Server fault early warning system and method for computer

Abstract

The invention relates to the technical field of computer fault early warning, in particular to a server fault early warning system and method for a computer. Comprising a multi-source signal acquisition module, a data preprocessing module, a cross-domain signal correlation analysis module, a nonlinear dynamic modeling module, a frequency domain feature extraction module, an intermittent fault prediction model and a multi-level fault early warning module. According to the method, through technologies of multi-source signal fusion acquisition, nonlinear dynamic modeling, frequency domain feature extraction, Bayesian reasoning and the like, accurate prediction and multi-level early warning of intermittent faults of the server are realized, and the accuracy and real-time response capability of fault detection are improved, so that stable operation of the server is effectively guaranteed.

Description

Technical Field

[0001] The present invention relates to the technical field of computer fault warning, and particularly to a server fault warning system and method for computers. Background Art

[0002] With the rapid development of computer and information technologies, servers play a crucial role in the business operations of various enterprises and institutions. The stability and efficient operation of servers directly affect the performance and business continuity of information systems. However, during long-term operation, servers may malfunction due to reasons such as hardware aging, environmental changes, and load fluctuations, especially intermittent faults, which are often difficult to predict and may lead to serious business interruptions.

[0003] Existing server fault detection methods mostly focus on post-mortem analysis, usually relying on log records or simple threshold alarm mechanisms, and it is difficult to effectively analyze the dynamic associations between complex and multi-source signals. At the same time, these methods often cannot accurately identify the occurrence time of intermittent faults, and lack accurate evaluation means when facing the coupling effect between server operating states and environmental signals.

[0004] Therefore, there is an urgent need for a comprehensive fault warning system and method to improve the accuracy and lead time of server fault detection, and ensure business continuity and stability. Summary of the Invention

[0005] Based on the above purpose, the present invention provides a server fault warning system and method for computers.

[0006] A server fault warning system for computers includes a multi-source signal acquisition module, a data preprocessing module, a cross-domain signal correlation analysis module, a nonlinear dynamics modeling module, a frequency-domain feature extraction module, an intermittent fault prediction model, and a multi-level fault warning module; where: The multi-source signal acquisition module: is used to collect multi-dimensional operating signals from server hardware and the environment, and the operating signals include CPU load, memory usage rate, disk read and write rate, network traffic, power input noise, fan speed fluctuation, and server physical vibration, and output the collected signals in the form of a time series; The data preprocessing module: receives the time series data from the multi-source signal acquisition module, and performs preprocessing operations such as data cleaning, denoising, and time alignment; The cross-domain signal correlation analysis module: based on the preprocessed data, uses a multivariate autoregressive model to analyze the time-domain correlation between different signals, generates a signal correlation matrix, and is used to identify the dynamic coupling feature data between the server operating state and external environmental signals; Nonlinear dynamics modeling module: Based on the linkage feature data output by the cross-domain signal correlation analysis module, the server's operating status is modeled using the Lyapunov exponent and strange attractor model to analyze the nonlinear behavior of the server when it transitions from a stable state to a fault state, so as to identify the critical state of the server operation and generate nonlinear feature data related to intermittent faults; Frequency domain feature extraction module: receives the preprocessed data from the data preprocessing module, and uses Fourier transform and wavelet analysis methods to convert the time domain data into frequency domain data to extract the frequency features of the server operation; Intermittent fault prediction model: Combines the data output by the nonlinear dynamics modeling module and the frequency domain feature extraction module, performs a comprehensive assessment of the server's health status based on the Bayesian reasoning model, and calculates the probability of intermittent faults. Multi-level fault warning module: used to divide the risk level of intermittent faults into multiple levels according to the probability of fault occurrence, and generate corresponding warning signals.

[0007] Optionally, the multi-source signal acquisition module includes a CPU load acquisition unit, a memory usage acquisition unit, a disk read / write rate acquisition unit, a network traffic acquisition unit, a power input noise acquisition unit, a fan speed fluctuation acquisition unit, and a server physical vibration acquisition unit; wherein: CPU load collection unit: used to collect the server's CPU usage in real time by reading the server's hardware performance monitoring interface. The collected data is recorded in units of one second. Memory usage collection unit: obtains the current memory usage by accessing the server's memory management library and records it at fixed time intervals; Disk read / write rate acquisition unit: used to obtain the current disk read / write operation rate from the server storage device interface. The rate is stored in a time series manner and output with a timestamp; Network traffic collection unit: monitors the current network data transmission rate through the server network interface controller, and outputs the collected traffic data according to the traffic fluctuation situation per second; Power input noise acquisition unit: used to obtain high-frequency noise signals at the power input end of the server and use them as important signals of the operating status. Specific data is captured through frequency sampling and output in a time series. Fan speed fluctuation collection unit: monitors fan speed fluctuations in real time by reading the server fan controller interface, and stores the acquired speed data in a time series format; Server physical vibration collection unit: uses built-in vibration sensors to capture physical vibration data of server chassis and hardware components. The vibration data is recorded at millisecond time intervals and output in the form of time series.

[0008] Optionally, the data preprocessing module includes a data cleaning unit, a denoising unit, and a time alignment unit; where: Data cleaning unit: used to identify and eliminate abnormal data and missing data collected by the multi-source signal acquisition module. By setting the range of abnormal value thresholds, it automatically filters out sudden abnormal values in the signal, and linearly interpolates and completes the missing data based on historical data to ensure data integrity; Denoising unit: used to perform noise processing on the collected signal. By analyzing the frequency components in the signal, it removes high-frequency or low-frequency noise signals, especially the interference parts in the power input noise and physical vibration signals, thereby improving the purity of the effective signal; Time alignment unit: used to perform time series alignment on the collected signals, and mark different signals in the multi-source signal acquisition module with a unified time reference.

