A comprehensive safety inspection method and system for a smart power plant

Abstract

The application discloses a kind of comprehensive safety inspection methods and systems of wisdom power plant, specifically related to wisdom power plant inspection technical field, system uses five-layer architecture and digital twin hub integrated design, perception layer constructs "air-ground-indoor and outdoor-fixed mobile" three-dimensional perception network, digital twin collaborative layer realizes terminal dynamic scheduling, data fusion layer is fused multi-source data by improving attention mechanism's CNN-LSTM model, decision-making disposal layer realizes hierarchical early warning, automatic disposal and block chain storage. Method includes model initialization, multi-terminal collaborative inspection, multi-source data fusion analysis, hierarchical closed-loop disposal and model optimization five steps.The application solves the pain points of traditional inspection blind area, low efficiency, high degree of artificial dependence, improves the overall and intelligent level of inspection, realizes early warning and full-process traceability of fault, can be seamlessly connected with existing system of power plant, can be implemented, is easy to scale, reduces operation and maintenance cost.

Description

Technical Field

[0001] This invention relates to the field of smart power plant inspection technology, and more specifically, to a comprehensive safety inspection method and system for smart power plants. Background Technology

[0002] Smart power plants are the core direction of the digital and intelligent development of the power industry. As a key link in the operation and maintenance of power plant equipment and risk prevention and control, safety inspection is directly related to the safe and stable operation of power plants. At present, power plant inspection has gradually upgraded from manual inspection to intelligent inspection, and has begun to use intelligent equipment such as drones, inspection robots, and various sensors to replace manual labor. This has enabled basic functions such as equipment temperature detection, simple defect identification, and environmental parameter collection, which has improved inspection efficiency to a certain extent.

[0003] However, existing smart power plant inspection technologies still suffer from many technical pain points, such as incomplete sensing network coverage, weak multi-terminal collaborative scheduling capabilities, insufficient depth of multi-source data fusion, passive inspection processes without closed-loop evidence storage mechanisms, poor compatibility with existing power plant systems, and high transformation costs. These make it difficult to meet the high-precision and high-efficiency safety inspection needs of complex smart power plant scenarios.

[0004] Therefore, in view of the above situation, the present invention provides a comprehensive safety inspection method and system for smart power plants. Summary of the Invention

[0005] In order to overcome the above-mentioned defects of the prior art, the present invention provides a comprehensive safety inspection method and system for smart power plants to solve the problems mentioned in the background art.

[0006] To achieve the above objectives, the present invention provides the following technical solution: On the one hand, this invention provides a comprehensive safety inspection method for smart power plants, including the following steps: S1. Model initialization and task deployment: Construct a 1:1 digital twin model of the power plant, import equipment ledgers, inspection points and hazardous area boundary data, set routine inspection task parameters and fault judgment thresholds, and automatically decompose tasks to each sensing terminal. S2. Multi-terminal collaborative inspection: Each sensing terminal performs inspection according to the initial path. The digital twin model maps the terminal location, operating status and collected data in real time. The dynamic collaborative scheduling engine monitors the integrity of inspection coverage, terminal operating status and fault location information in real time, and automatically completes blind spot filling, low-battery terminal charging path planning and multi-terminal linkage verification of fault locations. S3. Multi-source data fusion analysis: Time and space alignment of sensing data, equipment ledger data and environmental time series data. By improving the attention mechanism of the CNN-LSTM fusion model, the correlation between fault features and multi-dimensional data is mined to identify equipment defects, unsafe personnel behaviors and environmental risks, and generate fault analysis reports and prediction results. S4. Tiered handling and closed-loop review: The corresponding handling strategy is executed according to the fault level. Minor faults automatically trigger preset handling actions. Emergency and major faults are linked to audible and visual alarms, personnel positioning and equipment start and stop control. After the handling is completed, the dispatch terminal reviews and updates the system data after confirming that the hidden danger has been eliminated, forming a closed loop of the whole process. S5. Data review and model optimization: Regularly analyze inspection and handling data, iteratively optimize AI recognition model parameters, dynamic scheduling strategies and handling strategy library to achieve continuous upgrading of inspection capabilities.

