Equipment fault prediction method based on intelligent equipment health degree analysis

Abstract

The invention relates to a health monitoring and fault prediction technology of intelligent equipment, in particular to an equipment fault prediction method based on health degree analysis of the intelligent equipment. According to the method, multiple sensors are embedded in the intelligent device, data of the operation state of the device are continuously collected, and primary processing and abnormal value elimination are carried out through an edge computing node. The processed data is transmitted to a cloud server, and machine learning models such as a deep neural network, a Bayesian classifier and a support vector machine are utilized to analyze the equipment health degree trend, and a potential fault mode is identified. And intelligently generating a fault prediction report according to an analysis result, and notifying an operation and maintenance team to take measures through an operation and maintenance management system. In addition, the system further comprises a historical fault database, a scoring system, an online review group and other functional modules, the accuracy and reliability of prediction are ensured, and the equipment maintenance efficiency and safety are improved.

Description

Technical Field

[0001] The present application relates to health monitoring and fault prediction technology for smart devices, and in particular to a device fault prediction method based on health analysis of smart devices. Background Art

[0002] Equipment failure prediction is an important part of industrial production and maintenance management, aiming to reduce unplanned downtime, extend equipment life and reduce maintenance costs by identifying potential failures at an early stage. With the development of smart manufacturing and Internet of Things technologies, more and more smart devices are deployed on production lines. These devices not only improve production efficiency, but also provide a rich data source, making failure prediction possible. However, traditional failure prediction methods often rely on manual experience and regular inspections, making it difficult to achieve real-time monitoring and accurate prediction of equipment status.

[0003] At present, in order to meet this challenge, the methods commonly used in the industry include but are not limited to: 1) rule-based fault detection, that is, pre-defining a series of fault judgment rules and identifying faults by matching the collected data; 2) statistical analysis, using statistical methods to analyze historical data, find out the laws of fault occurrence, and then make predictions; 3) expert systems, using the knowledge base and experience of domain experts, combined with actual data for reasoning and decision-making. Although these methods can improve the accuracy of fault prediction to a certain extent, they still have certain limitations when facing complex and changeable actual working conditions.

[0004] Existing fault prediction methods mainly rely on a single type of sensor data and lack the ability to comprehensively analyze multiple sensor data, resulting in insufficient accuracy and timeliness of fault prediction. In addition, traditional methods usually require a lot of manual configuration and debugging, and cannot adapt to the rapidly changing production environment. Therefore, there is an urgent need for a fault prediction method that can integrate multiple sensor data and use advanced data analysis technology to improve the effectiveness of equipment health management. Summary of the invention

[0005] The purpose of this application is to overcome the above technical problems and provide a device failure prediction method based on intelligent device health analysis. A device failure prediction method based on intelligent device health analysis includes the following steps: Embed a health monitoring sensor in the smart device, wherein the health monitoring sensor includes at least one of a temperature and humidity sensor, a vibration sensor, a current sensor, and a pressure sensor, and is used to continuously collect various indicator data reflecting the operating status of the device; Receive and process the collected data through edge computing nodes, perform preliminary filtering and remove outliers; Including: sorting the received data by timestamp, eliminating duplicate data, marking the data exceeding the preset threshold as an outlier and excluding it, and retaining the valid data; the edge computing node adopts a sliding window mechanism to check the difference between adjacent data points one by one, and if the difference exceeds the set threshold, it is marked as an outlier; The processed data is transmitted to a cloud server, and the equipment health trend is analyzed using a machine learning model to identify potential failure modes. The machine learning model uses a deep neural network and continuously optimizes the model parameters through a back-propagation algorithm. The method comprises: inputting the processed data into a deep neural network, generating an intermediate layer representation through forward propagation, and adjusting the weights and optimizing the model parameters through back propagation until a preset convergence standard is reached; the deep neural network adopts a structure combining a convolutional layer and a fully connected layer to extract the spatial and temporal features of the data; Based on the analysis results, a fault prediction report is intelligently generated, and the operation and maintenance team is notified to take measures through the operation and maintenance management system; It includes: determining the possibility of a fault occurring based on the probability distribution of the model output, generating a fault prediction report when the probability exceeds a preset threshold, and notifying the operation and maintenance team to take corresponding measures through the message push function of the operation and maintenance management system; the operation and maintenance management system has a log recording function to record the time, object and content of each notification.

[0006] By adopting the above technical solutions, comprehensive monitoring of the health of smart devices and fault prediction can be achieved. First, by embedding various types of health monitoring sensors in smart devices, various indicator data reflecting the operating status of the equipment can be continuously collected to ensure the diversity and accuracy of the data. Secondly, the edge computing nodes are used to perform preliminary filtering and outlier removal on the received data to ensure the quality and validity of the data. Then, the processed data is transmitted to the cloud server, and the health trend of the equipment is analyzed with the help of a deep neural network model to identify potential failure modes, and the model parameters are continuously optimized through the back propagation algorithm to improve the accuracy and reliability of fault prediction. Finally, a fault prediction report is automatically generated based on the analysis results, and the operation and maintenance team is promptly notified through the operation and maintenance management system to take measures, which effectively reduces the incidence of equipment failures and improves the reliability and operational efficiency of the equipment.

[0007] Preferably, the method further comprises the following steps: Collect all fault records and detailed maintenance reports of similar equipment in the past ten years to build a detailed historical fault database, which is stored on a cloud server and supports multi-user access and data sharing, facilitating collaboration between operation and maintenance teams in different regions. Including: obtaining fault records and maintenance reports from operation and maintenance teams in various places, organizing them into a unified format, and uploading them to the historical fault database on the cloud server; the historical fault database adopts distributed storage technology to ensure data consistency and availability; Design a specialized algorithm to extract key parameters from each record and convert them into a standardized format that can be used for model training; It includes: parsing text information in fault records, extracting keywords and numerical parameters, and converting them into a machine-readable standard format; the algorithm uses natural language processing technology to automatically identify and extract key parameters; Update the model regularly to incorporate the latest failure cases and keep the prediction logic in line with the actual changes; The method includes: extracting the latest fault cases from the historical fault database at fixed time intervals, retraining the machine learning model, and optimizing the model parameters; using an incremental learning method in the model update process to gradually incorporate new data and reduce training time.

