A special bolt-based full-life-cycle intelligent manufacturing system

Abstract

This invention discloses an intelligent manufacturing system based on the entire lifecycle of special bolts, belonging to the field of intelligent manufacturing technology. It includes a digital twin construction module for receiving orders, calling a pre-set working condition-performance mapping model, generating initial performance targets based on the target working condition information in the order, and creating an initial digital twin with a unique digital identifier for each order. The initial digital twin integrates design specifications, material properties, and the initial performance targets. This invention significantly improves product performance consistency, manufacturing efficiency, and resource utilization by constructing a digital twin and intelligent closed-loop system covering the entire lifecycle of special bolts. It achieves transparency in the manufacturing process and precise control over product quality. This effectively avoids early failures caused by improper processes, reduces rework and scrap, and thus lowers production costs and delivery cycles while meeting high-performance indicators.

Description

Technical Field

[0001] This invention relates to the field of intelligent manufacturing technology, specifically to an intelligent manufacturing system based on the entire life cycle of special bolts. Background Technology

[0002] In the field of high-end equipment manufacturing, the reliability of special bolts is directly related to the safety of the overall structure. Currently, the core contradiction facing intelligent manufacturing lies in the disconnect between data flow and decision-making flow. On the one hand, the actual service performance of bolts under extreme working conditions (such as stress distribution under dynamic loads and material degradation rate in corrosive environments), as the most critical feedback for optimized design and processes, often forms data silos due to a lack of online monitoring methods or scattered and isolated data, making it impossible to effectively collect and aggregate high-value information. On the other hand, the process knowledge base at the manufacturing end is unable to receive and digest this real service data, remaining in a static and universal state for a long time. Its optimization and iteration are severely lagging behind, relying solely on costly physical experiments and expert experience, which restricts breakthroughs in product reliability.

[0003] The root cause of this predicament lies in an inherent flaw in the existing technological architecture: it constructs an open-loop, unidirectional manufacturing-service chain. The manufacturing system outputs standardized products according to fixed rules, while the differentiated, collective failure data generated at the service end cannot be systematically and structurally fed back to the design and process origins. This circular barrier between data and knowledge deprives the manufacturing system of its core driving force for continuous self-optimization based on real-world performance.

[0004] Therefore, the lack of an integrated intelligent system that can span the entire life cycle and enable service data to drive manufacturing knowledge in reverse and achieve closed-loop optimization has become a key bottleneck restricting the personalized precision manufacturing and predictive reliable operation and maintenance of special bolts. Summary of the Invention

[0005] The purpose of this invention is to provide an intelligent manufacturing system based on the entire life cycle of special bolts to address the shortcomings of the prior art.

[0006] To achieve the above objectives, the present invention provides the following technical solution: an intelligent manufacturing system based on the entire life cycle of special bolts, comprising:

[0007] The digital twin construction module is used to receive orders, call the pre-set working condition-performance mapping model, generate initial performance targets based on the target working condition information in the order, and create an initial digital twin with a unique digital identifier for each order. The initial digital twin integrates design specifications, material properties and the initial performance targets.

[0008] The personalized process planning module is used to match process units from the modular process knowledge base according to the initial performance target, and generate a personalized process route bound to the initial digital twin through finite element simulation and multi-objective iterative optimization.

[0009] The flexible manufacturing execution module is used to parse and execute the personalized process route, schedule flexible production line resources to complete bolt manufacturing, and collect manufacturing process data in real time and synchronize it to the initial digital twin, driving it to evolve into a manufacturing-state digital twin containing a complete manufacturing history.

[0010] The service intelligent operation and maintenance module is used to access the service monitoring data of the corresponding physical bolt, call the manufacturing history in the manufacturing state digital twin, perform life analysis by integrating the physical model and the data-driven degradation prediction model, and generate active operation and maintenance instructions when the prediction results reach the warning threshold.

[0011] In a preferred embodiment, the digital twin building module includes:

[0012] The target mapping unit is used to call the working condition-performance mapping knowledge base built based on historical data and physical mechanisms, and map the working condition parameter set containing load spectrum and environmental medium into a performance index set containing tensile strength and fatigue life to generate the initial performance target.

[0013] The twin generation unit is used to generate a globally unique digital identifier for each order and create an initial digital twin data structure that includes a geometric model, material genes, performance targets, and standardized data interfaces.

[0014] In a preferred embodiment, the personalized process planning module includes:

[0015] The process knowledge base unit stores a modular material property library, a unit process model library, and a process combination rule engine.

[0016] The simulation optimization unit is used to load the material properties and target working condition load spectrum of the initial digital twin, with the optimization goal of reducing the stress concentration factor of the threaded pair and extending the fatigue life. It iteratively simulates the heat treatment and rolling strengthening process parameters matched from the process knowledge base unit, and outputs the final personalized process route card.

[0017] In a preferred embodiment, the flexible manufacturing execution module includes:

[0018] The production line scheduling unit is used to parse the personalized process route card, generate equipment control commands and material dynamic routing information, and send them to the flexible production line composed of several pluggable dedicated process modules.

[0019] The twin synchronization unit is used to associate the digital twin with the bolt blank through the carrier RFID or visual identification code when the bolt blank flows through each process module, and to collect processing parameters, online detection data and equipment status in real time, and write them back to the corresponding digital twin to form a manufacturing history data layer.

[0020] In a preferred embodiment, the service intelligent operation and maintenance module includes:

[0021] The multi-source data access unit is used to receive real-time service status data from miniature sensors embedded in bolts or connection structures, or from indirect monitoring devices that perform inversion analysis on vibration signals of the main equipment.

[0022] The intelligent decision-making unit is used to call the manufacturing digital twin, compare the expected performance curve with the performance degradation trajectory calculated based on real-time service data, calculate the remaining life or preload degradation state through the degradation prediction hybrid model, and generate maintenance or replacement instructions when it exceeds the safety threshold.

[0023] In a preferred embodiment, it further includes:

[0024] The full lifecycle data traceability center, built on blockchain technology, is used to encrypt and store the entire chain of data related to the design, process, manufacturing and service of each digital twin in chronological order, forming an immutable digital passport for bolts.

[0025] In a preferred embodiment, it further includes:

[0026] The process knowledge adaptive optimization module is used to periodically collect and analyze the service performance data and failure modes of multiple batches of bolts under the same target working condition.

