A multi-agent collaborative operation and maintenance system and method for pharmaceutical packaging production lines

By using a multi-agent collaborative operation and maintenance system, the system monitors and provides early warnings of equipment status in real time, solving the problems of insufficient equipment status perception and fault early warning in pharmaceutical packaging production lines. It achieves equipment status visualization and accurate fault early warning, improving operation and maintenance efficiency and the reliability of maintenance records, and meeting GMP compliance requirements.

CN123070934APending Publication Date: 2026-11-10宋欣欣
View PDF 0 Cites 0 Cited by

Patent Information

Application Number
CN202611370110.6
Authority / Receiving Office
CN · China
Patent Type
Applications(China)
Current Assignee / Owner
Filing Date
2026-09-04
Publication Date
2026-11-10

AI Technical Summary

Technical Problem

The pharmaceutical packaging production line has weak equipment status perception capabilities, insufficient fault early warning capabilities, difficulty in accumulating maintenance experience, and unreliable quality traceability, resulting in serious losses from unplanned downtime and GMP compliance risks.

Method used

The system employs a multi-agent collaborative operation and maintenance system, comprising a perception layer, a data layer, and an analysis layer. It utilizes IoT sensors and an AI vision system to collect data in real time, combined with a big data storage module, a blockchain evidence storage module, and an operation and maintenance knowledge base. Through a fault prediction engine and professional intelligent agents, it monitors and warns about equipment status, generates diagnostic reports, and establishes reliable maintenance records.

Benefits of technology

It enables visualization of equipment status and accurate early warning of faults, improves operation and maintenance efficiency and the reliability and traceability of maintenance records, meets GMP compliance requirements, and reduces losses from unplanned downtime.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN123070934A_ABST
    Figure CN123070934A_ABST
Patent Text Reader

Abstract

This invention discloses a multi-agent collaborative operation and maintenance system and method for pharmaceutical packaging production lines. The system includes a perception layer, a data layer, and an analysis layer. The perception layer includes IoT sensors and an AI vision system for collecting operating parameter data and drug location data of pharmaceutical packaging production line equipment. The data layer includes a big data storage module, a blockchain evidence storage module, and an operation and maintenance knowledge base. The big data storage module stores the operating parameter data and drug location data transmitted from the perception layer. The blockchain evidence storage module stores key evidence data from the maintenance process. The operation and maintenance knowledge base stores equipment manuals, standard operating procedures (SOPs), and maintenance case documents. The beneficial effects of this invention are: improved equipment status visualization, fault early warning accuracy, and maintenance traceability reliability, meeting GMP compliance requirements.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] This invention relates to the field of packaging line operation detection system technology, specifically a multi-agent collaborative operation and maintenance system and method for pharmaceutical packaging production lines. Background Technology

[0002] Pharmaceutical manufacturing is a strategic industry vital to the national economy and people's livelihood. As the final stage of pharmaceutical production, the operational stability of pharmaceutical packaging equipment directly impacts drug quality and patient safety. Currently, the main problems in the operation and maintenance of pharmaceutical packaging production lines are as follows: 1. Weak equipment status perception: Current pharmaceutical packaging production lines mainly rely on manual inspections and regular maintenance. Key parameters such as equipment temperature and vibration cannot be obtained in real time. Often, the problem is only discovered when a fault occurs, resulting in serious losses from unplanned downtime. 2. Insufficient fault early warning capability: Existing systems generally adopt the "post-event maintenance" mode, lacking the ability to continuously analyze and predict trends of equipment operation data, making it impossible to identify abnormal signs of equipment in advance and difficult to achieve predictive maintenance; Third, difficulty in accumulating maintenance experience: key maintenance knowledge mainly relies on the experience accumulation of senior technicians, and there is a lack of a systematic knowledge management platform; Fourth, quality traceability is unreliable: More than 70% of enterprises still store their maintenance records in paper files or Excel spreadsheets. The data is easy to tamper with and difficult to trace. Once a quality accident occurs, a complete chain of evidence cannot be provided, which may lead to the risk of production stoppage, rectification, or even revocation of GMP certificate.

[0003] Therefore, in order to address the above-mentioned technical problems, it is necessary to provide a multi-agent collaborative operation and maintenance system and method for pharmaceutical packaging production lines. Summary of the Invention

[0004] To address the shortcomings of the existing technology, the purpose of this invention is to provide a multi-agent collaborative operation and maintenance system and method for pharmaceutical packaging production lines, thereby solving the problems mentioned in the background.

[0005] To solve the above-mentioned technical problems, the present invention adopts the following technical solution: This invention provides a multi-agent collaborative operation and maintenance system for pharmaceutical packaging production lines, comprising a perception layer, a data layer, and an analysis layer. The perception layer includes IoT sensors and an AI vision system for collecting operating parameter data and drug location data from pharmaceutical packaging production line equipment. The data layer includes a big data storage module, a blockchain evidence storage module, and an operation and maintenance knowledge base. The big data storage module stores the operating parameter data and drug location data transmitted from the perception layer; the blockchain evidence storage module stores key evidence data from the maintenance process; and the operation and maintenance knowledge base stores equipment manuals, standard operating procedures (SOPs), and maintenance case documents. The analysis layer includes a central intelligence module, a fault prediction engine, and specialized intelligent agents. The central intelligence module is communicatively connected to both the fault prediction engine and the specialized intelligent agents. The fault prediction engine is used to analyze historical data from the big data storage module... The system uses line data and real-time operational data to predict equipment failure risks and output early warning information. The intelligent brain central hub is used to call corresponding professional intelligent agents for collaborative analysis and generate diagnostic reports and treatment plans based on the early warning information. The application layer includes a digital twin visualization screen, an intelligent operation and maintenance APP, and an AI voice assistant. The digital twin visualization screen is communicatively connected to the perception layer and the data layer, and is used to map the movement status of physical equipment on the production line in real time based on the operational parameter data and drug location data. The intelligent operation and maintenance APP is communicatively connected to the analysis layer, and is used to receive and display diagnostic reports and treatment plans. The AI ​​voice assistant is integrated into the intelligent operation and maintenance APP, and is used to receive natural language voice commands and call corresponding professional intelligent agents to perform corresponding operations. The blockchain evidence storage module is communicatively connected to the application layer, and is used to store the entire maintenance process data.