[0008] By adopting the above technical solutions, a detailed historical fault database is built to support multi-user access and data sharing, which facilitates collaboration between operation and maintenance teams in different regions. Fault records and maintenance reports are obtained from operation and maintenance teams in various regions, organized into a unified format, and uploaded to the historical fault database on the cloud server. Distributed storage technology is used to ensure data consistency and availability, and improve data reliability and security. Special algorithms are designed to parse text information in fault records, extract keywords and numerical parameters, and convert them into a machine-readable standard format. Natural language processing technology is used to automatically identify and extract key parameters, which improves the efficiency and accuracy of data processing. The model is updated regularly, and the latest fault cases are extracted from the historical fault database at fixed time intervals. The machine learning model is retrained, the model parameters are optimized, and the incremental learning method is used to gradually integrate new data, which reduces training time and ensures the real-time and accuracy of the model.

[0009] Preferably, the method further comprises the following steps: Establish an online review team composed of senior maintenance engineers, who are responsible for secondary verification of emergency alarms issued by the model; Including: when the model prediction results show a high-risk fault, an alarm is automatically triggered to notify the online review team members to log in to the system to view detailed information; the online review team members can remotely access the system through a mobile terminal to view the alarm information and related data in real time; The system design includes a scoring system that comprehensively assesses the risk level based on multiple dimensions such as failure probability, equipment type, and impact range; Including: according to the probability value output by the model, the importance of the equipment and the impact range of the fault, the corresponding weights are assigned respectively to calculate the total risk score; the scoring system adopts the hierarchical analysis method to ensure the objectivity and accuracy of the scoring; When the prediction score exceeds the predetermined threshold and is approved by the review team, an early warning notification is pushed to the on-site operation and maintenance personnel; Including: If the risk score is higher than the preset threshold and approved by the review team, an early warning notification will be sent to the designated operation and maintenance personnel through the message push function of the operation and maintenance management system; the early warning notification will include the specific fault location, expected time of occurrence and related recommended measures.

[0010] By adopting the above technical solutions, an online review team composed of senior maintenance engineers was established, which can perform secondary verification when the model predicts a high-risk fault to ensure the accuracy of the alarm. At the same time, the system has designed a scoring system that comprehensively considers multiple dimensions such as fault probability, equipment type and impact range to objectively evaluate the risk level. When the risk score exceeds the predetermined threshold and is reviewed and approved by the review team, an early warning notification containing the specific fault location, expected time of occurrence and related recommended measures will be pushed to the on-site operation and maintenance personnel, thereby improving the timeliness and effectiveness of fault handling, reducing unnecessary maintenance work caused by misjudgment, and improving the reliability and safety of equipment operation.

[0011] Preferably, the health monitoring sensor further includes at least one of an acoustic wave sensor and an infrared temperature sensor to enhance the monitoring capability of the equipment's operating status: the acoustic wave sensor is used to detect the noise level inside the equipment, and the infrared temperature sensor is used to measure the temperature distribution on the equipment surface, and the acquired data is transmitted together with other sensor data to the edge computing node for processing; the acoustic wave sensor is made of piezoelectric ceramic material; and the infrared temperature sensor uses a thermopile detector.

[0012] By adopting the above technical solutions, the monitoring capability of the equipment operation status is enhanced. The acoustic wave sensor can detect the noise level inside the equipment, which helps to detect problems such as wear and looseness of mechanical parts at an early stage; the infrared temperature sensor can measure the temperature distribution on the surface of the equipment, provide more intuitive thermal map information, and help identify hot spots and poor heat dissipation. These data are transmitted to the edge computing node together with the data from the temperature and humidity sensor, vibration sensor, current sensor and pressure sensor for processing, which improves the accuracy and reliability of fault prediction.

[0013] Preferably, the edge computing node has a self-diagnosis function, can automatically detect its own performance status and start a redundant backup mechanism when necessary to ensure the continuity and reliability of data processing; Including: the edge computing node regularly checks its own memory usage, CPU utilization and network connection status. Once an abnormality is found, it immediately switches to the backup node to continue processing data; the self-diagnosis function adopts a heartbeat detection mechanism, sends heartbeat signals regularly, and monitors the status of each component.

[0014] By adopting the above technical solutions, the self-diagnosis function of the edge computing node can monitor its own performance status in real time to ensure the continuity and reliability of data processing. This function regularly checks the memory usage, CPU utilization and network connection status. Once an abnormality is found, it immediately switches to the backup node to continue processing data, thereby avoiding data processing interruptions caused by single point failures and improving the stability and robustness of the entire system.

[0015] Preferably, the machine learning model not only uses a deep neural network, but also combines a Bayesian classifier and a support vector machine. The processed data is input into the deep neural network, the Bayesian classifier and the support vector machine in sequence, and the output results of the three models are integrated through a voting mechanism as the final fault prediction conclusion; the Bayesian classifier adopts a naive Bayes algorithm, which is suitable for learning small sample data; the support vector machine adopts a linear kernel function, which is suitable for rapid classification of large-scale data.

[0016] By adopting the above technical solutions, the accuracy and robustness of equipment fault prediction can be significantly improved. Specifically, by combining three different machine learning models, deep neural networks, Bayesian classifiers, and support vector machines, their respective advantages can be fully utilized. Deep neural networks are good at capturing complex data features, Bayesian classifiers are suitable for processing small sample data, and support vector machines perform well in large-scale data classification. By combining the output results of the three models through a voting mechanism, the deviations and errors that may be caused by a single model can be effectively reduced, thereby improving the reliability and stability of fault prediction.

[0017] Preferably, the historical fault database further includes relevant information such as the time, location, and environmental conditions of the fault, so as to more fully understand the cause of the fault; Including: When collating fault records, in addition to extracting key parameters, it is also necessary to record the specific time and location of the fault, as well as the environmental conditions at the time (such as temperature, humidity, etc.); the environmental condition information uses data from meteorological stations to ensure the accuracy and timeliness of the data.

[0018] By adopting the above technical solutions, the historical fault database not only contains the key parameters of the fault, but also adds information such as the time, location and environmental conditions of the fault, making the fault cause analysis more comprehensive and accurate. This additional information helps to better understand the background and context of the fault, thereby improving the accuracy and reliability of fault prediction. At the same time, the use of data from meteorological stations ensures the accuracy and timeliness of environmental condition information, further improving the data quality of the entire system.

[0019] Preferably, the scoring system also includes equipment importance and maintenance cost factors. When calculating the risk score, the importance and maintenance cost of the equipment are taken into consideration, and a higher weight is given to failures of important equipment or high maintenance costs; the equipment importance is measured by the frequency of use of the equipment and the downtime loss, and the maintenance cost is estimated by the average maintenance cost and maintenance difficulty.

