the real-time pressure pattern features and the baseline pressure pattern features, it is judged whether the running state of the heat dissipation unit is abnormal;

[0009] In response to the abnormal running of the heat dissipation unit, it is judged whether the heat dissipation unit has a resonance risk according to the frequency spectrum comparison result of the real-time frequency spectrum and the resonance frequency spectrum;

[0010] In response to the existence of the resonance risk, a fusion adjustment instruction is generated based on the feature comparison result and the frequency spectrum comparison result and is sent to the heat dissipation unit.

[0011] The application also provides an electronic device, comprising a memory for storing a computer program, and a processor for executing the computer program to implement the steps of any of the heat dissipation unit control methods.

[0012] According to the application, since the acoustic signals and vibration signals of the heat dissipation unit in multiple operating states are acquired, the baseline pressure pattern features and resonance spectrum are determined, and a heat dissipation acoustic archive is established, and then the real-time acoustic signals and vibration signals are acquired during the server operation process for comparison, the dynamic monitoring and intelligent identification of the operating state of the heat dissipation unit can be realized, and the early warning of the abnormal operating state and the resonance risk can be realized. At the same time, since the abnormal response control strategy and the resonance adjustment instruction are respectively issued after the abnormal pressure pattern features or the resonance risk are detected, the adjustment can be performed before the noise or the resonance is significantly amplified, thus overcoming the deficiencies of the prior art that only relies on temperature feedback for speed adjustment after the event or only relies on hardware improvement, so that the noise generated by the abnormal operation or resonance of the heat dissipation unit can be effectively reduced, the server noise can be suppressed, and the failure of the heat dissipation unit caused by excessive noise can be avoided, so that the heat dissipation unit can be found and handled in time, and the operating reliability of the heat dissipation unit can be improved. BRIEF DESCRIPTION OF DRAWINGS

[0013] In order to more clearly illustrate the embodiments of the application, the drawings needed in the embodiments will be briefly introduced. Obviously, the drawings in the following description are only some embodiments of the application, and other drawings can be obtained by those skilled in the art without creative labor.

[0014] Figure 1 An application environment schematic diagram provided for the embodiments of the application;

[0015] Figure 2 A heat dissipation unit control method flowchart provided for the embodiments of the application;

[0016] Figure 3 A heat dissipation unit control device structure block diagram provided for the embodiments of the application;

[0017] Figure 4 A feature analysis workflow schematic diagram provided for the embodiments of the application;

[0018] Figure 5 A hardware connection method schematic diagram for heat dissipation unit control provided for the embodiments of the application;

[0019] Figure 6 An electronic device schematic diagram provided for the embodiments of the application. DETAILED DESCRIPTION

[0020] With reference to the accompanying drawings, the technical solutions in the embodiments of the present application will be described clearly and completely. Obviously, the described embodiments are only part of the embodiments of the present application, rather than all the embodiments of the present application. Based on the embodiments in the present application, all the other embodiments obtained by those of ordinary skill in the art without creative work fall within the protection scope of the present application.

[0021] It should be noted that, in the description of the present application, the terms "comprise", "contain" or any other variants thereof are intended to cover non-exclusive inclusion, so that the process, method, article or equipment comprising a series of elements not only includes those elements, but also includes other elements not explicitly listed or inherent to such process, method, article or equipment. The terms "first", "second" and the like in the present application are used to distinguish similar objects, and are not used to describe a specific order or sequence.

[0022] In order for those skilled in the art to better understand the present application, the present application will be further described in detail below with reference to the accompanying drawings and specific embodiments.

[0023] The heat dissipation unit control method provided by the present application can be applied to the application environment as shown in Figure 1 The terminal 102 communicates with the server 104 through the network to realize remote monitoring and control of the server running state. The terminal 102 can be various electronic devices with data interaction function, such as personal computer, notebook computer, smart phone, tablet computer or portable wearable device, etc.; the server 104 can be a single independent server, or a server cluster composed of multiple servers, which is internally provided with multiple heat dissipation units for system heat dissipation, each heat dissipation unit at least comprising one or more fans, heat dissipation fins and related driving circuits.

[0024] In the embodiment, the heat dissipation unit control method can be deployed in the controller of the server 104 to realize real-time sensing and regulation of the operating state, acoustic characteristics and vibration characteristics of the plurality of heat dissipation elements in the server. Specifically, the server 104 acquires acoustic signals and vibration signals of the heat dissipation unit in a plurality of operating states during operation, determines baseline pressure pattern characteristics and resonance spectrum of the heat dissipation unit through feature extraction and spectrum analysis, and establishes a heat dissipation acoustic archive library accordingly; during actual operation of the server, real-time acoustic signals and real-time vibration signals are continuously acquired, which are analyzed to obtain real-time pressure pattern characteristics and real-time spectrum, and the real-time pressure pattern characteristics are compared with the baseline pressure pattern characteristics to determine whether the operating state of the heat dissipation unit is abnormal; if an abnormality is detected, the real-time spectrum is compared with the resonance spectrum to identify whether there is a risk of resonance, and if there is, a fusion adjustment instruction is generated to make the operating parameters of the heat dissipation unit deviate from the resonance working condition interval and tend to the abnormal response interval, so as to realize noise suppression and stable heat dissipation. Through the scheme, the application realizes algorithm-level monitoring and active regulation of the acoustic and vibration state of the heat dissipation unit in the server, and noise prevention, abnormality detection and resonance adjustment can be completed without adding additional hardware structure, thereby improving the acoustic stability and system reliability of the server operation.

[0025] In one embodiment, as shown in FIG. 1, Figure 2 The heat dissipation unit control method provided by the application comprises the following steps:

[0026] Step 201, acquiring acoustic signals and vibration signals of the heat dissipation unit in a plurality of operating states in the server, determining baseline pressure pattern characteristics and resonance spectrum of the heat dissipation unit, and establishing a heat dissipation acoustic archive library;

[0027] Step 202, in response to the operation of the server, acquiring and analyzing real-time acoustic signals and real-time vibration signals of the heat dissipation unit to obtain real-time pressure pattern characteristics and real-time spectrum;

[0028] Step 203, determining whether the operating state of the heat dissipation unit is abnormal according to the feature comparison result of the real-time pressure pattern characteristics and the baseline pressure pattern characteristics;

[0029] Step 204, in response to abnormal operation of the heat dissipation unit, determining whether the heat dissipation unit has a risk of resonance according to the spectrum comparison result of the real-time spectrum and the resonance spectrum;

[0030] Step 205, in response to the risk of resonance, generating a fusion adjustment instruction based on the feature comparison result and the spectrum comparison result and issuing the instruction to the heat dissipation unit.

[0031] Specifically, the application provides a heat dissipation unit control method, which can realize dynamic monitoring and intelligent identification of the running state of the heat dissipation unit by acquiring acoustic signals and vibration signals of the heat dissipation unit in multiple running states, determining baseline pressure pattern features and resonance frequency spectrum and establishing a heat dissipation acoustic archive, and then comparing real-time acoustic signals and vibration signals acquired during server operation, so as to realize early warning of abnormal running states and resonance risks. Meanwhile, by respectively issuing abnormal response control strategies and resonance adjustment instructions after detecting abnormal pressure pattern features or resonance risks, the heat dissipation unit can be adjusted before noise or resonance is significantly amplified, thereby overcoming the shortcomings of the prior art that relies only on temperature feedback for post-speed adjustment or relies only on hardware improvement, and effectively reducing noise generated due to abnormal running or resonance of the heat dissipation unit, so as to suppress server noise. In addition, the heat dissipation unit faults can be avoided due to excessive noise, so that the heat dissipation unit can be found and handled in time, and the operation reliability of the heat dissipation unit can be improved.

