A device maintenance management system and method based on a low-code platform and AI intelligent glasses

Abstract

The application discloses a kind of based on low code platform and AI intelligent glasses equipment maintenance management system and method, it is related to industrial equipment maintenance technical field, including low code equipment maintenance work order configuration module, intelligent glasses end interactive guidance module, equipment maintenance process digitization acquisition module and maintenance data closed loop management module;Solve the existing equipment maintenance work order configuration inflexible, guiding interaction is poor, process digitization is insufficient, safety early warning lag, data no closed loop problem, realize maintenance process quick adaptation, intelligent guidance, safety early warning, whole-process traceability and optimization.

Description

Technical Field

[0001] This invention relates to the field of industrial equipment maintenance technology, specifically to an equipment maintenance management system and method based on a low-code platform and AI smart glasses. Background Technology

[0002] Routine maintenance of industrial equipment is a core element in ensuring stable operation, reducing downtime risks, and extending equipment lifespan. The standardization, efficiency, and digitalization of equipment maintenance directly impact production safety and operating costs. However, traditional equipment maintenance methods have the following drawbacks, specifically:

[0003] Firstly, existing equipment maintenance work orders mostly rely on traditional code development or customized development models, resulting in rigid processes that are difficult to adjust and require professional technical personnel to modify. This makes it difficult to quickly adapt to the differentiated needs of different equipment models and maintenance scenarios. Furthermore, the work order templates have poor reusability, and the development cycle for new scenario work orders can be as long as several weeks, making it difficult to quickly respond to the needs of equipment iteration and maintenance process optimization.

[0004] Secondly, although AR glasses have been introduced to assist in operation guidance in some maintenance scenarios, they can only achieve simple graphic and text overlay display, lacking AI technology support and unable to provide intelligent Q&A, fault diagnosis suggestions and other auxiliary capabilities; at the same time, the AR guidance system and the back-end work order system are independent of each other, the guidance process cannot be linked with the work order steps, the operation guidance is disconnected from the actual maintenance task, and operation deviation is prone to occur, resulting in insufficient guidance accuracy and practicality.

[0005] Thirdly, in the traditional maintenance process, key operation records, status verification, and evidence retention data mostly rely on manual paper records or post-event filling, which has problems such as record lag, information omission, and large human error; moreover, there is no automated collection mechanism, which cannot capture the data of the entire maintenance process in real time, multimedia evidence cannot be linked with operation steps, the authenticity of data is difficult to verify, and subsequent traceability and responsibility determination are difficult.

[0006] Fourth, the images, videos, and text records collected during the maintenance process are mostly stored on ordinary local devices or unencrypted servers. There is no unified mechanism for spatiotemporal traceability and integrity verification, making the data easy to be tampered with or lost. At the same time, the data is not bound to key information such as the operator, time, and location, making it impossible to form a traceable chain of evidence and failing to meet the compliance and auditing management requirements of the industrial sector. Summary of the Invention

[0007] This invention provides a device maintenance management system and method based on a low-code platform and AI smart glasses, to achieve rapid adaptation and intelligent guidance of maintenance processes.

[0008] This invention provides the following technical solution:

[0009] In a first aspect, the present invention provides a device maintenance management system based on a low-code platform and AI smart glasses, comprising:

[0010] The low-code equipment maintenance work order configuration module is used to quickly configure and generate structured work order JSON templates containing step sequences, component types, business attributes, and strong control tags through a visual drag-and-drop interface and a pre-built component library;

[0011] The smart glasses-side interactive guidance module is used to receive the work order JSON template and dynamically render the corresponding AR operation guidance overlaid on the real device screen according to the configuration information of the current work order step node. At the same time, it realizes hands-free operation and intelligent knowledge assistance through multimodal AI interaction.

[0012] The equipment maintenance process digital acquisition module is used to automatically trigger the camera to collect multimedia evidence according to the acquisition type configured in the work order when the work order step node is configured as a strong control node, and to attach an immutable digital watermark consisting of NTP / BeiDou time synchronization timestamp, GPS / BeiDou dual-mode positioning coordinates and SHA-256 hash check value to the collected data.

[0013] The maintenance data closed-loop management module is used to realize status-driven workflow updates and multi-dimensional statistical analysis based on accumulated maintenance data, and to automatically adjust the cycle of preventive maintenance plans according to historical failure frequency and average repair time.

[0014] As a further improvement of the present invention, the low-code device maintenance work order configuration module includes:

[0015] The visual interactive interface unit includes a canvas area, a component library, and an attribute configuration area. The canvas area displays the maintenance step sequence based on a flow layout and uses drag-and-drop component technology to adjust the order of step nodes. The component library provides pre-set standardized maintenance action components. The attribute configuration area is used to dynamically load configuration forms and verify the legality of configuration logic in real time when a canvas node is clicked.

[0016] The template data structure and storage unit are defined by the TaskTemplate entity storage process on the backend, and the DOM nodes on the canvas are converted into structured data for persistence on the frontend through JSON serialization technology.

[0017] The integrity verification unit is used to automatically traverse all nodes and check whether the required fields are complete before the template is saved or published.

[0018] As a further improvement of the present invention, the smart glasses-side interactive guidance module includes:

[0019] The task precision distribution unit is used to push work orders to the smart glasses of designated maintenance personnel based on personnel skill matrix and geographical location, and supports two modes: order grabbing and order dispatching. It uses WebSocket to maintain real-time communication between the end and the cloud.