[0020] By adopting the above technical solutions, the scoring system is more comprehensive and accurate. First, considering the importance of the equipment, the frequency of use and downtime loss of the equipment are used to ensure that key equipment is focused on, which improves the priority and response speed of fault prediction. Secondly, the maintenance cost factor is added, and the average maintenance cost and maintenance difficulty are used to estimate, making risk assessment more economical and reasonable, avoiding unnecessary waste of resources. In summary, this scoring system can more effectively identify and respond to high-risk faults, and improve the reliability and operation and maintenance efficiency of the system.

[0021] A device failure prediction system based on intelligent device health analysis, used for a device failure prediction method based on intelligent device health analysis, comprising: A smart device with a built-in health monitoring sensor for continuously collecting data on the device's operating status, wherein the health monitoring sensor includes at least one of a temperature and humidity sensor, a vibration sensor, a current sensor, a pressure sensor, an acoustic wave sensor, and an infrared temperature sensor; The edge computing node is connected to the smart device and is used to receive and process data, perform preliminary filtering and remove outliers. The edge computing node adopts a distributed architecture, and multiple edge computing nodes work together to share the data processing load and improve data processing efficiency. The edge computing nodes communicate with each other through message queues to ensure data synchronization and consistency. The cloud server communicates with the edge computing node to store and analyze the processed data and generate a fault prediction report. The machine learning model uses a deep neural network and continuously optimizes the model parameters through a back propagation algorithm. The cloud server adopts a microservice architecture to support elastic scaling and high concurrent processing. The operation and maintenance management system is connected to the cloud server and is used to receive fault prediction reports and notify the operation and maintenance team to take measures. The operation and maintenance management system receives the fault prediction reports generated by the cloud server and promptly notifies the operation and maintenance team to take corresponding measures through the message push function; the operation and maintenance management system has authority management and auditing functions to ensure the security and compliance of operations.

[0022] By adopting the above technical solutions, various health monitoring sensors in smart devices can continuously collect multiple indicator data of the device's operating status. Edge computing nodes achieve efficient data processing and synchronization through distributed architecture and message queues. Cloud servers use deep neural networks to predict faults and generate reports. The operation and maintenance management system ensures operational safety through permission management and auditing functions. The system not only improves the accuracy and response speed of fault prediction, but also enhances the reliability and stability of the overall system, effectively reducing the economic losses and operational interruption risks caused by equipment failures.

[0023] Preferably, the operation and maintenance management system further includes: The fault database module is used to store historical fault records and maintenance reports. Fault records and maintenance reports are obtained from operation and maintenance teams in various places, sorted into a unified format, and uploaded to the fault database module on the cloud server. The fault database module adopts a relational database to support complex queries and statistical analysis. The prediction model update module is used to regularly update the machine learning model, incorporate the latest fault cases, extract the latest fault cases from the fault database module at regular intervals, retrain the machine learning model, and optimize the model parameters; the prediction model update module uses an automated script to simplify the model update process; The expert review module is used for the online review team members to conduct secondary verification of the emergency alarm issued by the model. When the model prediction results show a high-risk failure, an alarm is automatically triggered to notify the online review team members to log in to the system to view detailed information; the expert review module has task allocation and progress tracking functions to ensure the efficient execution of the review work; The self-diagnosis module is used to monitor the working status of each component of the system and automatically start the redundant backup mechanism when an abnormality is found. The self-diagnosis module regularly checks the status of each component of the system. Once an abnormality is found, it immediately switches to the backup component to continue normal operation; the self-diagnosis module adopts real-time monitoring technology and alarm mechanism to ensure the stable operation of the system.

[0024] By adopting the above technical solutions, the operation and maintenance management system can not only efficiently store and manage historical fault records and maintenance reports, but also regularly update the machine learning model to ensure that the prediction logic keeps up with the actual changes. At the same time, the system supports the online review team to conduct secondary verification of high-risk fault alarms, which improves the accuracy and reliability of fault judgment. In addition, the self-diagnosis module monitors the status of each component of the system in real time and activates the redundant backup mechanism in time, which ensures the stable operation of the system and improves the overall operation and maintenance efficiency.

[0025] In summary, the present application includes at least one of the following beneficial technical effects: 1. By embedding a variety of health monitoring sensors (such as temperature and humidity sensors, vibration sensors, current sensors, and pressure sensors) in smart devices, multi-dimensional data on the operating status of the equipment can be collected in real time to improve the accuracy and comprehensiveness of fault prediction; 2. Use edge computing nodes to perform preliminary data processing and remove outliers to ensure the quality of data transmitted to the cloud, while reducing the burden on cloud servers and improving the response speed and stability of the overall system; 3. Combining deep neural networks with other machine learning models (such as Bayesian classifiers and support vector machines) for comprehensive analysis can not only extract the spatial and temporal characteristics of the data, but also obtain more reliable fault prediction results through a voting mechanism, significantly improving the prediction accuracy. BRIEF DESCRIPTION OF THE DRAWINGS

[0026] Figure 1 It is a flowchart of the device failure prediction method based on smart device health analysis in Example 1 of the present application.

[0027] Figure 2 It is a principle block diagram of the equipment failure prediction system based on intelligent equipment health analysis in Example 1 of the present application.

[0028] Figure 3 It is a principle block diagram of the operation management system in Example 1 of the present application. DETAILED DESCRIPTION

[0029] The following is combined with Figure 1-Figure 3 This application is described in further detail. Example

[0030] Reference Figure 1 The device failure prediction method based on smart device health analysis provided in the embodiment of the present application includes the following steps: S1. Embed a health monitoring sensor in a smart device. The health monitoring sensor includes at least one of a temperature and humidity sensor, a vibration sensor, a current sensor, and a pressure sensor, and is used to continuously collect various indicator data reflecting the operating status of the device.

[0031] Among them, these sensors can be widely used in various mechanical equipment, such as CNC machine tools, industrial robots, compressors, etc. For example, the temperature and humidity sensor can be resistive or capacitive, used to detect the ambient temperature and humidity around the equipment; the vibration sensor can be an accelerometer or a gyroscope, used to detect the vibration of the equipment; the current sensor can be a Hall effect sensor or a mutual inductor, used to detect the current change of the equipment; the pressure sensor can be a piezoresistive or capacitive, used to detect the pressure condition of the equipment. These sensors transmit data to the edge computing node through wireless communication modules (such as Wi-Fi, Bluetooth or Zigbee).

