Laser welding simulation method, device, equipment and product based on digital twinning

Abstract

The application provides a laser welding simulation method, device, equipment and product based on digital twinning. The method constructs a digital twinning model based on an adaptive grid optimization mechanism, receives and deploys initial welding parameters including at least laser power, welding speed and defocusing amount, performs multi-physical field coupling simulation on a welding process, obtains welding prediction results including at least molten pool shape, temperature distribution, stress distribution and weld forming effect, and constructs an association graph of welding process and defects based on a preset defect quantification judgment standard and a natural language processing technology, so as to generate a visual risk mark at a corresponding position in a three-dimensional space of the digital twinning model. The method can predict, identify and locate key physical states and potential defects in the welding process before trial welding or in a welding analysis stage, thereby improving the predictability, monitorability and traceability of the laser welding process, and improving the process debugging efficiency and welding quality stability.

Description

Technical Field

[0001] This application relates to the field of laser welding technology, and in particular to a laser welding simulation method, apparatus, equipment and product based on digital twins. Background Technology

[0002] In high-end manufacturing fields such as welding of new energy vehicle battery packs and joining of precision aerospace components, laser welding technology has become a core process due to its high energy density and rapid prototyping capabilities. However, the laser welding process is extremely sensitive to parameters (such as power, speed, and defocusing amount), and the welding quality directly affects product performance and safety.

[0003] With existing laser welding technology, engineers need to rely on experience to repeatedly adjust parameters (such as power and speed), obtain welding information through surface imaging, and store welding parameters, equipment status, environmental data, etc. in different systems to complete laser welding.

[0004] However, the above technical solutions rely on manual experience to adjust parameters, resulting in poor welding quality stability. Summary of the Invention

[0005] This application provides a laser welding simulation method, apparatus, equipment and products based on digital twins, to solve problems such as low efficiency of parameter setting, insufficient process status perception and lag in defect prediction in existing laser welding processes.

[0006] In a first aspect, this application provides a laser welding simulation method based on digital twins for verifying process parameters and predicting defects before trial welding, including:

[0007] A digital twin model is constructed based on an adaptive mesh optimization mechanism. The digital twin model is generated based on a three-dimensional model of the welded workpiece and welding data. The welding data includes laser welding parameters, material properties, and environmental parameters. The adaptive mesh optimization mechanism uses the multiphysics solution results as feedback to adaptively and dynamically adjust the mesh.

[0008] Receive initial welding parameters input by the user, which include at least laser power, welding speed, and defocusing amount;

[0009] The initial welding parameters are deployed into the digital twin model, and the welding process is simulated using multiphysics coupling to obtain welding prediction results. The welding prediction results include at least the molten pool morphology, temperature distribution, stress distribution, and weld formation effect.

[0010] Based on the preset defect quantification judgment standard and natural language processing technology, the welding prediction results are processed for defect identification to construct a correlation map between welding process and defects.

[0011] Based on the correlation map between the welding process and defects, a visual risk marker is generated at the corresponding position in the three-dimensional space of the digital twin model, so that the user can adjust the initial welding parameters based on the visual risk marker to complete the laser welding simulation.

[0012] In one possible implementation, a digital twin model is constructed based on an adaptive grid optimization mechanism, including:

[0013] Based on geometric replication and mesh generation, an initial digital twin model is generated from the 3D model of the welded workpiece.

[0014] The welding data is input into the initial digital twin model to obtain the updated digital twin model;

[0015] By using multiphysics coupling modeling, the temperature gradient, molten pool velocity gradient, and stress gradient are solved in real time to obtain gradient feedback data; a digital twin model is constructed based on the adaptive mesh optimization mechanism and the gradient feedback data.

[0016] In one possible implementation, a digital twin model is constructed based on an adaptive grid optimization mechanism and the gradient feedback data, including:

[0017] Based on the adaptive grid optimization mechanism, a preset gradient threshold is determined;

[0018] The gradient feedback data is compared with the preset gradient threshold to obtain the comparison result;

[0019] If the comparison result indicates that any of the temperature gradient, the melt pool velocity gradient, and the stress gradient does not meet the preset gradient threshold, then mesh update and partitioning are performed.

