Multi-physical-form ultra-high-voltage converter station operation and maintenance safety control Al recorder and method adapting to strong electric field environment

Abstract

The invention discloses a multi-physical-form ultra-high-voltage converter station operation and maintenance safety control Al recorder and method suitable for a strong electric field environment. According to the method, through an AI recorder, in combination with noise, environmental electromagnetic intensity, temperature and humidity and other multi-sensor data, intelligent identification of converter station equipment types and operation objects is achieved, and patrol tasks are automatically matched. In the inspection process, the AI recorder collects multi-mode data such as audios, pictures and videos in real time, and in combination with an abnormity linkage detection algorithm, equipment abnormity is intelligently captured and a maintainer is prompted in time. A maintainer can interact with the AI through voice to obtain operation guidance and emergency disposal suggestions. And all data are automatically encrypted and synchronized to the cloud platform, so that the whole-process data security archiving is realized. The AI recorder can be hung in the chest, worn on the head or matched with a safety belt, is provided with an LED illumination structure, a semi-arc shell structure and the like, is suitable for various operation and maintenance scenes, and greatly improves the inspection safety and intelligent level of the ultrahigh-voltage converter station.

Description

Technical Field

[0001] This invention belongs to the field of artificial intelligence and converter station operation and maintenance, and more specifically relates to an AI recorder and method for safety management and control of operation and maintenance of ultra-high voltage converter stations with multiple physical forms that are adapted to strong electric field environments. Background Technology

[0002] With the transformation of my country's energy structure and the widespread application of ultra-high voltage direct current (UHVDC) transmission technology, ultra-high voltage converter stations, as key hubs in the power grid, undertake the important task of large-scale, long-distance, and efficient power transmission. The internal equipment structure of converter stations is complex, including core components such as valve halls, cooling systems, and high-voltage electrical equipment. Especially in multi-physical environments characterized by strong electric fields, strong electromagnetic interference, and high noise, monitoring equipment operation status and troubleshooting are extremely difficult. In recent years, although some converter stations have adopted digital operation and maintenance management systems and wearable recorders, existing technologies are mostly limited to environmental monitoring or simple data collection, failing to achieve intelligent fusion of multi-source heterogeneous information and real-time risk management of the operation process.

[0003] Currently, routine inspections and maintenance of converter stations rely primarily on manual experience. Maintenance personnel operate in harsh environments with high voltage, strong electromagnetic fields, and high noise levels, leading to cumbersome procedures, delayed risk warnings, and untimely handling of anomalies. Traditional maintenance record-keeping methods, typically involving manual form filling, paper records, or single audio / video capture, struggle to capture multimodal information in a timely manner and ensure data integrity and accuracy. In practice, when equipment malfunctions, maintenance personnel often lack quick access to historical case studies and operational guidelines, delaying response times and potentially causing safety accidents.

[0004] In the operation and maintenance of existing converter stations, the on-site modification of air-cooled systems also faces extremely high safety risks and technical challenges. Air-cooled systems are typically located in open or semi-open high-pressure environments with dense equipment and piping, presenting challenges such as confined working spaces, frequent high-altitude work, and strong electromagnetic interference. Traditional modification work often lacks real-time risk monitoring and intelligent process control, relying on manual inspections and verbal coordination, which can easily lead to omissions, unclear hazard identification, and delayed emergency response. Existing information technology methods are insufficient for multimodal data acquisition and real-time anomaly warnings throughout the entire operation process, especially in the disassembly, wiring, and testing of critical air-cooled system equipment, making it difficult to promptly identify and address potential risks. With the increasing demand for converter station equipment upgrades and system modifications, leveraging intelligent AI recorders and multi-sensor fusion perception to achieve safe control and efficient archiving of the entire air-cooled system modification process has become a crucial direction for achieving inherent safety and intelligent operation and maintenance of converter stations.

[0005] With the development of next-generation information technologies such as artificial intelligence, sensing, and the Internet of Things, integrating multimodal sensing, intelligent recognition, voice interaction, and big data analysis into the operation and maintenance of converter stations has become a hot topic in industry research and technological innovation. Collecting on-site audio and video data and environmental parameters through portable intelligent terminals, combined with AI algorithms, enables automatic equipment status analysis, rapid anomaly detection, and intelligent risk warning, significantly improving maintenance efficiency and safety levels, and promoting the intelligent operation and maintenance upgrade of converter stations. However, there is currently a lack of professional AI recorders and their safety management systems that can adapt to complex environments with strong electric fields, possess multi-physical quantity fusion capabilities, and offer intelligent voice interaction. There is an urgent need to research and develop new intelligent equipment and methods that can provide real-time assistance in maintenance and ensure operational safety.

[0006] Existing technologies mainly suffer from two problems: firstly, the level of informatization and intelligence in on-site inspection methods is not high, data collection and analysis are lagging, making it difficult to detect potential faults in a timely manner; secondly, there is a lack of personalized intelligent inspection solutions for the complex operating conditions and diverse cooling types of converter station cooling systems. How to integrate wearable AI devices, edge intelligent analysis, multi-sensor information fusion, and cloud-based big data archiving and knowledge base reasoning to construct a full-process intelligent inspection and maintenance method for cooling systems has become an urgent problem to be solved in the digital and intelligent upgrading of converter stations.

[0007] Therefore, this paper proposes an intelligent inspection and maintenance method for converter station cooling systems based on badge-type or head-mounted AI recorders. Combining multi-sensor real-time perception, AI-assisted anomaly detection, voice-interactive operation guidance, and secure data archiving, this method effectively improves the intelligence level, operational efficiency, and safety of cooling system inspection and maintenance. This new method aligns with the trend of digital and intelligent transformation in power operation and maintenance, significantly reducing reliance on manual labor and operational risks, and providing solid technical support for the long-term safe and stable operation of converter stations. Summary of the Invention

[0008] This invention addresses the problems existing in the operation and maintenance of ultra-high voltage converter stations, such as low efficiency of manual inspections under high electric field intensity and multi-physical disturbance environments, high operational risks, untimely anomaly detection, incomplete and difficult-to-trace data collection, and lack of operational guidance. It provides a multi-physical-mode ultra-high voltage converter station operation and maintenance safety management method and matching intelligent AI recorder that can realize multi-modal perception information fusion, automatic identification of inspection objects, intelligent anomaly early warning, AI-assisted operation guidance, and safety data archiving. It aims to improve the level of intelligent and refined management and control of the maintenance process, and comprehensively ensure the safety of maintenance personnel and the integrity of operation and maintenance data.

[0009] To achieve the above objectives, the present invention employs the following technical solution: the method comprises:

[0010] Before the inspection, maintenance personnel enter the equipment area of ​​the converter station, wear a badge-style or head-mounted AI recorder, and automatically synchronize the current inspection task, the list of key inspection objects, and precaution information to the maintenance personnel interface.

[0011] The converter station equipment type and inspection object are automatically identified. The AI ​​recorder combines noise spectrum, environmental electromagnetic intensity, temperature and humidity data from multiple sensors to automatically determine whether the current work area is the cooling system, electrical equipment, power lines, or other equipment, and automatically matches the core equipment to be inspected based on the on-site scan.

[0012] On-site inspection and process information collection: maintenance personnel check each pipeline step by step according to the inspection route under guidance. The AI ​​recorder automatically collects and stores multimodal data such as audio, pictures and videos of the inspection process. When an abnormality is found, it automatically associates fault photos / video clips through working voice notes to improve the accuracy and completeness of the record.

[0013] Intelligent anomaly detection and real-time risk alerts input audio, video, image, and various environmental sensor data into the anomaly linkage detection algorithm in real time, automatically comparing the operating noise of each device, environmental electromagnetic intensity, and air temperature and humidity information; for detected anomalies, the equipment immediately alerts the maintenance personnel to pay attention through voice / vibration.

[0014] AI-assisted question answering and operation guidance: On-site maintenance personnel can ask questions via voice. The AI ​​recorder, based on the currently collected information and the converter station's dedicated knowledge base, uses a context fusion reasoning algorithm to provide maintenance personnel with accurate operating procedures, historical case references, or emergency handling suggestions.

[0015] Data security synchronization and process archiving: All data during the inspection process is automatically classified, encrypted, and synchronized to the secure cloud platform.

[0016] In one scheme, the self-identification of converter station equipment type and inspection object includes: comprehensive perception of the current work area; the equipment has multiple built-in sensors to collect environmental noise signals, air and environmental electromagnetic intensity parameters and temperature time series data, and generate a comprehensive feature vector.