[0032] S2. Receive and process the collected data through the edge computing node, perform preliminary filtering and remove outliers, including: sorting the received data by timestamp, removing duplicate data, marking the data that exceeds the preset threshold as outliers and excluding it, and retaining valid data; the edge computing node uses a sliding window mechanism to check the differences between adjacent data points one by one, and if the difference exceeds the set threshold, it is marked as an outlier.

[0033] Among them, the edge computing node is a miniaturized computer system, which is usually installed close to the device to reduce data transmission delay. The edge computing node uses a sliding window mechanism to check the differences between adjacent data points one by one. For example, if the vibration data in a certain period of time suddenly increases and exceeds the set threshold, it will be marked as an outlier and excluded. In addition, the edge computing node will also deduplicate the data to ensure the validity of the data. The edge computing node can also be equipped with a self-diagnosis function to regularly check its own memory usage, CPU utilization and network connection status. Once an abnormality is found, it will immediately switch to the backup node to continue processing the data. Specifically, the edge computing node will use the sliding window mechanism to compare the change amplitude of adjacent data points during data processing. If the change amplitude exceeds the preset threshold \(\Delta T\), it will be determined as an outlier and eliminated. Here, \(\Delta T\) can be set according to actual needs, such as \(\Delta T= 0.5g\) for vibration data. At the same time, the edge computing node will record each excluded outlier and its corresponding original data for subsequent audit and review.

[0034] S3. The processed data is transmitted to the cloud server, and the machine learning model is used to analyze the equipment health trend and identify potential failure modes. The machine learning model uses a deep neural network and continuously optimizes the model parameters through the back propagation algorithm; including: inputting the processed data into the deep neural network, generating the intermediate layer representation through forward propagation, and then adjusting the weights through back propagation to optimize the model parameters until the preset convergence standard is reached; the deep neural network adopts a structure that combines convolutional layers and fully connected layers to extract the spatial and temporal characteristics of the data.

[0035] The cloud server is a high-performance server cluster that supports large-scale data storage and processing. The machine learning model uses a deep neural network and continuously optimizes the model parameters through the back-propagation algorithm. The deep neural network uses a structure that combines convolutional layers and fully connected layers to extract the spatial and temporal features of the data. For example, the time series features of vibration data can be captured through the convolutional layer, and multiple sensor data can be integrated through the fully connected layer to generate a comprehensive health assessment result.

[0036] In addition, the cloud server also supports a multi-tenant architecture, allowing multiple companies or departments to share resources and improve resource utilization. Specifically, the cloud server aggregates data from multiple edge computing nodes and processes them in batches to improve data processing efficiency. At the same time, the cloud server performs quality control on each batch of data to ensure data consistency and accuracy. During the training of a deep neural network, the error \(E\) is calculated as follows: \[ E = \frac{1}{2} \sum_{i=1}^{n} (y_i t_i)^2 \] where \(y_i\) is the predicted value output by the model, \(t_i\) is the actual target value, and \(n\) is the number of samples. Through the back-propagation algorithm, the weight matrix \(W\) and the bias term \(b\) are adjusted to minimize the error.

[0037] S4. Based on the analysis results, a fault prediction report is intelligently generated, and the operation and maintenance team is notified to take measures through the operation and maintenance management system; including: determining the possibility of a fault based on the probability distribution of the model output, and generating a fault prediction report when the probability exceeds the preset threshold, and notifying the operation and maintenance team to take corresponding measures through the message push function of the operation and maintenance management system; the operation and maintenance management system has a log recording function to record the time, object and content of each notification.

[0038] Among them, the operation and maintenance management system is an integrated operation platform that supports the generation and push of fault prediction reports. When the probability distribution output by the model shows that the probability of a fault is high, a fault prediction report will be generated, and the operation and maintenance team will be notified to take corresponding measures through the message push function of the operation and maintenance management system. The operation and maintenance management system has a log recording function to record the time, object and content of each notification, which is convenient for subsequent auditing and tracing. Specifically, the operation and maintenance management system will determine whether a fault prediction report needs to be generated based on the fault probability distribution output by the model. If the fault probability exceeds the preset threshold \(P_t\), the system will automatically generate a report containing the fault location, expected time of occurrence and related recommended measures, and notify the operation and maintenance team by email, SMS or mobile phone application. The operation and maintenance team can take preventive measures quickly based on the information in the report to avoid equipment failure. The preset threshold \(P_t\) here can be set according to the importance of different devices. For example, \(P_t = 0.8\) means that the system will only generate a report when the probability of failure is greater than 80%.

[0039] Reference Figure 2 and Figure 3 The present application also discloses a device failure prediction system based on smart device health analysis, including the following: Smart devices: built-in health monitoring sensors to continuously collect data on the operating status of the device. Health monitoring sensors include at least one of temperature and humidity sensors, vibration sensors, current sensors, pressure sensors, acoustic wave sensors, and infrared temperature sensors. For example, the temperature and humidity sensor can be resistive or capacitive, used to detect the ambient temperature and humidity around the device; the vibration sensor can be an accelerometer or a gyroscope, used to detect the vibration of the device; the current sensor can be a Hall effect sensor or a mutual inductor, used to detect the current change of the device; the pressure sensor can be piezoresistive or capacitive, used to detect the pressure condition of the device; the acoustic wave sensor can be made of piezoelectric ceramic material, used to detect abnormal noise inside the device; the infrared temperature sensor can use a thermopile detector, used to measure the temperature distribution on the surface of the device.

[0040] Edge computing nodes: connected to smart devices, used to receive and process data, perform preliminary filtering and remove outliers. Edge computing nodes adopt a distributed architecture, and multiple edge computing nodes work together to share the data processing load and improve data processing efficiency. Edge computing nodes communicate through message queues to ensure data synchronization and consistency. For example, edge computing nodes regularly check their own memory usage, CPU utilization, and network connection status. Once an abnormality is found, they immediately switch to the backup node to continue processing data. Edge computing nodes use a heartbeat detection mechanism to send heartbeat signals regularly to monitor the status of each component.

[0041] Cloud server: communicates with edge computing nodes to store and analyze processed data and generate fault prediction reports. The machine learning model uses a deep neural network and continuously optimizes model parameters through the back-propagation algorithm. The cloud server adopts a microservice architecture to support elastic scaling and high concurrency processing. For example, the cloud server aggregates data from multiple edge computing nodes and processes them in batches to improve data processing efficiency. At the same time, the cloud server performs quality control on each batch of data to ensure data consistency and accuracy.