[0009] Optionally, the cross-domain signal correlation analysis module includes a data input unit, a multivariate autoregressive analysis unit, a signal correlation matrix generation unit, and a dynamic linkage characteristic identification unit; where: Data input unit: receives the preprocessed multi-source signal data output by the data preprocessing module, and inputs all the data into the multivariate autoregressive model according to the time series; Multivariate autoregressive analysis unit: based on the time series data provided by the data input unit, adopts a multivariate autoregressive model, models and analyzes the time changes of each signal, comprehensively estimates the linkage effect between signals, and generates an autoregressive coefficient matrix describing the dynamic correlation between signals; Signal correlation matrix generation unit: based on the autoregressive coefficient matrix generated by the multivariate autoregressive analysis unit, generates a signal correlation matrix. Each element in the correlation matrix represents the time-domain influence intensity and direction of the corresponding signal on other signals, and is used to represent the dynamic relationship between the server operating state and the external environment signal; Dynamic linkage characteristic identification unit: by analyzing the signal correlation matrix, identifies the dynamic linkage characteristics between the server operating state and the external environment signal, including the influence of power input noise on CPU load and the linkage effect of physical vibration on fan speed, and generates linkage characteristic data.

[0010] Optionally, the signal correlation matrix generation unit specifically includes: Receive autoregressive coefficient matrix: first receive the autoregressive coefficient matrix generated by the multivariate autoregressive analysis unit, and the autoregressive coefficient matrix is generated by the autoregressive model; Calculate the total correlation degree between signals: using the received autoregressive coefficient matrix, calculate the signals by accumulating the coefficients of all regression orders and signals The total correlation ; Calculate the directionality of signal correlation: Based on the calculation of the total correlation, analyze the directionality of signal correlation, which is determined by the sign of each regression coefficient, that is, a positive coefficient indicates a positive impact, and a negative coefficient indicates a negative impact; Construct signal correlation matrix: Generate signal correlation matrix based on the calculation results of total correlation and directionality , each element of the matrix Indicates signal Signal The temporal domain affects the intensity and direction of the

[0011] Optionally, the nonlinear dynamics modeling module includes a Lyapunov exponent calculation unit, a strange attractor modeling unit, a nonlinear behavior analysis unit and a nonlinear feature generation unit; wherein: Lyapunov index calculation unit: used to receive the linkage feature data output by the cross-domain signal correlation analysis module, and use the Lyapunov index to analyze the stability of the server operation status. The Lyapunov index is calculated by the following formula: ,in, represents the Lyapunov exponent, and Represent the system initial state and time respectively If the state changes when , indicating that the system is in an unstable state; if , indicating that the system is in a stable state; Strange attractor modeling unit: Based on the analysis results of the Lyapunov exponent calculation unit, the nonlinear characteristics of server operation are modeled using the strange attractor model. The server's time series data is mapped to a multidimensional space through phase space reconstruction technology. The attractor trajectory of the server system is formed by reconstructing the phase space to capture the dynamic evolution trajectory from stability to failure during server operation. Nonlinear behavior analysis unit: Analyzes the server operation model constructed by the strange attractor modeling unit to identify the nonlinear behavior of the server when it transitions from a stable state to a fault state, including bifurcation of the operation trajectory, period doubling, and chaos phenomena; Non-linear feature generation unit: Based on the results of the non-linear behavior analysis unit, non-linear feature data related to intermittent failures of the server is generated.

[0012] Optionally, the frequency domain feature extraction module includes a data input unit, a Fourier transform unit, a wavelet analysis unit and a frequency feature extraction unit; wherein: Data input unit: Receives the preprocessed signal data from the data preprocessing module; Fourier transform unit: Used to perform Fourier transform on the time-domain data received by the data input unit, converting the signal from the time domain to the frequency domain; Wavelet analysis unit: Based on the Fourier transform, uses wavelet analysis to perform multi-scale decomposition on the non-stationary signals during server operation to extract more detailed frequency features; Frequency feature extraction unit: Based on the frequency-domain signals extracted by the Fourier transform and wavelet analysis units, extracts the frequency features during server operation, including high-frequency noise, instantaneous electromagnetic interference, and abnormal frequency components in physical vibration.

[0013] Optionally, the intermittent fault prediction model includes a data fusion unit, a Bayesian inference unit, and a fault occurrence probability calculation unit; where: Data fusion unit: Receives the non-linear feature data output from the non-linear dynamics modeling module and the frequency feature data extracted by the frequency-domain feature extraction module, and performs data fusion to form a comprehensive feature vector of the server operation state; Bayesian inference unit: Based on the comprehensive feature vector provided by the data fusion unit, uses the Bayesian inference model to comprehensively evaluate the server health state; Fault occurrence probability calculation unit: Based on the output of the Bayesian inference unit, refines the occurrence probability of intermittent faults, specifically by performing weighted combination on different non-linear features and frequency features to generate the comprehensive probability of fault occurrence .

[0014] Optionally, the multi-level fault warning module includes a risk level classification unit and a warning signal generation unit; where: Risk level classification unit: Based on the fault occurrence probability , used to perform multi-level classification on the risk level of the server intermittent fault, specifically classified into three levels, namely low risk level, medium risk level, and high risk level; specifically when is within it is the low risk level; when is within it is the medium risk level; when is within it is the high risk level; Early warning signal generation unit: It is used to generate corresponding early warning signals according to the output results of the risk level division unit. Specifically, when the risk level is low risk, it generates a slight fluctuation record to record the historical data of the server status change, but does not trigger an alarm; when the risk level is medium risk, it generates a moderate intermittent anomaly warning to prompt the system administrator that there are potential problems with the server and recommends an inspection; when the risk level is high risk, it generates a severe fault early warning signal, requiring the administrator to take emergency measures immediately.

