Wheel hub processing multi-station collaborative control system based on industrial internet

Abstract

The application discloses a kind of based on industrial internet's wheel hub processing multi-station collaborative control system, it is related to wheel hub processing field;Multi-source data acquisition module accurately collects multiple parameters and spatiotemporal alignment, digital twin modeling module simulates processing and optimizes parameter, multi-station collaborative scheduling module uses quantum heuristic algorithm to improve equipment utilization, quality closed loop control module detects defect and automatically repairs, and industrial internet interface module interfaces enterprise system to realize whole-process digital management.The application improves wheel hub processing collaborative efficiency, reduces order postponement rate;Strengthen quality control, improve yield rate;Optimize equipment health, reduce downtime and unit energy consumption;Efficiency operation and maintenance, reduce cost, support low-carbon certification, promote intelligent transformation.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of hub processing, and in particular to a hub processing multi-station collaborative control system based on an industrial internet. BACKGROUND

[0002] Hub processing is a key link in automobile manufacturing, involving multi-station collaborative operation of lathes, milling machines, grinding machines, etc., with high process complexity and strict quality requirements. Traditional hub processing relies on manual experience for equipment control and station scheduling, facing problems such as low matching efficiency of processing parameters, insufficient multi-device collaboration, etc. For example, information transmission between different stations lags, often leading to repeated processing paths or idle equipment, making it difficult to meet the flexible production demand of multiple varieties and small batches. At the same time, equipment state monitoring relies on regular inspection and cannot capture subtle abnormalities such as cutting force mutations and tool wear in real time, easily causing processing precision to decline or even equipment failure, affecting production continuity.

[0003] With the development of industrial internet technology, some enterprises have tried to introduce Internet of Things sensors to collect equipment data, but problems such as multi-source data time and space asynchronization and non-uniform protocols are prominent. For example, the difference in collection frequency between vibration sensors and visual detection equipment leads to fragmented data, making it difficult to build a complete processing process model. In addition, existing systems lack intelligent scheduling algorithms for cross-station collaboration, and in the face of emergency order insertion or equipment failure, manual rescheduling takes as long as several hours, leading to high order delay rate. In terms of process optimization, traditional trial-and-error methods rely on technical personnel's experience to adjust cutting parameters, and the introduction of new products takes several weeks, which cannot meet the market's rapidly changing demand.

[0004] In terms of green manufacturing and cost control, the traditional processing mode lacks correlation analysis of energy consumption and process parameters, and the unit energy consumption is high. At the same time, quality detection relies on manual sampling inspection, with a high rate of missed detection, and the cost of waste product disposal is expensive. With the implementation of international standards such as carbon tariffs, enterprises urgently need to establish a green manufacturing system covering energy consumption monitoring and carbon footprint tracing, but existing systems cannot provide full-process low-carbon process optimization support. The present application integrates multi-station data through industrial internet technology to build an intelligent collaborative control system, solving the core pain points of traditional processing modes in terms of efficiency, quality, energy consumption, etc. SUMMARY

[0005] The hub processing multi-station collaborative control system based on an industrial internet is proposed to solve the problems mentioned in the prior art.

[0006] To achieve the above purpose, the present application adopts the following technical scheme: a hub processing multi-station collaborative control system based on an industrial internet, comprising:

[0007] Multi-source data acquisition module: Deploy distributed edge nodes on equipment, synchronize time through IEEE 1588v2 protocol, collect parameters with dual-antenna RTK-GNSS spatiotemporal alignment, and use compressed sensing technology to sparsely reconstruct vibration signals.

[0008] Digital twin modeling module: Introduce simplified machining error formula Build dynamic twin, E is the simulation and actual machining error, k is the process correction coefficient, The difference between the theoretical and actual cutting force is, The material property matching deviation; Integrate physical engine to simulate cutting force-thermal coupling process, optimize tool path through reinforcement learning algorithm; AR real-time superposition, adjust twin model parameters through gestures, and synchronously drive physical equipment;

[0009] Multi-station collaborative scheduling module: Use quantum annealing algorithm to solve multi-objective optimization problem, and construct scheduling performance formula , P is the comprehensive scheduling performance, α is the algorithm efficiency weight, T QA is the quantum annealing algorithm solving time, T GA is the traditional genetic algorithm solving time, β is the energy optimization weight, E min is the minimum energy consumption of current scheduling, E avg is the historical average energy consumption, dynamically schedule stations; Establish a processing task "digital token" circulation mechanism, and automatically execute process handover through smart contract;

