Nitrogen generation device fault monitoring system based on AI edge calculation

Abstract

The invention discloses a nitrogen generation device fault monitoring system based on AI edge calculation, and relates to the technical field of artificial intelligence. The system is composed of a device layer, a data acquisition layer, an edge calculation layer and a control execution layer. The equipment layer collects operation parameters and physical quantities of the nitrogen generation device in real time through a PLC and a sensor; the data acquisition layer realizes accurate acquisition and uniform format processing of multi-source data by using a multi-protocol adaptation module and an extended sensor interface. The edge calculation layer adopts a lightweight AI model to perform efficient data aggregation and preprocessing, and performs model reasoning in combination with LSTM and XGBoost algorithms to predict potential faults. The control execution layer is responsible for fault early warning rule configuration, fault prediction processing, adaptive adjustment and model training optimization, and supports a reinforcement learning engine to automatically optimize a fault prediction strategy. The operation reliability, the fault prediction precision and the operation and maintenance efficiency of the nitrogen making device are improved, and the manual inspection cost is reduced.

Description

TECHNICAL FIELD

[0001] The application belongs to the technical field of artificial intelligence, and particularly relates to a nitrogen making device fault monitoring system based on AI edge computing. BACKGROUND

[0002] In the industrial production process, the nitrogen making device is an indispensable part of many industries, such as chemical industry, electronics, food and beverage, and medicine, etc. These devices provide high-purity nitrogen supply by separating nitrogen from air, and their running efficiency and reliability are directly related to the stability of the production line and the product quality. However, due to the long-term work of the nitrogen making device in harsh environments such as high pressure and high temperature, the frequent faults caused by equipment aging, wear and tear of parts, etc. seriously affect the continuity and safety of production.

[0003] Traditionally, the fault monitoring of the nitrogen making device mainly relies on manual inspection and regular maintenance. This method not only consumes time and effort, but also is difficult to achieve early warning of potential faults, and often only measures can be taken after the fault occurs, which greatly increases the maintenance cost and downtime. In addition, with the development of Internet of Things (IoT), sensor technology and artificial intelligence (AI), more and more intelligent monitoring means are applied in the industrial field, but how to effectively integrate these advanced technologies to improve the running reliability and maintenance efficiency of the nitrogen making device is still a problem to be solved.

[0004] In recent years, edge computing as a new computing mode can perform real-time data processing and analysis near the data source, providing a new idea for solving the above problems. Combined with lightweight AI models such as long short-term memory network (LSTM) and extreme gradient boosting (XGBoost), potential faults of the nitrogen making device can be quickly and accurately predicted on the edge side, thereby significantly improving the timeliness and accuracy of fault detection. However, the existing edge computing and AI combined fault monitoring system still faces challenges such as difficulty in multi-protocol data acquisition, low data quality, limited processing capacity of nonlinear relationship, etc.

[0005] Therefore, the present application aims to propose a nitrogen making device fault monitoring system based on AI edge computing, which realizes comprehensive and accurate data acquisition of the nitrogen making device, efficient data aggregation and preprocessing, effective capture of complex nonlinear relationship, and thus improves the fault prediction accuracy and operation and maintenance efficiency. At the same time, the system also supports flexible configuration of early warning rules and instant notification function, ensuring quick response to various fault conditions and adapting to changing working environments. SUMMARY

[0006] In order to overcome the shortcomings and deficiencies of the prior art, the present application adopts the following technical solutions: A fault monitoring system for a nitrogen generator based on AI edge computing includes an equipment layer, a data acquisition layer, an edge computing layer, and a control execution layer. The equipment layer includes a nitrogen generator PLC, which controls the operation of the nitrogen generator and collects key parameters. It also deploys sensors / terminal devices to monitor the pressure, flow rate, purity, and temperature of the nitrogen generator in real time. The data acquisition layer includes a multi-protocol adapter module and an extended sensor interface. The multi-protocol adapter module is connected to the nitrogen generator PLC to collect and analyze the data in the filtered PLC. The extended sensor interface is adapted to different types of sensors / terminal devices to collect and analyze real-time data. The edge computing layer transmits data through an industrial-grade gateway, uses lightweight AI models to aggregate and preprocess edge data, and predicts the operating status of the nitrogen generation unit through model inference. The control execution layer includes a fault early warning rule configuration module, a fault prediction and processing module, and a fault early warning notification module. It also includes a real-time monitoring module, an adaptive adjustment module, a model training module, and a reinforcement learning engine module. The fault early warning rule configuration module sets fault judgment rules and early warning levels. The fault prediction and processing module compares the model inference results with preset rules and generates control commands and alarm notifications. The fault early warning notification module sends early warning information to maintenance personnel. The real-time monitoring module displays operating parameters and status. The adaptive adjustment module fine-tunes operating parameters. The model training module optimizes the model. The reinforcement learning engine module automatically optimizes fault prediction strategies and control command generation rules.

