Device fault identification method, system and storage medium

Abstract

The application discloses a kind of equipment failure identification method, system and storage medium, the method includes the original fault data set of equipment operation data acquisition, using principal component analysis method to carry out dimension reduction processing, by feature weight initialization mode and regularization term coefficient, extract key features, obtain the fault feature set after dimension reduction, reduce feature redundancy;For the rare fault mode of determination, SMOTE oversampling technique is used to generate synthetic samples, balance the distribution of data set through feature correlation analysis, obtain the balanced fault data set, improve the feature sparsity;From the simulated annealing algorithm after optimization, obtain the fault discrimination result, through feature explainability analysis, if there is rare fault mode in discrimination result, then trigger early warning mechanism, according to node fault recovery strategy generates targeted maintenance suggestion.The method can effectively identify and handle rare failure, improve the accuracy and efficiency of fault diagnosis.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of equipment fault identification, and in particular to an equipment fault identification method, system and storage medium. BACKGROUND

[0002] In equipment fault identification, the simulated annealing algorithm faces the problem of insufficient discrimination ability. Because the feature differences of different fault data sets are large, directly inputting them into the algorithm as if they were the same will lead to low discrimination accuracy for some fault modes. Although the importance of different data sets can be distinguished by assigning weights, how to reasonably set the weights is also a thorny problem. In addition, the current fault feature extraction method and objective function design are not accurate enough, and cannot fully mine the information contained in the fault data, which further limits the discrimination ability of the algorithm.

[0003] At the same time, in practical application, the types and causes of equipment faults are complex, including common wear and tear, fatigue and other problems, as well as some rare special faults. For these rare fault modes, the existing simulated annealing algorithm is difficult to give accurate discrimination results. How to balance the discrimination accuracy and generalization ability in the algorithm for common faults and rare faults is a problem worthy of in-depth study.

[0004] In addition, equipment fault identification often has high real-time requirements, and the algorithm needs to quickly give a discrimination result in order to take timely maintenance measures. However, the simulated annealing algorithm involves complex optimization calculations, and faces the problem of high time cost when dealing with large-scale data. How to improve the calculation efficiency of the algorithm while ensuring the discrimination accuracy is also a technical challenge that needs to be solved urgently. SUMMARY

[0005] The present application provides an equipment fault identification method, system and storage medium, which can effectively identify and handle rare faults, improve the accuracy and efficiency of fault diagnosis, provide timely and accurate decision support for equipment maintenance, and thus improve the overall equipment operation reliability.

[0006] In a first aspect, to solve the above technical problems, the present application provides an equipment fault identification method, comprising:

[0007] Obtaining an original fault data set in equipment operation data, performing dimensionality reduction processing by using a principal component analysis method to obtain a reduced fault feature set;

[0008] According to the reduced fault feature set, using a density clustering algorithm to classify the fault modes, and through feature distribution difference judgment, if the number of samples in the clustering result is lower than a preset early stopping mechanism threshold, it is determined as a rare fault mode, otherwise as a common fault mode;

[0009] According to the rare fault mode, a synthetic sample is generated by using a SMOTE oversampling technique, and a balanced dataset distribution operation is performed through feature correlation analysis to obtain a balanced fault dataset;

[0010] According to the balanced fault dataset, an adaptive weight distribution strategy is used to calculate the importance weight of each fault mode;

[0011] The importance weight is integrated into a preset objective function, and a weighted objective function is obtained through optimization algorithm selection and batch size setting, which is used to optimize the discriminant ability of the simulated annealing algorithm;

[0012] According to the weighted objective function, the calculation process of the simulated annealing algorithm is optimized by using a parallel computing framework, the number of parallel computing nodes is dynamically adjusted through node load balancing and communication delay optimization, and if the calculation time exceeds the preset time threshold, the parallel computing nodes are increased through task scheduling strategy and resource allocation algorithm;

[0013] The fault discriminant result is obtained from the optimized simulated annealing algorithm, and if the rare fault mode exists in the discriminant result, the early warning mechanism is triggered, and the maintenance suggestion is generated according to the node fault recovery strategy.

[0014] In a second aspect, the present application provides a device fault identification system, comprising:

[0015] The dimensionality reduction processing module is used to obtain the original fault dataset in the device operation data, and a principal component analysis method is used for dimensionality reduction processing to obtain the fault feature set after dimensionality reduction;

[0016] The fault classification module is used to classify the fault mode according to the fault feature set after dimensionality reduction by using a density clustering algorithm, and through feature distribution difference judgment, if the number of samples in a certain class in the clustering result is lower than the preset early stop mechanism threshold, it is determined as a rare fault mode, otherwise it is a common fault mode;

[0017] The data balancing module is used to generate a synthetic sample by using a SMOTE oversampling technique according to the rare fault mode, and a balanced dataset distribution operation is performed through feature correlation analysis to obtain a balanced fault dataset;

[0018] The weight distribution module is used to calculate the importance weight of each fault mode according to the balanced fault dataset by using an adaptive weight distribution strategy;

[0019] The function optimization module is used to integrate the importance weight into a preset objective function, and a weighted objective function is obtained through optimization algorithm selection and batch size setting, which is used to optimize the discriminant ability of the simulated annealing algorithm;

[0020] A parallel optimization module is configured to optimize the calculation process of the simulated annealing algorithm by using a parallel computing framework according to a weighted objective function, dynamically adjust the number of parallel computing nodes through node load balancing and communication delay optimization, and increase the number of parallel computing nodes by using a task scheduling strategy and a resource allocation algorithm if the calculation time exceeds a preset time threshold.

