Network card vibration detection method and device, equipment and storage medium

Abstract

The invention provides a network card vibration detection method, device and equipment and a storage medium, and the method comprises the steps: deploying a plurality of groups of three-axis acceleration sensor arrays in different function regions of a network card, and collecting vibration data when the network card is not powered on; after the network card is powered on, the substrate management controller reads vibration data, vibration spectrum characteristics are extracted by using a time-frequency analysis method, and dangerous frequency components are identified in combination with server heat dissipation system characteristics and network card physical parameters; based on the vibration data and the danger frequency component, vibration source positioning and classification are carried out by combining electromagnetic characteristic changes before and after the network card is powered on, and a vibration threat level is determined; and selecting and executing corresponding protection measures according to the vibration source information and the real-time communication state of the network card. According to the method, the problems of a spectrum analysis blind area and vibration source recognition fuzziness are solved, and network card data and communication stability are effectively protected through targeted protection measures.

Description

Technical Field

[0001] The present invention relates to the field of data processing technology, and in particular to a network card vibration detection method, device, equipment and storage medium. Background Art

[0002] With the continuous advancement of data center and server technology, the reliability of network interface cards (NICs), critical network interface devices in servers, directly impacts overall system stability and data transmission security. Current data center environments are complex and volatile. Factors such as server rack handling, high-speed cooling fan operation, and vibration from adjacent equipment often cause NICs to experience complex mechanical vibrations. Existing NIC vibration detection methods primarily rely on simple acceleration threshold monitoring when the device is powered on. These methods are unable to capture vibration conditions during periods when the device is not powered on. This makes it difficult for maintenance personnel to fully understand the NIC's vibration history, impacting the accuracy of fault diagnosis.

[0003] Furthermore, existing technologies have significant shortcomings in vibration source identification, making it difficult to effectively distinguish between different types of vibration sources. Especially when multiple vibration sources coexist, it's impossible to determine whether the vibration is caused by a loose network card itself or by external vibration transmission. Furthermore, traditional methods lack in-depth analysis of vibration spectrum characteristics, easily overlooking specific frequency components that are small in amplitude but potentially harmful to the network card structure, creating detection blind spots. These technical issues result in a lack of targeted vibration protection measures for network cards, impacting their stable operation in complex environments. Summary of the Invention

[0004] The main purpose of the present invention is to solve the technical problems of spectrum analysis blind spots and vibration source identification ambiguity in existing network card vibration detection methods; A first aspect of the present invention provides a network card vibration detection method, the network card vibration detection method comprising: When the network card is not powered on, multiple sets of three-axis acceleration sensor arrays deployed in different functional areas of the network card are used to collect vibration data from each functional area of the network card. After the network card is powered on, the vibration data is read by the baseboard management controller, and the vibration spectrum characteristics of the vibration data are extracted using a time-frequency analysis method. The dangerous frequency components are identified by combining the characteristics of the server cooling system and the physical parameters of the network card; Based on the vibration data and the dangerous frequency components, combined with the changes in electromagnetic characteristics before and after the network card is powered on, the vibration source is located and classified, and the vibration threat level is determined to obtain vibration source information; According to the vibration source information and combined with the real-time communication status monitoring of the network card, corresponding protection measures are selected and executed to protect the network card data and communication stability.

[0005] Optionally, in a first implementation of the first aspect of the present invention, after the network card is powered on, reading the vibration data through the baseboard management controller, extracting vibration spectrum characteristics of the vibration data using a time-frequency analysis method, and combining the characteristics of the server cooling system and the physical parameters of the network card to identify dangerous frequency components includes: After the network card is powered on, the vibration data is read by the baseboard management controller, and the vibration data is decomposed into time and frequency domains using a multi-resolution wavelet transform to obtain the time domain distribution of different frequency components; Obtain historical cooling fan speed data from the server management module, calculate the fan blade pass frequency and harmonic components, and construct a cooling noise spectrum template; Applying an adaptive spectrum subtraction algorithm to the time domain distribution of the different frequency components according to the heat dissipation noise spectrum template to remove the vibration components caused by the heat dissipation system and obtain a vibration spectrum after removal; The natural frequency and dangerous resonance mode of the network card are calculated based on the thickness, elastic modulus, size, and fixing point positions of the network card PCB, forming a spectrum diagram of the network card vulnerability. By calculating the degree of overlap between the vibration spectrum after stripping and the network card vulnerability spectrum diagram, the dangerous frequency component that poses the greatest threat to the structural integrity of the network card is determined.

[0006] Optionally, in a second implementation of the first aspect of the present invention, applying an adaptive spectral subtraction algorithm to the time domain distribution of the different frequency components according to the heat dissipation noise spectrum template to strip off the vibration components caused by the heat dissipation system to obtain the stripped vibration spectrum includes: The time domain distribution of the different frequency components is converted to the frequency domain using fast Fourier transform, and the original vibration spectrum matrix corresponding to frequency and energy is constructed, wherein the matrix elements represent the energy distribution of each frequency point in different time windows; Based on the fan blade passing frequency and harmonic components in the heat dissipation noise spectrum template, the correlation with the original vibration spectrum is calculated to determine the pollution degree of the heat dissipation noise at each frequency point, and the adaptive gain factor is calculated segmentally according to the pollution degree; Applying the adaptive gain factor to each frequency point of the original vibration spectrum matrix, performing a point-by-point spectrum subtraction operation, and obtaining processed spectrum matrix data; The inverse fast Fourier transform is applied to the processed spectrum matrix data to reconstruct the signals of each frequency band into the time domain. The instantaneous frequency characteristics of the signal are extracted through Hilbert transform. Combined with the amplitude information, the vibration spectrum after stripping off the vibration components caused by the heat dissipation system is obtained.

[0007] Optionally, in a third implementation of the first aspect of the present invention, the vibration source is located and classified based on the vibration data and the dangerous frequency component, and the vibration threat level is determined in combination with changes in electromagnetic characteristics before and after the network card is powered on. The vibration source information obtained includes: Analyze the changes in the electromagnetic characteristics of the high-speed PHY chip and FPGA before and after the network card is powered on. By comparing the differences in the frequency response functions before and after power-on, identify electromagnetically sensitive areas. By using the time difference of the vibration data captured in each functional area and applying a modified triangulation algorithm, the coordinates of the vibration source are located on the three-dimensional spatial model of the network card; Classifying the vibration source according to the correlation between the vibration source coordinates, the dangerous frequency components, and the electromagnetic sensitive area to obtain a vibration source classification result, wherein the vibration source classification result includes one or more of an internal structure loose type, an external mechanical conduction type, a heat dissipation system induction type, an electromagnetic interference coupling type, and a combined type; The vibration threat level is calculated based on the overlap between the dangerous frequency component and the network card vulnerability spectrum, vibration data, and the distance between the vibration source and key components, and the vibration source classification result and the vibration threat level are combined to form vibration source information.

[0008] Optionally, in a fourth implementation of the first aspect of the present invention, analyzing changes in electromagnetic characteristics of the high-speed PHY chip and the FPGA before and after the network card is powered on, and identifying electromagnetically sensitive areas by comparing differences in frequency response functions before and after power-on includes: Calculate the frequency response function matrix around the high-speed PHY chip and the FPGA in the unpowered state of the network card based on the vibration data; During the network card power-up process, the post-power-on vibration data of the high-speed PHY chip and FPGA were collected in idle mode, standard load mode, and high load mode. Calculating a frequency response function for the vibration data after power-on, performing a difference operation on the frequency response function matrix in the non-power-on state, and obtaining a frequency response difference matrix caused by the change in electromagnetic state; The main change mode is extracted from the frequency response difference matrix, and an electromagnetic sensitivity distribution map is constructed on the three-dimensional model of the network card to mark the electromagnetic sensitive areas.

[0009] Optionally, in a fifth implementation of the first aspect of the present invention, classifying the vibration source according to the correlation between the vibration source coordinates, the dangerous frequency component, and the electromagnetic sensitive area to obtain a vibration source classification result includes: Calculating the distance relationship between the coordinates of the vibration source and the position of the internal structural components of the network card to determine whether the vibration source is located around the key components inside the network card; Analyze the spectral correlation between the dangerous frequency component and the server fan speed, and the positional relationship between the vibration source coordinates and the PCIe gold finger area, to determine whether the vibration is caused by external conduction or the heat dissipation system; Calculating the spatial overlap between the vibration source coordinates and the electromagnetic sensitive area, and combining the vibration characteristics changes before and after power-on to determine whether the vibration is related to electromagnetic interference coupling; Based on the confidence scores of the above judgment results, a multi-feature fusion decision algorithm is applied to classify the vibration sources into internal structure loose type, external mechanical conduction type, heat dissipation system induced type, electromagnetic interference coupling type or composite type, and calculate the contribution weight of each type.