[0010] Quality closed-loop control module: Deploy line laser radar and hyperspectral camera, and construct "digital fingerprint" on hub surface; Use Transformer visual model to identify microscopic defects, combine knowledge graph to infer defect causes, trigger robot to perform laser repair, and form "detection-diagnosis-repair" closed loop;

[0011] Industrial Internet interface module: Based on digital thread technology, connect ERP / MES / PLC systems, establish process parameter-energy consumption-carbon emission correlation model; Apply Bayesian optimization algorithm to dynamically adjust cutting parameters, trace carbon footprint in real time, and generate ISO14067 standard report.

[0012] Further, it also includes:

[0013] Equipment health management module: Process edge node vibration data with differential privacy, aggregate cross-enterprise fault features through federated transfer learning, and train to generate general equipment health index GHI model; When the GHI of the station bearing exceeds the threshold, the system automatically calls the digital twin to reproduce the fault, predicts the probability of failure, and automatically replenishes the spare parts supply chain.

[0014] Further, it also includes:

[0015] Immersion operation module: build a digital twin of the machining scene, remotely control the equipment through a VR terminal; use a haptic feedback glove to simulate tool contact force and debug virtual clamping; combine digital twin mirror image rehearsal function to optimize changeover time and key operation compliance rate.

[0016] Further, the multi-source heterogeneous data acquisition module uses photoelectric composite sensor fusion technology to synchronously measure the cutting temperature field and tool displacement, and eliminates noise interference through a space-time joint filtering algorithm.

[0017] Further, the digital twin evolution module establishes a tool wear prediction LSTM-Attention model and combines historical machining parameter data to predict tool remaining life.

[0018] Further, the multi-station collaborative scheduling module introduces the concept of dynamic division of "machining domain", classifies similar process hubs into the same virtual machining domain, and schedules through domain resource pooling.

[0019] Further, the quality self-recovery control module develops a defect generalization detection model, identifies new types of defects through samples, and optimizes the model update cycle.

[0020] Further, the industrial internet interface module integrates a blockchain storage function, records the execution process of a low-carbon process scheme on the chain, and forms a green manufacturing certificate.

[0021] Further, the equipment health management module designs a cross-enterprise data contribution incentive mechanism, and enterprises can obtain "computing power points" by sharing effective fault data and exchange cloud AI training resources.

[0022] Further, the immersion operation module uses twin scene construction technology and combines eye tracking technology to automatically focus on key parameters.

[0023] Compared with the prior art, the beneficial effects of the present application are:

[0024] Through the quantum heuristic scheduling engine, the machining path is optimized, the order is dynamically allocated to the optimal station, multiple orders are processed in parallel, and the equipment utilization is optimized. In view of the dynamic disturbance (such as equipment failure) in the machining process, the system can complete scheduling adjustment in a very short time through a local search algorithm, and the order delay is reduced.

[0025] The multi-source data acquisition module with accurate space-time synchronization can collect high-frequency machining parameters and equipment states, and combine the digital twin model driven by physical constraints to keep the error between virtual machining and actual machining within a small range. The quality self-recovery control module with cognitive enhancement can identify fine cracks with the help of line laser radar and hyperspectral cameras, effectively improving the yield rate. The automatic repair function builds a "detection-diagnosis-repair" closed loop, reduces the scrap rate, and significantly reduces the cost of manual detection.

[0026] The federal learning driven device health management module aggregates cross-enterprise fault data, can early warn potential faults such as bearing wear, and reduces unplanned downtime. The energy-process collaborative optimization module dynamically adjusts the cutting parameters by using the Bayesian algorithm to reduce unit energy consumption, and integrates blockchain technology to realize real-time tracing of carbon footprint, generate ISO14067 standard reports, and help enterprises meet international low-carbon certification requirements and obtain carbon trading income.

[0027] The 5G-MEC enabled immersive operation and maintenance module relies on VR / AR technology to carry out virtual clamping debugging and real-time operation guidance, helps operation and maintenance personnel improve decision-making efficiency, and also shortens the training period of new employees. The system is fully automated, reduces manual intervention, reduces operation and maintenance costs, and at the same time, through data sharing and process knowledge sedimentation, shortens the new process debugging period, and builds a low-cost and efficient intelligent production system for enterprises.