[0006] In one or more embodiments of the present invention, the Internet of Things sensor includes a temperature and humidity sensor, a vibration transmitter and an industrial noise transmitter. The vibration transmitter and the industrial noise transmitter form a data acquisition network through a data acquisition unit and a LoRa gateway. The LoRa gateway encrypts and transmits the acquired operating parameter data to the data layer. The AI ​​vision system includes an industrial area scan camera and an RGB-D depth camera. The RGB-D depth camera is mounted on the end of a robotic arm and is used to acquire images of pharmaceutical packaging.

[0007] In one or more embodiments of the present invention, the big data storage module is built on a Hadoop distributed environment and is used to store massive amounts of device operation data collected by the perception layer; The blockchain evidence storage module is built on the FISCOBCOS consortium blockchain, and the maintenance process data includes equipment number, fault cause, maintenance method, maintenance result, operator and timestamp; The operation and maintenance knowledge base is used to store unstructured equipment manuals, SOP operating procedures, and maintenance case documents, providing reasoning basis for the professional intelligent agent.

[0008] In one or more embodiments of the present invention, the fault prediction engine is used to read no less than 30 days of historical operating data from the big data storage module, integrate equipment mechanism knowledge, train a time series prediction model, perform predictive analysis on equipment operating trends, and output early warning information 24 hours in advance. The specialized intelligent agents include a data credibility inspection intelligent agent, an equipment status inspection intelligent agent, an anomaly evidence interpretation intelligent agent, an operation and maintenance knowledge Q&A intelligent agent, and a management situation report intelligent agent.

[0009] In one or more embodiments of the present invention, the digital twin visualization screen is built based on the Unity3D engine and is used to realize 1:1 digital twin model mapping and virtual-real linkage display of physical equipment on the production line; The AI ​​voice assistant is used to receive natural language questions and answers and voice commands from operations and maintenance personnel, call the operations and maintenance knowledge question and answer intelligent agent to retrieve matching solutions from the operations and maintenance knowledge base, and output the search results to the intelligent operations and maintenance APP.

[0010] A multi-agent collaborative operation and maintenance method for a pharmaceutical packaging production line includes the following steps: Step S1: Real-time collection of operating parameter data of pharmaceutical packaging production line equipment through IoT sensors in the perception layer, collection of drug location data through AI vision system in the perception layer, and encryption and transmission of the operating parameter data and drug location data to the data layer via LORA gateway; Step S2: Construct a digital twin model of the production line based on the Unity3D engine, and map the motion status of the physical equipment on the production line in real time through the digital twin visualization screen at the application layer; Step S3: The fault prediction engine of the analysis layer reads the historical and real-time operating data stored in the big data storage module of the data layer, performs fusion analysis on the historical and real-time operating data, predicts equipment failure risks, and outputs early warning information. In step S4, the intelligent brain center of the analysis layer receives the warning information output in step S3, and calls the corresponding professional intelligent agent to perform collaborative analysis based on the warning information. The professional intelligent agent reads the equipment manual, SOP operation procedure and maintenance case document from the operation and maintenance knowledge base of the data layer as the basis for reasoning, and generates a diagnostic report and treatment plan. Step S5: The application layer's intelligent operation and maintenance APP receives and displays the diagnostic report and handling plan generated in step S4. The operation and maintenance personnel confirm and execute the handling plan through the intelligent operation and maintenance APP, and call the execution intelligent agent to complete the equipment operation according to the confirmation instruction. Step S6: The maintenance data from the entire process from steps S1 to S5 is automatically uploaded to the blockchain evidence storage module of the data layer for on-chain evidence storage, forming an immutable and trustworthy maintenance file.

[0011] In one or more embodiments of the present invention, in step S1, the operating parameter data includes equipment temperature data, vibration data, noise data, and current data; The AI ​​vision system uses the YOLO target detection algorithm to identify the location of the drug and output pixel coordinates. The pixel coordinates are then converted into three-dimensional grasping coordinates in the robotic arm's base coordinate system using a hand-eye calibration matrix.

[0012] In one or more embodiments of the present invention, in step S3, the fault prediction engine reads no less than 30 days of historical operating data from the big data storage module, integrates equipment mechanism knowledge, trains a time series prediction model, and performs predictive analysis on the equipment operating trend. When the time-series prediction model detects that the equipment operating parameters show an abnormal trend and the prediction reaches the warning value, the fault prediction engine automatically generates the warning information, which includes the expected time to reach the warning value.

[0013] In one or more embodiments of the present invention, in step S4, the intelligent brain central system calls the abnormal evidence interpretation intelligent agent, which reads real-time operating data, compares historical operating data, and matches the equipment manuals, SOP operating procedures and maintenance case documents in the operation and maintenance knowledge base, analyzes the cause of the abnormality and generates a diagnostic report. The intelligent brain central system calls the device status inspection intelligent agent, which evaluates the device health based on the operating parameter data and outputs a health score. The intelligent brain central unit integrates the diagnostic report and health score and pushes them to the intelligent operation and maintenance APP.

