Wheel flash welding optimization system and method based on digital twinning

Abstract

The invention discloses a wheel flash welding optimization system and method based on digital twinning, and belongs to the crossing field of intelligent manufacturing, welding processes and artificial intelligence. The system comprises a material attribute database module, a multi-source data acquisition and preprocessing module, a digital twin module integrating heat-force-microstructure and other multi-physics field models, an AI prediction and multi-objective optimization module, an advanced visualization and interaction module and a closed-loop control module. The method comprises the following steps: initializing and matching material parameters; collecting data in real time to update the twinborn model; running simulation and AI to predict the welding temperature, the structure and the performance; if the prediction does not reach the standard, starting a multi-objective optimization algorithm to generate an optimal process parameter set; and executing optimization parameters through closed-loop control and performing real-time visual feedback. According to the method, full-chain real-time mapping, transparent monitoring and intelligent optimization of technology-structure-performance in the welding process are achieved, the performance stability of joints is remarkably improved, and the efficiency and the product quality are optimized.

Description

Technical Field

[0001] This invention belongs to the interdisciplinary field of intelligent manufacturing, welding technology and artificial intelligence. Specifically, it relates to a visualization and multi-objective optimization system and method for flash welding of wheels based on digital twins. It is applicable to the flash welding process control of high-precision wheels in the fields of automobiles, high-speed rail, aviation and other fields, and realizes real-time simulation, intelligent optimization, dynamic monitoring and immersive visualization of welding process. Background Technology

[0002] Flash welding, as a highly efficient and reliable solid-state joining technology, is widely used in the manufacture of critical load-bearing components such as wheels and rails. Its process involves complex coupling of multiple physical fields, including heat, force, electricity, and magnetism, as well as the dynamic evolution of the material's microstructure. The final joint quality is significantly affected by process parameters. To improve welding quality and stability, existing technologies mainly explore process optimization, process monitoring, and the application of auxiliary physical fields, but the following limitations still exist:

[0003] Traditional process control relies on experience and lacks precise data-driven approaches. The setting and adjustment of traditional wheel flash welding process parameters depend heavily on manual experience, making it difficult to respond in real-time to microscopic structural changes such as material phase transformations and dynamic recrystallization during welding. This results in significant fluctuations in joint performance (such as strength and fatigue life) (typically exceeding ±15%), making it difficult to consistently improve product yield. This is a typical "black box" trial-and-error model, characterized by long debugging cycles and high costs.

[0004] Process monitoring and quality assessment systems are limited in function and lack real-time simulation and prediction capabilities. Existing welding process monitoring systems (such as the rail flash welding detection and evaluation device disclosed in CN119237892A) mainly collect macroscopic parameters such as current, voltage, and upsetting force through sensors, and use machine learning and neural network algorithms for post-process quality assessment or limited parameter correction. Although such systems achieve data acquisition and analysis based on historical data, they are essentially still "observations" and "post-process evaluations" of the actual welding process, lacking a "digital twin" model that can map the actual welding process 1:1 in real time and perform multi-physics field coupled simulation and advanced prediction. Therefore, it is impossible to predict the evolution of temperature and stress fields in real time during welding, let alone correlate and predict microstructure (such as grain size and phase composition) with final mechanical properties, making it difficult to achieve true closed-loop control and online optimization of the entire "process-structure-performance" chain.

[0005] Emerging physics-assisted processes focus on local optimization and lack global multi-objective coordination. To improve welding quality, process methods that introduce external physical fields (such as magnetic fields) have emerged. For example, CN120791096A discloses a traveling wave magnetic field-assisted flash welding method and equipment, which improves heat transfer, metal flow, and refines grain size by applying a magnetic field, and uses a genetic algorithm to optimize the magnetic field and welding parameters. However, this method has a relatively singular optimization objective (mainly minimizing grain size) and fails to systematically consider multiple objectives such as joint strength, welding energy consumption, production efficiency, and deformation control. In addition, this method does not construct a virtual twin that integrates real-time data; its optimization is based on offline models or experimental data, which cannot be dynamically adjusted and adaptively controlled during the welding process, and it lacks immersive visualization support for the entire welding process, resulting in insufficient process transparency.

[0006] The visualization and interactivity capabilities are weak, resulting in low process transparency. Most existing technologies display welding data and results using two-dimensional curves, charts, or simple models, lacking advanced visualization methods that can intuitively and immersively present the three-dimensional formation of welds, the dynamic expansion of the heat-affected zone, the evolution of microstructure, and the spatial distribution of multi-physics fields (temperature, stress). This makes it difficult for process engineers to intuitively understand the intrinsic connections of complex processes, hindering rapid commissioning and accurate decision-making.