[0020] The grid density is dynamically adjusted based on the updated grid division to construct a digital twin model.

[0021] In one possible implementation, the method of constructing a digital twin model further includes:

[0022] In the offline state before welding begins, the digital twin model is calibrated based on historical test welding data and error compensation algorithm to obtain a digital twin model that meets the preset simulation welding requirements.

[0023] During the welding process, welding data is collected in real time based on multiple sensors;

[0024] According to the timestamp synchronization algorithm, the welding data is timestamped using a unified clock reference, and the communication link delay is compensated to obtain updated welding data;

[0025] The updated welding data is mapped to the digital twin model in real time, thereby completing the real-time calibration of the digital twin model.

[0026] In one possible implementation, the initial welding parameters are deployed into the digital twin model, and a multiphysics coupled simulation of the welding process is performed to obtain welding prediction results, including:

[0027] A multiphysics solver is determined, which includes at least a thermal solver, a fluid solver, and a structural solver;

[0028] The initial welding parameters are deployed into the digital twin model, and the multiphysics solver is called to perform multiphysics coupling simulation to obtain the welding prediction results.

[0029] The multiphysics coupling simulation includes: the temperature field output by the thermal solver is used as the basis for correcting the physical property parameters of the fluid solver; the velocity field of the molten pool obtained by the fluid solver is fed back to the structural solver to update the local load distribution; the deformation results calculated by the structural solver correct the position of the subsequent heat source, forming a closed-loop coupling relationship.

[0030] In one possible implementation, based on a preset defect quantification judgment standard and natural language processing technology, the welding prediction results are subjected to defect identification processing to construct a correlation map between welding processes and defects, including:

[0031] Based on natural language processing technology, the welding prediction results are processed for defect identification to determine multidimensional defect features, which include at least penetration depth features, stress features and temperature features.

[0032] Based on the multidimensional defect characteristics and the preset defect quantification judgment criteria, a correlation map between welding process and defects is constructed.

[0033] The correlation diagram between welding process and defects represents the relationship between welding parameters, operating conditions and defect types.

[0034] In one possible implementation, before deploying the initial welding parameters into the digital twin model and performing multiphysics coupling simulation of the welding process to obtain the welding prediction results, the method further includes:

[0035] Real-time data collection of molten pool, stress, and temperature during welding;

[0036] Based on the current position of the welding torch, the workpiece coordinate system, and the pre-established geometric registration relationship, the data on the molten pool, stress, and temperature are mapped to the digital twin model to generate real-time monitoring data.

[0037] Secondly, this application proposes a laser welding simulation device based on digital twins for verifying process parameters and predicting defects before trial welding, including:

[0038] The model building module is used to build a digital twin model based on an adaptive mesh optimization mechanism. The digital twin model is generated based on a 3D model of the welded workpiece and welding data. The welding data includes laser welding parameters, material properties and environmental parameters. The adaptive mesh optimization mechanism uses the multiphysics solution results as feedback to adaptively and dynamically adjust the mesh.

[0039] The parameter receiving module is used to receive the initial welding parameters input by the user, which include at least the laser power, welding speed and defocusing amount;

[0040] The multiphysics coupling simulation module is used to deploy the initial welding parameters into the digital twin model, perform multiphysics coupling simulation on the welding process, and obtain welding prediction results. The welding prediction results include at least the molten pool morphology, temperature distribution, stress distribution, and weld formation effect.

[0041] The welding defect determination module is used to perform defect identification processing on the welding prediction results based on preset defect quantification judgment standards and natural language processing technology, and to construct a correlation map between welding process and defects.

[0042] The defect marking module is used to generate a visual risk mark at the corresponding position in the three-dimensional space of the digital twin model based on the correlation map between the welding process and the defect, so that the user can adjust the initial welding parameters based on the visual risk mark to complete the laser welding simulation.