[0027] Based on the analysis results, common weaknesses are identified, and optimization suggestions for specific process unit parameters or combination rules in the modular process knowledge base are generated in reverse, driving the online update and upgrade of the process knowledge base.

[0028] In a preferred embodiment, the digital twin adopts a unified data structure with standardized data interfaces to ensure that the information flow is continuous, correlated, and traceable when it is transmitted between the digital twin construction module, personalized process planning module, flexible manufacturing execution module, and service intelligent operation and maintenance module.

[0029] In a preferred embodiment, each pluggable dedicated process module in the flexible production line is provided with a corresponding module digital twin;

[0030] During the processing, the digital twin of the module and the digital twin of the bolt interact in real time, and the process parameters are fine-tuned and compensated online according to the actual processing status.

[0031] In a preferred embodiment, the degradation prediction hybrid model is a fusion model that integrates physical failure mechanisms and machine learning algorithms, and its input features simultaneously include personalized manufacturing history data and real-time service monitoring data from the manufacturing digital twin.

[0032] The technical effects and advantages provided by the present invention in the above technical solution are as follows:

[0033] This invention significantly improves product performance consistency, manufacturing efficiency, and resource utilization by constructing a digital twin and intelligent closed-loop system covering the entire lifecycle of special bolts. The digital twin construction module accurately maps customer operating conditions into quantifiable performance targets and creates a unique digital starting point for each order, ensuring the accurate transmission of design intent. The personalized process planning module, based on a modular process knowledge base and multi-objective simulation optimization, generates a customized optimal process route for each bolt, fundamentally solving the problem of traditional mass production's inability to meet special operating conditions. The flexible manufacturing execution module, through virtual-physical linkage and online compensation control, ensures high-fidelity execution of personalized process routes and synchronizes manufacturing process data to the digital twin in real time, achieving transparency in the manufacturing process and precise control over product quality. This effectively avoids early failures due to improper processes, reduces rework and scrap, thereby lowering production costs and delivery cycles while meeting high-performance indicators.

[0034] This invention achieves intelligent closed-loop management from manufacturing to service, significantly improving product reliability and lifecycle value. The intelligent operation and maintenance module, by integrating real-time monitoring data with a digital twin containing manufacturing history, constructs a personalized degradation prediction hybrid model. This model can accurately predict the remaining lifespan of bolts and trigger proactive maintenance, transforming reactive maintenance into predictive maintenance and avoiding significant losses from unplanned downtime. The lifecycle data traceability center, based on blockchain technology, establishes an immutable digital passport for bolts, ensuring reliable traceability and auditing of data across the entire design, manufacturing, and operation and maintenance chain. This provides a reliable data foundation for zero-quality issues, responsibility definition, and a circular economy. More importantly, the operation and maintenance feedback and failure knowledge generated during service can drive online optimization of the process knowledge base, forming a continuous improvement closed loop of design-manufacturing-service-optimization. This enables the system to have self-learning and self-evolution capabilities, continuously accumulating and solidifying domain knowledge, ultimately achieving continuous improvement in product reliability and optimization of total lifecycle cost. Attached Figure Description

[0035] To more clearly illustrate the technical solutions in the embodiments of this application or the prior art, the drawings used in the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments recorded in this invention. For those skilled in the art, other drawings can be obtained based on these drawings.

[0036] Figure 1 This is a system block diagram of the present invention. Detailed Implementation

[0037] To make the objectives, technical solutions, and advantages of the embodiments of the present invention clearer, 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, 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.

[0038] In the context of this specification, for ease of understanding and to ensure precise wording, the following terms are clearly defined:

[0039] Bolt blanks: refer to raw materials (such as bars) or semi-finished products (such as forged blanks) that have been preliminarily shaped and processed in the flexible manufacturing execution module of this system.

[0040] Bolt: A general term for the manufactured objects in this system. When not specifically referring to the processing stage, it refers to the final product; when it involves full lifecycle management, its meaning covers the complete product entity from the creation of the digital twin to its final disposal.

[0041] Materials Gene: In this system, this refers to a structured dataset used to uniquely characterize and trace specific material batches, supporting process simulation. It includes at least the material's grade, standard chemical composition range, key process performance parameters (such as continuous cooling transition curves), and a traceable code assigned by the system that associates it with its original production batch. This dataset serves as the fundamental attribute of a digital twin and is a key input driving personalized process simulation and optimization.

[0042] OPCUA (Open Platform Unified Communication Architecture): A cross-platform, service-oriented communication architecture for industrial automation to ensure secure, reliable, and semantically consistent data exchange between different components in a manufacturing system.

[0043] MEMS (Micro-Electro-Mechanical Systems): refers to the technology of integrating micro-mechanical components, sensors, actuators, and electronic circuits onto a single chip. In this invention, it refers to a micro-sensor pre-embedded in a bolt.

[0044] FBG (Fiber Bragg Grating): An optical sensor that achieves this through periodic modulation of the refractive index inside an optical fiber. It is sensitive to strain and temperature and can be used to directly measure the strain of a bolt.

[0045] LoRa (Long Range Radio): A low-power, long-range wireless communication technology suitable for low-power data backhaul of the embedded sensor in this invention.

[0046] Example 1, please refer to Figure 1 As shown in this embodiment, a smart manufacturing system based on the entire life cycle of special bolts includes: a digital twin construction module, a personalized process planning module, a flexible manufacturing execution module, and an intelligent operation and maintenance module for service.

[0047] In one possible implementation, a digital twin building module is used to receive orders, call a pre-set working condition-performance mapping model, generate initial performance targets based on the target working condition information in the order, and create an initial digital twin with a unique digital identifier for each order. The initial digital twin integrates design specifications, material properties, and the initial performance targets.

[0048] The digital twin building module includes:

[0049] The target mapping unit is used to call the working condition-performance mapping knowledge base built based on historical data and physical mechanisms, and map the working condition parameter set containing load spectrum and environmental medium into a performance index set containing tensile strength and fatigue life to generate the initial performance target.

[0050] The twin generation unit is used to generate a globally unique digital identifier for each order and create an initial digital twin data structure that includes a geometric model, material genes, performance targets, and standardized data interfaces.

