Fault detection visualization processing system and method for industrial switch

Abstract

The invention discloses a fault detection visualization processing system and method for an industrial switch, and belongs to the technical field of artificial intelligence. The method comprises the following steps: collecting operation data of an industrial switch from different data sources, performing multi-source data fusion, and constructing a multi-source data fusion data set; extracting abnormal features, training a fault diagnosis model by using a random forest algorithm, inputting real-time data into the model, and performing fault prediction and diagnosis; designing a visual interface, and displaying the operation state and fault information of the industrial switch; functions of real-time refreshing, history backtracking and fault positioning are provided; after a fault is found, a fault source is automatically tracked, a fault propagation path is analyzed, based on a fault type and historical data, a Bayesian network is used to establish a correlation model based on the historical data, and repair suggestions and prevention measures are provided; and analyzing data needing preliminary fault detection, performing preliminary fault detection locally on the industrial switch, and synchronizing edge and cloud data in real time.

Description

Technical Field

[0001] The present invention relates to the field of artificial intelligence technology, and in particular to a fault detection visualization processing system and method for an industrial switch. Background Art

[0002] Industrial switches play a key role in industrial automation and control systems. They are used to connect and manage various devices and sensors in industrial networks to ensure reliable data transmission. However, with the improvement of industrial automation level and the expansion of industrial network scale, the stability and reliability of industrial switches face increasing challenges. Once an industrial switch fails, it may paralyze the entire industrial system and cause huge losses. Therefore, real-time monitoring and diagnosis of the operating status of industrial switches and timely detection and handling of faults have become a research hotspot in the field of industrial automation.

[0003] Many existing systems can only diagnose faults after they occur, lacking early warning and prediction capabilities, making it difficult to take preventive measures in time to reduce the impact of faults. Existing systems usually perform all data processing and model training locally, with limited computing resources and slow processing speeds, making it difficult to cope with the processing needs of large amounts of data in real time. Many existing systems lack intelligent analysis capabilities and are unable to automatically track the source of faults, analyze fault propagation paths, and provide repair suggestions and preventive measures based on historical data. Summary of the invention

[0004] The object of the present invention is to provide a visual processing system and method for fault detection of industrial switches to solve the problems raised in the above background technology.

[0005] In order to solve the above technical problems, the present invention provides the following technical solutions: A visual processing method for fault detection of an industrial switch, the method comprising: S100, collecting operation data of the industrial switch from different data sources, where the data sources include sensor data, log data, and flow data, performing multi-source data fusion, constructing a multi-source data fusion data set, and recording it into historical data; S200, extracts abnormal features from the historical data of industrial switches, uses the random forest algorithm to train the fault diagnosis model, inputs real-time data into the model, and performs fault prediction and diagnosis; S300, based on user needs, designs a visual interface, uses several graphics to display the operating status and fault information of industrial switches; provides real-time refresh, historical backtracking and fault location functions; S400, after discovering a fault, automatically traces the source of the fault, analyzes the fault propagation path, and uses a Bayesian network to establish a correlation model between sensor data, log data, and traffic data based on historical data based on the fault type and historical data, and provides repair suggestions and preventive measures; S500: Analyze the data that needs preliminary fault detection, perform preliminary fault detection locally on the industrial switch to reduce data transmission delay, filter out normal data, move the remaining data and model training tasks to the cloud, and synchronize edge and cloud data in real time.

[0006] According to step S100, sensor data is collected from various sensors installed on the industrial switch, the sensors including temperature sensors, voltage sensors and vibration sensors; log data is collected from the system log of the switch, the system log includes a startup log, an error log and an event log, the log data records the operation history and abnormal events of the switch; flow data is collected from the network interface of the switch, the flow data reflects the operation status and performance of the network; The collected raw data is cleaned to remove noise data and outliers, data from different sources are converted into a unified format, the timestamps of each data source are aligned, multi-source data is fused, and a multi-source data fusion dataset is constructed.

[0007] According to step S100, temperature data is collected through a temperature sensor to record time and temperature values; voltage data is collected through a voltage sensor to record time and voltage values; vibration data is collected through a vibration sensor to record time and vibration intensity values; the startup time of the switch and events during the startup process are recorded through a startup log; error information occurring during the operation of the switch is recorded through an error log; and various events of the industrial switch system are recorded through an event log; all data are unified into a JSON format, and each record contains a timestamp and a data value; When performing multi-source data fusion, it is necessary to extract features from the data. The average value, maximum value, minimum value, standard deviation, rate of change, peak duration, and number of abnormalities of temperature, voltage, and vibration are extracted from sensor data. The time, event type, and error code of key events are extracted from log data. The total amount, peak value, average value, and transmission rate of traffic are extracted from traffic data. A database is created, wherein the table in the database contains all feature fields, and for each time point, the feature values ​​in different data sources are matched and integrated to construct a multi-source data fusion data set.

[0008] According to step S200, the steps of extracting abnormal features from the historical data of the industrial switch are as follows: S201-1. Evaluate the thermal stability of the industrial switch based on the average temperature over a period of time; detect whether the temperature exceeds the safety range based on the highest and lowest temperatures; evaluate the temperature stability based on the standard deviation of temperature fluctuations; identify temperature mutations based on the temperature change rate; identify overheating based on the duration of the peak value when the temperature exceeds a certain threshold; determine whether the industrial switch has abnormal thermal phenomena based on the number of abnormal temperature occurrences, where the number of abnormal occurrences indicates the number of times the temperature exceeds a preset threshold; S201-2. Evaluate the power supply stability based on the average voltage over a period of time; detect whether the voltage exceeds the safety range based on the highest and lowest voltages; evaluate the power supply stability based on the standard deviation of voltage fluctuations; identify voltage mutations based on the voltage change rate; identify overvoltage or undervoltage based on the duration of the peak value when the voltage exceeds a certain threshold; and determine whether the industrial switch has abnormal power supply based on the number of abnormal voltages; S201-3. Evaluate the vibration intensity of the industrial switch based on the average vibration value over a period of time; detect whether the vibration exceeds the safe range based on the highest and lowest vibrations; evaluate the vibration stability based on the standard deviation of the vibration fluctuations; identify the sudden change of vibration based on the rate of change of vibration; identify the over-vibration based on the duration of the peak value of vibration exceeding a certain threshold; determine whether the industrial switch has abnormal vibration based on the abnormal number of vibrations; S201-4. Extract the time and type of events during the startup process for each startup of the switch, and identify abnormal behavior during the startup process; extract the error type and error code for the error occurrence time in the error log, identify the error type, count the frequency of errors within a period of time, and evaluate the stability of the industrial switch; extract the event type and description for the occurrence time of various events in the industrial switch system, identify common and abnormal events, and count the frequency of events within a period of time; S201-5. Evaluate the network load based on the total traffic volume over a period of time; identify the peak hours of network traffic based on the traffic peak over a period of time; evaluate the daily use of the network based on the average traffic volume over a period of time; and evaluate the network performance based on the data transmission rate over a period of time.