[0043] Thirdly, this application provides a laser welding simulation device based on digital twins, including: a memory and a processor;

[0044] The memory stores computer-executed instructions;

[0045] The processor executes computer execution instructions stored in the memory, causing the processor to perform the first aspect and / or various possible implementations of the first aspect as described above.

[0046] Fourthly, this application provides a computer program product, including a computer program that, when executed by a processor, implements the first aspect and / or various possible implementations of the first aspect.

[0047] This application provides a laser welding simulation method, apparatus, equipment, and product based on digital twins. The method constructs a digital twin model based on an adaptive mesh optimization mechanism, receives and deploys initial welding parameters including at least laser power, welding speed, and defocusing amount, performs multi-physics coupling simulation on the welding process, and obtains welding prediction results including at least molten pool morphology, temperature distribution, stress distribution, and weld formation effect. Based on preset defect quantification judgment criteria and natural language processing technology, it constructs a correlation map between welding process and defects, so as to generate visual risk markers at corresponding positions in the three-dimensional space of the digital twin model. It can predict, identify, and locate key physical states and potential defects in the welding process before trial welding or during the welding analysis stage, thereby improving the predictability, monitorability, and traceability of the laser welding process, and improving process debugging efficiency and welding quality stability. Attached Figure Description

[0048] The accompanying drawings, which are incorporated in and form part of this specification, illustrate embodiments consistent with this application and, together with the description, serve to explain the principles of this application.

[0049] Figure 1 A flowchart illustrating an embodiment of the laser welding simulation method based on digital twins provided in this application;

[0050] Figure 2 A schematic flowchart of Embodiment 2 of the laser welding simulation method based on digital twin provided in this application;

[0051] Figure 3 A schematic flowchart of Embodiment 3 of the laser welding simulation method based on digital twin provided in this application;

[0052] Figure 4 A schematic diagram of the structure of the laser welding simulation device based on digital twin provided in this application;

[0053] Figure 5 A schematic diagram of the structure of the laser welding simulation equipment based on digital twin provided in this application.

[0054] The accompanying drawings illustrate specific embodiments of this application, which will be described in more detail below. These drawings and descriptions are not intended to limit the scope of the concept in any way, but rather to illustrate the concept of this application to those skilled in the art through reference to particular embodiments. Detailed Implementation

[0055] Exemplary embodiments will now be described in detail, examples of which are illustrated in the accompanying drawings. When the following description relates to the drawings, unless otherwise indicated, the same numbers in different drawings denote the same or similar elements. The embodiments described in the following exemplary embodiments do not represent all embodiments consistent with this application. Rather, they are merely examples of apparatuses and methods consistent with some aspects of this application as detailed in the appended claims.

[0056] This application primarily targets laser welding processes in high-end manufacturing, such as welding of new energy vehicle battery packs and joining precision structural components in aerospace. The system is deployed in the industrial network environment of the welding workshop, interconnecting with laser welding machines, sensors (infrared thermal imagers, high-speed cameras, force sensors), and other equipment via industrial protocols to establish real-time data interaction between the digital twin model and the physical welding system. The digital twin model runs on a local server or cloud computing platform, displaying the welding status through multi-terminal visual interfaces such as industrial large screens and workshop terminals. It supports real-time intervention by engineers to optimize parameters and stores the entire process data in a structured database, providing a basis for subsequent process iterations and defect analysis.

[0057] In existing technologies, laser welding processes lack systematic simulation and verification methods. Engineers must rely on experience to repeatedly adjust parameters (such as power and speed), resulting in high trial-and-error costs and low efficiency. During welding, the molten pool is in a high-temperature, high-speed dynamic state. Traditional monitoring methods can only obtain limited information through surface imaging and cannot directly observe key parameters such as internal temperature and stress distribution of the molten pool. Defects (such as incomplete fusion and porosity) are often only discovered after welding through non-destructive testing, leading to high rework costs. Furthermore, welding parameters, equipment status, and environmental data are scattered across different systems (such as MES, PLC, and testing equipment), lacking unified management. This makes defect root cause analysis difficult, and process optimization relies on manual experience, making it difficult to form a closed-loop iteration. These problems are particularly prominent in scenarios with extremely high welding quality requirements, such as welding of new energy vehicle batteries and joining precision components in aerospace.