[0051] In one possible implementation, the digital twin building module is responsible for transforming the operational requirements in customer orders into a digital starting point for the entire lifecycle of bolt products, i.e., an initial digital twin. This module generates an evolvable digital entity with complete performance definitions for each order through a generative engine integrating domain knowledge and intelligent mapping. Specifically, this includes:

[0052] Mapping Model and Intelligent Inference Architecture: This is a work condition-performance mapping knowledge base that integrates data-driven and physical mechanisms. Its role is to transform engineering requirements into quantifiable and actionable performance metrics.

[0053] The knowledge base is composed of, specifically, the following:

[0054] Historical Case Library: Stores complete processed project history data, including input conditions for historical orders, materials and processes used, measured performance data of manufactured bolts, and key service feedback.

[0055] Materials Genome Library: This library constructs digital archives of materials using their grades as an index. It records chemical composition, standard mechanical properties, and correlates them with processing response characteristics, such as continuous cooling transition curves at specific cooling rates and tempering hardness change models, providing material models for performance simulation.

[0056] Physical rule model library: Integrates classical mechanics models, fatigue damage accumulation models, finite element analysis templates, and engineering correction factors. Used to calculate the theoretical boundaries and constraints of performance based on fundamental principles.

[0057] The workflow of the intelligent inference engine specifically includes:

[0058] Once a new order is entered, the engine executes multiple analysis processes in parallel:

[0059] Case similarity retrieval: The feature vector matching algorithm is used to find the most similar cases in the historical case database and extract their successful solutions as candidates.

[0060] Physical boundary calculation: By calling the physical rule model and combining the load spectrum and environmental parameters in the order, the lower limit of performance that meets the safety threshold is simulated and calculated.

[0061] Multi-objective decision-making and optimization: The engine integrates case recommendation results with physical calculation results, and comprehensively considers objectives such as cost, delivery time, and reliability margin. Through a pre-set weighted scoring model, it outputs the optimal initial performance objective set. This objective set is a structured decision output, explicitly specifying: recommended material grade, target tensile strength range, target fatigue life, and key performance indicators.

[0062] The initial digital twin construction process: Based on the generated initial performance targets, the system executes a standardized digital entity creation process.

[0063] Step 1: The twin generation unit uses a composite algorithm based on timestamp, customer code, order serial number, and random factor to generate a globally unique numerical identifier. This identifier will become the primary key for all data associations related to this bolt throughout its entire lifecycle.

[0064] Step Two: Using this unique identifier as the core, the system creates and initializes a structured initial digital twin data object. This object is an extensible collection consisting of the following core data fields:

[0065] Geometric model domain: Associated with a parameterized 3D geometric model whose key dimensions have been preliminarily calculated and determined based on performance targets.

[0066] Materials Genome Domain: Fully records detailed material information mapped from the knowledge base, including grade, chemical composition range, desired microstructure target, and process response parameters.

[0067] Performance target domain: The various performance metrics and their thresholds generated by the inference engine for structured storage serve as criteria for subsequent manufacturing and verification activities.

[0068] Original demand spectrum domain: Establish the association with the set of working condition parameters in the original order to ensure the traceability of demand to target.

[0069] Standardized interface domain: Provides a set of predefined application programming interfaces for writing or reading data to or from the twin in subsequent stages, ensuring its interactivity and dynamic evolution capabilities throughout the entire process.

[0070] This twin is defined as a standardized data container in the system, and its logical structure is shown in Table 1 below:

[0071] Table 1

[0072] Data domain Data type / format Content Description Function and Purpose Unique Identifier String Example: BOLT-20231027-CUST001-0001 The core primary key for full lifecycle data association Geometric model Parametric CAD files Bolt 3D model, dimensional parameters can be driven Fundamental geometric definitions for design, simulation, and manufacturing Materials Gene Structured data Material grade, composition, microstructure target, and relevant curve references Provides digital definitions of material properties to support process simulation. Performance goals key-value pair set Tensile strength: ≥1200MPa, fatigue life: ≥1E7 cycles...} Defining the ultimate performance requirements of a product is the goal of all activities. Original requirements Data Link References to specific working condition parameter sets in the order database Maintain a complete traceability chain from demand to product Data Interface API List Provide methods for obtaining materials and updating process data. Ensure that the twin can interact with and be updated by external system modules.

[0073] The creation of the initial digital twin marks the establishment of the bolt product in the digital space. It is then published to the system's central data bus, triggering subsequent processes: its performance targets and material information are called by the process planning module; its standardized interface becomes the entry point for the manufacturing execution module to write measured data and the service monitoring module to write status data; this twin also serves as the original basis for subsequent full lifecycle data traceability.

[0074] Through this module, the system creates a unique digital definition for each special bolt, making it a calculable, simulable, traceable, and interactive digital entity from the conceptual stage, laying the foundation for precision manufacturing and intelligent services throughout its entire life cycle.

[0075] In one possible implementation, a personalized process planning module, connected to the digital twin construction module, is used to match process units from a modular process knowledge base based on the initial performance target, and generate a personalized process route bound to the initial digital twin through finite element simulation and multi-objective iterative optimization.

[0076] The personalized process planning module includes:

[0077] The process knowledge base unit stores a modular material property library, a unit process model library, and a process combination rule engine.

[0078] The simulation optimization unit is used to load the material properties and target working condition load spectrum of the initial digital twin, with the optimization goal of reducing the stress concentration factor of the threaded pair and extending the fatigue life. It iteratively simulates the heat treatment and rolling strengthening process parameters matched from the process knowledge base unit, and outputs the final personalized process route card.

[0079] Also includes:

[0080] The process knowledge adaptive optimization module is used to periodically collect and analyze the service performance data and failure modes of multiple batches of bolts under the same target working condition; based on the analysis results, it identifies common weak links and generates optimization suggestions for specific process unit parameters or combination rules in the modular process knowledge base, driving the online update and upgrade of the process knowledge base.

[0081] In one possible implementation, the personalized process planning module is responsible for transforming the performance targets in the digital twin into an executable, verifiable, and optimizable digital manufacturing instruction chain. This module, based on a modular knowledge-driven and simulation-optimized closed-loop decision-making system, ensures that the manufacturing plan for each bolt is tailor-made for it. Specific implementation includes the following aspects:

[0082] The modular process knowledge base uses a three-layer modular structure to encapsulate and manage manufacturing knowledge.

[0083] Material Property Library: Indexed by material grade, this library stores digital archives of special alloys. Each record contains a range of mechanical properties and their dynamic process response parameters, such as phase transformation kinetic data, thermodynamic and mechanical constitutive models, and is associated with the material's actual performance distribution in historical production.

