Server model and fault identification method and system

Abstract

The invention discloses a server model and fault identification method and system, and relates to the technical field of machine learning, and the method comprises the steps: collecting a power signal, a heat dissipation system signal and an electromagnetic radiation signal of a server through a signal sensor; filtering and denoising preprocessing operation is carried out on an original digital signal, the signal quality is improved, and signal enhancement processing is carried out; extracting signal features from the preprocessed signals by using a time domain analysis signal processing technology; using the training set to train the constructed product fault identification model; verifying the model by using the verification set, and adjusting hyper-parameters of the model based on a verification result; and evaluating the trained model by using the test set. The model identification module, the fault identification module and the model application module are arranged, so that the identification efficiency and accuracy are improved, diversified requirements of server models can be met, and intelligent and automatic fault identification is realized.

Description

Technical Field

[0001] The present invention relates to the technical field of machine learning, and in particular to a server model and fault identification method and system. Background Art

[0002] With the rapid development of information technology, the demand for servers in data centers and cloud computing services is growing. As the core equipment for data storage and processing, the stability and reliability of servers are crucial. However, servers may encounter various faults during operation, such as hardware damage, overheating, connection problems, etc. If these faults are not discovered and handled in time, they may lead to serious consequences such as data loss and service interruption.

[0003] The existing manual identification method requires professionals to check the servers one by one, which is not only inefficient but also easily affected by human factors, resulting in low identification accuracy. Although the monitoring method based on hardware sensors can monitor certain status parameters of the server in real time, it cannot fully cover all possible fault types. With the continuous advancement and upgrading of server technology, there are more and more server models, each of which may have different signal characteristics and fault manifestations. Traditional identification methods are difficult to adapt to such diverse needs. Summary of the invention

[0004] In order to solve the above technical problems, a server model and fault identification method and system are provided. This technical solution solves the problems raised in the above background technology.

[0005] In order to achieve the above purpose, the technical solution adopted by the present invention is: A server model and fault identification method, comprising: The power supply signal, cooling system signal and electromagnetic radiation signal of the server are collected respectively through signal sensors to obtain original digital signals; Perform pre-processing operations such as filtering and denoising on the original digital signal to improve the signal quality and perform signal enhancement processing; The signal features are extracted from the preprocessed signal using time domain analysis signal processing technology, and the extracted signal features are amplified and output as product feature vectors. The signal features include peak value, mean value, variance, spectral component and energy distribution. Identify server models based on differences in product feature vectors; Obtain historical failure data containing different server models, annotate the historical failure data with product model information, and divide the historical failure data into training set, validation set, and test set; The constructed product fault recognition model is trained using the training set. The mapping relationship between product feature vectors and product faults is learned by adjusting the weights and bias parameters of the convolution kernel in the model. During the training process, the model is validated using the validation set, and the hyperparameters of the model are adjusted based on the validation results, including the learning rate, the number of layers, and the number of neurons; Use the test set to evaluate the trained model, and evaluate the comprehensive performance indicators of the model by calculating the accuracy and recall rate of the model; The extracted product feature vector is input into the product fault identification model, and based on the server product model identification, the mapping relationship between the product feature vector and the product fault is obtained, and the identified server product fault is output.

[0006] Preferably, the preprocessing operation of filtering and denoising the original digital signal to improve the signal quality and perform signal enhancement processing specifically includes: Based on the signal characteristics and noise distribution of the original digital signal, a bandpass filter is designed for filtering and denoising; Set the center frequency of the bandpass filter to the frequency point where the maximum frequency density in the signal is located, and set the bandwidth of the bandpass filter to the absolute value of the difference between the frequency corresponding to the frequency point where the maximum frequency density in the signal is located and the maximum frequency; The original digital signal is input into a bandpass filter, which filters the signal based on set parameters to remove noise components that are not within the set frequency range; The quality of the filtered signal is evaluated by calculating the ratio index of the available signal to the noise signal in the filtered signal, and the bandwidth parameter of the bandpass filter is adjusted by feedback; The filtered signal is enhanced to improve the signal amplitude and contrast.