[0021] A fault discrimination module is configured to obtain a fault discrimination result from the optimized simulated annealing algorithm, trigger a warning mechanism if there is a rare fault mode in the discrimination result, and generate a maintenance suggestion according to a node fault recovery strategy.

[0022] In a third aspect, the present application further provides an electronic device comprising a processor, a memory, and a computer program stored in the memory and configured to be executed by the processor, wherein the processor implements the device fault identification method according to any one of the above-mentioned aspects when executing the computer program.

[0023] In a fourth aspect, the present application further provides a computer-readable storage medium comprising a stored computer program, wherein the computer-readable storage medium controls the device on which the computer-readable storage medium is located to execute the device fault identification method according to any one of the above-mentioned aspects when the computer program runs.

[0024] Compared with the prior art, the present application has the following beneficial effects:

[0025] The present application discloses a device fault identification method. The method first performs dimensionality reduction processing on original fault data to extract key features, then classifies fault modes by using a density clustering algorithm to identify rare faults, balances the data set by using an oversampling technique for the rare faults, then calculates the importance weight of each fault mode by using an adaptive weight allocation strategy and integrates the weight information into an improved objective function, then optimizes the simulated annealing algorithm by using a parallel computing framework to improve the calculation efficiency, and finally generates a maintenance suggestion based on the discrimination result and updates the device operating state by using real-time data stream processing technology. The present application can effectively identify and handle rare faults, improve the accuracy and efficiency of fault diagnosis, provide timely and accurate decision support for device maintenance, and thus improve the overall device operating reliability. BRIEF DESCRIPTION OF DRAWINGS

[0026] Figure 1 is a device fault identification method flowchart provided by the first embodiment of the present application;

[0027] Figure 2 is a device fault identification system structure diagram provided by the second embodiment of the present application. DETAILED DESCRIPTION

[0028] With reference to the drawings of the embodiments of the present application, the technical solutions in the embodiments of the present application will be clearly and completely described, obviously, the described embodiments are only a part of the embodiments of the present application, rather than all the embodiments. Based on the embodiments in the present application, all other embodiments obtained by those skilled in the art without creative labor fall within the scope of protection of the present application.

[0029] With reference to Figure 1 The first embodiment of the present application provides a device fault identification method, comprising the following steps:

[0030] S11, obtaining an original fault data set in device operation data, performing dimension reduction processing by using a principal component analysis method to obtain a fault feature set after dimension reduction;

[0031] S12, according to the fault feature set after dimension reduction, classifying fault modes by using a density clustering algorithm, and judging by feature distribution difference, if the number of samples in the clustering result is lower than a preset early stop mechanism threshold, determining it as a rare fault mode, otherwise as a common fault mode;

[0032] S13, according to the rare fault mode, generating synthetic samples by using a SMOTE oversampling technology, and balancing the distribution of the data set by feature correlation analysis to obtain a balanced fault data set;

[0033] S14, according to the balanced fault data set, calculating the importance weight of each fault mode by using an adaptive weight allocation strategy;

[0034] S15, integrating the importance weight into a preset objective function, obtaining a weighted objective function by optimization algorithm selection and batch size setting, which is used to optimize the discriminant ability of the simulated annealing algorithm;

[0035] S16, according to the weighted objective function, optimizing the calculation process of the simulated annealing algorithm by using a parallel computing framework, dynamically adjusting the number of parallel computing nodes by node load balancing and communication delay optimization, if the calculation time exceeds a preset time threshold, increasing the parallel computing nodes by task scheduling strategy and resource allocation algorithm;

[0036] S17, obtaining a fault discriminant result from the optimized simulated annealing algorithm, if there is a rare fault mode in the discriminant result, triggering an early warning mechanism, and generating a maintenance suggestion according to a node fault recovery strategy.

[0037] In step S11, an original fault data set in device operation data is obtained, and a principal component analysis method is used for dimension reduction processing to obtain a fault feature set after dimension reduction, including:

[0038] According to the device operation data, the original fault data set is obtained;

[0039] For the original fault data set, a principal component analysis method is adopted to perform dimension reduction processing through a feature weight initialization manner and a regularization term coefficient, to obtain a dimension reduction processing result;

[0040] Key features of the data set are extracted from the dimension reduction processing result, and a dimension-reduced fault feature set is obtained according to the extracted key features;

[0041] For the dimension-reduced fault feature set, it is determined whether the feature redundancy is lower than a preset redundancy threshold;

[0042] If the feature redundancy is lower than the preset redundancy threshold, the dimension-reduced fault feature set is output;

[0043] If the feature redundancy is higher than the preset redundancy threshold, the step of performing dimension reduction processing through the feature weight initialization manner and the regularization term coefficient is returned to, until the feature redundancy meets the requirement.