[0017] The AI ​​model uses a pre-trained discriminator to identify the converter station equipment type label from the environmental feature vector;

[0018] The AI ​​recorder's camera activates target detection and text recognition algorithms. Using on-site QR codes or equipment nameplate information, it extracts the location of the equipment and obtains its unique identification code. By dynamically comparing the equipment identification with the inspection task list, it automatically matches the set of core equipment that needs to be inspected in this inspection in real time.

[0019] In one solution, the intelligent anomaly capture and real-time risk alert includes: the AI ​​recorder sends the audio, video, images and various environmental sensor data collected on-site to the device anomaly linkage detection algorithm for analysis in real time;

[0020] Audio anomalies are detected by comparing healthy samples with spectrograms; anomalies in temperature and humidity sensor data and environmental electromagnetic intensity data are detected by comparing moving mean and standard deviation; and appearance anomalies are detected in images and videos by using deep convolutional neural networks.

[0021] When a certain modality is abnormal, the linkage algorithm dynamically adjusts the sampling frequency and analysis sensitivity of relevant data, improves image resolution, and guides maintenance personnel to supplement the collected information. After the multimodal abnormality scores are weighted and fused, risk warnings are given through voice broadcast, interface highlighting, or vibration, and detailed abnormality records are guided.

[0022] In one solution, the AI-assisted question answering and operation guidance includes: using a localized speech recognition model to transcribe the maintenance personnel's on-site questions into text, and combining anomaly detection scores, real-time collected data, equipment operating status and historical inspection logs to form a comprehensive scene description vector;

[0023] By using a context fusion reasoning algorithm, the question text, scene data vector and knowledge graph are jointly encoded, and the question-knowledge matching score and the current anomaly detection distribution probability are weighted and sorted to select the operation steps, historical cases or handling suggestions most relevant to the problem and equipment status.

[0024] The AI-assisted question-and-answer and operation guidance include: broadcasting instructions to maintenance personnel in the form of natural language, and supplementing the image acquisition process and additional instructions based on the on-site matching degree.

[0025] In one solution, the data security synchronization and process archiving include: the AI ​​recorder manages all data collected during the inspection process in a unified manner, classifies and labels the raw data, and synchronizes the data based on user permissions and enterprise security policies.

[0026] The synchronized data is automatically organized into structured inspection logs, generating event indexes and summaries, and compiling multimedia reports. Archived content is saved locally and synchronized to the cloud-based intelligent knowledge base.

[0027] In one solution, the on-site inspection and process information collection includes: an AI recorder automatically generates the optimal inspection path based on the synchronized inspection task and the type of converter station equipment, and guides the maintenance personnel to check the equipment step by step through voice, display or vibration; a multimodal acquisition module automatically records audio, video and photos and marks the time, geographical location and equipment number to realize abnormal information aggregation and event tracing; all data is encrypted and stored locally and synchronized to the cloud platform.

[0028] On the other hand, an AI recorder for safety management and control of operation and maintenance of ultra-high voltage converter stations with multiple physical forms that are adapted to strong electric field environments is provided. The AI ​​recorder is applicable to the method described above. The AI ​​recorder includes: a camera, a control module, a power module, a display screen, and a clip.

[0029] The control module includes: a wireless transmission module, a positioning module, a voice module, and a processing module;

[0030] The camera, positioning module, voice module, and display screen are all electrically connected to the processing module. The output of the processing module is electrically connected to the wireless transmission module, and the wireless transmission module is connected to the background database.

[0031] The power module provides power to the recorder, which is worn via a clip.

[0032] On the other hand, there is a badge-style AI recorder, which is used in conjunction with work clothes via a buckle, and the outer shell is shaped like a badge.

[0033] On another front, a head-mounted AI recorder, the head-mounted AI recorder being based on the AI ​​recorder, the head-mounted AI recorder further comprising;

[0034] The lighting module uses LED lighting beads and is connected to the power module. The recorder is connected to the helmet via a clip.

[0035] The AI ​​recorder has a semi-circular arc-shaped shell, which conforms to the shape of a helmet edge. The camera and lighting module are located at opposite ends of the shell, so they do not interfere with each other's operation.

[0036] On another front, a seatbelt-type AI recorder is provided, which is based on the AI ​​recorder and is used in conjunction with a seatbelt. The seatbelt-type AI recorder is engaged with the seatbelt via a buckle.

[0037] Beneficial effects of this invention:

[0038] This invention integrates a badge-style or head-mounted AI recorder, multiple environmental sensors, and AI intelligence. First, through the real-time acquisition and automatic archiving of multimodal data by the AI ​​recorder, the comprehensiveness and accuracy of inspection data are significantly improved, effectively avoiding problems such as information omissions and incomplete records during manual inspections.

[0039] Secondly, by utilizing intelligent anomaly detection algorithms and multi-sensor fusion analysis, real-time perception of the operating status of key equipment in the cooling system and fault linkage early warning can be achieved, greatly improving the timeliness and reliability of discovering hidden dangers and anomalies during inspections.

[0040] Third, by providing on-site AI-assisted Q&A and operation process guidance, it provides maintenance personnel with personalized and contextualized technical support and emergency response suggestions, reducing the risk of misoperation and knowledge gaps during operations. In addition, it supports automatic classification and tagging of inspection data, and encrypts and synchronizes it to the cloud according to permissions and security policies, which not only ensures data security, but also facilitates subsequent knowledge accumulation, incident tracing and management decision-making.

[0041] Finally, by automatically generating structured inspection logs and multimedia reports, the standardization, transparency, and traceability of maintenance work are greatly improved, providing solid support for the construction of an intelligent converter station operation and maintenance system. Attached Figure Description

[0042] Figure 1 This is a flowchart of the method of the present invention;

[0043] Figure 2 This is a block diagram of the AI ​​recorder system of the present invention;

[0044] Figure 3 This is a schematic diagram of wearing the head-mounted AI recorder of the present invention;

[0045] Figure 4 This is a structural diagram of the head-mounted AI recorder of the present invention;

[0046] Figure 5 This is a structural diagram of the badge-style AI recorder of the present invention;

[0047] Figure 6 This is a structural diagram of the seatbelt-type AI recorder of the present invention;

[0048] Figure 7 This is a side view of the AI ​​recorder of the present invention;

[0049] Figure 8 This is a flowchart illustrating the workflow of the AI ​​recorder of this invention.

[0050] Figure 9 This is a flowchart of the AI ​​recorder-assisted maintenance process of the present invention;

[0051] Figure 10 This is a diagram of the built-in neural network model of the AI ​​recorder of this invention.

[0052] In the diagram, 1-Power module, 2-Camera, 3-Lighting module, 4-Control module, 5-Display screen, 6-Snap fastener, 41-Wireless transmission module, 42-Positioning module, 43-Voice module, 44-Processing module. Detailed Implementation

[0053] To facilitate understanding of the present invention, a more complete description will be given below with reference to the accompanying drawings. Typical embodiments of the invention are shown in the drawings. However, the invention can be implemented in many different forms and is not limited to the embodiments described herein. Rather, these embodiments are provided so that this disclosure will be thorough and complete.

[0054] Unless otherwise defined, all technical and scientific terms used in this invention have the same meaning as understood by one of ordinary skill in the art to which this invention pertains. The terminology used in this specification is for the purpose of describing particular embodiments only and is not intended to limit the invention. To facilitate understanding, the invention will now be described more fully with reference to the accompanying drawings. Typical embodiments of the invention are shown in the drawings. However, the invention can be implemented in many different forms and is not limited to the embodiments described herein. Rather, these embodiments are provided to make the disclosure of the invention more thorough and complete.

[0055] Example 1:

[0056] like Figure 1 As shown, this invention addresses the technical challenges of traditional maintenance recording methods, such as incomplete information collection, inefficient management, high risk of data loss, and insufficient intelligent assistance. It proposes an intelligent inspection and maintenance method for converter station cooling systems based on badge-type or head-mounted AI recorders.

[0057] Step 1: Pre-inspection preparation and equipment initialization

[0058] This method is applicable to the safety management of operation and maintenance of multi-physical ultra-high voltage converter stations in strong electric field environments, during the pre-inspection preparation and equipment initialization phases. Maintenance personnel first enter the relevant work areas of the cooling system, including pump rooms, fan rooms, and other strong electric field locations, according to the prescribed attire requirements.