[0007] Preferably, the method further includes: acquiring historical fault data, which includes the model of the nitrogen generator and the fault type; dividing the historical fault data based on the equipment model to obtain multiple historical data sets; grouping the historical data sets based on the fault type to obtain multiple historical data combinations; labeling the historical data combinations according to the fault type; and training the fault detection model based on the labeled historical data combinations.

[0008] Preferably, the model inference of the edge computing layer uses long short-term memory networks to capture the time-series features of the data and combines extreme gradient boosting algorithms to process nonlinear relationships, thereby inferring the operating status of the nitrogen generation unit.

[0009] Preferably, the fault prediction and processing module of the control execution layer compares the model inference results of the edge computing layer with preset rules. If an anomaly is detected, a control command is generated and an alarm notification is issued. The control command includes adjusting operating parameters and starting backup equipment.

[0010] Preferably, the fault condition determination includes: preprocessing the first image data of the key parts of the nitrogen generator to obtain the second image data, the preprocessing including denoising, normalization and image enhancement; extracting fault features from the second image data based on feature extraction algorithms, the feature extraction algorithms including edge detection algorithms, color analysis technology and texture analysis algorithms, fusing the extracted edge features, color features and texture features to obtain fault features; and inputting the fault features into the fault detection model to obtain the first fault condition.

[0011] Preferably, the fault determination further includes: comparing the detection data with preset data to determine abnormal detection data; determining the cause of the abnormality based on the abnormal detection data; determining the sensor corresponding to the abnormal detection data; determining the abnormal location based on the sensor's location; and determining a second fault condition based on the cause of the abnormality, the abnormal location, and the abnormal detection data.

[0012] Preferably, the generation of early warning information in the control execution layer includes: analyzing historical fault data to obtain historical fault characteristics, including high-incidence periods of faults; counting the number of historical fault data in each historical data combination; determining the fault level of each fault type based on the fault type and quantity; if the quantity is greater than a first preset quantity and / or the fault severity is a serious fault, then the fault level is the first fault level; if the quantity is less than or equal to a first preset quantity and greater than a second preset quantity, and the fault severity is a minor or moderate fault, then the fault level is the second fault level; if the quantity is less than or equal to a second preset quantity, and the fault severity is a minor or moderate fault, then the fault level is the third fault level; and generating early warning information based on historical fault characteristics, fault level, and fault condition.

[0013] Preferably, the early warning information includes the most frequent fault type and its corresponding fault level within the time period, and a repair strategy is determined based on the fault situation. If it is an automatic repair, the repair strategy is sent, the repair result is obtained, and early warning information is generated based on historical fault characteristics, repair results, fault level, and fault situation.

[0014] Preferably, the method further includes acquiring the target fault point to be monitored in the nitrogen generator; obtaining the corresponding operation reference document and the required monitoring operation data based on the target fault point; obtaining the health compliance rate of the target fault point based on the operation reference document, and obtaining the fault risk score of the target fault point based on the operation data; determining the fault monitoring result of the target fault point based on the health compliance rate and the fault risk score; the operation data includes temperature, pressure, flow rate, purity, vibration, energy consumption, valve status, control system response time, and alarm frequency; obtaining the fault risk score of the target fault point based on the operation data includes: when the target fault point is the compression system, obtaining the deviation index of the compression system based on temperature deviation, pressure deviation, and vibration deviation; and obtaining the first fault risk score of the compression system according to the formula. When the target failure point is the adsorption system, the adsorption quality indicators of the adsorption system are obtained based on purity deviation, flow rate deviation, and energy consumption deviation; the second failure risk score of the adsorption system is obtained according to the formula. When the target failure point is the output system, the output quality indicators of the output system are obtained based on purity, flow rate, and pressure stability; the third failure risk score of the output system is obtained according to the formula. Before obtaining the failure risk score for the compression system, adsorption system, or output system, the following steps are also included: obtaining preset weights according to the formula. When the target fault point is the control system, the control quality indicators of the control system are obtained based on valve state deviation, control system response time deviation, and alarm frequency; the fourth fault risk score of the control system is obtained according to the formula. When the target failure point is the cooling system, the cooling quality indicators of the cooling system are obtained based on temperature stability, energy consumption deviation, and vibration deviation; the fifth failure risk score of the cooling system is obtained according to the formula. When the target failure point is the entire system, the comprehensive health index of the entire system is obtained based on purity, flow stability, and energy consumption; the sixth failure risk score of the entire system is obtained according to the formula. The health compliance rate of the target fault point is obtained based on the running reference document, including: extracting the corresponding standard parameter document from the running database according to the document type of the running reference document; comparing the real-time running data and the standard parameter document, calculating the proportion of parameters within the preset range, and determining this proportion as the health compliance rate.

