UV printing remote monitoring and maintenance system based on Internet of Things

Abstract

The invention relates to the technical field of intelligent maintenance of the Internet of Things, and discloses a UV printing remote monitoring and maintenance system based on the Internet of Things, and the system comprises a data collection and preprocessing module which collects the operation parameters and environment data of UV printing equipment, and carries out the data preprocessing and state evaluation; the maintenance scheduling and resource coordination module is used for dividing the UV printing equipment network into a plurality of regional maintenance centers, constructing a deep Q network maintenance scheduling model, learning an optimal maintenance strategy and realizing resource coordination among regions; the intelligent contract and path optimization module automatically distributes and executes maintenance tasks and plans maintenance paths of technicians; the maintenance quality evaluation module is used for evaluating the maintenance task completion degree and updating the reputation value of a technician; the knowledge sharing and model optimization module is used for realizing maintenance knowledge sharing and model optimization among equipment; according to the invention, a decentralized architecture is adopted to eliminate the single-point fault risk, and the stability and reliability of the whole system are ensured.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of Internet of Things intelligent maintenance, more particularly, it relates to a UV printing remote monitoring and maintenance system based on Internet of Things. BACKGROUND

[0002] With the wide application of UV printing technology in industrial production, advertising production, packaging printing and other fields, the number of UV printing equipment presents an explosive growth, forming a large-scale distributed equipment network. However, the existing UV printing equipment maintenance technology has the following technical problems: Firstly, the traditional UV printing equipment maintenance scheduling system usually adopts a centralized architecture, all maintenance requests and decisions need to be processed through a central server, when the number of devices increases to a certain scale, the processing capacity of the central server becomes the performance bottleneck of the system, resulting in increased response delay, at the same time, once the central node fails, the entire maintenance system will not run normally, there is a single point failure risk; Secondly, the existing maintenance scheduling system mostly uses preset fixed rules for resource allocation, such as first-come-first-served, priority sorting, etc., such algorithms cannot effectively cope with complex and variable maintenance requirements, it is difficult to reasonably allocate limited maintenance resources when device failures occur intensively, resulting in delayed maintenance of critical devices and low efficiency; Thirdly, the traditional scheduling system is difficult to comprehensively consider multiple factors such as device importance, fault severity, geographical location, technical personnel expertise matching degree, etc., often causing unbalanced allocation of maintenance resources, high-value devices are not maintained in time, and technical personnel spend too much time moving between devices, resulting in low efficiency; In addition, there are many types of UV printing equipment, and the use scenarios are different, under the existing technology, the maintenance experience and knowledge of different equipment cannot be effectively shared, each device needs to be learned and adapted individually, resulting in low maintenance efficiency of new networked devices and small batch devices, at the same time, due to data privacy concerns, it is difficult for enterprises to share and analyze equipment operation data, which restricts the development of collaborative maintenance; Finally, the traditional maintenance system lacks an objective evaluation mechanism for maintenance quality, making it difficult to accurately evaluate the maintenance effect of technical personnel, affecting the continuous improvement of maintenance quality.

[0003] Therefore, a new technical solution is needed to solve the above problems and realize the intelligent, distributed remote monitoring and maintenance of UV printing equipment. SUMMARY

[0004] The present application provides a UV printing remote monitoring and maintenance system based on Internet of Things, which solves the technical problems of low maintenance efficiency of UV printing equipment, inability to realize real-time remote monitoring, unreasonable maintenance resource allocation and lack of intelligent decision support in related technologies.

[0005] The present application provides a UV printing remote monitoring and maintenance system based on Internet of Things, which includes: The data acquisition and preprocessing module collects the operating parameters and environmental data of the UV printing equipment through Internet of Things sensors and performs data preprocessing and state evaluation using an edge computing unit. The maintenance scheduling and resource coordination module divides the UV printing equipment network into multiple regional maintenance centers, constructs a deep Q network maintenance scheduling model based on the output of the edge computing unit, learns the optimal maintenance strategy, and realizes resource coordination between regions through a lightweight consensus mechanism. The smart contract and path optimization module automatically allocates and executes maintenance tasks through a smart contract based on the output of the deep Q network maintenance scheduling model, and uses a multi-objective constraint optimization algorithm to plan the maintenance path of technicians. The maintenance quality evaluation module evaluates the completion of maintenance tasks based on the results of maintenance task execution, updates the reputation value of technicians, and updates the reputation value of technicians. The knowledge sharing and model optimization module realizes maintenance knowledge sharing and model optimization between devices based on the evaluation results of maintenance tasks and the update of the reputation value of technicians.

[0006] Further, the operating parameters of the UV printing equipment collected by the Internet of Things sensors include UV lamp intensity, temperature, humidity, current, ink flow, print head position, air pressure and vibration, and the environmental data includes environmental temperature, humidity, air pressure and dust concentration. The data preprocessing and state evaluation using the edge computing unit includes filtering and normalizing the collected raw data, calculating the parameter deviation value based on statistical analysis and lightweight machine learning algorithm, calculating the device state comprehensive score based on the parameter importance weight, and determining the maintenance priority by considering the device state score and device importance.