[0058] To address the aforementioned technical challenges, this application constructs a digital twin model of laser welding to achieve closed-loop control of the entire welding process, encompassing simulation prediction, real-time monitoring, automatic parameter optimization, and data traceability. Based on multiphysics coupling modeling and combining real-time sensor data with simulation results, this concept proposes a dynamic parameter adjustment method based on particle swarm optimization (PSO) algorithm. Furthermore, it achieves full-process data traceability through a unified data link design. This technical concept overcomes the limitations of traditional welding processes, which rely on manual experience, lack process visibility, and suffer from fragmented data. It solves core problems such as poor welding quality stability, high trial-and-error costs, and difficulty in tracing defect sources.

[0059] The technical solution of this application and how the technical solution of this application solves the above-mentioned technical problems are described in detail below with specific embodiments. These specific embodiments can be combined with each other, and the same or similar concepts or processes may not be described again in some embodiments. The embodiments of this application will now be described with reference to the accompanying drawings.

[0060] Figure 1 This is a flowchart illustrating an embodiment of the laser welding simulation method based on digital twins provided in this application. Figure 1 As shown, the method includes:

[0061] S101. Construct a digital twin model based on an adaptive grid optimization mechanism.

[0062] Among them, the digital twin model is generated based on the three-dimensional model of the welded workpiece and welding data. It refers to a computable model that is established for the physical object to be welded and can reflect the workpiece's geometry, material behavior, process input and environmental boundary conditions in virtual space. It is not just a static three-dimensional display model, but a composite data model that includes geometric, material, process and environmental layers.

[0063] The 3D model of the welded workpiece can be derived from a 3D solid file, surface file, or mesh file exported from a computer-aided design system. After parsing the file, information such as welding path, joint type, plate thickness distribution, boundary contour, and clamping contact area can be extracted.

[0064] Welding data includes laser welding parameters, material properties, and environmental parameters. Laser welding parameters characterize the process input applied to the workpiece by the welding equipment during processing; material properties characterize the response characteristics of the workpiece material under thermal and stress effects; and environmental parameters characterize the external conditions at the welding site that affect heat exchange and process stability.

[0065] The adaptive mesh optimization mechanism uses the multiphysics solution results as feedback to adaptively and dynamically adjust the mesh. This mechanism uses the gradient data of the real-time multiphysics solution results as the core feedback basis, optimizes the traditional static fixed mesh generation mode, realizes dynamic fine adjustment and sparsity adaptation of the mesh across the entire welding domain, and balances simulation calculation accuracy and computational efficiency. It solves the industry pain points of insufficient calculation accuracy in the molten pool area, waste of computing power in non-critical areas, and distortion of dynamic welding process simulation caused by the fixed simulation mesh in traditional simulation.

[0066] The specific implementation method of this step includes:

[0067] Based on geometric replication and mesh generation, an initial digital twin model is generated from the 3D model of the welded workpiece.

[0068] Welding data is input into the initial digital twin model to obtain an updated digital twin model;

[0069] By using multiphysics coupling modeling, the temperature gradient, molten pool velocity gradient, and stress gradient are solved in real time to obtain gradient feedback data;

[0070] A digital twin model is constructed based on an adaptive grid optimization mechanism and gradient feedback data.

[0071] Specifically, in this embodiment, the executing entity can be a simulation server, process design workstation, or cloud-edge collaborative processing platform deployed in the industrial network of the welding workshop. The executing entity first receives a 3D model of the welding workpiece. In one possible embodiment, the 3D model can be directly retrieved through a data interface with the product design system; in another embodiment, it can be imported by manually uploading the model file.