[0009] According to step S200, the steps of using the random forest algorithm to train the fault diagnosis model and perform fault prediction and diagnosis on real-time data are as follows: S202-1. Label each record in the data set according to historical data, and divide it into "normal state" and "fault state", and divide the data set into training set and test set in proportion; S202-2. Set hyperparameters of the random forest, where the hyperparameters include the number of decision trees, the maximum depth of each tree, and the minimum number of sample splits, and use the set hyperparameters to initialize the random forest model; S202-3. Randomly sample from the training set to generate multiple groups of sample subsets, each group of subsets is used to train a decision tree; in the splitting process of each decision tree, a part of the features are randomly selected from all the features for splitting to reduce correlation and overfitting; S202-4. Calculate the Gini index or information gain of each split point based on the selected features and sample subsets, select the best split point, and build a decision tree layer by layer until the stop condition is met; S202-5, combining all constructed decision trees into a random forest model, and synthesizing the prediction results of each decision tree through voting; S202-6. Input the test set data into the trained random forest model to perform fault prediction; calculate the accuracy, recall rate and F1 score of the model to evaluate the model performance; generate a confusion matrix to analyze the classification effect of the model and identify misjudgments and missed judgments; S202-7. Extract the importance score of each feature from the random forest model and identify the features that have a significant impact on fault diagnosis; optimize feature extraction and selection based on the results of feature importance analysis and remove redundant features; S202-8. Use grid search to adjust the hyperparameters of random forest and find the best parameter combination; perform k-fold cross validation to evaluate the performance of the model under different parameter combinations and select the optimal model; prevent model overfitting by pruning and adding regularization; update the random forest model based on the optimization results; S202-9. Input the features extracted in real time into the trained random forest model, and the model outputs the prediction results to determine whether there are fault precursors or abnormal behaviors in the current state; the prediction results are fed back to the monitoring system for real-time warning and fault diagnosis.

[0010] According to step S202-8, when using grid search to adjust the hyperparameters of the random forest, first, determine the hyperparameters that need to be adjusted in the random forest model, including the number of trees n_estimators, the maximum tree depth max_depth, the minimum sample split number min_samples_split, the minimum sample leaf number min_samples_leaf, etc. Among them, n_estimators: from 50 to 200, take a value every 50; max_depth: from 10 to 50, take a value every 10; min_samples_split: from 2 to 10, take a value every 2; min_samples_leaf: from 1 to 5, take a value every 1; by traversing the value range of all hyperparameters, all possible parameter combinations are generated; for each parameter combination, the random forest model is trained, and the model performance is evaluated using a predetermined evaluation index.

[0011] When performing k-fold cross-validation, the data set is divided into k subsets of equal size. For each parameter combination, k training and validation are performed. Each time, k-1 of the k subsets are used for training, and the remaining 1 subset is used for validation. After each training and validation, the performance indicators of the model are recorded, and the results of the k validations are averaged to obtain the average performance indicators of the parameter combination. The average performance indicators of all parameter combinations are compared, and the parameter combination with the best performance is selected.

[0012] When building a decision tree, set the maximum tree depth max_depth to limit the growth of the tree and prevent the generation of too deep trees. Set the minimum number of sample splits min_samples_split to ensure that each split has at least a certain number of samples, thereby reducing the complexity of the tree. Set the minimum number of sample leaves min_samples_leaf to ensure that each leaf node has at least a certain number of samples to avoid overfitting. By increasing the number of trees n_estimators, the model is made more stable and the risk of overfitting of a single tree is reduced. Use the bagged Bootstrap sampling technique to use a different sample subset when training each tree, so that each tree has a certain degree of randomness and improves the generalization ability of the model. Introduce feature selection to make the model simpler by limiting the number of features considered at each split max_features, which helps prevent overfitting.

[0013] According to step S300, the visualization interface includes the following areas: the dashboard area displays real-time data, the historical data chart area displays the historical change trend of various indicators, the fault information area displays fault records and detailed information, and the operation area provides user interaction functions; Regularly obtain the latest sensor data, log data, and flow data through real-time communication with the data interface of the industrial switch; update the latest data to the visualization interface, refresh charts and dashboards; allow users to set the automatic refresh frequency; Provides date and time range selection function, users can select a specific time period and view historical data within that period; according to the time range selected by the user, query the corresponding historical data from the database and display the historical change trend of various indicators in the chart area; provides time axis sliding function, users can browse historical data at different time points by sliding the time axis; In the fault information area, the latest fault records are displayed in real time. Users can click on a fault record to view detailed information about the fault. A fault trend analysis chart is provided to show the historical frequency and change trend of different types of faults. If the industrial switch is installed in different geographical locations, a fault location map is provided to show the location and distribution of each fault.