[0007] Preferably, the method of extracting signal features from the preprocessed signal using time domain analysis signal processing technology, amplifying the extracted signal features, and outputting the product feature vector specifically includes: By scanning the signal, the positive peak value and the negative peak value in the preprocessed signal are obtained, and the average value and variance of the signal are calculated; Determine whether the signal type is a continuous signal. If so, use Fourier integral transform; if not, use discrete Fourier transform; Perform Fourier transform. For continuous signals, use the Fourier integral transform formula to calculate and obtain the continuous signal frequency domain transformed from the signal time domain; Perform Fourier transform, and for discrete signals, preset the number of signal samples for discrete Fourier transform; The discrete Fourier transform formula is used to perform weighted summation on the discrete sampling points of the signal to calculate the discrete signal frequency domain transformed from the signal time domain; In the signal frequency domain, analyzing the spectrum components of the signal, the spectrum components include a primary frequency, a secondary frequency and a spectrum bandwidth; Calculate the amplitude square signal of each frequency component in the spectrum and output it as the energy distribution of the signal at each frequency; Assign weights to each signal feature based on its importance in the product feature vector; Based on the weights, a weighted matrix is ​​constructed, each feature is input into each row of the weighted matrix, each sample is input into each column of the weighted matrix, and the weights corresponding to the features and samples are input into the elements in the weighted matrix; Multiply the feature matrix of the original digital signal by the weighting matrix to obtain a weighted feature matrix; Linearly amplify the weighted feature matrix, and combine the amplified feature matrix with other unweighted signal components to form a processed signal; Combine the amplified features to form and output a product feature vector; The Fourier integral transform formula is: , In the formula, is a continuous time signal, is a continuous signal in the frequency domain, is the angular frequency of the signal, is an imaginary unit, is the time parameter; The discrete Fourier transform formula is: , In the formula, is a discrete-time signal, is the discrete signal frequency domain, is the number of signal samples, is the frequency index, is the number of the discrete-time signal.

[0008] Preferably, the step of using the training set to train the constructed product fault identification model and adjusting the weight and bias parameters of the convolution kernel in the model to learn the mapping relationship between the product feature vector and the product fault specifically includes: S101: Randomly select initial values ​​from Gaussian distribution, initialize the weight and bias parameters of the convolution kernel, and build an initial product fault recognition model; S102: Input the samples in the training set into the product fault recognition model, perform forward propagation calculation through the convolution layer, and obtain the prediction result of the model for each sample; S103: Convert the actual labels in the training set into a one-hot encoding form. The label of each sample is represented as a vector with the same number of fault categories, where the position of the correct category is 1 and the other positions are 0; S104: Calculate the cross entropy loss between the prediction result of the product fault recognition model and the actual label using the cross entropy loss function formula; S105: Determine whether the cross entropy loss is higher than a preset threshold. If so, use the back propagation method to calculate the gradient of the cross entropy loss function with respect to the model parameters, update the model parameters based on the calculated gradient, and return to step S102. If not, output the current product fault identification model. The cross entropy loss function formula is: , In the formula, is the cross entropy loss between the prediction results of the product fault recognition model and the actual label, is the number of samples in the training set, is the actual label value of the i-th sample, The probability that the i-th sample predicted by the model is a positive class.

[0009] Preferably, during the training process, the model is verified using a verification set, and the hyperparameters of the model are adjusted based on the verification results, specifically including: In the preset model hyperparameter space, a grid search method is used to obtain a combination of at least two hyperparameters; The product fault identification model is verified using the verification set, and based on the verification results, the learning rate, number of layers, and number of neurons are gradually adjusted; Determine whether the verification loss decrease rate is lower than the preset speed threshold. If so, increase the learning rate. If not, do not output. Determine whether the cross entropy loss of the product fault identification model in the validation set is higher than the preset threshold. If so, increase the number of model layers to enhance the model's expressiveness. If not, reduce the number of model layers to reduce the computational cost and improve the generalization ability. When each layer of the model is overfitting, the number of neurons in this layer is reduced; when each layer of the model is underfitting, the number of neurons in this layer is increased.