[0028] The industrial internet middle platform integrates ERP / MES / PLC systems, builds process parameter-quality-energy consumption correlation models, supports similar process intelligent recommendation and cross-platform data interaction. Enterprises can improve the first piece qualification rate and shorten the new product introduction cycle by using digital twin pre-rehearsal functions. The cross-enterprise data sharing mechanism improves the accuracy of fault diagnosis through federal learning and incentive mechanism, and forms a benign ecological environment for industry cooperation and innovation.

[0029] Through technical innovation and mode reconstruction, the present application promotes the transformation of hub processing from "experience-driven" to "data-driven", and provides an efficient, accurate and green intelligent solution for the automobile parts manufacturing industry. BRIEF DESCRIPTION OF DRAWINGS

[0030] Figure 1 A schematic block diagram of a hub processing multi-station collaborative control system based on an industrial internet is provided for the present application;

[0031] Figure 2 A performance comparison diagram for a multi-source heterogeneous data acquisition module is provided;

[0032] Figure 3 A quantum heuristic scheduling engine optimization effect comparison diagram is provided;

[0033] Figure 4 A quality closed-loop control module defect detection rate comparison diagram is provided;

[0034] Figure 5 An energy-process collaborative optimization effect diagram is provided;

[0035] Figure 6 A comprehensive benefit comparison diagram before and after the implementation of the system is provided. DETAILED DESCRIPTION

[0036] With reference to the drawings of the embodiments of the present application, the technical solutions in the embodiments of the present application will be clearly and completely described. Obviously, the described embodiments are only a part of the embodiments of the present application, rather than all the embodiments of the present application. Based on the embodiments of the present application, all the other embodiments obtained by a person of ordinary skill in the art without creative work fall within the protection scope of the present application.

[0037] In the description of the present application, it should be understood that the terms "center", "longitudinal", "transverse", "length", "width", "thickness", "upper", "lower", "front", "back", "left", "right", "vertical", "horizontal", "top", "bottom", "inner", "outer", "clockwise", "counterclockwise" and the like indicate the orientation or positional relationship shown in the drawings, and are only used to facilitate the description of the present application and simplify the description, and do not indicate or imply that the device or element referred to must have a particular orientation, be constructed and operated in a particular orientation, and therefore cannot be understood as a limitation of the present application.

[0038] In addition, the terms "first", "second" are only used for description purpose, and cannot be understood as indicating or implying relative importance or implicitly indicating the number of the technical features indicated. Therefore, the features defined with "first", "second" can explicitly or implicitly include one or more of the features. In the description of the present application, the meaning of "multiple" is two or more, unless otherwise specifically limited. In addition, the terms "mounting", "connecting", "connection" should be broadly understood, for example, it can be fixed connection, or detachable connection, or integral connection; it can be mechanical connection, or electrical connection; it can be direct connection, or indirect connection through intermediate medium, or internal communication of two elements. For a person of ordinary skill in the art, the specific meaning of the above terms in the present application can be understood according to the specific circumstances, and the present application will be further described in detail with reference to the drawings.

[0039] With reference to Figures 1 to 6 A hub processing multi-station collaborative control system based on industrial internet, comprising:

[0040] Multi-source data acquisition module: Deploy distributed edge nodes on key equipment such as lathes, milling machines, and grinding machines in the wheel hub processing workshop. Use a customized acquisition terminal (model: DX-Edge8000) with ARM Cortex-A72+FPGA architecture. The terminal integrates the IEEE1588v2 precision clock synchronization protocol stack, and through the fiber ring network, the clock deviation of the equipment in the workshop is ≤100ns. Install three-axis acceleration sensors (sensitivity 100mV / g, range ±50g) on the key parts of the main shaft, combined with fiber Bragg temperature sensors (measurement range -40~200℃, accuracy ±1℃), to realize synchronous acquisition of vibration and temperature data. Tool condition monitoring uses an industrial-grade line laser displacement sensor (model: Keyence LK-G5000, measurement accuracy ±2μm) to measure tool wear in real time, and a high-resolution industrial camera (Basler ace2, 20 million pixels) is deployed to capture images of the cutting area. All sensor data are preprocessed by the edge node, and the compressed sensing (CS) algorithm is used to sparsely sample the vibration signal, reducing the original sampling rate of 20kHz to 8kHz while maintaining the integrity of the signal features. The preprocessed data are encrypted and transmitted to the workshop MEC server through the MQTT-SN protocol, reducing transmission bandwidth usage by 60% and end-to-end delay ≤20ms.