[0084] Unit Process Model Library: Abstracts key processes in the manufacturing flow into independent, parameterized digital process unit models. Each unit model is executable; for example, the controlled atmosphere heat treatment unit model can predict workpiece hardness gradients and residual stress fields; the rolling strengthening unit model can simulate and calculate the residual compressive stress field introduced at the thread root and the fatigue strength enhancement factor. Process Combination Rule Engine: Encodes domain expert knowledge and manufacturing constraints in the form of rules to ensure the feasibility and robustness of the process chain. Rules include sequence constraints, compatibility constraints, and optimization-oriented rules.

[0085] Simulation optimization loop and personalized route generation: Using the initial digital twin as input, an automated virtual manufacturing, performance verification and parameter optimization closed loop is initiated.

[0086] Step S1: The process planning engine receives the target material grade and key performance indicators from the initial digital twin. Under the constraints of the process combination rule engine, the engine retrieves and combines a basic process chain template from the unit process model library. The initial parameters of each process unit are preset according to the recommended values ​​in the material property library.

[0087] Step S2: The system automatically calls the integrated finite element analysis solver to perform key analyses: import the bolt geometric model and apply the target working condition load spectrum; set parameters according to the current process chain to drive the material model to calculate the local material properties after virtual manufacturing; accurately calculate the local stress peak value and theoretical stress concentration factor at the first engagement point of the bolt thread pair.

[0088] Step S3: Activate the optimization algorithm to form an automatic optimization closed loop: Set key process parameters as optimization variables; establish an optimization function with the objectives of minimizing the stress concentration factor and maximizing the predicted fatigue life; the iterative process includes generating new parameters, updating the process model, rerunning the finite element analysis and calculating the predicted life, evaluating the objective, and generating the next generation of parameters. After reaching the convergence condition, the system selects the best equilibrium scheme from the optimal solution set, solidifies all determined process units and their optimized parameters into a structured, personalized process route card, and binds it to the digital twin of the bolt.

[0089] The adaptive optimization module for process knowledge enables the self-evolution of process knowledge.

[0090] Data aggregation and case library construction: The module periodically retrieves the complete data chain of bolts that have ended their service life, forming a structured service and failure case library.

[0091] Common weaknesses identified in group analysis: Clustering algorithms are used to analyze failure cases and automatically identify clusters of high-frequency failure modes. These failure clusters are then compared with those in safe service to identify features that show significant differences in manufacturing history data.

[0092] Root cause inversion and knowledge generation: Based on failure physics, root cause hypotheses are proposed, simulation optimization units are invoked, and specific parameters or rules in the process knowledge base are optimized in reverse to eliminate common weak links as constraints, generating optimization suggestions.

[0093] Secure and reliable knowledge updates: Once the generated optimization suggestions are confirmed, they are formalized into new rules and used to update the process knowledge base. This update process is auditable.

[0094] In one possible implementation, a flexible manufacturing execution module is connected to the personalized process planning module. It is used to parse and execute the personalized process route, schedule flexible production line resources to complete bolt manufacturing, and simultaneously collect manufacturing process data in real time and synchronize it to the initial digital twin, driving it to evolve into a manufacturing-state digital twin containing a complete manufacturing history.

[0095] The flexible manufacturing execution module includes:

[0096] The production line scheduling unit is used to parse the personalized process route card, generate equipment control commands and material dynamic routing information, and send them to the flexible production line composed of several pluggable dedicated process modules.

[0097] The twin synchronization unit is used to associate the digital twin with the carrier's radio frequency identification tag or visual identification code when the bolt blank flows through each process module, and to collect processing parameters, online detection data and equipment status in real time, and write them back to the corresponding digital twin to form a manufacturing history data layer.

[0098] Each pluggable dedicated process module in the flexible production line is equipped with a corresponding module digital twin. During the processing, the module digital twin interacts with the digital twin of the bolt in real time, and performs online fine-tuning and compensation control of process parameters based on the actual processing status.

[0099] In one possible implementation, the flexible manufacturing execution module is responsible for converting the personalized process route card generated in the digital space into physical bolts in a high-fidelity, highly flexible, and fully traceable manner. This module constructs a closed-loop execution system with precise instruction issuance, real-time status awareness, bidirectional data synchronization, and dynamic parameter compensation. Specific implementation includes the following aspects:

[0100] The production line scheduling unit interprets the structured process route card into a sequence of instructions that can be executed by the equipment, and realizes the optimal dynamic allocation of production resources.

[0101] Process route card parsing and instruction generation: The scheduling engine parses the process route card and converts it into a standardized equipment instruction set based on the unit process model called for each process step. Flexible production line resource modeling and dynamic scheduling: This unit maintains a real-time updated digital image of the production line, dynamically mapping the status of all available resources. A dynamic path planning algorithm based on real-time status dynamically calculates and allocates the optimal execution resources according to process sequence, resource status, order priority, etc.

[0102] The twin synchronization unit establishes a data pipeline from the physical world to the digital space, ensuring the high-fidelity evolution of the digital twin from the design state to the manufacturing state.

[0103] Identity recognition and data association: Through identification tags on the vehicle, the identity of bolts is automatically identified at key workstations, and their associated digital twins are activated.

[0104] Real-time acquisition and injection of multi-source data: During the processing, the sensor network collects process parameters, online detection data and equipment status data, and adds them to the manufacturing history data layer of the digital twin in real time.

[0105] Modular digital twin interaction and closed-loop process control: Each pluggable dedicated process module is equipped with a modular digital twin, which serves as its dynamic agent in the digital space.

[0106] The composition and function of the module twin: including the precise equipment model, the process model, and the health status model.

[0107] Real-time interactive and compensation control process: When the bolt arrives at the process module, its digital twin sends encrypted parameter instructions to the module twin; the module twin performs feedforward simulation based on the equipment status, predicts deviations, and generates feedforward compensation parameters; during the execution of the physical equipment, sensor data flows into the module twin in real time for comparison, and if an offset is detected, feedback control instructions are dynamically generated to correct the parameters; after the process is completed, a high-fidelity process history data package is sent back to the bolt digital twin.