[0010] Preferably, the trained model is evaluated using a test set, and the comprehensive performance indicators of the model are evaluated by calculating the accuracy and recall of the model. Specifically, the comprehensive performance indicators of the model include: Get the actual label of each sample in the test set; Based on the prediction results and actual labels, calculate the number of true positives, false positives, true negatives, and false negatives for each sample in the test set; The accuracy of the product fault identification model is obtained by calculating the ratio of the sum of the number of true positive examples and true negative examples to the total number of test set samples; The recall rate of the product fault recognition model is obtained by calculating the ratio of the number of true positive examples to the sum of the number of true positive examples and false negative examples. Based on the accuracy and recall rate of the product fault identification model, the performance comprehensive index formula is used to evaluate the model's comprehensive performance index; The performance comprehensive index formula is: , In the formula, is the comprehensive performance index of the product fault identification model. is the accuracy of the model, is the recall rate of the model.

[0011] Furthermore, a server model and fault identification system is proposed, which is used to implement the server model and fault identification method as described above, including: A model identification module, which is used to collect the power supply signal, cooling system signal and electromagnetic radiation signal of the server through a signal sensor, obtain the original digital signal, perform preprocessing operations of filtering and denoising the original digital signal to improve the signal quality, perform signal enhancement processing, extract signal features from the preprocessed signal using time domain analysis signal processing technology, amplify the extracted signal features, output them as product feature vectors, and identify the server model based on the difference in product feature vectors; A fault identification module, which is used to obtain historical fault data containing different server models, and annotate the historical fault data with product model information, divide the historical fault data into a training set, a validation set, and a test set, use the training set to train the constructed product fault identification model, and learn the mapping relationship between product feature vectors and product faults by adjusting the weights and bias parameters of the convolution kernel in the model. During the training process, the model is verified using the validation set, and the hyperparameters of the model are adjusted based on the verification results. The trained model is evaluated using the test set, and the comprehensive performance indicators of the model are evaluated by calculating the accuracy and recall rate of the model; The model application module is used to input the extracted product feature vector into the product fault identification model, obtain the mapping relationship between the product feature vector and the product fault based on the server product model identification, and output the identified server product fault.

[0012] Compared with the prior art, the present invention has the following beneficial effects: By setting up model identification module, fault identification module and model application module, the efficiency and accuracy of identification are improved. It can also adapt to the diverse needs of server models and realize intelligent and automatic fault identification. It can also monitor the status of the server in real time, promptly discover and warn of potential faults, and provide strong technical support for the operation and maintenance management of the data center, meeting the needs of modern data centers for efficient operation and maintenance. BRIEF DESCRIPTION OF THE DRAWINGS

[0013] Figure 1 A flow chart of the server model and fault identification method of the present invention; Figure 2 The present invention is a flow chart of a method for preprocessing the original digital signal by filtering and denoising, improving the signal quality, and performing signal enhancement processing; Figure 3 A flow chart of a method for extracting signal features from a preprocessed signal using a time domain analysis signal processing technique according to the present invention; Figure 4 A flow chart of a method for learning a mapping relationship between a product feature vector and a product fault according to the present invention; Figure 5 A flow chart of a method for adjusting the hyperparameters of a model based on verification results of the present invention; Figure 6 It is a flow chart of the performance comprehensive index method of the evaluation model of the present invention. DETAILED DESCRIPTION

[0014] The following description is used to disclose the present invention so that those skilled in the art can implement the present invention. The preferred embodiments described below are only examples, and those skilled in the art may think of other obvious variations.