[0010] Optionally, in a sixth implementation of the first aspect of the present invention, selecting and executing corresponding protective measures based on the vibration source information in combination with real-time communication status monitoring of the network card to protect network card data and communication stability includes: Monitor the communication traffic, bit error rate, retransmission rate, and link quality indicators of the network card, perform correlation analysis with the vibration source information, and construct a vibration-performance impact map; Selecting corresponding levels of protective measures based on the vibration threat level in the vibration-performance impact map and the vibration source information; According to the execution results of the protective measures and the changes in the communication status of the network card, the vibration source information is updated to complete the dynamic adjustment of the protective measures.

[0011] A second aspect of the present invention provides a network card vibration detection device, the network card vibration detection device comprising: The acquisition module is used to collect vibration data from various functional areas of the network card when the network card is not powered on, using multiple sets of three-axis acceleration sensor arrays deployed in different functional areas of the network card; An analysis module is configured to read the vibration data through a baseboard management controller after the network card is powered on, extract vibration spectrum characteristics of the vibration data using a time-frequency analysis method, and identify dangerous frequency components based on characteristics of the server cooling system and physical parameters of the network card; a positioning module for locating and classifying the vibration source based on the vibration data and the dangerous frequency components, in combination with changes in electromagnetic characteristics before and after the network card is powered on, and determining the vibration threat level to obtain vibration source information; The protection module is used to select and execute corresponding protection measures based on the vibration source information and combined with the real-time communication status monitoring of the network card to protect the network card data and communication stability.

[0012] A third aspect of the present invention provides a network card vibration detection device, comprising: a memory and at least one processor, wherein the memory stores instructions, and the memory and the at least one processor are interconnected via a line; the at least one processor calls the instructions in the memory so that the network card vibration detection device executes the steps of the above-mentioned network card vibration detection method.

[0013] A fourth aspect of the present invention provides a computer-readable storage medium, wherein the computer-readable storage medium stores instructions, which, when executed on a computer, enable the computer to execute the steps of the above-mentioned network card vibration detection method.

[0014] The above-mentioned network card vibration detection method, device, equipment, and storage medium deploy multiple sets of three-axis acceleration sensor arrays in different functional areas of the network card to collect vibration data when the network card is not powered on. After the network card is powered on, the baseboard management controller reads the vibration data, uses time-frequency analysis methods to extract vibration spectrum characteristics, and identifies dangerous frequency components based on the characteristics of the server cooling system and the physical parameters of the network card. Based on the vibration data and dangerous frequency components, combined with the changes in electromagnetic characteristics before and after the network card is powered on, the vibration source is located and classified to determine the vibration threat level. Based on the vibration source information and the real-time communication status of the network card, corresponding protective measures are selected and implemented. This invention solves the problems of spectrum analysis blind spots and vibration source identification ambiguity, effectively protecting network card data and communication stability through targeted protective measures.

[0015] Other features and advantages of the present invention will be described in the following description, and in part will become apparent from the description, or understood by practicing the present invention. The purposes and other advantages of the present invention are realized and obtained by the structures particularly pointed out in the description, claims and drawings.

[0016] In order to make the above-mentioned objects, features and advantages of the present invention more obvious and easy to understand, preferred embodiments are given below and described in detail with reference to the accompanying drawings. BRIEF DESCRIPTION OF THE DRAWINGS

[0017] Figure 1 Schematic diagram of a first embodiment of a network card vibration detection method according to an embodiment of the present invention; Figure 2 A schematic diagram of an embodiment of a network card vibration detection device according to an embodiment of the present invention; Figure 3 FIG. 1 is a schematic diagram of an embodiment of a network card vibration detection device according to an embodiment of the present invention. DETAILED DESCRIPTION

[0018] To make the objectives, technical solutions, and advantages of the embodiments of the present invention more clear, the technical solutions of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the embodiments described are only part of the embodiments of the present invention, not all of them. All other embodiments obtained by ordinary technicians in this field based on the embodiments of the present invention without making any creative efforts shall fall within the scope of protection of the present invention.

[0019] The terms "including," "having," and any variations thereof, as used in the embodiments of the present invention are intended to cover non-exclusive inclusions. For example, a process, method, system, product, or device comprising a series of steps or units is not limited to the listed steps or units, but may optionally include other steps or units not listed, or may optionally include other steps or units inherent to the process, method, product, or device.

[0020] To facilitate understanding of this embodiment, a network card vibration detection method disclosed in an embodiment of the present invention is first described in detail. Figure 1 As shown, this method includes the following steps: 101. When the network card is not powered on, multiple sets of three-axis acceleration sensor arrays are deployed in different functional areas of the network card to collect vibration data of each functional area of the network card; In one embodiment of the present invention, for a typical high-performance PCIe network card, a triaxial microelectromechanical (MEMS) accelerometer array is deployed in multiple key functional areas of its mainboard and daughterboard. Specifically, sensor nodes are arranged in the PCIe gold finger area of the network card, around the high-speed PHY chip, in the FPGA area, around the RJ45 / SFP+ interface, and at the connection between the mainboard and daughterboard, forming a full-coverage monitoring network. Each monitoring point is equipped with a pair of sensors with complementary frequency response characteristics, one optimized for low-frequency response (0-500Hz) and the other optimized for high-frequency response (500-5000Hz), to address the frequency response limitations of a single sensor. These sensors adopt an ultra-low power design and are powered by the network card backup power supply or an independent battery to ensure continuous operation when the server is completely powered off. When the network card is not powered on, the sensor operates with an adaptive sampling strategy: when the vibration amplitude exceeds a preset threshold (such as 0.05g), the sampling rate is automatically increased to 10kHz to capture the entire vibration event; during periods of vibration calm, the sampling rate is reduced to 100Hz to extend the monitoring time. Each sensor node collects three-axis (X, Y, Z) vibration acceleration data and records metadata such as the acquisition timestamp, sampling rate, and sensitivity setting. The collected raw vibration data is initially processed by the onboard low-power DSP, including signal conditioning, digital filtering, and data compression, reducing storage space requirements by approximately 75%. In addition, the sensor array also synchronously collects ambient temperature, humidity, and air pressure data as an auxiliary reference for vibration data. All collected data is transmitted to the independent flash memory on the network card via a low-power bus, forming a complete vibration history record. The collected vibration data includes the vibration characteristics of the network card in various non-powered scenarios such as server transportation, installation, and storage, and records key information such as vibration amplitude, duration, main frequency components, and propagation characteristics.

[0021] 102. After the network card is powered on, the vibration data is read through the baseboard management controller, and the vibration spectrum characteristics of the vibration data are extracted using the time-frequency analysis method. The dangerous frequency components are identified by combining the characteristics of the server cooling system and the physical parameters of the network card; In one embodiment of the present invention, after the network card is powered on, the vibration data is read by the baseboard management controller, the vibration spectrum characteristics of the vibration data are extracted using a time-frequency analysis method, and the dangerous frequency components are identified in combination with the characteristics of the server cooling system and the physical parameters of the network card. The method includes: after the network card is powered on, the vibration data is read by the baseboard management controller, and the vibration data is decomposed in the time-frequency domain using a multi-resolution wavelet transform to obtain the time domain distribution of different frequency components; historical speed data of the cooling fan is obtained from the server management module, the fan blade pass frequency and harmonic components are calculated, and a cooling noise spectrum template is constructed; an adaptive spectral subtraction algorithm is applied to the time domain distribution of the different frequency components based on the cooling noise spectrum template to remove the vibration components caused by the cooling system and obtain a stripped vibration spectrum; the natural frequency and dangerous resonance mode of the network card are calculated based on the thickness, material elastic modulus, size, and fixed point position of the network card PCB to form a network card vulnerability spectrum diagram; and the dangerous frequency component that poses the greatest threat to the structural integrity of the network card is determined by calculating the overlap between the stripped vibration spectrum and the network card vulnerability spectrum diagram.