[0014] According to step S400, relevant sensor data, log data and flow data are extracted from the fault record, abnormal features before the fault occurs are analyzed, and the source of the fault is determined; multi-source data association analysis is performed on the sensor data, log data and flow data to identify events and behaviors related to the fault; time series data before the fault occurs is analyzed to identify the time point and path of fault propagation; Use Bayesian networks to build association models between sensor data, log data, and traffic data based on historical data, and represent the associations between different data sources; when a fault is detected, analyze the relationships between data sources through the association model to identify the fault propagation path; determine which data features and events are interrelated, and generate the corresponding time sequence; in the association model, identify key nodes, including abnormal sensors, error logs, and traffic peaks; track the fault propagation path based on key nodes and time series data; Faults are classified according to their types, with each type of fault corresponding to different repair methods and preventive measures; the current fault is matched with the fault records in the historical data to find out the handling experience and results of similar faults; based on the fault type and historical data, repair suggestions are generated using predefined rules; the random forest algorithm is used to automatically generate repair suggestions based on historical fault data and repair records.

[0015] When using the Bayesian network to build an association model, each data feature is defined as a node in the Bayesian network, the prior dependencies between the nodes are defined, and the K2 algorithm is used to learn the dependencies between the nodes from historical data to build the Bayesian network structure. Using historical data and Bayesian estimation, the conditional probability table of each node in the Bayesian network is calculated, and the parameters of the Bayesian network model are optimized. The Bayesian network is used to analyze the fault propagation path, infer the causal relationship between temperature anomalies, voltage fluctuations and vibration anomalies, and determine the fault propagation path. In the real-time fault detection process, the Bayesian network model is dynamically updated according to the latest data, and the propagation path and source of the current fault are determined through real-time reasoning. The accuracy and performance of the Bayesian network model are verified using the test data set, and the effect of fault source identification and propagation path analysis is evaluated. Based on the verification results, the Bayesian network model is optimized, and structural adjustments and parameter updates are performed.

[0016] According to step S500, preliminary fault detection is performed using predefined thresholds and rules, and preliminary fault detection is performed on key features using simple linear regression. When the temperature exceeds the safety range, the voltage fluctuates, or the vibration intensity exceeds the standard, fault detection is triggered; the detected faults and abnormal data are marked, and the marked data is saved locally; normal data is filtered out, and only the marked data and related features are transmitted to the cloud; A two-way data synchronization mechanism is established between the edge device and the cloud server, which adopts an incremental synchronization method and only transmits new and changed data. The edge device and the cloud server share fault detection information, and the suggestions and preventive measures generated by the cloud server are fed back to the edge device through the synchronization mechanism.

[0017] A visual processing system for fault detection of an industrial switch, comprising: Multi-source data fusion module: including: data acquisition unit and data fusion unit; wherein the data acquisition unit is responsible for collecting the operation data of the industrial switch from different data sources, the data sources include sensor data, log data and flow data, and the data fusion unit is responsible for multi-source data fusion, constructing a multi-source data fusion data set, and recording into historical data; Fault diagnosis module: including: feature extraction unit, model training unit and real-time prediction unit; the feature extraction unit is responsible for extracting abnormal features from the historical data of the industrial switch, the model training unit uses the random forest algorithm to train the fault diagnosis model, and the real-time prediction unit is responsible for inputting real-time data into the model for fault prediction and diagnosis; Visualization interface module: including: user interface unit and function interaction unit; the user interface unit designs a visualization interface according to user needs, using several graphics to display the operating status and fault information of the industrial switch, and the function interaction unit provides real-time refresh, history backtracking and fault location functions; Fault analysis module: including: fault tracking unit and Bayesian network unit; the fault tracking unit automatically tracks the fault source and analyzes the fault propagation path after discovering the fault; the Bayesian network unit uses the Bayesian network to establish a correlation model between sensor data, log data and traffic data based on the fault type and historical data, and provides repair suggestions and preventive measures; Edge computing module: includes: preliminary detection unit, data transmission unit and data synchronization unit; the preliminary detection unit analyzes the data that needs preliminary fault detection and performs preliminary fault detection locally on the industrial switch to reduce data transmission delay. The data transmission unit is responsible for filtering out normal data and moving the remaining data and model training tasks to the cloud. The data synchronization unit is responsible for real-time synchronization of edge and cloud data.

[0018] Compared with the prior art, the beneficial effects achieved by the present invention are: The present invention uses a random forest algorithm to train a fault diagnosis model, inputs real-time data into the model, performs fault prediction and diagnosis, provides early warning, and reduces the impact of faults.

[0019] The present invention performs preliminary fault detection locally on the industrial switch, reduces data transmission delays, and improves real-time performance; it moves complex data processing and model training tasks to the cloud for execution, utilizes the powerful computing power of the cloud, and improves processing efficiency.

[0020] After a fault is discovered, the present invention automatically tracks the source of the fault, analyzes the fault propagation path, establishes a data association model based on the fault type and historical data using a Bayesian network, provides repair suggestions and preventive measures, and improves the intelligent level of fault handling. BRIEF DESCRIPTION OF THE DRAWINGS

[0021] The accompanying drawings are used to provide a further understanding of the present invention and constitute a part of the specification. Together with the embodiments of the present invention, they are used to explain the present invention and do not constitute a limitation of the present invention. In the accompanying drawings: Figure 1 It is a schematic diagram of the steps of a visual processing method for fault detection of an industrial switch of the present invention; Figure 2 The present invention is a system structure diagram of a fault detection visualization processing system for industrial switches. DETAILED DESCRIPTION

[0022] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0023] See also Figure 1 and Figure 2 , the present invention provides a technical solution: According to one embodiment of the present invention, Figure 1 A step diagram of a method for visualizing fault detection of an industrial switch is shown in the figure. A method for visualizing fault detection of an industrial switch comprises: S100, collecting operation data of the industrial switch from different data sources, where the data sources include sensor data, log data, and flow data, performing multi-source data fusion, constructing a multi-source data fusion data set, and recording it into historical data; S200, extracts abnormal features from the historical data of industrial switches, uses the random forest algorithm to train the fault diagnosis model, inputs real-time data into the model, and performs fault prediction and diagnosis; S300, based on user needs, designs a visual interface, uses several graphics to display the operating status and fault information of industrial switches; provides real-time refresh, historical backtracking and fault location functions; S400, after discovering a fault, automatically traces the source of the fault, analyzes the fault propagation path, and uses a Bayesian network to establish a correlation model between sensor data, log data, and traffic data based on historical data based on the fault type and historical data, and provides repair suggestions and preventive measures; S500: Analyze the data that needs preliminary fault detection, perform preliminary fault detection locally on the industrial switch to reduce data transmission delay, filter out normal data, move the remaining data and model training tasks to the cloud, and synchronize edge and cloud data in real time.