[0041] Digital twin modeling module: Introduce a simplified machining error formula Construct a dynamic twin body, E is the simulation and actual machining error, k is the process correction coefficient, is the difference between the theoretical and actual cutting force, is the material property matching deviation. Based on the Python+PyTorch framework, a wheel hub processing full-process digital twin is constructed, a parameterized modeling method is used to create a wheel hub three-dimensional model, and the material properties are set to aluminum alloy 6061-T6 (elastic modulus 68.9GPa, Poisson's ratio 0.33). The BulletPhysics physics engine is integrated to simulate the cutting process, and the contact stiffness of the tool and workpiece is set to 1×10 7N / m, friction coefficient 0.3. In the cutting force simulation, the Johnson-Cook constitutive model is considered, and the strain rate sensitivity coefficient C=0.015 and the thermal softening coefficient m=1.02 are set. The digital twin interacts with the physical device in real time through the OPCUA protocol, and updates the state data every 50 ms. The PPO (Proximal Policy Optimization) reinforcement learning algorithm is used, and the reward function is set as the weighted sum of the machining time, surface roughness and tool wear (weights are 0.5, 0.3 and 0.2 respectively). During the training process, the agent performs 100,000 iterations of learning in the digital twin environment, and finally generates the optimal machining strategy. A Unity3D application is developed to realize the virtual-real fusion display of the digital twin through Hololens2. The operator can adjust the cutting parameters through gestures, and the system updates the twin model in real time and feeds back the response of the physical device. For example, when the operator adjusts the feed rate from 0.2 mm / r to 0.15 mm / r in the AR environment, the digital twin displays that the cutting force is reduced by 18%, and the surface roughness Ra value is improved from 1.2 μm to 0.8 μm, while the physical device synchronously executes the parameter adjustment.

[0042] Multi-station collaborative scheduling module: based on D-Wave quantum annealing processor to construct scheduling optimization model, convert multi-station collaborative scheduling problem into quadratic unconstrained binary optimization (QUBO) problem. Construct scheduling performance formula , P is the comprehensive scheduling performance, α is the algorithm efficiency weight, T QA is the quantum annealing algorithm solving time, T GA is the traditional genetic algorithm solving time, β is the energy consumption optimization weight, E min is the minimum energy consumption of current scheduling, E avg is the historical average energy consumption. Set each task can only be assigned to one station, equipment processing capacity constraint and process sequence constraint. Map the problem to the quantum bits of the quantum processor, and find the optimal solution through the quantum annealing process. Process dynamic disturbance (such as equipment failure, emergency order insertion), design two-stage scheduling strategy. The first stage generates an initial scheduling scheme based on quantum algorithm; the second stage, when disturbance occurs during execution, uses a local search algorithm to quickly adjust the affected tasks, with an adjustment time ≤100 ms.

[0043] Quality closed-loop control module: A detection system composed of a line laser radar (model: SICK LMS511, scanning frequency 50 Hz, angle resolution 0.25°) and a hyperspectral camera (Headwall Nano-Hyperspec, spectral range 400-1000 nm, spectral resolution 5 nm) is deployed at the end of the hub machining line. The line laser radar is used to measure the hub geometric dimensions (roundness, cylindricity, etc.), with a measurement accuracy of ±5 μm; the hyperspectral camera is used for surface defect detection, which can identify micro-cracks ≤0.2 mm. A deep learning detection model based on VisionTransformer (ViT) is developed, which is fine-tuned using 5000 hub defect samples based on ImageNet pre-training. The model input is multi-modal data (point cloud + hyperspectral image), and the output is a 12-class defect probability distribution. To solve the small sample problem, the meta-learning (MAML) algorithm is used, which can quickly adapt to only 5 new type defect samples, and the model update cycle is shortened from 7 days to 2 hours. When a defect is detected, the system infers the cause of the defect through a knowledge graph. For example, when fish-scale-like lines are detected on the hub surface, the system retrieves the knowledge base and associates it with possible causes such as too low cutting speed and too large feed rate, and further analyzes the historical process parameters to determine the most likely cause. Then, the system automatically generates a repair scheme, and a six-axis robot (repeat positioning accuracy ±0.03 mm) carrying a laser repair head is used to repair the defect, and after repair, the defect is detected again to form a closed-loop control.