[0020] The work order step-driven AR augmented reality guidance unit is used to parse the work order JSON template in real time, dynamically render AR operation guidance including arrows, highlight boxes, and parameter threshold prompts according to the type and strong control mark of the current work order step node, and use head posture tracking to ensure that the guidance information is always in the user's comfortable field of vision. After confirmation, it automatically advances to the next step node.

[0021] The multimodal AI interaction unit integrates an automatic speech recognition engine to support hands-free voice commands; and an integrated text-to-speech engine to convert text operation steps in work order steps into speech broadcasts.

[0022] The AI ​​knowledge base companion unit is used to automatically retrieve relevant technical documents, drawings, and historical maintenance cases from the knowledge base based on the equipment type bound to the current work order step node, and use them as context input for large language model reasoning.

[0023] As a further improvement of the present invention, the accompanying unit of the AI ​​knowledge base includes:

[0024] The knowledge base construction sub-unit is used to perform semantic slicing on documents such as maintenance manuals and equipment manuals. The slices are converted into vectors through an embedding model, stored in a vector database, and an equipment type index is built.

[0025] The retrieval subunit is used to construct a query vector by combining the user's question transcribed by automatic speech recognition with the current work order device context, perform an approximate nearest neighbor retrieval, and recall Top-K relevant document slices;

[0026] The large language model context injection subunit is used to inject work order step information, equipment parameters and recall document slices into the large language model through structured prompt templates, generate troubleshooting suggestions, and broadcast them in text-to-speech format.

[0027] As a further improvement of the present invention, the voice interaction processing method of the smart glasses terminal interaction guidance module includes:

[0028] Input an audio stream and output unstructured text to achieve automatic speech recognition and transcription;

[0029] The automatic speech recognition text, along with the device type and normal range parameters of the current work order step node, is injected into the large language model to extract the device name, abnormal value, and abnormal description.

[0030] The extracted values ​​are compared with the thresholds configured in the work order step nodes, and the pass or fail status is automatically marked, with the judgment result announced by voice.

[0031] If an anomaly is detected, the system will automatically trigger a camera to take a picture for evidence based on the strong control flag configuration of the work order steps, and bind it to the step result identifier.

[0032] As a further improvement of the present invention, the digital acquisition module for the equipment maintenance process includes:

[0033] The multimedia evidence chain acquisition unit is used to automatically trigger camera acquisition based on the acquisition type configured in the work order when a work order step node is configured as a strong control node, ensuring a strict binding relationship between each piece of multimedia evidence and the corresponding standard operating procedure step.

[0034] The temporal and spatial immutable digital watermarking unit is used to automatically add a standard UTC timestamp synchronized by NTP or BeiDou time service, a spatial positioning coordinate with a horizontal accuracy of better than 5 meters by GPS or BeiDou dual-mode positioning, and a file integrity check value calculated by SHA-256 hash algorithm to all collected data.

[0035] The intelligent association and binding unit is used to manage unstructured data through backend object storage, automatically binding the collected files with work order identifiers and step result identifiers, and writing the hash value into the step result record to form a structured digital archive.

[0036] The AI-powered anomaly intelligent classification unit is used to automatically identify the device status in uploaded photos using computer vision models, and to analyze text records using large language models, automatically extracting keywords and converting unstructured descriptions into standardized fault classification labels.

[0037] As a further improvement of the present invention, the digital acquisition module for equipment maintenance process further includes a target equipment identification unit, and the identification method of the target equipment identification unit includes:

[0038] The smart glasses terminal continuously scans Bluetooth beacons in the work area, reads signal strength values, determines whether personnel have entered the 3-5 meter working range of a specific device, and automatically preloads the device data for that area.

[0039] The glasses camera is activated to capture video streams. Key points are extracted using ORB or AKAZE feature extraction algorithms and compared with device reference images in the local database. When the matching degree exceeds a preset threshold, the target device identifier is locked and a guide icon is superimposed on the screen.

[0040] As a further improvement of the present invention, the maintenance data closed-loop management module includes:

[0041] The status-driven workflow unit is used to trigger the backend event bus when the work order status changes, and automatically update the device's master data status.

[0042] The multi-dimensional reporting unit is used to statistically analyze the average maintenance time of different personnel and different machine models and identify efficiency bottlenecks, and generate heat maps that show the equipment parts and time periods with high failure rates.

[0043] The knowledge accumulation and reuse unit is used to transform historical anomaly records into micro-knowledge points and support precise retrieval. Based on historical fault frequency and average repair time statistics, it automatically adjusts the cycle of preventive maintenance plans, realizing the transformation from on-time maintenance to on-demand maintenance.

[0044] Secondly, a device maintenance and management method based on a low-code platform and AI smart glasses, applicable to the aforementioned system, includes the following steps:

[0045] Step S1: The cloud server parses the work order JSON template generated by the low-code platform, associates the task tree with the corresponding subset of the exception status library, packages it and pushes it to the smart glasses.

[0046] Step S2: When personnel enter the work area, the smart glasses sense the Bluetooth signal, and the camera detects the equipment nameplate or appearance features. The screen then displays the AR operation guide for the first step node according to the work order step configuration.

[0047] Step S3, voice input and image verification are driven by the strong control mark configuration of the work order step node, and the collected content is bound to the step result identifier and spatiotemporal hash watermark for evidence storage;

[0048] Step S4: If the step is determined to be abnormal, the system will automatically trigger the collection and uploading of the abnormal package to the cloud, and generate a maintenance work order to trigger the next level of work order process.