[0108] The real-time interaction and compensation control between the module digital twin and the bolt digital twin is achieved through the following standardized protocol and logic:

[0109] Interaction Protocol and Message Definition: The system defines a set of information models based on OPCUA (Open Platform Communication Architecture) to ensure interoperability and semantic consistency. Interaction messages are encapsulated in JSON format data frames and published / subscribed via the MQTT protocol on the Manufacturing Execution System (MES) bus. Key interaction messages include:

[0110] Process instruction issuance message: issued by the bolt digital twin.

[0111] Device status synchronization messages: periodically published by the module digital twin.

[0112] Process data write-back message: Published by the module digital twin after the process is completed. The message body contains complete, timestamped actual process curve data and final compensation parameter records.

[0113] Online fine-tuning and compensation control logic: The compensation control adopts a composite strategy of feedforward compensation as the main method and feedback fine-tuning as a supplementary method, which is executed in the controller embedded in the module's digital twin.

[0114] Feedforward compensation calculation: When the module digital twin receives the process instruction, it immediately calls its embedded precise equipment model (such as the furnace heat transfer model and the rolling mill stiffness model), and combines it with the current equipment state (such as wear_compensation_factor) to simulate the execution process and predict the potential deviation Δ between the output and the instruction target. The controller pre-calculates the compensation amount based on Δ and converts it into equipment control parameter offset, which is then issued before physical execution begins.

[0115] Feedback fine-tuning trigger: During physical execution, sensor data flows into the module's digital twin in real time. The controller compares the actual values ​​with the commanded values ​​and the feedforward predicted values. When the deviation exceeds a dynamic threshold (e.g., temperature deviation > 5°C for 10 seconds), the feedback fine-tuning loop is triggered.

[0116] Feedback control algorithm: The feedback loop uses a proportional-integral (PI) control algorithm. Controller output:

[0117] u(t) = ×e(t)+ × dt;

[0118] Where u(t): the control quantity output by the controller at time t (e.g., the adjustment value of heating power, the compensation amount of rolling pressure).

[0119] e(t): The real-time deviation between the set value of the process parameter and the measured value of the sensor at time t (e.g., the difference between the target temperature and the actual temperature).

[0120] and Gain coefficient of proportional-integral (PI) control algorithm. The proportional gain determines the strength of the response to the current deviation e(t); (Integral gain) is used to eliminate the cumulative effect of historical bias, thereby reducing steady-state error.

[0121] Fine-tuning instructions are sent to the PLC in real time via a dedicated device interface (such as ModbusTCP).

[0122] Data packet solidification: All feedforward compensation parameters, feedback fine-tuning events, and the final actual curve are encrypted and encapsulated in the process data write-back message, becoming an immutable part of the manufacturing history.

[0123] In one possible implementation, the service intelligent operation and maintenance module is connected to the digital twin construction module and the flexible manufacturing execution module, respectively. It is used to access the service monitoring data of the corresponding physical bolt, call the manufacturing history in the manufacturing state digital twin, perform life analysis by integrating the physical model and the data-driven degradation prediction model, and generate active operation and maintenance instructions when the prediction result reaches the warning threshold.

[0124] The intelligent operation and maintenance module includes:

[0125] The multi-source data access unit is used to receive real-time service status data from miniature sensors embedded in bolts or connection structures, or from indirect monitoring devices that perform inversion analysis on vibration signals of the main equipment.

[0126] The intelligent decision-making unit is used to call the manufacturing digital twin, compare the expected performance curve with the performance degradation trajectory calculated based on real-time service data, calculate the remaining life or preload degradation state through the degradation prediction hybrid model, and generate maintenance or replacement instructions when it exceeds the safety threshold.

[0127] The construction and operation mechanism of the degradation prediction hybrid model is as follows:

[0128] Model input and data preprocessing:

[0129] The model's input feature vector consists of two parts:

[0130] Personalized manufacturing history data: extracted from the manufacturing digital twin, including but not limited to: actual heat treatment curves (used to calculate theoretical microstructure and hardness distribution), actual pressure and feed rate of thread rolling (used to initialize the residual stress field model), and final measured geometric dimensions (used to correct the finite element mesh).

[0131] Real-time service monitoring data: Time-series data processed by the multi-source data access unit, including but not limited to: axial stress amplitude sequence obtained from strain inversion, preload decay curve, vibration health index, and ambient temperature sequence.

[0132] Before the model runs, the system performs time alignment, normalization and feature engineering on the above data to extract more physically meaningful features such as the number of load spectral block cycles, the preload decay rate and the energy change rate of vibration characteristic frequency bands.

[0133] Model framework and interaction process: The framework adopts a physical mechanism model first, and data-driven online model calibration.

[0134] Physical mechanism model initialization: The system first calls a physical model based on fracture mechanics and fatigue damage mechanics. A specific example illustrates this: the model uses the bolt's personalized manufacturing history data as input. For instance, based on the actual parameters of the rolling process, it calculates the initial residual compressive stress field at the thread root using a preset finite element template. ;

[0135] Fatigue strength coefficient in the material gene Based on the load spectrum, the theoretical initial fatigue life is calculated using the modified Goodman formula. .this As a baseline for prediction.

[0136] Data-driven model online learning and deviation capture: In parallel, a time series prediction model (such as LSTM) begins to learn from real-time service monitoring data. Instead of directly predicting lifespan, this model learns the changing patterns of monitored features (such as the Vibration Health Index, HI) during normal degradation. When real-time data exhibits patterns not anticipated by the physical model (e.g., abnormal preload decay rates due to abnormal wear on assembly surfaces), the data-driven model can capture these deviation trends more quickly.

[0137] Bayesian calibration of the fusion decision layer: The fusion layer is the core of the model. It treats the physical model as a parameterized, adjustable model, where some parameters (such as the equivalent initial defect size a0 and the uncertainty of the material constant C in the Paris formula) are treated as random variables with prior distributions. Abnormal trends in monitoring data captured by the data-driven model, as well as periodic (e.g., monthly) measured preload data, are used as observational evidence. The system uses Bayesian updates or sequential Monte Carlo methods to dynamically update the posterior distributions of these key parameters. Continuing the example: If the measured preload decay is faster than the physical model prediction for several consecutive cycles, the fusion layer updates the estimates of the joint friction coefficient or load spectrum severity factor, thereby calibrating the physical model's output in real time and obtaining a new remaining life prediction value that reflects the actual state of the individual bolt. .

[0138] Output and confidence level: The final output of the model is the calibrated predicted remaining lifetime. The model provides a prediction confidence interval. This interval is wider in the early stages of model operation, but will gradually narrow as service data accumulates and Bayesian updates are performed, reflecting the improvement in prediction accuracy.