[0022] Specifically, after the network card is powered on, the baseboard management controller (BMC) first establishes a communication connection with the independent flash memory on the network card and reads the vibration data stored in the flash memory via an I2C or SPI bus interface. The reading process is divided into two stages: first, the vibration data metadata is read, including the data acquisition time point, sampling rate, compression method, and total data length; then, based on this metadata, the vibration data is read and decompressed block by block to reconstruct a complete time-domain vibration dataset. For a typical network card, this dataset includes three-axis vibration time series records from multiple monitoring points, such as the PCIe gold finger area, the area around the high-speed PHY chip, and the FPGA area. After reading, the BMC loads the vibration data into its internal RAM for subsequent processing. At this point, the BMC uses a multi-resolution wavelet transform (MRWT) to decompose the vibration data in the time-frequency domain. In the specific implementation, the Daubechies-8 (db8) wavelet basis function is selected, and a six-level wavelet decomposition is performed on the vibration data at each monitoring point, generating a set of detailed coefficients (D1-D6) and an approximate coefficient (A6). These coefficients correspond to different frequency bands: D1 for 2500-5000Hz, D2 for 1250-2500Hz, D3 for 625-1250Hz, D4 for 312.5-625Hz, D5 for 156.25-312.5Hz, D6 for 78.125-156.25Hz, and A6 for 0-78.125Hz. By calculating the energy distribution and time-varying characteristics of each coefficient, a time-frequency diagram of the vibration signal is constructed. This diagram reflects the energy distribution of different frequency components along the time axis, enabling the BMC to identify transient and persistent vibration events. This time-frequency analysis method is more suitable for analyzing non-stationary vibration signals in network card environments than traditional fast Fourier transforms. It can accurately capture the time location and frequency characteristics of vibration, and obtain the different frequency components of vibration and their time domain distribution.

[0023] After obtaining the time-frequency characteristics of the vibration, the system needs to distinguish between background vibration caused by the cooling system and other vibration sources. To this end, the BMC obtains historical operating data of the cooling fans from the server management module via the IPMI protocol, including fan model, quantity, number of blades, speed records, and control strategy. For example, a typical 2U server is typically equipped with 6-8 fans, each with 7-9 blades and a speed range of 800-12000 RPM. Based on this information, the BMC calculates the fan blade pass frequency (fundamental frequency = speed × number of blades / 60) and its harmonic components (double, triple, etc.). For a 6000 RPM, 7-blade fan, its fundamental frequency is 700Hz, with major harmonic components at 1400Hz and 2100Hz. To account for fan speed fluctuations, the BMC creates a ±5% bandwidth of interest for each frequency, while also accounting for interactions and resonance effects between multiple fans. By summarizing the frequency contributions of all fans at different speeds, the system constructs a comprehensive "heat dissipation noise spectrum template." This template describes the energy distribution characteristics and time-varying patterns of the cooling system in the frequency domain, providing a reference for subsequent noise extraction. This spectrum template construction method, based on actual cooling system parameters, resolves the unclear characteristics of heat dissipation noise in traditional vibration analysis and improves the accuracy of identifying vibrations caused by the cooling system.

[0024] After obtaining the heat dissipation noise spectrum template, the system applies an adaptive spectral subtraction algorithm to the vibration signal to remove vibration components caused by the heat dissipation system. This algorithm first converts the time-domain distribution of the different frequency components obtained by MRWT decomposition into the frequency domain to construct a time-frequency energy matrix. Then, for each characteristic frequency in the heat dissipation noise spectrum template and its surrounding bandwidth, the average energy and standard deviation of the corresponding frequency point in the time-frequency energy matrix are calculated. Based on these statistical characteristics, the system constructs an adaptive gain factor matrix: for frequencies that highly overlap with the heat dissipation noise spectrum, the gain factor is set to 0.1-0.3; for frequencies with no significant correlation with the heat dissipation frequency, the gain factor is set to 0.8-1.0; and for frequencies in the transition region, the gain factor is calculated inversely proportional to the heat dissipation noise energy contribution. In frequency bands with significant time-varying characteristics, the system uses adaptive Wiener filtering for enhancement. This technique dynamically adjusts the filter strength based on the local signal-to-noise ratio, preserving the transient characteristics of non-heat dissipation vibrations. The adaptive gain factor matrix is applied to the original time-frequency energy matrix, performing point-by-point spectral subtraction, effectively suppressing the vibration components caused by the heat dissipation system. Finally, the processed spectrum data is reconstructed into a time-domain signal through an inverse transform. The Hilbert transform is then applied to extract the signal's instantaneous frequency characteristics. Combined with the amplitude information, the complete vibration spectrum is obtained, stripped of the cooling system's vibrations. This adaptive spectral subtraction method overcomes the limitations of traditional fixed-threshold filtering, which is unable to handle complex cooling noise. This method enables the system to extract the vibration signatures that pose a true threat to the network card from background noise.

[0025] After removing the vibration components caused by the cooling system, the system needs to identify vibration frequencies that pose a potential threat to the NIC structure. To this end, it calculates its natural frequencies and dangerous resonant modes based on the NIC's physical parameters. The system first accesses the NIC's physical parameter database to obtain key parameters: a typical PCIe NIC uses an FR-4 epoxy resin substrate with a thickness of 1.6mm or 2.0mm, a copper cladding thickness of 1-2oz (35-70μm), board dimensions of approximately 200mm × 120mm, an elastic modulus of approximately 24GPa, and a Poisson's ratio of 0.18. The locations of the fixing points include the coordinates of the PCIe gold finger connections, the baffle screw fixing points, and the inter-board connections. Based on these parameters, the system calculates the natural frequency of the NIC PCB using the Rayleigh-Ritz method, while also considering the influence of the mass distribution of key components on the board (such as the FPGA, large transformer, and heat sink). For complex structures, the system pre-establishes a parametric model using finite element analysis software, generating a frequency response lookup table for different parameter combinations. Real-time calculations require only interpolation to obtain the results, improving computational efficiency. Through these calculations, the system identified several critical frequency ranges for network cards. Generally speaking, the first-order bending mode frequency of PCIe network cards is between 80-120 Hz, the torsional mode is between 350-400 Hz, and the higher-order modes are between 800-900 Hz. The system identifies these frequencies and their ±15% bandwidth as highly sensitive areas, creating a NIC vulnerability spectrum. This spectrum reflects the sensitivity of the NIC structure at different frequencies, indicating which vibration frequencies are most likely to cause structural resonance and fatigue damage.

[0026] Finally, the system calculates the overlap between the stripped vibration spectrum and the NIC vulnerability spectrum to identify the most hazardous frequency components posing the greatest threat to the NIC's structural integrity. This calculation involves three dimensions of matching analysis: frequency matching, energy matching, and duration matching. Frequency matching calculates the degree of overlap between the energy peak in the stripped vibration spectrum and the sensitive frequency band in the vulnerability spectrum; energy matching assesses the ratio of the vibration energy within the sensitive frequency band to the vulnerability threshold; and duration matching considers the cumulative duration of the vibration within the sensitive frequency band. The system calculates a comprehensive matching score for these three dimensions for each monitoring point and takes a weighted average based on the structural importance of the monitoring point to generate a global risk score. Based on this score, the system identifies and ranks the frequency components that pose the greatest threat to the NIC, including frequency value, amplitude, duration, and spatial distribution characteristics. For example, the system may identify 92Hz (the first bending mode of the motherboard), 376Hz (the resonance frequency of the PCIe interface), and 843Hz (the local resonance frequency of the FPGA area) as the three most hazardous frequency components. These identified hazardous frequency components directly reflect the vibration risk points faced by network cards in actual use environments, providing a precise basis for locating vibration sources and formulating protective measures. This refined frequency risk analysis overcomes the blind spots in spectrum analysis used in traditional vibration monitoring, enabling the identification of specific frequencies with low amplitude but potentially severe consequences.

[0027] Furthermore, the adaptive spectrum subtraction algorithm is applied to the time domain distribution of the different frequency components according to the heat dissipation noise spectrum template to remove the vibration components caused by the heat dissipation system to obtain the vibration spectrum after removal, including: converting the time domain distribution of the different frequency components into the frequency domain using fast Fourier transform, constructing the original vibration spectrum matrix corresponding to frequency-energy, wherein the matrix elements represent the energy distribution of each frequency point in different time windows; calculating the correlation with the original vibration spectrum according to the fan blade passing frequency and harmonic components in the heat dissipation noise spectrum template, determining the pollution degree of the heat dissipation noise at each frequency point, and calculating the adaptive gain factor in segments according to the pollution degree; applying the adaptive gain factor to each frequency point of the original vibration spectrum matrix, performing point-by-point spectrum subtraction operation, and obtaining processed spectrum matrix data; applying inverse fast Fourier transform to the processed spectrum matrix data, reconstructing the signals of each frequency band into the time domain, and extracting the instantaneous frequency characteristics of the signal through Hilbert transform, and combining the amplitude information to obtain the vibration spectrum after removing the vibration components caused by the heat dissipation system.

[0028] Specifically, the system converts the time-domain distribution of different frequency components obtained through multi-resolution wavelet transforms into frequency-domain processing. A fast Fourier transform is applied to the multi-level wavelet decomposition results of each sensor node, using a piecewise windowing approach. A Hamming window is selected as the window function with a window length of 1024 points and a 50% overlap between adjacent windows. Longer window lengths are used for low-frequency signals to improve frequency resolution. After the transformation, the system constructs a three-dimensional frequency-energy-time matrix, or the original vibration spectrum matrix. This matrix's dimensions include frequency, time, and the number of sensor nodes. Each matrix element represents the vibration energy value at a specific time window, at a specific sensor node, and at a specific frequency point. For example, an element represents the energy value at a frequency of 100 Hz in the 30th time window at sensor position 2. This approach integrates vibration information, previously scattered across the time and frequency domains, into a unified data structure, facilitating subsequent noise analysis and processing of specific frequency components. This original vibration spectrum matrix not only reflects the frequency composition of the vibration but also records the temporal evolution of these frequency components, providing a complete data foundation for subsequent spectral subtraction.