[0044] Specifically, equipment operation data is a crucial source of information in industrial production, and by analyzing these data, potential failure risks can be identified in a timely manner. Obtaining the original failure dataset is the first step in failure diagnosis and prediction. For example, in a large chemical plant, various sensor data such as temperature, pressure, flow rate, etc. can be collected, which constitutes the original failure dataset. Principal Component Analysis (PCA) is a commonly used dimensionality reduction method that can effectively reduce the complexity of data. When applying PCA, the initialization of feature weights and the selection of regularization term coefficients are crucial. For example, for the data of a chemical plant, according to the initialized feature weights, higher weights can be given to temperature and pressure, while lower weights can be given to some secondary parameters. The regularization term coefficient can be determined by cross-validation and other methods to balance the complexity and generalization ability of the model. After dimensionality reduction processing by PCA, the key features of the dataset can be extracted. In the example of a chemical plant, it is found that temperature variation, pressure fluctuation, and the concentration of certain chemicals are the most important features. These key features constitute the failure feature set after dimensionality reduction, greatly simplifying the subsequent analysis work. Feature redundancy is an important indicator of the quality of a feature set. If the feature redundancy is lower than the preset threshold, it indicates that the dimensionality reduction effect is good, and this feature set can be directly used for model training. For example, if the original data has 100 features, and only 10 features are retained after dimensionality reduction, and these 10 features can explain 95% of the data variance, then the redundancy of this feature set is very low. However, if the feature redundancy is higher than the preset threshold, it is necessary to return to adjust the feature weights and the regularization term coefficients. For example, you can try to increase the regularization strength or adjust the initial weight distribution to give some previously overlooked features higher weights. This process needs to be iterated multiple times until a satisfactory result is obtained. The advantage of this method is that it can effectively handle high-dimensional data and extract the most critical information. In industrial production, this can help engineers more quickly identify potential problems and improve the accuracy and efficiency of failure diagnosis.

[0045] For example, in a chemical plant, if it is found that temperature anomalies are the most important features, then temperature-related parameters can be monitored and analyzed first, thereby more quickly discovering and solving problems. In addition, this method also has strong adaptability. Different industrial scenarios have different feature importance, and by adjusting the feature weights and regularization parameters, the model can better adapt to specific application scenarios. For example, in a system where pressure control is the main concern, higher weights should be given to pressure-related features. In summary, this PCA-based failure feature extraction method, through iterative optimization, can greatly reduce the dimensionality of data while preserving key information, laying a solid foundation for subsequent failure diagnosis and prediction.

[0046] In step S12, according to the reduced fault feature set, a density clustering algorithm is used to classify the fault modes, and through feature distribution difference judgment, if the number of samples in a certain class in the clustering result is lower than the preset early stopping mechanism threshold, it is determined as a rare fault mode, otherwise as a common fault mode, including:

[0047] For the reduced fault feature set, according to the density connectivity between samples, the fault features are clustered and analyzed to obtain the clustering results of multiple fault modes;

[0048] The sample number distribution of each class in the clustering result is analyzed, and compared with the preset rare fault threshold to determine whether there is a rare fault mode;

[0049] If the number of samples in a certain class is lower than the preset early stopping mechanism threshold, it is determined that this class is a rare fault mode; if the number of samples in a certain class is not lower than the preset early stopping mechanism threshold, it is determined that this class is a common fault mode.

[0050] In this embodiment, the density clustering algorithm such as DBSCAN can effectively identify clusters of different shapes. In fault diagnosis, this feature is particularly important. For example, a certain type of bearing fault shows irregular shape distribution in the feature space. DBSCAN can divide these density-connected points into a cluster and identify noise points by setting the neighborhood radius and minimum sample number. The identification of rare faults is a difficult point in fault diagnosis. Assuming that a certain fault occurs only 5 times in 1000 samples. By setting a suitable threshold, such as less than 1% of the total samples, this type of fault can be marked as a rare fault.

[0051] In step S13, according to the rare fault mode, SMOTE oversampling technology is used to generate synthetic samples, and feature correlation analysis is used to balance the dataset distribution to obtain a balanced fault dataset.

[0052] Specifically, in the field of fault diagnosis, handling rare fault patterns is a significant challenge. To improve the recognition ability of rare faults, the SMOTE oversampling technique can be used to balance the dataset. For example, suppose there is a fault dataset containing 1000 samples, of which only 20 samples belong to a certain rare fault type. Through the SMOTE algorithm, new synthetic samples can be generated by interpolating between the minority class samples, increasing the number of rare fault samples to 100, making the dataset more balanced. Feature extraction is a key step in fault diagnosis. Taking industrial equipment as an example, multi-dimensional features such as vibration, temperature, and pressure can be extracted. By calculating the correlation coefficients between these features, a feature correlation matrix can be constructed. Suppose the correlation coefficient between vibration and temperature features is 0.8, while the correlation coefficient between vibration and pressure is only 0.2, indicating that there is some information redundancy between vibration and temperature. To reduce feature redundancy, the minimum redundancy maximum correlation criterion can be used for feature selection. In the above example, the vibration and pressure features can be selected while the temperature feature is discarded. This not only reduces the computational complexity but also improves the generalization ability of the model.

[0053] To continuously improve system performance, a dynamic updating mechanism can also be established. Whenever the system detects new fault cases, especially rare fault cases, these data will be added to the fault knowledge base. For example, if the system successfully predicts and confirms 3 new rare fault cases in a month, the feature data of these cases will be used to expand the original dataset. By regularly updating, the system's recognition ability for various fault patterns, especially for rare faults that are difficult to capture, can be continuously improved.

[0054] In step S14, according to the balanced fault dataset, an adaptive weight allocation strategy is used to calculate the importance weight of each fault pattern, including:

[0055] According to each fault pattern in the balanced fault dataset, the frequency of its occurrence in historical data is counted to obtain a fault pattern frequency list;

[0056] According to the fault pattern frequency list, an adaptive weight allocation strategy is used to initialize the importance weight of each fault pattern through learning rate adjustment strategy and weight decay factor;

[0057] For each fault pattern, it is determined whether its frequency is higher than a preset frequency threshold. If it is higher than the frequency threshold, the importance weight of the fault pattern is multiplied by a factor greater than 1 to obtain a first adjusted weight;

[0058] If it is lower than the frequency threshold, the importance weight of the fault pattern is multiplied by a factor less than 1 to obtain a second adjusted weight;

[0059] output the first adjustment weight or the second adjustment weight as a final importance weight of each failure mode.