[0139] Model training and tuning:

[0140] The initial construction of this hybrid model relies on a historical case library. The parameters of the physical model are calibrated using historical experimental data. The training of the data-driven model uses two types of samples from historical service data: normal service to planned replacement and premature failure, to learn the characteristics of health degradation and anomalous predictors. The prior distribution parameters in the fusion rules are also derived from historical data statistics. During system operation, the model can incrementally learn through newly added, validated closed-loop cases (tracing back from failure to manufacturing), achieving continuous performance optimization.

[0141] The specific implementation and collaborative working mechanism of the degradation prediction hybrid model are as follows:

[0142] Input parameters and construction of the physical mechanism model: This physical model is driven by personalized manufacturing history data, and its input parameters include:

[0143] Intrinsic parameters of the material: obtained from the material gene domain of the digital twin, including the material's elastic modulus, Poisson's ratio, and fatigue strength coefficient. ), fatigue ductility coefficient ( ) and crack propagation parameters (such as C, m in the Paris formula).

[0144] Initial state induced by manufacturing: Calculated based on historical manufacturing data, this is a key personalized input. For example, based on actual rolling process parameters (pressure, speed, number of cycles), the initial residual compressive stress field distribution at the thread root is calculated using a pre-stored finite element response surface model. (x, y, z);

[0145] (x, y, z): Represents the residual compressive stress value at the spatial coordinates (x, y, z) at the root of the bolt thread, in megapascals (MPa). This field is introduced by the rolling hardening process and is key to improving fatigue performance.

[0146] and : These are the fatigue strength coefficient and fatigue ductility coefficient of the material, respectively, and are intrinsic parameters describing the strain-life (ε-N) behavior of the material.

[0147] C and m: Material constants in the Paris formula describing the fatigue crack propagation law of materials. Crack propagation rate. It is usually expressed as = , where ΔK is the stress intensity factor amplitude.

[0148] Based on the actual heat treatment temperature curve, the microstructure (such as martensite content) and initial hardness gradient of the core and surface layer are calculated using a phase transformation kinetic model.

[0149] Theoretical load spectrum: derived from the initial target load condition of the order, serving as input for theoretical stress calculation.

[0150] Based on the above inputs, the physical model calculates the theoretical stress concentration factor through finite element analysis, and applies the local stress-strain method and damage accumulation model (such as Miner's rule) to calculate the theoretical fatigue life baseline of the bolt under ideal working conditions. .

[0151] The data-driven model employs a multi-channel input Long Short-Term Memory (LSTM) network structure to learn temporal degradation patterns in the monitoring data.

[0152] Input feature dimensions: The input feature vector at each time step includes: processed real-time preload. Vibration Health Index Ambient temperature The feature dimensions, as well as the statistics of the aforementioned features within the most recent time window (such as mean, standard deviation, and rate of change), can be expanded based on the sensor configuration.

[0153] Network Structure and Output: The LSTM network layer is used to capture temporal dependencies, and the fully connected layer is used for output. The training objective of this model is to learn the mapping from monitoring features to the deviation of health status. Its training data comes from the complete monitoring time series data of normally serving bolts in the historical case library, as well as data from specific time periods before the known failure point.

[0154] Online learning: The model supports online incremental learning. Once the system has accumulated enough data from new service cycles, the network weights can be fine-tuned in offline mode using the new data to adapt to new operating conditions.

[0155] The specific calibration process of the fusion decision layer: The fusion layer realizes dynamic Bayesian calibration of the physical model parameters, and its specific process is as follows:

[0156] Key parameters in the physical model that vary with individual differences and service process are defined as state variables θ (e.g., equivalent initial defect size a0, load spectrum severity correction factor S).

[0157] Based on historical statistics, θ is assigned a prior probability distribution P(θ), such as a normal distribution or a log-normal distribution.

[0158] In each calibration cycle (e.g., every time monthly measured preload data is received or a health index aberration occurs), the health status deviation output by the data-driven model and the measured preload attenuation ΔF are used as observational evidence Z.

[0159] Applying Bayes' theorem, the posterior probability distribution P(θ|Z) of a state variable is jointly determined by the prior distribution P(θ) and the likelihood function P(Z|θ), and their relationship is expressed as:

[0160] P(θ|Z)∝P(Z|θ)×P(θ);

[0161] Wherein, θ represents the set of state variables to be estimated. For example, in the degradation prediction hybrid model of the present invention, θ may include the equivalent initial defect size (a0), load spectrum severity correction factor (S), etc.

[0162] Z: Represents the set of observed evidence data, such as deviation from health status and measured preload decay (ΔF).

[0163] P(θ): The prior probability distribution of the state variable θ, representing the initial knowledge before obtaining new observation data.

[0164] P(Z|θ): The likelihood probability of observing data Z given state θ.

[0165] P(θ|Z): The posterior probability distribution of the state variable θ after obtaining the observational evidence Z, i.e., the latest knowledge obtained after Bayesian update.

[0166] The system can use the particle filtering (sequential Monte Carlo) method for efficient computation to handle non-Gaussian distributions.

[0167] Using the updated posterior distribution, Monte Carlo simulations were performed on the physical model to obtain the predicted remaining lifetime. The probability distribution is used as the expected value of the final predicted point, and its variance (or 95% confidence interval) is used as a measure of prediction uncertainty.

[0168] In one possible implementation, the intelligent operation and maintenance module is responsible for monitoring the condition of in-service bolts, predicting their individual lifespan, and proactively intervening. This module constructs an analysis engine based on manufacturing history knowledge, real-time condition awareness, and hybrid decision-making. Specific implementation includes the following aspects:

[0169] Multi-source heterogeneous service status awareness network: adopts a strategy that combines embedded direct sensing with system-level indirect inversion.

[0170] Direct Sensing Layer: For core bolts, miniature sensor modules are pre-embedded during the manufacturing stage, integrating microelectromechanical system strain sensors, temperature and corrosion sensors, and self-powered and wireless transmission units. These are used to directly measure parameters such as strain and temperature, and the preload is calculated through a calibration model. Indirect Sensing Layer: For most bolts, vibration sensors are deployed in key parts of the main equipment to collect broadband vibration signals. Through signal processing and a pre-trained deep learning model, damage-sensitive features associated with bolt loosening or cracking are automatically identified from the vibration spectrum, outputting a health index.