[0015] A method for warning of server failures in a computer is implemented by the above-mentioned server failure warning system for a computer, and includes the following steps: S1: Collect multi-dimensional operation signals from the hardware subsystem and external environment of the server, including signals such as CPU load, memory usage rate, disk read and write speed, network traffic, power input noise, fan speed fluctuation, and server physical vibration. S2: Perform data preprocessing on the collected time series signals, and the preprocessing includes data cleaning, denoising, and time alignment operations. S3: Based on the preprocessed data, use a multi-variable autoregressive model to analyze the time-domain correlation between different signals, and generate a signal correlation matrix, which is used to reveal the dynamic linkage characteristics between the server operation state and external environment signals. S4: Based on the linkage feature data obtained from cross-domain signal correlation analysis, use the Lyapunov exponent and strange attractor model to perform non-linear dynamics modeling on the server operation state, and extract non-linear features reflecting intermittent faults by identifying the non-linear behavior when the server transfers from a stable state to a fault state. S5: Perform Fourier transform and wavelet analysis on the preprocessed signal data, convert the time-domain data into frequency-domain data, and extract the frequency characteristics of high-frequency noise, transient electromagnetic interference, and physical vibration existing during the server operation. S6: Combine the feature data obtained from non-linear dynamics modeling and frequency-domain feature extraction, and use a Bayesian inference model to evaluate the health status of the server and calculate the probability of the server having an intermittent fault. S7: According to the occurrence probability of the intermittent fault, divide the fault risk into multiple levels and generate corresponding early warning signals.

[0016] Advantages of the present invention: In the present invention, by introducing the fusion acquisition technology of multi-source signals, multi-dimensional operation signals can be collected from server hardware and the external environment. Combining data preprocessing, cross-domain signal correlation analysis, non-linear dynamics modeling, and frequency-domain feature extraction technologies, comprehensive monitoring of the complex state of the server is achieved. Through these technologies, potential fault features during server operation can be deeply mined. Especially for intermittent faults, their early features can be effectively captured, and the occurrence of server faults can be predicted in advance, improving the accuracy and real-time performance of fault detection.

[0017] In the present invention, by combining a Bayesian inference model with various feature data, the probability of server fault occurrence is accurately calculated, and the risk level is refined based on the evaluation of the fault probability, so as to achieve multi-level fault early warning. This not only improves the accuracy of fault prediction during server operation but also reduces false alarms, ensuring timely response in high-risk states and enhancing the stability of the server and the ability to guarantee business continuity. BRIEF DESCRIPTION OF THE DRAWINGS

[0018] To more clearly illustrate the technical solutions in the present invention or the prior art, the following will briefly introduce the drawings required for use in the description of the embodiments or the prior art. Obviously, the drawings in the following description are only those of the present invention. For those of ordinary skill in the art, without creative efforts, other drawings can also be obtained based on these drawings.

[0019] Figure 1 It is a schematic diagram of the server fault early warning system according to an embodiment of the present invention; Figure 2 It is a schematic diagram of the server fault early warning method flow according to an embodiment of the present invention. DETAILED DESCRIPTION OF THE EMBODIMENTS

[0020] The present invention will be described in detail below in conjunction with the drawings and specific embodiments. At the same time, it should be noted here that in order to make the embodiments more detailed, the following embodiments are the best and preferred embodiments. For some well-known technologies, those skilled in the art can also adopt other alternative methods for implementation; and the drawings are only for more specific description of the embodiments and are not intended to specifically limit the present invention.

[0021] It should be pointed out that in the specification, when referring to "an embodiment", "embodiments", "exemplary embodiments", "some embodiments", etc., it indicates that the described embodiments may include specific features, structures, or characteristics, but not necessarily every embodiment includes such specific features, structures, or characteristics. In addition, when combining embodiments to describe specific features, structures, or characteristics, implementing such features, structures, or characteristics in combination with other embodiments (whether explicitly described or not) should be within the knowledge scope of those skilled in the relevant art.

[0022] Generally, terms can be understood at least in part from their use in context. For example, depending at least in part on the context, the term "one or more" as used herein can be used to describe any feature, structure, or property in the singular sense, or can be used to describe a combination of features, structures, or properties in the plural sense. Additionally, the term "based on" can be understood as not necessarily intended to convey an exclusive set of factors, but rather can alternatively, depending at least in part on the context, allow for the existence of other factors that are not necessarily explicitly described.

[0023] As Figure 1 shown, a server fault warning system for a computer includes a multi-source signal acquisition module, a data preprocessing module, a cross-domain signal correlation analysis module, a nonlinear dynamics modeling module, a frequency domain feature extraction module, an intermittent fault prediction model, and a multi-level fault warning module; wherein: Multi-source signal acquisition module: used to acquire multi-dimensional operation signals from the server hardware and environment, the operation signals including CPU load, memory usage rate, disk read / write rate, network traffic, power input noise, fan speed fluctuation, and server physical vibration, and output the acquired signals in the form of a time series; Data preprocessing module: receives the time series data from the multi-source signal acquisition module, and performs preprocessing operations such as data cleaning, denoising, and time alignment to ensure the accuracy and consistency of the data; Cross-domain signal correlation analysis module: based on the preprocessed data, uses a multivariate autoregressive model (MVAR) to analyze the time domain correlation between different signals, generates a signal correlation matrix, and is used to identify the dynamic linkage feature data between the server operation state and external environment signals; Nonlinear dynamics modeling module: based on the linkage feature data output by the cross-domain signal correlation analysis module, uses the Lyapunov exponent and the strange attractor model to model the operation state of the server, and is used to analyze the nonlinear behavior when the server transfers from a stable state to a fault state, so as to identify the critical state of the server operation and generate nonlinear feature data related to intermittent faults for fault prediction; Frequency domain feature extraction module: receives the preprocessed data from the data preprocessing module, and uses Fourier transform and wavelet analysis methods to convert the time domain data into frequency domain data to extract the frequency features during the server operation; Intermittent fault prediction model: combines the data output by the nonlinear dynamics modeling module and the frequency domain feature extraction module, comprehensively evaluates the health state of the server based on the Bayesian inference model, and calculates the occurrence probability of intermittent faults for use by the warning module; Multi-level Fault Warning Module: It is used to multi-level classify the risk level of intermittent faults according to the occurrence probability of faults, and generate corresponding warning signals. The warning signals are divided into three categories, including slight fluctuation records, moderate intermittent anomaly alarms, and severe intermittent fault warnings.