[0024] According to step S100, sensor data is collected from various sensors installed on the industrial switch, the sensors including temperature sensors, voltage sensors and vibration sensors; log data is collected from the system log of the switch, the system log includes a startup log, an error log and an event log, the log data records the operation history and abnormal events of the switch; flow data is collected from the network interface of the switch, the flow data reflects the operation status and performance of the network; The collected raw data is cleaned to remove noise data and outliers, data from different sources are converted into a unified format, the timestamps of each data source are aligned, multi-source data is fused, and a multi-source data fusion dataset is constructed.

[0025] The experimental environment of this embodiment is as follows: data collection cycle: 1 second; data set time span: 1 month; data source: temperature sensor, voltage sensor, vibration sensor, system log, flow data; data recording format: JSON.

[0026] According to step S100, temperature data is collected through a temperature sensor to record time and temperature values; voltage data is collected through a voltage sensor to record time and voltage values; vibration data is collected through a vibration sensor to record time and vibration intensity values; the startup time of the switch and events during the startup process are recorded through a startup log; error information occurring during the operation of the switch is recorded through an error log; and various events of the industrial switch system are recorded through an event log; all data are unified into a JSON format, and each record contains a timestamp and a data value; When performing multi-source data fusion, it is necessary to extract features from the data. The average value, maximum value, minimum value, standard deviation, rate of change, peak duration, and number of abnormalities of temperature, voltage, and vibration are extracted from sensor data. The time, event type, and error code of key events are extracted from log data. The total amount, peak value, average value, and transmission rate of traffic are extracted from traffic data. A database is created, wherein the table in the database contains all feature fields, and for each time point, the feature values ​​in different data sources are matched and integrated to construct a multi-source data fusion data set.

[0027] Temperature sensor: records the temperature value every second. The data format is as follows: {"timestamp": "2023-06-01T00:00:00Z", "temperature": 25.4}; Voltage sensor: records the voltage value every second. The data format is as follows: {"timestamp": "2023-06-01T00:00:00Z", "voltage": 220}; Vibration sensor: records the vibration intensity every second. The data format is as follows: {"timestamp": "2023-06-01T00:00:00Z", "vibration":0.002}.

[0028] Boot log: records the time of each boot and the events during the process. The data format is such as {"timestamp":"2023-06-01T00:00:00Z", "event": "boot", "detail": "success"}; Error log: records the error information that occurred. The data format is such as {"timestamp": "2023-06-01T00:00:00Z", "error_code":404, "message": "not found"}; Event log: records various events in the system. The data format is such as {"timestamp": "2023-06-01T00:00:00Z", "event_type": "update", "detail": "software update"}.

[0029] Total flow: records the flow value per second, the data format is such as {"timestamp": "2023-06-01T00:00:00Z", "flow_total": 100}; Peak flow: records the peak flow per second, the data format is such as {"timestamp": "2023-06-01T00:00:00Z", "flow_peak": 120}; Average flow: records the average flow per second, the data format is such as {"timestamp": "2023-06-01T00:00:00Z", "flow_average": 80}; Transfer rate: records the transfer rate per second, the data format is such as {"timestamp": "2023-06-01T00:00:00Z", "transfer_rate": 1000}.

[0030] According to step S200, the steps of extracting abnormal features from the historical data of the industrial switch are as follows: S201-1. Evaluate the thermal stability of the industrial switch based on the average temperature over a period of time; detect whether the temperature exceeds the safety range based on the highest and lowest temperatures; evaluate the temperature stability based on the standard deviation of temperature fluctuations; identify temperature mutations based on the temperature change rate; identify overheating based on the duration of the peak value when the temperature exceeds a certain threshold; determine whether the industrial switch has abnormal thermal phenomena based on the number of abnormal temperature occurrences, where the number of abnormal occurrences indicates the number of times the temperature exceeds a preset threshold; S201-2. Evaluate the power supply stability based on the average voltage over a period of time; detect whether the voltage exceeds the safety range based on the highest and lowest voltages; evaluate the power supply stability based on the standard deviation of voltage fluctuations; identify voltage mutations based on the voltage change rate; identify overvoltage or undervoltage based on the duration of the peak value when the voltage exceeds a certain threshold; and determine whether the industrial switch has abnormal power supply based on the number of abnormal voltages; S201-3. Evaluate the vibration intensity of the industrial switch based on the average vibration value over a period of time; detect whether the vibration exceeds the safe range based on the highest and lowest vibrations; evaluate the vibration stability based on the standard deviation of the vibration fluctuations; identify the sudden change of vibration based on the rate of change of vibration; identify the over-vibration based on the duration of the peak value of vibration exceeding a certain threshold; determine whether the industrial switch has abnormal vibration based on the abnormal number of vibrations; S201-4. Extract the time and type of events during the startup process for each startup of the switch, and identify abnormal behavior during the startup process; extract the error type and error code for the error occurrence time in the error log, identify the error type, count the frequency of errors within a period of time, and evaluate the stability of the industrial switch; extract the event type and description for the occurrence time of various events in the industrial switch system, identify common and abnormal events, and count the frequency of events within a period of time; S201-5. Evaluate the network load based on the total traffic volume over a period of time; identify the peak hours of network traffic based on the traffic peak over a period of time; evaluate the daily use of the network based on the average traffic volume over a period of time; and evaluate the network performance based on the data transmission rate over a period of time.

[0031] According to step S200, the steps of using the random forest algorithm to train the fault diagnosis model and perform fault prediction and diagnosis on real-time data are as follows: S202-1. Label each record in the data set according to historical data, and divide it into "normal state" and "fault state", and divide the data set into training set and test set in proportion; S202-2. Set the hyperparameters of the random forest, which include the number of decision trees, the maximum depth of each tree, and the minimum number of sample splits. Use the set hyperparameters to initialize the random forest model. In this embodiment, the number of trees is set to 100, 150, 200, the maximum tree depth is set to 10, 20, 30, the minimum number of sample splits is set to 2, 4, 6, and the minimum number of sample leaves is set to 1, 2, 3.