[0014] In one or more embodiments of the present invention, in step S5, the operation and maintenance personnel input natural language voice commands through the AI ​​voice assistant in the intelligent operation and maintenance APP. The AI ​​voice assistant calls the operation and maintenance knowledge question and answer intelligence to retrieve the handling plan corresponding to the diagnostic report from the operation and maintenance knowledge base, and outputs the retrieval results to the intelligent operation and maintenance APP for the operation and maintenance personnel to confirm.

[0015] The beneficial effects of this invention are as follows: 1. Visualized Equipment Status: Achieve 1:1 virtual-physical linkage of the production line through a digital twin screen, making equipment health and operating parameters clear at a glance, completely solving the problem of "invisible equipment status"; 2. Accurate fault early warning: Based on the collaborative architecture of "intelligent brain center + multi-intelligent agent", the equipment status inspection intelligent agent continuously monitors the health of the equipment, and the fault prediction engine driven by the Qianwen big model predicts equipment risks 24 hours in advance, with an early warning accuracy rate of 91.2%, effectively avoiding unplanned downtime losses. 3. Significantly improved operation and maintenance efficiency: Troubleshooting time reduced by 65%, repair response time reduced by 60%, and first-time repair rate increased by 35%; 4. Digital transfer of maintenance experience: Based on the natural language question-and-answer capabilities of the maintenance knowledge question-and-answer intelligent agent, it solves the problem of lack of maintenance experience for new employees; 5. Reliable and traceable maintenance records: Based on the data verification of the intelligent agent for data credibility inspection and the full-process on-chain storage of evidence on the FISCOBCOS consortium blockchain, GMP compliance requirements are met, and a complete chain of evidence is provided for quality audits; 6. Intelligent Agent Collaborative Closed-Loop Operation and Maintenance: Construct a six-in-one closed loop of "perception-mapping-analysis-decision-execution-evidence storage", with the central intelligent brain uniformly scheduling multiple professional intelligent agents to work collaboratively, realizing a paradigm leap from "passive maintenance" to "predictive maintenance" and then to "trustworthy collaborative maintenance". Attached Figure Description

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

[0017] Figure 1 This is an architecture diagram of a multi-agent collaborative operation and maintenance system for a pharmaceutical packaging production line according to one embodiment of the present invention; Figure 2 This is a flowchart of a multi-agent collaborative operation and maintenance method for a pharmaceutical packaging production line according to an embodiment of the present invention. Detailed Implementation

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

[0019] Example 1: like Figure 1 and Figure 2As shown in one embodiment of the present invention, a multi-agent collaborative operation and maintenance system and method for a pharmaceutical packaging production line is provided. The system adopts a four-layer architecture design, consisting of a perception layer, a data layer, an analysis layer, and an application layer from bottom to top. Data interaction between the layers is achieved through industrial Ethernet and wireless communication networks.

[0020] The sensing layer is deployed on the pharmaceutical packaging production line and consists of two main parts: IoT sensors and an AI vision system. The IoT sensors specifically include temperature and humidity sensors, LORA vibration transmitters, and Type 485 industrial noise transmitters. LORA vibration transmitters are installed at key equipment on the production line, such as the front-end belt conveyor, pillow bag machine, rear-end belt conveyor, and the bearing housings, motor bases, and reducer housings of the Yuejiang Nova2 six-axis robotic arm (which has 6 movable joints, a maximum load of 10kg, and a maximum speed of 1.6m / s), to collect three-axis vibration data (including vibration acceleration, vibration velocity, and vibration displacement) during equipment operation. Temperature and humidity sensors are deployed at key locations in the production line environment to collect ambient temperature and humidity data. Type 485 industrial noise transmitters are deployed near the robotic arm joint motors and conveyor drive motors to collect sound pressure level and spectrum data of equipment operating noise. LORA vibration transmitters and Type 485 industrial noise transmitters collect data via the LORA485 data acquisition unit, and then transmit the collected operating parameter data (including temperature, vibration, noise, and current data) to the data layer in an encrypted manner via the LORA gateway. The LORA gateway uses LoRa modulation technology, operates in the 470MHz frequency band, has a communication distance of over 2km, and features spread spectrum communication and AES-128 encrypted transmission capabilities, ensuring data security and anti-interference during wireless transmission.

[0021] The AI ​​vision system comprises an industrial area scan camera and an RGB-D depth camera. The RGB-D depth camera is mounted at the end effector of the Yuejiang Nova2 six-axis robotic arm, moving with the arm to acquire real-time image data of the pharmaceutical packaging. The industrial area scan camera is fixedly mounted above key workstations on the production line for monitoring the packaging status from a global perspective. The AI ​​vision system uses the YOLO object detection algorithm to process the acquired pharmaceutical packaging images in real time, identifying the position of the packaging and outputting its pixel coordinates in the pixel coordinate system. Then, through a pre-calibrated hand-eye calibration matrix (i.e., the coordinate transformation matrix between the camera coordinate system and the robotic arm's base coordinate system), the pixel coordinates are converted into 3D grasping coordinates in the robotic arm's base coordinate system, with positioning accuracy controlled within ±0.1mm. Data such as the pharmaceutical location coordinates, recognition confidence, and grasping success rate identified by the vision system, along with the robotic arm's motion data, are simultaneously uploaded to the big data platform and the digital twin system, providing the analysis layer with "visible" decision-making support and serving as a trigger signal source for the virtual-real linkage of the digital twin.

[0022] On the group, a three-tiered security system is constructed, which includes three main modules: big data storage module, blockchain evidence storage module, and operation and maintenance knowledge base.