[0059] Maintenance personnel wear badge-style or head-mounted AI recorders. These devices are equipped with identity sensing and automatic power-on functions, and can adapt to special on-site environments with high voltage and strong electromagnetic interference. Once the device detects the maintenance personnel's arrival at the designated work area, it automatically connects to the converter station's wireless network or dedicated communication link and synchronizes data with the backend management system. After startup, the device pushes the currently assigned inspection tasks, a list of key equipment to be inspected in each area, specific inspection items, and on-site safety precautions to the maintenance personnel through a visual interface or voice broadcast, ensuring they fully understand the work content and safety points of this maintenance task.

[0060] Meanwhile, the AI ​​recorder automatically performs a self-check, including checking storage space, battery level, and sensor functionality, reminding maintenance personnel to complete relevant confirmation steps to ensure smooth inspection and data collection in strong electric field environments. This process not only enables standardized information dissemination and reception in remote multi-physical environments but also improves operational safety and data integrity through equipment initialization checks, laying the foundation for subsequent intelligent inspections and efficient recording.

[0061] Step 2: Self-identification of converter station equipment type and inspection objects

[0062] The AI ​​recorder activates a multimodal environment recognition algorithm and automatically matches the core equipment to be inspected based on on-site scanning (QR code / identification nameplate).

[0063] In the type identification and inspection object self-identification stages, this method is applicable to the operation and maintenance safety management of the entire ultra-high voltage converter station with multiple physical forms under strong electric field environments. The AI ​​recorder first activates a multi-modal environment recognition algorithm to comprehensively perceive and adapt to the environment of various key equipment involved in the current work area, including but not limited to cooling systems, electrical equipment, busbars, surge arresters, auxiliary power, and other operating locations. Multiple sensors built into the device collect noise signals from the surrounding environment. Environmental electromagnetic intensity With water vapor parameters, and temperature The time-series data is used, and real-time data preprocessing is performed in conjunction with the influence of complex physical fields such as strong electric fields.

[0064] For different system or device types, the algorithm calculates a comprehensive feature vector.

[0065]

[0066] Each feature is obtained through sensor signal acquisition and normalization processing.

[0067]

[0068] This represents the average noise level per unit time. (e.g., average relative humidity).

[0069] The AI ​​model uses a pre-trained discriminator. By leveraging lightweight convolutional networks, environmental feature vectors are mapped to target device or system type labels:

[0070]

[0071] in Activation is achieved through softmax, enabling automatic identification of regional cooling methods and the operating conditions of other key equipment, thus enhancing the robustness of the algorithm in regions coupled with high-voltage electric fields and multi-physics fields.

[0072] Meanwhile, the AI ​​recorder's camera activates object detection and text recognition algorithms. By recognizing information such as QR codes and equipment nameplates on site, it uses a YOLO object detection network to locate the device area. Then, the unique identifier of the device is extracted using OCR. By dynamically matching this identifier with the inspection task list (e.g., through hash mapping or SQLite local database queries), the system can identify key equipment and components requiring immediate inspection during the current inspection, and organize the process accordingly. Waiting for the inspection team.

[0073] During the self-identification and discrimination process, the AI ​​system automatically switches and activates detection and positioning modules adapted to various equipment types based on the judgment results. For example, for converter valves, busbars, or high-voltage electrical equipment, the system will guide maintenance personnel to focus on key nodes, insulation parts, and partial discharge characteristics. When inspecting the cooling system, it will highlight parts such as fans, water pumps, and pipeline connections. This multi-source feature fusion and deep learning discrimination mechanism enables flexible and dynamic adaptation and efficient positioning of key inspection environments and objects in ultra-high-voltage converter stations with strong electric fields and multiple physical forms. While improving the automation and accuracy of inspections, it comprehensively ensures safety compliance and data reliability during the operation process.

[0074] Step 3: On-site inspection and process information collection

[0075] Guided by the multimodal operation of the AI ​​recorder and guided by the automatically generated optimal inspection path, maintenance personnel sequentially inspect all types of high-voltage equipment and systems within the entire station, including but not limited to cooling devices, converter valves, busbars, surge arresters, auxiliary power supplies, and control cabinets. The AI ​​recorder automatically adapts to the type of equipment being inspected and the task priority, utilizing multi-source sensors and depth perception algorithms to dynamically adjust the data collection strategy, ensuring coverage of equipment operating characteristics and key maintenance parameters under multi-physics environments.

[0076] During the inspection, the AI ​​recorder's multimodal acquisition module operated continuously and efficiently, automatically capturing and storing high-resolution images of key components' appearance and the environment, including maintenance personnel's voice conversations, fault discussions, video streams of on-site operations, and data from sensors such as temperature, humidity, current, and vibration. The system precisely labels all data with time, geographical coordinates, and equipment numbers, automatically mapping task lists to unique equipment identifiers to achieve data collection and task tracking across multiple objects and scenarios.

[0077] When maintenance personnel proactively report anomalies, describe equipment status, or record inspection comments via voice, the AI ​​will transcribe the voice content in real time and intelligently correlate it with currently collected images, videos, and sensor anomaly logs to form a structured record of anomaly events. If the system's multimodal anomaly linkage detection module automatically determines that the current equipment shows signs of risk, it will also immediately capture and aggregate the current multimodal data, supplemented by automatically generated inspection voice prompts, to remind maintenance personnel to focus on supplementing detailed data collection or providing further explanations.

[0078] All raw data is stored locally with encryption and synchronized to a unified cloud-based operations and maintenance platform periodically or in real-time, depending on the on-site network conditions. During data transmission and storage, the system incorporates a hierarchical access control and traceability management mechanism to ensure data integrity, confidentiality, and traceability, providing comprehensive and reliable data support for subsequent comprehensive analysis, knowledge iteration, expert review, and post-event accountability.

[0079] This multimodal, real-time, and high-precision inspection process information collection mechanism not only comprehensively improves the inspection quality and data accuracy of high-risk, heavy-load equipment under strong electric field environments, but also lays a solid foundation for AI-assisted anomaly detection, intelligent Q&A guidance, and full life-cycle equipment management, greatly enhancing the standardization, closed-loop nature, and intelligence of maintenance work.

[0080] Step 4: Intelligent Anomaly Detection and Real-time Risk Alerts

[0081] In the intelligent anomaly detection and real-time risk alert phase, this method is comprehensively applicable to the operation and maintenance of various key equipment in multi-physical ultra-high voltage converter stations under strong electric field environments. The AI ​​recorder inputs real-time audio, video, image, and various environmental sensor data collected on-site into the anomaly detection algorithm. Combined with the operational characteristics of different systems such as air-cooled, water-cooled, converter valves, busbars, and auxiliary equipment, multi-modal anomaly identification is performed. For noise and vibration signals from equipment such as fans, motors, valves, and transformers, the system uses Short-Time Fourier Transform (STFT) to obtain real-time acoustic spectrograms. and historical health samples Compare and calculate the spectral anomaly:

[0082]

[0083] when Exceeding the empirical threshold At that time, the system determines that there is an acoustic or mechanical anomaly in the detection area.

[0084] For multi-dimensional sensor sequence data such as temperature, humidity, current, and partial discharge signals The moving average and standard deviation are used to detect sudden anomalies or drift. Specifically:

[0085]

[0086] It can trigger abnormal alarms such as temperature exceeding limits, and the same applies to humidity, current and other physical quantities unique to high-voltage equipment.

[0087] like Figure 10 As shown, in the image and video monitoring stage, a deep convolutional neural network (ResNet) is used to perform intelligent visual anomaly detection on key areas such as equipment appearance, electrical connection points, and insulation components. The current image is encoded as a feature vector. and the known normal sample center Calculate the Euclidean distance:

[0088]

[0089] like Exceeding the threshold It will automatically detect visual abnormalities such as corrosion, damage, discharge, water leakage, and insulation deterioration.

[0090] The innovation of this method lies in the linkage analysis and dynamic response mechanism of multimodal anomaly signals. When any signal source (such as abnormal noise) shows suspicious activity, the system immediately increases the data sampling frequency and analysis sensitivity of relevant video, temperature, vibration, and other modalities. For example, if high-frequency noise anomaly is detected, the system automatically increases the camera resolution, guiding maintenance personnel to focus on specific details, add supplementary shots, and prompts the system via voice assistant to collect more operational environment information. The weighted fusion of multimodal anomaly scores is as follows:

[0091]

[0092] When the total score Exceeding the comprehensive risk threshold In such cases, the AI ​​recorder immediately alerts maintenance personnel to abnormal risks through voice broadcasts, interface highlighting, directional navigation, or vibration, and guides them to supplement high-quality detailed data (images, audio, and text descriptions), achieving comprehensive, traceable recording of major anomalies and providing support for security control.