[0015] Reference Figure 1 As shown, a server model and a fault identification method include: The power supply signal, cooling system signal and electromagnetic radiation signal of the server are collected respectively through signal sensors to obtain original digital signals; Perform pre-processing operations such as filtering and denoising on the original digital signal to improve the signal quality and perform signal enhancement processing; The signal features are extracted from the preprocessed signal using time domain analysis signal processing technology, and the extracted signal features are amplified and output as product feature vectors. The signal features include peak value, mean value, variance, spectral component and energy distribution. Identify server models based on differences in product feature vectors; Obtain historical failure data containing different server models, annotate the historical failure data with product model information, and divide the historical failure data into training set, validation set, and test set; The constructed product fault recognition model is trained using the training set. The mapping relationship between product feature vectors and product faults is learned by adjusting the weights and bias parameters of the convolution kernel in the model. During the training process, the model is validated using the validation set, and the hyperparameters of the model are adjusted based on the validation results, including the learning rate, the number of layers, and the number of neurons; Use the test set to evaluate the trained model, and evaluate the comprehensive performance indicators of the model by calculating the accuracy and recall rate of the model; The extracted product feature vector is input into the product fault identification model, and based on the server product model identification, the mapping relationship between the product feature vector and the product fault is obtained, and the identified server product fault is output.

[0016] Reference Figure 2 As shown, the original digital signal is filtered and denoised for preprocessing to improve signal quality, and the signal enhancement process specifically includes: Based on the signal characteristics and noise distribution of the original digital signal, a bandpass filter is designed for filtering and denoising; Set the center frequency of the bandpass filter to the frequency point where the maximum frequency density in the signal is located, and set the bandwidth of the bandpass filter to the absolute value of the difference between the frequency corresponding to the frequency point where the maximum frequency density in the signal is located and the maximum frequency; The original digital signal is input into a bandpass filter, which filters the signal based on set parameters to remove noise components that are not within the set frequency range; The quality of the filtered signal is evaluated by calculating the ratio index of the available signal to the noise signal in the filtered signal, and the bandwidth parameter of the bandpass filter is adjusted by feedback; The filtered signal is enhanced to improve the signal amplitude and contrast.

[0017] Since the filter may cover outside the boundary of the signal, it is necessary to fill a circle of frequency points with a signal frequency of 0 outside the signal boundary (called zero filling) to ensure that the filter can completely cover all frequency points. The ratio of the available signal to the noise signal in the filtered signal is used to evaluate the quality of the filtered signal. This indicator is the signal-to-noise ratio.

[0018] Reference Figure 3 As shown, the signal features are extracted from the preprocessed signal using time domain analysis signal processing technology, and the extracted signal features are amplified and output as product feature vectors, which specifically include: By scanning the signal, the positive peak value and the negative peak value in the preprocessed signal are obtained, and the average value and variance of the signal are calculated; Determine whether the signal type is a continuous signal. If so, use Fourier integral transform; if not, use discrete Fourier transform; Perform Fourier transform. For continuous signals, use the Fourier integral transform formula to calculate and obtain the continuous signal frequency domain transformed from the signal time domain; Perform Fourier transform, and for discrete signals, preset the number of signal samples for discrete Fourier transform; The discrete Fourier transform formula is used to perform weighted summation on the discrete sampling points of the signal to calculate the discrete signal frequency domain transformed from the signal time domain; In the signal frequency domain, analyzing the spectrum components of the signal, the spectrum components include a primary frequency, a secondary frequency and a spectrum bandwidth; Calculate the amplitude square signal of each frequency component in the spectrum and output it as the energy distribution of the signal at each frequency; Assign weights to each signal feature based on its importance in the product feature vector; Based on the weights, a weighted matrix is ​​constructed, each feature is input into each row of the weighted matrix, each sample is input into each column of the weighted matrix, and the weights corresponding to the features and samples are input into the elements in the weighted matrix; Multiply the feature matrix of the original digital signal by the weighting matrix to obtain a weighted feature matrix; Linearly amplify the weighted feature matrix, and combine the amplified feature matrix with other unweighted signal components to form a processed signal; Combine the amplified features to form and output a product feature vector; The Fourier integral transform formula is: , In the formula, is a continuous time signal, is a continuous signal in the frequency domain, is the angular frequency of the signal, is an imaginary unit, is the time parameter; The discrete Fourier transform formula is: , In the formula, is a discrete-time signal, is the discrete signal frequency domain, is the number of signal samples, is the frequency index, is the number of the discrete-time signal.