[0007] In summary, existing wheel flash welding technologies either rely on experience, focus on post-process evaluation, or are limited to a single physical field and optimization objective, failing to construct a comprehensive intelligent system integrating "high-fidelity real-time simulation (digital twin), multi-physics coupled prediction, AI-assisted decision-making, multi-objective collaborative optimization, immersive visual interaction, and online closed-loop control." Therefore, an innovative solution is urgently needed to achieve full transparency, intelligence, and global optimization of the wheel flash welding process. Summary of the Invention

[0008] To address the aforementioned issues, this invention provides a digital twin-based system and method for optimizing wheel flash welding. This system integrates a material database, multiphysics modeling, real-time data acquisition, AI prediction, multi-objective optimization, and advanced visualization modules to construct a digital twin of the entire wheel flash welding process, enabling dynamic optimization of process parameters and transparent control of the welding process.

[0009] The present invention adopts the following technical solution:

[0010] A digital twin-based wheel flash welding optimization system includes:

[0011] The material property database module is used to store the basic parameters, microstructure parameters and welding process reference parameters of steel used for wheels, and provides a data interface;

[0012] The multi-source data acquisition and preprocessing module is used to acquire temperature, pressure, current and displacement data of the welding process in real time through the sensor unit, and to preprocess the data.

[0013] The multiphysics modeling and digital twin module is used to integrate thermo-mechanical coupling models, microstructure evolution models, and electromagnetic-fluid coupling models to construct and update a virtual twin of the wheel flash welding process in real time.

[0014] The AI ​​prediction and multi-objective optimization module is used to predict welding temperature, microstructure, defects and mechanical properties based on artificial intelligence models, and to generate an optimized set of process parameters using multi-objective optimization algorithms.

[0015] The advanced visualization and interaction module is used to provide multi-view visualization of the welding process, physical field distribution, microstructure evolution, and optimization results. Figure 3 3D visualization and interaction;

[0016] The closed-loop control module is used to dynamically adjust the process parameters of the welding equipment based on the predicted deviation between the virtual twin and the real-time welding process.

[0017] This solution is the first to construct a complete system architecture that integrates "real-time data perception, multi-physics digital twin simulation, AI prediction and multi-objective collaborative optimization, immersive visualization and online closed-loop control", realizing a fundamental transformation of wheel flash welding from "experience trial and error" to "data-driven, model prediction and intelligent decision-making" closed-loop optimization of the entire process.

[0018] Furthermore, in the aforementioned system, the basic parameters of wheel steel stored in the material property database module include: steel composition, thermophysical parameters, mechanical property parameters, and phase transformation temperature parameters. By constructing a standardized material property database, accurate and unified initial conditions and property basis are provided for downstream multiphysics simulation and AI prediction models. This serves as the data foundation for achieving precise mapping and optimization of process, microstructure, and performance, overcoming the problems of scattered and inconvenient material parameters in traditional processes.

[0019] Furthermore, in the aforementioned system, the sensor unit of the multi-source data acquisition and preprocessing module includes: an infrared thermometer, a pressure sensor, a Hall current sensor, and a displacement sensor. By integrating multiple types of high-response sensors, synchronous real-time capture of key physical quantities such as heat, force, electricity, and displacement during the welding process is achieved. This provides a high-fidelity, highly synchronous real-time data source for the digital twin, which is the foundation for ensuring the real-time mapping and synchronous evolution of the virtual model and the physical process.

[0020] Furthermore, in the aforementioned system, the microstructure evolution model in the multiphysics modeling and digital twin module is constructed based on the Johnson-Mehl-Avrami equation, the dynamic recrystallization model, and the Ostwald ripening model. By deeply coupling macroscopic thermodynamic field simulation with microstructure evolution models based on classical theory (JMA equation, dynamic recrystallization, etc.), cross-scale quantitative prediction of microstructure evolution from macroscopic process parameters is achieved, establishing the intrinsic correlation between "process-structure" and providing mechanistic model support for controlling the final performance through process optimization.

[0021] Furthermore, in the aforementioned system, the AI ​​prediction and multi-objective optimization module includes an artificial intelligence model comprising a neural network for temperature prediction, a convolutional neural network for microstructure identification, a target detection model for defect detection, and a random forest model for comprehensive performance prediction; the multi-objective optimization algorithm is the NSGA-II genetic algorithm. Combining multiple specialized AI models (for fast, high-precision prediction) with the NSGA-II multi-objective optimization algorithm forms a core intelligent decision-making mechanism of "fast prediction - multi-objective trade-offs." This combination leverages the efficiency of AI models for real-time state assessment and seeks the global optimal solution under multiple constraints such as intensity, energy consumption, and efficiency through optimization algorithms, achieving intelligent, multi-objective automatic optimization of process parameters.