[0093] Step 5: AI-assisted Q&A and operation guidance

[0094] In the AI-assisted question-and-answer and operation guidance section, this system targets the station-wide operation and maintenance needs of multi-physical ultra-high voltage converter stations. It fully integrates the multi-modal detection results, real-time sensing data and historical operation records from the aforementioned "intelligent anomaly capture and real-time risk warning" section to provide intelligent and personalized support for all key equipment (such as converter valves, busbars, cooling systems, transformers and relay protection).

[0095] In the AI-assisted question-and-answer and operation guidance process, a localized speech recognition model is used to transcribe the maintenance personnel's on-site questions (such as "How to troubleshoot water pump flow fluctuations") into text. The transcription results will be combined with the current inspection context of the system—including anomaly detection scores, real-time collected data, equipment operating status, and historical inspection logs—to form a comprehensive scenario description vector. In addition, the system possesses a structured knowledge graph of cooling system experts. , where nodes Represents entities such as equipment, fault type, diagnostic method, and operating procedure, with edges It indicates the relationship between entities.

[0096] The core of personalized guidance is the context fusion reasoning algorithm. First, the system will process the question text... Scene data vector It is jointly encoded with knowledge graphs. A multimodal embedding approach is used to integrate... and Input a context-aware encoder (a self-attention network fusing BERT and temporal data), output a representation vector. At the same time, for knowledge graph entity sets All candidate node embeddings are obtained by relying on graph convolutional networks (GCNs) or knowledge embedding models (such as TransE). .

[0097] System calculation problem - knowledge matching score:

[0098]

[0099] in This is either cosine similarity or a weighted inner product. Scores above a threshold are selected. The algorithm uses a set of nodes to obtain the most relevant operational steps, historical cases, or handling suggestions related to the current problem and equipment status. To improve the real-time performance and relevance of the reasoning, the algorithm also considers the current anomaly detection probability distribution in the filtered results. Weighted sorting:

[0100]

[0101] The top k highest-scoring results are generated into natural language instructions, which are then synthesized via TTS and broadcast to the maintenance personnel. If the problem closely matches the on-site anomaly, and abnormal vibration is detected and the question is strongly correlated with "water pump flow fluctuations," the system will automatically supplement the image acquisition process, add additional instructions ("It is recommended to check the inlet, filter, and flow meter in sequence"), and retrieve historical case videos and documents to assist in troubleshooting.

[0102] Voice prompts include: "It is recommended to check if the inlet filter is clogged, confirm the pump body vibration status, and take additional photos of the pump body nameplate and control panel." All interactive content, along with anomaly detection results, equipment data, and supplementary on-site instructions, are archived together to provide a complete clue for anomaly tracing and subsequent operation and maintenance.

[0103] This AI question-and-answer mechanism ensures the dynamic integration of context and knowledge experience from the actual inspection site, providing maintenance personnel with personalized, timely, and highly relevant support, greatly improving the efficiency of front-line operation and maintenance response and fault location.

[0104] Step 6: Secure Data Synchronization and Process Archiving

[0105] All data during the inspection process (including abnormal segments, key images, and operation audio) is automatically categorized, encrypted, and then synchronized to the secure cloud platform. The process employs industry-standard data transmission protocols, such as GB28181 or an intranet secure tunnel, to ensure the security and compliance of sensitive maintenance data.

[0106] The system automatically generates structured inspection logs, anomaly summaries, and multimedia reports, which are then archived in both local and cloud-based intelligent knowledge bases. For new types of faults and frequently occurring problems, the cloud platform automatically compiles and pushes handling suggestions to similar sites, enabling efficient experience sharing and collaborative team growth.

[0107] In the data security synchronization and process archiving stages, the AI ​​recorder provides unified management and high-level security protection for all data collected during the inspection process. The system first categorizes and organizes the raw data, automatically tagging and classifying anomaly detection segments, key images, maintenance personnel operation audio, on-site questions, and system responses. For each type of data, the platform uses industry-recognized encryption algorithms for local encryption to prevent unauthorized access.

[0108] During data synchronization, based on user permissions and enterprise security policies, the system employs the GB28181 video transmission standard or an SSL / TLS-based intranet security tunnel to ensure end-to-end static and dynamic encryption of data throughout the transmission link, reducing the risk of data leakage and tampering. For sensitive operational data or information with privacy markings, the system can also enable multi-authorization and access auditing mechanisms to ensure compliance.

[0109] The synchronized data is automatically organized into structured inspection logs. The system automatically generates event indexes and summaries based on the collection time, location, inspection object, and anomaly type, thus forming electronic archives that can be quickly retrieved and traced. In addition to basic text logs, the AI ​​system also automatically generates multimedia reports, integrating key images, audio clips, detection curves, and processing suggestions into a unified format for management, training, or subsequent traceability. These archived contents are not only stored on local terminals for emergency offline access but are also synchronized to a cloud-based intelligent knowledge base, supporting large-scale data aggregation and intelligent analysis.

[0110] For newly detected fault modes and frequently occurring issues, the cloud platform possesses automatic analysis and knowledge organization capabilities, enabling it to promptly compile optimal handling procedures, risk warnings, and operational recommendations. Through an intelligent distribution mechanism, the system proactively pushes these new experiences and typical cases to similar sites and relevant team members' terminals, helping enterprises quickly achieve experience sharing, knowledge updates, and collaborative improvement of team capabilities, ultimately promoting the continuous progress and intelligent development of the entire operations and maintenance system.

[0111] Example 2:

[0112] like Figure 2 As shown, an AI recorder for safety management and control of operation and maintenance of ultra-high voltage converter stations with multiple physical forms that are adapted to strong electric field environments is described. The AI ​​recorder is applicable to the method described above. The AI ​​recorder includes: a camera 2, a control module 4, a power module 1, a display screen 5, and a buckle 6.

[0113] The control module 4 includes: a wireless transmission module 41, a positioning module 42, a voice module 43, and a processing module 44;

[0114] The camera 2, positioning module 42, voice module 43, and display screen 5 are all electrically connected to the processing module 44. The output of the processing module 44 is electrically connected to the wireless transmission module 41, and the wireless transmission module is connected to the background database.

[0115] The power module 1 provides power to the recorder, which is worn via a buckle.

[0116] like Figure 5 As shown, a name tag-style AI recorder is used to pair with work clothes via a buckle 6, and the outer shell is shaped like a name tag.

[0117] like Figure 3 and Figure 4 As shown, a head-mounted AI recorder is provided, the head-mounted AI recorder being based on the AI ​​recorder, and the head-mounted AI recorder further includes;

[0118] The lighting module 3 uses LED lighting beads and is connected to the power module 1. The recorder is connected to the helmet via the buckle 6.

[0119] The AI ​​recorder has a semi-circular arc-shaped shell, which conforms to the shape of a helmet edge. The camera and lighting module are located at opposite ends of the shell, so they do not interfere with each other's operation.

[0120] like Figure 6 and Figure 7 As shown, a seat belt-type AI recorder is provided. The seat belt-type AI recorder is based on the AI ​​recorder and is used in conjunction with a seat belt. The seat belt-type AI recorder is connected to the seat belt by a buckle.

[0121] Headset-mounted and badge-style AI recorders securely connect to the intelligent access gateway via VPN and TF encryption cards, ensuring the compliance and security of data interaction during the maintenance and inspection of the converter station cooling system. On-site audio and video data are accessed to a unified video platform using the GB28181 national standard protocol or a power industry-specific B interface, enabling centralized storage, real-time access, and intelligent analysis of audio and video data throughout the entire cooling system operation and maintenance process. In terms of business integration, the intelligent inquiry module connects to the intelligent inquiry data microservice via API, ensuring real-time linkage between on-site maintenance inquiries and business data such as cooling system assets and operating status. Based on the converter station operation and maintenance management system, a closed-loop business model of "risk alarm - result subscription - algorithm upload" is constructed—realizing real-time push of cooling equipment alarm information, on-demand subscription of AI analysis results, synchronous upload of algorithm detection results, and deep integration with the converter station SCADA system and cooling system substations.