[0032] S202-3. Randomly sample from the training set to generate multiple groups of sample subsets, each group of subsets is used to train a decision tree; in the splitting process of each decision tree, a part of the features are randomly selected from all the features for splitting to reduce correlation and overfitting; S202-4. Calculate the Gini index or information gain of each split point based on the selected features and sample subsets, select the best split point, and build a decision tree layer by layer until the stop condition is met; S202-5, combining all constructed decision trees into a random forest model, and synthesizing the prediction results of each decision tree through voting; S202-6. Input the test set data into the trained random forest model to perform fault prediction; calculate the accuracy, recall rate and F1 score of the model to evaluate the model performance; generate a confusion matrix to analyze the classification effect of the model and identify misjudgments and missed judgments; S202-7. Extract the importance score of each feature from the random forest model and identify the features that have a significant impact on fault diagnosis; optimize feature extraction and selection based on the results of feature importance analysis and remove redundant features; S202-8. Use grid search to adjust the hyperparameters of random forest and find the best parameter combination; perform k-fold cross validation to evaluate the performance of the model under different parameter combinations and select the optimal model; prevent model overfitting by pruning and adding regularization; update the random forest model based on the optimization results; S202-9. Input the features extracted in real time into the trained random forest model, and the model outputs the prediction results to determine whether there are fault precursors or abnormal behaviors in the current state; the prediction results are fed back to the monitoring system for real-time warning and fault diagnosis.

[0033] According to step S202-8, when using grid search to adjust the hyperparameters of the random forest, first, determine the hyperparameters that need to be adjusted in the random forest model, including the number of trees n_estimators, the maximum tree depth max_depth, the minimum sample split number min_samples_split, the minimum sample leaf number min_samples_leaf, etc. Among them, n_estimators: from 50 to 200, take a value every 50; max_depth: from 10 to 50, take a value every 10; min_samples_split: from 2 to 10, take a value every 2; min_samples_leaf: from 1 to 5, take a value every 1; by traversing the value range of all hyperparameters, all possible parameter combinations are generated; for each parameter combination, the random forest model is trained, and the model performance is evaluated using a predetermined evaluation index.

[0034] When performing 5-fold cross-validation, the data set is divided into 5 subsets of equal size. For each parameter combination, training and validation are performed 5 times, each time using 4 of the 5 subsets for training and the remaining 1 subset for validation. After each training and validation, the performance indicators of the model are recorded, and the results of the 5 validations are averaged to obtain the average performance indicators of the parameter combination. The average performance indicators of all parameter combinations are compared to select the parameter combination with the best performance.

[0035] When building a decision tree, set the maximum tree depth max_depth to limit the growth of the tree and prevent the generation of too deep trees. Set the minimum number of sample splits min_samples_split to ensure that each split has at least a certain number of samples, thereby reducing the complexity of the tree. Set the minimum number of sample leaves min_samples_leaf to ensure that each leaf node has at least a certain number of samples to avoid overfitting. By increasing the number of trees n_estimators, the model is made more stable and the risk of overfitting of a single tree is reduced. Use the bagged Bootstrap sampling technique to use a different sample subset when training each tree, so that each tree has a certain degree of randomness and improves the generalization ability of the model. Introduce feature selection to make the model simpler by limiting the number of features considered at each split max_features, which helps prevent overfitting.

[0036] Model evaluation on the test set: accuracy: 98%, recall: 95%, F1 score: 96%, generate confusion matrix, analyze the classification effect of the model, identify misclassification and omission. Extract the importance score of each feature, identify the features that have a significant impact on fault diagnosis, and optimize feature extraction and selection.

[0037] For feature importance, temperature change rate: 35%, voltage peak duration: 25%, vibration standard deviation: 20%, error log frequency: 15%, and total flow: 5%.

[0038] According to step S300, the visualization interface includes the following areas: the dashboard area displays real-time data, the historical data chart area displays the historical change trend of various indicators, the fault information area displays fault records and detailed information, and the operation area provides user interaction functions; Regularly obtain the latest sensor data, log data, and flow data through real-time communication with the data interface of the industrial switch; update the latest data to the visualization interface, refresh charts and dashboards; allow users to set the automatic refresh frequency; Provides date and time range selection function, users can select a specific time period and view historical data within that period; according to the time range selected by the user, query the corresponding historical data from the database and display the historical change trend of various indicators in the chart area; provides time axis sliding function, users can browse historical data at different time points by sliding the time axis; In the fault information area, the latest fault records are displayed in real time. Users can click on a fault record to view detailed information about the fault. A fault trend analysis chart is provided to show the historical frequency and change trend of different types of faults. If the industrial switch is installed in different geographical locations, a fault location map is provided to show the location and distribution of each fault.

[0039] According to step S400, relevant sensor data, log data and flow data are extracted from the fault record, abnormal features before the fault occurs are analyzed, and the source of the fault is determined; multi-source data association analysis is performed on the sensor data, log data and flow data to identify events and behaviors related to the fault; time series data before the fault occurs is analyzed to identify the time point and path of fault propagation; Use Bayesian networks to build association models between sensor data, log data, and traffic data based on historical data, and represent the associations between different data sources; when a fault is detected, analyze the relationships between data sources through the association model to identify the fault propagation path; determine which data features and events are interrelated, and generate the corresponding time sequence; in the association model, identify key nodes, including abnormal sensors, error logs, and traffic peaks; track the fault propagation path based on key nodes and time series data; Faults are classified according to their types, with each type of fault corresponding to different repair methods and preventive measures; the current fault is matched with the fault records in the historical data to find out the handling experience and results of similar faults; based on the fault type and historical data, repair suggestions are generated using predefined rules; the random forest algorithm is used to automatically generate repair suggestions based on historical fault data and repair records.