[0007] Preferably, in step S2, the scheduling logic of the dynamic collaborative scheduling engine is as follows: terminal resources are allocated according to the priority of fault level; for emergency faults, at least two terminals of different types are scheduled for joint inspection; for general faults, a single terminal is allocated according to the principle of optimal inspection efficiency; and for major faults, terminal inspection and emergency start / stop of associated equipment are triggered simultaneously.

[0008] Preferably, in step S3, the sensing data includes infrared temperature measurement data, laser point cloud data, audio feature data, ultrasonic flaw detection data, and hazardous gas concentration data, and the environmental time series data includes at least the temperature and humidity, dust concentration change trends, and extreme weather warning information for the past 72 hours.

[0009] Preferably, in step S4, the fault level is divided into three levels: general faults are instrument reading deviations and minor equipment dust accumulation; emergency faults are equipment overheating and dangerous gas concentration exceeding the standard; and major faults are belt tearing, fire and equipment leakage. Different levels correspond to clearly differentiated alarm intensity, response time and the authority of the handling body.

[0010] On the other hand, the present invention provides a comprehensive safety inspection system for smart power plants to implement the above-mentioned method. It adopts a five-layer architecture and a digital twin central integrated design. Each layer achieves low-latency interaction through industrial Ethernet and 5G slicing network, including a perception layer, a digital twin collaboration layer, a data fusion layer, a decision processing layer and an application layer. The perception layer constructs a three-dimensional perception network of air-ground, indoor-outdoor, fixed-mobile, including a high-altitude UAV equipped with a dual-light gimbal and lidar, a ground inspection unit with IP67 protection and ultrasonic flaw detection module, a fixed monitoring unit integrating lidar and UWB base station, and a personnel intelligent terminal with UWB positioning and one-click emergency call function. The digital twin collaboration layer is the core layer, with a built-in dynamic collaborative scheduling engine. It realizes dynamic resource replenishment based on fault level and terminal coverage capability, and updates the mapping relationship between the digital twin model and the terminal status in real time. The data fusion layer is equipped with a multi-source heterogeneous data deep fusion engine, which integrates perception data, ledger data and environmental time series data to complete data cleaning, correlation and feature extraction. The decision-making and handling layer constructs a hierarchical early warning, automatic handling and self-healing closed-loop mechanism, configures a preset handling strategy library, and supports fault tracing and blockchain evidence storage; The application layer provides a B / S architecture visualization platform that supports seamless integration with existing SCADA systems and equipment management platforms in power plants.

[0011] Preferably, the ground inspection unit includes a track-mounted robot and an ultra-thin mini robot. The track-mounted robot is equipped with a fire extinguishing bomb module, and the ultra-thin mini robot has a thickness of ≤80mm, which is suitable for the inspection needs of specific narrow spaces such as the bottom of the belt and the gaps in the pipe gallery.

[0012] Preferably, the digital twin model supports full lifecycle status mapping of equipment, updates equipment maintenance records, fault history and operating parameters in real time, and has the functions of automatic blind spot identification and dynamic correction of inspection paths.

[0013] Preferably, the data fusion layer uses a timestamp alignment algorithm to achieve spatiotemporal synchronization of multi-source data, and strengthens the fault feature weights by improving the attention mechanism of the CNN-LSTM model to improve the accuracy of fault prediction.

[0014] Preferably, the blockchain-stored evidence in the decision-making and handling layer includes inspection data, fault analysis reports, handling process records, and verification results, ensuring that the entire process is traceable and tamper-proof.

[0015] Preferably, the application layer supports automatic generation of inspection logs, hierarchical management of personnel permissions, and three-dimensional situation visualization. The frequency of inspection tasks and fault judgment thresholds can be customized according to operation and maintenance needs.

