Assembly equipment health management system based on industrial internet of things

Abstract

The invention discloses an assembly equipment health management system based on industrial Internet of Things, and particularly relates to the field of intelligent maintenance of industrial equipment, comprising a distributed sensing module, a cross-production-line transfer learning module, a dynamic maintenance strategy optimization module, a multi-modal man-machine cooperation module and an enhanced execution terminal module, multi-source sensor clusters and intelligent edge nodes are deployed, multi-dimensional operation data are collected and preprocessed at high frequency, an encrypted global fault feature library is constructed by using a federal transfer learning algorithm, cross-production-line group intelligent advanced early warning is realized, and the equipment health degree, the production plan and the resource state are comprehensively considered based on a deep reinforcement learning decision-making agent. A priority maintenance strategy is output, a digital twinning and AR technology is fused to generate an enhanced maintenance guidance package, real-time operation verification and health re-evaluation are realized through an execution terminal, accurate and efficient closed-loop health management is formed, and the equipment reliability and the maintenance intelligence level are remarkably improved.

Description

Technical Field

[0001] This invention relates to the field of intelligent maintenance technology for industrial equipment, and more specifically, to a health management system for assembly equipment based on the Industrial Internet of Things. Background Technology

[0002] In the industrial manufacturing sector, the stability and reliability of assembly equipment directly affect the overall efficiency of the production line and product quality. With the rapid development of industrial Internet of Things (IoT) technology, traditional regular maintenance and post-repair methods are no longer sufficient to meet the stringent requirements of modern intelligent manufacturing for high equipment availability and low downtime. Therefore, data-driven predictive health management has gradually become a key focus of the industry.

[0003] The existing technology is a predictive maintenance system for equipment based on a single data source. Typically, vibration or temperature sensors are installed on a single device to collect operational data, which is then transmitted to a cloud platform. Machine learning algorithms are used for fault diagnosis and remaining life prediction. When an anomaly is detected, alarm information is sent to maintenance personnel. The system judges the health status of the equipment by setting fixed thresholds or simple models, and generates maintenance suggestions based on the diagnostic results to perform health management of the assembled equipment.

[0004] However, in actual use, it still has some shortcomings, such as relying on data from a single device, making it difficult to capture weak or early fault characteristics, resulting in delayed early warnings and a high rate of missed reports. Maintenance decisions do not comprehensively consider multi-dimensional constraints such as production plans, resource inventory, and personnel status. The generated maintenance plans are often out of sync with the actual production rhythm and resource conditions, resulting in poor executability. The maintenance execution process lacks effective human-machine collaboration and real-time guidance, mainly relying on traditional paper work instructions or two-dimensional electronic manuals, which can easily lead to operational errors and make it difficult to quantify and verify the maintenance effect. It is impossible to form a closed-loop health management system from status perception and intelligent decision-making to precise execution. Summary of the Invention

[0005] In order to overcome the above-mentioned deficiencies of the prior art, embodiments of the present invention provide an assembly equipment health management system based on the Industrial Internet of Things to solve the problems mentioned in the background art.

[0006] To achieve the above objectives, the present invention provides the following technical solution: an assembly equipment health management system based on the Industrial Internet of Things, including a distributed sensing module, which deploys a sensor cluster and intelligent edge nodes on the assembly equipment on the production line to collect and preprocess equipment operating status data at high frequency; The cross-production line transfer learning module establishes an encrypted global fault precursor feature library. Through federated transfer learning algorithms, it mines the correlation of weak fault features among similar equipment on different production lines and performs early fault warning based on swarm intelligence. The dynamic maintenance strategy optimization module is a decision-making agent based on deep reinforcement learning. It sets the state space to include equipment health, production plan, resource inventory and personnel status, and evaluates the equipment health index through action space and reward function. The multimodal human-machine collaboration module generates maintenance instructions and synchronously integrates the equipment's digital twin model, maintenance knowledge base, and real-time data to generate an enhanced maintenance guidance package that includes AR visualization guidance, 3D operation animation, standard operating procedures, and key parameter thresholds. The enhanced execution terminal module receives the enhanced maintenance guidance package, collects maintenance operation data and equipment feedback parameters in real time, compares and verifies them with the standard operating procedures and key parameter thresholds in the guidance package in real time, and automatically triggers the equipment health status reassessment process.