[0007] Further, the construction of the deep Q network maintenance scheduling model includes defining the state space as the set of device states within the region, the state of the maintenance personnel and the location, and the maintenance task queue, defining the action space as the task allocation decision, the maintenance path planning and the resource scheduling, and defining the reward function as the weighted combination of maintenance timeliness, resource utilization efficiency and device importance. The deep Q network includes an input layer, a feature extraction layer, a fully connected layer and an output layer, and the network is trained through experience replay technology to store historical decision experience.

[0008] Further, the resource coordination between regions through a lightweight consensus mechanism includes that each regional maintenance center regularly exchanges state information and resource conditions, realizes resource allocation decision through resource state broadcast, resource demand broadcast and two-phase commit protocol, and processes resource allocation requests according to the emergency level of maintenance demand using a differentiated consensus mechanism.

[0009] Further, the automatic allocation and execution of maintenance tasks through the smart contract comprises: constructing a smart contract structure containing contract identification, task description, execution condition, completion standard, time constraint, reward mechanism, task priority and state variable, and triggering contract execution according to device state exception, predictive maintenance demand, periodic maintenance need or manual creation; The maintenance path planning using the multi-objective constraint optimization algorithm comprises: modeling the maintenance path planning as a vehicle path problem with time window constraints, defining an optimization objective function considering total travel time, device urgency and skill matching degree, and solving the optimal path using an improved ant colony algorithm.

[0010] Further, the multi-dimensional evaluation mechanism introduced to evaluate the completion degree of the maintenance task specifically comprises: constructing an evaluation model containing three core dimensions of timeliness, quality and satisfaction, the timeliness score considering response time and repair completion time, the quality score considering problem solving thoroughness, device recovery state and subsequent failure frequency, and the satisfaction score considering device operator feedback and user evaluation; calculating a comprehensive score by weighted average, and dynamically adjusting the weight coefficient according to the device type and the maintenance task; The updating of the reputation value of the technical personnel comprises: using an exponentially weighted average method to calculate a new reputation value based on historical reputation values and recent task scores, and establishing a special skill scoring system to form a technical personnel expertise portrait.

[0011] Further, the implementation of maintenance knowledge sharing and model optimization between devices is achieved by using a federated learning method and a transfer learning method, specifically comprising: Each regional maintenance center locally saves device operation and maintenance data, trains a maintenance model using the local data, uploads the trained model parameters instead of the original data to a central server, the central server generates a global model by weighted averaging the model parameters of each region according to the data volume of each region, and distributes the updated global model to each regional maintenance center; The model optimization using the transfer learning method comprises: for new network devices or small sample devices, selecting a source model based on device similarity, using feature transfer, instance transfer and parameter transfer strategies, and quickly adapting to the characteristics of the target device through a fine-tuning method.

[0012] Further, the federated learning method further comprises privacy protection enhancement measures: adding carefully calibrated noise before parameter upload to achieve differential privacy protection, using encryption technology to achieve secure aggregation of model parameters, and reducing the precision and quantity of uploaded parameters to reduce the risk of privacy leakage.

[0013] Further, the fine-tuning method of the transfer learning method comprises: adopting a hierarchical learning rate setting for the source model layer and the new added layer, adopting a progressive unfreezing strategy of freezing all pre-training layers first and then unfreezing layer by layer, and applying different intensity of regularization constraints to the source model parameters and the target model parameters.

[0014] The application provides a computer medium, comprising a memory and one or more processors, the memory stores executable code, and the one or more processors execute the executable code to implement the above-mentioned UV printing remote monitoring and maintenance system based on Internet of Things.

[0015] The application has the beneficial effects that: the decentralized architecture eliminates the risk of single point failure, even if part of the regional maintenance center is offline, the system can continue to run, ensuring the stability and reliability of the overall system. Through the combination of intelligent scheduling algorithm and path optimization, the maintenance response time can be effectively reduced, the work efficiency of technicians can be improved, the downtime of key equipment can be reduced, through predictive maintenance, the occurrence of sudden failures can be reduced, and the proportion of planned maintenance can be improved. Through the multi-agent collaborative mechanism, the reasonable allocation of maintenance resources is realized, the over-concentration or idling of resources is avoided, the resource utilization rate is improved, the maintenance cost is reduced, and through dynamic adjustment of task priority, timely maintenance of key equipment is ensured. Through the reinforcement learning method, the system can continuously learn and improve from historical data, adapt to environmental changes and new demands, the federal learning and transfer learning mechanism enables the system to quickly adapt to new equipment, improves the convergence speed of the maintenance model, and the system can easily support large-scale equipment, and the performance remains good scalability with the increase of the number of equipment. Through the multi-dimensional evaluation mechanism, an objective maintenance quality measurement standard is provided, the maintenance quality is improved, the customer satisfaction is improved, the service life of the equipment is prolonged, the reputation mechanism is used to guide technicians to improve the maintenance quality, and a virtuous circle is formed. BRIEF DESCRIPTION OF DRAWINGS

[0016] Figure 1 is a module diagram of the UV printing remote monitoring and maintenance system based on Internet of Things in the application; Figure 2 is a column chart comparing the system availability of the traditional centralized architecture and the decentralized architecture of the application under different failure rates; Figure 3 is a radar chart comparing the five key maintenance efficiency indicators of the application scheme and the traditional scheme; Figure 4 is a line chart showing the trend of fault prediction accuracy, response time optimization and resource utilization rate with the increase of training rounds in federal learning; Figure 5This is a bubble chart showing the performance metrics of UV printing equipment networks of different sizes using different algorithms; Figure 6 It is an area graph comparing the performance of the new device using transfer learning methods with traditional learning methods during model training. Detailed Implementation

[0017] The subject matter described herein will now be discussed with reference to exemplary embodiments. It should be understood that these embodiments are discussed only to enable those skilled in the art to better understand and implement the subject matter described herein, and changes may be made to the function and arrangement of the elements discussed without departing from the scope of this specification. Various processes or components may be omitted, substituted, or added as needed in the examples. Furthermore, some features described in the examples may be combined in other examples.