[0032] In one embodiment, the acoustic signals and vibration signals of the heat dissipation unit in the server in multiple running states are acquired, comprising:

[0033] An operation parameter space is constructed, and the operation parameter space at least includes heat dissipation unit operation parameters, server load and environmental temperature. Preferably, the heat dissipation unit operation parameters include fan speed, duty cycle, control mode, etc., the server load includes CPU / GPU utilization and power consumption, and the environmental temperature includes air inlet temperature and computer room temperature.

[0034] Based on parameter representativeness and collection efficiency, one or more test state sequences are determined in the operation parameter space, so that the test state sequences represent multiple state running conditions of the heat dissipation unit. The test state sequences can be determined based on hierarchical sampling or Latin hypercube.

[0035] According to the test state sequence, corresponding test control parameters are generated and issued to the heat dissipation unit to make the heat dissipation unit run based on the test control parameters.

[0036] In response to the running of the heat dissipation unit, the acoustic signals and vibration signals of the heat dissipation unit are collected and stored in association with the corresponding test control parameters to construct a multi-condition acoustic data set of the heat dissipation unit.

[0037] Specifically, in the present embodiment, the operation parameter space including heat dissipation unit operation parameters, server load and environmental temperature is constructed, and the test state sequence is determined based on parameter representativeness and collection efficiency, so that the collected acoustic data can comprehensively cover the multiple state running conditions of the heat dissipation unit. By associating and storing the collected acoustic signals with the test control parameters to construct a multi-condition acoustic data set, the systematicness and consistency of data collection are improved, and high-quality and traceable data support is provided for subsequent feature extraction and baseline establishment, thereby ensuring the representativeness and reliability of the acoustic archive.

[0038] In one embodiment, the baseline acoustic features of the heat dissipation unit are determined, including:

[0039] The multi-condition acoustic data set of the heat dissipation unit is preprocessed and classified to remove environmental noise and distinguish acoustic samples under different operating conditions, to obtain multi-state acoustic samples, wherein the multi-state acoustic samples are obtained by grouping different acoustic samples according to the working condition label;

[0040] Based on the multi-state acoustic samples, the multi-state acoustic features and the multi-state spectral features of the heat dissipation unit are extracted to generate a multi-condition feature set;

[0041] Based on the multi-condition feature set, the feature stability of the sound pressure and the acoustic fingerprint of the heat dissipation unit under different operating conditions is determined, and based on the feature stability, the baseline acoustic features representing the normal operating characteristics of the heat dissipation unit are obtained.

[0042] According to the spectral energy distribution in the multi-condition feature set, the frequency components are statistically analyzed and peak values are detected to determine the resonance spectrum of the heat dissipation unit.

[0043] Specifically, in this embodiment, by preprocessing and classifying the multi-condition acoustic data set, background noise and non-stable samples can be effectively removed, so that the acoustic samples are more distinguishable under different conditions. Further, by extracting the acoustic features and spectral features from the multi-state acoustic samples and generating a multi-condition feature set, and then analyzing the feature stability to obtain baseline acoustic features representing normal operating characteristics, and simultaneously determining the resonance spectrum through spectral energy distribution statistics and peak value detection, the features stored in the acoustic archive can be both stable and sensitive, thereby significantly improving the accuracy of subsequent anomaly recognition and resonance determination.

[0044] In one embodiment, based on the multi-condition feature set, the feature stability of the sound pressure and the acoustic fingerprint of the heat dissipation unit under different operating conditions is determined, and based on the feature stability, the baseline acoustic features representing the normal operating characteristics of the heat dissipation unit are obtained, including:

[0045] The multi-state acoustic features of the heat dissipation unit are analyzed to obtain the multi-state acoustic features;

[0046] The multi-state acoustic features are analyzed to obtain a plurality of acoustic fingerprint feature components and sound pressure feature components to represent the operating characteristics of the heat dissipation unit under different operating conditions;

[0047] According to the feature similarity indexes of the plurality of acoustic fingerprint feature components and the plurality of sound pressure feature components, a plurality of acoustic fingerprint feature stability parameters and a plurality of sound pressure feature stability parameters are obtained;

[0048] From the plurality of voiceprint feature components and sound pressure feature components, a plurality of target voiceprint feature components and target sound pressure feature components with feature stability parameters greater than or equal to a preset stability threshold are screened out, and the plurality of target voiceprint feature components and target sound pressure feature components are aggregated to obtain a baseline pressure pattern feature.

[0049] It is worth noting that in one specific embodiment, the feature similarity index is used to measure the consistency or similarity of the same feature component under different working conditions, which can be obtained by comparing the difference, change amplitude or correlation of the feature component under different samples or working conditions, for example, according to the mean, variance, correlation coefficient or similarity coefficient, etc. When the change of a certain feature component under multiple working conditions is small and the similarity is high, it means that the component is more stable.

[0050] Based on the feature similarity index of each feature component, a corresponding stability parameter can be determined to reflect the overall stability level of the component under different working conditions. The higher the value of the stability parameter, the smaller the fluctuation of the component under multiple working conditions, and the stronger the stability.

[0051] The plurality of target voiceprint feature components and target sound pressure feature components are aggregated to generate a baseline pressure pattern feature for representing the normal operating state of the heat dissipation unit, wherein the aggregation processing is to integrate or statistically analyze the typical values of the target feature components under multiple working conditions, for example, by taking the average, taking the median or using weighted average, etc. The generated baseline pressure pattern feature can accurately reflect the acoustic feature distribution of the heat dissipation unit under normal state.

[0052] Specifically, in this embodiment, by analyzing the multi-state pressure pattern feature, extracting the voiceprint feature components and sound pressure feature components, and calculating their feature similarity indexes to obtain the feature stability parameters, the stability degree of feature change under different working conditions can be quantified. Further, by screening the target features with stability parameters greater than or equal to the preset threshold and aggregating them, the atypical features affected by random noise or short-term disturbance can be effectively excluded, so that the baseline pressure pattern feature obtained can more accurately reflect the acoustic feature regularity of the heat dissipation unit under normal state, thereby improving the accuracy of abnormal detection and the generalization ability of the model.

[0053] In one embodiment, according to the spectral energy distribution in the multi-condition feature set, the frequency components are statistically analyzed and peak detection is performed to determine the resonance spectrum of the heat dissipation unit, including:

[0054] Analyzing the multi-condition feature set to obtain multi-state spectral features of the heat dissipation unit;

[0055] According to the multi-state spectral features, a plurality of frequency component amplitude statistics are obtained, and the amplitude statistics at least include average amplitude and amplitude fluctuation index.

[0056] Based on a plurality of amplitude statistics, local peak amplitudes are identified in the multi-state spectral feature, and the local peak amplitudes are compared with a preset amplitude threshold;

[0057] In response to the local peak amplitude being greater than the preset amplitude threshold, it is determined that the corresponding local peak frequency is a candidate resonance frequency;

[0058] According to the peak stability and amplitude significance across multiple working conditions, a plurality of candidate resonance frequencies are screened to obtain a target resonance frequency set;

[0059] The target resonance frequency set is mapped to the operating parameter space of the heat dissipation unit to generate a resonance spectrum, and a resonance interval table for operating adjustment of the heat dissipation unit is obtained.

[0060] In one specific embodiment, it is worth noting that the peak stability across multiple working conditions is used to measure whether the same frequency component continuously appears peak characteristics under different working conditions. When a certain frequency shows a high amplitude and a small position change in multiple working condition tests, it is considered that the frequency has a high peak stability. This index can reflect whether the frequency corresponding to the structural vibration characteristics is related to the inherent mode of the heat dissipation unit, so as to determine whether it belongs to the potential resonance frequency. The amplitude significance is used to represent the prominence of the peak value relative to the background spectrum. Specifically, the difference between the peak amplitude and the average amplitude or noise baseline amplitude in the adjacent frequency bandwidth can be compared to determine when the peak amplitude is significantly higher than the background level. This indicates that the frequency component is more prominent in energy and has a higher significance.