[0015] In summary, due to the adoption of the above technical solution, the beneficial effects of the present invention are: 1. This invention employs multi-protocol adaptation technology, supporting data acquisition using multiple protocols. This breaks away from the traditional system's reliance on a single protocol, improving the flexibility and compatibility of data acquisition. Simultaneously, the expanded sensor interface design allows the system to easily connect to different types of sensors, further enhancing the comprehensiveness and accuracy of data acquisition. 2. This invention combines LSTM and XGBoost algorithms, fully leveraging the advantages of LSTM in processing time-series data and the ability of XGBoost to handle nonlinear relationships, thus achieving accurate prediction of the operating status of nitrogen generators. Furthermore, the system further improves the accuracy and comprehensiveness of fault prediction through multi-dimensional fault analysis driven by image features and sensor data. 3. This invention introduces a reinforcement learning engine, enabling the system to automatically optimize fault prediction strategies and control command generation rules, adapting to different operating environments without manual intervention. Furthermore, the system supports a model training module that continuously trains and optimizes the LSTM / XGBoost hybrid model using historical and real-time feedback data, constantly improving prediction accuracy and achieving self-learning and self-optimization. Attached Figure Description

[0016] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0017] Figure 1 A block diagram of a nitrogen generator fault monitoring system based on AI edge computing according to the present invention is shown. Figure 2 A flowchart of the operation of the system of the present invention is shown. Detailed Implementation

[0018] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0019] Furthermore, the described features, structures, or characteristics can be combined in any suitable manner in one or more exemplary embodiments. Numerous specific details are provided in the following description to give a full understanding of exemplary embodiments of this disclosure. However, those skilled in the art will recognize that the technical solutions of this disclosure can be practiced with one or more of the specific details omitted, or other methods, components, steps, etc., can be employed. In other instances, well-known structures, methods, implementations, or operations are not shown or described in detail to avoid obscuring various aspects of this disclosure. Example

[0020] See Figure 1 As shown in this embodiment, a nitrogen generator fault monitoring system based on AI edge computing includes an equipment layer, a data acquisition layer, an edge computing layer, and a control execution layer. Equipment layer: Includes the nitrogen generator PLC, used to control the operation of the nitrogen generator and collect key operating parameters; deploys sensors / terminal devices to monitor physical quantities such as pressure, flow rate, purity, and temperature of the nitrogen generator in real time, providing basic data for fault monitoring.

[0021] Data Acquisition Layer: The multi-protocol adapter module connects to the nitrogen generator PLC, acquiring data from the PLC through various protocols, performing data parsing and filtering to remove noise and invalid data, ensuring the accuracy and stability of the acquired data; it also expands the sensor interface to adapt to different types of sensors / terminal devices, acquiring real-time data such as pressure and flow, and performing data parsing and filtering to unify the data format for easy subsequent processing.

[0022] Edge computing layer: Data transmission is achieved through industrial-grade gateways, and edge data is aggregated using lightweight AI models. Data collected by multi-protocol adaptation modules and extended sensor interfaces is integrated to form a unified dataset, providing a complete data foundation for subsequent processing. Edge data preprocessing involves cleaning and normalizing the aggregated data to improve its quality and make it meet the model input requirements. Model inference uses long short-term memory networks to capture the time-series features of the data and combines extreme gradient boosting (XGBoost) algorithms to handle nonlinear relationships, inferring the operating status of the nitrogen generation unit and predicting potential faults.