[0049] As a further improvement of the present invention, in step S3, the smart glasses and the cloud use the JSON cloud-to-device communication protocol for data interaction:

[0050] Downlink work order configuration protocol: The low-code platform compiles the standard operation process into a TaskTemplate object and issues it. The structure includes step identifier, component type, attribute information and strong control flag field;

[0051] Uplink operation result protocol: After execution by the smart glasses, a structured result is returned, including step result identifier, judgment result, structured measurement value, standard text converted by automatic speech recognition, list of evidence files, NTP or BeiDou time stamp, GPS or BeiDou positioning coordinates, and SHA-256 file hash value field.

[0052] The present invention has the following beneficial effects:

[0053] 1. It adopts a low-code visual drag-and-drop interface and a pre-built standardized component library, which can quickly configure maintenance work order templates without professional coding skills. It supports free adjustment of step sequence and custom configuration of business attributes. The template has strong reusability, and the process adjustment cycle is shortened from several weeks to hours. It can quickly adapt to the differentiated needs of different equipment models and maintenance scenarios, and significantly reduce the development and maintenance costs of work orders.

[0054] 2. Based on the work order JSON template to drive AR dynamic rendering, virtual guidance such as arrows, highlight boxes, and parameter threshold prompts are accurately superimposed on the real device screen. With head posture tracking, the guidance is kept in a comfortable field of vision. It integrates multimodal AI interaction, supports hands-free voice commands and step broadcasts, and relies on the AI ​​knowledge base to retrieve technical documents and historical cases in real time and generate fault suggestions, completely freeing up maintenance personnel's hands, reducing operational errors, and significantly improving the accuracy of maintenance guidance.

[0055] 3. Work order control nodes automatically trigger multimedia evidence collection, ensuring that evidence is strictly bound to operation steps; the collected data is automatically appended with NTP / BeiDou time stamps, GPS / BeiDou dual-mode positioning coordinates and SHA-256 hash check values, forming an immutable digital watermark for spatiotemporal traceability and integrity verification; the data is automatically archived and bound to work orders and step identifiers, realizing that the entire maintenance process data is searchable, traceable and verifiable, reducing the cost and error of manual recording. Attached Figure Description

[0056] Figure 1 This is an architecture diagram of the equipment maintenance management system in this invention.

[0057] Figure 2 This is a system interface diagram of the canvas area in this invention.

[0058] Figure 3 This is a system interface diagram of the component library in this invention.

[0059] Figure 4 This is a system interface diagram of the attribute configuration area in this invention.

[0060] Figure 5 This is a system interface diagram of the integrity verification unit in this invention.

[0061] Figure 6 This is a system interface diagram of the task precise distribution unit in this invention.

[0062] Figure 7 This is a diagram of the inspection interface of the smart glasses in this invention.

[0063] Figure 8 This is a question-and-answer diagram for maintenance knowledge in this invention.

[0064] Figure 9 This is a screenshot of the camera interface of the digital acquisition module for equipment maintenance process in this invention.

[0065] Figure 10 This is a screenshot of the inspection results interface of the digital acquisition module for equipment maintenance process in this invention.

[0066] Figure 11 This is a diagram of the work order status interface of the maintenance data closed-loop management module in this invention.

[0067] Figure 12 This is a multi-dimensional statistical interface diagram of the maintenance data closed-loop management module in this invention.

[0068] Figure 13 This is a diagram of the work order anomaly retrieval and image retrieval interface of the maintenance data closed-loop management module in this invention.

[0069] Figure 14 This is a flowchart of the equipment maintenance and management method in this invention. Detailed Implementation

[0070] The technical solutions of the embodiments of this specification will be explained and described below with reference to the accompanying drawings. However, the following embodiments are only preferred embodiments of this specification and not all of them. Other embodiments obtained by those skilled in the art based on the embodiments in the implementation methods without creative effort are all within the protection scope of this specification.

[0071] Example 1

[0072] Please see Figures 1-13 As shown, this invention provides a device maintenance and management system based on a low-code platform and AI smart glasses, comprising:

[0073] The low-code equipment maintenance work order configuration module is used to quickly configure and generate structured work order JSON templates containing step sequences, component types, business attributes, and strong control tags through a visual drag-and-drop interface and a pre-built component library.

[0074] Specifically, the low-code device maintenance work order configuration module includes:

[0075] First, the visual interactive interface unit. It includes a canvas area, a component library, and a property configuration area.

[0076] like Figure 2As shown, the canvas area displays the maintenance step sequence based on a fluid layout. It utilizes drag-and-drop (vuedraggable) component technology to adjust the order of step nodes, allowing users to intuitively adjust the inspection order. The canvas area employs a drag-and-drop sorting index update algorithm for step nodes, ensuring that the visual display order is completely consistent with the data sorting index, providing ordered step data for backend storage.

[0077] like Figure 3 As shown, the component library provides pre-built standardized maintenance action components, including: a value recording component (ValueRecordType) for inputting voltage, temperature, etc., supporting configuration of minimum / maximum value thresholds; a result judgment component (ResultRecordType) for quick "qualified / unqualified" determination; a universal recording component (UniversalRecordType) for text description; and business-specific components such as goods inbound for specific scenarios.

[0078] like Figure 4 As shown, the attribute configuration area is used to dynamically load the configuration form and verify the legality of the configuration logic in real time when a canvas node is clicked. The attribute configuration area adopts a real-time component attribute legality verification algorithm. Through the instant verification of the attribute configuration area, it prevents users from entering incorrect or invalid component configurations, ensuring the legality of the configuration logic of a single node.