[0061] When screening the candidate resonance frequencies, both the peak stability and the amplitude significance are considered: only when a certain candidate frequency maintains a high stability under multiple working conditions and its amplitude significance exceeds a preset determination threshold, the frequency is determined as a target resonance frequency.

[0062] In another specific embodiment, a resonance interval table is used to represent the frequency range and corresponding operating conditions in which the heat dissipation unit may resonate under different operating parameters (such as rotational speed, load, or ambient temperature, etc.); specifically, the resonance interval table can be obtained by mapping the target resonance frequency to the operating parameters of the heat dissipation unit, for example, when there is a fixed proportional relationship between the target resonance frequency and the rotational speed of the fan, it can be determined that the operating interval of the fan near the corresponding speed of the frequency is the resonance interval; the resonance interval table can include the following information: the rotational speed interval corresponding to each target resonance frequency, the sound pressure or vibration energy level in the interval, the resonance intensity index or risk level, and the recommended operating adjustment strategy (such as avoiding the speed range or adjusting the control parameters); by establishing the resonance interval table, the current operating state can be quickly compared in real-time operation to determine whether it enters the potential resonance interval, and the operating parameters can be automatically optimized according to the preset adjustment strategy, thereby avoiding the noise or structural fatigue problems caused by resonance, and improving the operating stability and reliability of the heat dissipation unit.

[0063] Specifically, in this embodiment, by analyzing the polymorphic spectral characteristics, calculating the average amplitude and amplitude fluctuation index of multiple frequency components, and identifying significant frequency components based on local peak detection, the high-amplitude frequencies that stably exist under multiple working conditions can be accurately captured; further, by filtering candidate resonance frequencies through cross-condition peak stability and amplitude significance and mapping them to the operating parameter space, a resonance spectrum and a resonance interval table are generated, which can clearly define the resonance risk distribution area in the frequency dimension, thereby providing a quantitative basis for the operating adjustment of the heat dissipation unit and realizing the visualization and controllability of the resonance risk.

[0064] In one embodiment, real-time acoustic signals and real-time vibration signals of the heat dissipation unit are obtained and analyzed to obtain real-time pressure pattern features and real-time spectrum, including:

[0065] The real-time acoustic signals and real-time vibration signals are obtained by the signal acquisition unit arranged around the heat dissipation unit;

[0066] The real-time acoustic signals and real-time vibration signals are respectively converted from time domain signals to frequency domain signals to obtain real-time acoustic frequency domain signals and real-time vibration frequency domain signals;

[0067] Based on a preset sound pressure weight coefficient, the weighted sound pressure level of the real-time acoustic frequency domain signal is determined, and a real-time sound pressure feature component is generated;

[0068] The Mel cepstrum coefficients of the real-time acoustic frequency domain signal are extracted to generate a real-time voiceprint feature component, which is fused with the real-time sound pressure feature component to obtain a real-time pressure pattern feature;

[0069] The real-time vibration frequency domain signal is subjected to spectral peak detection to identify real-time local peak frequencies, and based on the real-time local peak frequencies, a real-time spectrum is obtained.

[0070] Specifically, in the present embodiment, real-time acoustic signals and vibration signals are obtained by arranging signal acquisition units around the heat dissipation unit, and are converted from time domain to frequency domain signals, which can ensure that the signal analysis has high frequency resolution; at the same time, by calculating the weighted sound pressure level based on the preset sound pressure weight and combining the mel cepstral coefficient to extract the voiceprint feature, and then fusing to obtain the real-time voiceprint feature, the running state can be reflected in both sound pressure amplitude and acoustic texture; further, the real-time frequency spectrum can be obtained by peak detection on the real-time vibration frequency domain signal, which can reflect the structure vibration characteristics in real time; thus, the present embodiment can realize high-precision real-time acoustic vibration monitoring of the running state of the heat dissipation unit, and provide fast and reliable feature input for subsequent abnormal and resonance judgment.

[0071] In one specific embodiment, the real-time acoustic signal and the real-time vibration signal are respectively converted from time domain signals to frequency domain signals to obtain real-time acoustic frequency domain signals and real-time vibration frequency domain signals, and short-time Fourier transform is used, which is represented as:

[0072] ;

[0073] Wherein, X(k,m) represents the complex spectrum coefficient at the kth frequency component of the mth frame, that is, the real-time acoustic frequency domain signal, and similarly, the implementation vibration frequency domain signal is Y(k,m); x[n] represents the sampling sequence of the real-time acoustic time domain signal, w[n] represents the window function, for example, Hanning window, Hamming window, which is used to reduce frequency leakage, H represents the frame shift length, and N represents the number of sampling points per frame; j represents the imaginary unit, that is ;

[0074] Based on the preset sound pressure weight coefficient, the weighted sound pressure level of the real-time acoustic frequency domain signal is determined, and the real-time sound pressure feature component is generated, which is represented as:

[0075] ;

[0076] Wherein, L p (m) represents the weighted sound pressure level of the mth frame of the acoustic frequency domain signal, and the average value of the weighted sound pressure level of the continuous frame constitutes the real-time sound pressure feature component; W(k) is the preset sound pressure weight coefficient, which is preferably the A-weighted curve weight; |X(k,m)| 2 is the power spectral density of the mth frame; p0 is the reference sound pressure, which is preferably 20μPa;

[0077] The mel cepstral coefficient of the real-time acoustic frequency domain signal is extracted to generate the real-time voiceprint feature component, which is represented as:

[0078] ;

[0079] wherein c represents a mel-cepstral coefficient vector, i.e. a real-time voiceprint feature component; DCT represents a discrete cosine transform, F mel is a Mel filter bank matrix for mapping an acoustic frequency domain signal as a linear spectrum to a Mel frequency domain;

[0080] is fused with the real-time sound pressure feature component to obtain a real-time pressure-print feature, denoted as:

[0081] ;

[0082] wherein f pw represents a real-time pressure-print feature vector for comprehensively representing the acoustic state of the heat dissipation unit, f spl represents a real-time sound pressure feature component, and a and b are fusion weight coefficients of the sound pressure and the voiceprint for balancing the contribution of the amplitude and the texture information, and are preferably 0.45 and 0.55;

[0083] The real-time vibration frequency domain signal is subjected to a spectrum peak value detection to identify real-time local peak frequencies, including: scanning the amplitude spectrum of each time frame to find local peak values, denoted as:

[0084] P peak ={f k ,A k ∣A k >A k-1 &A k >A k+1 &A k >A threshold};

[0085] wherein P peak represents a set of identified real-time local peak frequencies, f k represents a frequency corresponding to the peak value, obtained by a ratio of a sampling rate and a number of sampling points per frame, A k represents an amplitude at the frequency f k , i.e. |X(k,m)|, and A threshold represents an amplitude threshold value, preferably 5% of an average value of the amplitude spectrum.

[0086] In one embodiment, according to a feature comparison result of the real-time pressure-print feature and the baseline pressure-print feature, it is determined whether the operating state of the heat dissipation unit is abnormal, including:

[0087] The real-time pressure-print feature and the baseline pressure-print feature are subjected to a standardization preprocessing to obtain a real-time feature vector and a baseline feature vector;

[0088] Based on a component dimension determined by the baseline feature vector, a plurality of real-time dimension components in the real-time feature vector are extracted;

[0089] According to the similarity index and the deviation measure between the plurality of real-time dimension components and the corresponding baseline dimension components, a comparison measure set is obtained, and the comparison measure set is weighted and fused to obtain an abnormal score of the heat dissipation unit;

[0090] The abnormal score of the heat dissipation unit is compared with an abnormal threshold value determined by the baseline embossing feature;

[0091] In response to the abnormal score of the heat dissipation unit being greater than or equal to the abnormal threshold value, it is determined that the operating state of the heat dissipation unit is abnormal, and the corresponding real-time embossing feature is determined as an abnormal embossing feature;

[0092] In response to the abnormal score of the heat dissipation unit being less than the abnormal threshold value, it is determined that the operating state of the heat dissipation unit is normal.