[0016] The technical effects and advantages of this invention are as follows: 1. This invention utilizes a three-dimensional perception network of "air-ground-indoor-outdoor-fixed-mobile" in the perception layer, combined with the collaboration of multiple types of terminals such as ultra-thin Mini robots and high-altitude drones, to achieve comprehensive inspection of narrow / hidden areas such as high-altitude cables, belt bottoms, and gaps in pipelines in complex scenarios such as coal conveyor corridors in power plants. At the same time, it uses a digital twin model to map the inspection status in real time and dynamically correct the inspection path, eliminating blind spots in traditional inspections, improving the comprehensiveness and safety of inspections, reducing the safety risks and labor intensity of manual inspections, and solving the technical pain points of existing power plant inspections, such as many blind spots, incomplete coverage, and high dependence on manual labor. 2. This invention improves the attention mechanism of the CNN-LSTM fusion model through the data fusion layer, realizing deep fusion of multi-source data and accurate extraction of fault features, significantly improving the accuracy of fault prediction, and enabling early warning of potential faults. Combined with the hierarchical early warning and automatic handling mechanism of the decision-making and handling layer and the dynamic scheduling engine of the digital twin collaboration layer, it realizes rapid fault response, hierarchical handling and closed-loop management of the whole process, shortens the fault handling response time, avoids the fault from escalating and causing production interruption, improves the intelligence and efficiency of power plant operation and maintenance, and improves the intelligence level of power plant inspection and fault handling efficiency. 3. This invention uses a blockchain-based evidence storage module in the decision-making and handling layer to store all data throughout the entire process, including inspection data, fault analysis reports, and handling process records. This ensures that the data is traceable and tamper-proof, meeting the compliance requirements of power plant operation and maintenance. Furthermore, the system adopts a five-layer architecture and an integrated digital twin hub design, allowing for seamless integration with existing SCADA systems and equipment management platforms in power plants without requiring overall modifications. It adapts to the inspection needs of various smart power plants, facilitating large-scale deployment and application. This significantly reduces power plant operation and maintenance costs, improves operation and maintenance management, and achieves full traceability and tamper-proof data throughout the inspection process. It also possesses strong feasibility and promotional value. Attached Figure Description

[0017] Figure 1 This is a flowchart of the method of the present invention.

[0018] Figure 2 This is a system diagram of the present invention.

[0019] The attached diagram is labeled as follows: 1. Perception layer; 2. Digital twin collaboration layer; 3. Data fusion layer; 4. Decision-making and processing layer; 5. Application layer; 101. High-altitude unmanned aerial vehicle; 102. Ground inspection unit; 103. Fixed monitoring unit; 104. Personnel intelligent terminal. Detailed Implementation

[0020] The present invention will be further described in detail below with reference to the accompanying drawings, using a typical example of a coal conveying pipeline inspection scenario in a thermal power plant. This embodiment is only used to explain the present invention and is not intended to limit its scope of protection. The present invention is also applicable to the full-scenario safety inspection of various smart power plants such as hydropower plants, photovoltaic power plants, and nuclear power plants.

[0021] Example 1 As attached Figure 2As shown, this embodiment provides a comprehensive safety inspection system for smart power plants, using a 300MW coal conveyor corridor as an application scenario. This corridor is 800 meters long and includes a belt conveyor section, transfer station, cable trays, and a gas monitoring area. It suffers from drawbacks such as blind spots in high-altitude cable inspections, difficulties in detecting the bottom of the conveyor belts, and the easy accumulation of hazardous gases. This embodiment deploys the comprehensive safety inspection system of this invention to execute corresponding inspection methods, achieving comprehensive safety management of equipment, personnel, and the environment within the corridor, thus solving the problems of low efficiency, numerous blind spots, and slow response in traditional inspections.

[0022] In this embodiment, the hardware selection and software configuration of each layer of the system meet the requirements of industrial-grade operation and maintenance, and can be directly adapted to the existing SCADA system and equipment management platform of the power plant without overall modification, thus possessing strong feasibility.