[0024] The multi-source signal acquisition module includes a CPU load acquisition unit, a memory usage acquisition unit, a disk read / write rate acquisition unit, a network traffic acquisition unit, a power input noise acquisition unit, a fan speed fluctuation acquisition unit, and a server physical vibration acquisition unit; among them: CPU Load Acquisition Unit: It is used to collect the CPU usage rate of the server in real time by reading the hardware performance monitoring interface of the server, and the collected data is recorded per second. Memory Usage Acquisition Unit: It obtains the current memory occupancy by accessing the memory management library of the server, and specifically records it at fixed time intervals. Disk Read / Write Rate Acquisition Unit: It is used to obtain the current disk read / write operation rate from the server storage device interface. The rate is stored in a time series manner and output with a timestamp marker. Network Traffic Acquisition Unit: It monitors the current network data transmission rate through the server network interface controller. The collected traffic data is output according to the traffic fluctuation per second to ensure the timing synchronization of the signals. Power Input Noise Acquisition Unit: It is used to obtain the high-frequency noise signal at the power input end of the server and regard it as an important signal of the operating state. The specific data is captured through frequency sampling and output in a time series. Fan Speed Fluctuation Acquisition Unit: It monitors the fan speed fluctuation in real time by reading the server fan controller interface, and the obtained speed data is stored in a time series manner. Server Physical Vibration Acquisition Unit: It uses a built-in vibration sensor to capture the physical vibration data of the server chassis and hardware components. The vibration data is recorded at millisecond-level time intervals and output in a time series form. The signals collected by the above units are all marked through the internal time synchronization mechanism and output to the data preprocessing module in a time series form to ensure the timing consistency of various signals, which is convenient for subsequent correlation analysis and fault detection.

[0025] The data preprocessing module includes a data cleaning unit, a denoising unit, and a time alignment unit; among them: Data Cleaning Unit: It is used to identify and eliminate the abnormal data and missing data collected by the multi-source signal acquisition module. By setting the abnormal value threshold range, it automatically filters out the sudden abnormal values in the signals, and linearly interpolates and fills in the missing data according to the historical data to ensure the integrity of the data. Denoising Unit: It is used to process the noise of the collected signals. By analyzing the frequency components in the signals, it removes high-frequency or low-frequency noise signals, especially the interference parts in the power input noise and physical vibration signals, thereby improving the purity of the effective signals. This unit performs multi-level filtering through a step-by-step iterative method to ensure the accuracy of the signals; Time Alignment Unit: It is used to align the collected signals in time series, mark different signals in the multi-source signal acquisition module with a unified time reference, and ensure that each signal can be compared and analyzed on the same time axis. The time alignment unit standardizes signals with different sampling frequencies through timestamp calibration technology to make them comparable; Through the collaborative work of the above units, after the data cleaning unit removes outliers and fills in data, it outputs to the denoising unit for noise filtering, and finally the time alignment unit ensures the timing consistency. The preprocessed data is output to the subsequent analysis module to ensure the accuracy and consistency of the data.

[0026] The cross-domain signal correlation analysis module includes a data input unit, a multivariate autoregressive analysis unit, a signal correlation matrix generation unit, and a dynamic linkage characteristic identification unit; among them: Data Input Unit: It receives the preprocessed multi-source signal data output by the data preprocessing module. The data includes operating signals such as CPU load, memory usage rate, disk read and write speed, network traffic, power input noise, fan speed fluctuation, and server physical vibration, and inputs all data into the multivariate autoregressive model according to the time series; Multivariate Autoregressive Analysis Unit: Based on the time series data provided by the data input unit, it adopts the multivariate autoregressive (MVAR) model. By modeling and analyzing the time changes of each signal, it comprehensively estimates the linkage effect between signals and generates an autoregressive coefficient matrix describing the dynamic correlation between signals; The expression of the multivariate autoregressive model is: , where, represents the signal vector collected at time , is the autoregressive coefficient matrix, indicating the time-domain correlation intensity between signals, is the regression order, is the error term. The multivariate autoregressive analysis unit solves the autoregressive equation to obtain the linkage effect of each signal on other signals, and then generates an autoregressive coefficient matrix. Each element in the matrix represents the influence degree and direction of a certain signal on other signals in the time series; Signal Correlation Matrix Generation Unit: Based on the autoregressive coefficient matrix generated by the multivariate autoregressive analysis unit, it generates a signal correlation matrix. Each element in the correlation matrix characterizes the time-domain influence intensity and direction of the corresponding signal on other signals, and is used to represent the dynamic relationship between the server operating state and external environment signals; Dynamic linkage characteristic recognition unit: By analyzing the signal correlation matrix, it recognizes the dynamic linkage characteristics between the server operating state and external environmental signals, including the impact of power input noise on CPU load and the linkage effect of physical vibration on fan speed, and generates linkage feature data for subsequent use by the fault prediction module. The above units work together. The data input unit receives multi-dimensional signals, the multi-variable autoregressive analysis unit analyzes the dynamic correlation between the signals, the signal correlation matrix generation unit generates the correlation matrix, and finally the dynamic linkage characteristic recognition unit recognizes the linkage characteristics between the server state and environmental signals and outputs the linkage feature data.