[0022] Furthermore, in the aforementioned system, the optimization objectives of the multi-objective optimization algorithm include: maximizing joint strength, minimizing energy consumption, minimizing deformation, maximizing production efficiency, and minimizing defect rate.

[0023] Furthermore, in the aforementioned system, the views provided by the advanced visualization and interaction module include: a main view displaying the three-dimensional model of the wheel welding, a cross-sectional view displaying the weld section, a physical field view displaying the temperature field or stress field distribution, and a trend view displaying the quality or performance trend.

[0024] This invention also discloses an optimization method for flash welding of wheels based on digital twins, employing the system described in any of the above claims, and comprising the following steps:

[0025] S1. Based on the target wheel steel material, match relevant parameters from the material property database and initialize the system;

[0026] S2. During the welding process, welding data is collected in real time, and the boundary conditions of the digital twin model are updated after preprocessing;

[0027] S3. Run multiphysics coupling simulation and call AI prediction model to obtain prediction results of welding temperature field, stress field, microstructure, defects and mechanical properties;

[0028] S4. If the prediction results do not meet the preset performance requirements, the multi-objective optimization algorithm is started to generate the Pareto optimal solution set based on the current parameters and constraints, and the optimal combination of process parameters is selected from it.

[0029] S5. Output the optimized process parameters to the welding equipment for execution, update the visualization interface in real time, and store the welding process data.

[0030] The above method defines a closed-loop process control approach based on digital twins and multi-objective optimization. This method achieves online dynamic optimization and adaptive adjustment of the welding process through an iterative cycle of "real-time acquisition - simulation prediction - intelligent evaluation - multi-objective optimization - closed-loop execution," transforming the traditional static, offline process planning mode into a dynamic, online, and self-optimizing intelligent control mode.

[0031] Furthermore, in the above method, in step S4, the multi-objective optimization algorithm is the NSGA-II algorithm, and its optimization objective function includes at least joint strength, energy consumption, and deformation.

[0032] Furthermore, in step S5 of the above method, the closed-loop control module dynamically adjusts the upsetting force, sintering current, or pressure of the cooling medium based on the deviation between the predicted residual stress, grain size, or final cooling temperature and the target threshold.

[0033] Furthermore, the above method can be applied to high-precision, high-efficiency welding of wheels in fields such as automobiles and high-speed trains.

[0034] The beneficial effects of this invention are:

[0035] 1. Improved process controllability: Through 1:1 real-time mapping of digital twins and multi-physics field coupled simulation, the welding process is made "transparent", the adjustment of process parameters is changed from "experience-driven" to "data-driven", and the joint performance fluctuation is reduced from ±15% to ±5%.

[0036] 2. Improved efficiency: The NSGA-II multi-objective optimization algorithm combined with AI prediction reduces optimization time to ≤10s, reducing trial and error costs by more than 30% compared to traditional trial and error methods, and improving production efficiency by 20%.

[0037] 3. Accurate quality prediction: The AI ​​model (temperature prediction accuracy ≥94.2%, defect detection accuracy ≥85%) enables performance prediction and defect warning during the welding process, improving the pass rate to over 99%.

[0038] 4. Enhanced Visual Experience: Multi-view Figure 3 The 3D visualization and interactive functions intuitively present weld formation, microstructure and physical field distribution, shortening the process debugging cycle by 40%.

[0039] 5. Support for green manufacturing: Multi-objective optimization balances energy consumption (reduced by 20%) and performance, reduces reliance on welding preheating / postheating processes, and helps achieve the "dual carbon" target. Attached Figure Description

[0040] Figure 1 The system architecture diagram illustrates the interactive relationships between modules including the materials database, data acquisition, digital twin, AI optimization, visualization, and closed-loop control.

[0041] Figure 2 It is a multi-view layout diagram, including the main view of wheel welding, cross-sectional view, temperature field view, stress field view and phase distribution view;

[0042] Figure 3 The Pareto front curves of the NSGA-II algorithm are used to demonstrate the multi-objective optimization relationship between strength, energy consumption, and deformation.