[0093] Specifically, in the present embodiment, by standardizing the real-time embossing feature and the baseline embossing feature, and calculating the similarity index and the deviation measure of the two in a plurality of dimension components, the difference between the current operating feature of the heat dissipation unit and the normal state can be quantitatively described; further, by weighting and fusing the comparison measure set to obtain the abnormal score, and comparing it with the abnormal threshold value, the adaptive judgment of the abnormal state can be realized, so as to accurately distinguish the normal fluctuation from the real abnormality, reduce the misjudgment, improve the sensitivity and stability of the abnormality recognition, and thus realize the early detection and rapid response of the abnormal state of the heat dissipation unit.

[0094] In one embodiment, in response to the absence of resonance risk, an abnormal type is determined based on the abnormal embossing feature, and a corresponding abnormal response control strategy is determined and issued to the heat dissipation unit, including:

[0095] The abnormal embossing feature is compared with a pre-stored set of historical abnormal embossing features to determine an abnormal type matching the abnormal embossing feature, wherein the set of historical abnormal embossing features at least includes a historical abnormal type and a corresponding historical abnormal embossing feature; specifically, the comparison between the abnormal embossing feature and the historical sample can be realized by using a feature similarity matching algorithm, for example, by calculating the cosine similarity or dynamic time warping (DTW) distance between the feature vectors to evaluate the matching degree of the two groups of voiceprint features; when the similarity value is greater than a preset threshold value, it is determined that the current abnormality is the same as the abnormal type corresponding to the historical abnormal embossing feature; when all matching degrees are lower than the threshold value, the abnormality is marked as an unknown abnormal type, and a default conservative control strategy is triggered to prevent further spread;

[0096] According to the abnormal type, a corresponding abnormal response control strategy is selected from a preset abnormal response control strategy library, wherein the abnormal response control strategy library is designed for the abnormal state reflected by the abnormal pressure pattern feature, and each strategy template in the strategy library includes: a heat dissipation unit operating parameter field (such as a target rotating speed interval or a rotating speed correction amount Δn) that needs to be adjusted, a necessary control mode description (such as smooth speed reduction or step speed reduction), and a flag indicating whether to trigger an alarm or a maintenance prompt; further, the strategy call can adaptively fine-tune the strategy parameters with reference to the current operating parameters, server load and environmental temperature, so as to balance the heat dissipation performance and risk mitigation; in addition, if the abnormality is determined to be an unknown abnormal type, a conservative strategy (such as reducing the rotating speed and reporting an alarm) is preferentially selected to ensure system safety, and relevant acoustic samples are recorded for offline analysis and subsequent labeling;

[0097] Based on the abnormal response control strategy, control instructions are generated and sent to the heat dissipation unit for execution, and the abnormal type and its control execution result are stored in the heat dissipation acoustic archive library for subsequent updating, for example, during the control execution process, the acoustic features and temperature feedback after the control are continuously monitored, and the strategy effect is judged based on the trend of the fusion change rate of the soundprint sound pressure and the temperature change rate; if the abnormal pressure pattern feature does not converge as expected after the control, a predefined secondary strategy or parameter fine-tuning can be started, and the control results of this process and the corresponding acoustic samples are recorded together to form closed-loop feedback data that can be used for evaluation and optimization;

[0098] Specifically, in the embodiment, the collected abnormal pressure pattern feature is compared with the pre-stored historical abnormal pressure pattern feature set, and the abnormal type is determined by combining the feature similarity matching algorithm, thereby realizing the rapid identification and accurate positioning of the abnormal state of the heat dissipation unit. Further, in the embodiment, the strategy corresponding to the abnormal type is selected from the preset abnormal response control strategy library, and adaptive fine-tuning is performed in combination with the current operating parameters, server load and environmental temperature, so that the heat dissipation unit can adjust the operating parameters or trigger an alarm in a differentiated manner, thereby effectively intervening in the abnormality at an early stage. For unknown abnormal types, the embodiment can preferentially execute a conservative strategy, while recording relevant acoustic samples for offline analysis and strategy optimization, forming a closed-loop self-learning mechanism. By storing the abnormal type and its control execution result in the heat dissipation acoustic archive library and evaluating and adjusting the control effect in combination with real-time feedback, the embodiment can continuously optimize the abnormal response strategy, improve the accuracy of abnormality determination and the timeliness of response, thereby effectively suppressing the noise and risk caused by abnormal operation of the heat dissipation unit, and improving the stability and reliability of the server heat dissipation system.

[0099] In one embodiment, according to the frequency spectrum comparison result of the real-time frequency spectrum and the resonance frequency spectrum, it is judged whether the heat dissipation unit has a resonance risk, including:

[0100] The real-time frequency is obtained by analyzing the real-time spectrum, and the current operating condition of the heat dissipation unit is determined, the current operating condition including at least one or more of the current operating parameter of the heat dissipation unit, the current load of the server and the current temperature of the environment;

[0101] The resonance frequency and the corresponding resonance condition are obtained by analyzing the resonance spectrum, and a plurality of resonance frequency intervals and the corresponding resonance condition intervals are generated in combination with a preset risk reduction coefficient;

[0102] In response to the real-time frequency falling into any resonance frequency interval and / or the current operating condition falling into the corresponding resonance condition interval, it is determined that the heat dissipation unit has a resonance risk;

[0103] In response to the resonance risk, a fusion adjustment instruction is generated based on the feature comparison result and the spectrum comparison result and is issued to the heat dissipation unit, including:

[0104] Based on the abnormal type and the abnormal score represented by the feature comparison result, an abnormal influence weight is determined;

[0105] Based on the resonance frequency interval and the amplitude significance represented by the spectrum comparison result, a resonance risk weight is determined;

[0106] The abnormal influence weight and the resonance risk weight are weighted and fused to obtain a target adjustment coefficient;

[0107] According to the target adjustment coefficient, a fusion adjustment parameter is generated to make the operating parameter of the heat dissipation unit deviate from the resonance condition interval and tend to an abnormal response target interval corresponding to the abnormal type;

[0108] The fusion adjustment parameter is converted into a fusion adjustment instruction and is issued to the heat dissipation unit for execution, and an execution feedback result of the heat dissipation unit is obtained to update the fusion adjustment strategy.

[0109] Specifically, in the present embodiment, not only is a single adjustment based on spectral anomalies, but also the abnormality degree of acoustic embossing features and the resonance frequency significance are further combined to construct abnormality impact weight and resonance risk weight, and the two are weighted and fused to generate a target adjustment coefficient. In this way, multi-source information fusion regulation of abnormal type recognition and resonance risk identification is realized, so that the heat dissipation unit can not only timely escape from the resonance region, but also perform precise optimization control according to the abnormal operation mode, avoiding the decline of cooling performance or excessive regulation caused by a single regulation strategy. Compared with the traditional scheme of triggering fast speed regulation or simply reducing speed according to the resonance frequency, the present embodiment can adaptively generate fusion regulation parameters according to comprehensive factors such as operating condition intensity, spectral peak significance, and acoustic fingerprint deviation degree, and has higher regulation accuracy and device adaptability. At the same time, through feedback learning of the execution of the fusion regulation instruction, the strategy can be dynamically updated to realize continuous adaptation to long-term running state and environmental changes, improving the stability and energy efficiency performance of the heat dissipation system.