[0029] Specifically, after constructing the original vibration spectrum matrix, the system calculates the correlation between the original vibration spectrum and the heat dissipation noise characteristics based on the heat dissipation noise spectrum template. This calculation is divided into two levels: local frequency band correlation and global spectrum pattern correlation. Local frequency band correlation analysis is performed for each characteristic frequency in the heat dissipation noise template. Taking the fan fundamental frequency as an example, the system extracts data within a certain bandwidth near the fundamental frequency from the original vibration spectrum matrix and calculates the energy variation curve of this frequency band over time. It also obtains the fan speed change record during this period. By calculating the correlation coefficient between the two time series, it quantifies the correlation between the vibration in this frequency band and fan operation. Global spectrum pattern correlation examines whether the harmonic structure in the original vibration spectrum matches the characteristic harmonic pattern of heat dissipation noise. It calculates the ratio of each harmonic energy to the fundamental frequency energy and compares it with the typical harmonic ratio pattern of the heat dissipation system. Based on these two analyses, the system calculates a "heat dissipation noise contamination level" value for each frequency point in the spectrum matrix, ranging from 0 to 1. Then, adaptive gain factors are calculated segment by segment based on the pollution level: high-pollution frequencies are assigned low gain factors for strong suppression; medium-pollution frequencies are assigned medium gain factors; and low-pollution frequencies are assigned high gain factors to largely preserve information. This method achieves refined noise control for different frequencies.

[0030] Specifically, after calculating the adaptive gain factor, the system applies it to the original vibration spectrum matrix, performing a point-by-point spectrum subtraction operation. For each element in the spectrum matrix, the system multiplies it by the corresponding gain factor to obtain the processed spectrum value. This processing method achieves selective suppression of thermal noise in the frequency domain while preserving the spectral characteristics of vibrations from non-heat-dissipating sources. To avoid the artifacts introduced by simple multiplication, the system employs an improved minimum mean square error spectrum enhancement algorithm. This algorithm considers the relationship between adjacent frequency points and time windows and adjusts the application of the gain factor by optimizing an objective function. For frequencies with significant time-varying characteristics, the system introduces an adaptive Wiener filter, dynamically adjusting the filter parameters based on the local signal-to-noise ratio to more effectively preserve transient vibration characteristics. The system also utilizes a Kalman filter for time-domain smoothing to address the temporal correlation of thermal noise, reducing temporal discontinuities that may be introduced by spectrum subtraction. By combining these algorithms, the system suppresses thermal noise while maximally preserving the spectral characteristics of other vibration sources, resulting in processed spectrum matrix data that provides clearer spectral information for subsequent analysis.

[0031] Specifically, after obtaining the processed spectrum matrix data, the system applies an inverse fast Fourier transform (IFFT) to reconstruct the frequency domain data into a time domain signal. Similar to the forward transform, the inverse transform also employs a segmented approach, using the same window function and overlap ratio. To avoid edge effects caused by segmented processing, the system uses a weighted overlap-add method to synthesize the complete time domain signal, ensuring smooth transitions at the window boundaries. The system performs reconstruction for each frequency band, obtaining the time domain signal after stripping away the heat dissipation noise. The system then applies the Hilbert transform to the reconstructed signals in each frequency band to extract the instantaneous frequency characteristics of the signal. The Hilbert transform creates an analytic signal for the real signal. By calculating the phase change of the analytic signal, the system obtains the temporal variation of the instantaneous frequency of the signal. This process is particularly effective in identifying frequency modulation and transient frequency variations, revealing vibration characteristics that are difficult to detect using simple spectrum analysis. The system also combines the amplitude information of the analytic signal to construct a complete time-frequency energy distribution map, forming the vibration spectrum after stripping away the influence of the heat dissipation system. This spectrum clearly demonstrates the frequency composition, energy distribution, and temporal evolution of vibrations from non-heat dissipation sources, highlighting potentially hazardous vibration modes.

[0032] 103. Based on the vibration data and dangerous frequency components, combined with the changes in electromagnetic characteristics before and after the network card is powered on, the vibration source is located and classified, and the vibration threat level is determined to obtain vibration source information; In one embodiment of the present invention, the vibration source is located and classified based on the vibration data and the dangerous frequency component, and the vibration threat level is determined based on the changes in electromagnetic characteristics before and after the network card is powered on. Obtaining vibration source information includes: analyzing the changes in electromagnetic characteristics of the high-speed PHY chip and the FPGA before and after the network card is powered on, and identifying electromagnetic sensitive areas by comparing the differences in frequency response functions before and after power-on; utilizing the time differences in vibration data captured by each functional area and applying a modified triangulation algorithm to locate the vibration source coordinates on the three-dimensional spatial model of the network card; classifying the vibration source based on the correlation between the vibration source coordinates, the dangerous frequency component, and the electromagnetic sensitive area to obtain a vibration source classification result, wherein the vibration source classification result includes one or more of an internal structure loose type, an external mechanical conduction type, a heat dissipation system induced type, an electromagnetic interference coupling type, and a composite type; calculating the vibration threat level based on the overlap between the dangerous frequency component and the network card vulnerability spectrum, the vibration data, and the distance between the vibration source and key components, and combining the vibration source classification result and the vibration threat level to form vibration source information.

[0033] Specifically, the system first extracts vibration data collected from sensors around the high-speed PHY chip and FPGA when the network card is powered off from flash memory and uses this data as baseline data. This baseline data reflects purely mechanical vibration characteristics and is unaffected by electromagnetic fields. Then, after the network card is powered on, the system controls the card to enter three different operating modes: idle mode (minimal circuit activity), standard load mode (50% network throughput), and high load mode (90% network throughput). Vibration data from sensors around the high-speed PHY chip and FPGA are collected in each mode. For each data set, a frequency response function (FRF) is calculated, which describes the system's response to different frequency inputs. To perform the calculation, the system applies a fast Fourier transform to the time-domain data from each sensor node to obtain its spectral characteristics. The complete FRF is then constructed through autocorrelation and cross-correlation analysis. Next, the system compares the baseline FRF before power-on with the FRFs in each operating mode to calculate a difference matrix. Frequency bands and locations with significant differences represent areas where electromagnetic field activity affects mechanical vibration. By performing principal component analysis and clustering on this difference data, the system identifies areas of the network card's PCB with the strongest electromagnetic-mechanical coupling effects, known as electromagnetically sensitive areas, on the 3D model. These areas are typically located around high-speed PHY chips, FPGA power inputs, and high-speed data lines, providing important references for identifying electromagnetic interference-related vibrations. This process overcomes the limitation of traditional vibration analysis, which ignores electromagnetic influences, and provides a foundation for accurately distinguishing electromagnetically induced vibrations from purely mechanical vibrations.

[0034] Specifically, after identifying electromagnetically sensitive areas, the system uses the time differences in vibration data captured by each functional area to locate the coordinates of the vibration source. The system first extracts the temporal characteristics of the vibration signals at each monitoring point, focusing specifically on the propagation characteristics of transient vibration events. For each significant vibration event, the system accurately records its arrival time at different sensor nodes and constructs a vibration time difference of arrival (TDOA) matrix. Because the propagation velocity of vibration waves in a PCB is dependent on material properties and propagation direction, the system employs a modified triangulation algorithm to locate the vibration source. This algorithm first establishes a vibration wave propagation velocity model within the network card based on the elastic parameters (e.g., the elastic modulus of FR-4 substrate is 24 GPa) and geometric dimensions of the PCB, accounting for propagation velocity variations in different directions within the board. Then, based on the vibration time difference of arrival and propagation velocity model, the system solves for the vibration source coordinates using a least-squares optimization method. To improve positioning accuracy, the algorithm introduces weighting coefficients to adjust the contribution of each measurement based on signal strength and sensor reliability. Furthermore, the system considers the effects of reflected waves and multipath propagation, using time-frequency analysis to distinguish between direct and reflected waves to prevent reflections from interfering with the positioning results. After multiple iterations of optimization, the system ultimately determined the precise coordinates of the vibration source on the network card's 3D spatial model, achieving a positioning accuracy of ±5mm. These coordinates are directly annotated on the 3D model, visually displaying the vibration source's location and providing a spatial basis for subsequent vibration source classification. This precise positioning method based on time difference eliminates the problem of ambiguity in vibration source location in traditional vibration analysis.