[0018] At least one embodiment of the present invention discloses a remote monitoring and maintenance system for UV printing based on the Internet of Things, such as Figure 1 As shown, it includes: The data acquisition and preprocessing module collects the operating parameters and environmental data of the UV printing equipment through IoT sensors, and uses the edge computing unit to perform data preprocessing and status assessment. Operating data from the UV printing equipment is collected via IoT sensors and preliminarily processed using edge computing units to generate equipment status assessment results. Specifically, this includes: Step 1.1: Collect the operating parameters and environmental data of the UV printing equipment in real time through IoT sensors; Equipment operating parameters include UV lamp intensity, temperature, humidity, current, ink flow rate, printhead position, air pressure, vibration, and other parameters; Environmental data includes ambient temperature, humidity, air pressure, dust concentration, etc.

[0019] Step 1.2: Perform data preprocessing and preliminary analysis using edge computing units; The collected raw data is preprocessed by filtering, normalization, etc. Detect outliers in the data and remove noise interference; Generate basic equipment status indicators.

[0020] Step 1.3: Construct a lightweight anomaly detection model for real-time status assessment; An anomaly detection model based on statistical analysis and lightweight machine learning algorithms is adopted; The deviation value of each parameter is calculated to indicate the degree of abnormality of the parameter. The specific method is to divide the difference between the current value of the parameter and the historical mean by the historical standard deviation to obtain a standardized deviation index. Based on the deviation of each parameter, and combined with the weight of the parameter importance, a comprehensive score of equipment status is calculated. The comprehensive score reflects the overall degree of abnormality of the equipment. Optionally, an adaptive threshold mechanism can be introduced into the anomaly detection model to dynamically adjust the threshold based on the equipment's operating environment and workload, thereby improving the accuracy of anomaly detection. For example, automatic calibration can be performed to determine the normal range of UV lamp intensity under different ambient temperatures.

[0021] In some implementations, the local anomaly factor algorithm can be used instead of simple deviation calculation. By calculating the local density ratio of the samples, anomaly points can be identified more accurately, which is particularly suitable for anomaly detection in multidimensional parameter space.

[0022] Step 1.4: Prioritize maintenance based on equipment status score and equipment importance assessment; Maintenance priority is determined by a weighted combination of equipment status score and equipment importance index, with the weighting coefficients adjusted according to actual application needs; The assessment results are transmitted to the regional maintenance center as input data for maintenance scheduling.

[0023] The maintenance scheduling and resource coordination module divides the UV printing equipment network into multiple regional maintenance centers, constructs a deep Q-network maintenance scheduling model based on the output of the edge computing unit, learns the optimal maintenance strategy, and achieves inter-regional resource coordination through a lightweight consensus mechanism. The UV printing equipment network is divided into multiple regional maintenance centers, and an intelligent scheduling system based on deep Q networks is built to achieve distributed maintenance decision-making. Specifically, this includes: Step 2.1: Construct a multi-agent collaborative maintenance architecture; The UV printing equipment network is divided into multiple regions based on geographical location; Each region is configured with an independent maintenance center, acting as an intelligent agent; Each regional maintenance center is responsible for monitoring the status of equipment and making maintenance decisions within its designated area.

[0024] Step 2.2: Construct and maintain a scheduling model using a deep Q-network; Define the state space, including the set of device states within the area, the status and location of maintenance personnel, and the maintenance task queue, etc. Define the action space, including task allocation decisions, maintenance path planning, resource scheduling, etc. Define a reward function that comprehensively considers factors such as maintenance timeliness, resource utilization efficiency, and equipment importance, and calculates the total reward value through a weighted method; Constructing a deep Q-network structure for estimating the state-action value function includes: Input layer: Receives state vectors, which include features such as device status, technician status, and task information; Feature extraction layer: Contains multiple convolutional layers to extract spatial features of the device state; Fully connected layers: Contain 3 to 4 fully connected layers, each using the ReLU activation function for deep feature fusion; Output layer: Q-value corresponding to each possible maintenance scheduling action; The goal of network training is to minimize the temporal difference error, which is the mean square error between the current Q-value estimate and the target Q-value.

[0025] In a practical implementation, the specific structure of a deep Q-network can be: Input layer: Receives a length of The state vector, The number of equipment and technical personnel in the region will be dynamically adjusted. First hidden layer: 256 neurons, ReLU activation function; Second hidden layer: 128 neurons, ReLU activation function; Third hidden layer: 64 neurons, ReLU activation function; Output layer: neurons of action space size, with linear activation functions.

[0026] In a scenario with 20 UV printers and 5 technicians in the area, the state vector dimension might be 100 to 200, while the action space dimension (possible combinations of task assignments) is approximately 100. Through continuous learning, the network can capture the complex relationships between equipment status, fault characteristics, and optimal maintenance decisions.