[0079] Employing a low-code visual drag-and-drop interface and a pre-built standardized component library, maintenance work order templates can be quickly configured without professional coding skills; the templates are highly reusable, and the process adjustment cycle is shortened from several weeks to hours, which can quickly adapt to the differentiated needs of different equipment models and maintenance scenarios, significantly reducing the development and maintenance costs of work orders.

[0080] Second, template data structure and storage unit.

[0081] The backend uses TaskTemplate entities to store process definitions, including process types (such as PATROL / REPAIR) and an ordered list of steps.

[0082] The front-end uses JSON serialization technology to convert DOM (Document Object Model) nodes on the canvas into structured data for persistence. The structured data includes: tag (component identifier), properties (business attributes), sort (sorting index), and is_mandatory (strong control tag).

[0083] Third, the integrity verification unit is used to automatically traverse all nodes before the template is saved or published to check whether the required fields are complete. That is, the system has a built-in step data verification algorithm that automatically traverses all nodes before the template is saved / published to check whether the data unit is missing, whether the judgment standard is not set, and other required fields, to ensure the logical closed loop of the work order sent to the glasses.

[0084] The smart glasses-side interactive guidance module is used to receive the work order JSON template and dynamically render the corresponding AR operation guidance overlaid on the real device screen according to the configuration information of the current work order step node. At the same time, it realizes hands-free operation and intelligent knowledge assistance through multimodal AI interaction.

[0085] Specifically, the smart glasses-side interactive guidance module includes:

[0086] First, the precise task distribution unit. For example... Figure 6 As shown, the task precision distribution unit is used to push work orders to the smart glasses of designated maintenance personnel based on the personnel skill matrix and geographical location. It supports both order-grabbing and order-assignment modes, and utilizes WebSocket (full-duplex communication protocol) to maintain real-time communication between the end and the cloud. This unit employs a personnel skill-geographical location weighted order-assignment matching algorithm, pushing work orders to the glasses based on skill matching degree and distance. The specific execution steps are as follows:

[0087] (1) Data collection: skill tag matrix of maintenance personnel, real-time GPS location, on-duty status, and historical maintenance qualification rate.

[0088] (2) Work order binding: Extract the device type, required skill level, and work site coordinates of the work order device_type.

[0089] (3) Calculate the total matching score using a two-dimensional weighted scoring method. The calculation formula is as follows: Where S is the overall personnel matching score, S skill The similarity score is the skill matching score (range 0~1), where D is the straight-line distance between the person and the work point. max To maximize the effective dispatch distance, S rate This represents the historical maintenance completion pass rate. α, β, and γ are weighting coefficients, where α + β + γ = 1.

[0090] (4) Score sorting, automatic order assignment prioritizes high-scoring personnel; order grabbing mode allows personnel to apply for orders independently.

[0091] (5) WebSocket establishes a long connection between the end and the cloud, and sends work orders to the smart glasses terminal in real time.

[0092] Second, the AR (Augmented Reality) guidance unit driven by work order steps. For example... Figure 7As shown, the AR augmented reality guidance unit is used to parse the work order JSON template in real time. Based on the type of the current work order step node and the strong control mark, it dynamically renders AR operation guidance including arrows, highlight boxes, and parameter threshold prompts. It also uses head posture tracking to ensure that the guidance information is always in the user's comfortable field of vision. After confirmation, it automatically advances to the next step node.

[0093] The smart glasses parse the work order JSON template generated by the low-code platform in real time. Based on the type of the current work order step node (text / image / video / form) and the `is_mandatory` strong control flag, it dynamically renders the corresponding AR operation guidance, overlaying virtual guidance content (arrows, highlighted boxes, parameter threshold prompts) onto the real device screen, achieving a strict correspondence between the work order step sequence and the AR guidance content. The display of AR content at each step is determined by the work order step configuration, and automatically advances to the next step node after confirmation. Eye-tracking: Head tracking is used to ensure that the guidance information is always within the user's comfortable field of vision. The above employs an AR virtual-real space coordinate mapping overlay algorithm, with the spatial mapping formula as follows: , where P v P is a virtual guide pixel coordinate; r is the pixel coordinates of the real-world device image; M is the spatial mapping transformation matrix; B is the visual offset compensation vector.

[0094] Third, the multimodal AI interaction unit.

[0095] It integrates an automatic speech recognition engine to support hands-free voice commands; it supports hands-free voice commands such as "Next," "Take a photo," and "Call an expert," completely freeing your hands. The automatic speech recognition engine uses the ASR (Automatic Speech Command Acoustic Feature Matching) algorithm.

[0096] An integrated text-to-speech (TTS) engine is used to convert text-based operation steps in work order steps into spoken audio. The engine employs an industrial-scenario TTS prosody normalization synthesis algorithm to achieve text-to-speech audio transmission of work order steps.

[0097] Fourth, the AI ​​knowledge base accompanying unit is used to automatically retrieve related technical documents, drawings and historical maintenance cases from the knowledge base based on the device type (device_type field) bound to the current work order step node, as the context input for the large language model reasoning. This function is driven by the device context in the work order step, without the need for visual monitoring based on voiceprint / hand posture recognition tools. Its advantages are: (1) It has strong adaptability to industrial sites and strong anti-interference ability. Because industrial sites are noisy, have complex lighting, and random and irregular operation actions, it is easy to cause voiceprint misidentification and hand posture recognition failure. However, the context driving is not affected by environmental noise, personnel actions, obstruction and posture. The triggering logic is stable and reliable, and it is suitable for all scenarios of workshops, factories and computer rooms. (2) It has low code and fast adaptation and high implementation efficiency. Because all driving rules only need to be modified in the background work order template configuration, the front-end interaction logic can be changed by adjusting the equipment parameters and process flow. There is no need to modify the terminal program and the underlying algorithm, and it can quickly adapt to the maintenance business of multiple categories and models of equipment.