[0060] Specifically, in the fault diagnosis system, the next step after balancing the dataset is to analyze the failure mode frequency. This step is crucial for improving diagnostic accuracy, as the frequency of occurrence of different failure modes often varies significantly. For example, on a production line in a certain factory, bearing wear is a high-frequency failure, while motor burnout is a low-frequency failure. By analyzing historical data, a failure mode frequency list can be obtained. Suppose that in the past year, bearing wear occurred 200 times, accounting for 40% of the total number of failures; motor burnout occurred only 5 times, accounting for 1%. Based on this frequency list, an adaptive weight allocation strategy is adopted. The core idea of this strategy is to dynamically adjust the model's attention to different failures based on their frequency.

[0061] Specifically, a frequency threshold can be set, such as 10%. For failure modes above the threshold, such as bearing wear, their initial weights are multiplied by a factor greater than 1, such as 1.5; while for failure modes below the threshold, such as motor burnout, they are multiplied by a factor less than 1, such as 0.8. The purpose of this is to ensure that the model has good recognition ability for high-frequency failures while not ignoring low-frequency but serious failures. Learning rate and weight decay factor play an important role in the weight adjustment process. Learning rate determines the step size of parameter update at each iteration, while weight decay factor helps prevent overfitting of the model. For example, a larger learning rate, such as 0.1, can be set for high-frequency failures to quickly converge; while a smaller learning rate, such as 0.01, can be set for low-frequency failures to avoid overfitting to a small number of samples. At the same time, for low-frequency failures, the weight decay factor can be increased, such as from 0.001 to 0.005, to enhance the model's generalization ability. Through this method, a more balanced and practical fault diagnosis model can be trained. This model not only accurately identifies common failures, but also remains sensitive to rare but serious failures. In practical applications, this method can significantly improve the accuracy and timeliness of fault warning, thereby reducing equipment downtime, reducing maintenance costs, and improving production efficiency.

[0062] In step S15, the importance weight is integrated into the preset objective function, and a weighted objective function is obtained through optimization algorithm selection and batch size setting, which is used to optimize the discriminant ability of the simulated annealing algorithm, including:

[0063] An initial objective function is constructed, and the importance weight is introduced as a parameter into the initial objective function to obtain an objective function expression containing weight information;

[0064] A gradient descent algorithm is used to optimize and solve the initial objective function to obtain the optimal weight parameter;

[0065] Substitute the optimal weight parameters into the simulated annealing algorithm, and dynamically adjust the acceptance probability and annealing strategy during the simulated annealing process according to the weight information;

[0066] During the simulated annealing iteration process, the weight parameters and weight update frequency are adaptively adjusted according to the quality of the current solution and the historical optimal solution. The weights are appropriately scaled and regularized according to the sensitivity of the weight parameters.

[0067] The optimal hyperparameter combination is selected through cross-validation to obtain the weighted objective function;

[0068] Among them, the weighted objective function is expressed as:

[0069]

[0070] Where, represents the weighted objective function, Indicates the The weight of each sub-goal, Indicates the sub-objective functions, is the decision variable vector, is the number of sub-goals, is the regularization parameter.

[0071] Specifically, this formula represents an objective function that includes weight information. The first term represents the sum of weighted sub-objective functions, and the second term is an L1 regularization term used to control the sparsity of the weights. Using optimization algorithms such as gradient descent, the weighted objective function is optimized by setting an appropriate batch size and number of iterations to obtain the optimal solution for the weight parameters. The resulting optimal weight parameters are then substituted into a simulated annealing algorithm. The acceptance probability and annealing strategy are dynamically adjusted during the simulated annealing process based on the weight information, improving the algorithm's discriminative capabilities. During the simulated annealing iterations, the weight parameters are adaptively adjusted based on the quality of the current solution and the historical optimal solution, accelerating weight convergence and preventing the algorithm from falling into local optima. Regarding the weight update frequency, an adaptive weight update strategy is designed to dynamically adjust the weight update frequency based on the algorithm convergence status and problem size, ensuring convergence speed while reducing computational overhead. During the algorithm iterations, changes in the weight parameters are monitored in real time. Based on the sensitivity of the weight parameters, appropriate weight scaling and regularization are performed to improve the algorithm's generalization and robustness.

[0072] In step S16, based on the weighted objective function, the parallel computing framework is used to optimize the calculation process of the simulated annealing algorithm. The number of parallel computing nodes is dynamically adjusted through node load balancing and communication delay optimization. If the calculation time exceeds the preset time threshold, additional parallel computing nodes are added through the task scheduling strategy and resource allocation algorithm, including:

[0073] According to the weighted objective function, a parallel computing model of the simulated annealing algorithm is constructed, and a parallel computing framework is designed in a master-slave architecture, wherein the master node is responsible for task scheduling and global parameter updating, and the slave node is responsible for local computing tasks;

[0074] Load conditions and communication delay data of each parallel computing node are acquired, and the computing task allocation of each node is dynamically adjusted through a load balancing algorithm;

[0075] During the parallel computing process, the computing time of each node is continuously monitored, and if it is found that the computing time exceeds a preset time threshold, a dynamic node adjustment mechanism is triggered, and the number of parallel computing nodes is dynamically increased according to a task scheduling strategy and a resource allocation algorithm.