[0044] Industrial Internet interface module: Based on the Kubernetes container orchestration platform, an industrial internet middle platform is built, and a micro-service architecture is used to decouple each functional module. The middle platform integrates Apache Kafka message queue (throughput 100,000 pieces / second) to process real-time data streams, uses Elasticsearch to store time series data (supports PB-level data storage), and uses Apache Spark for batch data analysis. An associated model of process parameters-energy consumption-carbon emissions is established, 1000 groups of energy consumption data under different process parameters are collected, and a random forest algorithm is used to train the parameter-energy consumption mapping model, with a model determination coefficient R 2 =0.92. In actual production, when a new order is received, the system predicts the energy consumption and carbon emissions of different process schemes based on the model, and uses a Bayesian optimization algorithm to find the optimal parameter combination. To realize carbon footprint tracing, the Hyperledger Fabric blockchain platform is used to record the production process data (raw material source, machining process, energy consumption, etc.) of each hub. When a customer needs a carbon footprint report, the system automatically generates a document that meets the ISO 14067 standard, and the report generation time is shortened from 3 days to 2 hours.

[0045] In the present application, the following modules are also included:

[0046] Device health management module: TensorFlowLite lightweight framework is deployed on each enterprise edge node to preprocess vibration data. Differential privacy technology is used to add Laplace noise to feature vectors to protect enterprise data privacy. Each node is connected to the federated learning server through the MQTT protocol, and the server aggregates model parameters using the FedAvg algorithm. In cross-enterprise fault feature learning, device fault data from 5 hub manufacturing enterprises (including 8 types of faults such as bearing failure, tool wear, and spindle imbalance) are collected. Each enterprise retains 70% of the data for local training and 30% for testing. After 100 rounds of federated learning iterations, the average accuracy of the general device health index (GHI) model on each enterprise test set reaches 92%. When the GHI value of a device in an enterprise exceeds the warning threshold (e.g., 0.7), the system automatically calls the digital twin of the device to reproduce the fault. For example, when the GHI value of a lathe spindle is detected to be 0.85, the system simulates different degrees of imbalance fault in the digital twin environment, compares the actual vibration characteristics, and determines that the fault is caused by a spindle dynamic balance deviation of 20g・mm. The system then generates a maintenance work order, recommends a balancing weight scheme, and automatically allocates balancing blocks through the spare parts supply chain.

[0047] In the present application, the following modules are also included:

[0048] Immersive operation and maintenance module: A 5G private network (frequency band 3.5GHz, bandwidth 100MHz) is deployed in the workshop, using ultra-dense networking (UDN) technology to achieve seamless coverage of 5G signals within the workshop, with a downlink rate ≥1Gbps, an uplink rate ≥100Mbps, and an air interface delay ≤10ms. The MEC server is deployed on the edge of the workshop, within 50 meters of the device, to achieve low-latency data processing. An immersive operation and maintenance application based on Unity3D is developed, supporting the HTC Vive Focus 3 VR headset and CyberGlove III haptic feedback gloves. After wearing the VR device, the operator can enter a 1:1 scale virtual workshop environment and operate the virtual control panel through hand gestures. The haptic feedback gloves can simulate force feedback of 0.1N-10N, allowing the operator to feel the real operating resistance in the virtual environment. In complex equipment changeover scenarios, the system first rehearses the changeover process in the digital twin environment and generates the optimal operation scheme. The operator completes the virtual changeover operation in the VR environment according to the instructions, and the system records the operation process and evaluates compliance. During actual changeover, the AR glasses (HoloLens 2) superimpose virtual operation steps on the real equipment to guide the operator to complete the changeover. For example, when switching from a 16-inch to an 18-inch hub production line, the system guides the operator to adjust the chuck position (accuracy ±0.5mm) and replace the tool (torque control ±5%).