[0098] The accompanying unit of the AI ​​knowledge base includes:

[0099] The knowledge base construction sub-unit is used to semantically slice documents such as maintenance manuals and equipment manuals (divided by paragraph / chapter, chunk_size approximately 512 tokens). The slices are converted into vectors through an embedding model (such as text-embedding-ada-002 or locally deployed BAAI / bge-m3), stored in a vector database (such as Milvus / Qdrant), and an equipment type index is built.

[0100] The retrieval subunit is used to construct a query vector by combining the user's question transcribed from automatic speech recognition with the current work order device context, and then perform an Approximate Nearest Neighbor (ANN) retrieval to recall Top-K relevant document slices. The execution steps of the ANN approximate nearest neighbor vector retrieval algorithm are as follows:

[0101] (1) Construct query text by combining the current work order device_type and the user's voice question;

[0102] (2) Generate query vector ;

[0103] (3) Perform ANN fast search in the vector library;

[0104] (4) Recall Top-K highly correlated document slices by sorting by similarity;

[0105] (5) Filter out irrelevant equipment data and retain maintenance data of the same type.

[0106] The formula for calculating similarity is:

[0107] ,in These are vector slices of documents from the knowledge base. The closer the similarity value is to 1, the stronger the content correlation.

[0108] The large language model context injection subunit is used to inject work order step information, equipment parameters and recall document slices into the large language model through structured prompt templates, generate troubleshooting suggestions, and broadcast them in text-to-speech format.

[0109] Driven by a work order JSON template, AR dynamic rendering accurately overlays virtual guidance elements such as arrows, highlighted boxes, and parameter threshold prompts onto the real equipment screen. Head posture tracking ensures the guidance remains within a comfortable field of view. Multimodal AI interaction is integrated, supporting hands-free voice commands and step-by-step instructions. Leveraging an AI knowledge base, it retrieves technical documents and historical cases in real time and generates fault suggestions, completely freeing up maintenance personnel's hands, reducing operational errors, and significantly improving the accuracy of maintenance guidance. Integrating computer vision and large language models, it automatically identifies equipment status, extracts fault keywords, and transforms unstructured records into standardized fault labels. Combined with vector databases and RAG retrieval technology, it accurately retrieves relevant technical information, quickly locates the root cause of faults, and outputs troubleshooting solutions, greatly improving the efficiency of anomaly identification and problem-solving, and reducing reliance on the experience of senior maintenance personnel.

[0110] The digital data acquisition module for equipment maintenance processes is used to automatically trigger the camera to collect multimedia evidence based on the acquisition type configured in the work order when the work order step node is configured as a strong control node. It also adds an immutable digital watermark to the collected data, consisting of an NTP / BeiDou time stamp, GPS / BeiDou dual-mode positioning coordinates, and SHA-256 hash check value.

[0111] Specifically, the digital data acquisition module for the equipment maintenance process includes:

[0112] First, the multimedia evidence chain acquisition unit is used to automatically trigger camera acquisition based on the acquisition type (photo / video) configured in the work order when the work order step node is configured as a strong control node, ensuring a strict binding relationship between each piece of multimedia evidence and the corresponding standard operating procedure step.

[0113] Second, the spatiotemporally immutable digital watermarking unit is used to automatically add a standard UTC timestamp synchronized by NTP or BeiDou time service (accurate to the millisecond level to prevent tampering with the device's local clock), spatial positioning coordinates with a horizontal accuracy better than 5 meters from GPS or BeiDou dual-mode positioning (supporting indoor Bluetooth beacon-assisted positioning), and a file integrity check value calculated by the SHA-256 hash algorithm to all collected data. These are written together with the spatiotemporal metadata into the work order step result record to realize file integrity verification and ensure that the data uploaded to OSS cannot be tampered with.

[0114] Third, the intelligent association and binding unit is used to manage unstructured data through backend object storage. It automatically binds the collected files with the work order identifier ID (task_id) and the step result identifier ID (check_item_id), and writes the hash value into the step result record to form a structured digital archive.

[0115] Fourth, the AI ​​anomaly intelligent classification unit is used to automatically identify the equipment status (such as instrument readings and degree of corrosion) in uploaded photos using computer vision models, and to analyze text records using large language models to automatically extract keywords (such as "overheating" and "abnormal noise") and convert unstructured descriptions into standardized fault classification labels.

[0116] Fifth, the target device identification unit, whose identification methods include:

[0117] Level 1 Area Awareness. The smart glasses terminal continuously scans Bluetooth beacons within the work area, reads signal strength values, determines whether personnel have entered the 3-5 meter working range of specific equipment, and automatically preloads equipment data for that area;

[0118] Secondary precise recognition (image matching). The glasses' camera is activated to capture video streams, and key points are extracted using ORB or AKAZE feature extraction algorithms. These points are compared with device reference images in the local database. When the matching degree exceeds a preset threshold (e.g., 0.8), the target device identifier is locked and a guide icon is superimposed on the screen.