[0072] Upon receiving the 3D model, geometric replication and mesh generation are performed. Geometric replication refers to virtually mapping the geometric boundaries of the workpiece based on the 3D model, ensuring that the final digital twin model corresponds to the physical workpiece in shape and position. For example, geometric replication ensures this correspondence by analyzing the boundary contours and clamping contact areas of the 3D model of the welded workpiece. Mesh generation refines the mesh in the molten pool and weld areas through discretization of computational units to improve computational accuracy, thereby supporting numerical solutions for local temperature, flow, and stress.

[0073] Meshing refers to discretizing a three-dimensional model into multiple computational units to enable numerical solutions for local temperature, flow, and stress. Denser meshes can be used in the molten pool and weld regions to improve computational accuracy. In practical applications, other mesh types can also be selected.

[0074] For welding data, it is used to drive the state correction of the initial digital twin model. Its contents can include laser power, welding speed, defocusing amount, material thermal properties, and ambient temperature and humidity parameters. After the welding data is input into the initial digital twin model, the boundary conditions, thermal input items, and material parameters can be updated simultaneously, so that the updated digital twin model can reflect the real-time state of the workpiece to be welded under the current process settings.

[0075] Unlike traditional static modeling methods, this application, based on an updated digital twin model, initiates real-time multiphysics coupling and simultaneous computation of three core gradient data: a heat conduction model is used to solve the global temperature gradient of the workpiece, accurately locating areas of rapid high-temperature changes and heat concentration; a molten pool dynamics model is used to solve the velocity gradient of the liquid metal inside the molten pool, capturing areas of intense molten pool flow and abrupt changes in surface fluctuations; and a stress-strain model is used to solve the transient stress gradient during welding, identifying high-risk areas of stress concentration and rapid stress changes. Finally, the data is integrated to obtain global gradient feedback data, providing a quantitative basis for dynamic mesh optimization.

[0076] Based on this, an adaptive mesh optimization mechanism is used to dynamically adjust the mesh. The core optimization logic is as follows: based on the magnitude and rate of change of the gradient feedback data, the mesh in different regions is adjusted adaptively. For core regions with large and rapidly changing temperature gradients, flow velocity gradients, and stress gradients (i.e., the core area of ​​the molten pool, the weld fusion zone, and the high-risk area of ​​stress concentration), the mesh is automatically refined, adaptively increasing the mesh density to the 0.1mm level. This accurately captures the subtle physical processes such as the dynamic evolution of the molten pool, sudden temperature changes, and stress abrupt changes, preventing insufficient simulation accuracy and missed defects in critical areas. For non-critical regions with small gradient values ​​and stable physical states (i.e., the low-temperature steady-state region of the workpiece substrate far from the weld), the mesh is automatically thinned, adaptively relaxing the mesh density to the 1mm level. This significantly reduces invalid computational units, lowers simulation computational power consumption, and shortens the simulation iteration cycle.

[0077] Meanwhile, this mechanism supports dynamic real-time mesh adjustment throughout the entire process. It can follow the movement of the laser welding heat source, the evolution of the molten pool shape, and the dynamic accumulation of thermal stress, update the gradient feedback data step by step over time, and iterate the mesh density in real time. This achieves a closed-loop adaptive iteration of "simulation solution - gradient feedback - mesh optimization - re-solution", which completely solves the defects of traditional fixed meshes that cannot adapt to the dynamic changes in the welding process and the imbalance between steady-state and transient simulation accuracy.

[0078] On the other hand, after generating the initial digital twin model, temperature distribution data of the molten pool area can be collected by an infrared thermal imager, and the temperature matrix can be mapped to the three-dimensional surface mesh of the digital twin model. Based on the real-time temperature data, the mesh is refined in the high-temperature area of ​​the molten pool (temperature above the liquidus temperature) (mesh density increased to 0.1 mm level), and the mesh is sparsed in the low-temperature area far from the molten pool (mesh density relaxed to 1 mm level). In subsequent multiphysics coupling simulation, the mesh density of the molten pool dynamics, heat conduction and stress-strain models can be dynamically updated to ensure high-precision prediction of molten pool morphology, temperature distribution and stress distribution.