[0110] In one specific embodiment, in response to the existence of resonance risk, a fusion regulation instruction is generated based on the feature comparison result and the spectrum comparison result and issued to the heat dissipation unit, including:

[0111] The abnormal type and abnormal score represented in the feature comparison result are analyzed, and an abnormality impact weight is determined through a preset weight mapping model to reflect the influence degree of the abnormal type on the stability of the heat dissipation unit, wherein the abnormal type can include fan shaft deviation, fan blade imbalance, fluid turbulence anomaly, etc., and the abnormal score is calculated according to the similarity deviation between the real-time embossing feature and the baseline embossing feature;

[0112] The resonance frequency interval and amplitude significance represented in the spectrum comparison result are analyzed to determine a resonance risk weight to represent the contribution degree of different resonance components to the overall noise and vibration energy distribution;

[0113] The abnormality impact weight and the resonance risk weight are weighted and fused to obtain a target adjustment coefficient to reflect the abnormal severity and resonance risk intensity of the current heat dissipation unit;

[0114] The fusion regulation parameters are generated according to the target adjustment coefficient, specifically, the fusion regulation parameters can include the correction amount of fan speed, driving frequency, motor phase difference, PWM duty cycle, etc. running parameters, to make the running parameters of the heat dissipation unit deviate from the resonance working condition interval and tend to the abnormal response target interval corresponding to the abnormal type, so as to suppress resonance while correcting abnormal operation state;

[0115] The fusion regulation parameters are converted into fusion regulation instructions and issued to the heat dissipation unit for execution. Preferably, during the execution process, the acoustic and vibration feedback signals of the heat dissipation unit are monitored in real time to update the target adjustment coefficient and the fusion regulation parameters, forming a closed-loop adaptive regulation.

[0116] Through the above-mentioned embodiments, the present embodiment can realize joint regulation for resonance risk and operation abnormity, and give consideration to acoustic stability and operation reliability. Compared with the existing scheme of single speed reduction or frequency offset control based on resonance frequency, the present embodiment utilizes multi-source feature fusion and dynamic weight distribution mechanism, can effectively avoid over-regulation or lag problem, improve sensitivity and precision of resonance suppression, and thus realize intelligent adaptive stable operation of the server heat dissipation unit.

[0117] In one embodiment, the method further comprises:

[0118] updating the heat dissipation acoustic archive according to the running acoustic detection result of the heat dissipation unit;

[0119] based on the updated heat dissipation acoustic archive, adaptively updating the baseline voiceprint feature and the resonance frequency spectrum.

[0120] In one specific embodiment, updating the heat dissipation acoustic archive according to the running acoustic detection result of the heat dissipation unit comprises:

[0121] in response to completion of one running detection of the heat dissipation unit, analyzing and evaluating the collected acoustic feature data;

[0122] when the detection result shows that the current running state is within the normal range, and the voiceprint feature and the sound pressure feature do not appear abnormal deviation, marking the acoustic feature data of this detection as valid sample and storing it in the heat dissipation acoustic archive; further, according to the new and old weights of the data, dynamically managing the historical samples in the archive, deleting expired or low confidence samples, so as to maintain the representativeness and timeliness of the archive data;

[0123] in response to completion of the heat dissipation acoustic archive update, re-counting the stability parameters of each voiceprint feature component and sound pressure feature component, aggregating and correcting the baseline voiceprint feature; and combining the frequency spectrum feature of the new sample, re-verifying and screening the original resonance frequency, to generate the updated resonance frequency spectrum and resonance interval table.

[0124] Preferably, the archive update can be performed according to a periodic strategy or a triggered strategy: under the periodic strategy, the update is automatically performed when the cumulative sample number reaches a preset threshold or the running time exceeds a set period; under the triggered strategy, when baseline drift, noise distribution change or stability parameter decrease is detected, the archive update and baseline recalculation process are automatically started.

[0125] Specifically, in the present embodiment, the heat dissipation acoustic archive can continuously learn and optimize with the long-term operation of the device, so that the baseline voiceprint feature and the resonance frequency spectrum always reflect the latest acoustic state of the heat dissipation unit, thereby improving the accuracy of abnormal detection and the adaptive ability of the system.

[0126] In one embodiment, the heat dissipation unit comprises at least two heat dissipation elements arranged at different positions in the server, and the method further comprises:

[0127] The acoustic signals of the multiple elements are acquired by the acoustic signal acquisition units arranged at adjacent positions of the multiple heat dissipation elements respectively, and are fused to obtain a fused acoustic signal;

[0128] The fused acoustic signal is analyzed to obtain a fused dimple feature, and the fused dimple feature is compared with a fused baseline dimple feature stored in the heat dissipation acoustic archive to obtain a fused feature comparison result;

[0129] According to the fused feature comparison result, it is determined whether the operation state of the heat dissipation unit is abnormal, and if so, the element dimple features obtained from the element acoustic signals are compared to determine abnormal element dimple features and abnormal heat dissipation elements;

[0130] The fused vibration signal is obtained by the vibration signal acquisition unit arranged in the server, and the fused frequency spectrum is obtained by analysis;

[0131] The fused frequency spectrum is compared with the element cooperative resonance frequency spectrum stored in the heat dissipation acoustic archive to determine whether the multiple heat dissipation elements have a cooperative resonance risk, and if so, multiple element fused adjustment instructions are generated based on the abnormal dimple features and the fused frequency spectrum and are respectively sent to the multiple abnormal heat dissipation elements.

[0132] It is worth noting that in one specific embodiment, the determination of abnormal element dimple features and corresponding abnormal heat dissipation elements can be achieved by a sound source positioning algorithm, which includes: after detecting that the fused dimple feature deviates significantly from the fused baseline dimple feature, performing spatial inversion processing on the abnormal acoustic signal, using the time domain signal collected by the multi-point array microphone, and determining the spatial coordinates of the abnormal sound source by the time difference of arrival (TDOA) algorithm, beamforming or acoustic imaging algorithm (such as MUSIC, GCC-PHAT); the position of the sound source located is matched with the structure coordinates of each heat dissipation element to determine the target element that produces abnormal acoustic features, and the abnormal dimple features corresponding to the element are extracted as the basis for abnormal identification and subsequent fault determination.

[0133] Specifically, in the embodiment when the heat dissipation unit includes multiple heat dissipation elements, the fused texture features are generated by fusing the acoustic signals of the multiple elements, and compared with the fused baseline texture features in the archive, so that the overall heat dissipation unit operation state can be comprehensively judged; if an abnormality is detected, the abnormal element can be further determined through unit acoustic feature comparison to realize accurate positioning of the abnormal source; at the same time, the fused frequency spectrum is formed by fusing the vibration signals and compared with the cooperative resonance frequency spectrum, so that the cooperative resonance risk between multiple elements can be identified, and differentiated adjustment instructions are issued to each element, thereby the embodiment can realize global cooperative monitoring and distributed control in a multiple-element heat dissipation system, and improve the anti-resonance ability and noise control level of the system.