[0007] Preferably, the sensor cluster includes vibration sensors, temperature sensors, current sensors, acoustic sensors, machine vision sensors, and torque sensors. The operating status data includes time-domain peak / valley / mean / kurtosis / frequency-domain power spectral density (PSD), characteristic frequency, real-time temperature value, three-phase current RMS value, current harmonics, starting current peak value, sound pressure level and spectrogram, equipment image, tightening torque, and rotation angle.

[0008] Preferably, the preprocessing includes data cleaning, feature extraction, and data filtering. The data cleaning uses wavelet thresholding for noise reduction, the feature extraction includes time-domain features, frequency-domain features, and time-series features, and the data filtering deploys a lightweight anomaly detection model to make a preliminary judgment on the preprocessed features.

[0009] Preferably, the global fault precursor feature library includes a three-layer structure of a basic feature layer, a precursor feature layer, and a fault feature layer, and adopts dual protection of "federated encryption + differential privacy". The federated transfer learning algorithm includes domain partitioning and initialization, local training, domain adaptation, global model update and weak fault mining. The domain partitioning is to divide the production line into a source domain and a target domain based on the equipment runtime and fault cases.

[0010] Preferably, the action space includes maintenance type decision, resource allocation decision, and personnel allocation decision. The maintenance type decision includes predictive maintenance, preventive maintenance, and post-failure repair. The reward function includes positive reward items and negative reward items. Positive reward items include health improvement rewards, production assurance rewards, resource efficiency rewards, and personnel efficiency rewards. Negative reward items include maintenance delay penalties and production loss penalties.

[0011] Preferably, the digital twin model of the device includes geometric information, physical information, and motion information, and the maintenance knowledge base includes structured data, unstructured data, and knowledge updates.

[0012] Preferably, the AR visualization guidance includes fault location, operation guidance, and parameter monitoring; the 3D operation animation includes equipment breakdown animation and assembly animation; and the standard operating procedure includes basic information about the assembly equipment, step descriptions, and acceptance criteria.

[0013] Preferably, the maintenance operation data includes operation process data and manually entered data, and the equipment feedback parameters include trial operation data and assembly quality data.

[0014] Preferably, the real-time comparison and verification includes operation step verification, operation accuracy verification, and equipment parameter verification, and the health status reassessment includes triggering conditions, assessment models, and application of assessment results.

[0015] The technical effects and advantages of this invention are as follows: 1. This invention constructs a distributed sensing system by deploying a multi-source sensor cluster and intelligent edge nodes, realizing high-frequency acquisition and preprocessing of multi-dimensional status information of equipment vibration, temperature, current, acoustics and assembly quality. Based on the federated transfer learning architecture, it encrypts, shares and mines weak fault features of similar equipment across production lines, and establishes an early fault warning mechanism driven by collective intelligence, which significantly improves the timeliness and accuracy of fault detection and overcomes the bottleneck of limited data sensing capabilities and delayed early warning of traditional single equipment. 2. This invention constructs a decision-making intelligent agent by employing deep reinforcement learning technology. It integrates multi-dimensional constraints such as real-time equipment health, dynamic production planning, resource inventory, and personnel skill load. Driven by a reward function aimed at overall system benefits, it autonomously outputs collaborative maintenance instructions and resource allocation schemes. This achieves adaptive optimization of maintenance strategies, production cycle time, and resource status, improving the scientific nature and feasibility of maintenance decisions, effectively avoiding conflicts between maintenance activities and production plans, and reducing losses caused by unplanned downtime. 3. This invention deeply integrates digital twin models, AR visualization guidance, and real-time data verification to generate an enhanced maintenance guidance package. This enables real-time comparison and closed-loop verification of the maintenance process, greatly reducing the reliance on personnel experience in maintenance operations and improving the standardization and accuracy of operations. Through the re-evaluation of the health status of equipment after maintenance, a closed-loop health management system is formed, from intelligent early warning and optimized decision-making to precise execution and effect feedback, fundamentally improving the reliability and management level of equipment. Attached Figure Description

[0016] Figure 1 This is a schematic diagram of the overall structure of the present invention; Figure 2 This is a schematic diagram of the cross-production line migration learning and early warning system of the present invention; Figure 3 This is a schematic diagram illustrating the dynamic maintenance decision-making and AR execution of the present invention. Detailed Implementation

[0017] 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.