[0042] Operation and maintenance management system: connected to the cloud server, used to receive fault prediction reports and notify the operation and maintenance team to take measures. The operation and maintenance management system receives the fault prediction reports generated by the cloud server, and promptly notifies the operation and maintenance team to take corresponding measures through the message push function. The operation and maintenance management system has permission management and audit functions to ensure the safety and compliance of operations. For example, the operation and maintenance management system will determine whether it is necessary to generate a fault prediction report based on the fault probability distribution output by the model. If the fault probability exceeds the preset threshold \(P_t\), the system will automatically generate a report containing the fault location, expected time of occurrence and related recommended measures, and notify the operation and maintenance team by email, SMS or mobile phone application. The operation and maintenance team can take preventive measures quickly based on the information in the report to avoid equipment failure. The preset threshold \(P_t\) here can be set according to the importance of different devices. For example, \(P_t = 0.8\) means that the system will only generate a report when the fault probability is greater than 80%.

[0043] Furthermore, the operation and maintenance management system also includes: The fault database module is used to store historical fault records and maintenance reports. Fault records and maintenance reports are obtained from operation and maintenance teams in various locations, organized into a unified format, and uploaded to the fault database module on the cloud server. The fault database module uses a relational database to support complex queries and statistical analysis. The prediction model update module is used to regularly update the machine learning model and incorporate the latest fault cases. At fixed time intervals, the latest fault cases are extracted from the fault database module to retrain the machine learning model and optimize the model parameters. The prediction model update module uses automated scripts to simplify the model update process. The expert review module is used by online review team members to conduct secondary verification of emergency alarms issued by the model. When the model prediction results show high-risk failures, an alarm is automatically triggered to notify online review team members to log in to the system to view detailed information. The expert review module has task allocation and progress tracking functions to ensure the efficient execution of the review work. The self-diagnosis module is used to monitor the working status of each component of the system and automatically start the redundant backup mechanism when an abnormality is found. The self-diagnosis module regularly checks the status of each component of the system. Once an abnormality is found, it immediately switches to the backup component to continue normal operation. The self-diagnosis module uses real-time monitoring technology and alarm mechanism to ensure the stable operation of the system.

[0044] The implementation principle of this embodiment is: By introducing various types of health monitoring sensors and combining the powerful data processing capabilities of edge computing nodes and cloud servers, comprehensive monitoring and real-time analysis of the equipment's operating status are achieved. This method not only improves the accuracy and timeliness of fault prediction, but also significantly reduces manual intervention and reduces maintenance costs. Especially in large-scale manufacturing companies and critical infrastructure, the application of this fault prediction method can significantly improve the reliability and safety of equipment.

[0045] Correspondingly, the operation and maintenance management system can not only efficiently store and manage historical fault records and maintenance reports, but also regularly update machine learning models to ensure that the prediction logic keeps up with real-world changes. At the same time, the system supports the online review team to conduct secondary verification of high-risk fault alarms, which improves the accuracy and reliability of fault judgment. In addition, the self-diagnosis module monitors the status of each component of the system in real time and promptly activates the redundant backup mechanism to ensure the stable operation of the system and improve the overall operation and maintenance efficiency. Example

[0046] The present application is different from Embodiment 1 in that, after step S4, the following steps are further included: S5. Collect all fault records and detailed maintenance reports of similar equipment in the past ten years to build a detailed historical fault database. The historical fault database is stored on the cloud server and supports multi-user access and data sharing to facilitate collaboration between operation and maintenance teams in different regions. This includes: obtaining fault records and maintenance reports from operation and maintenance teams in various regions, organizing them into a unified format, and uploading them to the historical fault database on the cloud server. The historical fault database uses distributed storage technology to ensure data consistency and availability.

[0047] Specifically, the historical fault database stores all fault records and detailed maintenance reports of similar equipment in the past ten years, and supports multi-user access and data sharing. Fault records and maintenance reports are obtained from operation and maintenance teams in various locations, organized into a unified format, and uploaded to the historical fault database on the cloud server. The historical fault database uses distributed storage technology to ensure data consistency and availability. For example, Hadoop Distributed File System (HDFS) or other similar distributed storage solutions can be used to ensure high consistency and availability of data.

[0048] Through the above steps, more failure cases can be accumulated to help the machine learning model better understand and predict failure modes. Specifically, the historical failure database will record the specific time, location, environmental conditions and related maintenance reports of each failure. These data will be cleaned and standardized to form a unified format for subsequent analysis and modeling. Environmental condition information uses data from meteorological stations to ensure data accuracy and timeliness.

[0049] Furthermore, after step S5, the following steps are also included: S6. Design a special algorithm to extract the key parameters in each record and convert them into a standardized format that can be used for model training; including: parsing the text information in the fault record, extracting keywords and numerical parameters, and converting them into a machine-readable standard format; the algorithm uses natural language processing technology to automatically identify and extract key parameters.

[0050] The algorithm uses natural language processing technology to automatically identify and extract key parameters. It parses the text information in the fault record, extracts keywords and numerical parameters, and converts them into a machine-readable standard format. For example, named entity recognition (NER) technology can be used to automatically extract key parameters such as fault type, occurrence time, location, and environmental conditions.

[0051] Therefore, the above steps ensure the quality and consistency of historical fault data and improve the effect of model training. Specifically, the process of data standardization involves a lot of text processing and data cleaning. For example, by performing operations such as word segmentation, stemming, and removing stop words on the text information in the fault record, key information fragments are extracted and mapped to pre-defined standard fields. These standardized data can be directly used for model training without additional manual intervention.

[0052] After step S6, the following steps are included: S7, regularly update the model, incorporate the latest fault cases, and keep the prediction logic in line with real changes; including: extracting the latest fault cases from the historical fault database at fixed time intervals, retraining the machine learning model, and optimizing model parameters; using incremental learning methods in the model update process to gradually incorporate new data and reduce training time.

[0053] Specifically, at fixed time intervals (such as once a month), the latest fault cases are extracted from the historical fault database, the machine learning model is retrained, and the model parameters are optimized. Incremental learning methods are used in the model update process to gradually incorporate new data and reduce training time. For example, online learning algorithms such as stochastic gradient descent (SGD) can be used to gradually update model parameters without stopping existing services. This helps the model to evolve continuously and adapt to new failure modes and technological developments. Specifically, the model's hyperparameters are dynamically adjusted during the model update process to adapt to the new data distribution. For example, the optimal hyperparameter combination can be selected through cross-validation methods to improve the model's generalization ability and prediction accuracy. At the same time, the results of the model update will be rigorously tested and verified to ensure its stability and reliability.