[0043] Figure 4 The microstructure comparison diagram shows the changes in grain size and phase distribution before and after optimization. Detailed Implementation

[0044] (I) Visualization and Multi-Objective Optimization System for Flash Welding of Wheels Based on Digital Twin

[0045] This system integrates a materials database, multiphysics modeling, real-time data acquisition, AI prediction, multi-objective optimization, and advanced visualization modules to construct a digital twin of the entire flash welding process for automobile wheels, enabling dynamic optimization of process parameters and transparent control of the welding process. The system specifically includes the following core modules:

[0046] 1. Material Property Database Module

[0047] Storage data types:

[0048] 1) Basic parameters of steel for wheels: including steel composition (e.g., C 0.45%, Mn 0.65%, Si 0.25%, P ≤ 0.025%, S ≤ 0.025%), thermal properties (thermal conductivity 45 W / (m·K), specific heat capacity 460 J / (kg·K), density 7850 kg / m³). 3 The coefficient of thermal expansion is 12×10. -6 / K), mechanical property parameters (elastic modulus 210GPa, Poisson's ratio 0.3, melting point 1500℃, phase transformation temperature Ac1=723℃, Ac3=850℃, Ms=350℃).

[0049] 2) Microstructure parameters: including the hardness range (e.g., ferrite 120-180HV, martensite 500-700HV), strength range (e.g., pearlite 600-900MPa, martensite 1000-1800MPa) and dynamic recrystallization parameters (activation energy, Avrami parameter, recrystallization initiation stress) of each phase (ferrite, pearlite, bainite, martensite, austenite).

[0050] 3) Welding process reference parameters: including the reference parameter ranges for the preheating (temperature 600-1000℃, heating rate 50-150℃ / s), flash (speed 1.5-3.5mm / s, current 2500-3500A), and upsetting (force 60000-100000N, speed 5-12mm / s) stages.

[0051] Module functions: Provides data retrieval, visualization (such as CCT / TTT curve chart display), parameter comparison and interaction interface with downstream simulation modules, and supports dynamic adaptation of material properties and process requirements.

[0052] 2. Multi-source data acquisition and preprocessing module

[0053] Sensor unit:

[0054] 1) Infrared thermometer: Real-time acquisition of welding zone temperature (300-1200℃), sampling rate 100Hz;

[0055] 2) Pressure sensor: Real-time acquisition of upsetting force (200-600kN), accuracy ±1%;

[0056] 3) Hall current sensor: Real-time acquisition of welding current (0-6000A), response time ≤1ms;

[0057] 4) Displacement sensor: Real-time acquisition of upsetting speed (1.0-2.5mm / s), resolution 0.01mm;

[0058] 5) Acoustic sensor: Collects acoustic emission signals from the welding area to monitor spatter and defect formation.

[0059] Preprocessing unit: Filters (Kalman filter), denoises (wavelet denoising) and calibrates the raw data, and extracts characteristic parameters such as temperature change rate, force-displacement curve slope, and current fluctuation amplitude.

[0060] Monitoring and storage unit: The host computer software displays temperature, current, and upsetting force curves in real time, and triggers audible and visual alarms when parameters exceed limits; the processed data is stored in CSV / SQL format, and supports historical data query and trend analysis.

[0061] 3. Multiphysics Modeling and Digital Twin Module

[0062] Core model integration:

[0063] 1) Thermo-mechanical coupling model: Based on the finite element method, using a 200×200×50 mesh, the heat conduction equation is solved.

[0064] Where Q represents the Joule heat and heat of plastic deformation, ρ (density), c (specific heat capacity), T (temperature), t (time), and k (thermal conductivity) are all constants. (Gradient / divergence operator), ∂ is a partial differential operator) and the equations of mechanical equilibrium. The temperature field, stress field (von Mises stress 0-500MPa), and displacement field are predicted in real time (where σ (stress tensor), f (volume force), ρ (density), u (displacement vector), and t (time)).

[0065] 2) Microstructure evolution model: Integrated Johnson-Mehl-Avrami phase transition model (X(t)=1-exp(-kt)) n (n=1.5, k=0.1), dynamic recrystallization model (critical strain ε) c =0.65εp, k=0.012, n=1.2) and the Ostwald ripening model were used to predict the grain size (45±10μm (ferrite), 25±5μm (pearlite)) and phase distribution (ferrite 0.65±0.1, pearlite 0.30±0.05).

[0066] 3) Electromagnetic-fluid coupling model: Calculate the current density and metal vapor flow field in the welding zone (velocity 0-10mm / s), and simulate the plasma arc morphology and metal loss (2-6mm).