[0119] Work order control nodes automatically trigger multimedia evidence collection, ensuring that evidence is strictly bound to operation steps; collected data is automatically appended with NTP / BeiDou time stamps, GPS / BeiDou dual-mode positioning coordinates, and SHA-256 hash check values, forming an immutable digital watermark for spatiotemporal traceability and integrity verification; data is automatically archived and bound to work orders and step identifiers, enabling data to be searchable, traceable, and verifiable throughout the maintenance process, reducing manual recording costs and errors.

[0120] The maintenance data closed-loop management module is used to realize status-driven workflow updates and multi-dimensional statistical analysis based on accumulated maintenance data, and to automatically adjust the cycle of preventive maintenance plans according to historical failure frequency and average repair time.

[0121] Specifically, the maintenance data closed-loop management module includes:

[0122] The status-driven workflow unit is used to trigger the backend event bus when the work order status changes, and automatically update the device's master data status.

[0123] The multi-dimensional reporting unit is used to statistically analyze the average maintenance time of different personnel and different machine models and identify efficiency bottlenecks, and generate heat maps that show the equipment parts and time periods with high failure rates.

[0124] The knowledge accumulation and reuse unit is used to transform historical anomaly records into micro-knowledge points and support precise retrieval. Based on historical fault frequency and average repair time statistics, it automatically adjusts the cycle of preventive maintenance plans, realizing the transformation from on-time maintenance to on-demand maintenance.

[0125] Build a closed-loop management system for maintenance data to achieve automatic updates of workflows driven by work order status; conduct multi-dimensional statistical analysis of personnel efficiency and equipment failure patterns to generate visual reports and heat maps; accumulate historical maintenance knowledge to form a searchable knowledge base, and automatically adjust preventive maintenance cycles based on failure frequency and repair duration, upgrading from traditional "on-time maintenance" to data-driven "on-demand maintenance," reducing maintenance costs and improving equipment operational reliability.

[0126] The voice interaction processing method of the smart glasses terminal interaction guidance module mentioned above includes:

[0127] The first step is to input an audio stream and output unstructured text T_raw (e.g., "The hydraulic pump pressure gauge shows 35, which is a bit high") to achieve automatic speech recognition and transcription.

[0128] The second step involves injecting the automatically recognized speech text along with the device type and normal range parameters of the current work order step node into the large language model to extract the device name, abnormal value, and abnormal description. During LLM inference, the work order step configuration is used as the anchor point, rather than generalized knowledge inference.

[0129] The third step is to compare the extracted values ​​with the thresholds configured in the work order step nodes, automatically mark the pass or fail status, and announce the judgment result by voice.

[0130] Fourth step: If an anomaly is detected, the system will automatically trigger the camera to take a picture for evidence based on the mandatory control flag (is_mandatory) of the work order step and bind it to the step result identifier.

[0131] In this embodiment, multimodal AI identifies equipment anomalies and violations during maintenance in real time, broadcasts the judgment results via voice and triggers warnings; it also links with an AR interface to highlight risk areas, replacing traditional manual warnings, thereby achieving real-time identification and accurate warnings of safety risks, reducing the incidence of safety accidents at industrial maintenance sites, and ensuring the personal safety of maintenance personnel.

[0132] Example 2

[0133] Please see Figure 14As shown, a device maintenance management method based on a low-code platform and AI smart glasses is applicable to the aforementioned system and includes the following steps:

[0134] Step S1: The cloud server parses the work order JSON template generated by the low-code platform, associates the task tree with the corresponding subset of the exception status library, packages it, and pushes it to the smart glasses.

[0135] Specifically, the cloud server parses the work order JSON template generated by the low-code platform, associates it with the corresponding device's abnormal status library subset, and packages it to generate a work order data package containing a step sequence, AR guidance parameters, strong control markers, and knowledge base indexes; it is then pushed to the smart glasses of the designated maintenance personnel via WebSocket, where the glasses complete the data package parsing, local caching, and preload the corresponding device's basic data.

[0136] Step S2: When personnel enter the work area, the smart glasses sense the Bluetooth signal, and the camera detects the equipment nameplate or appearance features. The screen then displays the AR operation guide for the first step node according to the work order step configuration.

[0137] Specifically, after maintenance personnel enter the work area, the smart glasses continuously scan the Bluetooth beacon to determine that they have entered a 3-5 meter working range of the target equipment; the camera is activated to capture video streams, and the equipment reference image is matched through the ORB feature extraction algorithm. After locking the target equipment, the first step node of the work order is automatically loaded, and AR operation guidance (such as screw removal arrows and instrument observation highlight boxes) is rendered on the real equipment screen, while the operation requirements are broadcast in voice at the same time.

[0138] Step S3, voice input and image verification, is driven by the strong control flag configuration of the work order step node, and the collected content is bound to the step result identifier and spatiotemporal hash watermark for evidence storage.

[0139] Specifically, maintenance personnel perform maintenance operations through hands-free voice commands (such as "start testing" and "value reading"). The voice data is transcribed into text by an automatic speech recognition engine. The system injects the text, equipment parameters, and step thresholds into a large language model, extracts abnormal values ​​and descriptions, automatically determines the results, and broadcasts them via voice.

[0140] If the step is a non-mandatory node, it will automatically proceed to the next process after the result is confirmed.

[0141] If the step is a strongly controlled node, the camera will automatically take a picture / record video for evidence storage, regardless of whether the result is abnormal. The collected data will be attached with a spatiotemporal hash watermark and bound to the step result identifier.

[0142] The smart glasses and the cloud use the JSON cloud-to-device communication protocol for data exchange.