[0018] As attached Figure 1 Appendix Figure 2 Appendix Figure 3 The assembly equipment health management system based on the Industrial Internet of Things shown includes a distributed sensing module that deploys sensor clusters and intelligent edge nodes on the assembly equipment on the production line to collect and preprocess equipment operating status data at high frequency.

[0019] It should be specifically noted that the sensor cluster includes vibration sensors, temperature sensors, current sensors, acoustic sensors, machine vision sensors, and torque sensors. The operating status data includes time-domain peak / valley / mean / kurtosis / frequency-domain power spectral density (PSD), characteristic frequency, real-time temperature value, three-phase current RMS value, current harmonics, starting current peak value, sound pressure level and spectrogram, equipment image, tightening torque, and rotation angle.

[0020] The preprocessing includes data cleaning, feature extraction, and data filtering. Data cleaning uses wavelet thresholding for noise reduction. Feature extraction includes time-domain features, frequency-domain features, and time-series features. Data filtering deploys a lightweight anomaly detection model to make a preliminary judgment on the preprocessed features.

[0021] It should be further explained that a multi-type sensor cluster is deployed to target the core fault points of the assembly equipment. The specific configuration is as follows: Vibration sensors are piezoelectric triaxial accelerometers, mounted on the surface of transmission components, with a sampling frequency of 20kHz. Acquiring parameters include time-domain peak / valley / mean / kurtosis / frequency-domain power spectral density (PSD) and characteristic frequency, used to detect fault characteristics such as mechanical wear and imbalance. Temperature sensors are resistance sensors with an accuracy of ±0.1℃, deployed in the motor stator windings, bearing lubrication system, and hydraulic oil tank, acquiring real-time temperature values ​​and temperature fluctuation amplitude within 10 minutes, used to identify faults such as motor overload, lubrication failure, and hydraulic system leakage. Current sensors are Hall effect current sensors, connected in series in the motor power supply circuit, acquiring the three-phase current RMS value, current harmonic distortion rate (THD), and starting current peak value, used to detect electrical... For electrical faults such as machine stall, winding short circuit, and abnormal load, the acoustic sensor uses a MEMS microphone array with a sampling rate of 48kHz, installed within 1m of the equipment casing, to collect sound pressure level and spectrogram, supplementing high-frequency faults that the vibration sensor cannot cover. The machine vision sensor uses an industrial camera with a resolution of 20 million pixels and a frame rate of 30fps, paired with a ring light source, to photograph the assembly interface. Image processing is used to extract part positioning deviation (±0.02mm), fitting gap (≤0.05mm), and bolt exposed length (±0.5mm) to identify potential faults caused by insufficient assembly accuracy. The torque sensor is integrated into the actuator of the tightening machine and press-fitting machine to collect tightening torque (range 0-500Nm, accuracy ±1%), rotation angle (0-360°), and torque-rotation angle curve to determine whether the bolt tightening meets the standard.

[0022] The reason for deploying sensor clusters and collecting related data is that a single sensor cannot cover the multiple types of faults in assembly equipment. The six types of sensors in the sensor cluster form a multi-dimensional data matrix of "mechanical-electrical-assembly", covering more than 95% of equipment fault types and reducing the rate of missed fault feature collection.

[0023] Each production line is equipped with one edge computing gateway to achieve localized "acquisition-preprocessing-screening" processing, reducing bandwidth waste caused by full data transmission. The specific processing flow includes: data cleaning, using wavelet threshold denoising to remove electromagnetic interference and mechanical vibration noise from the workshop, retaining fault characteristic signals, and feature extraction (time domain features, frequency domain features, and time series features). Time domain features: calculating the peak factor of the vibration signal. Pulse factor The slope of the temperature signal trend Slope during overload Frequency domain features: The vibration signal is converted to the frequency domain by FFT, and the PSD peak value corresponding to the feature frequency is extracted. Time series features: The 10-minute trend change of the current signal is extracted by using a sliding window to identify the load fluctuation pattern. Data is filtered, a lightweight anomaly detection model is deployed, and the pre-processed features are initially judged. If the feature is within the normal threshold range, it is stored locally for 7 days. If it exceeds the threshold, it is marked as "suspected fault data", encrypted and uploaded to the cloud.

[0024] The cross-production line transfer learning module establishes an encrypted global fault precursor feature library. Through federated transfer learning algorithms, it mines the correlation of weak fault features among similar equipment on different production lines and performs early fault warning based on swarm intelligence.