[0122] The local model on the terminal side is continuously trained and optimized through field samples, and is primarily applied to two key scenarios: First, the identification of safety attire and compliance of work procedures for maintenance personnel. Images of maintenance personnel are collected in real time on-site, and the intelligent AI model automatically identifies the wearing status of personal protective equipment such as safety helmets, insulating gloves, and protective shoes. If compliance is not found, a local audible and visual warning is automatically triggered, and the violation information is simultaneously sent to the backend to improve work safety. Second, the identification of defects in the cooling system inspection, such as abnormal cooling pumps, cooling tower icing, valve corrosion, pipeline leaks, water pump leaks, cooling fan failures, and equipment nameplate detachment. Through image recognition and multimodal data analysis, it assists personnel in proactively discovering and labeling various defects. The identification results are labeled with key tags to facilitate subsequent closed-loop management and accountability. The terminal is equipped with multiple types of sensors and customized algorithms to achieve safety monitoring of cooling station maintenance operations: such as using infrared and attitude algorithms to detect helmet detachment, using electric field strength and temperature and humidity monitoring to help determine abnormal working environment of cooling equipment, using acceleration sensors to detect emergency situations such as falls or knocks, and using sensors such as air pressure, smoke, and hydrogen to provide safety warnings for special working conditions. In case of emergencies, it automatically contacts emergency contacts and uploads the site location and images, while providing voice reminders of standard operations and risk points. The on-site operation status is automatically synchronized to the intelligent platform.

[0123] In terms of backend model application, firstly, it relies on intelligent question-and-answer scenario microservices to build an integrated maintenance question-and-answer function, eliminating the need to repeatedly build a question-and-answer engine and implementing a closed-loop "question-answer-action" application, making it convenient for maintenance personnel to quickly obtain knowledge services such as standard operation of the cooling system, fault handling, and historical cases on-site; secondly, it calls on mature algorithms of the cloud artificial intelligence platform (including real-time behavior analysis of cooling equipment, equipment defect detection, audio and video structure extraction, etc.) and continuously adapts to changes in on-site business through the platform's iteration mechanism, achieving efficient technology upgrades and reuse; thirdly, based on speech-to-text conversion technology, it builds an intelligent recognition system for sensitive words and safety terms, which analyzes the on-site audio and video streams of maintenance personnel in real time, enabling multi-dimensional analysis such as safety announcements, distress calls, new specification reminders, and emotion recognition. The platform can reply to voice content in a friendly style, relieving the psychological pressure of maintenance and improving the working environment.

[0124] At the business integration level, the intelligent inquiry module connects to the intelligent inquiry data microservice via API to achieve comprehensive linkage between the maintenance site and the core business data of the cooling system; risk alarms, result subscriptions and algorithm uploads for cooling equipment support real-time push notifications of abnormal operation of equipment such as cooling water pumps, fans, and valves, analysis results are categorized and subscribed, algorithm recognition results are synchronously transmitted back, and the unified video platform and the converter station cooling system automation subsystem achieve deep integration of data and business.

[0125] like Figure 8 and Figure 9As shown, the intelligent maintenance method and system of the present invention, based on a badge-type or head-mounted AI recorder, adopts the following specific implementation steps:

[0126] 1. Hardware preparation and allocation

[0127] Maintenance personnel uniformly receive and activate badge-style or head-mounted AI recorders. These devices integrate high-sensitivity microphones, high-definition cameras, displays, data communication modules, AI chips, and various types of sensors, meeting the requirements of a three-tiered "cloud-edge-device" collaborative architecture. By binding personnel identification and maintenance task orders to the recorders, data access management and traceability are achieved. The devices support high-definition recording (1080P), wide-angle lenses, electronic image stabilization, 256GB storage expansion, dual-battery operation, and a comfortable, seamless wearing design for on-site work. Full 4G network compatibility and satellite positioning ensure data interconnectivity and dynamic personnel monitoring.

[0128] 2. Intelligent collection of on-site information

[0129] During maintenance, personnel activate the recorder via voice or button to capture all audio, video, image, and voice data in real time. The equipment automatically formats and stores the multimodal data from the cooling system maintenance process, supporting structured tag input for information such as equipment type and fault category, greatly improving the standardization, completeness, and usability of maintenance data.

[0130] 3. AI-assisted troubleshooting and operation Q&A

[0131] When encountering difficult or special equipment during maintenance, the AI ​​assistant can be consulted on-site via voice to obtain information on the cooling system's principles, specifications, operating procedures, or past maintenance cases. This provides immediate structured answers or documents, allowing for quick identification of problems and steps, and improving on-site handling efficiency.

[0132] 4. Real-time data upload and cloud synchronization

[0133] All field data is automatically uploaded to the intelligent operation and maintenance cloud platform after encryption, ensuring data security and real-time synchronization. The cloud side archives, retrieves, and intelligently analyzes the data, automatically generating standardized maintenance logs, statistical reports, and experience cases, and continuously improving the cooling system maintenance knowledge base.

[0134] 5. Centralized Management and Collaborative Operation and Maintenance

[0135] The intelligent platform centrally manages all on-site maintenance and AI Q&A results for the cooling system, supports multi-dimensional access control, historical work review, and sharing of typical cases, and helps professional teams collaborate and pass on knowledge between new and old employees, thereby improving the overall capabilities of the cooling system maintenance team.

[0136] 6. Security and Privacy Protection Mechanisms

[0137] The system has strict personnel approval and sensitive data anonymization processes, and adopts end-to-end encryption. All collected, transmitted, and stored data strictly comply with the power industry's information security standards, fully protecting the privacy and security of users and core production data.

[0138] Example 3:

[0139] A certain 500kV ultra-high voltage DC converter station has a dense concentration of electrical equipment and strong environmental electromagnetic interference, making operation and maintenance (O&M) operations challenging. To improve the intelligence level and safety management capabilities of O&M operations, the converter station has fully deployed the multi-physical-form ultra-high voltage converter station O&M safety management AI system described in this invention. The system hardware consists of head-mounted and badge-style AI recorders, environmental multi-sensor modules, and a safety cloud platform. The software covers functions such as AI multimodal recognition, anomaly analysis, and real-time interactive guidance.

[0140] 1. Preparations before the inspection

[0141] The maintenance team (3 people in total) checked in at the duty room, received and put on the name tag-style AI recorder, and the device was automatically bound to their identity information via Bluetooth.

[0142] The recorder automatically initializes, loading the day's inspection tasks, equipment list, and operational risk warnings. Staff check the AI ​​recorder (battery ≥95%, storage space ≥80%, network signal normal). After the system self-check passes, it automatically pushes operational safety notices to the maintenance personnel's interface.

[0143] The maintenance worker wore a safety helmet, and the AI ​​recorder was secured to the helmet with a buckle. The lighting function tested normally, and the voice interaction wake-up word test passed.

[0144] 2. Entry into the work area and equipment self-identification

[0145] When the maintenance personnel arrive at the valve hall, the AI ​​recorder activates multi-sensor data collection to detect ambient noise, electromagnetic field strength, and air temperature and humidity, and automatically identifies the type of the current work area (valve hall / cooling system / DC field / AC field, etc.).

[0146] By quickly scanning the equipment nameplate with a camera and combining environmental parameters with the equipment knowledge base, the system identifies the key equipment that needs to be inspected (such as converter transformers, circuit breakers, cooling pump sets, etc.) and highlights them on the maintenance personnel's interface to remind them to check them first.

[0147] Table 1: Area and Device Identification Data

[0148]

[0149] 3. On-site inspection and multimodal information collection

[0150] Guided by the AI ​​system, maintenance personnel inspected the equipment one by one along the optimal route. During the inspection, the AI ​​recorder automatically collected audio, images, video, and environmental data, recording the entire operation process.

[0151] If a sudden equipment malfunction is detected (such as a sudden change in the sound of the cooling pump), the maintenance worker can make a voice note saying "abnormal noise from the cooling pump." The AI ​​system will automatically capture the audio and video clips from that time period and mark the fault information.

[0152] When the system detects that a maintenance worker is approaching a high-voltage energized area and the ambient electromagnetic intensity exceeds the threshold, the AI ​​recorder immediately issues a voice and vibration alert to prompt the maintenance worker to take evasive action.

[0153] Table 2: Multimodal Data Acquisition

[0154]

[0155] 4. Intelligent anomaly detection and risk alerts

[0156] The built-in anomaly detection algorithm compares the currently captured audio and video with standard features. For example, if coolant pump A detects mechanical noise of 92dB, which exceeds the normal range (70~85dB), it will immediately issue a "Suspected malfunction of coolant pump A" warning.

[0157] Analysis revealed that the noise spectrum of cooling pump A was different from that of normal equipment. Combined with a slight increase in temperature (from 19.2℃ to 22.9℃), the system simultaneously issued a risk alarm and recommended that the maintenance personnel perform a shutdown inspection of the cooling pump.