[0027] Step 2.3: Train a deep Q-network to learn the optimal maintenance strategy; Utilizing experience replay technology to store historical decision-making experience: ; in Indicates the current state. Indicates the current action. Indicates the current reward. Indicates the next state.

[0028] Network parameters are updated using random mini-batch samples. Balance exploration and utilization through an ε-greedy strategy; Apply the target network to the stable training process.

[0029] Optionally, different variants of deep reinforcement learning algorithms can be used for UV printing device networks of different sizes: For small-scale networks (number of devices <50), a basic deep Q-learning algorithm can be used; For medium-sized networks (50 to 200 devices), the Double Deep Q-Learning (DDQN) algorithm can be used to reduce the problem of Q-value overestimation. For large-scale networks (number of devices > 200), the Distributed Deep Deterministic Policy Gradient (D4PG) algorithm can be used to improve training efficiency and policy quality.

[0030] Step 2.4: Implement a lightweight consensus mechanism between regions; The regional maintenance centers regularly exchange status information and resource availability. Coordinating cross-regional resource allocation through partial consensus algorithms, specifically including: Resource Status Broadcast: Each regional maintenance center periodically broadcasts a summary of its resource status, including the number of available technical personnel, their skill types, and load levels. Resource demand broadcasting: Regions with insufficient resources broadcast their resource needs, including the type, quantity, and urgency of the need.

[0031] Two-phase commit protocol: An improved two-phase commit protocol is used to complete resource allocation decisions. Phase 1 (Pre-submission): Potential resource providers send a pre-allocation plan to resource requesters; Phase Two (Confirmation): The resource requester selects the optimal solution from the received pre-allocation schemes and sends a confirmation message to the relevant region.

[0032] Lightweight verification mechanism: Only local nodes need to participate in verification, rather than the entire network consensus, reducing communication overhead.

[0033] Different consensus mechanisms are designed to address maintenance needs of varying urgency: High urgency: Fast consensus mode, simplified verification steps, and priority allocation of resources; Medium urgency: Standard consensus model, balancing decision-making speed and global optimality; Low urgency: Full consensus mode, fully considering global resource optimization; When resources are insufficient in a region, resources can be allocated from neighboring regions through a consensus mechanism. To ensure the overall optimality of global resource allocation, the fairness and efficiency of resource allocation are achieved by dynamically adjusting the resource weight coefficients of each region.

[0034] The smart contract and path optimization module, based on the output of the deep Q-network maintenance scheduling model, automatically allocates and executes maintenance tasks through smart contracts, and uses a multi-objective constraint optimization algorithm to plan the maintenance path for technicians; Maintenance tasks are automatically assigned and executed through smart contracts, and optimal maintenance paths are planned based on multi-objective constraint optimization algorithms. Specifically, this includes: Step 3.1: Construct a smart contract system for maintenance tasks; Define the structure of the smart contract for the maintenance task, including the following core components: Contract Identifier: A unique identifier used to track the contract execution process; Task description: Includes device ID, fault type, maintenance details, etc. Execution conditions: The set of conditions that trigger task execution; Completion criteria: Define the acceptance conditions for successful task completion; Time constraints: response time limits, completion deadlines, etc.; Reward mechanism: Rewards are calculated based on completion quality and timeliness; Task priority: determines the order in which resource conflicts are processed; State variables: record the current execution state of the contract.

[0035] Contract triggering conditions include: Equipment status abnormal: key parameters exceed thresholds; Predictive maintenance requirements: based on fault prediction results; Regular maintenance: Automatically triggered based on equipment usage time; Manually created: Maintenance tasks created manually by the administrator.

[0036] The contract execution process includes: Task allocation: The system automatically selects the most suitable technical personnel; Task Acceptance Confirmation: Technical personnel confirm receipt of the task; Execution progress report: Records each stage of task execution; Verification complete: The task has been confirmed to be completed through multiple verifications. Rewards Settlement: Calculate and distribute task rewards.

[0037] A typical implementation of smart contracts uses a state machine model to maintain the lifecycle of a task, including the following states: created, assigned, confirmed, executing, pending verification, completed, and canceled. Contracts can transition between these states based on preset conditions or external events.

[0038] An example structure for a maintenance task smart contract can be represented as follows: Contract ID: MT-20230615-0023; Equipment Information: {Equipment ID: UV-P-12345, Model: XYZ-2000, Location: Workshop A, Unit 3, Factory}; Fault information: {Type: Abnormal UV lamp intensity, Description: Lamp intensity is 20% lower than the standard value}; Maintenance requirements: {Response time: 4 hours, Completion time: 24 hours, Required skills: UV lamp repair}; Priority: High; Rewards: {Base value: 100 points, Time-based reward: up to 50 points, Quality reward: up to 50 points}; Current status: Allocated; Assignment Information: {Technician ID: TP-0089, Assignment Time: 2023-06-15 10:30:45}.

[0039] Step 3.2: Plan the maintenance path based on a multi-objective constraint optimization algorithm; The maintenance route planning is modeled as a variant of the vehicle routing problem with time window constraints. Its formal representation includes the node set, time window constraints, service time and travel time matrix. Define an optimization objective function that comprehensively considers total travel time, equipment urgency-weighted waiting time, and skill matching degree, and balances the importance of each factor through weighting coefficients. Consider multiple constraints: Time window constraint: Maintenance must begin within the specified time window; Capacity constraints: Technicians have limited working hours per day; Skills matching constraint: Specific equipment must be maintained by technicians with the corresponding skills; Priority constraint: High-priority tasks must be scheduled first; An improved ant colony algorithm is used to find the optimal path, including steps such as initialization, solution construction, local search optimization, and pheromone update.