[0023] This system adopts a five-layer architecture and an integrated digital twin hub design. Each layer achieves low-latency interaction with a 5G industry slice network with a latency of no more than 20ms via gigabit fiber optic industrial Ethernet. The specific deployment is as follows: Perception Layer 1 constructs a three-dimensional perception network encompassing "open ground - indoor and outdoor - fixed and mobile," precisely configuring terminal equipment for coal conveying pipeline corridor scenarios, as detailed below: High-altitude UAV 101: This is an industrial-grade multi-rotor UAV with a payload of no less than 5kg and a flight time of no less than 40 minutes. It is equipped with a dual-light gimbal and a 16-line LiDAR. The dual-light gimbal has an infrared resolution of 640×512 and a visible light resolution of 4K. The 16-line LiDAR has a ranging range of 0.5-100m and an accuracy controlled within ±2cm. This UAV has centimeter-level RTK positioning capabilities and is deployed at the inspection port on the top of the utility tunnel to inspect cables and cable trays. It can autonomously avoid obstacles such as beams and pipes within the utility tunnel.

[0024] Ground inspection unit 102 includes a track-mounted robot deployed along a pre-set I-beam guide rail on the side wall of the utility tunnel. It has an IP67 protection rating and is equipped with an infrared temperature sensor, an ultrasonic flaw detection module, and a dry powder fire extinguishing bomb module. The infrared temperature sensor has a temperature range of -20℃ to 150℃ with an accuracy of ±0.5℃. The ultrasonic flaw detection module has a detection depth of no more than 50mm. The dry powder fire extinguishing bomb module has a capacity of 1kg and an effective spray distance of 3m. One unit is deployed every 50 meters to inspect the equipment on the side wall of the utility tunnel, the side of the conveyor belt, and the transfer station. It also includes an ultra-thin Mini robot, 75mm thick (meeting the design requirement of no more than 80mm) and weighing no more than 3kg. Equipped with a high-definition camera and vibration sensor, it can move along the bottom track of the conveyor belt, adapting to the bottom of the conveyor belt and specific narrow spaces in the gaps of the utility tunnel. It can accurately detect belt joint wear and roller abnormalities. One unit is deployed every 100 meters and can autonomously recharge via a wireless charging base.

[0025] Fixed monitoring unit 103: At the entrance of the utility tunnel, transfer station and areas prone to gas accumulation, one integrated monitoring device is deployed every 20 meters. It integrates lidar and UWB base station. The lidar is used for personnel and equipment positioning, and the UWB base station positioning accuracy is ±10cm. The device synchronously collects environmental parameters such as temperature, humidity, dust concentration and gas concentration. The temperature and humidity ranges are 0-60℃ and 0-95%RH, respectively. The dust concentration range is 0-100mg / m³, and the gas concentration range is 0-5%VOL. The data sampling frequency is 1 time / second.

[0026] Personnel Intelligent Terminal 104: Inspection personnel wear intelligent safety helmets that integrate UWB positioning modules, sound and light alarms, and one-click emergency call buttons. The terminal can upload personnel posture (standing, bending over, falling) and safety helmet wearing status in real time. When personnel enter a dangerous area, the terminal automatically triggers sound and light alarms and synchronizes them to the dispatch platform.

[0027] Digital twin collaboration layer 2 is the core layer of the system. It builds a 1:1 digital twin model of the coal conveying pipeline based on the Unity3D engine, imports basic data such as pipeline structure, equipment ledger, inspection points, and dangerous area boundaries, and updates the model 10 times / second. It maps the location, operating status (battery power, sensor accuracy, task progress) and equipment operating parameters of each terminal in the perception layer 1 in real time. The digital twin collaboration layer 2 has a built-in dynamic collaborative scheduling engine, developed in Python and deployed on an industrial server. The server is equipped with a CPU and 64GB of memory. The scheduling logic is preset inside the engine: terminal resources are allocated according to the fault level priority (major fault > emergency fault > general fault). Among them, emergency faults prioritize the scheduling of at least two terminals of different types for joint inspection. General faults allocate a single terminal according to the principle of optimal inspection efficiency. Major faults simultaneously trigger terminal inspection and emergency start and stop of associated equipment.