[0134] In one embodiment, before comparing the fused texture features with the fused baseline texture features stored in the heat dissipation acoustic archive, the method further comprises:

[0135] The acoustic signals of the multiple heat dissipation elements under standard working conditions are collected, and after time-frequency transformation, feature extraction and standardization processing, multiple element acoustic feature vectors are obtained, which include but are not limited to element voiceprint feature components, element sound pressure feature components and element frequency spectrum energy distribution, for representing the running acoustic characteristics of the elements;

[0136] According to the structural position, mounting method and sound field propagation path of the corresponding element in the heat dissipation unit or server, the spatial coordinates of the corresponding element are determined, and the element acoustic feature vector is spatially corrected to obtain an element spatial acoustic feature vector, wherein the spatial correction can correct the amplitude based on the propagation distance and medium attenuation coefficient, compensate the phase based on the spatial path difference, and apply azimuth weight to the acoustic features of sound waves incident in different directions, so as to form the element spatial acoustic feature vector after spatial correction, so as to compensate the feature deviation caused by the sound field propagation difference;

[0137] According to the element spatial acoustic feature vector, the acoustic data of the element in multiple time periods is statistically analyzed to extract stability features such as mean, variance, principal component and energy center frequency, and a stable acoustic behavior feature cluster center is identified by means of a clustering algorithm, and then the baseline texture features of the corresponding heat dissipation element are determined to represent the typical acoustic response behavior of the element under standard working conditions;

[0138] According to the spatial coordinates of the heat dissipation element and the acoustic energy contribution, signal-to-noise ratio or spatial coupling degree of the baseline texture features thereof, weights are applied to each feature and fused to obtain a first fused texture feature; wherein the fusion method can use weighted feature fusion or principal component analysis method to extract the overall acoustic principal component to represent the theoretical fused acoustic characteristics of the multiple heat dissipation elements under standard working conditions;

[0139] Based on the aforementioned collected original acoustic signals, direct signal fitting is performed using least square fitting or neural network regression model, the acoustic signals of multiple elements are time domain superimposed and sound field regression is performed to obtain fused acoustic signals, and through the same time-frequency analysis and feature extraction process, a second fused texture feature is formed to reflect the fused acoustic performance of the heat dissipation element in the actual sound field environment;

[0140] The first fused texture feature and the second fused texture feature are combined to generate a similarity index (such as cosine similarity or Euclidean distance) therebetween, and a final target fused texture feature is formed by weighted combination according to the deviation correction principle, wherein the weight parameter used in the weighted combination can be adaptively adjusted according to the signal-to-noise ratio or the similarity of the two features;

[0141] The fused baseline texture feature is stored in the heat dissipation acoustic archive.

[0142] Specifically, in the present embodiment, the acoustic feature fusion method based on spatial position correction is introduced to realize high-precision modeling and standardized description of the acoustic behavior of the heat dissipation unit. Through time-frequency transformation, feature extraction and standardization processing of the acoustic signals of each heat dissipation element, combined with spatial correction of the features by structure position, installation method and sound field propagation path, the feature deviation caused by installation difference, structure shielding and sound wave propagation delay is effectively eliminated, so that the acoustic features of different elements can be accurately compared in the same reference coordinate system. Further, the present embodiment establishes a baseline element texture feature and generates a fused baseline texture feature by using a double-path fusion mechanism, wherein the first fused texture feature is used to represent the theoretical acoustic response under standard working conditions, and the second fused texture feature is used to reflect the comprehensive acoustic performance of the actual sampling signal. The joint analysis of the two can effectively suppress the deviation that may exist in single-path feature calculation. The target fused texture feature determined by weighting or similarity correction can more truly and stably reflect the overall acoustic state of the heat dissipation unit.

[0143] In one embodiment, the fused spectrum is compared with the element cooperative resonance spectrum stored in the heat dissipation acoustic archive to determine whether the multiple heat dissipation elements have a cooperative resonance risk; in response to yes, multiple element fusion adjustment instructions are generated based on the abnormal texture feature and the fused spectrum and are respectively sent to the multiple abnormal heat dissipation elements, including:

[0144] The fused spectrum signal is subjected to frequency spectrum matching analysis to calculate the similarity between the current spectrum and the baseline cooperative resonance spectrum in the archive, and the similarity can be measured by cosine similarity, cross-correlation coefficient or spectral overlap degree;

[0145] When the similarity exceeds a set threshold, the abnormal influence weight of each element is determined in combination with the abnormal texture feature type and the abnormal score of the corresponding abnormal heat dissipation element.

[0146] Further based on the energy contribution coefficient of each element at the cooperative resonance frequency, a resonance risk weight is determined;

[0147] The abnormal influence weight and the resonance risk weight are weighted and fused to obtain a target adjustment coefficient of each element;

[0148] According to the target adjustment coefficient, a corresponding element fusion adjustment parameter is generated, and the fusion adjustment parameter is used to comprehensively adjust the operation parameters of the element, such as the driving frequency, the rotating speed, the motor phase difference, or the duty cycle, so that the high-contribution element deviates from the resonance interval and tends to the stable working area corresponding to the abnormal type;

[0149] The fusion adjustment parameter is converted into an element fusion adjustment instruction and is issued to each abnormal heat dissipation element for execution;

[0150] In a specific implementation, the closed-loop adjustment control strategy can be used to monitor the fusion frequency spectrum change after adjustment in real time. When the main frequency peak value amplitude drops below a threshold value, the adjustment instruction is stopped and the fusion adjustment strategy is updated.

[0151] In the above manner, the embodiment can consider the abnormal characteristics of each heat dissipation element and the resonance energy distribution when detecting the cooperative resonance risk, realize adaptive dynamic regulation based on multi-dimensional feature fusion, and thus more accurately suppress the cooperative resonance effect, reduce the noise superposition and vibration amplification phenomenon, and improve the stability and reliability of the overall server heat dissipation system.

[0152] Further, to accurately identify the cooperative resonance risk of multiple heat dissipation elements, a baseline cooperative resonance frequency spectrum is constructed in the archive library establishment stage. The baseline cooperative resonance frequency spectrum is used to represent the acoustic resonance characteristics that the multiple heat dissipation elements may produce under standard structure configuration and normal working conditions. The establishment method includes:

[0153] Perform an acoustic excitation test on the heat dissipation unit under standard environmental conditions. By controlling the driving frequency and operating power of the multiple heat dissipation elements, gradually scan a preset frequency range (for example, 100 Hz to 20 kHz), and synchronously collect the sound pressure response signals of the overall sound field;

[0154] Perform fast Fourier transform (FFT) on the collected signals to obtain the frequency spectrum distribution, and identify the frequency interval with significantly enhanced amplitude or energy concentrated distribution on the frequency spectrum;

[0155] Combine the structure model to calculate the acoustic energy coupling relationship of each heat dissipation element at different frequencies, construct a cooperative coupling matrix between the elements, and determine the main resonance mode and corresponding frequency distribution of the overall sound field of the server by performing eigenvalue decomposition or principal component analysis on the matrix;

[0156] Extract the frequency components with the maximum amplitude or the most concentrated energy in the plurality of resonance modes, and form a baseline collaborative resonance spectrum of the heat dissipation unit.

[0157] In another implementation, the natural frequency of the heat dissipation structure and the corresponding acoustic radiation response can be calculated based on a finite element simulation (FEA) or an acoustic numerical simulation model, the main resonance frequency point obtained by simulation is fused with the measured data to form a calibrated baseline collaborative resonance spectrum, and the spectrum is stored in a heat dissipation acoustic archive for subsequent comparison and risk detection.

[0158] Specifically, the baseline collaborative resonance spectrum established in the embodiment not only reflects the single resonance characteristics of each heat dissipation element, but also comprehensively considers the structural coupling and acoustic field superposition effect, and can accurately depict the resonance characteristics of the heat dissipation unit as a whole under standard working conditions, thereby providing a reliable reference for resonance risk judgment.

[0159] Through the description of the above embodiments, those skilled in the art can clearly understand that the method according to the above embodiments can be realized by means of software and the necessary general hardware platform, of course, it can also be realized by hardware, but in many cases the former is a better implementation.

[0160] As shown in Figure 3 The embodiment of the present application also provides a heat dissipation unit control device, which comprises a multi-modal acoustic perception layer, an intelligent processing layer and an execution layer, wherein,

[0161] The multi-modal acoustic perception layer comprises an acoustic signal acquisition module, a vibration signal acquisition module and an environment data acquisition module.

[0162] The intelligent processing layer comprises a feature analysis module, an archive establishment module and a control decision module.