[0076] Specifically, constructing a parallel computing model for simulated annealing algorithm is an important means to improve algorithm efficiency. Using master-slave architecture to design parallel computing framework can make full use of distributed computing resources. For example, in a large-scale path optimization problem, the master node can be responsible for dividing the entire path network into multiple sub-regions and assigning these sub-regions to slave nodes for local optimization. The master node is also responsible for collecting the optimization results of each slave node and updating the global parameters. In order to achieve efficient parallel computing, factors such as load balancing and communication delay need to be considered. By monitoring the CPU usage, memory usage and other indicators of each node in real time, task allocation can be dynamically adjusted. For example, if the CPU usage of a slave node is found to be continuously above 90%, while the CPU usage of other nodes is only around 50%, some tasks can be transferred from the high-load node to the low-load node. At the same time, by optimizing the communication strategy between nodes, such as using asynchronous communication or batch communication, the impact of communication delay on computing performance can be effectively reduced. In the process of parallel computing, dynamic node adjustment mechanism can effectively deal with computing bottlenecks. Suppose in a complex financial model optimization task, initially 10 computing nodes are set, but as the computing scale increases, it is found that the average computing time exceeds the preset 5-minute threshold. At this time, the system can automatically trigger the node expansion mechanism, dynamically increase 5 computing nodes according to the current load and available resources, and reduce the average computing time to an acceptable range. Adaptive annealing temperature adjustment strategy is the key to improving the efficiency of simulated annealing algorithm. In the optimization process, the annealing temperature can be dynamically adjusted according to the quality change of the solution in continuous multiple iterations. For example, if the quality of the solution in 100 consecutive iterations does not improve significantly, the annealing temperature can be appropriately increased to increase the probability of the algorithm jumping out of the local optimum; conversely, if the solution quality continues to improve, the annealing temperature can be reduced to speed up the convergence speed. Efficient parallel random number generation algorithm is crucial to ensure the randomness and parallelism of simulated annealing algorithm. A hybrid random number generation method based on timestamp and node ID can be used to ensure that the random number sequences generated by each node are independent of each other and have good statistical properties. This can avoid different nodes generating the same random number sequence, thereby improving the exploration ability of the algorithm and the quality of the solution. Incremental global parameter update strategy can effectively reduce the communication overhead between nodes. In a large-scale image recognition task, each slave node can independently process a batch of image data and calculate the local gradient. After processing 1000 images, the slave node sends the accumulated gradient information to the master node. The master node aggregates the gradient information from all slave nodes, updates the global model parameters, and then broadcasts the updated parameters to all slave nodes. This approach ensures the consistency of the model while reducing the communication frequency.

[0077] In addition, designing a fault-tolerant mechanism with strong adaptability is crucial to ensure the reliability of parallel computing. Regular checkpointing technology can be used to save the computing state to persistent storage. If a node fails, the system can quickly recover the computing state from the latest checkpoint and redistribute the tasks of the failed node to other available nodes. In addition, by setting up a heartbeat mechanism, the master node can discover abnormal nodes in time and take appropriate fault-tolerant measures, such as restarting the node or redistributing tasks, to ensure the robustness and reliability of the entire parallel computing process.

[0078] In step S17, the fault discrimination result is obtained from the optimized simulated annealing algorithm. If there is a rare fault mode in the discrimination result, the early warning mechanism is triggered, and maintenance suggestions are generated according to the node fault recovery strategy, including:

[0079] The optimized simulated annealing algorithm is used to analyze the equipment operation data and obtain the fault discrimination result;

[0080] The key features in the fault discrimination result are interpreted to determine whether there is a rare fault mode;

[0081] If there is a rare fault mode in the discrimination result, the early warning mechanism is triggered, and early warning information is sent to the system administrator;

[0082] According to the early warning information, combined with the pre-established node fault recovery strategy knowledge base, a targeted node fault recovery scheme is generated;

[0083] The generated node fault recovery scheme is sent to the fault node to guide the self-repair of the fault node;

[0084] If the node fault recovery scheme fails to execute, the node operation log and fault information are used to generate node maintenance suggestions through a decision tree algorithm.

[0085] Specifically, the optimized simulated annealing algorithm can be used to analyze system operation data and obtain fault diagnosis results. This algorithm simulates the metal cooling process to find the global optimal solution in the search space. For example, for a large data center, this algorithm can be used to analyze the performance indicators of the server cluster, such as CPU usage, memory occupancy, network latency, etc., to identify potential faulty nodes. Feature interpretability analysis techniques help explain key features in fault diagnosis results and determine whether there are rare failure patterns. For example, if the analysis result shows that the CPU temperature of a server is abnormally high, while other indicators are normal, it indicates that the cooling system has failed. Through interpretability analysis, it can be determined that the abnormal temperature is the key factor causing the failure. When a rare failure pattern is found, the early warning mechanism sends an early warning message to the system administrator. In the data center scenario, the early warning message contains the ID of the faulty server, the abnormal indicator, the type of failure, and other key information. In this way, the administrator can quickly locate the problem and take appropriate measures. Case-based reasoning algorithms can be used to generate targeted node failure recovery plans. This algorithm is based on a pre-established knowledge base of node failure recovery strategies and generates solutions by matching similar cases. For example, for a server with an abnormal CPU temperature, the system will recommend checking the cooling fan, cleaning dust, replacing thermal paste, and other specific operation steps. The generated node failure recovery plan is sent to the faulty node to guide its self-repair.