[0028] Data fusion layer 3 deploys a deep fusion engine for multi-source heterogeneous data, running on edge computing nodes with a computing power of no less than 20 TOPS, integrating and processing three types of data, as detailed below: Sensing data includes infrared temperature measurement and laser point cloud data collected by UAVs, ultrasonic flaw detection and vibration data collected by ground robots, environmental parameter data collected by fixed monitoring unit 103, and positioning and attitude data uploaded by personnel terminals. The data is uniformly converted into JSON format before transmission. Equipment ledger data: Import the full life cycle records of equipment such as belts, idlers, and motors in the coal conveyor corridor, including structured data such as factory parameters, installation time, maintenance records of the past 3 years, and fault history of the past 5 years, and store them in a MySQL database; Environmental time series data: Collect temperature, humidity and dust concentration trend data for nearly 72 hours, and simultaneously access extreme weather warning information such as rainstorm, strong wind and high temperature from the power plant meteorological station. Multi-source data spatiotemporal synchronization is achieved through a timestamp alignment algorithm with an accuracy of no more than 1ms. An improved attention mechanism-based CNN-LSTM fusion model is adopted. The model input is multi-dimensional fused data, and the output is fault features and prediction results. The CNN module is responsible for extracting spatial features such as equipment appearance defects and environmental distribution differences, while the LSTM module is responsible for mining temporal features such as parameter change trends. The attention mechanism strengthens the weight of fault-related features. The model training dataset uses inspection and fault data from the power plant over the past year, with a sample size of no less than 100,000, and the fault prediction accuracy is no less than 92%.

[0029] Decision-making and processing layer 4, deployed on the power plant's central control room server, establishes a hierarchical early warning, automatic response, and self-healing closed-loop mechanism, with the following specific configuration: Tiered early warning mechanism: A three-level fault judgment standard and corresponding handling strategy library are preset and stored in a Redis cache database, supporting dynamic updates: General faults: When the instrument reading deviation is within ±5% or there is slight dust accumulation on the equipment, only the corresponding maintenance personnel's mobile terminal is pushed; Emergency faults: When the equipment temperature exceeds 80℃ or the gas concentration is not lower than 1%VOL, an audible and visual alarm is triggered in the pipe gallery, with the alarm volume not lower than 100dB and the light flashing frequency at 2 times / second, and is simultaneously pushed to the central control room and the inspection personnel's terminal; Major faults: Belt tear, fire, equipment leakage, in addition to the above alarms, automatically trigger the emergency start and stop of related equipment, such as belt shutdown and cutting off the area power supply.

[0030] Automatic handling module: It is linked with the fire extinguishing bomb module and belt speed regulation system of the track-mounted robot. For minor faults (such as local dust accumulation), the robot can be scheduled to start the high-pressure blowing module to clean it. For emergency faults (such as minor fires), the fire extinguishing bomb module will be automatically deployed.

[0031] Blockchain Evidence Storage Module: Adopting a consortium blockchain architecture, the evidence storage content includes inspection data, fault analysis reports, handling process records and review results. Each block contains a timestamp and hash value to ensure that the data is tamper-proof and the entire process is traceable. The evidence storage nodes are deployed on the power plant's local server to avoid data leakage.

[0032] Application layer 5 provides a B / S architecture visualization platform, supporting access via Chrome and Edge browsers. The interface includes 3D situation display, fault tracing, automatic generation of inspection logs, and hierarchical personnel access control. The platform seamlessly integrates with the power plant's existing SCADA system and equipment management platform via the OP CUA protocol, enabling real-time retrieval of equipment operating parameters and historical maintenance records. It also supports customizable inspection task frequencies, such as belt conveyor inspections every 2 hours and cable inspections every 4 hours. Fault judgment thresholds can also be adjusted, for example, setting the gas concentration alarm threshold to 0.8% VOL.

[0033] Example 2 As attached Figure 1 As shown, based on the above system deployment, the inspection method in this embodiment is executed according to the following steps to achieve full-process safety inspection of the coal conveying pipeline: S1. Model Initialization and Task Deployment Operations and maintenance personnel complete the initial configuration through the application layer 5 visualization platform, specifically as follows: Load a 1:1 digital twin model of the coal conveyor corridor, import equipment ledgers such as belts, idlers, and motors, mark 32 inspection points, including 20 regular points and 12 dangerous area points, and delineate 8 dangerous area boundaries, including gas accumulation areas and high-voltage cable areas. Routine inspection parameters are set as follows: the high-altitude drone 101 inspects the cables on the top of the pipe gallery once every 4 hours, the track-mounted robot inspects the equipment on the side wall of the pipe gallery once every 2 hours, the Mini robot inspects the bottom of the conveyor belt once every 1 hour, and the fixed monitoring unit 103 collects environmental parameters in real time. Set fault judgment thresholds: such as equipment over-temperature threshold of 80℃, gas concentration alarm threshold of 1%VOL, and instrument reading deviation threshold of ±5%. The system will automatically decompose the task to each sensing terminal, and the terminal will enter the inspection waiting state after receiving the task.