[0163] The execution layer comprises a heat dissipation unit driving module and an alarm output module.

[0164] In the multi-modal acoustic perception layer, the acoustic signal acquisition module is used to acquire acoustic signals of the heat dissipation unit in the server under a plurality of operating states; the vibration signal acquisition module is used to acquire vibration signals of the heat dissipation unit under a plurality of operating states; and the environment data acquisition module is used to acquire environment data in the server.

[0165] In the intelligent processing layer, the feature analysis module is configured to determine the baseline embossed feature and the resonance spectrum of the heat dissipation unit, and to analyze the real-time acoustic signal and the real-time vibration signal of the heat dissipation unit to obtain the real-time embossed feature and the real-time spectrum; the archive establishment module is configured to store the baseline embossed feature and the resonance spectrum of the heat dissipation unit as a heat dissipation acoustic archive; and the control decision module is configured to determine whether the running state of the heat dissipation unit is abnormal according to the feature comparison result of the real-time embossed feature and the baseline embossed feature, and in response to the determination, to determine whether the heat dissipation unit has resonance risk according to the spectrum comparison result of the real-time spectrum and the resonance spectrum, and in response to the determination, to generate a fusion adjustment instruction and send the instruction to the heat dissipation unit.

[0166] In the execution layer, the heat dissipation driving module is configured to receive the abnormal response control strategy or the resonance adjustment instruction sent by the intelligent processing layer, and convert the abnormal response control strategy or the resonance adjustment instruction into driving parameters of the heat dissipation unit to control the running of the heat dissipation unit; and the alarm generation module is configured to generate corresponding alarm information according to the feature comparison result or the spectrum comparison result, and send the alarm information to the client or the central control server, etc., to alert the system or the operation and maintenance personnel.

[0167] As shown in FIG. 1, Figure 4 In one embodiment, the feature analysis workflow performed by the feature analysis module includes the following steps:

[0168] First, the collected acoustic signal and vibration signal are preprocessed to remove background noise, direct current bias and invalid frequency band signal, and obtain an effective signal sequence for feature extraction;

[0169] Then, the preprocessed acoustic signal and vibration signal are subjected to short-time Fourier transform (STFT) to obtain time-frequency domain features, and a corresponding complex matrix is generated to provide basic data for subsequent feature extraction;

[0170] Subsequently, the feature analysis module performs multi-dimensional feature extraction branches based on the complex matrix, including:

[0171] The spectrum feature extraction branch: local peak frequencies are identified through spectrum peak value detection, which are used to reflect the resonance frequency point distribution of the heat dissipation unit under different running states, and the resonance spectrum is updated based on the change trend of the peak frequencies;

[0172] The voiceprint feature extraction branch: mel-frequency cepstral coefficients (MFCC) are extracted to generate voiceprint feature components representing the running characteristics of the heat dissipation unit, and the baseline voiceprint feature is updated based on the components;

[0173] The sound pressure feature extraction branch: sound pressure level calculation is performed on the acoustic signal to generate sound pressure feature components, which are used to represent the sound power level of the heat dissipation unit, and the baseline sound pressure feature is updated based on the components;

[0174] The acoustic texture feature is formed after the fusion of the voiceprint feature and the sound pressure feature. The fusion can be realized through a multi-feature weighting algorithm, feature vector splicing, principal component analysis (PCA), or a feature mapping model based on similarity constraints, to establish a joint representation relationship between the acoustic texture and the sound pressure energy, thereby enhancing the sensitivity and distinguishability of the operating state of the heat dissipation unit.

[0175] After the feature extraction and fusion are completed, the feature analysis module updates and stores the acoustic texture feature, the resonance spectrum, and the corresponding baseline parameters, forming a heat dissipation acoustic archive, which provides a feature benchmark for subsequent real-time monitoring, anomaly detection, and resonance risk assessment.

[0176] Through the process, multi-dimensional self-learning and modeling of acoustic and vibration signals of the heat dissipation unit can be realized. Without changing the hardware structure, dynamic pre-control and suppression of the noise characteristics of the server heat dissipation element can be realized through algorithms.

[0177] As shown in Figure 5 The hardware connection method for heat dissipation unit control provided by the embodiments of the present application includes the required hardware, which includes a main processor, a microphone array, a vibration sensor, a temperature sensor, a fan driving circuit, and a server fan.

[0178] The main processor is preferably an MCU or Soc, which is connected to the microphone array composed of multiple MEMS microphones through an I2S interface, connected to the MEMS accelerometer as the vibration sensor through an ADC interface, connected to the temperature sensor through an I2C interface, and electrically connected to the fan driving circuit through a PWM control signal. The fan driving circuit is connected to the 4-wire PWM fan as the server fan through a driving power supply.

[0179] As shown in Figure 6 The embodiments of the present application also provide an electronic device, which includes a memory and a processor. The memory stores a computer program, and the processor is configured to run the computer program to execute the steps in any of the above-mentioned heat dissipation unit control method embodiments.

[0180] The skilled person can further realize that the units and algorithm steps of the examples described in conjunction with the embodiments disclosed herein can be realized in electronic hardware, computer software, or a combination of both. To clearly illustrate the interchangeability of hardware and software, the components and steps of the examples have been described in general terms in the above description. Whether the functions are performed in hardware or software depends on the specific application and design constraints of the technical solution. The skilled person can use different methods to implement the described functions for each specific application, but such implementation should not be considered beyond the scope of the present application.

[0181] The above describes in detail the heat dissipation unit control method and the electronic device provided by the present application. The principles and implementation manners of the present application are described by applying specific examples, and the above description of the embodiments is only applicable to help understand the method of the present application and the core idea thereof. It should be pointed out that, for those skilled in the art, some improvements and modifications can be made to the present application without departing from the principles of the present application, and these improvements and modifications also fall within the protection scope of the present application.

Claims

1. A method for controlling a heat dissipation unit, characterized in that, include: Acoustic and vibration signals of the heat dissipation unit in the server under various operating conditions are obtained, and the baseline embossing characteristics and resonance spectrum of the heat dissipation unit are determined to establish a heat dissipation acoustic archive. In response to the operation of the server, the real-time acoustic signal and real-time vibration signal of the heat dissipation unit are acquired and analyzed to obtain real-time embossing features and real-time spectrum. Based on the feature comparison results between the real-time embossing features and the baseline embossing features, it is determined whether the operating status of the heat dissipation unit is abnormal. In response to the abnormal operation of the heat dissipation unit, the risk of resonance of the heat dissipation unit is determined based on the spectrum comparison result between the real-time spectrum and the resonance spectrum. In response to the existence of resonance risk, a fusion adjustment command is generated and sent to the heat dissipation unit based on the feature comparison result and the spectrum comparison result; The baseline embossing features and the resonance spectrum are determined in the following manner: The multi-condition acoustic dataset of the heat dissipation unit is preprocessed and classified to remove environmental noise and distinguish acoustic samples under different operating conditions, thus obtaining polymorphic acoustic samples. Based on the polymorphic acoustic samples, the polymorphic embossing features and polymorphic spectral features of the heat dissipation unit are extracted to generate a multi-condition feature set. Based on the spectral energy distribution in the multi-condition feature set, the frequency components are statistically analyzed and peaks are detected to determine the resonant spectrum of the heat dissipation unit. Based on the multi-condition feature set, the characteristic stability of sound pressure and acoustic signature of the heat dissipation unit under different operating conditions is determined, and based on the characteristic stability, the baseline embossing feature characterizing the normal operating characteristics of the heat dissipation unit is obtained, including: parsing the multi-condition feature set to obtain the polymorphic embossing feature of the heat dissipation unit; parsing the polymorphic embossing feature to obtain multiple acoustic signature feature components and sound pressure feature components to characterize the operating characteristics of the heat dissipation unit under different operating conditions; obtaining multiple acoustic signature feature stability parameters and multiple sound pressure feature stability parameters according to the feature similarity index of multiple acoustic signature feature components and the feature similarity index of multiple sound pressure feature components; selecting several target acoustic signature feature components and target sound pressure feature components from the multiple acoustic signature feature components and the sound pressure feature components whose acoustic signature feature stability parameters and sound pressure feature stability are greater than or equal to a preset stability threshold; and performing aggregation processing on the several target acoustic signature feature components and the target sound pressure feature components to obtain the baseline embossing feature.