[0143] Downlink Schema: The low-code platform compiles the standard operation process into a TaskTemplate object and issues it. The structure includes a step identifier (step_id), component type (IMAGE_TEXT / VIDEO / FORM), attribute information (title / guidance_text / reference_image_url / threshold), and a mandatory flag (is_mandatory) field.

[0144] Uplink Schema: After execution by the smart glasses, a structured result is returned, including step result identifier (check_item_id), judgment result (Pass / Fail), structured measurement value (measure_result), standard text converted from automatic speech recognition (description), list of evidence files (evidence_photos), NTP or BeiDou time stamp (timestamp), GPS or BeiDou positioning coordinates (location), and SHA-256 file hash value field (file_hash).

[0145] Step S4: If the step is determined to be abnormal, the system will automatically trigger the collection and uploading of the abnormal package to the cloud, and generate a maintenance work order to trigger the next level of work order process.

[0146] Specifically, when the system determines that the result of the step is abnormal, it automatically packages the abnormal data (including voice recordings, on-site photos, and equipment parameters) and uploads it to the cloud; the cloud parses the abnormal data, matches it with the fault knowledge base to generate a maintenance work order, and automatically triggers the next level of maintenance process; after the maintenance is completed, all data is archived to the maintenance database, and the equipment health status is updated synchronously to provide support for subsequent data analysis and maintenance plan optimization.

[0147] This equipment maintenance management method leverages low-code technology and AI smart glasses to achieve standardized, intelligent, and digital implementation of the entire equipment maintenance process, yielding significant results. Real-time edge-cloud collaboration accurately distributes work orders, quickly locating target equipment using Bluetooth area sensing and image feature matching. Work order-driven AR guidance overlays real-world visuals, coupled with voice broadcasts and hands-free interaction, significantly lowers the barrier to on-site operation and reduces human error. Control nodes automatically trigger multimedia data collection, adding time stamps and hash watermarks to ensure the authenticity, integrity, immutability, and full traceability of process evidence. Multimodal AI analyzes voice in real time, identifies anomalies, and broadcasts results. Anomaly data is automatically uploaded and linked to generate maintenance work orders, quickly triggering subsequent processes and shortening fault handling cycles. A closed-loop data system accumulates maintenance experience, providing data support for preventative maintenance optimization, effectively improving maintenance efficiency, ensuring operational safety, reducing operating costs, and driving the transformation of maintenance models towards intelligence and refinement.

[0148] The embodiments described above are merely preferred embodiments of this specification and are not intended to limit the scope of this specification. Any modifications and improvements made by those skilled in the art to the technical solutions of this specification without departing from the spirit of this specification should fall within the protection scope defined by the claims of this specification.

Claims

1. A device maintenance and management system based on a low-code platform and AI smart glasses, characterized in that, include: The low-code equipment maintenance work order configuration module is used to quickly configure and generate structured work order JSON templates containing step sequences, component types, business attributes, and strong control tags through a visual drag-and-drop interface and a pre-built component library; The smart glasses-side interactive guidance module is used to receive the work order JSON template and dynamically render the corresponding AR operation guidance overlaid on the real device screen according to the configuration information of the current work order step node. At the same time, it realizes hands-free operation and intelligent knowledge assistance through multimodal AI interaction. The equipment maintenance process digital acquisition module is used to automatically trigger the camera to collect multimedia evidence according to the acquisition type configured in the work order when the work order step node is configured as a strong control node, and to attach an immutable digital watermark consisting of NTP / BeiDou time synchronization timestamp, GPS / BeiDou dual-mode positioning coordinates and SHA-256 hash check value to the collected data. The maintenance data closed-loop management module is used to realize status-driven workflow updates and multi-dimensional statistical analysis based on accumulated maintenance data, and to automatically adjust the cycle of preventive maintenance plans according to historical failure frequency and average repair time. The maintenance data closed-loop management module includes: The status-driven workflow unit is used to trigger the backend event bus when the work order status changes, and automatically update the device's master data status. The multi-dimensional reporting unit is used to statistically analyze the average maintenance time of different personnel and different machine models and identify efficiency bottlenecks, and generate heat maps that show the equipment parts and time periods with high failure rates. The knowledge accumulation and reuse unit is used to transform historical anomaly records into micro-knowledge points and support precise retrieval. Based on historical fault frequency and average repair time statistics, it automatically adjusts the cycle of preventive maintenance plans, realizing the transformation from on-time maintenance to on-demand maintenance.

2. The equipment maintenance management system based on a low-code platform and AI smart glasses as described in claim 1, characterized in that, The low-code device maintenance work order configuration module includes: The visual interactive interface unit includes a canvas area, a component library, and an attribute configuration area. The canvas area displays the maintenance step sequence based on a flow layout and uses drag-and-drop component technology to adjust the order of step nodes. The component library provides pre-set standardized maintenance action components. The attribute configuration area is used to dynamically load configuration forms and verify the legality of configuration logic in real time when a canvas node is clicked. The template data structure and storage unit are defined by the TaskTemplate entity storage process on the backend, and the DOM nodes on the canvas are converted into structured data for persistence on the frontend through JSON serialization technology. The integrity verification unit is used to automatically traverse all nodes and check whether the required fields are complete before the template is saved or published.