[0034] S2, Multi-terminal Collaborative Inspection Each sensing terminal executes its inspection task according to the initial path. The digital twin model maps the terminal's location, operating status, and collected data in real time, as detailed below: The drone flies along a pre-set path on the top of the utility tunnel, avoids obstacles on the crossbeams using lidar, and simultaneously collects data on cable temperature and appearance. The track-mounted robot moves along the guide rail to perform infrared temperature measurement and ultrasonic flaw detection on the side of the belt and the motor housing. The mini robot moves along the bottom track of the belt to collect data on belt joint wear and idler roller vibration. The fixed monitoring unit 103 uploads real-time data on temperature, humidity, dust, and gas concentration, while the personnel terminal uploads data on the location and posture of the inspection personnel. The dynamic collaborative scheduling engine monitors the inspection coverage in real time, as detailed below: When the Mini robot detects an abnormal temperature difference at the bottom of a conveyor belt roller, reaching 78°C and approaching the alarm threshold, it determines it as a potential emergency fault. It automatically dispatches a nearby track-mounted robot 15 meters away from the fault location in the pipe gallery to the side of the roller to verify the temperature. At the same time, it dispatches a drone to adjust its path and take pictures of the surrounding environment of the roller from the top of the pipe gallery, forming a multi-terminal linkage inspection. When the battery level of a certain tracked robot drops below 20%, the engine automatically plans the optimal charging path and dispatches the robot to the nearest charging station 8 meters away to replenish its power. At the same time, it assigns adjacent tracked robots to cover the unfinished inspection tasks to ensure the continuity of inspection.

[0035] S3, Multi-source data fusion analysis Data fusion layer 3 performs time and space alignment on the data collected by each terminal: the fault location data collected by drones, tracked robots and mini robots are synchronized to the same time axis through the timestamp alignment algorithm, and the multi-terminal data are mapped to the same spatial location of the digital twin model based on the lidar positioning data. An improved attention mechanism CNN-LSTM fusion model performs in-depth data analysis: integrating 78℃ idler roller temperature data, vibration data with an amplitude of 0.3mm (exceeding the normal range of 0.1mm), equipment ledger data (the idler roller has been running for 18 months, and the last overhaul was 6 months ago), and environmental time series data (temperature and humidity in the pipe gallery have been rising in the past 72 hours, and dust concentration is normal), mining data correlations, determining that the idler roller has a wear fault, and the fault trend is aggravating, predicting that there may be a risk of jamming within 24 hours, generating a fault analysis report (including fault location, type, severity, and prediction results) and uploading it to the decision-making and handling layer 4.

[0036] S4. Tiered handling and closed-loop review Based on the fault analysis report, the decision-making and handling layer 4 determines that the idler roller wear is an emergency fault and executes the corresponding handling strategy, as follows: Trigger the audible and visual alarms in the corresponding area of ​​the utility tunnel, and simultaneously push fault information (location, type, and prediction results) to the central control room platform and the terminals of the two inspection personnel responsible for the area, clarifying the response time of arriving at the scene within 30 minutes and the main authority of the maintenance team leader to take the lead in handling the situation; The tracked robot is deployed to stand by around the faulty idler, the fire extinguishing module is activated for backup, and the SCADA system is linked to closely monitor the idler's operating parameters to prevent the fault from escalating. After the inspection personnel arrive at the site, they replace the worn idler roller. After the repair is completed, they upload the repair record through the terminal, including the replacement time, the new idler roller model, and the personnel involved. The system automatically dispatches a Mini robot to the original fault location to verify the temperature and vibration data of the idler roller. The verified temperature is 45℃, which is normal, and the amplitude is 0.08mm. After confirming that the hidden danger has been eliminated, the system updates the operating status and ledger data of the idler roller in the digital twin model, forming a closed loop of the entire process of "inspection-analysis-early warning-repair-verification".