[0023] Control Execution Layer: The fault early warning rule configuration module sets fault judgment rules and early warning levels, and configures different parameter thresholds (such as pressure anomaly threshold, flow deviation threshold, etc.) based on historical fault data and industry standards; fault prediction processing compares the model inference results of the edge computing layer with preset rules. If an anomaly is detected, control commands are generated (such as adjusting operating parameters or starting standby equipment), and alarm notifications are sent simultaneously; the fault early warning notification module sends early warning information to maintenance personnel in a timely manner through a visual interface, SMS, email, etc., informing them of the fault type, location, and severity; the real-time monitoring module displays the operating parameters and status of the nitrogen generator in real time, facilitating manual inspection; the adaptive adjustment module fine-tunes the unit's operating parameters and optimizes the operating status based on monitoring data and fault feedback; the model training module continuously trains and optimizes the LSTM / XGBoost hybrid model using historical data and real-time feedback data to improve prediction accuracy; the reinforcement learning engine module automatically optimizes fault prediction strategies and control command generation rules through reinforcement learning algorithms to adapt to complex and ever-changing operating environments.

[0024] The beneficial effects of this embodiment are as follows: the fault monitoring system achieves comprehensive and accurate data acquisition through multi-protocol adaptation and sensor expansion; it improves data quality by using a lightweight AI model for efficient data aggregation and preprocessing at the edge; it effectively captures time series features and complex nonlinear relationships by combining LSTM and XGBoost algorithms, thereby improving fault prediction accuracy; it configures flexible early warning rules and supports multi-channel real-time notifications to ensure rapid response; it has adaptive adjustment and continuous model optimization capabilities, and the reinforcement learning engine automatically optimizes strategies to adapt to changing environments, significantly improving the operational reliability, prediction accuracy, and maintenance efficiency of the nitrogen generation unit. Example

[0025] See Figure 2 As shown, the workflow of a nitrogen generator fault monitoring system based on AI edge computing in this embodiment is as follows: Step 1: Acquisition of basic data.

[0026] Acquire nitrogen generator test data, including physical parameters such as pressure, flow rate, purity, and equipment temperature; collect first-image data of key parts of the nitrogen generator (such as images of components like the compressor chamber and filter membrane assembly).

[0027] Step 2: Fault detection model construction.

[0028] S21. Historical data preparation.

[0029] Obtain historical fault data, including nitrogen generator model, fault type, fault occurrence time, maintenance records, etc. S22, Data Classification and Labeling.

[0030] By equipment model: Historical data are grouped according to the model of the nitrogen generator, forming multiple historical data sets (data of the same model are grouped together). Grouping by fault type: For each historical data set, further grouping is done according to fault type (such as filter blockage, abnormal pressure, substandard purity, etc.) to form historical data combinations (data of the same fault type are grouped together).

[0031] Labeling process: Label each historical data combination with the corresponding fault type (such as filter blockage, compressor overheating, etc.). S23, Model Training.

[0032] Fault detection models (such as deep learning models, machine learning models, etc.) are trained based on the combination of labeled historical data. Step 3: Determine the real-time fault status.

[0033] S31. Data Input and Processing.

[0034] Input real-time detection data (pressure / flow rate / purity / temperature) and first image data. S32. Multi-dimensional fault analysis (divided into two fault scenarios) Case A: Image feature-driven first failure case (applicable to visualization component failure).

[0035] Image preprocessing: The first image data is denoised, normalized, and enhanced to obtain the second image data; Feature extraction: Edge detection algorithms extract the contour edge features of the component; Color analysis technology extracts color features such as abnormal color patches (e.g., reddish areas in high-temperature regions) and stains; Texture analysis algorithms extract texture features such as surface wear and cracks; By fusing edge, color, and texture features, a fault feature vector is formed. Model inference: Input the fault features into the fault detection model and output the first fault condition (including fault type and preliminary level). Case B: Sensor data-driven second fault case (applicable to parameter anomaly type faults).

[0036] Abnormal data identification: Compare real-time detection data with preset standard data (such as pressure rating and purity threshold) to identify abnormal detection data (such as pressure exceeding limits or sudden drop in purity). Anomaly localization: Identify the sensor corresponding to the abnormal data (such as pressure sensor, purity sensor); locate the location of the anomaly based on the sensor's installation location (such as compressor No. 1, filter unit No. 2). Root cause analysis: Analyze the causes of the anomalies (such as sensor failure or component aging) by combining historical fault knowledge base. Fault determination: Based on the cause, location, and data of the anomaly, a second fault condition (including fault type and preliminary level) is generated. Step 4: Fault Level Assessment.