[0023] The big data storage module is built on a Hadoop distributed environment and deployed in a 3-node high-availability cluster. It stores massive amounts of equipment operation data collected by the sensing layer, including equipment temperature time-series data, vibration waveform data, noise spectrum data, current load data, and drug location data. The module uses the HDFS distributed file system for persistent data storage and is configured with an HBase columnar database for efficient reading and writing of time-series data. Data retention is at least 3 years, meeting GMP requirements for data integrity and traceability.

[0024] The blockchain evidence storage module is built on the FISCOBCOS consortium blockchain, configuring FISCOBCOS consortium blockchain nodes on a server cluster to establish an enterprise-level consortium blockchain network. The consortium blockchain network adopts the PBFT consensus mechanism, with participating nodes including production line maintenance nodes, quality management nodes, and regulatory audit nodes. Each node achieves access control through certificate authentication. The blockchain evidence storage module stores key evidentiary data of the maintenance process, specifically including full-process maintenance data: equipment number, cause of failure, maintenance method, maintenance result, operator, timestamp, and hash values ​​of on-site photos. After each maintenance is completed, the system automatically packages the above data into a transaction, which is then confirmed by consensus among all consortium blockchain nodes and written to the blockchain, forming an immutable and trustworthy maintenance record. The blockchain evidence storage module also provides a data traceability interface, supporting the querying of complete maintenance history records by equipment number or time range, providing a complete chain of evidence for GMP compliance audits.

[0025] The operations and maintenance (O&M) knowledge base stores unstructured equipment manuals, standard operating procedures (SOPs), and maintenance case documents. The knowledge base uses Elasticsearch as its search engine, performing full-text indexing and vectorized embedding of documents, supporting both semantic and keyword search modes. Equipment manuals include installation manuals, operation manuals, and maintenance manuals for each piece of equipment; SOPs include standardized operating instructions, safe operating procedures, and equipment inspection standards; maintenance case documents include the diagnostic process, handling solutions, repair results, and experience summaries of historical fault cases. The O&M knowledge base provides reasoning basis for various professional agents in the analysis layer. When an agent needs to query the fault handling methods for specific equipment, it can obtain relevant document fragments through the knowledge base's search interface.

[0026] The analysis layer is the core intelligent decision-making layer of the system, deployed on high-performance computing nodes of the server cluster. It includes the intelligent brain center, fault prediction engine, data credibility inspection intelligent agent, equipment status inspection intelligent agent, abnormal evidence interpretation intelligent agent, operation and maintenance knowledge Q&A intelligent agent, and management status reporting intelligent agent.

[0027] The central AI system is responsible for task scheduling and system security control. It employs a microservice architecture and features functions such as task decomposition and distribution, agent lifecycle management, cross-agent communication coordination, and system security control. The central AI system has a built-in task orchestration engine. Upon receiving warning information from the fault prediction engine, it automatically generates a task execution plan based on the warning type and urgency, decomposing complex tasks into several sub-tasks and distributing them to the corresponding specialized agents according to a preset collaboration process. The central AI system also maintains the registration information, health status, and load of each agent, enabling dynamic scheduling and failover of agents.

[0028] The fault prediction engine is a fault prediction engine driven by the Qianwen big data model and deployed on GPU computing nodes. The fault prediction engine first reads no less than 30 days of historical operating data (including multi-dimensional time-series data such as temperature, vibration, noise, and current) from the big data storage module, integrates equipment mechanism knowledge (including prior knowledge such as equipment rated parameters, failure modes, and life curves), and uses a time-series prediction model for model training.

[0029] After training, the fault prediction engine reads the latest operational data uploaded by the perception layer in real time, integrates and analyzes it with historical data, predicts the future operating trend of the equipment, and outputs early warning information 24 hours in advance. When the prediction model detects that the operating parameters of a certain piece of equipment show an abnormal trend (such as a continuous rise in the temperature of the robotic arm or a gradual increase in the amplitude of the vibration characteristic frequency) and is expected to reach the warning value within 8 to 24 hours, the fault prediction engine automatically generates early warning information, which includes the expected time of reaching the warning value, the type of abnormal parameter, and the risk level.

[0030] The data credibility inspection intelligent agent is responsible for verifying the quality of the data uploaded by the perception layer, including detecting data missing rate, outliers, data latency and communication interruption, scoring the data credibility, and feeding back the verification results to the intelligent brain central hub and blockchain storage module to ensure the authenticity and integrity of the data on the chain.

[0031] The intelligent agent for equipment status inspection is responsible for continuously monitoring the health of each device. Based on the multi-dimensional operating parameter data collected by the perception layer, combined with the device operating threshold and health assessment model, it calculates the health score of each device in real time (the score range is 0 to 100 points, and the lower the score, the worse the health status), and pushes the assessment results to the digital twin screen for visualization.

[0032] The abnormal evidence interpretation agent is responsible for conducting in-depth analysis of the early warning information output by the fault prediction engine. The specific method is as follows: read the real-time operating data of the abnormal device, compare it with the historical normal operating data of the device, match the device manual and maintenance cases in the operation and maintenance knowledge base, analyze the root cause of the abnormality, and generate a diagnostic report that includes fault cause analysis, impact scope assessment and handling suggestions.

[0033] The operation and maintenance knowledge Q&A agent is responsible for responding to the natural language queries of operation and maintenance personnel. Based on the equipment manuals, SOP operating procedures and maintenance case documents in the operation and maintenance knowledge base, combined with the real-time operating data of the current equipment, it generates personalized diagnostic reports and handling suggestions.