[0067] Digital twin construction: By integrating multiphysics simulation data and real-time sensor data, a virtual twin of wheel flash welding is constructed to achieve a 1:1 real-time mapping of the welding process (preheating-flash-upsetting) with a time synchronization error of ≤0.1s.

[0068] 4. AI Prediction and Multi-Objective Optimization Module

[0069] AI prediction submodule:

[0070] 1) Temperature prediction neural network: Input 8-dimensional process parameters (preheating temperature, flash current, etc.), predict the temperature of the welding zone through 3 hidden layers (20-15-10 nodes), with an accuracy of ≥94.2%;

[0071] 2) Microstructure prediction model: Based on ResNet-18CNN, given a 64×64×3 tissue image, predict the volume fraction of 5 phases with a confidence level ≥91.8%;

[0072] 3) Defect detection model: Using the YOLOv5 algorithm, a 416×416×3 weld image is input to detect 6 types of defects, including porosity and inclusions, with an accuracy of ≥85%;

[0073] 4) Comprehensive performance prediction: Based on the random forest model, predict indicators such as tensile strength and fatigue life with an error of ≤5%.

[0074] Multi-objective optimization submodule:

[0075] 1) Optimization algorithm: NSGA-II genetic algorithm is adopted, with a population size of 100, 200 iterations, crossover probability of 0.9, and mutation probability of 0.1;

[0076] 2) Objective function: The objectives are to "maximize joint strength (≥550MPa), minimize energy consumption (≤1.5kWh), minimize deformation (≤0.5mm), maximize production efficiency (≥80%), and minimize defect rate (≤5%)".

[0077] 3) Input variables: preheating temperature (400-1000℃), flash current (2500-3500A), upsetting force (60000-100000N), cooling rate (10-80℃ / s);

[0078] 4) Output: Generates the Pareto optimal solution set, automatically selects the optimal parameter set with compromise, and optimizes in ≤10s.

[0079] 5. Advanced Visualization and Interaction Module

[0080] Multi-view display:

[0081] 1) Main view: 3D wheel welding model (outer diameter 600mm, inner diameter 580mm, thickness 20mm), real-time display of weld position and heat-affected zone;

[0082] 2) Cross-sectional view: showing weld penetration (4-8mm), grain size distribution, and defect location;

[0083] 3) Physical field view: temperature field (thermal map display, peak value at 1500℃), stress field (equivalent stress cloud map), phase distribution (color-coded phase regions);

[0084] 4) Trend View: Real-time plotting of quality scores (0-100 points) and performance prediction curves.

[0085] Interactive features: Supports model rotation (rotate3d), scaling (zoom), and panning (pan). The visualization effects are automatically updated when switching between welding stages (preheating-flash-upsetting) (e.g., displaying plasma arc particle effects during the flash stage and displaying deformation during the upsetting stage).

[0086] 6. Closed-loop control module

[0087] Based on the deviation (≤5%) between the digital twin and the actual welding process, the process parameters are dynamically adjusted:

[0088] 1) If the residual stress σ max ≥0.8σ y (σ) y If the yield strength of the base material is used, the upsetting force is increased to 550-580kN, and the sintering current is reduced to 2800-3000A;

[0089] 2) If the grain size d DRX For diameters ≥15μm, the upsetting speed increases to 2.4-2.5mm / s;

[0090] 3) During the cooling stage, the air jet pressure is adaptively adjusted (0.40-0.45MPa) according to the final cooling temperature (400-450℃).

[0091] (II) Visualization and Multi-Objective Optimization Method for Flash Welding of Wheels Based on Digital Twin

[0092] The specific steps for implementing flash welding control of wheels using the above system are as follows:

[0093] S1. System Initialization and Parameter Matching

[0094] 1) Based on a material property database, an incremental algorithm is used to match the phase transformation dynamics parameters of the target wheel steel (e.g., 490CL):

[0095] Search the database for materials with similar compositions and filter the initial candidate set;

[0096] The parameter combination is iteratively optimized through local search (parameter fine-tuning) and global search (random sampling);

[0097] By combining experimental data error analysis (deviation ≤3%) with feedback adjustment, high-precision parameters such as CCT curve and phase transition activation energy are output.

[0098] 2) Configure system parameters: Set simulation accuracy (high), rendering quality (ultra), real-time mode (on), and GPU acceleration (on).