[0158] Table 3: Anomaly Detection and Risk Warning Data

[0159]

[0160] 5. AI-assisted Q&A and operation guidance

[0161] When the maintenance worker encountered an abnormal noise from coolant pump A, he asked via voice prompts: "How do I troubleshoot and resolve the abnormal noise from coolant pump A?"

[0162] The AI ​​recorder, through contextual fusion analysis and combining historical device data and a case knowledge base, provides the following guidance:

[0163] (1) Stop the machine and disconnect the power, and check whether the mechanical parts of the pump body are loose;

[0164] (2) Check the lubricating oil level;

[0165] (3) Review the cooling pump operation log of the previous month and focus on replaying the previous similar alarm case (system push video V2024051512).

[0166] Following the instructions, the maintenance personnel quickly completed mechanical tightening and lubrication replenishment, provided real-time voice reports on the processing, and the AI ​​recorder simultaneously recorded video.

[0167] 6. Secure data synchronization and archiving

[0168] After the operation is completed, the AI ​​recorder encrypts and packages all audio, video, images, and environmental sensor data collected on site, and automatically synchronizes them to the secure cloud platform.

[0169] When the data is archived, it is automatically categorized as "2024-06-01 Inspection Log" and indexed according to categories such as equipment, anomalies, and maintenance procedures to support subsequent traceability.

[0170] The entire process of audio and video recordings and operational data is only accessible to the operations supervisor and safety inspector. The system automatically generates inspection reports and risk alarms, which are then pushed to the station manager and technical lead.

[0171] Table 4: Data Archiving and Access Control

[0172]

[0173] III. Implementation Advantages and Effects

[0174] Maintenance efficiency: The entire operation is automated, reducing maintenance time by about 30% and improving anomaly response speed by 50%.

[0175] Safety risk incident rate: Through AI early warning and on-site reminders, incidents of accidental equipment contact and misoperation have decreased significantly, and the number of annual safety accidents has decreased by 70%.

[0176] This embodiment fully demonstrates the application of the present invention in intelligent operation and maintenance in the complex and high-risk environment of ultra-high voltage converter stations. Through multi-sensor fusion, anomaly detection, intelligent guidance, and secure data archiving of the AI ​​recorder, the efficiency and safety level of on-site maintenance are greatly improved, providing solid support for the digital and intelligent upgrading of ultra-high voltage converter station operation and maintenance.

[0177] Example 4:

[0178] I. Implementation Background

[0179] A ±800kV DC converter station is equipped with multiple cooling systems, including both air-cooled and water-cooled systems, to provide reliable heat dissipation for converter valves, transformers, and other critical equipment. With long-term operation, the cooling systems are prone to various failures due to environmental pollution, equipment aging, pipeline leaks, and fan noise. Currently, under traditional operation and maintenance methods, cooling system anomalies can only be initially diagnosed visually and audibly, data acquisition lacks multi-dimensional integration, the maintenance process is simplistic, and there are numerous safety hazards.

[0180] To overcome the above challenges, this embodiment employs the AI ​​multi-physical intelligent recorder and safety management system of the present invention to intelligently manage the entire inspection process of air-cooled and water-cooled systems, enabling multi-modal acquisition of equipment status, automatic anomaly identification, intelligent assisted decision-making, and data traceability.

[0181] II. System Configuration and Maintenance Preparation

[0182] 1. Smart Device and Platform Configuration

[0183] AI smart recorder: available in head-mounted and chest-mounted versions, integrating a high-definition camera, environmental acoustic sensor, and multi-dimensional sensing modules for temperature, humidity, gas, vibration, flow, and current.

[0184] Data synchronization platform: Supports preliminary local analysis at the edge and in-depth diagnosis in the cloud, and has a historical case library and knowledge retrieval interface.

[0185] Inspection team: All three maintenance personnel have undergone AI-assisted operation and maintenance training.

[0186] 2. Preparations before inspection

[0187] The maintenance personnel confirmed the task and equipment location at the control center, and then received and checked the AI ​​recorder (battery level, cleanliness, and operation indicator lights were normal).

[0188] The work plan is automatically distributed, specifying the inspection route for the cooling system and key risk-prone equipment.

[0189] The system pushes the daily environmental monitoring report (including abnormal alerts for parameters such as wind and sand, temperature, humidity, and water quality).

[0190] III. Intelligent Inspection and Anomaly Detection Process of Cooling System

[0191] Step 1: System Self-Identification and Multimodal Acquisition

[0192] When the maintenance personnel enter the air-cooled unit area of ​​the main valve hall, the AI ​​recorder automatically senses the current environmental parameters, activates the air-cooled equipment inspection mode, and automatically locates the numbers and status of the main fans, air filters, and motor drive cabinets. When inspecting the water-cooled pump room, the system automatically switches to water-cooling mode and locates the locations of each water pump, cooling tower, filter, pressure gauge, etc.

[0193]

[0194] Step Two: Initial Anomaly Assessment and Risk Warning

[0195] The AI ​​recorder automatically detected that the noise level of fan A1-03 in the air-cooled area reached 92dB, which was significantly higher than the normal operating average (75~85dB). The vibration sensor also showed a reading of 3.4mm / s (significantly higher than the 1.8mm / s of other similar fans). Simultaneously, the video footage recorded occasional oscillations in the fan drive shaft. The system immediately issued a triple alert with vibration, voice, and image notifications, and displayed a message stating, "Fan A1-03 is suspected of bearing or balance abnormalities; please conduct a thorough inspection."

[0196] In the water-cooled pump area, the water pressure reading of the main pump B-02 dropped to 0.32 MPa, lower than the rated 0.45~0.55 MPa, accompanied by a slight increase in pump noise. A comparison with previous cases indicated that "the water circuit of the main pump B-02 may be blocked, leaking, or the filter element may be contaminated; it is recommended to check the inlet pipe and pressure gauge."

[0197]

[0198] Step 3: AI-assisted diagnosis and operation guidance

[0199] After receiving the abnormality alert, the maintenance technician used the voice command "Wind turbine A1-03 malfunction, how to quickly troubleshoot?"

[0200] The AI ​​recorder automatically pushes step-by-step troubleshooting suggestions (simultaneously displaying past case images and videos):

[0201] 1. Stop the operation of fan A1-03. After the movement stops, manually rotate the fan disc to test the bearing damping and record a video for uploading.

[0202] 2. Compare the wind turbine monitoring data and retrieve the wind turbine vibration trend chart of A1-03 for the past month.

[0203] 3. Check the temperature rise of the drive motor terminals and measure the actual temperature with an infrared thermometer.

[0204] 4. If the bearing is abnormally worn, replace it or add grease according to the standard, and fill in the maintenance record.

[0205] In the water-cooled area, the maintenance worker asked, "The water pressure in main pump B-02 is low. How should we handle this?"

[0206] The AI ​​system determined that the pump noise was slightly increased and there were no obvious abnormalities in the high-frequency vibration, inferring that it might be due to blockage or leakage, and then pushed process guidance:

[0207] 1. Turn off the main pump B-02 and slowly drain the inlet and outlet water pipes.

[0208] 2. The inlet filter element was disassembled and inspected, and photos were taken on site. It was found that there was obvious sediment deposits on the filter element.

[0209] 3. Clean the filter element with non-woven cloth, check the pipeline seal, and confirm that there is no leakage.

[0210] 4. After resetting, the water pressure was measured to be restored to 0.47MPa and the noise was reduced to 81dB. The system automatically determined that the fault was eliminated.

[0211]

[0212] Step 4: AI voice and gesture interaction, information recording

[0213] At critical junctures (such as replacing bearings or disassembling and assembling pump bodies), maintenance personnel will issue a voice announcement stating, "Bearing replacement of fan A1-03 is complete; start trial operation."

[0214] The system automatically records relevant audio and video and archives them synchronously with maintenance operations. If an operation requires confirmation from two people, the system automatically identifies the personnel on site and asks both to simultaneously make an OK gesture for confirmation, and then takes a photo for record-keeping.

[0215] If an operational risk is encountered, the AI ​​will automatically remind you, "No grounding wire is currently connected. Please pay attention to safety!" After the maintenance personnel complete the necessary safety measures, the system will provide a voice feedback, "Grounding confirmed," and the system will record the information.

[0216] Step 5: Secure Archiving and Analysis of Work Data

[0217] After maintenance, the AI ​​recorder automatically encrypts and summarizes the following data and uploads it to the cloud platform:

[0218] Full-process high-definition video (slow-motion replay of key operations such as bearing replacement and filter cleaning)

[0219] Multi-point environmental acoustic, vibration, temperature, water pressure and other parameter curves

[0220] On-site voice / gesture confirmation record

[0221] The causes of the faults and the repair conclusions are automatically added to the case knowledge base.