[0079] By adopting the above construction method, the digital twin model can gradually evolve from a geometric representation into a computationally capable welding virtual body, and achieve synchronous parameter updates after welding data input. Since heat conduction, molten pool dynamics, and stress-strain models are integrated into the same model, it can simultaneously reflect the temperature field, flow field, and mechanical response during the welding process, thereby improving the accuracy of predictions on weld penetration, forming quality, and deformation risk, and providing a consistent model basis for subsequent welding simulation and quality assessment.

[0080] Another aspect of this step includes:

[0081] In offline mode, the digital twin model is calibrated based on the test welding data and error compensation algorithm to obtain a digital twin model that meets the preset simulation welding requirements;

[0082] During the welding process, welding data is collected in real time based on multiple sensors;

[0083] Based on the timestamp synchronization algorithm, welding data is mapped to the digital twin model in real time, completing the real-time calibration of the digital twin model.

[0084] Among them, the test welding data is used to characterize the weld formation, penetration depth, heat input fluctuation and defect distribution information obtained under actual welding conditions, and the error compensation algorithm is used to correct the parameter deviation between the digital twin model and the real welding process so that the output of the final digital twin model is consistent with the physical welding result.

[0085] Multiple sensors may include infrared thermal imagers, high-speed cameras, temperature sensors, displacement sensors, and current and voltage sensors. Each sensor is used to collect status information such as molten pool temperature, weld seam image, workpiece deformation, and welding electrical parameters.

[0086] The timestamp synchronization algorithm is used to time-align welding data under different sampling frequencies and communication links, and to map the aligned data to the corresponding spatial location and physical state parameters of the digital twin model. For example, a unified clock reference is used to timestamp multi-sensor data (such as infrared thermal imager and high-speed camera) and compensate for communication link delays, so that data with different sampling frequencies are updated synchronously in the model along the time axis.

[0087] In practical implementation, the offline calibration process can first establish a reference dataset based on the weld cross-sectional dimensions, surface forming quality, and inspection results obtained from the test weld samples. Then, the reference dataset is input into an error compensation algorithm to correct the thermal conductivity coefficient, heat source distribution parameters, molten pool boundary parameters, and stress-strain parameters, ensuring that the digital twin model output meets the preset simulation welding requirements. The error compensation algorithm can employ a parameter correction method based on least squares fitting or a deviation learning method based on a regression model. In practical applications, this algorithm can also adaptively correct for material batch differences and equipment condition changes; however, this embodiment does not limit this approach.

[0088] During online calibration, each sensor uploads the collected data to the edge computing unit through the industrial communication network. The edge computing unit timestamps the data and compensates for delays based on a unified clock reference, and then writes the synchronized welding data into the corresponding node of the digital twin model in real time, thereby updating the temperature field, molten pool morphology and weld condition parameters in the model.

[0089] The aforementioned execution process ensures that the digital twin model possesses high simulation consistency before actual welding. Simultaneously, through real-time data acquisition from multiple sensors and synchronized timestamp mapping, dynamic corrections are achieved during the welding process, enabling the model to continuously reflect the actual welding state. Based on offline and online calibration modes, the digital twin model possesses static accuracy before welding and dynamic response capabilities during welding, reducing the impact of material differences, equipment drift, and environmental fluctuations on simulation deviations, improving the reliability of welding prediction results, and enhancing the real-time characterization capabilities of molten pool evolution, weld formation, and defect risks.

[0090] By constructing a digital twin model containing multi-dimensional information on geometry, process, materials, and environment, the steps in this application can form a virtual mapping basis consistent with the actual welding object in the pre-welding stage. This allows subsequent simulation results to no longer rely on simple empirical estimations, but to be based on traceable structures and boundary conditions, which provides a necessary prerequisite for predicting welding risks of complex structural workpieces.

[0091] S102, Receive the initial welding parameters input by the user.