[0027] The signal correlation matrix generation unit specifically includes: Autoregressive coefficient matrix receiver: First, it receives the autoregressive coefficient matrix generated by the multi-variable autoregressive analysis unit. The autoregressive coefficient matrix is generated by the autoregressive model, and each element of the matrix represents the influence intensity of signal at time on signal at time . Calculation of the total correlation degree between signals: Using the received autoregressive coefficient matrix, by accumulating the coefficients of all regression orders, calculate the total correlation degree between signal and signal ; Its calculation formula is: ; where, represents the total correlation degree of signal on signal on signal , represents the corresponding element in the -th order regression coefficient matrix. By considering the influence of multi-order regression, the total correlation between signals is obtained; Calculation of signal correlation directionality: On the basis of calculating the total correlation degree, analyze the directionality of signal correlation. The directionality is determined by the sign of each regression coefficient, that is, a positive coefficient indicates a positive influence, and a negative coefficient indicates a negative influence; Calculate the correlation directionality of signal on signal The formula is: ; where, The function is used to judge the positive and negative correlation between signals. If it is positive, it indicates a positive correlation. If it is negative, it indicates a negative correlation; Construct a signal correlation matrix: Based on the calculation results of the total correlation degree and directionality, generate a signal correlation matrix , and each element of the matrix represents signal The time-domain influence intensity and direction on the signal ; the matrix elements are calculated by the following formula: , where represents the influence intensity of the signal on the signal , represents the influence direction, and each element of the matrix fully characterizes the dynamic correlation between signals; through the above steps, by constructing a signal correlation matrix and considering multiple regression orders, the dynamic relationship between internal and external signals of the server can be comprehensively quantified, and the time-domain correlation intensity and directionality between signals can be clearly displayed, providing highly reliable and accurate correlation data for subsequent fault prediction, and significantly improving the analysis accuracy and dynamic response ability of the fault warning system.

[0028] The nonlinear dynamics modeling module includes a Lyapunov exponent calculation unit, a strange attractor modeling unit, a nonlinear behavior analysis unit, and a nonlinear feature generation unit; among them: Lyapunov exponent calculation unit: used to receive the linkage feature data output by the cross-domain signal correlation analysis module, and use the Lyapunov exponent to analyze the stability of the server operation state. The Lyapunov exponent is calculated by the following formula: , where represents the Lyapunov exponent, and represent the initial state of the system and the state change at time respectively. If , it indicates that the system is in an unstable state; if , it indicates that the system is in a stable state; Strange attractor modeling unit: Based on the analysis results of the Lyapunov exponent calculation unit, use the strange attractor model to model the nonlinear characteristics of the server operation, map the time series data of the server to a multi-dimensional space through phase space reconstruction technology, and define the mapping function: , where represents the signal vector collected at time , is the data value at time , is the time delay, is the embedding dimension, and by reconstructing the phase space, the attractor trajectory of the server system is formed to capture the dynamic evolution trajectory from stable to faulty during the server operation. The strange attractor reflects the nonlinear change of the system state in the phase space. Especially when chaos or bifurcation phenomena occur, the shape of the attractor will change significantly; Nonlinear Behavior Analysis Unit: Analyze the server operation model constructed by the Singular Attractor Modeling Unit, identify the nonlinear behaviors when the server transfers from a stable state to a fault state, including bifurcations, period doublings, and chaotic phenomena in the operation trajectories. This unit identifies the critical state according to the changing trend of the server operation state and predicts the upcoming anomalies of the system. Nonlinear Feature Generation Unit: Based on the results of the Nonlinear Behavior Analysis Unit, generate nonlinear feature data related to the intermittent faults of the server. This unit generates feature data reflecting the server health state by analyzing the dynamic changes of Lyapunov exponents and singular attractors of the server under different operation states and uses it for fault prediction. By introducing the singular attractor model and combining with the phase space reconstruction technology, the system can capture the dynamic nonlinear behaviors of the server under different operation states. Especially when the system is about to enter the fault state, the significant changes in the singular attractor morphology provide an important basis for prediction, thus effectively improving the identification accuracy and predictability of intermittent faults.

[0029] The frequency domain feature extraction module includes a data input unit, a Fourier transform unit, a wavelet analysis unit, and a frequency feature extraction unit; where: Data Input Unit: Receive the preprocessed signal data from the data preprocessing module. The signal data is the time-domain data of each operation state of the server, including CPU load, memory usage rate, disk read / write speed, network traffic, power input noise, fan speed fluctuation, and physical vibration, etc. Fourier Transform Unit: Used to perform Fourier transform on the time-domain data received by the data input unit, convert the signal from the time domain to the frequency domain, and complete the extraction of frequency components using the following formula: , where, is the signal in the frequency domain, is the time-domain signal, is the frequency. The Fourier transform extracts the main frequency features that appear during the server operation by analyzing the spectral components of the signal, such as high-frequency noise, instantaneous electromagnetic interference, etc. Wavelet Analysis Unit: On the basis of the Fourier transform, use wavelet analysis to perform multi-scale decomposition on the non-stationary signals during the server operation to extract more detailed frequency features. The wavelet transform formula is: , where, is the wavelet coefficient, is the scale parameter, is the translation parameter, is the mother wavelet function. This unit can effectively capture the transient vibrations and local signal anomalies during the server operation, such as the abnormal frequency components in power fluctuations and physical vibrations, through the multi-scale decomposition of the signal. Frequency feature extraction unit: Based on the frequency-domain signals extracted by the Fourier transform and wavelet analysis unit, it extracts the frequency features during the operation of the server, including high-frequency noise, instantaneous electromagnetic interference, and abnormal frequency components in physical vibration, providing basic data for subsequent fault prediction; Through the collaborative work of the above units, the data input unit transfers the time-domain signal to the Fourier transform unit for preliminary frequency-domain conversion, and then the wavelet analysis unit performs multi-scale processing on the non-stationary signal. Finally, the frequency feature extraction unit extracts the key frequency features of the server and provides them for the subsequent analysis module to use.