[0037] S41. Historical data statistics. Analyze historical fault data and extract historical fault characteristics such as high-incidence periods of faults (e.g., 2-4 AM on the 15th of each month); Count the number of failures in each historical data combination (e.g., filter clogging failures occurred 50 times in the past six months). S42, Level Determination Rules. Fault severity classification: Preset minor faults (such as sensor signal fluctuations), moderate faults (such as slight flow abnormalities), and severe faults (such as pressure over-limit alarms); Level determination logic: First fault level (high risk): If the number of faults exceeds the first preset number (e.g., 30 times / month), or the fault severity is a serious fault; Second fault level (medium risk): If the first preset quantity ≥ quantity > the second preset quantity (e.g., 10 times / month), and the fault severity is slight / moderate; Third fault level (low risk): If the quantity is less than or equal to the second preset quantity, and the severity of the fault is slight / moderate. S43. Obtain the target fault point to be monitored in the nitrogen generator; based on the target fault point, obtain the corresponding operation reference document and the operation data to be monitored. The health compliance rate of the target failure point is obtained based on the operation reference document, and the failure risk score of the target failure point is obtained based on the operation data. Based on health compliance rate and failure risk score, determine the failure monitoring results of the target failure point; operational data include temperature, pressure, flow rate, purity, vibration, energy consumption, valve status, control system response time, and alarm frequency.

[0038] The failure risk score of the target failure point is obtained based on the operating data, including: when the target failure point is the compression system, the deviation index of the compression system is obtained based on the temperature deviation, pressure deviation, and vibration deviation. According to the formula Obtain the first failure risk score of the compression system. ;in, Temperature deviation; For pressure deviation; For vibration deviation, Preset weights; When the target failure point is the adsorption system, the adsorption quality indicators of the adsorption system are obtained based on purity deviation, flow rate deviation, and energy consumption deviation; according to the formula... Obtain the second failure risk score of the adsorption system. ,in, For adsorption quality indicators; This is due to purity deviation; For flow deviation; When the target fault point is the output system, the output quality indicators of the output system are obtained based on purity, flow rate, and pressure stability; according to the formula... Obtain the third fault risk score of the output system. ,in, To output quality indicators; Purity; For traffic; Before obtaining a failure risk score for the compression system, adsorption system, or output system, the method also includes: according to the formula Obtain preset weights ,in, The amount of operational data used to calculate the fault risk score; For each running data value, .

[0039] When the target fault point is the control system, the control quality indicators of the control system are obtained based on the valve state deviation, control system response time deviation, and alarm frequency; according to the formula: Obtain the fourth failure risk score of the control system ,in, To control quality indicators; Valve status deviation; For response time deviation; When the target fault point is the cooling system, the cooling quality indicators of the cooling system are obtained based on temperature stability, energy consumption deviation, and vibration deviation; according to the formula: Obtain the fifth failure risk score for the cooling system. ,in, For cooling quality indicators; Energy consumption deviation; For vibration deviation; When the target failure point is the entire system, the comprehensive health index of the entire system is obtained based on purity, flow stability, and energy consumption; according to the formula... Obtain the sixth failure risk score for the overall system. ,in, For comprehensive health indicators.

[0040] The health compliance rate of the target fault point is obtained based on the operating reference document, including: extracting the corresponding standard parameter document from the operating database according to the document type of the operating reference document; comparing the real-time operating data and the standard parameter document, calculating the proportion of parameters within the preset range, and determining this proportion as the health compliance rate.

[0041] Based on the health compliance rate and fault risk score, determine the fault monitoring results of the target fault point, including: determining the warning level in the preset rating table based on the health compliance rate and fault risk score; and filling in the health compliance rate, fault risk score and warning level in the corresponding positions of the preset fault report template.

[0042] Step 5: Early Warning Information Generation and Response.

[0043] S51, Warning Triggering Conditions.

[0044] During periods of high failure incidence in historical statistics, early warning information is automatically generated. The warning content includes: the most frequent failure type (such as filter blockage) and its corresponding failure level (such as Level 1) during that period.

[0045] S52, Repair Strategy Linkage.