[0040] When using the Bayesian network to build an association model, each data feature is defined as a node in the Bayesian network, the prior dependencies between the nodes are defined, and the K2 algorithm is used to learn the dependencies between the nodes from historical data to build the Bayesian network structure. Using historical data and Bayesian estimation, the conditional probability table of each node in the Bayesian network is calculated, and the parameters of the Bayesian network model are optimized. The Bayesian network is used to analyze the fault propagation path, infer the causal relationship between temperature anomalies, voltage fluctuations and vibration anomalies, and determine the fault propagation path. In the real-time fault detection process, the Bayesian network model is dynamically updated according to the latest data, and the propagation path and source of the current fault are determined through real-time reasoning. The accuracy and performance of the Bayesian network model are verified using the test data set, and the effect of fault source identification and propagation path analysis is evaluated. Based on the verification results, the Bayesian network model is optimized, and structural adjustments and parameter updates are performed.

[0041] According to step S500, preliminary fault detection is performed using predefined thresholds and rules, and preliminary fault detection is performed on key features using simple linear regression. When the temperature exceeds the safety range, the voltage fluctuates, or the vibration intensity exceeds the standard, fault detection is triggered; the detected faults and abnormal data are marked, and the marked data is saved locally; normal data is filtered out, and only the marked data and related features are transmitted to the cloud; A two-way data synchronization mechanism is established between the edge device and the cloud server, which adopts an incremental synchronization method and only transmits new and changed data. The edge device and the cloud server share fault detection information, and the suggestions and preventive measures generated by the cloud server are fed back to the edge device through the synchronization mechanism.

[0042] According to another embodiment of the present invention, Figure 2 As shown in the system structure diagram of a fault detection visualization processing system for an industrial switch, a fault detection visualization processing system for an industrial switch is characterized in that it includes: Multi-source data fusion module: including: data acquisition unit and data fusion unit; wherein the data acquisition unit is responsible for collecting the operation data of the industrial switch from different data sources, the data sources include sensor data, log data and flow data, and the data fusion unit is responsible for multi-source data fusion, constructing a multi-source data fusion data set, and recording into historical data; Fault diagnosis module: including: feature extraction unit, model training unit and real-time prediction unit; the feature extraction unit is responsible for extracting abnormal features from the historical data of the industrial switch, the model training unit uses the random forest algorithm to train the fault diagnosis model, and the real-time prediction unit is responsible for inputting real-time data into the model for fault prediction and diagnosis; Visualization interface module: including: user interface unit and function interaction unit; the user interface unit designs a visualization interface according to user needs, using several graphics to display the operating status and fault information of the industrial switch, and the function interaction unit provides real-time refresh, history backtracking and fault location functions; Fault analysis module: including: fault tracking unit and Bayesian network unit; the fault tracking unit automatically tracks the fault source and analyzes the fault propagation path after discovering the fault; the Bayesian network unit uses the Bayesian network to establish a correlation model between sensor data, log data and traffic data based on the fault type and historical data, and provides repair suggestions and preventive measures; Edge computing module: includes: preliminary detection unit, data transmission unit and data synchronization unit; the preliminary detection unit analyzes the data that needs preliminary fault detection and performs preliminary fault detection locally on the industrial switch to reduce data transmission delay. The data transmission unit is responsible for filtering out normal data and moving the remaining data and model training tasks to the cloud. The data synchronization unit is responsible for real-time synchronization of edge and cloud data.

[0043] In this embodiment, after detecting a fault, the fault analysis module automatically tracks the source of the fault and determines the time and location of the fault by analyzing the sensor data and log data. At a certain moment, an abnormal temperature rise is detected, and at the same time, the log record shows a power failure. The fault tracking unit determines that the source of the fault is the power module.

[0044] For fault one, some of the tracing results are as follows: fault type: abnormal voltage fluctuation; fault source: power module; fault propagation path: voltage abnormality -> temperature increase -> system restart; tracing time: 1 second; for fault two, some of the tracing results are as follows: fault type: excessive network traffic; fault source: network interface overload; fault propagation path: traffic peak -> data transmission rate decrease -> network performance decrease; tracing time: 2 seconds.

[0045] For fault 1, some of the processing results of the Bayesian network unit are as follows: Historical data analysis: Based on the data of the past month, the correlation between voltage abnormality and temperature increase was analyzed; Model results: Abnormal voltage fluctuation -> temperature increase -> system restart; Repair suggestions: Check the power module to ensure voltage stability; Increase temperature monitoring; Preventive measures: Install a backup power module and set the temperature alarm threshold.

[0046] For Fault 2, some of the processing results of the Bayesian network unit are as follows: Historical data analysis: Based on the data of the past month, the correlation between excessive network traffic and system performance degradation was analyzed; Model results: Traffic peak -> Data transmission rate decreased -> Network performance decreased; Repair suggestions: Optimize network interface configuration and increase bandwidth; Preventive measures: Monitor traffic peaks and set traffic alarm thresholds.

[0047] It should be noted that, in this article, relational terms such as first and second, etc. are only used to distinguish one entity or operation from another entity or operation, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Moreover, the terms "include", "comprise" or any other variants thereof are intended to cover non-exclusive inclusion, so that a process, method, article or device including a series of elements includes not only those elements, but also other elements not explicitly listed, or also includes elements inherent to such process, method, article or device.

[0048] Finally, it should be noted that the above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Although the present invention has been described in detail with reference to the aforementioned embodiments, those skilled in the art can still modify the technical solutions described in the aforementioned embodiments or replace some of the technical features therein by equivalents. Any modification, equivalent replacement, improvement, etc. made within the spirit and principle of the present invention shall be included in the protection scope of the present invention.