[0030] The intermittent fault prediction model includes a data fusion unit, a Bayesian inference unit, and a fault occurrence probability calculation unit; Among them: Data fusion unit: Receives the non-linear feature data output from the non-linear dynamics modeling module and the frequency feature data extracted by the frequency-domain feature extraction module, and performs data fusion to form a comprehensive feature vector of the server operation state. The feature vector includes the dynamic change information of the server health state and its frequency-domain representation; Bayesian inference unit: Based on the comprehensive feature vector provided by the data fusion unit, it uses the Bayesian inference model to comprehensively evaluate the server health state. Bayesian inference is carried out through the following formula: , where, represents the posterior probability of server failure under the condition of observing the evidence (i.e., the non-linear features and frequency features of the server), is the prior probability of server failure, represents the probability of observing this evidence under the condition of server failure, is the total probability of the evidence appearing. This unit calculates the risk value of server failure by continuously updating the prior and posterior probabilities; Fault occurrence probability calculation unit: Based on the output of the Bayesian inference unit, it refines the occurrence probability of intermittent faults. Specifically, by performing weighted combination on different non-linear features and frequency features, it generates the comprehensive probability of fault occurrence , and the calculation formula is: , where, is the occurrence probability of intermittent faults, is the posterior probability of failure for the type of feature, The weight values for different features reflect the impact degree of these features on the occurrence of faults. By weighting different features, the probability of fault occurrence can more accurately reflect the current health status of the server. By combining the data from the non-linear dynamics modeling and the frequency domain feature extraction module, the intermittent fault prediction model can use the Bayesian inference method to accurately evaluate the health status of the server, and calculate the occurrence probability of intermittent faults by weighted combination of different feature data, ensuring more accurate and timely prediction of the intermittent faults of the server, thus enhancing the reliability and real-time performance of the overall fault warning system.

[0031] The multi-level fault warning module includes a risk level classification unit and a warning signal generation unit; among them: Risk level classification unit: Based on the probability of fault occurrence is used to perform multi-level classification on the risk level of the intermittent faults of the server, specifically divided into three levels, namely low risk level, medium risk level and high risk level; specifically when is within it is the low risk level, indicating that the server is in a normal or slightly fluctuating state with low risk; when is within it is the medium risk level, indicating that the server has abnormal features but has not reached the severe fault state; when is within it is the high risk level, indicating that the server is in a high risk state and is very likely to have intermittent faults about to occur; Warning signal generation unit: is used to generate corresponding warning signals according to the output results of the risk level classification unit. Specifically, when the risk level is low risk, generate a slight fluctuation record to record the historical data of the server status change, but do not trigger an alarm, only for subsequent reference; when the risk level is medium risk, generate a medium intermittent anomaly warning to prompt the system administrator that there are potential problems with the server, and it is recommended to check, and generate a log for subsequent analysis; when the risk level is high risk, generate a severe fault warning signal, requiring the administrator to take emergency treatment measures immediately to avoid business interruption or loss caused by server faults. By the above units outputting corresponding warning signals according to different risk levels, it is ensured that appropriate warning measures are taken under different risk situations.

[0032] As Figure 2 shown, a method for warning server faults for a computer is implemented by the above-mentioned warning system for server faults for a computer, and includes the following steps: S1: Collect multi-dimensional operation signals from the hardware subsystem and the external environment of the server, including signals such as CPU load, memory usage rate, disk read and write speed, network traffic, power input noise, fan speed fluctuation and server physical vibration, and record these signals in the form of a time series; S2: Perform data preprocessing on the collected time-series signals. The preprocessing includes data cleaning, denoising, and time alignment operations to eliminate abnormal data, filter out noise, and ensure that the signals are analyzed on a unified time axis; S3: Based on the preprocessed data, use a multivariate autoregressive model (MVAR) to analyze the time-domain correlation between different signals, generate a signal correlation matrix. The matrix is used to reveal the dynamic linkage characteristics between the server operating state and external environmental signals, and extract potential fault features; S4: Based on the linkage feature data obtained from cross-domain signal correlation analysis, use the Lyapunov exponent and strange attractor model to perform nonlinear dynamics modeling on the server operating state. By identifying the nonlinear behavior when the server transfers from a stable state to a fault state, extract the nonlinear features reflecting intermittent faults; S5: Perform Fourier transform and wavelet analysis on the preprocessed signal data, convert the time-domain data into frequency-domain data, and extract the frequency features of high-frequency noise, transient electromagnetic interference, and physical vibration existing during the server operation; S6: Combine the feature data obtained from nonlinear dynamics modeling and frequency-domain feature extraction, and use a Bayesian inference model to evaluate the server health state and calculate the probability of the server having intermittent faults; S7: According to the occurrence probability of intermittent faults, perform multi-level division of the fault risk and generate corresponding warning signals.

[0033] The present invention covers any substitutions, modifications, equivalent methods, and solutions made within the spirit and scope of the present invention. To enable the public to have a thorough understanding of the present invention, specific details are described in detail in the following preferred embodiments of the present invention. However, those skilled in the art can fully understand the present invention without these detailed descriptions. In addition, well-known methods, processes, procedures, components, and circuits are not described in detail to avoid unnecessary confusion to the essence of the present invention.

[0034] The above are only the preferred embodiments of the present invention. It should be noted that for those of ordinary skill in the art, without departing from the principle of the present invention, several improvements and refinements can be made, and these improvements and refinements should also be regarded as the protection scope of the present invention.