[0171] Data fusion and twin contextual association: All monitoring data are aggregated, aligned, and contextualized here.

[0172] Spatiotemporal alignment and context injection: timestamp and location label the monitoring data and synchronously associate it with the current service condition context.

[0173] Manufacturing history data retrieval: By using the unique identifier of the bolt, retrieve its manufacturing state digital twin and extract key manufacturing history data.

[0174] Structured data lake construction: Real-time monitoring data, operating condition context, and manufacturing history data are written together into the service history data layer of the bolt digital twin.

[0175] Construction and operation of hybrid degradation prediction model: This model is a deep coupling of physical mechanism and data-driven algorithm.

[0176] The model employs a layered architecture: the physical mechanism layer is based on fracture mechanics and fatigue damage accumulation theory. Using personalized manufacturing history data as the core input, it calculates the theoretical performance degradation baseline under ideal design conditions.

[0177] The data-driven layer employs a time-series prediction algorithm, taking real-time service monitoring data sequences as input, to learn their dynamic evolution patterns and their correlation with failure risks.

[0178] The fusion decision layer uses algorithms such as Bayesian updates or filtering to take real-time monitoring data as observations and continuously calibrate the key initial parameters or state variables of the physical mechanism model, thus personalizing the general physical model.

[0179] Example of a prediction process:

[0180] Actual process parameters are read from the digital twin, and the theoretical initial fatigue life is calculated using a physical model.

[0181] Analyze the measured load spectrum and health index trends to identify the actual load level and possible damage acceleration modes.

[0182] Using recent measured data, the estimates of load severity and initial material damage state in the physical model are updated, and the corrected personalized remaining life prediction and confidence interval are output.

[0183] Intelligent decision-making and proactive operation and maintenance instruction generation:

[0184] The prediction results are input into the rule engine and optimizer to generate operation and maintenance strategies.

[0185] The system is equipped with a risk-based multi-level early warning mechanism, and the threshold can be dynamically adjusted according to the criticality of the bolt.

[0186] The generated maintenance instructions are structured data packets containing object identifiers, suggested measures, and decision-making basis. The instructions are automatically dispatched through the maintenance management platform and simultaneously written back to the digital twin of the bolt.

[0187] In one possible implementation, the intelligent operation and maintenance module and the adaptive optimization module for process knowledge within the personalized process planning module form a cross-lifecycle intelligent optimization closed loop. Specifically, the operation and maintenance instructions (such as replacement instructions) generated by the intelligent decision-making unit and their associated failure mode analysis results are input as key feedback to the adaptive optimization module for process knowledge. This module not only analyzes common weaknesses based on historical service data but also initiates targeted root cause inversion for specific failure cases identified and confirmed in real time by the intelligent decision-making unit. For example, when multiple bolts from the same batch reach the warning threshold prematurely under similar operating conditions, the system can automatically associate their manufacturing history and use the simulation optimization unit to solve in reverse, generating correction suggestions for specific process parameters such as rolling pressure and heat treatment temperature. These confirmed optimization suggestions then drive the online update of the modular process knowledge base, thereby enabling continuous improvement of bolt process routes planned for the same or similar operating conditions, achieving a closed-loop improvement from service problem discovery to process knowledge optimization to manufacturing quality prevention.

[0188] In one possible implementation, the full lifecycle data traceability center is responsible for building and maintaining a consistent, tamper-proof, authoritative digital archive for each special bolt. Its core lies in utilizing the distributed consensus, cryptographic tamper-proof nature, and automatic execution characteristics of smart contracts in blockchain technology to anchor dispersed key data chronologically onto a trusted chain, achieving end-to-end data traceability and value transfer.

[0189] Permissioned Consortium Blockchain Network Architecture: Adopts a permissioned consortium blockchain as the underlying framework.

[0190] Node roles and consensus mechanisms include sorting nodes responsible for packaging blocks, endorsement nodes responsible for verifying data compliance, and lightweight clients for lightweight access to business systems.

[0191] Hybrid data storage model: adopts a hybrid strategy of hash value on-chain and secure original text storage.

[0192] On-chain storage: Only the cryptographic hash value and core metadata of the data are written to the immutable blockchain ledger. The hash value serves as the digital fingerprint of the data.

[0193] Off-chain storage: The original, full data is stored in a high-performance distributed file system or time-series database. On-chain hash values ​​serve as trusted pointers to the off-chain data and as integrity verification codes.

[0194] Data structure and genealogy definition of the bolt digital passport:

[0195] Each digital passport is a dynamically growing set of structured data with a bolt unique identifier as the primary key, and its core data fields constitute a complete lifecycle spectrum.

[0196] For example, when the twin synchronization unit writes a heat treatment actual temperature curve data, the system automatically generates the hash value H1 of the data, and stores it in the blockchain network as a transaction with information such as timestamp and device ID through a smart contract.

[0197] The genealogy dimensions include: identity and digest, requirements and design genealogy, manufacturing archive genealogy, and service and maintenance genealogy. The key data digest hash of each genealogy is anchored on-chain and associated with the complete original data off-chain.

[0198] Smart contract mechanism for data anchoring and circulation: Data on-chain is driven by a series of automated and strongly constrained smart contracts.

[0199] Automatically invoked after the initial twin is created, a standard-compliant, non-homogeneous digital passport is cast for the bolt.

[0200] Triggered upon completion of critical processes, and checks whether previous process records have been uploaded to the blockchain, forcibly ensuring that the physical order of the manufacturing process is consistent with the order of data records.

[0201] Before execution, verify the submitter's digital signature permissions to ensure the authenticity of the operation and maintenance data is linked to the responsible party.

[0202] In one possible implementation, the digital twin adopts a unified data structure and standardized interface as a standardized container and interaction protocol for cross-module and cross-stage data transmission, ensuring data semantic consistency, temporal continuity, and full traceability.

[0203] A hierarchical definition of a unified data structure: adopting a hierarchical model that separates core attributes from dynamic lineages.

[0204] The information is solidified at the time of twin creation as its immutable digital definition, including a globally unique identifier, a basic geometric model, material genome data, and initial performance targets.

[0205] It grows dynamically as its life cycle progresses, using a time-series structure to record manufacturing and service lineages, and includes data lineage.

[0206] Define the names, types, constraints, and access permissions of data fields that can be accessed by external modules in a machine-readable format.