[0099] S2. Real-time data acquisition and model update

[0100] 1) Activate the multi-source sensor to collect data on temperature, upsetting force, current, and displacement in the welding zone, with a sampling interval of 0.1s;

[0101] 2) Filter and denoise the raw data to extract features such as temperature change rate and current fluctuation;

[0102] 3) Input the processed data into the digital twin model and update the boundary conditions (such as heat source intensity and constraint conditions) synchronously to ensure that the virtual model is synchronized with the actual process in real time.

[0103] S3. Multiphysics Simulation and AI Prediction

[0104] 1) Run the multiphysics coupling model to predict the temperature field (peak 1500℃), stress field (maximum 300MPa), and displacement field (maximum 1mm) of the welding zone.

[0105] 2) Activate the AI ​​prediction model:

[0106] The temperature prediction model outputs the trend of temperature change in the welding zone in the next 5 seconds.

[0107] The microstructure model predicts the grain size (14±2μm) and phase distribution (65% ferrite, 30% pearlite) in the heat-affected zone.

[0108] The defect detection model analyzes weld images in real time to identify defects such as porosity and inclusions (accuracy ≥ 85%).

[0109] The performance prediction model outputs tensile strength (580±20MPa) and fatigue life (≥1×10). 6 Second-rate).

[0110] S4. Multi-objective optimization and decision-making

[0111] If the AI ​​prediction results do not meet the requirements (e.g., tensile strength < 550 MPa, defect rate > 5%), multi-objective optimization is triggered:

[0112] 1) Input the current process parameters and performance constraints, and start the NSGA-II algorithm;

[0113] 2) Generate the Pareto optimal solution set (50 candidate solutions) and calculate the weighted score of each solution (strength weight 0.3, energy consumption weight 0.25, deformation weight 0.2, efficiency weight 0.15, defect rate weight 0.1).

[0114] 3) Select the parameter group with the highest weighted score (such as preheating temperature 820℃, flash current 2800A, upsetting force 85000N, cooling rate 35℃ / s).

[0115] S5. Closed-loop control and visual feedback

[0116] 1) Output the optimized process parameters to the welding equipment and perform parameter adjustments;

[0117] 2) Real-time updates of the advanced visualization interface, displaying the optimized temperature field, stress field, microstructure, and quality score (≥90 points).

[0118] 3) Store the welding data (process parameters, performance results, simulation data) and support historical queries and process iteration optimization.

[0119] To make the objectives, technical solutions, and advantages of this invention clearer, the invention will be further described in detail below with reference to the accompanying drawings and specific embodiments.

[0120] Example:

[0121] 490CL wheel steel flash welding.

[0122] This embodiment takes the flash welding of 490CL wheel steel for automobiles as an example to illustrate the specific implementation process and effects of the system of the present invention.

[0123] 1. System initialization and parameter matching

[0124] like Figure 1 As shown, the system architecture of this invention includes a material property database module, a multi-source data acquisition and preprocessing module, a multiphysics modeling and digital twin module, an AI prediction and multi-objective optimization module, an advanced visualization and interaction module, and a closed-loop control module. Each module works collaboratively through data interfaces and communication protocols.

[0125] First, the operator selects the target material "490CL wheel steel" in the system. The material property database module automatically matches the material's composition parameters (C 0.45%, Mn 0.65%, Si 0.25%, P 0.02%, S 0.015%), thermal properties parameters (thermal conductivity 45 W / (m·K), specific heat capacity 460 J / (kg·K)), and phase transformation kinetic parameters (e.g., Ac1=723℃, Ac3=850℃, Ms=350℃). The system automatically loads the corresponding multiphysics model parameters based on the material data and initializes the simulation environment.

[0126] Next, configure the system operating parameters: set simulation accuracy to "High", rendering quality to "Ultra", enable real-time mode, and enable GPU acceleration. After system initialization, the interface will display as follows. Figure 2 The multi-view layout shown includes a main view of wheel welding, a cross-sectional view, a temperature field view, a stress field view, and a phase distribution view, providing comprehensive visual monitoring of the welding process.

[0127] 2. Welding process initiation and real-time data fusion

[0128] Set the initial process parameters: preheating temperature 800℃, flash current 3000 A, forging force 80000 N, cooling rate 50℃ / s. Start the welding equipment; the multi-source data acquisition module starts working synchronously.

[0129] The infrared thermometer collects the temperature of the welding area in real time at a sampling rate of 1000 Hz;

[0130] Pressure sensors monitor changes in upsetting force;

[0131] Hall current sensors acquire welding current signals;

[0132] Displacement sensors record upsetting speed;

[0133] Acoustic sensors monitor welding spatter and abnormal acoustic emissions.