[0054] The implementation principle of this embodiment is: By establishing a detailed historical fault database and regularly updating the model, not only the accuracy and robustness of fault prediction are enhanced, but also the performance of the equipment in different environments can be better coped with. Especially for those equipment that have been running for a long time and have many types of faults, this approach can significantly improve the reliability of fault prediction and reduce the risk of unexpected downtime. Example

[0055] The difference between this embodiment and the above embodiment is that the function of online review group is added. The details are as follows: That is, after step S7, the following steps are also included: S8. Establish an online review team composed of senior maintenance engineers, who are responsible for secondary verification of emergency alarms issued by the model; including: when the model prediction results show a high-risk fault, the alarm is automatically triggered to notify the online review team members to log in to the system to view detailed information; the online review team members can remotely access the system through mobile terminals to view alarm information and related data in real time.

[0056] Specifically, members of the online review team can access the system remotely through mobile terminals to view alarm information and related data in real time. When the model predicts that a certain device has a high probability of failure in the next week, the system will automatically send an email or text message to notify the review team members, reminding them to log in to the system to view detailed information. For example, review team members can use smartphones or tablets to log in to the system through a dedicated APP or web page to view alarm information, historical data, and fault prediction reports. Doing so can ensure that fault prediction results are verified in a timely manner and avoid unnecessary maintenance work caused by false alarms. Specifically, review team members will verify each prediction result one by one based on the alarm information provided by the system. They will compare historical data with the current operating status of the equipment to comprehensively judge the credibility of the prediction results. If there are any questions, review team members can also initiate discussions to discuss the best response strategies together.

[0057] S9. The system design includes a scoring system that comprehensively assesses the risk level based on multiple dimensions such as failure probability, equipment type, and impact scope. This includes: assigning corresponding weights based on the probability value output by the model, the importance of the equipment, and the impact scope of the failure, and calculating the total risk score. The scoring system uses the hierarchical analysis method to ensure the objectivity and accuracy of the scoring.

[0058] Among them, according to the probability value output by the model, the importance of the equipment and the scope of influence of the fault, the corresponding weights are assigned respectively to calculate the total risk score. The scoring system adopts the hierarchical analysis method to ensure the objectivity and accuracy of the scoring. For example, the importance of the equipment can be determined based on factors such as the frequency of use of the equipment and the loss of downtime; the scope of influence of the fault can be determined based on factors such as the downtime of the production line and economic losses that may be caused by the fault, which can help the operation and maintenance team prioritize high-risk faults and allocate resources reasonably.

[0059] Specifically, the scoring system will take into account multiple factors, such as the criticality of the equipment, the frequency of failures, the scope of impact of failures, etc., to form a multi-level evaluation framework. Each factor will have a fixed weight, and the final risk score is the weighted average of the scores of each factor. Such a scoring system can not only quantify the risk of failures, but also guide the operation and maintenance team to formulate reasonable emergency plans. Assuming that the weight of the importance of the equipment is \(w_1\), and the weight of the impact range of the failure is \(w_2\), the total risk score \(R\) is calculated as follows: \[ R = w_1 \times P + w_2 \times I\] Where \(P\) is the failure probability output by the model, and \(I\) is the impact range of the failure.

[0060] In addition, after step S9, the method further includes the following steps: S10. When the predicted score exceeds the preset threshold and is approved by the review group, a warning notification is pushed to the on-site operation and maintenance personnel; including: if the risk score is higher than the preset threshold and is approved by the review group, a warning notification is sent to the designated operation and maintenance personnel through the message push function of the operation and maintenance management system; the warning notification contains the specific fault location, expected time of occurrence and related recommended measures.

[0061] Specifically, if the risk score is higher than the preset threshold and approved by the review team, an early warning notification will be sent to the designated operation and maintenance personnel through the message push function of the operation and maintenance management system. The early warning notification contains the specific fault location, expected time of occurrence, and related recommended measures. For example, spare parts can be prepared in advance, and maintenance personnel can be arranged to perform preventive maintenance. Ensure that on-site operation and maintenance personnel respond quickly and take necessary preventive measures to prevent faults from occurring.

[0062] Specifically, the early warning notification will be sent in various forms, such as text messages, phone calls, emails, etc., to ensure that the information can be conveyed to relevant personnel in a timely manner. After receiving the notification, the operation and maintenance personnel will take prompt action according to the recommended measures in the early warning report to eliminate potential fault hazards.

[0063] The implementation principle of this embodiment is as follows: By setting up an online review team and scoring system, not only the credibility of fault prediction is improved, but also the operation and maintenance team can deal with high-risk faults more efficiently. This dual protection mechanism can minimize the losses caused by equipment failures without affecting normal production. Example

[0064] The difference between this embodiment and the above embodiment is that the types of health monitoring sensors are expanded. The details are as follows: Enhanced sensors: The health monitoring sensor further includes at least one of an acoustic wave sensor and an infrared temperature sensor. The acoustic wave sensor is used to detect the noise level inside the device, and the infrared temperature sensor is used to measure the temperature distribution on the surface of the device. The data of these two sensors are transmitted to the edge computing node for processing together with other sensor data, further enriching the evaluation indicators of the device health. For example, the acoustic wave sensor can be made of piezoelectric ceramic material to detect abnormal noise inside the device; the infrared temperature sensor can use a thermopile detector to measure the temperature distribution on the surface of the device.

[0065] Specifically, acoustic sensors can detect subtle noise changes in the early stages of equipment, which is very sensitive to wear and looseness of some mechanical parts. Infrared temperature sensors can provide accurate temperature measurements in high-temperature environments to help detect hot spots and uneven heat distribution. These new sensor data are combined with existing temperature, humidity, vibration, current, pressure and other data to form a multi-dimensional equipment health assessment system.

[0066] Self-diagnosis function: The edge computing node has a self-diagnosis function, which can automatically detect its own performance status and start the redundant backup mechanism when necessary to ensure the continuity and reliability of data processing. The edge computing node regularly checks its own memory usage, CPU utilization and network connection status. Once an abnormality is found, it immediately switches to the backup node to continue processing data. The self-diagnosis function uses a heartbeat detection mechanism to send heartbeat signals regularly to monitor the status of each component. For example, you can set it to send a heartbeat signal once a minute. If the master node does not receive the heartbeat signal within the specified time, it is considered that the master node has failed and automatically switches to the backup node to continue processing data. This ensures that data processing is uninterrupted even in extreme cases.