[0037] S5, Data Review and Model Optimization The system performs big data analysis on inspection data and fault handling data weekly, as detailed below: The incidence rates of various types of failures were statistically analyzed, with roller wear accounting for 35% and cable aging accounting for 20%. Based on this, the frequency of roller inspections was optimized and adjusted to once every 1.5 hours. Based on new fault data, the parameters of the CNN-LSTM model are iteratively optimized and improved to strengthen the correlation weights of roller vibration, temperature and wear faults, thereby improving the accuracy of fault prediction. The handling strategy library was updated to include quick changeover procedures for different idler roller models, shortening handling time. Through continuous review and optimization, inspection capabilities were iteratively upgraded.

[0038] In this embodiment, the system and method of this invention were deployed in the coal conveying pipeline of a 300MW thermal power plant. After three months of stable operation, compared with traditional manual inspection and existing intelligent inspection solutions, all operation and maintenance indicators were significantly improved. The specific results are as follows: Inspection efficiency has been improved by 45%, and the inspection time for the entire utility tunnel area has been reduced from 2 hours for traditional manual inspection to 45 minutes. Moreover, it can achieve uninterrupted inspection at all times without the need for manual duty. The accuracy of fault prediction was improved by 32%. During the period, 12 potential faults were successfully predicted 24 to 48 hours in advance, including 8 cases of roller wear, 3 cases of cable aging and 1 case of gas leakage hazard, which effectively prevented production interruptions caused by the expansion of faults. The response time for major faults has been reduced by 50%, from the traditional 1 hour to within 30 minutes, allowing sufficient time for fault handling. The cost of manual intervention is reduced by 60%. Minor faults can be handled automatically by the system without the need for human intervention. Only major and complex faults require human intervention, which greatly reduces the manpower required for operation and maintenance.

[0039] In summary, the method and system of this invention have strong feasibility and promotional value, and can effectively improve the safety, intelligence level and operation and maintenance efficiency of smart power plant inspections, and are fully adaptable to the inspection needs of various complex power plant scenarios.

[0040] The system is highly feasible and has great potential for widespread adoption. It can effectively improve the safety, intelligence, and operational efficiency of smart power plant inspections and is fully adaptable to the inspection needs of various complex power plant scenarios.

[0041] The above are merely preferred embodiments of the present invention and are not intended to limit the present invention. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the protection scope of the present invention.

Claims

1. A comprehensive safety inspection method for a smart power plant, characterized in that: Includes the following steps: S1. Model initialization and task deployment: Construct a 1:1 digital twin model of the power plant, import equipment ledgers, inspection points and hazardous area boundary data, set routine inspection task parameters and fault judgment thresholds, and automatically decompose tasks to each sensing terminal. S2. Multi-terminal collaborative inspection: Each sensing terminal performs inspection according to the initial path. The digital twin model maps the terminal location, operating status and collected data in real time. The dynamic collaborative scheduling engine monitors the integrity of inspection coverage, terminal operating status and fault location information in real time, and automatically completes blind spot filling, low-battery terminal charging path planning and multi-terminal linkage verification of fault locations. S3. Multi-source data fusion analysis: Time and space alignment of sensing data, equipment ledger data and environmental time series data. By improving the attention mechanism of the CNN-LSTM fusion model, the correlation between fault features and multi-dimensional data is mined to identify equipment defects, unsafe personnel behaviors and environmental risks, and generate fault analysis reports and prediction results. S4. Tiered handling and closed-loop review: The corresponding handling strategy is executed according to the fault level. Minor faults automatically trigger preset handling actions. Emergency and major faults are linked to audible and visual alarms, personnel positioning and equipment start and stop control. After the handling is completed, the dispatch terminal reviews and updates the system data after confirming that the hidden danger has been eliminated, forming a closed loop of the whole process. S5. Data review and model optimization: Regularly analyze inspection and handling data, iteratively optimize AI recognition model parameters, dynamic scheduling strategies and handling strategy library to achieve continuous upgrading of inspection capabilities.