[0134] The collected raw data is preprocessed using Kalman filtering and wavelet denoising to extract feature parameters (such as temperature change rate and current fluctuation amplitude), which are then input into the digital twin module in real time to update the boundary conditions and heat source intensity of the thermo-mechanical coupling model. The digital twin achieves a 1:1 real-time mapping with the actual welding process with a time synchronization error of ≤0.1 s.

[0135] 3. Multiphysics simulation and AI real-time prediction

[0136] The multiphysics modeling module, based on updated boundary conditions, runs coupled thermo-mechanical-electro-magnetic simulations to predict the temperature, stress, and displacement fields in the welding zone in real time. Simulation results show:

[0137] The peak temperature of the welding zone is 1480℃, and the width of the heat-affected zone is approximately 25mm.

[0138] The maximum residual stress is 180 MPa, which does not exceed the yield strength of the parent material (σ). y =500 MPa) 80%;

[0139] The microstructure evolution model predicts that the grain size of the heat-affected zone is 16 μm, with a ferrite content of 63% and a pearlite content of 32%.

[0140] At the same time, the AI ​​prediction submodule is activated:

[0141] The temperature prediction neural network outputs the temperature trend of the welding area within the next 5 seconds.

[0142] A CNN model based on ResNet-18 predicts the phase distribution with a confidence level of ≥91.8%.

[0143] The YOLOv5 defect detection model analyzed the weld image in real time and found no obvious defects.

[0144] The random forest performance prediction model outputs a tensile strength of 520 MPa, which is lower than the preset threshold (≥550 MPa).

[0145] 4. Multi-objective optimization and parameter re-decision

[0146] Since the predicted tensile strength does not meet the requirements, the system automatically triggers a multi-objective optimization process. The optimization objectives include: maximizing joint strength, minimizing energy consumption, minimizing deformation, maximizing production efficiency, and minimizing the defect rate. The NSGA-II algorithm uses the current process parameters as the initial population and generates the Pareto optimal solution set after 200 iterations.

[0147] like Figure 3 As shown, the Pareto front curve clearly illustrates the trade-off between strength, energy consumption, and deformation. The system calculates the comprehensive score of each solution based on preset weights (strength 0.3, energy consumption 0.25, deformation 0.2, production efficiency 0.15, defect rate 0.1), and finally selects the optimal parameter combination.

[0148] Preheating temperature 620℃;

[0149] Flash current 2800A;

[0150] Upsetting force 85000N;

[0151] Upsetting speed: 10.5 mm / s;

[0152] Cooling rate: 35℃ / s.

[0153] 5. Closed-loop control and effect verification

[0154] Optimized parameters are sent to the welding equipment in real time via the closed-loop control module for execution. The system continues to monitor the welding process and dynamically adjusts the parameters.

[0155] When the real-time predicted grain size is close to 14μm, the system fine-tunes the upsetting speed to 2.4mm / s;

[0156] During the final cooling stage, the jet pressure is adaptively adjusted to 0.42 MPa based on temperature feedback.

[0157] After welding, samples were taken from the joint for analysis. The actual test results are as follows:

[0158] Peak temperature 1500℃, heat-affected zone width 26mm;

[0159] Residual stress 150 MPa;

[0160] Grain size 14μm;

[0161] Tensile strength 580 MPa;

[0162] Defect rate: 0.2%.

[0163] Compared with the simulation prediction results, the deviations of all indicators are ≤2%, indicating that the digital twin model has high-precision prediction capabilities. Figure 4As shown, the optimized microstructure is significantly refined and the phase distribution is more uniform, forming a stark contrast with the unoptimized microstructure.

[0164] 6. Comprehensive Benefit Analysis

[0165] In this embodiment, the system completes multi-objective optimization and achieves closed-loop parameter adjustment within 10 seconds, shortening the overall welding cycle by approximately 20%. Energy consumption is reduced from 1.5 kWh to 1.2 kWh, a decrease of 20%. Joint performance is stable, the pass rate is increased to over 99%, and the process debugging cycle is shortened by approximately 40%.

[0166] Summary: This embodiment uses flash welding of 490CL wheel steel as an example to fully demonstrate the implementation process and significant effects of the system of this invention. The system first automatically matches and initializes parameters based on the material property database. Welding data is collected in real time through multi-source sensors, driving a multi-physics coupled digital twin model for 1:1 high-fidelity synchronous simulation. When the AI ​​prediction model identifies that the joint tensile strength (520MPa) is lower than the target value (≥550MPa), the system automatically triggers the NSGA-II multi-objective optimization algorithm, generating and recommending the optimal combination of process parameters (e.g., preheating temperature 820℃, flash current 2800 A) within 10 seconds. The optimized parameters are sent to the welding equipment for execution in real time via the closed-loop control module and dynamically adjusted. Actual welding results show that the joint tensile strength is increased to 580MPa, the grain size is refined to 14μm, energy consumption is reduced by 20%, and the deviation between the actual results and the simulation prediction is ≤2%. This embodiment fully verifies the effectiveness of the system in achieving transparency, intelligent optimization, and precise online control of the welding process, significantly improving welding quality, efficiency, and consistency.