[0067] Specifically, the self-diagnosis function of the edge computing node will regularly perform a series of health checks, including but not limited to memory usage, CPU load, disk space, network bandwidth, etc. If any indicator exceeds the preset threshold, the node will automatically enter the fault recovery mode and try to restart or switch to the backup node. This process is fully automated and does not require human intervention, which greatly improves the robustness and stability of the system.

[0068] The implementation principle of this embodiment is as follows: By introducing acoustic wave sensors and infrared temperature sensors, the comprehensiveness and accuracy of equipment health monitoring are further improved. At the same time, the self-diagnosis function of the edge computing node ensures the continuity and reliability of data processing, making the entire fault prediction system more robust. Example

[0069] The difference between this embodiment and the above embodiment is that a multi-model fusion method is adopted. The details are as follows: Multi-model fusion: The machine learning model not only uses deep neural networks, but also combines Bayesian classifiers and support vector machines. The processed data is input into these three models in turn, and the output results of the three models are integrated through a voting mechanism as the final fault prediction conclusion. Deep neural networks are mainly used to extract spatial and temporal features of data, Bayesian classifiers are suitable for learning small sample data, and support vector machines are suitable for fast classification of large-scale data.

[0070] For example, a deep neural network can use a structure that combines convolutional layers and fully connected layers, generate intermediate layer representations through forward propagation, and then adjust weights through backpropagation to optimize model parameters until the preset convergence standard is reached; a Bayesian classifier can use a naive Bayesian algorithm, which is suitable for learning small sample data, especially for the prediction of certain specific fault types; a support vector machine can use a linear kernel function, which is suitable for the rapid classification of large-scale data, especially for the prediction of common fault types. In this way, the characteristics of each model can be fully utilized to improve the generalization ability and scope of application of fault prediction.

[0071] Specifically, the multi-model fusion method will first divide the data into multiple subsets, and then send them to the deep neural network, Bayesian classifier and support vector machine for processing. Each model will output a prediction result, and finally the result recognized by the majority of models will be selected as the final fault prediction conclusion through a voting mechanism. This method can not only effectively utilize the advantages of each model, but also reduce the possible deviation of a single model and improve the overall prediction accuracy.

[0072] In addition, the deep neural network adopts a structure that combines convolutional layers and fully connected layers. It generates intermediate layer representations through forward propagation, and then adjusts weights through back propagation to optimize model parameters until the preset convergence standard is reached. The Bayesian classifier uses the naive Bayesian algorithm, which is suitable for learning small sample data, especially for the prediction of certain specific fault types. The support vector machine uses a linear kernel function, which is suitable for the rapid classification of large-scale data, especially for the prediction of common fault types. In this way, the advantages of each model can be fully utilized to improve the overall performance of fault prediction.

[0073] Specifically, the deep neural network will capture the time series characteristics of the data through the convolution layer, integrate multiple sensor data through the fully connected layer, and generate comprehensive health assessment results. The Bayesian classifier will use prior knowledge and posterior probability to efficiently classify small sample data. The support vector machine will find the optimal decision boundary through the maximum margin principle to achieve rapid classification of large-scale data. The joint use of these models enables the fault prediction system to more accurately identify various fault modes and improve the reliability and accuracy of the prediction.

[0074] For example, for a specific fault type, a naive Bayes classifier can be used for prediction because it performs well in this scenario, while for general fault types, a support vector machine can be used for classification because it has a fast speed and good performance on large-scale data.

[0075] The implementation principle of this embodiment is: Through the multi-model fusion method, not only can the advantages of each model be fully utilized, but also the shortcomings of a single model can be effectively compensated to improve the overall performance of fault prediction. This method is particularly suitable for complex and changeable actual working conditions and can better adapt to different equipment and application scenarios.

[0076] The above are all preferred embodiments of the present application, and the protection scope of the present application is not limited thereto. Therefore, any equivalent changes made according to the structure, shape, and principle of the present application should be included in the protection scope of the present application.

Claims

1. A device failure prediction method based on intelligent device health analysis, characterized in that: The following steps are involved: S1. Embed a health monitoring sensor in a smart device, wherein the health monitoring sensor includes at least one of a temperature and humidity sensor, a vibration sensor, a current sensor, and a pressure sensor, and is used to continuously collect various indicator data reflecting the operating status of the device; S2. Receive and process the collected data through the edge computing node, perform preliminary filtering and remove outliers, including: sorting the received data by timestamp, removing duplicate data, marking the data exceeding the preset threshold as outliers and excluding them, and retaining valid data; the edge computing node uses a sliding window mechanism to check the difference between adjacent data points one by one, and if the difference exceeds the set threshold, it is marked as an outlier; S3. Transmitting the processed data to a cloud server, analyzing the health trend of the equipment and identifying potential failure modes using a machine learning model, wherein the machine learning model uses a deep neural network and continuously optimizes the model parameters through a back-propagation algorithm; including: inputting the processed data into the deep neural network, generating an intermediate layer representation through forward propagation, and then adjusting the weights through back-propagation to optimize the model parameters until a preset convergence standard is reached; the deep neural network uses a structure combining convolutional layers and fully connected layers to extract spatial and temporal features of the data; S4. Based on the analysis results, a fault prediction report is intelligently generated, and the operation and maintenance team is notified to take measures through the operation and maintenance management system; including: determining the possibility of a fault according to the probability distribution of the model output, and when the probability exceeds a preset threshold, a fault prediction report is generated, and the operation and maintenance team is notified to take corresponding measures through the message push function of the operation and maintenance management system; the operation and maintenance management system has a log recording function to record the time, object and content of each notification.

2. The device failure prediction method based on intelligent device health analysis according to claim 1 is characterized in that: The following steps are also included: S5. Collect all fault records and detailed maintenance reports of similar equipment in the past ten years to build a detailed historical fault database, which is stored on a cloud server and supports multi-user access and data sharing, facilitating collaboration between operation and maintenance teams in different regions; including: obtaining fault records and maintenance reports from operation and maintenance teams in various regions, organizing them into a unified format, and uploading them to the historical fault database on the cloud server; the historical fault database uses distributed storage technology to ensure data consistency and availability; S6. Design a special algorithm to extract key parameters from each record and convert them into a standardized format that can be used for model training; including: parsing text information in the fault record, extracting keywords and numerical parameters, and converting them into a machine-readable standard format; the algorithm uses natural language processing technology to automatically identify and extract key parameters; S7. Update the model regularly to incorporate the latest fault cases and keep the prediction logic in line with real-world changes; including: extracting the latest fault cases from the historical fault database at fixed time intervals, retraining the machine learning model, and optimizing model parameters; using incremental learning methods in the model update process to gradually incorporate new data and reduce training time.