[0025] It should be specifically noted that the global fault precursor feature library includes a three-layer structure: a basic feature layer, a precursor feature layer, and a fault feature layer. It adopts dual protection of "federated encryption + differential privacy". The federated transfer learning algorithm includes domain partitioning and initialization, local training, domain adaptation, global model update, and weak fault mining. The domain partitioning is to divide the production line into source domain and target domain based on the equipment runtime and fault cases.

[0026] It needs further explanation that, for similar equipment on different production lines, an encrypted global fault precursor feature library is constructed. The specific design includes: a three-level structure for the feature library: a basic feature layer, a precursor feature layer, and a fault association layer. The basic feature layer stores multi-domain features pre-processed locally by each production line. Each feature is labeled with the equipment ID, production line number, acquisition time, and operating condition. The precursor feature layer uses a feature selection model trained on historical fault data (e.g., "vibration kurtosis > 4 and temperature trend slope > 0.3℃ / min" is a precursor feature of bearing lubrication failure). The fault association layer establishes a mapping relationship between precursor features and fault type and fault severity (levels 1-5, where level 1 is an early, weak fault and level 5 is a shutdown fault). The encryption scheme employs dual protection of "federated encryption + differential privacy." Each production line only uploads the encrypted precursor features, while the original data remains locally to avoid privacy leaks. Gaussian noise (noise intensity...) is added during feature library storage. ), satisfying differential privacy , Smaller size generally means stronger privacy protection; current technologies mostly provide this. This prevents attackers from using features to deduce the original data.

[0027] A hybrid algorithm combining federated averaging and domain-adaptive transfer learning is employed to uncover weak correlations in similar equipment across different production lines. Specific steps include: domain partitioning and initialization. Production lines with over 3 years of operation and more than 200 failure cases are designated as the "source domain," while newly established production lines with fewer than 50 failure cases are designated as the "target domain." A global model is initialized (a fault classification model based on ResNet-18, outputting fault severity levels 1-5), and distributed to local nodes on each production line. Each local node trains its model using local precursor feature data, employing a cross-entropy loss function (…). The actual fault level, For the predicted values, after training, the model parameters are uploaded to the cloud parameter server. The parameters are encrypted using elliptic curve cryptography to prevent parameter leakage. The cloud calculates the difference in feature distributions between the source and target domains using the Maximum Mean Difference (MMD) method. The specific analysis formula is as follows: in, Features of the source domain Features of the target domain.

[0028] Introducing a Domain Adaptive Network (DAN), the MMD (objective domain mismatch) is minimized through a gradient inversion layer (GRL). This approach maps the source and target domain features to the same feature space, resolving the "domain offset" problem caused by differences in operating conditions across production lines (e.g., the effective vibration threshold for production line A is 5 m / s², while for production line B it is 4.5 m / s²; the feature distributions of both are aligned using DAN). The parameter server aggregates the model parameters of each node to update the global model. For weak faults (Level 1 faults, feature signal-to-noise ratio <10 dB), a "feature enhancement + attention mechanism" is adopted. This mechanism enhances weak feature signals through wavelet packet transform and adds a channel attention layer to the model to focus on features highly correlated with the fault, thereby improving the identification rate of weak faults.

[0029] The dynamic maintenance strategy optimization module is a decision-making agent based on deep reinforcement learning. It sets the state space to include equipment health, production plan, resource inventory and personnel status, and evaluates the equipment health index through action space and reward function.

[0030] It should be specifically noted that the action space includes maintenance type decision, resource allocation decision, and personnel allocation decision. The maintenance type decision includes predictive maintenance, preventive maintenance, and post-failure repair. The reward function includes positive reward items and negative reward items. Positive reward items include health improvement rewards, production assurance rewards, resource efficiency rewards, and personnel efficiency rewards. Negative reward items include maintenance delay penalties and production loss penalties.