[0040] Step 3.3, Task Execution and Dynamic Adjustment; Maintenance personnel receive task details and route planning via mobile devices; Report progress and problems encountered in real time during task execution; Subsequent tasks will be dynamically adjusted based on the actual implementation situation; Path replanning when urgent tasks are inserted.

[0041] The maintenance quality assessment module, based on the results of maintenance task execution, introduces a multi-dimensional assessment mechanism to evaluate the completion of maintenance tasks and update the reputation value of technical personnel. A multi-dimensional evaluation mechanism is introduced to objectively assess the completion of maintenance tasks, which is used to update the reputation scores of technical personnel and adjust subsequent task allocation. Specifically, this includes: Step 4.1: Construct a multi-dimensional evaluation model for maintenance quality; Define evaluation dimensions, including three core dimensions: timeliness, quality, and satisfaction; Timeliness score ( Factors to consider: Response time, comparison of repair completion time with expected time; Quality rating ( Factors to consider: completeness of problem resolution, equipment recovery status, and frequency of subsequent failures; Satisfaction rating ( Factors to consider: feedback from equipment operators and user evaluations.

[0042] Step 4.2, Comprehensive score calculation and analysis; The overall score is calculated by weighting the timeliness score, quality score, and satisfaction score, with the sum of the weighting coefficients being 1. The weighting coefficients can be dynamically adjusted according to different types of equipment and maintenance tasks; Build a historical scoring database for technical personnel performance analysis and maintenance quality trend analysis.

[0043] Optionally, different weighting configurations can be used for different types of UV printing equipment: For critical production equipment: timeliness weight 0.4, quality weight 0.5, satisfaction weight 0.1, emphasizing maintenance quality; For time-sensitive equipment: timeliness weight 0.6, quality weight 0.3, satisfaction weight 0.1, emphasizing timeliness; For customer experience-sensitive devices: timeliness weight 0.3, quality weight 0.3, satisfaction weight 0.4, emphasizing satisfaction.

[0044] Step 4.3, update the technician's reputation value; The reputation score of technicians is updated based on the rating results of maintenance tasks. An exponential weighted average method is used, and the new reputation score is a proportional combination of the old reputation score and the score of the most recent task. Reputation score serves as an important reference factor when assigning tasks, increasing the probability that highly reputable technical personnel will receive tasks; Specialized skills assessments are conducted for specific equipment types to create a profile of the technical personnel's expertise.

[0045] The knowledge sharing and model optimization module enables maintenance knowledge sharing and model optimization among equipment based on maintenance task evaluation results and updates to technician reputation values. By leveraging federated learning and transfer learning methods, efficient sharing of maintenance knowledge and model updates can be achieved among devices, while protecting data privacy. Specifically, this includes: Step 5.1: Construct a federated learning framework to achieve knowledge sharing; Each regional maintenance center, as a participant in federated learning, locally stores equipment operation and maintenance data.

[0046] Locally trained and maintained models include the following typical models: Fault prediction model: Predicts the types and timing of potential equipment failures; Maintenance decision-making model: Optimize the allocation and scheduling of maintenance tasks; Life prediction model: Predicts the remaining service life of critical components.

[0047] Federated learning process implementation: Initialization: The central server creates an initial global model and distributes it to all participants; Local training: Each participant trains the model using local data. The training objective function is: ; in The minimization operator finds the parameters that minimize the objective function. These are model parameters; As the normalization factor, Indicates the first Local dataset of participants The number of samples; This represents the summation operator, which accumulates the loss for all samples in the dataset; For training sample pairs, As input features, For the corresponding tags; The "belongs to" operator indicates the sample. Belongs to dataset ; Indicates the first Local datasets of each participant Indicates the participant's number; Represents the loss function; The parameters represent the loss function. and These represent the input features and their corresponding labels, respectively. Model upload: Each participant uploads the trained model parameters (not the original data) to the central server; Model aggregation: The central server aggregates model parameters to generate a global model. The aggregation method is to weight the model parameters of each region based on the amount of data in each region. Model distribution: Distribute the updated global model to all participants; Iterative repetition: Repeat local training, model uploading, model aggregation, and model distribution until the model converges or reaches the preset number of iterations.

[0048] Enhanced privacy protection measures: Differential privacy: Add carefully calibrated noise before parameters are uploaded to protect sensitive information; Secure aggregation: Uses encryption technology to achieve secure aggregation of model parameters, preventing the central server from knowing the model parameters of individual participants; Model compression: Reduces the precision or number of uploaded parameters, further reducing the risk of privacy leaks.

[0049] In some implementations, different variants of federated learning can be selected based on data privacy requirements and network conditions: For scenarios with high privacy requirements, a vertical federated learning method can be used, which only shares gradients and not model parameters; For scenarios with limited communication, model distillation federated learning can be used to reduce communication overhead; For heterogeneous device scenarios, a personalized federated learning approach can be adopted, allowing each participant to maintain some personalized parameters.