[0046] Based on the fault condition, a preset repair strategy is matched: if the fault is automatically repairable (such as a minor sensor calibration abnormality), an automatic repair command is sent (such as restarting the sensor module); the repair result (success / failure) is recorded and added to the historical fault feature library. S53, Multi-channel reminders.

[0047] Warning information is pushed to maintenance personnel through visual interfaces, SMS, email, etc., including: fault type, fault level, abnormal location, suggested handling measures (such as manual inspection of filter element, replacement of sensor), health compliance rate, fault risk score and warning level.

[0048] The beneficial effects of this embodiment are as follows: the fault monitoring system constructs an accurate fault detection model through multi-dimensional data collection and preprocessing, analyzes image and sensor data in real time, and effectively identifies various faults; it scientifically assesses fault risks by combining historical data statistics and level judgment rules; it automatically generates early warning information and links repair strategies to improve response efficiency; it notifies operation and maintenance personnel through multiple channels to ensure timely handling, significantly improving the operational reliability, fault prediction accuracy, and operation and maintenance efficiency of the nitrogen generator, and reducing the cost of manual inspection.

[0049] The above description is only a preferred embodiment of the present invention, but the scope of protection of the present invention is not limited thereto. Any equivalent substitutions or modifications made by those skilled in the art within the scope of the technology disclosed in the present invention, based on the technical solution and inventive concept of the present invention, should be covered within the scope of protection of the present invention.

[0050] The preferred embodiments of the present invention disclosed above are merely illustrative of the invention. These preferred embodiments do not exhaustively describe all details, nor do they limit the invention to specific implementations. Clearly, many modifications and variations can be made based on the content of this specification. This specification selects and specifically describes these embodiments to better explain the principles and practical applications of the invention, thereby enabling those skilled in the art to better understand and utilize the invention. The invention is limited only by the claims and their full scope and equivalents.

Claims

1. An AI edge computing-based nitrogen production device fault monitoring system, characterized in that, The system comprises a device layer, a data acquisition layer, an edge computing layer and a control execution layer: The device layer comprises a nitrogen production device PLC for controlling the operation of the nitrogen production device and collecting key parameters, and also deploys sensor / terminal devices to monitor the pressure, flow, purity and temperature of the nitrogen production device in real time. The data acquisition layer comprises a multi-protocol adaptation module and an extended sensor interface, the multi-protocol adaptation module is connected with the nitrogen production device PLC, collects and analyzes the data in the filtered PLC, and the extended sensor interface adapts different types of sensors / terminal devices to collect and analyze real-time data. The edge computing layer transmits data through an industrial gateway, uses a lightweight AI model for edge data aggregation and preprocessing, and predicts the operating state of the nitrogen production device through model inference. The control execution layer comprises a fault early warning rule configuration module, a fault prediction processing module and a fault early warning notification module, and also comprises a real-time monitoring module, an adaptive adjustment module, a model training module and a reinforcement learning engine module. The fault early warning rule configuration module sets the fault judgment rules and warning levels, the fault prediction processing module compares the model inference results with the preset rules and generates control instructions and warning notifications, the fault early warning notification module sends warning information to the operation and maintenance personnel, the real-time monitoring module displays the operating parameters and state, the adaptive adjustment module fine-tunes the operating parameters, the model training module optimizes the model, and the reinforcement learning engine module automatically optimizes the fault prediction strategy and control instruction generation rules.

2. The nitrogen device fault monitoring system based on AI edge computing according to claim 1, characterized in that, Further comprising: Obtaining historical fault data, the historical fault data comprising the model of the nitrogen production device and the fault type; Dividing the historical fault data based on the device model to obtain a plurality of historical data sets; Grouping the historical data sets based on the fault type to obtain a plurality of historical data combinations; and labeling the historical data combinations according to the fault type; Training the fault detection model based on the labeled historical data combinations.

3. The AI edge computing-based nitrogen device fault monitoring system according to claim 1, characterized in that, The model inference of the edge computing layer uses a long short-term memory network to capture the time sequence features of the data, and combines an extreme gradient boosting algorithm to handle nonlinear relationships, and infers the operating state of the nitrogen production device.

4. The AI edge computing-based nitrogen device fault monitoring system according to claim 1, characterized in that, The fault prediction processing module of the control execution layer compares the model inference results of the edge computing layer with the preset rules, and if an anomaly is detected, generates a control instruction and sends an alarm notification, the control instruction comprising adjusting the operating parameters and starting the standby device.