[0049] In the application, the multi-source heterogeneous data acquisition module adopts photoelectric composite sensor fusion technology (fiber grating + vision). The fiber grating sensor uses the characteristic of grid period change caused by temperature to monitor the temperature field of the cutting area with an accuracy of ±1℃, and constructs a millimeter-level spatial resolution temperature model; the vision sensor relies on a high-frame-rate camera, a sub-pixel algorithm, and camera calibration parameters to realize tool displacement measurement with an accuracy of ±0.5μm, covering multi-dimensional displacement information, and the two complement each other. A space-time joint filtering algorithm (STF) is introduced to construct a two-dimensional space-time framework. The time dimension is predicted by Kalman filtering time series, and the space dimension is constructed by a neighborhood model based on cutting physical constraints. After space-time iterative optimization, the signal-to-noise ratio is improved by 15dB, and the data reliability is ensured. A high-precision space-time synchronization unit is provided, which realizes sub-microsecond-level acquisition time synchronization based on IEEE1588v2 protocol and hardware calibration, ensures the space-time alignment of data; and through PoE and EtherCAT protocols, a high-speed transmission channel is constructed, which lays a solid data foundation for real-time decision-making and closed-loop control of the system, and helps the industrial manufacturing to be intelligent and refined.

[0050] In the application, the digital twin evolution module focuses on tool life cycle management, and constructs an LSTM-Attention fusion model to realize intelligent wear prediction. The model integrates the advantages of long short-term memory network (LSTM) and attention mechanism (Attention), LSTM uses input, forget and output gate structure to deeply mine the long time series dependence of 180-day processing full historical data (covering cutting speed, feed rate, spindle load, cutting temperature and other multi-dimensional parameters), and accurately stores the wear influence information at different processing stages; the Attention mechanism automatically strengthens the weight distribution of key parameters (such as temperature fluctuation in high-speed cutting) and time intervals (run-in period, recession period, etc.) that are strongly related to tool wear, solving the problem of insufficient key information mining of traditional models. The data processing process is progressive, first, the 3σ principle and the isolation forest algorithm are used to eliminate sensor outliers, and then Z-Score normalization is used to unify the parameter dimension; then, a time series sample set containing the actual tool wear label (detected offline by an electron microscope and a roughness meter) is constructed by sliding block with a fixed window of 180 days, laying a solid data foundation for model training. In application, the trained model is connected to the real-time processing system to infer online after pre-processing real-time parameters, and the tool remaining life is predicted 5 processing cycles in advance, with an error of ≤5%, which promotes the innovation of tool replacement strategy from “periodic / fault replacement” to “on-demand precise replacement”, avoiding the risk of precision decline and workpiece scrap caused by excessive tool wear, and reducing the waste of resources caused by early tool replacement. Practice has proved that the tool replacement cost is directly reduced by 30%. In addition, the model is equipped with an online learning mechanism, which continuously optimizes parameters with new data input, adapts to various working conditions and tool types, provides long-term and accurate tool operation support for intelligent processing driven by digital twin, and helps the manufacturing process to be intelligent and efficient.

[0051] In the present application, the multi-station cooperative scheduling engine introduces the concept of dynamic division of "processing domain". Relying on process feature recognition algorithm, the cutting parameters, tool configuration and other dimensional data of hub processing technology are extracted, the process similarity is calculated by deep learning similarity measurement model, and the threshold is classified into the same virtual processing domain. The equipment working condition is also dynamically adjusted. When scheduling, the resource space is constructed by quantum bit coding in the domain, and the quantum annealing algorithm is used to solve the optimal matching of resources and tasks. The goal is to maximize efficiency, constrain load, delivery time, etc., so that the processing efficiency of similar workpieces is improved by 40%, and the time-consuming of equipment idling is reduced. Cross-domain switching relies on digital twin pre-play model to simulate the whole process, optimize the path and timing with reinforcement learning, and generate the optimal strategy. The integration of edge nodes synchronizes parameters to guide execution, reducing cross-domain switching time by 70%. At the same time, data is collected through the Internet of Things, and the feedback algorithm is calculated through edge computing to dynamically adjust the strategy, forming a closed-loop system to adapt to production disturbances and help enterprises upgrade flexible production.