[0034] The management status report agent is responsible for regularly summarizing production line operation and maintenance data and generating management status reports, including equipment health trend analysis, fault statistical analysis, maintenance work order completion rate, predictive maintenance effect evaluation, etc., to provide decision support for management.

[0035] The application layer includes the Ling Shu digital twin visualization screen, the intelligent operation and maintenance APP, and the "Ling Shu" AI voice assistant.

[0036] The Ling Shu digital twin visualization screen is built on the Unity3D engine. It utilizes 3D modeling technology to construct a 1:1 digital twin model of the pharmaceutical packaging production line, including virtual mappings of all physical equipment such as the front-end belt conveyor, pillow bag machine, back-end belt conveyor, and the Yuejiang Nova2 six-axis robotic arm. The digital twin screen receives real-time operational parameter data and drug location data uploaded from the perception layer, driving the 3D model to move synchronously with the physical equipment, achieving a synchronized display between the virtual and real worlds. Specifically, when the robotic arm moves in the physical world, the virtual robotic arm in the digital twin model synchronously executes the same trajectory; when parameters such as equipment temperature and vibration change, the corresponding equipment's color and labeling information in the digital twin model are updated synchronously, intuitively displaying the equipment's health status. The digital twin screen also integrates data visualization components, displaying real-time operating parameters and historical trends of the equipment in the form of dashboards, trend charts, and heat maps.

[0037] The intelligent operations and maintenance (O&M) app is an application that runs on mobile terminals (such as smartphones and tablets). It communicates with the analytics layer to receive and display diagnostic reports and handling solutions. After logging into the system through the app, O&M personnel can view maintenance work orders automatically dispatched by the system. The work order content includes the equipment name, fault description, urgency level, suggested handling solution, and estimated processing time. The app supports full-process management of work order reception, execution, feedback, and closure.

[0038] The "Ling Shu" AI voice assistant is integrated into the intelligent operation and maintenance (O&M) app. Based on speech recognition, natural language understanding, and speech synthesis technologies, it supports natural language question-and-answer and voice command execution. O&M personnel can use the voice assistant in the app to ask questions in natural language about equipment status, troubleshooting methods, and other related issues. The voice assistant automatically invokes an O&M knowledge-based intelligent system to retrieve matching solutions from the O&M knowledge base, combines real-time equipment data, and outputs a personalized diagnostic report, which is then broadcast to the O&M personnel via speech synthesis. O&M personnel can also use voice commands to confirm work orders and issue equipment operation instructions.

[0039] The perception layer collects production line operation parameters and drug location data through the LoRa gateway and AI vision system, which are then encrypted and transmitted to the big data storage module in the data layer for persistent storage. The fault prediction engine in the analysis layer reads historical and real-time operation data from the big data storage module for fusion analysis and outputs early warning information to the central AI system. Based on the early warning information, the central AI system calls on corresponding specialized intelligent agents for collaborative analysis. Each specialized intelligent agent reads equipment manuals, SOP operating procedures, and maintenance case documents from the maintenance knowledge base as reasoning basis to generate diagnostic reports and treatment plans, which are then pushed to the intelligent maintenance APP in the application layer. After receiving work orders through the APP, maintenance personnel can use the AI ​​voice assistant for natural language interaction to query and confirm instructions. The system automatically issues the confirmation instructions to the execution layer equipment to complete the operation. All maintenance process data is automatically uploaded to the blockchain evidence storage module for on-chain evidence storage and simultaneously synchronized to a digital twin screen for visualization, forming a complete data closed loop from perception to evidence storage.

[0040] Example 2: like Figure 2 As shown, this embodiment provides a multi-agent collaborative operation and maintenance method for pharmaceutical packaging production lines, which adopts a six-in-one digital diagnosis and treatment closed loop of "perception-mapping-analysis-decision-execution-evidence storage", specifically including the following steps.

[0041] Step S1: Real-time collection of operating parameter data of pharmaceutical packaging production line equipment through IoT sensors in the perception layer, collection of drug location data through AI vision system in the perception layer, and encryption transmission of operating parameter data and drug location data to the data layer via LORA gateway.

[0042] Specifically, during production line operation, LORA vibration transmitters deployed at measuring points such as the robotic arm bearing housing and motor base continuously collect triaxial vibration data at a sampling frequency of no less than 1 kHz; 485-type industrial noise transmitters deployed near the robotic arm joint motors collect sound pressure level data at a sampling frequency of no less than 20 kHz; temperature and humidity sensors deployed in the workshop environment collect ambient temperature and humidity data at a sampling frequency of 1 Hz; and the robotic arm controller transmits current and torque data of each joint motor in real time via the EtherCAT bus. All the above data is collected by the LORA485 data acquisition unit, encrypted by the LORA gateway using the AES-128 encryption algorithm, and then uploaded to the big data storage module of the data layer via the MQTT protocol.

[0043] Meanwhile, an RGB-D depth camera mounted on the end effector of the robotic arm acquires RGB color and depth images of the drug packaging in real time at a frame rate of no less than 30fps. The AI ​​vision system uses the YOLOv8 object detection algorithm to perform real-time inference on each frame, identifying the position of the drug packaging and outputting its bounding box coordinates in the pixel coordinate system. Subsequently, through a pre-calibrated hand-eye calibration matrix, the pixel coordinates (u, v) combined with the depth value d are converted into three-dimensional grasping coordinates (X, Y, Z) in the robotic arm's base coordinate system, with positioning accuracy controlled within ±0.1mm. The drug position coordinates identified by the vision system, the recognition confidence (ranging from 0 to 1, with detection results below 0.6 marked as low confidence and triggering secondary recognition), and the grasping success rate are simultaneously uploaded to the big data storage module and the digital twin system.