[0035] Specifically, after determining the coordinates of the vibration source, the system classifies the vibration source by combining the dangerous frequency components and the electromagnetic sensitive area information. The classification process uses a decision tree algorithm with multi-feature fusion, comprehensively considering four key features: vibration source location characteristics, frequency characteristics, propagation characteristics, and electromagnetic correlation. The vibration source location characteristics are evaluated by calculating the distance matrix between the vibration source coordinates and the key components of the network card (such as capacitors, connectors, chips, etc.); the frequency characteristics are determined by comparing the correlation between the vibration main frequency and characteristic frequencies such as the server fan speed and the natural frequency of the network card; the propagation characteristics are determined by analyzing whether the vibration propagates inward from the PCIe gold finger or from the inside to the outside; the electromagnetic correlation is determined by calculating the degree of spatial overlap between the vibration source and the electromagnetic sensitive area and the amplitude of the change in vibration characteristics before and after power-on. Based on these four sets of feature vectors, the system performs classification decisions: when the vibration source is located near the internal components of the network card and the frequency characteristics match the natural frequency of the components, it is judged as a loose internal structure type; when the vibration source is close to the PCIe gold finger and the propagation characteristics show that the vibration propagates from the outside to the inside, it is judged as an external mechanical conduction type; when the vibration frequency is highly correlated with the fan speed, it is judged as a cooling system induced type; when the vibration source highly overlaps with the electromagnetic sensitive area and the vibration characteristics change significantly before and after power-on, it is judged as an electromagnetic interference coupling type; when multiple features simultaneously meet different types of judgment conditions, the system calculates the confidence score of each type. If the confidence of multiple types exceeds the threshold, it is judged as a composite type and the contribution weight of each type is recorded. This multi-dimensional feature analysis vibration source classification method solves the problem of ambiguous vibration source type identification in traditional vibration monitoring.

[0036] Specifically, the system calculates the vibration threat level based on multiple parameters and generates complete vibration source information. The threat level calculation first assesses the overlap between the dangerous frequency components and the NIC vulnerability spectrum. This is done by calculating the energy contribution of the vibration signal within each sensitive frequency band in the vulnerability spectrum and summing them weighted by sensitivity. The sensitive frequency bands in the vulnerability spectrum are calculated based on the aforementioned physical parameters of the NIC and typically include the first-order bending mode (80-120Hz) and the torsional mode (350-400Hz). A higher overlap indicates that the vibration frequency is closer to the dangerous resonant frequency of the NIC and, therefore, the greater the threat. Second, the system analyzes the intensity and duration of the vibration data. Both short bursts of strong vibration and long-term accumulation of weak vibration can pose a threat. The system performs a comprehensive assessment by calculating the time-frequency integrated energy and combining it with the frequency overlap. Third, the system calculates the distance between the vibration source and key components, including the spatial distance from PCIe gold fingers, key chips, connectors, and other components. The closer the distance, the greater the threat. The system combines these three indicators through a weighted average to create a final vibration threat level score, which ranges from 1 to 10, with 1-3 indicating low risk, 4-7 indicating medium risk, and 8-10 indicating high risk. The system then combines the vibration source classification results with the vibration threat level score to generate complete vibration source information, including source type, location coordinates, primary frequency characteristics, threat level, and recommended protective measures. This comprehensive vibration source information provides a precise basis for formulating network card protection strategies.

[0037] Furthermore, the analysis of changes in electromagnetic characteristics of the high-speed PHY chip and FPGA before and after the network card is powered on, and identifying electromagnetic sensitive areas by comparing the differences in frequency response functions before and after power-on include: calculating a frequency response function matrix around the high-speed PHY chip and FPGA in an unpowered state of the network card based on the vibration data; during the power-on process of the network card, collecting post-power-on vibration data of the high-speed PHY chip and FPGA in idle mode, standard load mode, and high load mode respectively; calculating a frequency response function for the post-power-on vibration data, performing a difference operation with the frequency response function matrix in an unpowered state, and obtaining a frequency response difference matrix caused by changes in the electromagnetic state; extracting major change patterns from the frequency response difference matrix, and constructing an electromagnetic sensitivity distribution map on the three-dimensional model of the network card to mark the electromagnetic sensitive areas.

[0038] Specifically, the system reads vibration data collected while the network card is powered off from independent flash memory, focusing specifically on sensor data around the high-speed physical physical layer (PHY) chip and the FPGA. This data reflects purely mechanical vibration characteristics and is unaffected by electromagnetic interference. The system first preprocesses the raw vibration data, removing DC offset, applying a high-pass filter (5Hz cutoff frequency) to eliminate low-frequency drift, and a low-pass filter (2500Hz cutoff frequency) to suppress high-frequency noise. The system then segments the processed time-domain data into 2048 sampling points per segment, with 50% overlap between adjacent segments. A Hanning window function is applied to reduce spectral leakage. A fast Fourier transform is performed on each data segment to calculate power spectral density and phase information. Next, the system calculates the transfer function between each monitoring point and constructs a complete frequency response function matrix. This matrix is a three-dimensional data structure with dimensions F×N×N, where F represents the number of frequency points (0-2500Hz, resolution 2Hz) and N represents the number of sensor nodes. Each element in the matrix, FRF(f,i,j), represents the transfer function value from node i to node j at frequency f, describing the amplitude ratio and phase difference of vibration propagating from one point to another. In particular, the diagonal element, FRF(f,i,i), represents the self-frequency response characteristics at node i. These frequency response function matrices comprehensively characterize the mechanical vibration characteristics of the network card when unaffected by electromagnetic fields, providing benchmark data for subsequent comparisons.

[0039] Specifically, after obtaining the frequency response function matrix for the unpowered state, the system needs to collect vibration data when the NIC is powered on. The NIC power-up process is divided into three phases: power-on initialization, normal operation, and load variation. During the power-on initialization phase, the system records the vibration responses of the PHY chip and FPGA during the power sequence, clock startup, and initialization process. During the normal operation phase, the system controls the NIC to enter three typical operating modes: idle mode, standard load mode, and high load mode. In idle mode, the NIC maintains only basic functions, with near-zero network traffic and minimal circuit activity. In standard load mode, the NIC processes approximately 50% of its rated throughput, with moderate activity on the FPGA and PHY chip. In high load mode, the NIC processes nearly 90% of its rated throughput, with peak circuit activity. In each mode, the system collects vibration data using a sensor array surrounding the PHY chip and FPGA. The sampling rate is set to 10kHz to capture vibration frequency components up to 5kHz. The acquisition lasts for 120 seconds to ensure statistical significance. The system also records operating parameters such as power consumption, temperature variations, and clock frequency in each mode. These parameters are stored along with the vibration data for subsequent analysis of the relationship between electromagnetic field intensity and vibration response. This multi-modal vibration acquisition method captures the vibration characteristics of the network card under varying levels of electromagnetic activity, providing a comprehensive experimental dataset for identifying electromagnetically sensitive areas.

[0040] Specifically, the system calculates the frequency response function for the collected vibration data after power-on, using the same method used for processing unpowered data. Data from idle, standard load, and high load modes are processed separately, resulting in three sets of frequency response function matrices for the powered state. The system then performs a difference calculation between the frequency response function matrices for each mode and the baseline matrix for the unpowered state. This difference calculation includes two components: amplitude difference and phase difference. The amplitude difference reflects the vibration enhancement or suppression effect caused by electromagnetic activity, while the phase difference indicates the impact of electromagnetic activity on vibration propagation characteristics. The system performs this difference calculation for each frequency point and each pair of sensor nodes, forming a frequency response difference matrix caused by electromagnetic state changes. To facilitate analysis, the system normalizes the amplitude differences, calculating relative rates of change rather than absolute differences to eliminate the influence of variations in base vibration intensity at different locations. Furthermore, the system calculates differences between different load modes, such as the difference between high load and standard load. These difference matrices reveal the incremental impact of load changes (i.e., enhanced electromagnetic activity) on vibration characteristics. Through this multi-level difference analysis, the system can separate the changes in vibration characteristics caused by purely electromagnetic factors, effectively eliminating interference from non-electromagnetic factors such as temperature changes and mechanical looseness.