[0050] Step 5.2: Apply transfer learning to adapt to the new device; For newly connected devices or devices with a small sample size, transfer learning methods can be applied to quickly build maintenance models. The source model is selected based on device similarity measurement, and similarity is calculated by cosine similarity of feature vectors.

[0051] Implementation of transfer learning process: Source model selection: Based on the device similarity matrix, select the models of the top K devices with the highest similarity as source models, where K is a preset parameter for the number of candidate source models, usually ranging from 3 to 5; Domain Adaptation: The maximum mean difference minimization method is applied to reduce the difference in feature distribution between the source and target domains; Model structure adjustment: retain the feature extraction layer of the source model, and replace or fine-tune the classification / regression layer; Construction of hybrid training set: Combine relevant samples from the source domain with a small number of samples from the target domain to form a hybrid training set; Incremental training: Use a mixed training set to train the model and gradually increase the weight of the target domain data.

[0052] Specific implementation of the migration strategy: Feature transfer: Applicable to devices with similar structures but different usage scenarios; Instance migration: Suitable for scenarios where data distribution is similar but the quantity is insufficient; Parameter migration: Applicable to knowledge migration between devices with the same model architecture.

[0053] Details of the fine-tuning method: Layered learning rate settings: Use a smaller learning rate for the source model layer (e.g., 0.1 times the base learning rate), and use the standard learning rate for new layers; Progressive unfreezing: First, freeze all pre-trained layers and train only the newly added layers; then unfreeze them layer by layer, from the higher layers to the lower layers. Customized regularization: Apply regularization constraints of different strengths to the source model parameters and the target model parameters.

[0054] Optionally, for specific types of UV printing equipment, adversarial transfer learning methods can be employed. Adversarial training forces the features of the source and target domains to be distributed in the same space, further improving transfer performance. For example, for UV printing equipment with similar functions but significant brand differences, adversarial training can eliminate brand-specific data distribution differences.

[0055] Step 5.3: Implement continuous model optimization; Collect the differences between model predictions and actual results to serve as a basis for improvement; Regularly evaluate model performance, including metrics such as accuracy, recall, and F1 score (the harmonic mean of accuracy and recall). Design a time-based forgetting mechanism to reduce the impact of outdated data; Optimization of specialized models for specific types of faults; It should be noted that continuous model optimization can employ an incremental learning approach, using only new data for incremental training each time, avoiding the computational overhead of retraining the entire model, which is particularly suitable for resource-constrained edge devices.

[0056] A computer medium includes a memory and one or more processors, wherein executable code is stored in the memory, and when the one or more processors execute the executable code, it is used to implement the above-described IoT-based UV printing remote monitoring and maintenance system.

[0057] Here, the present invention provides an implementation example: This example is based on the remote monitoring and maintenance system of UV printing equipment of a chain advertising production company. The company has 20 branches across the country and uses a total of 120 UV printing devices, which are distributed in different geographical locations and are maintained by 40 technicians.

[0058] This chain advertising production company owns various models of UV printing equipment, mainly used for producing advertising signs, display boards, vehicle wraps, and other products. Due to its nationwide business, the traditional equipment maintenance model faces the following challenges: The equipment is scattered throughout the country, the number of technical personnel is limited, and the maintenance response time is long. Equipment failures and resulting downtime have severely impacted customer order delivery. Maintenance experience is difficult to share between equipment, and the adaptation period for new equipment or newly hired technicians is long; The quality of maintenance is inconsistent and there is a lack of objective evaluation standards.

[0059] The company decided to implement the IoT-based remote monitoring and maintenance method for UV printing provided in this application, and to upgrade and deploy the system on all 120 devices.

[0060] The company has installed an IoT sensor suite on each UV printer, including temperature sensors, humidity sensors, UV intensity sensors, current sensors, and vibration sensors. Each device is equipped with an edge computing unit at its edge, responsible for data preprocessing and anomaly detection.

[0061] For a UV printer of model XYZ-2000, the typical parameters collected by the system include: UV lamp intensity: 180-220mW / cm² (normal range); Printhead temperature: 55-65℃ (normal range); Ambient humidity: 30% to 50% (normal range); Motor current: 2.4 to 2.8A (normal range); Vibration index: 0.1 to 0.3g (normal range).

[0062] By analyzing the changing trends of these parameters in real time, the edge computing unit discovered that the intensity of the UV lamps in the device had been continuously decreasing over the past three days, from 210mW / cm² to 175mW / cm². Although still within the normal range, the deviation had reached the warning level. The system calculated the device's status score to be 0.72 (out of 1), and combined with the device's importance index of 0.85 (this device is responsible for the company's core business), calculated a maintenance priority of 0.78, triggering the predictive maintenance process.

[0063] The system divides the 20 branches nationwide into 5 regional maintenance centers, each responsible for managing the equipment of the surrounding branches. Regarding the equipment with abnormal UV lamp intensity, the East China regional maintenance center, upon receiving the maintenance request, performs decision analysis based on a deep Q-network model.

[0064] This model takes into account: Equipment status: UV lamp intensity has decreased, but it has not yet affected normal production; Equipment Importance: This equipment processes core business orders; Technical personnel: There are 3 technicians in the area who are qualified to maintain UV lamps; Geographic location: The nearest technician is approximately 35 kilometers from the equipment location; Other pending tasks: There are still 2 devices in the area with pending maintenance tasks.