[0019] When processing signals in practice, after sampling, only a series of discrete points can be obtained rather than a function with values ​​everywhere. If you want to use a computer to get actual numerical results, you can only calculate a series of discrete sampling points. In order to deal with this situation, you can use discrete Fourier transform. The frequency index in the discrete Fourier formula ranges from 0 to N-1, representing discrete frequency points in the frequency domain. The number of discrete time domain signals also ranges from 0 to N-1.

[0020] Reference Figure 4 As shown, the constructed product fault recognition model is trained using the training set. By adjusting the weights and bias parameters of the convolution kernel in the model, the mapping relationship between the product feature vector and the product fault is learned, including: S101: Randomly select initial values ​​from Gaussian distribution, initialize the weight and bias parameters of the convolution kernel, and build an initial product fault recognition model; S102: Input the samples in the training set into the product fault recognition model, perform forward propagation calculation through the convolution layer, and obtain the prediction result of the model for each sample; S103: Convert the actual labels in the training set into a one-hot encoding form. The label of each sample is represented as a vector with the same number of fault categories, where the position of the correct category is 1 and the other positions are 0; S104: Calculate the cross entropy loss between the prediction result of the product fault recognition model and the actual label using the cross entropy loss function formula; S105: Determine whether the cross entropy loss is higher than a preset threshold. If so, use the back propagation method to calculate the gradient of the cross entropy loss function with respect to the model parameters, update the model parameters based on the calculated gradient, and return to step S102. If not, output the current product fault identification model. The cross entropy loss function formula is: , In the formula, is the cross entropy loss between the prediction results of the product fault recognition model and the actual label, is the number of samples in the training set, is the actual label value of the i-th sample, The probability that the i-th sample predicted by the model is a positive class.

[0021] During the forward propagation process, the convolution layer applies the convolution kernel to slide on the input feature map, performs convolution operations, and extracts features. The pooling layer, such as maximum pooling or average pooling, is used to reduce the dimension of the feature map and reduce the amount of calculation. The fully connected layer maps the features to the fault category space.

[0022] Reference Figure 5As shown in the figure, during the training process, the model is verified using the verification set, and the hyperparameters of the model are adjusted based on the verification results, including: In the preset model hyperparameter space, a grid search method is used to obtain a combination of at least two hyperparameters; The product fault identification model is verified using the verification set, and based on the verification results, the learning rate, number of layers, and number of neurons are gradually adjusted; Determine whether the verification loss decrease rate is lower than the preset speed threshold. If so, increase the learning rate. If not, do not output. Determine whether the cross entropy loss of the product fault identification model in the validation set is higher than the preset threshold. If so, increase the number of model layers to enhance the model's expressiveness. If not, reduce the number of model layers to reduce the computational cost and improve the generalization ability. When each layer of the model is overfitting, the number of neurons in this layer is reduced; when each layer of the model is underfitting, the number of neurons in this layer is increased.

[0023] Grid search is an exhaustive search method that traverses all possible hyperparameter combinations and evaluates the performance of each combination through cross-validation. In this embodiment, the learning rate and number of layers can be first selected for grid search, the number of neurons is fixed to a certain initial value, and then the optimal learning rate and number of layers combination are determined based on the search results.