5. The AI edge computing-based nitrogen device fault monitoring system according to claim 1, characterized in that, The fault condition determination comprises: preprocessing first image data of key parts of the nitrogen production device to obtain second image data, the preprocessing comprising denoising, normalization and image enhancement; extracting fault features from the second image data based on a feature extraction algorithm, the feature extraction algorithm comprising an edge detection algorithm, a color analysis technique and a texture analysis algorithm, and fusing the extracted edge features, color features and texture features to obtain fault features; and inputting the fault features into the fault detection model to obtain a first fault condition.

6. The AI edge computing-based nitrogen device fault monitoring system according to claim 1, characterized in that, The fault condition determination further comprises: comparing the detection data with preset data to determine abnormal detection data; determining an abnormal cause based on the abnormal detection data; determining a sensor corresponding to the abnormal detection data; determining an abnormal position based on a position of the sensor; and determining a second fault condition based on the abnormal cause, the abnormal position, and the abnormal detection data.

7. The AI edge computing-based nitrogen device fault monitoring system according to claim 1, characterized in that, The pre-warning information generation of the control execution layer comprises: analyzing historical fault data to obtain historical fault features, the historical fault features including a high-fault period; counting the number of historical fault data in each historical data combination; determining a fault level of each fault type based on the fault type and the number, wherein if the number is greater than a first preset number and / or the fault severity is a serious fault, the fault level is a first fault level; if the number is less than or equal to the first preset number and greater than a second preset number, and the fault severity is a slight fault or a moderate fault, the fault level is a second fault level; if the number is less than or equal to the second preset number, and the fault severity is a slight fault or a moderate fault, the fault level is a third fault level; and generating pre-warning information based on the historical fault features, the fault level, and the fault condition.

8. The AI edge computing-based nitrogen device fault monitoring system according to claim 7, characterized in that, The pre-warning information includes the fault type with the most occurrences in the period and the corresponding fault level, and a repair strategy is determined based on the fault condition, wherein if the repair is automatic, the repair strategy is sent, a repair result is obtained, and pre-warning information is generated based on the historical fault features, the repair result, the fault level, and the fault condition.

9. The AI edge computing-based nitrogen device fault monitoring system according to claim 1, characterized in that, Further comprising obtaining a target fault point to be monitored of the nitrogen production device; obtaining a running reference document corresponding to the target fault point and running data needed to be monitored according to the target fault point; obtaining a health compliance rate of the target fault point according to the running reference document, and obtaining a fault risk score of the target fault point according to the running data; Determining a fault monitoring result of the target fault point according to the health compliance rate and the fault risk score; the running data includes temperature, pressure, flow, purity, vibration, energy consumption, valve state, control system response time, and alarm frequency; According to the operation data, a fault risk score of the target fault point is obtained, including: when the target fault point is a compression system, deviation indexes of the compression system are obtained according to temperature deviation, pressure deviation and vibration deviation; a first fault risk score of the compression system is obtained according to the formula ; when the target fault point is an adsorption system, adsorption quality indexes of the adsorption system are obtained according to purity deviation, flow deviation and energy consumption deviation; a second fault risk score of the adsorption system is obtained according to the formula ; when the target fault point is an output system, output quality indexes of the output system are obtained according to purity, flow and pressure stability; a third fault risk score of the output system is obtained according to the formula ; before the fault risk scores of the compression system, the adsorption system or the output system are obtained, a preset weight value is obtained according to the formula ; when the target fault point is a control system, control quality indexes of the control system are obtained according to valve state deviation, control system response time deviation and alarm frequency; a fourth fault risk score of the control system is obtained according to the formula ; when the target fault point is a cooling system, cooling quality indexes of the cooling system are obtained according to temperature stability, energy consumption deviation and vibration deviation; a fifth fault risk score of the cooling system is obtained according to the formula ; when the target fault point is a whole system, comprehensive health indexes of the whole system are obtained according to purity, flow stability and energy consumption; a sixth fault risk score of the whole system is obtained according to the formula ; a health compliance rate of the target fault point is obtained according to the operation reference document, including: according to the document type of the operation reference document, a corresponding standard parameter document is extracted from an operation database; real-time operation data and the standard parameter document are compared, a proportion of parameters within a preset range is calculated, and the proportion is determined as the health compliance rate.

Images