[0086] In the data center environment, this involves remote execution of scripts, adjustment of system parameters, or restart of specific services. Automated repair not only improves efficiency but also reduces human error. If automatic repair fails, decision tree algorithms can be used to generate node maintenance recommendations. This algorithm builds a logical decision-making process by analyzing node operation logs and failure information. For example, for persistent CPU temperature abnormalities, the decision tree will recommend checking hardware integrity, replacing the CPU, or considering server location adjustment for deeper solutions. Finally, the node maintenance recommendations are sent to the maintenance personnel to guide them in manual maintenance. These recommendations include detailed fault diagnosis steps, required tools and parts lists, safety precautions, etc. By providing structured maintenance guidelines, even complex failures can be effectively handled, ensuring the normal operation of the system.

[0087] In summary, the application discloses a device fault identification method. The method first performs dimensionality reduction processing on original fault data to extract key features, then classifies fault modes using a density clustering algorithm to identify rare faults, balances the data set using an oversampling technique for rare faults, then calculates the importance weight of each fault mode using an adaptive weight allocation strategy and integrates the weight information into an improved objective function, then uses a parallel computing framework to optimize the simulated annealing algorithm to improve computing efficiency, and finally generates maintenance recommendations based on the discrimination results and updates the device operating state using real-time data stream processing technology. The application can effectively identify and handle rare faults, improve the accuracy and efficiency of fault diagnosis, provide timely and accurate decision support for equipment maintenance, and thus improve the overall equipment operation reliability.

[0088] With reference to Figure 2 The second embodiment of the application provides a device fault identification system, comprising:

[0089] A dimensionality reduction processing module is configured to obtain an original fault data set in device operating data, perform dimensionality reduction processing using a principal component analysis method, and obtain a fault feature set after dimensionality reduction.

[0090] A fault classification module is configured to classify fault modes using a density clustering algorithm based on the fault feature set after dimensionality reduction, and determine a rare fault mode if the number of samples in a certain class in the clustering result is lower than a preset early stopping mechanism threshold, or a common fault mode otherwise, by judging the feature distribution difference.

[0091] A data balancing module is configured to generate synthetic samples using an SMOTE oversampling technique based on the rare fault mode, and balance the data set distribution by feature correlation analysis to obtain a balanced fault data set.

[0092] A weight allocation module is configured to calculate the importance weight of each fault mode using an adaptive weight allocation strategy based on the balanced fault data set.

[0093] A function optimization module is configured to integrate the importance weight into a preset objective function, select an optimization algorithm and set a batch size to obtain a weighted objective function, and use the weighted objective function to optimize the discrimination ability of the simulated annealing algorithm.

[0094] A parallel optimization module is configured to optimize the calculation process of the simulated annealing algorithm using a parallel computing framework based on the weighted objective function, dynamically adjust the number of parallel computing nodes by node load balancing and communication delay optimization, and increase the number of parallel computing nodes by a task scheduling strategy and a resource allocation algorithm if the calculation time exceeds a preset time threshold.

[0095] The fault discrimination module is configured to acquire a fault discrimination result from the optimized simulated annealing algorithm, and if a rare fault mode exists in the discrimination result, trigger a pre-warning mechanism, and generate a maintenance suggestion according to a node fault recovery strategy.

[0096] It should be noted that the device fault identification system provided by the embodiments of the present application is used to execute all process steps of the device fault identification method provided by the above embodiments, and the working principles and beneficial effects of the two are one-to-one correspondence, thus not being described in detail.

[0097] The embodiments of the present application further provide an electronic device. The electronic device comprises a processor, a memory, and a computer program, for example, a device fault identification program, stored in the memory and executable on the processor. The processor implements the steps in the above-mentioned various device fault identification method embodiments when executing the computer program, for example Figure 1 The processor implements the functions of the modules / units in the above-mentioned various system embodiments when executing the computer program, for example, a fault discrimination module.

[0098] For example, the computer program can be divided into one or more modules / units, which are stored in the memory and executed by the processor to complete the present application. The one or more modules / units can be a series of computer program instruction segments capable of completing a specific function, which are used to describe the execution process of the computer program in the electronic device.

[0099] The electronic device can be a desktop computer, a notebook, a palm computer, and a smart tablet, etc. The electronic device can include, but is not limited to, a processor, a memory. Those skilled in the art can understand that the above components are only examples of the electronic device, and do not constitute a limitation on the electronic device, and can include more or fewer components than the above, or combine certain components, or different components, for example, the electronic device can also include an input / output device, a network access device, a bus, etc.

[0100] The processor can be a central processing unit (CPU), and can also be other general-purpose processors, a digital signal processor (DSP), an application specific integrated circuit (ASIC), a field-programmable gate array (FPGA) or other programmable logic device, discrete gate or transistor logic, discrete hardware components, etc. The general-purpose processor can be a microprocessor or the processor can also be any conventional processor. The processor is a control center of the electronic device, and connects various parts of the electronic device through various interfaces and lines.

[0101] The memory can be used to store the computer program and / or modules, and the processor realizes various functions of the electronic device by running or executing the computer program and / or modules stored in the memory, and calling data stored in the memory. The memory can mainly include a program storage area and a data storage area, wherein the program storage area can store an operating system, at least one application program required by a function (such as a sound playing function, an image playing function, etc.), etc.; and the data storage area can store data created according to the use of the mobile phone (such as audio data, a phone book, etc.), etc. In addition, the memory can include a high-speed random access memory, and can also include a nonvolatile memory, for example, a hard disk, a memory, a plug-in hard disk, a smart media card (SMC), a secure digital (SD) card, a flash card, at least one disk storage device, a flash memory device, or other volatile solid-state memory device.