[0031] It should be further explained that the decision-making agent is built based on deep reinforcement learning (DRL), and the state space covers the entire dimension of "equipment-production-resources-personnel". Specific quantitative indicators include: equipment health. The LSTM-T hybrid model is used to predict the remaining useful life (RUL) and health status of the equipment. The range is 0-1, where 0 represents shutdown and 1 represents brand new. (For example, if the current RUL of the tightening machine bearing is 100h and the initial RUL is 500h, then...) (marked as "high risk"), production plan Including order priority ( High = 3, Medium = 2, Low = 1), Delivery time urgency ( Remaining days / Total construction period For emergency purposes, equipment production role (Key equipment = 2, standby equipment = 1, production line shutdown due to key equipment downtime) Current process progress Completed work / Total work (due to delays) resource inventory Spare parts inventory Current quantity / safety stock (Due to insufficient inventory) and spare parts replenishment cycle Number of days For long cycles) and maintenance tool availability : Number of available tools / Number required (For sufficient quantity), for example, bearing inventory. (Insufficient inventory), replenishment cycle If the resource status is "strained" (days or longer), then the personnel status is... Maintenance personnel skill level (Levels: High = 3, handles faults 1-5; Intermediate = 2, handles faults 1-3; Basic = 1, handles faults 1-2) Personnel workload. : Current number of tasks / Maximum number of tasks that can be carried (overload), personnel location Distance from the equipment per 100m (For close range).

[0032] The decision-making agent is designed using a proximal policy optimization algorithm, with the following core design: Action Space It includes three core actions and maintenance type decisions: predictive maintenance (for level 1-2 faults, performed during production breaks), preventive maintenance (for level 3 faults, scheduled for nighttime), and post-fault repair (for level 4-5 faults, performed with immediate shutdown). Resource allocation decisions: priority allocation of spare parts (e.g., allocating insufficient bearings to critical equipment) and tool scheduling (e.g., procuring dedicated torque wrenches from nearby workshops). Personnel allocation decisions: skill matching (senior technicians handle level 5 faults) and load balancing.

[0033] reward function Design a multi-objective weighted reward function to balance maintenance effectiveness and production efficiency: Positive Rewards: Health Improvement Rewards To maintain the increase in health afterward, Production assurance incentives: Prioritize the production of high-priority, urgent, and critical equipment, and reward efficient resource utilization. Encourage the use of readily available, short-lead-time spare parts, and reward efficient personnel: Encourage skill-matched and low-load personnel to participate; negative incentive: maintenance delay penalty. Production loss penalty: To avoid order delays due to maintenance, the total reward is: .

[0034] The equipment health index assessment breaks through the existing "single health" evaluation model and constructs a comprehensive index. : ,in, Weighting for device health , Historical failure frequency Weight , To maintain the effect ( Weight , Range 0-1, To be "excellent" For "good", The "Requires Attention" setting guides the maintenance priority ranking.

[0035] The multimodal human-machine collaboration module generates maintenance instructions and synchronously integrates the equipment's digital twin model, maintenance knowledge base, and real-time data to generate an enhanced maintenance guidance package that includes AR visualization guidance, 3D operation animations, standard operating procedures, and key parameter thresholds.

[0036] It should be specifically noted that the digital twin model of the equipment includes geometric information, physical information, and motion information, and the maintenance knowledge base includes structured data, unstructured data, and knowledge updates.

[0037] The AR visualization guidance includes fault location, operation guidance, and parameter monitoring; the 3D operation animation includes equipment breakdown animation and assembly animation; and the standard operating procedure includes basic information about the assembly equipment, step descriptions, and acceptance criteria.

[0038] It needs further explanation that the specific method for integrating the equipment digital twin model, maintenance knowledge base, and real-time data is as follows: A 1:1 high-fidelity twin model is constructed based on Unity3D, including geometric information such as component assembly relationships, dimensional parameters, and material properties. Physical information is linked in real-time with sensing module data. Equipment status and fault locations are dynamically labeled on the twin model. Motion information simulates equipment operation and maintenance actions. The maintenance knowledge base is constructed using a fusion of structured and unstructured data storage. The structured data includes 500... Historical repair cases are converted into a three-element group of "fault phenomenon - fault cause - repair steps - tools - spare parts - time" (e.g., "vibration kurtosis > 4 - insufficient bearing lubrication - add lithium-based grease - grease gun - grease type 3# - time 15min"). Unstructured data repair operation videos are played in slow motion to show key steps (e.g., wiring sequence when disassembling and assembling a motor), and expert voice guidance is provided (e.g., "tighten bolts in 3 stages, the first pre-tightening to 50% torque"). Knowledge updates use NLP technology to automatically parse new repair reports, and the knowledge base is updated monthly. Real-time data linkage uses the MQTT protocol to synchronize "suspected fault data" (e.g., vibration characteristics, temperature trends) from edge nodes and "urgency" of production plans (e.g., 2 days remaining for order delivery) to the guidance generation module, dynamically adjusting guidance priority (repair guidance for urgent orders is generated first).