2. The heat dissipation unit control method according to claim 1, characterized in that, Acquire acoustic and vibration signals from the server's heat dissipation unit under various operating conditions, including: Construct an operating parameter space, which includes at least one or more of the following: heat dissipation unit operating parameters, server load, and ambient temperature; Based on parameter representativeness and acquisition efficiency, one or more test state sequences are determined in the operating parameter space so that the test state sequences characterize the multi-state operating conditions of the heat dissipation unit. Based on the test state sequence, corresponding test control parameters are generated and sent to the heat dissipation unit so that the heat dissipation unit operates based on the test control parameters. In response to the operation of the heat dissipation unit, the acoustic and vibration signals of the heat dissipation unit are collected and associated with the corresponding test control parameters for storage, so as to construct a multi-condition acoustic dataset of the heat dissipation unit.

3. The heat dissipation unit control method according to claim 1, characterized in that, The step of performing statistical analysis and peak detection on frequency components based on the spectral energy distribution in the multi-condition feature set to determine the resonant spectrum of the heat dissipation unit includes: The multi-condition feature set is analyzed to obtain the multi-mode spectral features of the heat dissipation unit; Based on the polymorphic spectral characteristics, amplitude statistics of multiple frequency components are obtained, and the amplitude statistics include at least the average amplitude and amplitude fluctuation index. Based on multiple amplitude statistics, local peak amplitudes are identified in the polymorphic spectral features, and the local peak amplitudes are compared with preset amplitude thresholds; If the local peak amplitude is greater than the preset amplitude threshold, the corresponding local peak frequency is determined to be a candidate resonance frequency. Based on the peak stability and amplitude significance across multiple operating conditions, multiple candidate resonance frequencies are screened to obtain a set of target resonance frequencies; The target resonant frequency set is mapped to the operating parameter space of the heat dissipation unit to generate the resonant spectrum and obtain a resonant interval table for adjusting the operation of the heat dissipation unit.

4. The heat dissipation unit control method according to claim 1, characterized in that, The process of acquiring and analyzing the real-time acoustic and vibration signals of the heat dissipation unit to obtain real-time embossing features and real-time spectrum includes: The real-time acoustic signal and the real-time vibration signal are obtained by a signal acquisition unit located around the heat dissipation unit. The real-time acoustic signal and the real-time vibration signal are respectively converted from time-domain signals to frequency-domain signals to obtain real-time acoustic frequency-domain signals and real-time vibration frequency-domain signals; Based on the preset sound pressure weighting coefficient, the weighted sound pressure level of the real-time acoustic frequency domain signal is determined, and real-time sound pressure characteristic components are generated. Mel-spectral coefficients of the real-time acoustic frequency domain signal are extracted to generate real-time acoustic signature feature components, which are then fused with the real-time sound pressure feature components to obtain the real-time embossing feature. The real-time vibration frequency domain signal is subjected to spectral peak detection to identify real-time local peak frequencies, and the real-time spectrum is obtained based on the real-time local peak frequencies.

5. The heat dissipation unit control method according to claim 1, characterized in that, The step of determining whether the operating status of the heat dissipation unit is abnormal based on the feature comparison result between the real-time embossing feature and the baseline embossing feature includes: The real-time embossing features and the baseline embossing features are standardized and preprocessed to obtain real-time feature vectors and baseline feature vectors; Based on the component dimensions determined by the baseline feature vector, multiple real-time dimension components are extracted from the real-time feature vector. Based on the similarity index and deviation measure between multiple real-time dimension components and the corresponding baseline dimension components, a set of comparison measures is obtained, and the set of comparison measures is weighted and fused to obtain the heat dissipation unit anomaly score. The anomaly score of the heat dissipation unit is compared with the anomaly threshold determined by the baseline embossing features; If the abnormal score of the heat dissipation unit is greater than or equal to the abnormal threshold, the heat dissipation unit is determined to be operating abnormally, and the corresponding real-time embossing feature is identified as an abnormal embossing feature. If the abnormal score of the heat dissipation unit is less than the abnormal threshold, the heat dissipation unit is determined to be operating normally.

6. The heat dissipation unit control method according to claim 1, characterized in that, Based on the spectral comparison results between the real-time spectrum and the resonance spectrum, determine whether the heat dissipation unit has a resonance risk, including: The real-time frequency is obtained by analyzing the real-time spectrum, and the current operating condition of the heat dissipation unit is determined. The current operating condition includes at least one or more of the current operating parameters of the heat dissipation unit, the current load of the server, and the current ambient temperature. The resonance spectrum is analyzed to obtain multiple resonance frequencies and corresponding resonance conditions. Combined with a preset risk scaling factor, multiple resonance frequency ranges and corresponding resonance condition ranges are generated. If the real-time frequency falls into any of the resonant frequency ranges, and / or the current operating condition falls into the corresponding resonant operating condition range, then it is determined that the heat dissipation unit has a resonance risk. In response to the existence of resonance risk, a fusion adjustment command is generated and sent to the heat dissipation unit based on the feature comparison result and the spectrum comparison result, including: Based on the anomaly type and anomaly score represented by the feature comparison results, the anomaly impact weight is determined. Based on the resonance frequency range and amplitude significance characterized by the spectrum comparison results, the resonance risk weight is determined. The target adjustment coefficient is obtained by weighting and fusing the abnormal impact weight and the resonance risk weight. Based on the target adjustment coefficient, a fusion adjustment parameter is generated so that the operating parameters of the heat dissipation unit deviate from the resonance condition range and tend toward the abnormal response target range corresponding to the abnormal type. The fusion adjustment parameters are converted into fusion adjustment commands and sent to the heat dissipation unit for execution. The execution feedback results of the heat dissipation unit are obtained to update the fusion adjustment strategy.

7. The heat dissipation unit control method according to claim 1, characterized in that, The heat dissipation unit includes at least two heat dissipation elements, and the at least two heat dissipation elements are disposed in different locations within the server. The method further includes: Acoustic signal acquisition units, which are respectively set at adjacent positions of multiple heat dissipation elements, acquire and fuse the acoustic signals of multiple elements to obtain a fused acoustic signal. The fused acoustic signal is analyzed to obtain fused embossing features, and the fused embossing features are compared with the fused baseline embossing features stored in the thermal acoustic archive to obtain the fused feature comparison results; Based on the fusion feature comparison results, it is determined whether the operating status of the heat dissipation unit is abnormal. In response, the abnormal component embossing features and abnormal heat dissipation components are determined by comparing multiple component embossing features obtained from the component acoustic signals. By using a vibration signal acquisition unit installed in the server, a fused vibration signal is obtained, and the fused spectrum is obtained through analysis. The fused spectrum is compared with the component co-resonance spectrum stored in the thermal acoustic archive to determine whether there is a risk of co-resonance among the multiple heat dissipation components. In response, based on the abnormal embossing features and the fused spectrum, multiple component fusion adjustment commands are generated and sent to the multiple abnormal heat dissipation components respectively.

8. An electronic device, characterized in that, include: Memory, used to store computer programs; A processor, configured to implement the steps of the heat dissipation unit control method as described in any one of claims 1 to 7 when executing the computer program.

Images