[0102] The modules / units integrated in the electronic device, if realized in the form of software function units and sold or used as independent products, can be stored in a computer readable storage medium. Based on such understanding, all or part of the processes in the above-mentioned embodiment methods can also be completed by a computer program instructing related hardware, and the computer program can be stored in a computer readable storage medium. The computer program can implement the steps of each method embodiment when executed by a processor. The computer program includes computer program code, which can be in the form of source code, object code, executable files, or some intermediate forms, etc. The computer readable medium can include any entity or system, recording medium, U disk, mobile hard disk, magnetic disk, optical disk, computer memory, read-only memory (ROM), random access memory (RAM), electric carrier signal, telecommunication signal, and software distribution medium, etc. that can carry the computer program code. It should be noted that the contents included in the computer readable medium can be appropriately increased or decreased according to the requirements of legislation and patent practice in the jurisdiction, for example, in some jurisdictions, according to legislation and patent practice, the computer readable medium does not include electric carrier signals and telecommunication signals.

[0103] It should be noted that the above-described system embodiments are only schematic, and the units described as separate components can or can not be physically separated, and the components shown as units can or can not be physical units, that is, they can be located in one place or distributed on multiple network units. Part or all of the modules can be selected according to actual needs to achieve the purpose of the embodiment. In addition, the connection relationship between the modules in the system embodiments provided by the present application indicates that there is a communication connection between them, which can be realized as one or more communication buses or signal lines. Those skilled in the art can understand and implement it without creative labor.

[0104] The above-described specific embodiments further detail the purpose, technical solutions, and beneficial effects of the present application. It should be understood that the above-described specific embodiments are only for the specific embodiments of the present application and do not limit the protection scope of the present application. It is particularly pointed out that any modification, equivalent replacement, improvement, etc. made by those skilled in the art within the spirit and principles of the present application should be included in the protection scope of the present application.

Claims

1. A device failure identification method characterized by, The method comprises: obtaining an original fault data set in equipment operation data, performing dimension reduction processing on the original fault data set by using a principal component analysis method to obtain a fault feature set after dimension reduction; according to the fault feature set after dimension reduction, classifying fault modes by using a density clustering algorithm, and judging by feature distribution difference, if the number of samples in the clustering result is lower than a preset early stopping mechanism threshold, determining that it is a rare fault mode, otherwise, it is a common fault mode; according to the rare fault mode, generating a synthetic sample by using a SMOTE oversampling technology, and balancing the distribution of the data set by feature correlation analysis to obtain a balanced fault data set; according to the balanced fault data set, calculating the importance weight of each fault mode by using an adaptive weight distribution strategy; integrating the importance weight into a preset objective function, obtaining a weighted objective function by optimization algorithm selection and batch size setting, and using the weighted objective function to optimize the discriminant ability of the simulated annealing algorithm; according to the weighted objective function, optimizing the calculation process of the simulated annealing algorithm by using a parallel computing framework, dynamically adjusting the number of parallel computing nodes by node load balancing and communication delay optimization, and if the calculation time exceeds a preset time threshold, increasing the number of parallel computing nodes by task scheduling strategy and resource allocation algorithm; obtaining a fault discriminant result from the optimized simulated annealing algorithm, and if there is a rare fault mode in the discriminant result, triggering an early warning mechanism and generating a maintenance suggestion according to a node fault recovery strategy; wherein, according to the balanced fault data set, the importance weight of each fault mode is calculated by using an adaptive weight distribution strategy, which comprises: according to each fault mode in the balanced fault data set, counting its frequency in historical data to obtain a fault mode frequency list; according to the fault mode frequency list, initializing the importance weight of each fault mode by using an adaptive weight distribution strategy, a learning rate adjustment strategy and a weight decay factor; for each fault mode, judging whether its frequency is higher than a preset frequency threshold, if higher than the frequency threshold, multiplying the importance weight of the fault mode by a factor greater than 1 to obtain a first adjustment weight; if lower than the frequency threshold, multiplying the importance weight of the fault mode by a factor less than 1 to obtain a second adjustment weight; outputting the first adjustment weight or the second adjustment weight as the final importance weight of each fault mode; selecting the optimal combination of hyperparameters by cross-validation to obtain the weighted objective function; wherein, the weighted objective function is represented as: ； wherein represents a weighted objective function, represents the weight of the th sub-objective, represents the th sub-objective function, is a decision variable vector, is the number of sub-objectives, is a regularization parameter.

2. The method of claim 1, wherein, obtaining an original fault data set in equipment operation data, performing dimension reduction processing on the original fault data set by using a principal component analysis method to obtain a fault feature set after dimension reduction, which comprises: obtaining an original fault data set according to equipment operation data; performing dimension reduction processing on the original fault data set by using a principal component analysis method, and obtaining a dimension reduction processing result by feature weight initialization method and regularization term coefficient; extracting key features of the data set from the dimension reduction processing result, and obtaining a fault feature set after dimension reduction according to the extracted key features; judging whether the feature redundancy of the fault feature set after dimension reduction is lower than a preset redundancy threshold; If the feature redundancy is lower than the preset redundancy threshold, output the reduced fault feature set; If the feature redundancy is higher than the preset redundancy threshold, return to the step of re-performing the dimensionality reduction processing by the feature weight initialization method and the regularization term coefficient until the feature redundancy meets the requirement.