[0052] In the present application, the quality self-recovery control module aims at the quality control problem of small-batch customized production, and integrates meta-learning technology to build a defect generalization detection system. Model construction relies on the Meta-Learning framework and uses the Model-Agnostic Meta-Learning (MAML) algorithm. In the pre-training stage, collect multiple types of defect data in the whole process chain of the hub, build a meta-training set, let the model learn the general rules of defect features and the abstract knowledge associated with process parameters, and have the "meta-knowledge" basis for quickly adapting to new defect learning. Only 5 new defect samples are needed to fine-tune with gradient update strategy to identify new types of defects, breaking through the sample dependence bottleneck of traditional models. Data processing integrates multiple sensing devices to collect image, three-dimensional profile, temperature field, and multi-dimensional data such as processing and environment, which are preprocessed (image enhancement, point cloud optimization, time series feature construction) and input into the meta-learning model. A joint loss function (cross-entropy + contrastive learning loss) is used to strengthen the defect feature discrimination and generalization ability, and accurately capture the subtle differences of defects. In practical application, the model is deployed in a lightweight manner through edge computing, and the online learning algorithm reduces the update cycle from 7 days to 2 hours. In the face of small-batch customized detection, the "meta-knowledge" is used to adapt to the defect feature of customized process, combined with multi-modal data, to distinguish real defects from process differences, with an accuracy of 97%. At the same time, the closed-loop feedback mechanism returns the detection and production correction data back to the model for iterative optimization, providing intelligent support for quality self-recovery throughout the link, and promoting the high-quality development of customized manufacturing.

[0053] In the application, the industrial internet interface module focuses on the integration of green manufacturing and carbon management, and integrates the Hyperledger Fabric blockchain storage function. Based on the characteristics of modular architecture and PBFT consensus, carbon data storage is adapted. For low-carbon processes such as dry cutting, the whole process of planning, execution and detection is disassembled, and multi-dimensional data such as cutting parameters, equipment energy consumption and quality are collected. After edge gateway collection and preprocessing, ECDSA signature tagging is used, and the channel strategy is entered into the "green manufacturing channel". The intelligent contract captures the key nodes, and after hash calculation, it is chained. After PBFT consensus, it generates an unalterable storage certificate in seconds. Green manufacturing certificates rely on storage data, and according to ISO14064 standards, carbon emissions are automatically calculated. Carbon trading platform cross-chain query verification helps enterprises to convert income, and can also be connected to audit agencies for auditing. At the same time, the middle platform uses visual board to monitor data, combined with reinforcement learning algorithm to mine and analyze, optimize process parameters, realize the value jump from storage to closed-loop optimization, and promote the digital and green transformation of enterprises.

[0054] In the application, the equipment health management module designs a cross-enterprise data contribution incentive mechanism. Based on the federal learning "data available but invisible", it is clear that "effective fault data" must contain equipment information, running data at the time of failure, diagnosis conclusion and repair scheme, and JSON-Schema format specification and quality check rules are developed. Set "computing power points" as the incentive medium, and enterprises will get points for sharing 100 standard data. High-value equipment and rare fault data enjoy a 1.2-1.5 times coefficient. Points can be exchanged for multi-dimensional cloud AI resources: high-performance computing resources, GPU training time allocated according to points; high-quality training data sets, extract the same type of data after differential privacy processing; advanced AI model framework, exchange pre-trained model fine-tuning permissions. Relying on Hyperledger Fabric blockchain storage, set up a data sharing smart contract, and automatically issue points after uploading and verification. A dynamic adjustment mechanism is built to adjust the exchange ratio and package according to resource popularity and sharing activity. In practical application, the data sharing activity is improved by 300%, small and medium-sized enterprises make up for the shortcomings, and large enterprises expand the sample. The federal learning model absorbs multi-dimensional data, and the early fault identification accuracy rate is increased to 92%, and the remaining life prediction error is reduced to ±8%, promoting the upgrading of equipment health management in the industry.

[0055] In the application, the immersive operation and maintenance module adopts a full-process scheme based on physical rendering (PBR). The device and scene data are collected by a laser radar scanner (accuracy ±0.5mm), and after noise reduction and gridding processing, they are imported into a PBR material editing system to simulate the interaction between light and material, so that the material reflectivity error is ≤2%, and the real-time global lighting technology is integrated to ensure 60fps light rendering and create a realistic visual experience. A high-precision eye tracking device (accuracy ±1°) is installed to analyze eye movement data, and combined with the device operation and maintenance knowledge graph, the key parameters are intelligently focused. For example, when the operator focuses on the bearing, the system automatically locates and displays parameters such as temperature and vibration, simplifying the information acquisition process. The 5G network ensures large bandwidth and low latency data transmission, and the MEC node sinks the computing task to the edge side, so that the action instruction processing and scene feedback are completed within 20ms, which, combined with stable rendering frame rate, creates a "what you see is what you get" experience. In practical application, the scene realism reduces the operation and maintenance judgment error to ≤3%, the eye tracking shortens the information acquisition time by 70%, the 5G-MEC improves the decision-making efficiency by 50%, helps enterprises to reduce the equipment downtime by more than 15%, reduces the operation and maintenance cost by 20%, and promotes the upgrading of operation and maintenance mode.