[0039] An enhanced maintenance guidance package is generated, containing the following: AR Visualization Guidance: Based on Microsoft AR glasses, it achieves "virtual-reality" overlay, overlaying a red arrow on the real device to point to the fault location, overlaying virtual operation gestures, displaying the deviation between the current operation and standard procedures in real time, displaying device feedback parameters in real time on the AR interface, and issuing a red alarm when limits are exceeded; 3D Operation Animation: Creating animations of key steps, including breakdown animations demonstrating the disassembly sequence of components (e.g., "remove the motor end cover first - remove the bearing - clean the bearing housing"), marking common mistakes (e.g., "end cover bolts need to be disassembled diagonally to avoid deformation"); Assembly Animation: Demonstrating spare part installation requirements (e.g., "bearings need to be heated to 80℃ during installation, and cooled for 30 minutes after hot installation"), marking tolerance ranges (e.g., "bearing clearance 0.02-0.05mm"); and Standard Operating Procedures. The documentation uses Markdown format and includes basic information such as fault type, repair level, required tools (model + quantity), spare parts list (specifications + quantity), estimated time, step descriptions ("operation content - inspection points - anomaly handling"), acceptance criteria (requirements for post-repair equipment trial operation), and dynamic parameter thresholds (generated based on a fault feature library from a cross-production line migration learning module, including equipment-specific thresholds, normal thresholds, parameter ranges under healthy equipment conditions (e.g., "tightening machine torque 25-35Nm, current 5-8A"), alarm thresholds, parameter ranges indicating faults (e.g., "torque > 38Nm or < 22Nm, current > 10A or < 4A"), shutdown thresholds, and parameter ranges requiring immediate repair (e.g., "torque > 45Nm or < 18Nm, current > 12A or < 3A").

[0040] The enhanced execution terminal module receives the enhanced maintenance guidance package, collects maintenance operation data and equipment feedback parameters in real time, compares and verifies them with the standard operating procedures and key parameter thresholds in the guidance package in real time, and automatically triggers the equipment health status reassessment process.

[0041] It should be noted that the maintenance operation data includes operation process data and manually entered data, and the equipment feedback parameters include trial operation data and assembly quality data.

[0042] The real-time comparison and verification includes operation step verification, operation accuracy verification, and equipment parameter verification. The health status reassessment includes triggering conditions, assessment models, and application of assessment results.

[0043] It should be further explained that the deployment of the "AR glasses + handheld terminal + equipment sensor" three-in-one data acquisition terminal collects maintenance operation data and equipment feedback parameters. Through the AR glasses' camera and IMU sensor, the execution status of operation steps (such as "whether the bolts were disassembled in sequence"), operation accuracy, and operation time are collected. Maintenance personnel use the handheld terminal to record abnormal situations, spare parts replacement records, and tool usage. After maintenance, the equipment is run unloaded for 10 minutes and under load for 30 minutes to collect vibration (RMS value, kurtosis), temperature (real-time value, fluctuation amplitude), current (RMS value, THD), and acoustic (sound pressure level, spectrogram) parameters. Assembly interface parameters are collected through machine vision sensors, and tightening torque curves are collected through torque sensors.

[0044] The terminal has a built-in verification engine that performs multi-dimensional comparisons with the enhanced guidance package. This includes comparing the consistency between actual steps and SOP steps, comparing actual operating parameters with standard parameters, and comparing trial operation parameters with dynamic thresholds. It generates "qualified / unqualified / pending confirmation" reports. When unqualified, it automatically links to the knowledge base, recommends anomaly handling solutions, and automatically triggers a health status reassessment. This is triggered automatically after a qualified verification, or manually after a failed verification and anomaly handling. It calls the LSTM-T model of the dynamic maintenance module, inputs the equipment feedback parameters after maintenance, and recalculates the RUL and health level. ,like Upgrade to 0.8 or higher and update the device health index. The rating is "Excellent," and the repair records (steps, results) are stored in the knowledge base. If the improvement is less than 0.3, it is judged as "incomplete repair", triggering a secondary repair process (analyzing the reasons: such as spare parts quality problems, improper operation, and adjusting the guidance package). The evaluation results are synchronized to the cross-production line migration learning module to update the fault precursor feature library (such as "the repair characteristics of a certain type of fault can be used as new precursor features").