3. The device failure prediction method based on intelligent device health analysis according to claim 1 is characterized in that: The following steps are also included: S8. Establish an online review team composed of senior maintenance engineers, who are responsible for secondary verification of emergency alarms issued by the model; including: when the model prediction results show a high-risk fault, automatically trigger an alarm and notify the online review team members to log in to the system to view detailed information; the online review team members can remotely access the system through mobile terminals to view alarm information and related data in real time; S9. The system design includes a scoring system to comprehensively assess the risk level based on multiple dimensions such as failure probability, equipment type and impact scope; including: assigning corresponding weights according to the probability value output by the model, the importance of the equipment and the impact scope of the failure, and calculating the total risk score; the scoring system adopts the hierarchical analysis method to ensure the objectivity and accuracy of the scoring; S10. When the predicted score exceeds the preset threshold and is approved by the review group, a warning notification is pushed to the on-site operation and maintenance personnel; including: if the risk score is higher than the preset threshold and is approved by the review group, a warning notification is sent to the designated operation and maintenance personnel through the message push function of the operation and maintenance management system; the warning notification contains the specific fault location, expected time of occurrence and related recommended measures.

4. The device failure prediction method based on intelligent device health analysis according to claim 1 is characterized in that: The health monitoring sensor further includes at least one of an acoustic wave sensor and an infrared temperature sensor to enhance the monitoring capability of the equipment operation status: the acoustic wave sensor is used to detect the noise level inside the equipment, and the infrared temperature sensor is used to measure the temperature distribution on the surface of the equipment, and the acquired data is transmitted to the edge computing node together with other sensor data for processing; The acoustic wave sensor is made of piezoelectric ceramic material; the infrared temperature sensor is made of a thermopile detector.

5. The device failure prediction method based on intelligent device health analysis according to claim 1 is characterized in that: The edge computing node has a self-diagnosis function, which can automatically detect its own performance status and start a redundant backup mechanism when necessary to ensure the continuity and reliability of data processing; including: the edge computing node regularly checks its own memory usage, CPU utilization and network connection status, and once an abnormality is found, it immediately switches to the backup node to continue processing data; the self-diagnosis function adopts a heartbeat detection mechanism, sends heartbeat signals regularly, and monitors the status of each component.

6. The device failure prediction method based on intelligent device health analysis according to claim 1 is characterized in that: The machine learning model not only uses a deep neural network, but also combines a Bayesian classifier and a support vector machine. The processed data is input into the deep neural network, the Bayesian classifier and the support vector machine in sequence, and the output results of the three models are integrated through a voting mechanism as the final fault prediction conclusion; the Bayesian classifier adopts a naive Bayes algorithm, which is suitable for learning small sample data; the support vector machine adopts a linear kernel function, which is suitable for rapid classification of large-scale data.

7. The device failure prediction method based on intelligent device health analysis according to claim 2 is characterized in that: The historical fault database further includes relevant information on the time, location, and environmental conditions of the fault, so as to more fully understand the cause of the fault; including: when collating fault records, in addition to extracting key parameters, it is also necessary to record the specific time and location of the fault, as well as the environmental conditions at the time; the environmental condition information uses data from meteorological stations to ensure the accuracy and timeliness of the data.

8. The device failure prediction method based on intelligent device health analysis according to claim 3 is characterized in that: The scoring system also includes factors such as equipment importance and maintenance cost. When calculating the risk score, the importance and maintenance cost of the equipment are taken into consideration, and higher weights are given to failures of important equipment or high maintenance costs. The importance of the equipment is measured by the frequency of use and downtime loss of the equipment, and the maintenance cost is estimated by the average maintenance cost and difficulty of maintenance.

9. A device failure prediction system based on intelligent device health analysis, used to execute the device failure prediction method based on intelligent device health analysis as claimed in claim 8, characterized in that: include: A smart device with a built-in health monitoring sensor for continuously collecting data on the device's operating status, wherein the health monitoring sensor includes at least one of a temperature and humidity sensor, a vibration sensor, a current sensor, a pressure sensor, an acoustic wave sensor, and an infrared temperature sensor; An edge computing node is connected to the smart device and is used to receive and process data, perform preliminary filtering and remove outliers. The edge computing node adopts a distributed architecture, and multiple edge computing nodes work together to share the data processing load and improve data processing efficiency. The edge computing nodes communicate with each other through message queues to ensure data synchronization and consistency; The cloud server communicates with the edge computing node to store and analyze the processed data and generate a fault prediction report. The machine learning model uses a deep neural network and continuously optimizes the model parameters through a back propagation algorithm. The cloud server adopts a microservice architecture to support elastic scaling and high concurrent processing. The operation and maintenance management system is connected to the cloud server and is used to receive fault prediction reports and notify the operation and maintenance team to take measures. The operation and maintenance management system receives the fault prediction reports generated by the cloud server and promptly notifies the operation and maintenance team to take corresponding measures through the message push function; the operation and maintenance management system has authority management and auditing functions to ensure the security and compliance of operations.

10. The equipment failure prediction system based on intelligent equipment health analysis according to claim 9 is characterized in that: The operation and maintenance management system also includes: The fault database module is used to store historical fault records and maintenance reports. Fault records and maintenance reports are obtained from operation and maintenance teams in various places, sorted into a unified format, and uploaded to the fault database module on the cloud server. The fault database module adopts a relational database to support complex queries and statistical analysis. The prediction model update module is used to regularly update the machine learning model, incorporate the latest fault cases, extract the latest fault cases from the fault database module at regular intervals, retrain the machine learning model, and optimize the model parameters; the prediction model update module uses an automated script to simplify the model update process; The expert review module is used for the online review team members to conduct secondary verification of the emergency alarm issued by the model. When the model prediction results show a high-risk failure, an alarm is automatically triggered to notify the online review team members to log in to the system to view detailed information; the expert review module has task allocation and progress tracking functions to ensure the efficient execution of the review work; The self-diagnosis module is used to monitor the working status of each component of the system and automatically start the redundant backup mechanism when an abnormality is found. The self-diagnosis module regularly checks the status of each component of the system. Once an abnormality is found, it immediately switches to the backup component to continue normal operation; the self-diagnosis module adopts real-time monitoring technology and alarm mechanism to ensure the stable operation of the system.

Images