3. The method of claim 1, wherein, According to the reduced fault feature set, the density clustering algorithm is used to classify the fault modes, and the feature distribution difference is judged. If the number of samples in a certain class in the clustering result is lower than the preset early stopping mechanism threshold, it is determined as a rare fault mode, otherwise as a common fault mode, including: According to the reduced fault feature set, the density connectivity between samples is used to cluster and analyze the fault features, and the clustering results of multiple fault modes are obtained; The sample number distribution of each class in the clustering result is analyzed, and whether there is a rare fault mode is judged by comparing with the preset rare fault threshold; If the number of samples in a certain class is lower than the preset early stopping mechanism threshold, it is determined that the class is a rare fault mode; if the number of samples in a certain class is not lower than the preset early stopping mechanism threshold, it is determined that the class is a common fault mode.

4. The method of claim 1, wherein, The importance weight is integrated into the preset objective function, and the weighted objective function is obtained by optimization algorithm selection and batch size setting, which is used to optimize the discriminant ability of the simulated annealing algorithm, including: An initial objective function is constructed, the importance weight is introduced into the initial objective function as a parameter, and an objective function expression containing weight information is obtained; The gradient descent algorithm is used to optimize and solve the initial objective function, and the optimal weight parameter is obtained; The optimal weight parameter is substituted into the simulated annealing algorithm, and the acceptance probability and annealing strategy in the simulated annealing process are dynamically adjusted according to the weight information; In the simulated annealing iteration process, the weight parameter and the weight update frequency are adaptively adjusted according to the quality of the current solution and the historical optimal solution, and the weight is appropriately scaled and regularized according to the sensitivity of the weight parameter.

5. The method of claim 1, wherein, According to the weighted objective function, the calculation process of the simulated annealing algorithm is optimized by using a parallel computing framework, the number of parallel computing nodes is dynamically adjusted by optimizing node load balancing and communication delay, and if the calculation time exceeds the preset time threshold, the number of parallel computing nodes is increased by using task scheduling strategy and resource allocation algorithm, including: According to the weighted objective function, a parallel computing model of the simulated annealing algorithm is constructed, and a master-slave architecture is used to design a parallel computing framework, wherein the master node is responsible for task scheduling and global parameter updating, and the slave node is responsible for local computing tasks. The load condition and communication delay data of each parallel computing node are obtained, and the calculation task allocation of each node is dynamically adjusted by using a load balancing algorithm. In the parallel computing process, the calculation time of each node is continuously monitored, and if it is found that the calculation time exceeds the preset time threshold, the dynamic node adjustment mechanism is triggered, and the number of parallel computing nodes is dynamically increased according to the task scheduling strategy and resource allocation algorithm.

6. The method of claim 1, wherein, The fault discrimination result is obtained from the optimized simulated annealing algorithm, and if there is a rare fault mode in the discrimination result, the early warning mechanism is triggered, and the maintenance suggestion is generated according to the node fault recovery strategy, including: The optimized simulated annealing algorithm is used to analyze the equipment operation data to obtain a fault discrimination result. The key features in the fault discrimination result are interpreted to determine whether there is a rare fault mode. If there is a rare fault mode in the discrimination result, a warning mechanism is triggered to send a warning information to the system administrator. According to the warning information, a node fault recovery strategy knowledge base is established to generate a targeted node fault recovery scheme. The generated node fault recovery scheme is sent to the fault node to guide the self-repair of the fault node. If the node fault recovery scheme fails to execute, the node operation log and fault information are used to generate a node maintenance suggestion through a decision tree algorithm.

7. An apparatus failure identification system for implementing the apparatus failure identification method according to any one of claims 1 to 6, characterized by, The method comprises the following steps: a dimension reduction processing module is configured to obtain an original fault data set from the equipment operation data, and perform dimension reduction processing on the original fault data set by using a principal component analysis method to obtain a reduced fault feature set; a fault classification module is configured to classify fault modes by using a density clustering algorithm based on the reduced fault feature set, and determine whether a certain class of samples in the clustering result is a rare fault mode or a common fault mode by judging the feature distribution difference; a data balancing module is configured to generate synthetic samples by using a SMOTE oversampling technique based on the rare fault mode, and perform a balanced data set distribution operation on the synthetic samples by using feature correlation analysis to obtain a balanced fault data set; a weight allocation module is configured to calculate the importance weight of each fault mode by using an adaptive weight allocation strategy based on the balanced fault data set; a function optimization module is configured to integrate the importance weight into a preset objective function, select an optimization algorithm and set a batch size to obtain a weighted objective function, and use the weighted objective function to optimize the discrimination ability of the simulated annealing algorithm; a parallel optimization module is configured to optimize the calculation process of the simulated annealing algorithm by using a parallel computing framework based on the weighted objective function, dynamically adjust the number of parallel computing nodes by optimizing node load balancing and communication delay, and increase the number of parallel computing nodes by using a task scheduling strategy and a resource allocation algorithm if the calculation time exceeds a preset time threshold; a fault discrimination module is configured to obtain a fault discrimination result from the optimized simulated annealing algorithm, and trigger a warning mechanism to generate a maintenance suggestion based on a node fault recovery strategy if there is a rare fault mode in the discrimination result.

8. An electronic device, comprising: The computer readable storage medium comprises a stored computer program, wherein the computer readable storage medium controls the device to execute the device fault identification method when the computer program is running.

9. A computer readable storage medium, characterized in that, The computer readable storage medium comprises a stored computer program, wherein the computer readable storage medium controls the device to execute the device fault identification method when the computer program is running.

Images