[0024] Reference Figure 6 As shown in the figure, the trained model is evaluated using the test set. By calculating the accuracy and recall of the model, the comprehensive performance indicators of the evaluation model include: Get the actual label of each sample in the test set; Based on the prediction results and actual labels, calculate the number of true positives, false positives, true negatives, and false negatives for each sample in the test set; The accuracy of the product fault identification model is obtained by calculating the ratio of the sum of the number of true positive examples and true negative examples to the total number of test set samples; The recall rate of the product fault recognition model is obtained by calculating the ratio of the number of true positive examples to the sum of the number of true positive examples and false negative examples. Based on the accuracy and recall rate of the product fault identification model, the performance comprehensive index formula is used to evaluate the model's comprehensive performance index; The performance comprehensive index formula is: , In the formula, is the comprehensive performance index of the product fault identification model. is the accuracy of the model, is the recall rate of the model.

[0025] True positives: the number of samples predicted as positive and actually positive; false positives: the number of samples predicted as positive but actually negative; true negatives: the number of samples predicted as negative and actually negative; false negatives: the number of samples predicted as negative but actually positive.

[0026] Further, based on the same inventive concept as the above-mentioned server model and fault identification method, the present solution also proposes a server model and fault identification system, including: A model identification module, which is used to collect the power supply signal, cooling system signal and electromagnetic radiation signal of the server through a signal sensor, obtain the original digital signal, perform preprocessing operations of filtering and denoising the original digital signal to improve the signal quality, perform signal enhancement processing, extract signal features from the preprocessed signal using time domain analysis signal processing technology, amplify the extracted signal features, output them as product feature vectors, and identify the server model based on the difference in product feature vectors; A fault identification module, which is used to obtain historical fault data containing different server models, and annotate the historical fault data with product model information, divide the historical fault data into a training set, a validation set, and a test set, use the training set to train the constructed product fault identification model, and learn the mapping relationship between product feature vectors and product faults by adjusting the weights and bias parameters of the convolution kernel in the model. During the training process, the model is verified using the validation set, and the hyperparameters of the model are adjusted based on the verification results. The trained model is evaluated using the test set, and the comprehensive performance indicators of the model are evaluated by calculating the accuracy and recall rate of the model; The model application module is used to input the extracted product feature vector into the product fault identification model, obtain the mapping relationship between the product feature vector and the product fault based on the server product model identification, and output the identified server product fault.

[0027] Furthermore, the present solution also proposes a computer-readable storage medium on which a computer-readable program is stored. When the computer-readable program is called, the above-mentioned server model and fault identification method is executed.

[0028] It is understandable that the storage medium may be a magnetic medium, such as a floppy disk, a hard disk, or a magnetic tape; an optical medium, such as a DVD; or a semiconductor medium, such as a solid state drive (SSD).

[0029] To sum up, the advantages of the present invention are: by setting up a model identification module, a fault identification module and a model application module, the efficiency and accuracy of identification are improved, and it can adapt to the diversified needs of server models, realize intelligent and automatic fault identification, and can also monitor the status of the server in real time, promptly discover and warn of potential faults, provide strong technical support for the operation and maintenance management of the data center, and meet the needs of modern data centers for efficient operation and maintenance.

[0030] The above shows and describes the basic principles, main features and advantages of the present invention. Those skilled in the art should understand that the present invention is not limited to the above embodiments. The above embodiments and descriptions only describe the principles of the present invention. The present invention may be subject to various changes and improvements without departing from the spirit and scope of the present invention. These changes and improvements fall within the scope of the present invention. The scope of protection claimed by the present invention is defined by the attached claims and their equivalents.

Claims

1. A server model and fault identification method, characterized in that: include: The power supply signal, cooling system signal and electromagnetic radiation signal of the server are collected respectively through signal sensors to obtain original digital signals; Perform pre-processing operations such as filtering and denoising on the original digital signal to improve the signal quality and perform signal enhancement processing; The signal features are extracted from the preprocessed signal using time domain analysis signal processing technology, and the extracted signal features are amplified and output as product feature vectors. The signal features include peak value, mean value, variance, spectral component and energy distribution. Identify server models based on differences in product feature vectors; Obtain historical failure data containing different server models, annotate the historical failure data with product model information, and divide the historical failure data into training set, validation set, and test set; The constructed product fault recognition model is trained using the training set. The mapping relationship between product feature vectors and product faults is learned by adjusting the weights and bias parameters of the convolution kernel in the model. During the training process, the model is validated using the validation set, and the hyperparameters of the model are adjusted based on the validation results, including the learning rate, the number of layers, and the number of neurons; Use the test set to evaluate the trained model, and evaluate the comprehensive performance indicators of the model by calculating the accuracy and recall rate of the model; The extracted product feature vector is input into the product fault identification model, and based on the server product model identification, the mapping relationship between the product feature vector and the product fault is obtained, and the identified server product fault is output.