Claims

1. A server fault warning system for a computer, characterized in that, It includes a multi-source signal acquisition module, a data preprocessing module, a cross-domain signal correlation analysis module, a nonlinear dynamics modeling module, a frequency-domain feature extraction module, an intermittent fault prediction model, and a multi-level fault warning module; among which: The multi-source signal acquisition module: It is used to collect multi-dimensional operation signals from the server hardware and environment. The operation signals include CPU load, memory usage rate, disk read and write rate, network traffic, power input noise, fan speed fluctuation, and server physical vibration, and output the collected signals in the form of time series; The data preprocessing module: Receives the time series data of the multi-source signal acquisition module and performs preprocessing operations such as data cleaning, denoising, and time alignment; The cross-domain signal correlation analysis module: Based on the preprocessed data, uses a multi-variable autoregressive model to analyze the time-domain correlation between different signals, generates a signal correlation matrix, and is used to identify the dynamic linkage feature data between the server operation state and external environment signals; The nonlinear dynamics modeling module: Based on the linkage feature data output by the cross-domain signal correlation analysis module, uses the Lyapunov exponent and the strange attractor model to model the operation state of the server, and is used to analyze the nonlinear behavior when the server transfers from a stable state to a fault state, so as to identify the critical state of the server operation and generate nonlinear feature data related to intermittent faults; The frequency-domain feature extraction module: Receives the preprocessed data of the data preprocessing module, and uses Fourier transform and wavelet analysis methods to convert the time-domain data into frequency-domain data to extract the frequency features during the server operation; The frequency-domain feature extraction module includes a data input unit, a Fourier transform unit, a wavelet analysis unit, and a frequency feature extraction unit; among which: The data input unit: Receives the preprocessed signal data from the data preprocessing module; The Fourier transform unit: It is used to perform Fourier transform on the time-domain data received by the data input unit, and convert the signal from the time domain to the frequency domain; The wavelet analysis unit: On the basis of Fourier transform, uses wavelet analysis to perform multi-scale decomposition on the non-stationary signals during the server operation to extract more detailed frequency features; The frequency feature extraction unit: Based on the frequency-domain signals extracted by the Fourier transform and wavelet analysis units, extracts the frequency features during the server operation, including high-frequency noise, instantaneous electromagnetic interference, and abnormal frequency components in physical vibration; The intermittent fault prediction model: Combines the data output by the nonlinear dynamics modeling module and the frequency-domain feature extraction module, and comprehensively evaluates the health state of the server based on the Bayesian inference model, and calculates the occurrence probability of intermittent faults; The intermittent fault prediction model includes a data fusion unit, a Bayesian inference unit, and a fault occurrence probability calculation unit; among which: The data fusion unit: Receives the nonlinear feature data output by the nonlinear dynamics modeling module and the frequency feature data extracted by the frequency-domain feature extraction module, and performs data fusion to form a comprehensive feature vector of the server operation state; The Bayesian inference unit: Based on the comprehensive feature vector provided by the data fusion unit, uses the Bayesian inference model to comprehensively evaluate the health state of the server; Fault occurrence probability calculation unit: Based on the output of the Bayesian inference unit, refine the occurrence probability of intermittent faults, specifically by performing weighted combination on different non-linear features and frequency features to generate the comprehensive probability of fault occurrence ; Multi-level fault warning module: used to divide the risk level of intermittent faults into multiple levels according to the probability of fault occurrence, and generate corresponding warning signals.

2. The computer server fault warning system according to claim 1, characterized in that, The multi-source signal acquisition module includes a CPU load acquisition unit, a memory usage acquisition unit, a disk read and write rate acquisition unit, a network traffic acquisition unit, a power input noise acquisition unit, a fan speed fluctuation acquisition unit, and a server physical vibration acquisition unit; wherein: CPU load collection unit: used to collect the server's CPU usage in real time by reading the server's hardware performance monitoring interface. The collected data is recorded in units of one second. Memory usage collection unit: obtains the current memory usage by accessing the server's memory management library and records it at fixed time intervals; Disk read / write rate acquisition unit: used to obtain the current disk read / write operation rate from the server storage device interface. The rate is stored in a time series manner and output with a timestamp. Network traffic collection unit: monitors the current network data transmission rate through the server network interface controller, and outputs the collected traffic data according to the traffic fluctuation situation per second; Power input noise acquisition unit: used to obtain high-frequency noise signals at the power input end of the server and use them as important signals of the operating status. Specific data is captured through frequency sampling and output in a time series. Fan speed fluctuation collection unit: monitors fan speed fluctuations in real time by reading the server fan controller interface, and stores the acquired speed data in a time series format; Server physical vibration collection unit: uses built-in vibration sensors to capture physical vibration data of server chassis and hardware components. The vibration data is recorded at millisecond time intervals and output in the form of time series.

3. The server fault warning system for a computer according to claim 1, wherein, The data preprocessing module includes a data cleaning unit, a denoising unit and a time alignment unit; wherein: Data cleaning unit: used to identify and remove abnormal data and missing data collected in the multi-source signal acquisition module. By setting the abnormal value threshold range, it automatically filters out sudden abnormal values in the signal, and linearly interpolates and completes the missing data based on historical data to ensure data integrity. De-noising unit: used to process the noise of the collected signal, by analyzing the frequency components in the signal, removing high-frequency or low-frequency noise signals, especially the interference part in the power input noise and physical vibration signal, so as to improve the purity of the effective signal; Time alignment unit: used to align the time series of the collected signals and mark the different signals in the multi-source signal acquisition module with a unified time reference.