[0041] Specifically, the system performs advanced data analysis on the frequency response difference matrix to extract key variation patterns and construct an electromagnetic susceptibility distribution map. First, the system applies principal component analysis (PCA) to reduce the dimensionality of the difference matrix and extract the most significant variation patterns. PCA analysis identifies the most prominent groups of variation patterns at different frequencies and locations, typically retaining the top 3-5 principal components that explain more than 80% of the variance. Each principal component represents a typical electromagnetic-vibration coupling pattern, encompassing both frequency and spatial distribution characteristics. The system then performs cluster analysis on the principal components to identify clusters of regions with similar characteristics. These clusters typically correspond to specific types of electromagnetically active areas. The system then maps the analysis results onto a 3D CAD model of the network card. The 3D model contains detailed information such as the PCB stackup, component layout, and signal line distribution. The system then displays the electromagnetic susceptibility distribution using a heat map, with a color gradient from blue (low sensitivity) to red (high sensitivity), visually demonstrating the degree of electromagnetic impact in different areas. Electromagnetic sensitive areas are typically concentrated around high-speed PHY chips, near FPGA core power supplies, around clock generators, and along high-speed differential signal lines. In addition, the system also marks the main frequency characteristics and load correlation of each sensitive area. For example, a certain area is particularly sensitive to load changes in the 100-150Hz frequency band.

[0042] Furthermore, the vibration source is classified according to the correlation between the vibration source coordinates, the dangerous frequency components and the electromagnetic sensitive area, and the vibration source classification result is obtained, including: calculating the distance relationship between the vibration source coordinates and the position of the internal structural components of the network card, and judging whether the vibration source is located around the key components inside the network card; analyzing the spectral correlation between the dangerous frequency components and the server fan speed and the position relationship between the vibration source coordinates and the PCIe gold finger area, and judging whether the vibration is caused by external conduction or the heat dissipation system; calculating the spatial overlap degree between the vibration source coordinates and the electromagnetic sensitive area, and combining the changes in vibration characteristics before and after power-on to judge whether the vibration is related to electromagnetic interference coupling; according to the confidence score of the above judgment result, applying a multi-feature fusion decision algorithm, the vibration source is classified into internal structure loose type, external mechanical conduction type, heat dissipation system induced type, electromagnetic interference coupling type or composite type, and calculating the contribution weight of each type.

[0043] Specifically, the system first extracts PCB layout information from the network card design database and establishes a location database for the internal structural components of the network card. This database contains the precise coordinates of all key components, such as large capacitors (especially filter capacitors and decoupling capacitors), connectors (such as SFP+ interfaces and RJ45 interfaces), transformers, crystal oscillators, large chips (PHY chips, FPGAs, etc.), power management units, and various mechanical fixing points. The system then calculates the Euclidean distance between the vibration source coordinates obtained in the previous step and the center points of each key component to generate a distance vector. For components with a distance less than a set threshold (usually 15mm), the system further analyzes whether the vibration source coordinates fall within the boundary of the component or in the immediate vicinity. In addition, the system also comprehensively considers the relative positional relationship between the vibration source and multiple adjacent components and calculates the degree to which the vibration source is located in the component cluster. Combining these spatial relationship analyses, the system generates a component correlation score. A higher value indicates a stronger correlation between the vibration source and the internal structural components. When the score exceeds 0.75, the system preliminarily determines that the vibration source is located around the key components inside the network card. The system further analyzes whether the dangerous frequency components match the mechanical characteristics of the relevant components, for example, comparing whether the vibration frequency is close to the natural vibration frequency of large capacitors or connectors (usually in the range of 150-300Hz). By combining this analysis of location and frequency characteristics, the system can reliably determine whether the vibration is caused by structural looseness or resonance of the internal components of the network card.

[0044] Specifically, after determining the relationship between the vibration source and internal components, the system further analyzes whether the vibration is caused by external conduction or the cooling system. First, the system analyzes the spectral correlation between the critical frequency components and the server fan speed. The system obtains historical fan speed data from the server management module, including the time series of speed changes for each fan, and calculates the fan's fundamental frequency (speed × number of blades / 60) and its harmonic components. The system then calculates the time correlation coefficient and spectral similarity between the energy changes of the critical frequency components in the vibration signal and the fan speed changes. A high correlation (correlation coefficient > 0.7) indicates that the vibration is likely caused by the server cooling system. The system also analyzes the positional relationship between the vibration source coordinates and the PCIe gold finger area. The system calculates the distance from the vibration source to the PCIe gold finger area and analyzes the vibration propagation direction. The vibration propagation direction is determined by calculating the phase difference and propagation delay of the vibration signals at different monitoring points to form a vibration propagation vector. If the vibration source is close to the PCIe interface (distance < 25 mm) and the propagation vector indicates that the vibration is propagating inward from this area, the vibration is likely caused by external mechanical conduction. The system also analyzes the attenuation characteristics of the vibration propagation path. Externally conducted vibrations typically exhibit significant amplitude attenuation along the propagation path, while internally sourced vibrations exhibit amplitude peaks near the source. By comprehensively considering spectral correlation, positional relationships, and propagation characteristics, the system generates two judgment scores: a cooling system correlation score and an externally conducted correlation score, which quantify the vibration's degree of correlation with these two external sources, respectively.

[0045] Specifically, after assessing internal and external factors, the system calculates the correlation between the vibration source and the electromagnetically sensitive area, identifying electromagnetic interference-coupled vibration. The system first calculates the degree of spatial overlap between the vibration source coordinates and the electromagnetically sensitive area determined in the previous step. Specifically, the system represents the electromagnetically sensitive area as a Gaussian distribution in three-dimensional space, centered at the point of highest sensitivity and with decreasing intensity according to sensitivity level. The system then calculates the sensitivity value at the vibration source location to obtain a spatial overlap score. A high overlap score (>0.6) indicates that the vibration source is located in an area significantly affected by electromagnetic activity. Next, the system analyzes the degree of change in vibration characteristics before and after power-on. This analysis is based on the frequency response difference matrix calculated in the previous step. The system extracts the difference values near the vibration source location and calculates the change in vibration amplitude, frequency, and phase before and after power-on. Significant characteristic changes (rate of change >30%) indicate that the vibration is strongly affected by electromagnetic activity. Furthermore, the system analyzes the correlation between vibration intensity and changes in network card load, comparing the differences in vibration characteristics under different load modes. A high positive correlation between vibration intensity and load level further supports the identification of electromagnetic interference coupling. By integrating spatial overlap, characteristic variation, and load correlation, the system generates an EMI coupling correlation score that quantifies the degree of correlation between vibration and electromagnetic activity. This analysis overcomes the difficulty in distinguishing between electromagnetic and mechanical factors in traditional vibration monitoring.

[0046] Specifically, based on the various correlation scores generated in the first three steps, the system applies a multi-feature fusion decision algorithm for final classification. This algorithm first normalizes each classification feature so that all correlation scores fall within the range of 0-1. The system then constructs a feature vector, encompassing dimensions such as internal component correlation scores, cooling system correlation scores, external conduction correlation scores, and electromagnetic interference coupling scores. Next, the system evaluates the feature vector using a classifier that combines a decision tree and support vector machine. Based on a model pre-trained using finite element analysis and experimental data, the classifier accurately identifies the characteristic combinations of various vibration modes. The classification process not only outputs a judgment of the vibration source type but also generates a confidence score for each type to quantify the reliability of the classification result. When the confidence score of a particular type exceeds a threshold (typically 0.8) and is significantly higher than that of other types, the system classifies the vibration source as belonging to that single type. When the confidence scores of multiple types are close and all exceed a lower threshold (typically 0.5), the system classifies the vibration source as a composite type and calculates the contribution weights of each type. Contribution weights are calculated based on the normalized confidence scores, reflecting the relative importance of each vibration mechanism in the composite vibration. Ultimately, the system outputs a vibration source classification, including the type (internal structural looseness, external mechanical conduction, cooling system induction, electromagnetic interference coupling, or a combination), a confidence score, and the contribution weights of each type (for composite types).

[0047] 104. Based on the vibration source information and combined with the real-time communication status monitoring of the network card, select and implement corresponding protection measures to protect the network card data and communication stability.

[0048] In one embodiment of the present invention, the vibration source information is combined with the real-time communication status monitoring of the network card to select and execute corresponding protection measures to protect the network card data and communication stability, including: monitoring the communication traffic, bit error rate, retransmission rate, and link quality indicators of the network card, performing correlation analysis with the vibration source information, and constructing a vibration-performance impact map; selecting corresponding levels of protection measures according to the vibration-performance impact map and the vibration threat level in the vibration source information; updating the vibration source information according to the execution results of the protection measures and the changes in the communication status of the network card, and completing dynamic adjustment of the protection measures.