2. A server model and fault identification method according to claim 1, characterized in that: The preprocessing operation of filtering and denoising the original digital signal to improve the signal quality and perform signal enhancement processing specifically includes: Based on the signal characteristics and noise distribution of the original digital signal, a bandpass filter is designed for filtering and denoising; Set the center frequency of the bandpass filter to the frequency point where the maximum frequency density in the signal is located, and set the bandwidth of the bandpass filter to the absolute value of the difference between the frequency corresponding to the frequency point where the maximum frequency density in the signal is located and the maximum frequency; The original digital signal is input into a bandpass filter, which filters the signal based on set parameters to remove noise components that are not within the set frequency range; The quality of the filtered signal is evaluated by calculating the ratio index of the available signal to the noise signal in the filtered signal, and the bandwidth parameter of the bandpass filter is adjusted by feedback; The filtered signal is enhanced to improve the signal amplitude and contrast.

3. A server model and fault identification method according to claim 2, characterized in that: The method of extracting signal features from the preprocessed signal using the time domain analysis signal processing technology, amplifying the extracted signal features, and outputting the product feature vector specifically includes: By scanning the signal, the positive peak value and the negative peak value in the preprocessed signal are obtained, and the average value and variance of the signal are calculated; Determine whether the signal type is a continuous signal. If so, use Fourier integral transform; if not, use discrete Fourier transform; Perform Fourier transform. For continuous signals, use the Fourier integral transform formula to calculate and obtain the continuous signal frequency domain transformed from the signal time domain; Perform Fourier transform, and for discrete signals, preset the number of signal samples for discrete Fourier transform; The discrete Fourier transform formula is used to perform weighted summation on the discrete sampling points of the signal to calculate the discrete signal frequency domain transformed from the signal time domain; In the signal frequency domain, analyzing the spectrum components of the signal, the spectrum components include a primary frequency, a secondary frequency and a spectrum bandwidth; Calculate the amplitude square signal of each frequency component in the spectrum and output it as the energy distribution of the signal at each frequency; Assign weights to each signal feature based on its importance in the product feature vector; Based on the weights, a weighted matrix is ​​constructed, each feature is input into each row of the weighted matrix, each sample is input into each column of the weighted matrix, and the weights corresponding to the features and samples are input into the elements in the weighted matrix; Multiply the feature matrix of the original digital signal by the weighting matrix to obtain a weighted feature matrix; Linearly amplify the weighted feature matrix, and combine the amplified feature matrix with other unweighted signal components to form a processed signal; Combine the amplified features to form and output a product feature vector; The Fourier integral transform formula is: , In the formula, is a continuous time signal, is a continuous signal in the frequency domain, is the angular frequency of the signal, is an imaginary unit, is the time parameter; The discrete Fourier transform formula is: , In the formula, is a discrete-time signal, is the discrete signal frequency domain, is the number of signal samples, is the frequency index, is the number of the discrete-time signal.