[0065] The model outputs the following decision: Technician Li will be assigned to perform preventative maintenance the following morning, with a priority set to medium-high. Since resources are sufficient within the region, there is no need for cross-regional resource allocation; instead, the task assignment will be broadcast to other regional maintenance centers via a lightweight consensus mechanism to facilitate global task coordination.

[0066] The system automatically generated a maintenance task smart contract, which includes the following: Task ID: MT-20230712-0045; Equipment Information: Model XYZ-2000, ID: UV-P-10086, Location: Shanghai Branch, East China Region; Task Description: The UV lamp intensity is abnormal, requiring preventative maintenance; Maintenance includes: inspecting the UV lamps, cleaning the lamp surface, calibrating the lamp parameters, and replacing the lamps if necessary; Response timeframe: Within 24 hours; Completion timeframe: Within 48 hours; Required skills: UV lamp maintenance qualification; Priority: Medium to high.

[0067] After receiving the task via his mobile device, technician Li confirmed acceptance. The system then planned an optimal maintenance path, including the maintenance sequence and route for the three devices Li needed to handle that day. The total travel time was estimated to be 4.5 hours, saving 1.5 hours compared to traditional path planning.

[0068] Mr. Li completed the equipment maintenance according to the system's planned path, and the entire process was recorded in real time via a mobile device. It was ultimately confirmed that there was slight contamination on the surface of the UV lamp tube. After cleaning, the intensity was restored to 205mW / cm², which meets the normal working requirements, and there is no need to replace the lamp tube.

[0069] After the maintenance task is completed, the system automatically performs a multi-dimensional evaluation: Timeliness score: The task was completed within 18 hours of being assigned, demonstrating timely response, and received a score of 0.92. Quality score: The UV lamp intensity returned to normal and there was no recurrence within one week, with a score of 0.85; Satisfaction rating: The equipment operators gave positive feedback, with a rating of 0.90.

[0070] The system calculated a comprehensive score of 0.88 based on the weighted configuration of this type of equipment (timeliness 0.4, quality 0.5, satisfaction 0.1). This score was used to update Li's reputation score, raising it from 0.82 to 0.84, and also improving his specialized skills in UV lamp maintenance.

[0071] The complete data for this maintenance case is stored in the local database of the East China Regional Maintenance Center and used for local model training. Through the federated learning framework, the East China Regional Maintenance Center only shares the updated model parameters (not the original data) to the central server. The central server aggregates the model parameters from the five regional maintenance centers to generate a new global model.

[0072] A month later, the company added another UV printer of the same model in the Northwest region. Using transfer learning, the system quickly built a fault prediction model suitable for the new equipment based on the existing UV lamp maintenance model. This model, fine-tuned using only 30 sample data points from the new equipment's operation, achieved a fault prediction accuracy of over 80%, significantly shortening the adaptation period for the new equipment.

[0073] By implementing the technical solution of this application in the chain advertising production company, good technical results have been achieved: In terms of system reliability: In a regional server failure test, the South China regional maintenance center went offline, but the system was still able to maintain the normal operation of other regions through the decentralized architecture, and the overall maintenance capability remained above 85%, which proved the high reliability of the system.

[0074] Regarding maintenance efficiency: Six months after system implementation, the company's maintenance data is shown in Table 1: Table 1: Company Maintenance Data After 6 Months of Implementation Regarding resource utilization: The system intelligently schedules maintenance resources, achieving a more rational allocation of human resources. The comparison results of resource utilization optimization effects are shown in Table 2: Table 2: Comparison of Resource Utilization Optimization Effects The above data fully demonstrates the technical effectiveness of this implementation method in practical applications, and successfully solves the core technical problems in the remote monitoring and maintenance of large-scale UV printing equipment networks.

[0075] like Figures 2 to 6 The figures show: a bar chart comparing the system availability of the traditional centralized architecture and the decentralized architecture of this invention under different failure rates; a radar chart comparing the scheme of this invention and the traditional scheme on five key maintenance efficiency indicators; a line graph showing the changing trends of fault prediction accuracy, response time optimization, and resource utilization of federated learning as the number of training rounds increases; a bubble chart showing the performance indicators of UV printing equipment networks of different sizes using different algorithms; and an area graph comparing the performance of new equipment using transfer learning methods and traditional learning methods during model training.

[0076] The embodiments of the present invention have been described above. However, the embodiments are not limited to the specific implementation methods described above. The specific implementation methods described above are merely illustrative and not restrictive. Those skilled in the art can make more equivalent embodiments under the guidance of the present embodiments, and all of them are within the protection scope of the present embodiments.

Claims

1. A remote monitoring and maintenance system for UV printing based on the Internet of Things, characterized in that, include: The data acquisition and preprocessing module collects the operating parameters and environmental data of the UV printing equipment through IoT sensors, and uses the edge computing unit to perform data preprocessing and status assessment. The maintenance scheduling and resource coordination module divides the UV printing equipment network into multiple regional maintenance centers, constructs a deep Q-network maintenance scheduling model based on the output of the edge computing unit, learns the optimal maintenance strategy, and achieves inter-regional resource coordination through a lightweight consensus mechanism. The smart contract and path optimization module, based on the output of the deep Q-network maintenance scheduling model, automatically allocates and executes maintenance tasks through smart contracts, and uses a multi-objective constraint optimization algorithm to plan the maintenance path for technicians; The maintenance quality assessment module, based on the results of maintenance task execution, introduces a multi-dimensional assessment mechanism to evaluate the completion of maintenance tasks and update the reputation value of technical personnel. The knowledge sharing and model optimization module enables maintenance knowledge sharing and model optimization among equipment based on maintenance task evaluation results and updates to technician reputation values.