[0044] Step S2: Based on the running parameter data and drug location data transmitted to the data layer in step S1, a 1:1 digital twin model of the production line is constructed using the Unity3D engine. The motion status of the physical equipment on the production line is then mapped in real time through the digital twin visualization screen in the application layer.

[0045] Specifically, the Unity3D engine pre-built 3D models of all physical equipment in the pharmaceutical packaging production line, including the front-end belt conveyor, pillow bag machine, rear-end belt conveyor, and the Yuejiang Nova2 six-axis robotic arm, with a model accuracy reaching LOD3 level. The digital twin system subscribes to the operational parameter data uploaded by the perception layer in real time via the OPCUA protocol, including the motion and status parameters of each device. When new data is received, the Unity3D engine drives the corresponding components in the 3D model to perform corresponding pose transformations. For example, when joint 1 of the physical robotic arm rotates 15 degrees, joint 1 of the virtual robotic arm rotates 15 degrees synchronously; when the speed of the physical conveyor changes from 1.0 m / s to 1.2 m / s, the conveyor belt texture of the virtual conveyor accelerates synchronously. At the same time, the digital twin screen intuitively displays the health status of the equipment using color mapping: green for health scores above 80, yellow for 60 to 80, and red for below 60. Data labels pop up above the equipment displaying current key parameter values ​​such as temperature and vibration. The drug location data identified by the vision system is also synchronously mapped to the digital twin screen, displaying the three-dimensional markings of the drug packaging at the corresponding location in the virtual space, realizing a visual presentation of "dynamic recognition → three-dimensional mapping → virtual-real linkage".

[0046] Step S3: The fault prediction engine in the analysis layer reads the historical operating data stored in the big data storage module of the data layer and the real-time operating data collected in step S1, performs fusion analysis on the historical operating data and real-time operating data, predicts the equipment failure risk, and outputs early warning information.

[0047] Specifically, the fault prediction engine first reads at least 30 days of historical operating data from HBase, the big data storage module, including time-series temperature data of each joint of the robotic arm, spectral characteristics of triaxial vibration data, and RMS and peak values ​​of motor current. Then, the engine integrates prior information such as the design life of the robotic arm reducer, the rated load and failure cycle of the bearings, and the insulation class and temperature rise limits of the motor, and trains it using a time-series prediction model based on the Transformer architecture. During model training, historical time-series data is used as input, and the future state of the equipment (normal / abnormal) is used as a label. A multi-head self-attention mechanism is employed to capture long-range dependencies in the time-series data, learning the normal variation patterns and abnormal evolution laws of the equipment's operating parameters.

[0048] After model training is complete, the real-time inference phase begins. The fault prediction engine reads real-time operational data using a sliding time window, inputs it into the trained time-series prediction model, and outputs the predicted values ​​and confidence intervals of the operating parameters of each device for the next 24 hours. When the model detects an abnormal trend in the operating parameters of a device, such as the robotic arm joint temperature rising continuously from the normal value of 38℃ to 58.2℃ over two consecutive hours (while the normal operating temperature range for this device is 20℃ to 40℃), the model predicts that this temperature will reach the warning value within 8 to 24 hours, and the fault prediction engine automatically generates a warning message. This warning message includes: device name, device number, abnormal parameter type, current value, predicted time of reaching the warning value, risk level, and suggested handling measures.

[0049] In step S4, the intelligent brain center of the analysis layer receives the warning information output in step S3, and calls the corresponding professional intelligent agent to perform collaborative analysis based on the warning information. The professional intelligent agent reads the equipment manual, SOP operation procedure and maintenance case document from the operation and maintenance knowledge base of the data layer as the basis for reasoning, and generates a diagnostic report and treatment plan.

[0050] Specifically, after receiving a warning about abnormal temperature at joint 2 of the robotic arm, the central AI system initiates a task orchestration process, sequentially calling the following specialized AI agents according to preset collaboration rules: First, the central processing unit invokes the anomalous evidence interpretation agent. The anomalous evidence interpretation agent executes the following analysis process: Step S4.1: Read the temperature time series data of the robotic arm joint 2 for the most recent 7 days from the big data storage module and extract the temperature change trend features.

[0051] Step S4.2: Retrieve the equipment manual for this model of robotic arm from the maintenance knowledge base to obtain its normal operating temperature range, alarm temperature threshold, and shutdown protection temperature threshold.

[0052] Step S4.3: Search the historical maintenance cases of the equipment in the operation and maintenance knowledge base to find if there are similar temperature anomaly records and their handling solutions.

[0053] Step S4.4: Based on the above information, generate a diagnostic report, which includes: root cause analysis of the abnormality, assessment of the scope of impact, and treatment recommendations.

[0054] Subsequently, the central AI system invokes the equipment status inspection intelligent agent. Based on multi-dimensional operational parameter data, the agent comprehensively assesses the overall health of the robotic arm. The assessment model employs a weighted scoring method, with the weights of each parameter pre-set according to equipment mechanics. Scores are calculated for each parameter, then weighted and summed to output a health score. The calculation shows the robotic arm's current health score is 65, placing it in a "sub-healthy" state. The intelligent agent then pushes the score result to the central AI system.

[0055] Finally, the central processing unit integrates the diagnostic report generated by the abnormal evidence interpretation agent and the health score output by the equipment status inspection agent to generate a complete handling plan, which is then pushed to the intelligent operation and maintenance APP at the application layer. The handling plan is presented in the form of a work order, including equipment information, fault description, urgency level ("urgent"), suggested handling measures, estimated processing time, and a list of required spare parts.