4. A server model and fault identification method according to claim 3, characterized in that: The method of using the training set to train the constructed product fault identification model and adjusting the weight and bias parameters of the convolution kernel in the model to learn the mapping relationship between the product feature vector and the product fault specifically includes: S101: Randomly select initial values ​​from Gaussian distribution, initialize the weight and bias parameters of the convolution kernel, and build an initial product fault recognition model; S102: Input the samples in the training set into the product fault recognition model, perform forward propagation calculation through the convolution layer, and obtain the prediction result of the model for each sample; S103: Convert the actual labels in the training set into a one-hot encoding form. The label of each sample is represented as a vector with the same number of fault categories, where the position of the correct category is 1 and the other positions are 0; S104: Calculate the cross entropy loss between the prediction result of the product fault recognition model and the actual label using the cross entropy loss function formula; S105: Determine whether the cross entropy loss is higher than a preset threshold. If so, use the back propagation method to calculate the gradient of the cross entropy loss function with respect to the model parameters, update the model parameters based on the calculated gradient, and return to step S102. If not, output the current product fault identification model. The cross entropy loss function formula is: , In the formula, is the cross entropy loss between the prediction results of the product fault recognition model and the actual label, is the number of samples in the training set, is the actual label value of the i-th sample, The probability that the i-th sample predicted by the model is a positive class.

5. A server model and fault identification method according to claim 4, characterized in that: In the training process, the model is verified using the verification set, and the hyper parameters of the model are adjusted based on the verification results, specifically including: In the preset model hyperparameter space, a grid search method is used to obtain a combination of at least two hyperparameters; The product fault identification model is verified using the verification set, and based on the verification results, the learning rate, number of layers, and number of neurons are gradually adjusted; Determine whether the verification loss decrease rate is lower than the preset speed threshold. If so, increase the learning rate. If not, do not output. Determine whether the cross entropy loss of the product fault identification model in the validation set is higher than the preset threshold. If so, increase the number of model layers to enhance the model's expressiveness. If not, reduce the number of model layers to reduce the computational cost and improve the generalization ability. When each layer of the model is overfitting, the number of neurons in this layer is reduced; when each layer of the model is underfitting, the number of neurons in this layer is increased.

6. A server model and fault identification method according to claim 5, characterized in that: The trained model is evaluated using the test set. By calculating the accuracy and recall of the model, the comprehensive performance indicators of the evaluation model include: Get the actual label of each sample in the test set; Based on the prediction results and actual labels, calculate the number of true positives, false positives, true negatives, and false negatives for each sample in the test set; The accuracy of the product fault identification model is obtained by calculating the ratio of the sum of the number of true positive examples and true negative examples to the total number of test set samples; The recall rate of the product fault recognition model is obtained by calculating the ratio of the number of true positive examples to the sum of the number of true positive examples and false negative examples. Based on the accuracy and recall rate of the product fault identification model, the performance comprehensive index formula is used to evaluate the comprehensive performance index of the model; The performance comprehensive index formula is: , In the formula, is the comprehensive performance index of the product fault identification model. is the accuracy of the model, is the recall rate of the model.

7. A server model and fault identification system, used to implement the server model and fault identification method according to any one of claims 1 to 6, characterized in that: include: A model identification module, which is used to collect the power supply signal, cooling system signal and electromagnetic radiation signal of the server through a signal sensor, obtain the original digital signal, perform preprocessing operations of filtering and denoising the original digital signal to improve the signal quality, perform signal enhancement processing, extract signal features from the preprocessed signal using time domain analysis signal processing technology, amplify the extracted signal features, output them as product feature vectors, and identify the server model based on the difference in product feature vectors; A fault identification module, which is used to obtain historical fault data containing different server models, and annotate the historical fault data with product model information, divide the historical fault data into a training set, a validation set, and a test set, use the training set to train the constructed product fault identification model, and learn the mapping relationship between product feature vectors and product faults by adjusting the weights and bias parameters of the convolution kernel in the model. During the training process, the model is verified using the validation set, and the hyperparameters of the model are adjusted based on the verification results. The trained model is evaluated using the test set, and the comprehensive performance indicators of the model are evaluated by calculating the accuracy and recall rate of the model; The model application module is used to input the extracted product feature vector into the product fault identification model, obtain the mapping relationship between the product feature vector and the product fault based on the server product model identification, and output the identified server product fault.

Images