2. The IoT-based remote monitoring and maintenance system for UV printing according to claim 1, characterized in that, The operation parameters and environmental data of the UV printing equipment are collected through IoT sensors. The operation parameters of the UV printing equipment include UV lamp intensity, temperature, humidity, current, ink flow, printhead position, air pressure and vibration. The environmental data includes ambient temperature, humidity, air pressure and dust concentration. The data preprocessing and status assessment using edge computing units includes: filtering and normalizing the collected raw data, calculating the deviation values ​​of each parameter based on statistical analysis and lightweight machine learning algorithms, calculating the comprehensive equipment status score by combining the importance weights of the parameters, and determining the maintenance priority by comprehensively considering the equipment status score and equipment importance.

3. The IoT-based remote monitoring and maintenance system for UV printing according to claim 1, characterized in that, The construction of the deep Q network maintenance scheduling model includes: defining the state space as a set of equipment states, maintenance personnel states and locations, and maintenance task queues within the region; defining the action space as task allocation decision-making, maintenance path planning, and resource scheduling; and defining the reward function as a weighted combination of maintenance timeliness, resource utilization efficiency, and equipment importance. A deep Q-network consists of an input layer, a feature extraction layer, a fully connected layer, and an output layer. It uses experience replay technology to store historical decision-making experiences to train the network.

4. The IoT-based remote monitoring and maintenance system for UV printing according to claim 1, characterized in that, The lightweight consensus mechanism for inter-regional resource coordination includes: each regional maintenance center regularly exchanging status information and resource status; making resource allocation decisions through resource status broadcasting, resource demand broadcasting, and a two-phase commit protocol; and using differentiated consensus mechanisms to handle resource allocation requests based on the urgency of maintenance needs.

5. The IoT-based remote monitoring and maintenance system for UV printing according to claim 1, characterized in that, The automatic allocation and execution of maintenance tasks through smart contracts includes: constructing a smart contract structure containing contract identifier, task description, execution conditions, completion criteria, time constraints, reward mechanism, task priority and state variables, and triggering contract execution based on abnormal equipment status, predictive maintenance needs, periodic maintenance needs or manual creation; The method of planning maintenance routes using a multi-objective constraint optimization algorithm includes: modeling maintenance route planning as a vehicle routing problem with time window constraints, defining an optimization objective function that considers total travel time, equipment urgency, and skill matching, and using an improved ant colony algorithm to solve for the optimal route.

6. The IoT-based remote monitoring and maintenance system for UV printing according to claim 1, characterized in that, The introduction of a multi-dimensional evaluation mechanism to assess the completion of maintenance tasks specifically includes: constructing an evaluation model that includes three core dimensions: timeliness, quality, and satisfaction. The timeliness score considers response time and repair completion time; the quality score considers the thoroughness of problem resolution, equipment recovery status, and frequency of subsequent failures; and the satisfaction score considers feedback from equipment operators and user evaluations. A comprehensive score is calculated through a weighted average, with the weighting coefficients dynamically adjusted according to the equipment type and maintenance task. The updated technical personnel reputation value includes: using an index-weighted average method to calculate a new reputation value based on historical reputation values ​​and recent task scores, and establishing a specialized skills scoring system to form a profile of technical personnel's expertise.

7. The IoT-based remote monitoring and maintenance system for UV printing according to claim 1, characterized in that, The aforementioned achievement of inter-device maintenance knowledge sharing and model optimization utilizes federated learning and transfer learning methods, specifically including: Each regional maintenance center stores equipment operation and maintenance data locally, uses local data to train maintenance models, and uploads the trained model parameters, rather than the original data, to the central server. The central server generates a global model by weighting and averaging the model parameters of each region based on the amount of data in each region, and distributes the updated global model to each regional maintenance center. Model optimization using transfer learning methods includes: selecting source models based on device similarity for newly added devices or small sample devices, and employing feature transfer, instance transfer, and parameter transfer strategies to quickly adapt to the characteristics of the target device through fine-tuning methods.

8. The IoT-based remote monitoring and maintenance system for UV printing according to claim 7, characterized in that, The federated learning method also includes privacy protection enhancements: adding carefully calibrated noise before parameter upload to achieve differential privacy protection, using encryption technology to achieve secure aggregation of model parameters, and reducing the accuracy and number of uploaded parameters to reduce the risk of privacy leakage.

9. The IoT-based remote monitoring and maintenance system for UV printing according to claim 7, characterized in that, The fine-tuning method of the transfer learning method includes: setting a hierarchical learning rate for the source model layer and the newly added layer; a progressive unfreezing strategy that first freezes all pre-trained layers and trains only the newly added layer, and then unfreezes them layer by layer; and applying regularization constraints of different intensities to the source model parameters and the target model parameters.

10. A computer medium, characterized in that, It includes a memory and one or more processors, wherein the memory stores executable code, and the one or more processors execute the executable code to implement the IoT-based UV printing remote monitoring and maintenance system according to any one of claims 1-9.

Images