[0049] Specifically, the system continuously collects multiple communication performance metrics through the network card driver interface, including communication traffic (measured in Mbps), bit error rate (expressed as a 10^-12 bit error rate), retransmission rate (expressed as a percentage), and link quality indicators (such as signal quality and signal-to-noise ratio). This performance data is collected at second-level intervals, forming a continuous time series record. Simultaneously, the system reads the vibration source information obtained in the previous step and extracts key parameters such as vibration type, location, frequency characteristics, and threat level. Next, the system performs time series matching, aligning vibration events with communication performance fluctuations and calculating the temporal correlation between the two. For each performance metric, the system calculates its change trend before, during, and after the vibration occurs, quantifying the degree of vibration's impact on the metric. Furthermore, the system analyzes the differential impact of different types of vibration on each performance metric. For example, electromagnetic interference-coupled vibration typically has a significant impact on the bit error rate, while mechanical conduction vibration may primarily affect the retransmission rate. Based on this analysis, the system constructs a vibration-performance impact map, a multidimensional data structure that maps the quantitative relationship between vibration characteristics (type, frequency, and intensity) and network card performance metrics. The map contains multiple impact curves, each describing the expected change in a performance indicator under specific vibration conditions. This map enables the system to predict how network card performance will change under different vibration conditions, providing a basis for selecting protective measures.

[0050] Specifically, after constructing a vibration-performance impact map, the system selects a corresponding protection strategy based on the map's content and the vibration threat level. Protection strategies are divided into three main levels: low, medium, and high, corresponding to vibration events of varying threat levels. For low-level vibrations (threat levels 1-3), the system adjusts signal processing parameters, including fine-tuning the PHY chip's equalizer parameters (increasing the number of taps in the feedforward and decision feedback equalizers), optimizing clock recovery circuit parameters (adjusting the phase-locked loop bandwidth and phase detector gain), and enhancing the jitter tolerance of the signal sampling points. These adjustments are implemented by sending specific commands to the PHY chip via the NIC driver interface or the BMC. This enhances signal processing's resilience to vibration interference while having little impact on NIC performance. For medium-level vibrations (threat levels 4-7), the system signals the operating system via the BMC to negotiate a lower communication rate or modify link parameters. Specific measures include downshifting a 10Gbps link to 5Gbps or 1Gbps, increasing the interframe gap between transmit frames, enabling additional error correction coding, and adjusting the automatic repeat request (ARQ) strategy. These measures sacrifice some throughput in exchange for significantly improved link stability. For high-level vibrations with a threat level of 8-10, the system triggers a protective data preservation mechanism, rapidly dumping critical data from the network card memory buffer, DMA descriptor table, and status register to non-volatile memory. It also sends a warning message containing detailed vibration information to the system administrator. Furthermore, the system implements targeted protective measures based on the vibration type: for vibrations induced by the cooling system, a request is sent to the server management module to adjust fan speed to avoid dangerous resonant frequencies. For vibrations coupled with electromagnetic interference, the system dynamically adjusts the FPGA clock frequency or core voltage to mitigate electromagnetic-vibration coupling effects.

[0051] Specifically, after implementing protective measures, the system continuously monitors changes in the network card's communication status and the effectiveness of the protection measures, updating vibration source information and protection strategies accordingly. The system first evaluates performance changes before and after the protective measures are implemented, calculating the improvement rate, such as the percentage reduction in bit error rate and the degree of improvement in link stability. Simultaneously, the system monitors changes in vibration characteristics, including trends in amplitude, frequency distribution, and duration. Based on this real-time data, the system updates various parameters in the vibration source information: adjusting the vibration threat level score to reflect the latest observed actual threat level; updating the confidence score for vibration type judgment to incorporate newly acquired evidence of vibration-performance impact; and refining the vibration frequency characteristic description to highlight the frequency bands with the greatest actual impact on communication performance. Furthermore, the system optimizes the vibration-performance impact map, adjusting the impact curve based on new data to improve prediction accuracy. If the initial protective measures are ineffective (performance improvement is lower than expected), the system will automatically upgrade the protection level or adjust the protection strategy. For example, if fan speed adjustment for cooling system-induced vibration is found to be ineffective, the system may instead implement a communication parameter adjustment strategy. On the contrary, if the protection measures are excessive (such as unnecessarily reducing the communication rate), the system will appropriately downgrade the protection measures to find the optimal balance between performance and stability.

[0052] In this embodiment, multiple triaxial acceleration sensor arrays are deployed in different functional areas of the network card to collect vibration data when the card is not powered on. After the card is powered on, the baseboard management controller reads the vibration data, extracts the vibration spectrum characteristics using time-frequency analysis methods, and identifies dangerous frequency components based on the characteristics of the server cooling system and the physical parameters of the network card. Based on the vibration data and dangerous frequency components, combined with the changes in the electromagnetic characteristics before and after the network card is powered on, the vibration source is located and classified to determine the vibration threat level. Based on the vibration source information and the real-time communication status of the network card, appropriate protective measures are selected and implemented. This invention solves the problems of spectrum analysis blind spots and vibration source identification ambiguity, effectively protecting network card data and communication stability through targeted protective measures.

[0053] The above describes the network card vibration detection method according to the embodiment of the present invention. The following describes the network card vibration detection device according to the embodiment of the present invention. Figure 2 In one embodiment of the present invention, a network card vibration detection device includes: The acquisition module 201 is used to collect vibration data of each functional area of the network card by using multiple groups of three-axis acceleration sensor arrays deployed in different functional areas of the network card when the network card is not powered on; An analysis module 202 is configured to read the vibration data through a baseboard management controller after the network card is powered on, extract the vibration spectrum characteristics of the vibration data using a time-frequency analysis method, and identify dangerous frequency components based on the characteristics of the server cooling system and the physical parameters of the network card; A positioning module 203 is configured to locate and classify the vibration source based on the vibration data and the dangerous frequency components, combined with changes in electromagnetic characteristics before and after the network card is powered on, and determine the vibration threat level to obtain vibration source information; The protection module 204 is used to select and execute corresponding protection measures based on the vibration source information and in combination with the real-time communication status monitoring of the network card to protect the network card data and communication stability.

[0054] In an embodiment of the present invention, the network card vibration detection device implements the aforementioned network card vibration detection method. The device collects vibration data when the network card is not powered on by deploying multiple triaxial acceleration sensor arrays in different functional areas of the network card. After the network card is powered on, the baseboard management controller reads the vibration data, extracts vibration spectrum characteristics using a time-frequency analysis method, and identifies dangerous frequency components based on the characteristics of the server cooling system and the physical parameters of the network card. Based on the vibration data and dangerous frequency components, and in combination with changes in electromagnetic characteristics before and after the network card is powered on, the vibration source is located and classified to determine the vibration threat level. Based on the vibration source information and the real-time communication status of the network card, corresponding protective measures are selected and implemented. This present invention addresses the issues of blind spots in spectrum analysis and ambiguity in vibration source identification, effectively protecting network card data and communication stability through targeted protective measures.

[0055] above Figure 2 The network card vibration detection device in the embodiment of the present invention is described in detail from the perspective of modular functional entities. The network card vibration detection device in the embodiment of the present invention is described in detail from the perspective of hardware processing.

[0056] Figure 3 This is a schematic structural diagram of a network card vibration detection device provided by an embodiment of the present invention. The network card vibration detection device 300 may vary significantly due to different configurations or performance. It may include one or more processors (central processing units, CPUs) 310 (for example, one or more processors), a memory 320, and one or more storage media 330 (for example, one or more mass storage devices) storing application programs 333 or data 332. The memory 320 and storage medium 330 may be either transient or persistent storage. The program stored in the storage medium 330 may include one or more modules (not shown), each of which may include a series of instruction operations in the network card vibration detection device 300. Furthermore, the processor 310 may be configured to communicate with the storage medium 330, and execute the series of instruction operations in the storage medium 330 on the network card vibration detection device 300 to implement the steps of the aforementioned network card vibration detection method.

[0057] The network card vibration detection device 300 may further include one or more power supplies 340, one or more wired or wireless network interfaces 350, one or more input and output interfaces 360, and / or one or more operating systems 331, such as Windows Server, Mac OS X, Unix, Linux, FreeBSD, etc. It will be understood by those skilled in the art that Figure 3 The illustrated structure of the network card vibration detection device does not limit the network card vibration detection device provided by the present invention, and may include more or fewer components than illustrated, or combine certain components, or arrange the components differently.

[0058] The present invention also provides a computer-readable storage medium, which may be a non-volatile computer-readable storage medium or a volatile computer-readable storage medium. Instructions are stored in the computer-readable storage medium. When the instructions are executed on a computer, the computer executes the steps of the network card vibration detection method.

[0059] Those skilled in the art will clearly understand that, for the convenience and brevity of description, the specific working processes of the above-described systems, devices, and units can refer to the corresponding processes in the aforementioned method embodiments and will not be repeated here.

[0060] If the integrated unit is implemented as a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present invention, or the portion that contributes to the prior art, or all or part of the technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions for enabling a computer device (which can be a personal computer, server, or network device, etc.) to execute all or part of the steps of the method described in each embodiment of the present invention. The aforementioned storage medium includes various media that can store program code, such as a USB flash drive, a mobile hard drive, a read-only memory (ROM), a random access memory (RAM), a magnetic disk, or an optical disk.