[0056] In step S5, the application layer's intelligent operation and maintenance APP receives and displays the diagnostic report and handling plan generated in step S4. The operation and maintenance personnel confirm and execute the handling plan through the intelligent operation and maintenance APP, and the system calls the execution intelligent agent to complete the equipment operation according to the confirmation instruction.

[0057] Specifically, maintenance personnel receive automatically dispatched work order notifications via a smart maintenance app on their mobile phones. After viewing the work order details and confirming the diagnostic report and handling plan, the personnel click the "Ling Shu Voice Assistant" icon in the app and speak the voice command "Ling Shu, confirm execution plan one" into the phone's microphone. The AI ​​voice assistant converts the maintenance personnel's speech into text commands using a speech recognition engine, analyzes the command intent using a natural language understanding engine, and sends the analysis results to the AI ​​brain center. After verifying the maintenance personnel's permissions, the AI ​​brain center invokes the execution agent. The execution agent sends speed adjustment commands to the robotic arm controller via the OPCUA protocol, gradually reducing the overall operating speed of the robotic arm from the current 100% to 60%. Upon receiving the command, the robotic arm controller executes the speed reduction operation and transmits the execution results back to the system in real time.

[0058] After the deceleration operation was executed, the sensing layer continuously collected temperature data from joint 2 of the robotic arm and uploaded it to the system. The system detected a gradual decrease in temperature, and the temperature change trend was displayed in real time on the digital twin screen. The color of the robotic arm model on the screen gradually transitioned from a red alarm state to a yellow warning state, allowing maintenance personnel to visually monitor the temperature drop process through the screen.

[0059] Step S6: The maintenance data from the entire process from steps S1 to S5 is automatically uploaded to the blockchain evidence storage module in the data layer for on-chain evidence storage, forming an immutable and trustworthy maintenance file.

[0060] Specifically, after maintenance is completed, the system automatically collects key data from the entire maintenance process, including: equipment number, equipment name, fault symptoms, fault cause, handling plan, execution result, operator's employee number and name, operation time, work order number, and the SHA-256 hash value of on-site photos. This data is packaged according to the FISCOBCOS consortium blockchain transaction format, signed by the maintenance nodes, and broadcast to the consortium blockchain network. After consensus nodes reach an agreement through the PBFT consensus mechanism, the transaction is written to the blockchain, generating a unique transaction hash as proof of ownership. After proof of ownership is established, the system automatically generates a traceability QR code based on the transaction hash and equipment number. This QR code is printed and affixed to the equipment maintenance record sheet or entered into the electronic archive. Any authorized personnel can scan this QR code to query the complete maintenance history of the equipment from the blockchain, including the time of each fault occurrence, diagnostic process, handling measures, execution result, and operator, ensuring the authenticity, completeness, and immutability of the maintenance records, meeting GMP compliance requirements for "data integrity" and "traceability" of equipment maintenance records. The entire process data of each maintenance is simultaneously archived to the big data storage module, serving as the basic data source for subsequent fault prediction model training and knowledge base updates, thus realizing a complete data closed loop from early warning to evidence storage.

[0061] Obviously, the above-described embodiments are only used to illustrate the technical solutions of the present invention, and not to limit them. Although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some or all of the technical features therein; and these modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the scope of the technical solutions of the embodiments of the present invention. Thus, if these modifications and variations of the present invention fall within the scope of the claims of the present invention and their equivalents, the present invention also intends to include these modifications and variations.

Claims

1. A multi-agent collaborative operation and maintenance system for a pharmaceutical packaging production line, characterized in that, include: The perception layer includes IoT sensors and an AI vision system, used to collect operating parameter data and drug location data of pharmaceutical packaging production line equipment. The data layer includes a big data storage module, a blockchain evidence storage module, and an operation and maintenance knowledge base. The big data storage module is used to store the operating parameter data and drug location data transmitted by the perception layer. The blockchain evidence storage module is used to store key evidence data of the maintenance process. The operation and maintenance knowledge base is used to store equipment manuals, SOP operating procedures, and maintenance case documents. The analysis layer includes a central intelligent brain, a fault prediction engine, and a professional intelligent agent. The central intelligent brain is communicatively connected to the fault prediction engine and the professional intelligent agent. The fault prediction engine is used to predict equipment failure risks based on historical and real-time operating data in the big data storage module and output early warning information. The central intelligent brain is used to call the corresponding professional intelligent agent to perform collaborative analysis based on the early warning information and generate diagnostic reports and handling plans. The application layer includes a digital twin visualization screen, an intelligent operation and maintenance APP, and an AI voice assistant. The digital twin visualization screen is communicatively connected to the perception layer and the data layer, and is used to map the motion status of the physical equipment on the production line in real time based on the operating parameter data and drug location data. The intelligent operation and maintenance APP is communicatively connected to the analysis layer, and is used to receive and display diagnostic reports and treatment plans. The AI ​​voice assistant is integrated into the intelligent operation and maintenance APP, and is used to receive natural language voice commands and call the corresponding professional intelligent agents to perform corresponding operations. The blockchain evidence storage module communicates with the application layer and is used to store data from the entire maintenance process.

2. The multi-agent collaborative operation and maintenance system for a pharmaceutical packaging production line as described in claim 1, characterized in that, The IoT sensors include a temperature and humidity sensor, a vibration transmitter, and an industrial noise transmitter. The vibration transmitter and the industrial noise transmitter form a data acquisition network through a data acquisition unit and a LoRa gateway. The LoRa gateway encrypts and transmits the acquired operating parameter data to the data layer. The AI ​​vision system includes an industrial area scan camera and an RGB-D depth camera. The RGB-D depth camera is mounted on the end of a robotic arm and is used to acquire images of pharmaceutical packaging.