[0222] Abnormal data from wind turbine A1-03 and main pump B-02 are automatically labeled for use in subsequent AI training and experience backtesting.

[0223]

[0224] IV. Implementation Results and Advantages

[0225] Inspection efficiency: By adopting AI intelligent guidance and multimodal data integration, the time required for a single complete cooling system maintenance process is reduced by approximately 40% compared to traditional methods;

[0226] Abnormal response speed: Abnormalities in key components such as fans and water pumps are detected and intelligently classified in real time, with a response misjudgment rate of less than 1% and a major fault early warning rate of 96%;

[0227] Maintenance traceability: The entire process of audio and video recordings and gesture confirmation data ensures that the operation is compliant and traceable, greatly improving safety management and subsequent review capabilities;

[0228] User experience: AI voice Q&A accurately pushes operation steps, reducing reliance on personal experience and significantly improving the success rate of novice repairs.

[0229] This embodiment demonstrates the powerful capabilities of the present invention in intelligent maintenance of converter station cooling systems. The AI ​​recorder collects, integrates, judges, and provides intelligent operation guidance for multi-physical information from both air-cooled and water-cooled systems in real time, greatly improving equipment safety and personnel work efficiency, and achieving a new breakthrough in intelligent maintenance and management of converter station cooling systems under high-risk and complex environments.

[0230] Example 5:

[0231] I. Background

[0232] A ±800kV converter station contains main equipment including converter transformers, smoothing reactors, and GIS switchgear. The converter transformers are large and complex, operating under high voltage, high current, and strong electromagnetic environments for extended periods. Some maintenance work involves working at heights (such as inspecting pressure relief valves on the top of the oil tank and bushing maintenance) and in confined spaces (such as inspecting the inside of the transformer), posing high operational risks. Traditional inspection methods rely on manual visual inspection and paper records, which are not only inefficient but also prone to overlooking safety precautions and creating potential safety hazards when working close to energized conditions.

[0233] To address these issues, this embodiment employs the AI ​​seatbelt recorder system of the present invention to achieve comprehensive intelligent safety management and control of electrical equipment such as transformers in high-risk scenarios.

[0234] II. System Configuration and Operation Preparation

[0235] 1. Smart Devices and Configurations

[0236] AI-powered safety belt recorder: integrates a high-definition camera, infrared temperature measurement, microphone, vibration sensing, tilt angle (to prevent accidental operation due to falls from heights), posture recognition module, RFID positioning module, etc.; worn on the waist / shoulder of the maintenance personnel, freeing up their hands, and recording angles covering the entire maintenance process.

[0237] Intelligent cloud platform: Connects to recorder data and seamlessly links with station control center, work ticket system and equipment operation and maintenance big data database.

[0238] Safety auxiliary kit: safety belt for working at heights, insulated gloves, portable grounding wire, and emergency rescue positioning beacon.

[0239] 2. Homework Preparation

[0240] Each maintenance crew (4 people per group) must complete the daily safety and technical briefing and sign the digital work ticket.

[0241] Personnel wearing safety belts and AI recorders automatically read maintenance certificates, health self-assessments for the past three days, and safety training qualification information to enter the identity verification process.

[0242] The recorder's functions (camera, communication, angle sensing, infrared temperature measurement, emergency call) all passed self-tests, and the battery life was greater than 90%.

[0243] The AI ​​recorder automatically pushes the transformer number to be inspected, historical faults, key areas of concern, and EHS safety reminders (such as "fall prevention for working at heights", "maintaining live gaps", and "secondary grounding measures") according to the maintenance plan for the day.

[0244] III. Transformer Intelligent Inspection and Anomaly Detection Process

[0245] Step 1: Entering the work area and multimodal data acquisition

[0246] When maintenance personnel enter the converter transformer area, the safety belt AI recorder automatically senses environmental parameters (noise, electromagnetic intensity, ozone gas concentration, vibration, temperature, and maintenance personnel posture) and simultaneously records on-site video, audio, and work identity.

[0247]

[0248] Step Two: Safety Identification and Anomaly Warning in High-Altitude / Confined Spaces

[0249] The maintenance worker climbed to the top of the transformer tank to replace the pressure relief valve. The AI ​​recorder detected that the worker was wearing a safety belt and that the anchor point was compliant (RFID identification). The system then switched to high-altitude mode and displayed a pop-up message: "High-altitude operation has been detected, and fall protection measures are normal."

[0250] If the safety belt anchor point comes loose or an abnormal bending / falling posture is detected during the operation, the recorder will immediately issue a level 3 voice alarm and automatically upload the alarm recording.

[0251] Some maintenance workers entered the narrow space of the high-voltage terminal box. The AI ​​recorder issued a "Gas exceeding the standard in confined space operation, pay attention to ventilation" prompt based on posture sensing and ambient gas levels exceeding the warning value (ozone 0.19ppm, slightly higher than the 0.15ppm threshold).

[0252]

[0253] Step 3: Multimodal perception and anomaly detection during the operation process

[0254] Infrared temperature measurement inspection

[0255] The AI ​​recorder collects the surface temperature of different parts of the transformer bushing, leads, and tank in real time, and automatically compares it with the normal operating temperature rise curve.

[0256] The temperature measured at the top of phase A of the bushing is 38.2℃. The system identifies this as 2.5℃ higher than the historical value for the same period, marks it as "slightly abnormal", and sends a maintenance suggestion.

[0257] If there is no abnormality in the temperature of the middle part of the oil tank, the maintenance personnel can use the voice command "Analysis of the cause of temperature rise in phase A of the bushing" and the system will push similar past cases for judgment.

[0258] Vibration detection

[0259] During the inspection, the vibration sensor of the seat belt AI recorder recorded a peak vibration value of 1.2 mm / s on the surface of the fuel tank, which is lower than the limit of 2.0 mm / s, and there was no abnormal alarm.

[0260] Audio and video data synchronization

[0261] Key operations (such as draining oil, disassembling and assembling pressure relief valves, and sampling insulating oil) are recorded simultaneously in high definition, and AI automatically analyzes and tags key voice commands (such as "work completed" and "confirmed in place").

[0262]

[0263] Step 4: AI-powered intelligent operation guidance and decision support

[0264] During the operation, the maintenance worker asked via voice: "What are the precautions for replacing the pressure relief valve?"

[0265] AI recorder provides step-by-step guidance based on the on-site situation:

[0266] 1. Confirm that the corresponding busbar has been de-energized and secondary grounding has been completed (match the work order with the actual electrical isolation status).

[0267] 2. Avoid directly striking the valve body with hard tools during disassembly.

[0268] 3. The disassembly and installation process must be operated by two people. The system will automatically identify the identities and collaborative actions of the two maintenance personnel on site.

[0269] 4. Ensure safety belts are secure throughout the entire operation and report any emergencies immediately.

[0270] Maintenance personnel record audio and video of key processes, and the system pops up a reminder to record additional key parameters.

[0271] Example: "Please take photos of the nameplate and seal after the new valve is installed."

[0272]

[0273] Step 5: Automated Emergency Response and Secure Data Archiving

[0274] If any abnormality occurs during the maintenance process (fall, fainting, accidental contact with live parts), the seat belt AI recorder will immediately trigger an emergency alarm through posture recognition and heart rate / accelerometer sensing, and automatically upload key images to the control room and rescue personnel's terminals.

[0275] After the maintenance is completed, the AI ​​recorder will categorize and encrypt the audio and video, images, sensor data, work process and operation records of the entire inspection process and upload them to the cloud platform, automatically classifying them into the "2024-06-01 Transformer A Team Maintenance Log".

[0276] Maintenance cases are automatically integrated with the historical case database, facilitating subsequent intelligent retrieval, experience review, and training.

[0277]

[0278] IV. Implementation Results and Effects

[0279] Safety precautions: The AI ​​recorder provides real-time intelligent identification of high-altitude / confined spaces, achieves 100% implementation rate of safety measures, and automatically alarms for violations such as not wearing a safety belt, preventing "defective" operations.

[0280] Anomaly warning rate: Multimodal data accurately identifies early risks such as temperature rise and excessive gas levels in key components, increasing the anomaly warning rate to 98% and greatly reducing the probability of major failures.