[0061] As described above, the above embodiments are only used to illustrate the technical solutions of the present invention, rather than to limit the same. Although the present invention has been described in detail with reference to the above embodiments, those skilled in the art should understand that the technical solutions described in the above embodiments can still be modified, or some of the technical features thereof can be replaced by equivalents. However, these modifications or replacements do not deviate the essence of the corresponding technical solutions from the spirit and scope of the technical solutions of the embodiments of the present invention.

Claims

1. A network card vibration detection method, characterized in that: The network card vibration detection method comprises: When the network card is not powered on, multiple sets of three-axis acceleration sensor arrays deployed in different functional areas of the network card are used to collect vibration data from each functional area of the network card. After the network card is powered on, the vibration data is read by the baseboard management controller, and the vibration spectrum characteristics of the vibration data are extracted using a time-frequency analysis method. The dangerous frequency components are identified by combining the characteristics of the server cooling system and the physical parameters of the network card; Based on the vibration data and the dangerous frequency components, combined with the changes in electromagnetic characteristics before and after the network card is powered on, the vibration source is located and classified, and the vibration threat level is determined to obtain vibration source information; According to the vibration source information and combined with the real-time communication status monitoring of the network card, corresponding protection measures are selected and executed to protect the network card data and communication stability.

2. The network card vibration detection method according to claim 1, wherein: After the network card is powered on, the vibration data is read by the baseboard management controller, the vibration spectrum characteristics of the vibration data are extracted using a time-frequency analysis method, and the dangerous frequency components are identified by combining the characteristics of the server cooling system and the physical parameters of the network card. After the network card is powered on, the vibration data is read by the baseboard management controller, and the vibration data is decomposed into time and frequency domains using a multi-resolution wavelet transform to obtain the time domain distribution of different frequency components; Obtain historical cooling fan speed data from the server management module, calculate the fan blade pass frequency and harmonic components, and construct a cooling noise spectrum template; Applying an adaptive spectrum subtraction algorithm to the time domain distribution of the different frequency components according to the heat dissipation noise spectrum template to remove the vibration components caused by the heat dissipation system and obtain a vibration spectrum after removal; The natural frequency and dangerous resonance mode of the network card are calculated based on the thickness, elastic modulus, size, and fixing point positions of the network card PCB, forming a spectrum diagram of the network card vulnerability. By calculating the degree of overlap between the vibration spectrum after stripping and the network card vulnerability spectrum diagram, the dangerous frequency component that poses the greatest threat to the structural integrity of the network card is determined.

3. The network card vibration detection method according to claim 2, wherein: Applying an adaptive spectrum subtraction algorithm to the time domain distribution of the different frequency components according to the heat dissipation noise spectrum template to remove the vibration components caused by the heat dissipation system to obtain the stripped vibration spectrum includes: The time domain distribution of the different frequency components is converted to the frequency domain using fast Fourier transform, and the original vibration spectrum matrix corresponding to frequency and energy is constructed, wherein the matrix elements represent the energy distribution of each frequency point in different time windows; Based on the fan blade passing frequency and harmonic components in the heat dissipation noise spectrum template, the correlation with the original vibration spectrum is calculated to determine the pollution degree of the heat dissipation noise at each frequency point, and the adaptive gain factor is calculated segmentally according to the pollution degree; Applying the adaptive gain factor to each frequency point of the original vibration spectrum matrix, performing a point-by-point spectrum subtraction operation, and obtaining processed spectrum matrix data; The inverse fast Fourier transform is applied to the processed spectrum matrix data to reconstruct the signals of each frequency band into the time domain. The instantaneous frequency characteristics of the signal are extracted through Hilbert transform. Combined with the amplitude information, the vibration spectrum after stripping off the vibration components caused by the heat dissipation system is obtained.

4. The network card vibration detection method according to claim 2, wherein: Based on the vibration data and the dangerous frequency components, combined with the changes in electromagnetic characteristics before and after the network card is powered on, the vibration source is located and classified, and the vibration threat level is determined. The vibration source information obtained includes: Analyze the changes in the electromagnetic characteristics of the high-speed PHY chip and FPGA before and after the network card is powered on. By comparing the differences in the frequency response functions before and after power-on, identify electromagnetically sensitive areas. By using the time difference of the vibration data captured in each functional area and applying a modified triangulation algorithm, the coordinates of the vibration source are located on the three-dimensional spatial model of the network card; Classifying the vibration source according to the correlation between the vibration source coordinates, the dangerous frequency components, and the electromagnetic sensitive area to obtain a vibration source classification result, wherein the vibration source classification result includes one or more of an internal structure loose type, an external mechanical conduction type, a heat dissipation system induction type, an electromagnetic interference coupling type, and a combined type; The vibration threat level is calculated based on the overlap between the dangerous frequency component and the network card vulnerability spectrum, vibration data, and the distance between the vibration source and key components, and the vibration source classification result and the vibration threat level are combined to form vibration source information.

5. The network card vibration detection method according to claim 4, characterized in that: Analyzing the electromagnetic characteristics of the high-speed PHY chip and FPGA before and after the network card is powered on, and identifying electromagnetically sensitive areas by comparing the frequency response function differences before and after powering on, includes: Calculate the frequency response function matrix around the high-speed PHY chip and the FPGA in the unpowered state of the network card based on the vibration data; During the network card power-up process, the post-power-on vibration data of the high-speed PHY chip and FPGA were collected in idle mode, standard load mode, and high load mode. Calculating a frequency response function for the vibration data after power-on, performing a difference operation on the frequency response function matrix in the non-power-on state, and obtaining a frequency response difference matrix caused by the change in electromagnetic state; The main change mode is extracted from the frequency response difference matrix, and an electromagnetic sensitivity distribution map is constructed on the three-dimensional model of the network card to mark the electromagnetic sensitive areas.

6. The network card vibration detection method according to claim 4, characterized in that: The classifying the vibration source according to the correlation between the vibration source coordinates, the dangerous frequency components and the electromagnetic sensitive area to obtain the vibration source classification result includes: Calculating the distance relationship between the coordinates of the vibration source and the position of the internal structural components of the network card to determine whether the vibration source is located around the key components inside the network card; Analyze the spectral correlation between the dangerous frequency component and the server fan speed, and the positional relationship between the vibration source coordinates and the PCIe gold finger area, to determine whether the vibration is caused by external conduction or the heat dissipation system; Calculating the spatial overlap between the vibration source coordinates and the electromagnetic sensitive area, and combining the vibration characteristics changes before and after power-on to determine whether the vibration is related to electromagnetic interference coupling; Based on the confidence scores of the above judgment results, a multi-feature fusion decision algorithm is applied to classify the vibration sources into internal structure loose type, external mechanical conduction type, heat dissipation system induced type, electromagnetic interference coupling type or composite type, and calculate the contribution weight of each type.

7. The network card vibration detection method according to claim 1, characterized in that: The selecting and executing corresponding protective measures based on the vibration source information and combined with the real-time communication status monitoring of the network card to protect the network card data and communication stability includes: Monitor the communication traffic, bit error rate, retransmission rate, and link quality indicators of the network card, perform correlation analysis with the vibration source information, and construct a vibration-performance impact map; Selecting corresponding levels of protective measures based on the vibration threat level in the vibration-performance impact map and the vibration source information; According to the execution results of the protective measures and the changes in the communication status of the network card, the vibration source information is updated to complete the dynamic adjustment of the protective measures.

8. A network card vibration detection device, characterized in that: The network card vibration detection device comprises: The acquisition module is used to collect vibration data from various functional areas of the network card when the network card is not powered on, using multiple sets of three-axis acceleration sensor arrays deployed in different functional areas of the network card; An analysis module is configured to read the vibration data through a baseboard management controller after the network card is powered on, extract vibration spectrum characteristics of the vibration data using a time-frequency analysis method, and identify dangerous frequency components based on characteristics of the server cooling system and physical parameters of the network card; a positioning module for locating and classifying the vibration source based on the vibration data and the dangerous frequency components, in combination with changes in electromagnetic characteristics before and after the network card is powered on, and determining the vibration threat level to obtain vibration source information; The protection module is used to select and execute corresponding protection measures based on the vibration source information and combined with the real-time communication status monitoring of the network card to protect the network card data and communication stability.

9. A network card vibration detection device, characterized in that: The network card vibration detection device includes: a memory and at least one processor, wherein the memory stores instructions; The at least one processor calls the instruction in the memory to enable the network card vibration detection device to perform the steps of the network card vibration detection method according to any one of claims 1 to 7.

10. A computer-readable storage medium having instructions stored thereon, characterized in that: When the instructions are executed by the processor, the steps of the network card vibration detection method according to any one of claims 1 to 7 are implemented.

Images