2. The method according to claim 1, characterized in that: In step S2, the scheduling logic of the dynamic collaborative scheduling engine is as follows: terminal resources are allocated according to the fault level priority; for emergency faults, at least two terminals of different types are scheduled for joint inspection; for general faults, a single terminal is allocated according to the principle of optimal inspection efficiency; and for major faults, terminal inspection and emergency start / stop of associated equipment are triggered simultaneously.

3. The method according to claim 1, characterized in that: In step S3, the sensing data includes infrared temperature measurement data, laser point cloud data, audio feature data, ultrasonic flaw detection data, and hazardous gas concentration data. The environmental time series data includes at least the temperature and humidity, dust concentration change trends, and extreme weather warning information for the past 72 hours.

4. The method according to claim 1, characterized in that: In step S4, the fault level is divided into three levels: general faults include instrument reading deviation and minor equipment dust accumulation; emergency faults include equipment overheating and excessive concentration of hazardous gases; and major faults include belt tearing, fire and equipment leakage. Each level corresponds to a clear differentiation in alarm intensity, response time and the authority of the handling body.

5. A comprehensive safety inspection system for a smart power plant, used to implement the method described in any one of claims 1-4, characterized in that: The five-layer architecture and digital twin central design are adopted. Each layer achieves low-latency interaction through industrial Ethernet and 5G slicing network, including the perception layer (1), digital twin collaboration layer (2), data fusion layer (3), decision processing layer (4) and application layer (5). The perception layer (1) constructs an air-ground-indoor-outdoor-fixed-mobile three-dimensional perception network, including a high-altitude UAV (101) equipped with a dual-light gimbal and lidar, a ground inspection unit (102) with IP67 protection and ultrasonic flaw detection module, a fixed monitoring unit (103) integrating lidar and UWB base station, and a personnel intelligent terminal (104) with UWB positioning and one-click emergency call function. The digital twin collaboration layer (2) is the core layer, with a built-in dynamic collaborative scheduling engine. Based on the fault level and terminal coverage capability, it realizes dynamic resource replenishment and updates the mapping relationship between the digital twin model and the terminal status in real time. The data fusion layer (3) is equipped with a multi-source heterogeneous data deep fusion engine to integrate perception data, ledger data and environmental time series data to complete data cleaning, association and feature extraction; The decision-making and handling layer (4) constructs a hierarchical early warning, automatic handling and self-healing closed-loop mechanism, configures a preset handling strategy library, and supports fault tracing and blockchain evidence storage. The application layer (5) provides a B / S architecture visualization platform that supports seamless integration with the power plant's existing SCADA system and equipment management platform.

6. The system according to claim 5, characterized in that: The ground inspection unit (102) includes a track-mounted robot and an ultra-thin Mini robot. The track-mounted robot is equipped with a fire extinguishing bomb module, and the ultra-thin Mini robot has a thickness of ≤80mm, which is suitable for the inspection needs of specific narrow spaces such as the bottom of the belt and the gaps in the pipe gallery.

7. The system according to claim 5, characterized in that: The digital twin model supports full lifecycle status mapping of equipment, updates equipment maintenance records, fault history and operating parameters in real time, and has the functions of automatic blind spot identification and dynamic correction of inspection paths.

8. The system according to claim 5, characterized in that: The data fusion layer (3) uses a timestamp alignment algorithm to achieve spatiotemporal synchronization of multi-source data, and strengthens the fault feature weights by improving the attention mechanism of the CNN-LSTM model to improve the accuracy of fault prediction.

9. The system according to claim 5, characterized in that: The blockchain-based evidence storage content of the decision-making and handling layer (4) includes inspection data, fault analysis reports, handling process records and review results, ensuring that the entire process is traceable and tamper-proof.

10. The system according to claim 5, characterized in that: The application layer (5) supports automatic generation of inspection logs, hierarchical management of personnel permissions and three-dimensional situation visualization. The frequency of inspection tasks and fault judgment thresholds can be customized according to operation and maintenance needs.

Images