3. The multi-agent collaborative operation and maintenance system for a pharmaceutical packaging production line as described in claim 1, characterized in that, The big data storage module is built on a Hadoop distributed environment and is used to store the massive amounts of device operation data collected by the perception layer. The blockchain evidence storage module is built on the FISCOBCOS consortium blockchain, and the maintenance process data includes equipment number, fault cause, maintenance method, maintenance result, operator and timestamp; The operation and maintenance knowledge base is used to store unstructured equipment manuals, SOP operating procedures, and maintenance case documents, providing reasoning basis for the professional intelligent agent.

4. The multi-agent collaborative operation and maintenance system for a pharmaceutical packaging production line as described in claim 1, characterized in that, The fault prediction engine is used to read no less than 30 days of historical operating data from the big data storage module, integrate equipment mechanism knowledge, train a time series prediction model, predict and analyze the equipment operation trend, and output early warning information 24 hours in advance. The specialized intelligent agents include a data credibility inspection intelligent agent, an equipment status inspection intelligent agent, an anomaly evidence interpretation intelligent agent, an operation and maintenance knowledge Q&A intelligent agent, and a management situation report intelligent agent.

5. The multi-agent collaborative operation and maintenance system for a pharmaceutical packaging production line as described in claim 1, characterized in that, The digital twin visualization screen is built on the Unity3D engine and is used to realize 1:1 digital twin model mapping and virtual-real linkage display of physical equipment on the production line; The AI ​​voice assistant is used to receive natural language questions and answers and voice commands from operations and maintenance personnel, call the operations and maintenance knowledge question and answer intelligent agent to retrieve matching solutions from the operations and maintenance knowledge base, and output the search results to the intelligent operations and maintenance APP.

6. A multi-agent collaborative operation and maintenance method for a pharmaceutical packaging production line, applied to the multi-agent collaborative operation and maintenance system for a pharmaceutical packaging production line as described in any one of claims 1 to 5, characterized in that, Includes the following steps: Step S1: Real-time collection of operating parameter data of pharmaceutical packaging production line equipment through IoT sensors in the perception layer, collection of drug location data through AI vision system in the perception layer, and encryption and transmission of the operating parameter data and drug location data to the data layer via LORA gateway; Step S2: Construct a digital twin model of the production line based on the Unity3D engine, and map the motion status of the physical equipment on the production line in real time through the digital twin visualization screen at the application layer; Step S3: The fault prediction engine of the analysis layer reads the historical and real-time operating data stored in the big data storage module of the data layer, performs fusion analysis on the historical and real-time operating data, predicts equipment failure risks, and outputs early warning information. In step S4, the intelligent brain center of the analysis layer receives the warning information output in step S3, and calls the corresponding professional intelligent agent to perform collaborative analysis based on the warning information. The professional intelligent agent reads the equipment manual, SOP operation procedure and maintenance case document from the operation and maintenance knowledge base of the data layer as the basis for reasoning, and generates a diagnostic report and treatment plan. Step S5: The application layer's intelligent operation and maintenance APP receives and displays the diagnostic report and handling plan generated in step S4. The operation and maintenance personnel confirm and execute the handling plan through the intelligent operation and maintenance APP, and call the execution intelligent agent to complete the equipment operation according to the confirmation instruction. Step S6: The maintenance data from the entire process from steps S1 to S5 is automatically uploaded to the blockchain evidence storage module of the data layer for on-chain evidence storage, forming an immutable and trustworthy maintenance file.

7. The multi-agent collaborative operation and maintenance method for a pharmaceutical packaging production line as described in claim 6, characterized in that, In step S1, the operating parameter data includes equipment temperature data, vibration data, noise data, and current data. The AI ​​vision system uses the YOLO target detection algorithm to identify the location of the drug and output pixel coordinates. The pixel coordinates are then converted into three-dimensional grasping coordinates in the robotic arm's base coordinate system using a hand-eye calibration matrix.

8. The multi-agent collaborative operation and maintenance method for a pharmaceutical packaging production line as described in claim 6, characterized in that, In step S3, the fault prediction engine reads no less than 30 days of historical operating data from the big data storage module, integrates equipment mechanism knowledge, trains a time series prediction model, and performs predictive analysis on the equipment operating trend. When the time-series prediction model detects that the equipment operating parameters show an abnormal trend and the prediction reaches the warning value, the fault prediction engine automatically generates the warning information, which includes the expected time to reach the warning value.

9. A multi-agent collaborative operation and maintenance method for a pharmaceutical packaging production line as described in claim 6, characterized in that, In step S4, the intelligent brain central system calls the abnormal evidence interpretation intelligent agent, which reads real-time operating data, compares historical operating data, and matches the equipment manuals, SOP operating procedures and maintenance case documents in the operation and maintenance knowledge base to analyze the cause of the abnormality and generate a diagnostic report. The intelligent brain central system calls the device status inspection intelligent agent, which evaluates the device health based on the operating parameter data and outputs a health score. The intelligent brain central unit integrates the diagnostic report and health score and pushes them to the intelligent operation and maintenance APP.

10. A multi-agent collaborative operation and maintenance method for a pharmaceutical packaging production line as described in claim 6, characterized in that, In step S5, the maintenance personnel input natural language voice commands through the AI ​​voice assistant in the intelligent maintenance APP. The AI ​​voice assistant calls the maintenance knowledge question-and-answer intelligent agent to retrieve the corresponding handling plan from the maintenance knowledge base and outputs the retrieval results to the intelligent maintenance APP for the maintenance personnel to confirm.