3. The equipment maintenance management system based on a low-code platform and AI smart glasses as described in claim 1, characterized in that, The smart glasses-side interactive guidance module includes: The task precision distribution unit is used to push work orders to the smart glasses of designated maintenance personnel based on personnel skill matrix and geographical location, and supports two modes: order grabbing and order dispatching. It uses WebSocket to maintain real-time communication between the end and the cloud. The work order step-driven AR augmented reality guidance unit is used to parse the work order JSON template in real time, dynamically render AR operation guidance including arrows, highlight boxes, and parameter threshold prompts according to the type and strong control mark of the current work order step node, and use head posture tracking to ensure that the guidance information is always in the user's comfortable field of vision. After confirmation, it automatically advances to the next step node. The multimodal AI interaction unit integrates an automatic speech recognition engine to support hands-free voice commands; and an integrated text-to-speech engine to convert text operation steps in work order steps into speech broadcasts. The AI ​​knowledge base companion unit is used to automatically retrieve relevant technical documents, drawings, and historical maintenance cases from the knowledge base based on the equipment type bound to the current work order step node, and use them as context input for large language model reasoning.

4. The equipment maintenance management system based on a low-code platform and AI smart glasses according to claim 3, characterized in that, The accompanying unit of the AI ​​knowledge base includes: The knowledge base construction sub-unit is used to perform semantic slicing of maintenance manuals and equipment instruction manuals. The slices are converted into vectors through an embedding model, stored in a vector database, and an equipment type index is built. The retrieval subunit is used to construct a query vector by combining the user's question transcribed by automatic speech recognition with the current work order device context, perform an approximate nearest neighbor retrieval, and recall Top-K relevant document slices; The large language model context injection subunit is used to inject work order step information, equipment parameters and recall document slices into the large language model through structured prompt templates, generate troubleshooting suggestions, and broadcast them in text-to-speech format.

5. The equipment maintenance management system based on a low-code platform and AI smart glasses according to claim 4, characterized in that, The voice interaction processing methods of the smart glasses' interactive guidance module include: Input an audio stream and output unstructured text to achieve automatic speech recognition and transcription; The automatic speech recognition text, along with the device type and normal range parameters of the current work order step node, is injected into the large language model to extract the device name, abnormal value, and abnormal description. The extracted values ​​are compared with the thresholds configured in the work order step nodes, and the pass or fail status is automatically marked, with the judgment result announced by voice. If an anomaly is detected, the system will automatically trigger a camera to take a picture for evidence based on the strong control flag configuration of the work order steps, and bind it to the step result identifier.

6. The equipment maintenance management system based on a low-code platform and AI smart glasses according to claim 1, characterized in that, The digital data acquisition module for the equipment maintenance process includes: The multimedia evidence chain acquisition unit is used to automatically trigger camera acquisition based on the acquisition type configured in the work order when a work order step node is configured as a strong control node, ensuring a strict binding relationship between each piece of multimedia evidence and the corresponding standard operating procedure step. The temporal and spatial immutable digital watermarking unit is used to automatically add a standard UTC timestamp synchronized by NTP or BeiDou time service, a spatial positioning coordinate with a horizontal accuracy of better than 5 meters by GPS or BeiDou dual-mode positioning, and a file integrity check value calculated by SHA-256 hash algorithm to all collected data. The intelligent association and binding unit is used to manage unstructured data through backend object storage, automatically binding the collected files with work order identifiers and step result identifiers, and writing the hash value into the step result record to form a structured digital archive. The AI-powered anomaly intelligent classification unit is used to automatically identify the device status in uploaded photos using computer vision models, and to analyze text records using large language models, automatically extracting keywords and converting unstructured descriptions into standardized fault classification labels.

7. The equipment maintenance management system based on a low-code platform and AI smart glasses according to claim 6, characterized in that, The digital acquisition module for the equipment maintenance process also includes a target equipment identification unit, and the identification method of the target equipment identification unit includes: The smart glasses terminal continuously scans Bluetooth beacons in the work area, reads signal strength values, determines whether personnel have entered the 3-5 meter working range of a specific device, and automatically preloads the device data for that area. The glasses camera is activated to capture video streams. Key points are extracted using ORB or AKAZE feature extraction algorithms and compared with device reference images in the local database. When the matching degree exceeds a preset threshold, the target device identifier is locked and a guide icon is superimposed on the screen.

8. A device maintenance management method based on a low-code platform and AI smart glasses, the method being applicable to the system described in any one of claims 1-7, characterized in that, Includes the following steps: Step S1: The cloud server parses the work order JSON template generated by the low-code platform, associates the task tree with the corresponding subset of the exception status library, packages it and pushes it to the smart glasses. Step S2: When personnel enter the work area, the smart glasses sense the Bluetooth signal, and the camera detects the equipment nameplate or appearance features. The screen then displays the AR operation guide for the first step node according to the work order step configuration. Step S3, voice input and image verification are driven by the strong control mark configuration of the work order step node, and the collected content is bound to the step result identifier and spatiotemporal hash watermark for evidence storage; Step S4: If the step is determined to be abnormal, the system will automatically trigger the collection and uploading of the abnormal package to the cloud, and generate a maintenance work order to trigger the next level of work order process.

9. The device maintenance management method based on a low-code platform and AI smart glasses according to claim 8, characterized in that, In step S3, the smart glasses and the cloud use the JSON cloud-to-device communication protocol for data exchange. Downlink work order configuration protocol: The low-code platform compiles the standard operation process into a TaskTemplate object and issues it. The structure includes step identifier, component type, attribute information and strong control flag field; Uplink operation result protocol: After execution by the smart glasses, a structured result is returned, including step result identifier, judgment result, structured measurement value, standard text converted by automatic speech recognition, list of evidence files, NTP or BeiDou time stamp, GPS or BeiDou positioning coordinates, and SHA-256 file hash value field.

Images