4. A computer server fault warning system according to claim 1, characterized in that, The cross-domain signal correlation analysis module includes a data input unit, a multivariate autoregression analysis unit, a signal correlation matrix generation unit, and a dynamic linkage characteristic identification unit; wherein: Data input unit: receiving the preprocessed multi-source signal data output by the data preprocessing module, and inputting all the data into the multivariate autoregressive model according to the time series; Multivariate autoregressive analysis unit: Based on the time series data provided by the data input unit, using a multivariate autoregressive model, by modeling and analyzing the time changes of each signal, comprehensively estimating the linkage effect between signals, and generating an autoregressive coefficient matrix describing the dynamic association between signals; Signal correlation matrix generation unit: Based on the autoregressive coefficient matrix generated by the multivariate autoregressive analysis unit, generate a signal correlation matrix, and each element in the correlation matrix represents the intensity and direction of the time-domain influence of the corresponding signal on other signals, and is used to represent the dynamic relationship between the server operating state and external environment signals; Dynamic linkage characteristic identification unit: By analyzing the signal correlation matrix, identify the dynamic linkage characteristics between the server operating state and external environment signals, including the influence of power input noise on CPU load and the linkage effect of physical vibration on fan speed, and generate linkage characteristic data.

5. The computer server fault warning system according to claim 4, characterized in that, The signal correlation matrix generation unit specifically includes: Receiving autoregressive coefficient matrix: First, receive the autoregressive coefficient matrix generated by the multivariate autoregressive analysis unit, and the autoregressive coefficient matrix is generated by an autoregressive model; Calculate the total correlation degree between signals: Using the received autoregressive coefficient matrix, calculate the total correlation degree of signal and signal by accumulating the coefficients of all regression orders ; Calculating signal correlation directionality: On the basis of calculating the total correlation degree, analyze the directionality of signal correlation, and the directionality is determined by the sign of each regression coefficient, that is, a positive coefficient represents a positive influence, and a negative coefficient represents a negative influence; Construct a signal correlation matrix: Based on the calculation results of the total correlation degree and directivity, generate a signal correlation matrix , each element of the matrix represents the signal 's time-domain influence intensity and direction on the signal .

6. The computer server fault warning system according to claim 1, wherein, The non-linear dynamics modeling module includes a Lyapunov exponent calculation unit, a strange attractor modeling unit, a non-linear behavior analysis unit, and a non-linear feature generation unit; among them: Lyapunov exponent calculation unit: It is used to receive the linkage feature data output by the cross-domain signal correlation analysis module and analyze the stability of the server operation state using the Lyapunov exponent. The Lyapunov exponent is calculated by the following formula: , where represents the Lyapunov exponent, and represent the initial state of the system and the state change at time respectively; if , it indicates that the system is in an unstable state; if , it indicates that the system is in a stable state; Strange attractor modeling unit: Based on the analysis result of the Lyapunov exponent calculation unit, use a strange attractor model to model the non-linear characteristics of server operation, map the time series data of the server to a multi-dimensional space through phase space reconstruction technology, and form an attractor trajectory of the server system by reconstructing the phase space, which is used to capture the dynamic evolution trajectory from stable to faulty during server operation; Non-linear behavior analysis unit: Analyze the server operation model constructed by the strange attractor modeling unit, and identify the non-linear behaviors when the server transfers from a stable state to a faulty state, including bifurcation, period doubling, and chaotic phenomena of the operation trajectory; Non-linear feature generation unit: Based on the result of the non-linear behavior analysis unit, generate non-linear feature data related to server intermittent faults.

7. A computer server fault warning system according to claim 1, characterized in that, The multi-level fault warning module includes a risk level division unit and a warning signal generation unit; among them: Risk level division unit: Based on the probability of failure occurrence , which is used to conduct multi-level division of the risk level of server intermittent failures. Specifically, it is divided into three levels, namely low risk level, medium risk level and high risk level; specifically, when is within , it is a low risk level; when is within , it is a medium risk level; when is within , it is a high risk level; Warning signal generation unit: Used to generate corresponding warning signals according to the output result of the risk level division unit. Specifically, when the risk level is low risk, generate a slight fluctuation record to record the historical data of server state changes, but do not trigger an alarm; when the risk level is medium risk, generate a medium intermittent anomaly warning to prompt the system administrator that there are potential problems with the server and recommend an inspection; when the risk level is high risk, generate a severe fault warning signal, requiring the administrator to take emergency measures immediately.

8. A server fault warning method for a computer, implemented by a server fault warning system for a computer according to any one of claims 1-7, characterized in that, Including the following steps: S1: Collect multi-dimensional operation signals from the server's hardware subsystem and external environment, including signals of CPU load, memory usage rate, disk read / write speed, network traffic, power input noise, fan speed fluctuation, and server physical vibration; S2: Perform data preprocessing on the collected time series signals, and the preprocessing includes data cleaning, denoising, and time alignment operations; S3: Based on the preprocessed data, use a multi-variable autoregressive model to analyze the time-domain correlation between different signals, and generate a signal correlation matrix, which is used to reveal the dynamic linkage characteristics between the server operation state and external environment signals; S4: Based on the linkage feature data obtained from cross-domain signal correlation analysis, use the Lyapunov exponent and strange attractor model to perform non-linear dynamics modeling on the server operation state, and extract non-linear features reflecting intermittent faults by identifying the non-linear behavior when the server transfers from a stable state to a fault state; S5: Perform Fourier transform and wavelet analysis on the preprocessed signal data, convert the time-domain data into frequency-domain data, and extract the frequency characteristics of high-frequency noise, transient electromagnetic interference, and physical vibration existing during the server operation; S6: Combine the feature data obtained from non-linear dynamics modeling and frequency-domain feature extraction, and use a Bayesian inference model to evaluate the server health state and calculate the probability of the server having intermittent faults; S7: According to the occurrence probability of intermittent faults, perform multi-level partitioning on the fault risk and generate corresponding warning signals.

Images