[0281] Efficiency and compliance: Intelligent guidance throughout the process, automatic detection of key operations and reminders for supplementary recording reduce maintenance time by 35% and increase the operational compliance rate to 100%.

[0282] Data traceability: Multimodal and full-scenario data archiving throughout the maintenance process, allowing for rapid traceability and sharing of accidents / experiences, ensuring the safety of converter station operation and maintenance and the inheritance of technology.

[0283] This embodiment powerfully demonstrates the advantages of the seatbelt AI recorder system in intelligent safety operation and maintenance in complex and high-risk scenarios involving electrical equipment such as transformers in converter stations. Through multimodal perception, intelligent identification of high-risk operations, automatic anomaly analysis, and full-process operation tracking, it significantly improves the safety, standardization, and data value of operation and maintenance, providing strong support for the intelligent and digital safety management of converter stations.

[0284] Those skilled in the art will understand that all or part of the processes in the above embodiments can be implemented by a computer program instructing related hardware. The program can be stored in a computer-readable storage medium, and when executed, it can include the processes of the embodiments of the above methods. The storage medium can be a magnetic disk, optical disk, read-only memory (ROM), or random access memory (RAM), etc.

[0285] It should be understood that the above detailed description of the technical solutions of the present invention with reference to preferred embodiments is illustrative and not restrictive. Those skilled in the art can modify the technical solutions described in the embodiments or make equivalent substitutions for some of the technical features based on reading this specification; however, these modifications or substitutions do not cause the essence of the corresponding technical solutions to depart from the spirit and scope of the technical solutions of the embodiments of the present invention.

Claims

1. A method for safe operation and maintenance management of multi-physical-form ultra-high voltage converter stations adapted to strong electric field environments, characterized in that: The method includes: Before the inspection, maintenance personnel enter the equipment area of ​​the converter station, wear a badge-style or head-mounted AI recorder, and automatically synchronize the current inspection task, the list of key inspection objects, and precaution information to the maintenance personnel interface. The converter station equipment type and inspection object are automatically identified. The AI ​​recorder combines noise spectrum, environmental electromagnetic intensity, temperature and humidity data from multiple sensors to automatically determine whether the current work area is the cooling system, electrical equipment, power lines, or other equipment, and automatically matches the core equipment to be inspected based on the on-site scan. On-site inspection and process information collection: maintenance personnel check each pipeline step by step according to the inspection route under guidance. The AI ​​recorder automatically collects and stores multimodal data such as audio, pictures and videos of the inspection process. When an abnormality is found, it automatically associates fault photos / video clips through working voice notes to improve the accuracy and completeness of the record. Intelligent anomaly detection and real-time risk alerts input audio, video, image, and various environmental sensor data into the anomaly linkage detection algorithm in real time, automatically comparing the operating noise of each device, environmental electromagnetic intensity, and air temperature and humidity information; for detected anomalies, the equipment immediately alerts the maintenance personnel to pay attention through voice / vibration. AI-assisted question answering and operation guidance: On-site maintenance personnel can ask questions via voice. The AI ​​recorder, based on the currently collected information and the converter station's dedicated knowledge base, uses a context fusion reasoning algorithm to provide maintenance personnel with accurate operating procedures, historical case references, or emergency handling suggestions. Data security synchronization and process archiving: All data during the inspection process is automatically classified, encrypted, and synchronized to the secure cloud platform.

2. The method for safe operation and maintenance management of a multi-physical-form ultra-high voltage converter station adapted to strong electric field environments according to claim 1, characterized in that: The aforementioned self-identification of converter station equipment type and inspection object includes: comprehensive perception of the current work area; and the collection of environmental noise signals, air and environmental electromagnetic intensity parameters, and temperature time series data by multiple built-in sensors to generate a comprehensive feature vector. The AI ​​model uses a pre-trained discriminator to identify the converter station equipment type label from the environmental feature vector; The AI ​​recorder's camera activates target detection and text recognition algorithms. Using on-site QR codes or equipment nameplate information, it extracts the location of the equipment and obtains its unique identification code. By dynamically comparing the equipment identification with the inspection task list, it automatically matches the set of core equipment that needs to be inspected in this inspection in real time.

3. The method for safe operation and maintenance management of multi-physical-form ultra-high voltage converter stations adapted to strong electric field environments according to claim 1, characterized in that: The aforementioned intelligent anomaly capture and real-time risk alert includes: the AI ​​recorder sends the audio, video, images and various environmental sensor data collected on-site to the device anomaly linkage detection algorithm for analysis in real time; Audio anomalies are detected by comparing healthy samples with spectrograms; anomalies in temperature and humidity sensor data and environmental electromagnetic intensity data are detected by comparing moving mean and standard deviation; and appearance anomalies are detected in images and videos by using deep convolutional neural networks. When a certain modality is abnormal, the linkage algorithm dynamically adjusts the sampling frequency and analysis sensitivity of relevant data, improves image resolution, and guides maintenance personnel to supplement the collected information. After the multimodal abnormality scores are weighted and fused, risk warnings are given through voice broadcast, interface highlighting, or vibration, and detailed abnormality records are guided.

4. The method for safe operation and maintenance management of multi-physical-form ultra-high voltage converter stations adapted to strong electric field environments according to claim 1, characterized in that: The AI-assisted question-and-answer and operation guidance include: using a localized speech recognition model to transcribe the maintenance personnel's on-site questions into text, and combining anomaly detection scores, real-time collected data, equipment operating status and historical inspection logs to form a comprehensive scene description vector; By using a context fusion reasoning algorithm, the question text, scene data vector and knowledge graph are jointly encoded, and the question-knowledge matching score and the current anomaly detection distribution probability are weighted and sorted to select the operation steps, historical cases or handling suggestions most relevant to the problem and equipment status. The AI-assisted question-and-answer and operation guidance include: broadcasting instructions to maintenance personnel in the form of natural language, and supplementing the image acquisition process and additional instructions based on the on-site matching degree.

5. The method for safe operation and maintenance management of a multi-physical-form ultra-high voltage converter station adapted to strong electric field environments according to claim 1, characterized in that: The aforementioned data security synchronization and process archiving include: the AI ​​recorder uniformly manages all data collected during the inspection process, classifies and tags the raw data, and synchronizes the data based on user permissions and enterprise security policies. The synchronized data is automatically organized into structured inspection logs, generating event indexes and summaries, and compiling multimedia reports. Archived content is saved locally and synchronized to the cloud-based intelligent knowledge base.

6. The method for safe operation and maintenance management of a multi-physical-form ultra-high voltage converter station adapted to strong electric field environments according to claim 1, characterized in that: The on-site inspection and process information collection includes: the AI ​​recorder automatically generates the optimal inspection path based on the synchronized inspection task and the type of converter station equipment, and guides the maintenance personnel to check the equipment step by step through voice, display or vibration; the multimodal acquisition module automatically records audio, video and photos and marks the time, geographical location and equipment number to realize abnormal information aggregation and event tracing; all data is encrypted and stored locally and synchronized to the cloud platform.

7. An AI recorder for safety management and control of operation and maintenance of ultra-high voltage converter stations with multiple physical forms adapted to strong electric field environments, wherein the AI ​​recorder is applicable to the method as described in any one of claims 1-6, characterized in that: The AI ​​recorder includes: a camera, a control module, a power module, a display screen, and a clip; The control module includes: a wireless transmission module, a positioning module, a voice module, and a processing module; The camera, positioning module, voice module, and display screen are all electrically connected to the processing module. The output of the processing module is electrically connected to the wireless transmission module, and the wireless transmission module is connected to the background database. The power module provides power to the recorder, which is worn via a clip.

8. A badge-style AI recorder, wherein the badge-style AI recorder is based on the AI ​​recorder of claim 7, characterized in that: The aforementioned name tag-style AI recorder is used in conjunction with work clothes via a buckle, and its outer shell is shaped like a name tag.

9. A head-mounted AI recorder, wherein the head-mounted AI recorder is based on the AI ​​recorder of claim 7, characterized in that: The aforementioned head-mounted AI recorder also includes; The lighting module uses LED lighting beads and is connected to the power module. The recorder is connected to the helmet via a clip. The AI ​​recorder has a semi-circular arc-shaped shell, which conforms to the shape of a helmet edge. The camera and lighting module are located at opposite ends of the shell, so they do not interfere with each other's operation.

10. A seatbelt-type AI recorder, wherein the seatbelt-type AI recorder is based on the AI ​​recorder of claim 7, characterized in that: The aforementioned seatbelt-type AI recorder is used in conjunction with a seatbelt and is secured to the seatbelt via a buckle.

Images