[0167] Of course, the above description is not intended to limit the present invention, and the present invention is not limited to the examples given above. Any changes, modifications, additions or substitutions made by those skilled in the art within the scope of the present invention should also fall within the protection scope of the present invention.

Claims

1. A wheel flash welding optimization system based on digital twins, characterized in that, include: The material property database module is used to store the basic parameters, microstructure parameters and welding process reference parameters of steel used for wheels, and provides a data interface; The multi-source data acquisition and preprocessing module is used to acquire temperature, pressure, current and displacement data of the welding process in real time through the sensor unit, and to preprocess the data. The multiphysics modeling and digital twin module is used to integrate thermo-mechanical coupling models, microstructure evolution models, and electromagnetic-fluid coupling models to construct and update a virtual twin of the wheel flash welding process in real time. The AI ​​prediction and multi-objective optimization module is used to predict welding temperature, microstructure, defects and mechanical properties based on artificial intelligence models, and to generate an optimized set of process parameters using multi-objective optimization algorithms. The advanced visualization and interaction module is used to provide multi-view 3D visualization and interaction of the welding process, physical field distribution, microstructure evolution and optimization results; The closed-loop control module is used to dynamically adjust the process parameters of the welding equipment based on the predicted deviation between the virtual twin and the real-time welding process.

2. The system according to claim 1, characterized in that, The basic parameters of wheel steel stored in the material property database module include: steel composition, thermophysical parameters, mechanical property parameters, and phase transformation temperature parameters.

3. The system according to claim 1 or 2, characterized in that, The sensor units of the multi-source data acquisition and preprocessing module include: an infrared thermometer, a pressure sensor, a Hall current sensor, and a displacement sensor.

4. The system according to claim 1, characterized in that, In the multiphysics modeling and digital twin module, the microstructure evolution model is constructed based on the Johnson-Mehl-Avrami equation, the dynamic recrystallization model, and the Ostwald ripening model.

5. The system according to claim 1, characterized in that, In the AI ​​prediction and multi-objective optimization module, the artificial intelligence model includes a neural network for temperature prediction, a convolutional neural network for micro-tissue identification, a target detection model for defect detection, and a random forest model for comprehensive performance prediction; the multi-objective optimization algorithm is the NSGA-II genetic algorithm.

6. The system according to claim 5, characterized in that, The optimization objectives of the multi-objective optimization algorithm include: maximizing joint strength, minimizing energy consumption, minimizing deformation, maximizing production efficiency, and minimizing defect rate.

7. The system according to claim 1, characterized in that, The advanced visualization and interaction module provides the following views: a main view showing the 3D model of the wheel welding, a cross-sectional view showing the weld section, a physical field view showing the temperature field or stress field distribution, and a trend view showing the quality or performance trend.

8. An optimization method for flash welding of wheels based on digital twins, characterized in that, The system according to any one of claims 1 to 7 comprises the following steps: S1. Based on the target wheel steel material, match relevant parameters from the material property database and initialize the system; S2. During the welding process, welding data is collected in real time, and the boundary conditions of the digital twin model are updated after preprocessing; S3. Run multiphysics coupling simulation and call AI prediction model to obtain prediction results of welding temperature field, stress field, microstructure, defects and mechanical properties; S4. If the prediction results do not meet the preset performance requirements, the multi-objective optimization algorithm is started to generate the Pareto optimal solution set based on the current parameters and constraints, and the optimal combination of process parameters is selected from it. S5. Output the optimized process parameters to the welding equipment for execution, update the visualization interface in real time, and store the welding process data.

9. The method according to claim 8, characterized in that, In step S4, the multi-objective optimization algorithm is the NSGA-II algorithm, and its optimization objective function includes at least joint strength, energy consumption and deformation.

10. The method according to claim 8, characterized in that, In step S5, the closed-loop control module dynamically adjusts the upsetting force, sintering current, or cooling medium pressure based on the deviation between the predicted residual stress, grain size, or final cooling temperature and the target threshold.

Images