Claims

1. A visual processing method for fault detection of industrial switches, characterized in that: The method comprises: S100, collecting operation data of the industrial switch from different data sources, where the data sources include sensor data, log data, and flow data, performing multi-source data fusion, constructing a multi-source data fusion data set, and recording it into historical data; S200, extracts abnormal features from the historical data of industrial switches, uses the random forest algorithm to train the fault diagnosis model, inputs real-time data into the model, and performs fault prediction and diagnosis; S300, based on user needs, designs a visual interface, uses several graphics to display the operating status and fault information of industrial switches; provides real-time refresh, historical backtracking and fault location functions; S400, after discovering a fault, automatically traces the source of the fault, analyzes the fault propagation path, and uses a Bayesian network to establish a correlation model between sensor data, log data, and traffic data based on historical data based on the fault type and historical data, and provides repair suggestions and preventive measures; S500: Analyze the data that needs preliminary fault detection, perform preliminary fault detection locally on the industrial switch to reduce data transmission delay, filter out normal data, move the remaining data and model training tasks to the cloud, and synchronize edge and cloud data in real time.

2. The method for visualizing fault detection of an industrial switch according to claim 1, characterized in that: According to step S100, sensor data is collected from various sensors installed on the industrial switch, the sensors including temperature sensors, voltage sensors and vibration sensors; log data is collected from the system log of the switch, the system log including startup logs, error logs and event logs, and the log data records the operation history and abnormal events of the switch; Collect traffic data from the network interface of the switch, which reflects the operation status and performance of the network; The collected raw data is cleaned to remove noise data and outliers, data from different sources are converted into a unified format, the timestamps of each data source are aligned, multi-source data is fused, and a multi-source data fusion dataset is constructed.

3. The method for visualizing fault detection of an industrial switch according to claim 2, characterized in that: According to step S100, temperature data is collected through a temperature sensor to record time and temperature values, voltage data is collected through a voltage sensor to record time and voltage values, vibration data is collected through a vibration sensor to record time and vibration intensity values, startup time of the switch and events during the startup process are recorded through a startup log, error information occurring during the operation of the switch is recorded through an error log, and various events of the industrial switch system are recorded through an event log; All data is unified into JSON format, and each record contains a timestamp and data value; When performing multi-source data fusion, it is necessary to extract features from the data. The average value, maximum value, minimum value, standard deviation, rate of change, peak duration, and number of abnormalities of temperature, voltage, and vibration are extracted from sensor data. The time, event type, and error code of key events are extracted from log data. The total amount, peak value, average value, and transmission rate of traffic are extracted from traffic data. A database is created, wherein the table in the database contains all feature fields, and for each time point, the feature values ​​in different data sources are matched and integrated to construct a multi-source data fusion data set.

4. The method for visualizing fault detection of an industrial switch according to claim 1, characterized in that: According to step S200, the steps of extracting abnormal features from the historical data of the industrial switch are as follows: S201-1. Evaluate the thermal stability of the industrial switch based on the average temperature over a period of time; detect whether the temperature exceeds the safety range based on the highest and lowest temperatures; evaluate the temperature stability based on the standard deviation of temperature fluctuations; identify temperature mutations based on the temperature change rate; identify overheating based on the duration of the peak value when the temperature exceeds a certain threshold; determine whether the industrial switch has abnormal thermal phenomena based on the number of abnormal temperature occurrences, where the number of abnormal occurrences indicates the number of times the temperature exceeds a preset threshold; S201-2, evaluating the power supply stability based on the average voltage over a period of time; The highest and lowest voltages are used to detect whether the voltage exceeds the safety range; the standard deviation of voltage fluctuations is used to evaluate the power supply stability; the voltage change rate is used to identify voltage mutations; the duration of the peak voltage exceeding a certain threshold is used to identify overvoltage or undervoltage; the number of abnormal voltage occurrences is used to determine whether there is abnormal power supply in industrial switches; S201-3. Evaluate the vibration intensity of the industrial switch based on the average vibration value over a period of time; detect whether the vibration exceeds the safe range based on the highest and lowest vibrations; evaluate the vibration stability based on the standard deviation of the vibration fluctuations; identify the sudden change of vibration based on the rate of change of vibration; identify the over-vibration based on the duration of the peak value of vibration exceeding a certain threshold; determine whether the industrial switch has abnormal vibration based on the abnormal number of vibrations; S201-4. Extract the occurrence time and type of events during the startup process based on the startup time of each switch, and identify abnormal behaviors during the startup process; Based on the time of error occurrence in the error log, extract the error type and error code, identify the error type, count the frequency of error occurrence within a period of time, and evaluate the stability of the industrial switch; based on the time of occurrence of various events in the industrial switch system, extract the event type and description, identify common and abnormal events, and count the frequency of event occurrence within a period of time; S201-5. Evaluate the network load based on the total traffic volume over a period of time; identify the peak hours of network traffic based on the traffic peak over a period of time; evaluate the daily use of the network based on the average traffic volume over a period of time; and evaluate the network performance based on the data transmission rate over a period of time.

5. The method for visualizing fault detection of an industrial switch according to claim 4, characterized in that: According to step S200, the steps of using the random forest algorithm to train the fault diagnosis model and perform fault prediction and diagnosis on real-time data are as follows: S202-1. Label each record in the data set according to historical data, and divide it into "normal state" and "fault state", and divide the data set into training set and test set in proportion; S202-2. Set hyperparameters of the random forest, where the hyperparameters include the number of decision trees, the maximum depth of each tree, and the minimum number of sample splits, and use the set hyperparameters to initialize the random forest model; S202-3. Randomly sample from the training set to generate multiple groups of sample subsets, each group of subsets is used to train a decision tree; in the splitting process of each decision tree, a part of the features are randomly selected from all the features for splitting to reduce correlation and overfitting; S202-4. Calculate the Gini index or information gain of each split point based on the selected features and sample subsets, select the best split point, and build a decision tree layer by layer until the stop condition is met; S202-5, combining all constructed decision trees into a random forest model, and synthesizing the prediction results of each decision tree through voting; S202-6. Input the test set data into the trained random forest model to perform fault prediction; Calculate the model's accuracy, recall, and F1 score to evaluate model performance; generate a confusion matrix to analyze the model's classification effect and identify misclassifications and missed classifications; S202-7. Extract the importance score of each feature from the random forest model and identify the features that have a significant impact on fault diagnosis; According to the results of feature importance analysis, optimize feature extraction and selection and remove redundant features; S202-8. Use grid search to adjust the hyperparameters of random forest and find the best parameter combination; perform k-fold cross validation to evaluate the performance of the model under different parameter combinations and select the optimal model; prevent model overfitting by pruning and adding regularization; update the random forest model based on the optimization results; S202-9. Input the features extracted in real time into the trained random forest model, and the model outputs the prediction results to determine whether there are fault precursors or abnormal behaviors in the current state; the prediction results are fed back to the monitoring system for real-time warning and fault diagnosis.