[0056] The above is only a preferred specific embodiment of the present application, but the protection scope of the present application is not limited thereto, and any skilled person in the art can make equivalent replacement or change within the technical range disclosed by the present application according to the technical solution and inventive concept of the present application, which should be covered within the protection scope of the present application.

Claims

1. An industrial internet-based multi-station collaborative control system for hub machining, characterized in that, Comprise the following modules: Multi-source data acquisition module: Deploy distributed edge nodes on equipment, synchronize time through IEEE 1588v2 protocol, collect parameters through space-time alignment of dual-antenna positioning, and use compressed sensing technology to sparsely reconstruct vibration signals; Digital twin modeling module: Introduce a simplified machining error formula to build a dynamic twin, integrate a physical engine to simulate the cutting force-thermal coupling process, and optimize the tool path through reinforcement learning algorithms; AR real-time superposition, adjust the twin model parameters through gestures, and synchronize the driving of physical equipment; Multi-station collaborative scheduling module: Use quantum annealing algorithm to solve multi-objective optimization problem, construct scheduling performance formula, establish processing task "digital token" circulation mechanism, and automatically execute process handover through smart contract; Quality closed-loop control module: Deploy line laser radar and hyperspectral camera to build "digital fingerprint" on hub surface; use visual model to identify microscopic defects, infer defect causes with knowledge graph, trigger robot for laser repair, and form "detection-diagnosis-repair" closed loop; Industrial Internet interface module: Based on digital thread technology, connect ERP or MES or PLC systems, establish process parameter-energy consumption-carbon emission correlation model; apply Bayesian optimization algorithm to dynamically adjust cutting parameters, trace carbon footprint in real time, and generate standard report; Multi-source heterogeneous data acquisition module uses photoelectric composite sensor fusion technology to simultaneously measure cutting temperature field and tool displacement, and uses space-time joint filtering algorithm to eliminate noise interference; Digital twin evolution module establishes a tool wear prediction LSTM-Attention model, and predicts the remaining life of the tool based on historical processing parameter data; Quality self-healing control module develops a defect generalization detection model to identify new types of defects through samples and optimize the model update period; Industrial Internet interface module integrates blockchain storage function, records the execution process of low-carbon process scheme on the chain, and forms green manufacturing credentials. 2.The IIO-based hub machining multi-station collaborative control system according to claim 1, wherein, Also include: Equipment health management module: Differential privacy processing edge node vibration data, aggregating cross-enterprise fault features through federated transfer learning, training to generate general equipment health index GHI model; When the GHI of the workbench bearing exceeds the threshold, the system automatically calls the digital twin to reproduce the fault, predicts the probability of failure, and automatically replenishes the spare parts supply chain. 3.The IIO-based hub machining multi-station collaborative control system according to claim 1, wherein, Also include: Immersive operation and maintenance module: Build a digital twin of the machining site and remotely control the equipment through a VR terminal; Use haptic feedback gloves to simulate tool contact force and debug virtual clamping; Optimize changeover time and key operation compliance rate through digital Liang mirror image rehearsal function.

4. The IBB-based hub machining multi-station collaborative control system according to claim 1, characterized in that, Multi-station collaborative scheduling module introduces the concept of "machining domain" dynamic division, similar processes are classified into the same virtual machining domain through resource pooling scheduling within the domain. 5.The IIB-based hub machining multi-station collaborative control system according to claim 2, wherein, Equipment health management module designs a cross-enterprise data contribution incentive mechanism, and enterprises can earn "computing power points" by sharing effective fault data to exchange cloud AI training resources. 6.The IIB-based hub machining multi-station collaborative control system according to claim 3, wherein, Immersive operation and maintenance module uses twin scene construction technology combined with eye tracking technology to automatically focus on key parameters.

Images