[0045] Secondly: The accompanying drawings of the embodiments disclosed in this invention only involve the structures involved in the embodiments disclosed in this invention. Other structures can refer to the general design. In the absence of conflict, the same embodiment and different embodiments of this invention can be combined with each other. In conclusion, the above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the protection scope of the present invention.

Claims

1. A health management system for assembly equipment based on the Industrial Internet of Things, characterized in that, include: The distributed sensing module deploys sensor clusters and intelligent edge nodes on the production line assembly equipment to collect and preprocess equipment operating status data at high frequency. The cross-production line transfer learning module establishes an encrypted global fault precursor feature library. Through federated transfer learning algorithms, it mines the correlation of weak fault features among similar equipment on different production lines and performs early fault warning based on swarm intelligence. The dynamic maintenance strategy optimization module is a decision-making agent based on deep reinforcement learning. It sets the state space to include equipment health, production plan, resource inventory and personnel status, and evaluates the equipment health index through action space and reward function. The multimodal human-machine collaboration module generates maintenance instructions and synchronously integrates the equipment's digital twin model, maintenance knowledge base, and real-time data to generate an enhanced maintenance guidance package that includes AR visualization guidance, 3D operation animation, standard operating procedures, and key parameter thresholds. The enhanced execution terminal module receives the enhanced maintenance guidance package, collects maintenance operation data and equipment feedback parameters in real time, compares and verifies them with the standard operating procedures and key parameter thresholds in the guidance package in real time, and automatically triggers the equipment health status reassessment process.

2. The assembly equipment health management system based on the Industrial Internet of Things as described in claim 1, characterized in that: The sensor cluster includes vibration sensors, temperature sensors, current sensors, acoustic sensors, machine vision sensors, and torque sensors. The operating status data includes time-domain peak / valley / mean / kurtosis / frequency-domain power spectral density (PSD), characteristic frequency, real-time temperature value, three-phase current RMS value, current harmonics, starting current peak value, sound pressure level and spectrogram, equipment image, tightening torque, and rotation angle.

3. The assembly equipment health management system based on the Industrial Internet of Things as described in claim 1, characterized in that: The preprocessing includes data cleaning, feature extraction, and data filtering. Data cleaning uses wavelet thresholding for noise reduction. Feature extraction includes time-domain features, frequency-domain features, and time-series features. Data filtering deploys a lightweight anomaly detection model to make a preliminary judgment on the preprocessed features.

4. The assembly equipment health management system based on the Industrial Internet of Things as described in claim 1, characterized in that: The global fault precursor feature library includes a three-layer structure: a basic feature layer, a precursor feature layer, and a fault feature layer. It adopts dual protection of "federated encryption + differential privacy". The federated transfer learning algorithm includes domain partitioning and initialization, local training, domain adaptation, global model update, and weak fault mining. Domain partitioning is to divide the production line into source domain and target domain based on equipment runtime and fault cases.

5. The assembly equipment health management system based on the Industrial Internet of Things as described in claim 1, characterized in that: The action space includes maintenance type decision, resource allocation decision, and personnel allocation decision. Maintenance type decision includes predictive maintenance, preventive maintenance, and post-failure repair. The reward function includes positive reward items and negative reward items. Positive reward items include health improvement reward, production assurance reward, resource efficiency reward, and personnel efficiency reward. Negative reward items include maintenance delay penalty and production loss penalty.

6. The assembly equipment health management system based on the Industrial Internet of Things as described in claim 1, characterized in that: The digital twin model of the equipment includes geometric information, physical information, and motion information, while the maintenance knowledge base includes structured data, unstructured data, and knowledge updates.

7. The assembly equipment health management system based on the Industrial Internet of Things as described in claim 1, characterized in that: The AR visualization guidance includes fault location, operation guidance, and parameter monitoring; the 3D operation animation includes equipment breakdown animation and assembly animation; and the standard operating procedure includes basic information about the assembly equipment, step descriptions, and acceptance criteria.

8. The assembly equipment health management system based on the Industrial Internet of Things as described in claim 1, characterized in that: The maintenance operation data includes operation process data and manually entered data, and the equipment feedback parameters include trial operation data and assembly quality data.

9. The assembly equipment health management system based on the Industrial Internet of Things as described in claim 1, characterized in that: The real-time comparison and verification includes operation step verification, operation accuracy verification, and equipment parameter verification. The health status reassessment includes triggering conditions, assessment models, and application of assessment results.

Images