6. The method for visualizing fault detection of an industrial switch according to claim 1, characterized in that: According to step S300, the visualization interface includes the following areas: the dashboard area displays real-time data, the historical data chart area displays the historical change trend of various indicators, the fault information area displays fault records and detailed information, and the operation area provides user interaction functions; Regularly obtain the latest sensor data, log data, and flow data through real-time communication with the data interface of the industrial switch; update the latest data to the visualization interface, refresh charts and dashboards; allow users to set the automatic refresh frequency; Provides date and time range selection function, users can select a specific time period and view historical data within that period; according to the time range selected by the user, query the corresponding historical data from the database and display the historical change trend of various indicators in the chart area; provides time axis sliding function, users can browse historical data at different time points by sliding the time axis; In the fault information area, the latest fault records are displayed in real time. Users can click on a fault record to view the detailed information of the fault. Provides fault trend analysis charts to show the historical frequency and changing trends of different types of faults; if the industrial switches are installed in different geographical locations, provides fault location maps to show the location and distribution of each fault.

7. The method for visualizing fault detection of an industrial switch according to claim 1, characterized in that: According to step S400, relevant sensor data, log data and flow data are extracted from the fault record, abnormal characteristics before the fault occurs are analyzed, and the source of the fault is determined; multi-source data association analysis is performed by combining the sensor data, log data and flow data to identify events and behaviors related to the fault; Analyze the time series data before the fault occurs to identify the time point and path of fault propagation; Use Bayesian networks to build association models between sensor data, log data, and traffic data based on historical data, and represent the associations between different data sources. When a fault is detected, use the association model to analyze the relationships between data sources and identify the fault propagation path. Determine which data features and events are related to each other and generate the corresponding time sequence; In the correlation model, key nodes are identified, including abnormal sensors, error logs, and traffic peaks; fault propagation paths are tracked based on key nodes and time series data; According to the fault type, the faults are classified, and each fault type corresponds to different repair methods and preventive measures; Match the current fault with the fault records in the historical data to find out the handling experience and results of similar faults; generate repair suggestions based on the fault type and historical data using predefined rules; Use the random forest algorithm to automatically generate repair suggestions based on historical failure data and repair records.

8. The method for visualizing fault detection of an industrial switch according to claim 7, characterized in that: When using the Bayesian network to build an association model, each data feature is defined as a node in the Bayesian network, the prior dependencies between the nodes are defined, and the K2 algorithm is used to learn the dependencies between the nodes from historical data to build the Bayesian network structure. Using historical data and Bayesian estimation, we calculated the conditional probability table of each node in the Bayesian network and optimized the parameters of the Bayesian network model. We used the Bayesian network to analyze the fault propagation path, infer the causal relationship between temperature anomalies, voltage fluctuations, and vibration anomalies, and determine the fault propagation path. In the real-time fault detection process, the Bayesian network model is dynamically updated according to the latest data, and the propagation path and source of the current fault are determined through real-time reasoning; The test data set is used to verify the accuracy and performance of the Bayesian network model, and to evaluate the effectiveness of fault source identification and propagation path analysis. Based on the verification results, the Bayesian network model is optimized, and structural adjustments and parameter updates are performed.

9. The method for visualizing fault detection of an industrial switch according to claim 1, characterized in that: According to step S500, preliminary fault detection is performed using predefined thresholds and rules, and preliminary fault detection is performed on key features using simple linear regression, and fault detection is triggered when the temperature exceeds a safe range, the voltage fluctuates, or the vibration intensity exceeds the standard; Mark the detected faults and abnormal data and save the marked data locally; Filter out normal data and only transmit labeled data and relevant features to the cloud; A two-way data synchronization mechanism is established between the edge device and the cloud server, which adopts an incremental synchronization method and only transmits new and changed data. The edge device and the cloud server share fault detection information, and the suggestions and preventive measures generated by the cloud server are fed back to the edge device through the synchronization mechanism.

10. A visual processing system for fault detection of industrial switches, using a visual processing method for fault detection of industrial switches according to any one of claims 1 to 9, characterized in that: include: Multi-source data fusion module: including: data acquisition unit and data fusion unit; wherein the data acquisition unit is responsible for collecting the operation data of the industrial switch from different data sources, the data sources include sensor data, log data and flow data, and the data fusion unit is responsible for multi-source data fusion, constructing a multi-source data fusion data set, and recording into historical data; Fault diagnosis module: including: feature extraction unit, model training unit and real-time prediction unit; the feature extraction unit is responsible for extracting abnormal features from the historical data of the industrial switch, the model training unit uses the random forest algorithm to train the fault diagnosis model, and the real-time prediction unit is responsible for inputting real-time data into the model for fault prediction and diagnosis; Visualization interface module: including: user interface unit and function interaction unit; the user interface unit designs a visualization interface according to user needs, using several graphics to display the operating status and fault information of the industrial switch, and the function interaction unit provides real-time refresh, history backtracking and fault location functions; Fault analysis module: including: fault tracking unit and Bayesian network unit; the fault tracking unit automatically tracks the fault source and analyzes the fault propagation path after discovering the fault; the Bayesian network unit uses the Bayesian network to establish a correlation model between sensor data, log data and traffic data based on the fault type and historical data, and provides repair suggestions and preventive measures; Edge computing module: includes: preliminary detection unit, data transmission unit and data synchronization unit; the preliminary detection unit analyzes the data that needs preliminary fault detection and performs preliminary fault detection locally on the industrial switch to reduce data transmission delay. The data transmission unit is responsible for filtering out normal data and moving the remaining data and model training tasks to the cloud. The data synchronization unit is responsible for real-time synchronization of edge and cloud data.

Images