[0207] Design and implementation of standardized data interfaces: The system defines a set of standard application programming interfaces.

[0208] A status query interface for obtaining core data or process history of twins;

[0209] The interface for writing and updating data for appending process records or updating status;

[0210] An event subscription and notification interface for receiving state change events in real time;

[0211] And a verification interface for verifying data integrity.

[0212] Interface security and access control: All interface calls must be authenticated and authorized, and different data access roles and permissions are assigned to different business modules.

[0213] Mechanism for continuity and traceability of cross-module information flow:

[0214] A unified structure, interface, unique identifier, and global timing service together construct a seamless information flow.

[0215] From the moment the twin is created, its unique identifier becomes the data association key that runs through all business systems.

[0216] All write operations to the twin dynamic spectrum are timestamped globally by a unified high-precision timing service, ensuring that events are recorded strictly in the order they occur in the physical world.

[0217] By locating a digital twin using a unique identifier, its complete lineage can be traced back to the point of anomalous events, and further investigation can be conducted by combining data lineage relationships. The integrity of the data used in the tracing process can be verified by comparing it with the hash value stored on the blockchain via an interface.

[0218] The above description is merely a specific embodiment of this application, but the scope of protection of this application is not limited thereto. Any variations or substitutions that can be easily conceived by those skilled in the art within the technical scope disclosed in this application should be included within the scope of protection of this application. Therefore, the scope of protection of this application should be determined by the scope of the claims.

Claims

1. A special bolt full life cycle intelligent manufacturing system based on, characterized in that, include: The digital twin construction module is used to receive orders, call the pre-set working condition-performance mapping model, generate initial performance targets based on the target working condition information in the order, and create an initial digital twin with a unique digital identifier for each order. The initial digital twin integrates design specifications, material properties and the initial performance targets. The personalized process planning module is used to match process units from the modular process knowledge base according to the initial performance target, and generate a personalized process route bound to the initial digital twin through finite element simulation and multi-objective iterative optimization. The flexible manufacturing execution module is used to parse and execute the personalized process route, schedule flexible production line resources to complete bolt manufacturing, and collect manufacturing process data in real time and synchronize it to the initial digital twin, driving it to evolve into a manufacturing-state digital twin containing a complete manufacturing history. The service intelligent operation and maintenance module is used to access the service monitoring data of the corresponding physical bolt, call the manufacturing history in the manufacturing state digital twin, perform life analysis by integrating the physical model and the data-driven degradation prediction model, and generate active operation and maintenance instructions when the prediction results reach the warning threshold.

2. The full life cycle intelligent manufacturing system based on special bolts according to claim 1, characterized in that: The digital twin building module includes: The target mapping unit is used to call the working condition-performance mapping knowledge base built based on historical data and physical mechanisms, and map the working condition parameter set containing load spectrum and environmental medium into a performance index set containing tensile strength and fatigue life to generate the initial performance target. The twin generation unit is used to generate a globally unique digital identifier for each order and create an initial digital twin data structure that includes a geometric model, material genes, performance targets, and standardized data interfaces.

3. The full life cycle intelligent manufacturing system based on special bolts according to claim 2, characterized in that: The personalized process planning module includes: The process knowledge base unit stores a modular material property library, a unit process model library, and a process combination rule engine. The simulation optimization unit is used to load the material properties and target working condition load spectrum of the initial digital twin, with the optimization goal of reducing the stress concentration factor of the threaded pair and extending the fatigue life. It iteratively simulates the heat treatment and rolling strengthening process parameters matched from the process knowledge base unit, and outputs the final personalized process route card.

4. The full life cycle intelligent manufacturing system based on special bolts according to claim 3, characterized in that: The flexible manufacturing execution module includes: The production line scheduling unit is used to parse the personalized process route card, generate equipment control commands and material dynamic routing information, and send them to the flexible production line composed of several pluggable dedicated process modules. The twin synchronization unit is used to associate the digital twin with the bolt blank through the carrier RFID or visual identification code when the bolt blank flows through each process module, and to collect processing parameters, online detection data and equipment status in real time, and write them back to the corresponding digital twin to form a manufacturing history data layer.

5. The full life cycle intelligent manufacturing system based on special bolts according to claim 4, characterized in that: The intelligent operation and maintenance module includes: The multi-source data access unit is used to receive real-time service status data from miniature sensors embedded in bolts or connection structures, or from indirect monitoring devices that perform inversion analysis on vibration signals of the main equipment. The intelligent decision-making unit is used to call the manufacturing digital twin, compare the expected performance curve with the performance degradation trajectory calculated based on real-time service data, calculate the remaining life or preload degradation state through the degradation prediction hybrid model, and generate maintenance or replacement instructions when it exceeds the safety threshold.

6. The full life cycle intelligent manufacturing system based on special bolts according to claim 1, characterized in that: Also includes: The full lifecycle data traceability center, built on blockchain technology, is used to encrypt and store the entire chain of data related to the design, process, manufacturing and service of each digital twin in chronological order, forming an immutable digital passport for bolts.

7. The full life cycle intelligent manufacturing system based on special bolts according to claim 3, characterized in that: Also includes: The process knowledge adaptive optimization module is used to periodically collect and analyze the service performance data and failure modes of multiple batches of bolts under the same target working condition. Based on the analysis results, common weaknesses are identified, and optimization suggestions for specific process unit parameters or combination rules in the modular process knowledge base are generated in reverse, driving the online update and upgrade of the process knowledge base. 8.The special bolt full life cycle intelligent manufacturing system according to claim 1, characterized in that: The digital twin adopts a unified data structure with standardized data interfaces to ensure that the information flow is continuous, associative, and traceable when it is transmitted between the digital twin construction module, personalized process planning module, flexible manufacturing execution module, and service intelligent operation and maintenance module.

9. The full life cycle intelligent manufacturing system based on special bolts according to claim 4, characterized in that: Each pluggable dedicated process module in the flexible production line is equipped with a corresponding module digital twin; During the processing, the digital twin of the module and the digital twin of the bolt interact in real time, and the process parameters are fine-tuned and compensated online according to the actual processing status.

10. The full life cycle intelligent manufacturing system based on special bolts according to claim 5, characterized in that: The degradation prediction hybrid model is a fusion model that integrates physical failure mechanisms and machine learning algorithms. Its input features include personalized manufacturing history data and real-time service monitoring data from the manufacturing digital twin.

Images