[0092] In this step, the initial welding parameters refer to the first set of process parameters to be evaluated that the user inputs into the simulation system before process verification. These parameters are used to drive the digital twin model to perform one or more welding process solutions and include at least laser power, welding speed, and defocusing amount.

[0093] Among these parameters, laser power characterizes the energy level of the laser beam output to the workpiece surface per unit time, typically affecting the weld pool size, weld depth, and evaporation state; welding speed characterizes the linear velocity of the heat source moving along the welding path, typically affecting the heat input per unit length of weld; and defocusing amount characterizes the axial offset of the focal spot relative to the workpiece surface, directly affecting the spot energy density and energy penetration depth. Since laser welding quality is highly sensitive to the coupling relationship among these three parameters, this step not only receives the parameters but also performs standardized analysis and validity verification.

[0094] Specifically, in this embodiment, users can manually input initial welding parameters through the process parameter input interface, or input parameters by importing process card files, reading equipment recipe files, or retrieving templates from a historical process database. In one possible embodiment, the interface includes a parameter input field, a unit conversion module, and a parameter integrity check module. After inputting the laser power, the process engineer can select watts or kilowatts as the unit, and the system will automatically complete the unified conversion; the welding speed can be input in millimeters per second or meters per minute, and the system will convert it according to a preset standard unit; the defocusing amount can be input in millimeters and marked as positive or negative defocusing.

[0095] In addition, to prevent abnormal parameters from causing subsequent simulations to fail to converge or to distort the simulation results, the system can perform validity checks upon receiving parameters. For example, it can verify whether the laser power is within the equipment's rated power range, whether the welding speed is higher than the equipment's minimum stable operating speed but lower than the trajectory tracking upper limit, and whether the defocusing amount is within the focal depth range allowed by the optical system. If the parameters are out of range, a prompt message is returned to the user, requesting re-entry or confirmation of whether to use the extended verification mode.

[0096] In addition to the basic parameters, this embodiment can also receive supplementary parameters associated with the initial welding parameters in the same input session, such as the laser mode (continuous or pulsed), shielding gas type and flow rate, oscillation amplitude and frequency, welding direction, arc initiation and termination strategies, etc., and store them in the parameter set as extended solution conditions. To facilitate subsequent parameter tracking and result comparison, the system can generate a unique parameter version identifier for each set of initial welding parameters and record the input source, input time, operator information, and corresponding workpiece number, thereby forming a traceable simulation task record. Detailed steps will be described in Embodiment 2.

[0097] The result of this step is to obtain a structured, verifiable, and traceable initial set of welding parameters and establish its correlation with the current digital twin model. Based on the above analysis, it is evident that by standardizing the reception and constraint checks of laser power, welding speed, and defocusing amount, solution deviations caused by parameter input errors and unit inconsistencies can be reduced, providing a clear input basis for pre-welding simulation. Simultaneously, through parameter versioning management, consistent data can be provided for the comparative evaluation of different process combinations, thereby improving the standardization and traceability of the process development process.

[0098] S103. Deploy the initial welding parameters into the digital twin model, perform multi-physics coupling simulation of the welding process, and obtain the welding prediction results.

[0099] In this step, parameter deployment refers to mapping the verified initial welding parameters to the heat source model, boundary condition model, and path control model corresponding to the digital twin model, so that they become the actual inputs driving the evolution of the virtual welding process.

[0100] Multiphysics coupling simulation refers to the joint solution process that simultaneously considers heat conduction, material melting and solidification, thermo-elastic-plastic stress-strain evolution, and, when necessary, the effects of fluid behavior.

[0101] The welding prediction results include at least the molten pool morphology, temperature distribution, stress distribution, and weld formation effect. Among them, the molten pool morphology is used to characterize the spatial geometry of the liquid metal region at the welding moment or in the welding section; the temperature distribution is used to characterize the temperature field at different locations of the workpiece over time; the stress distribution is used to characterize the residual stress and transient stress state inside the workpiece during welding and cooling; and the weld formation effect is used to characterize evaluable indicators such as weld width, penetration depth, reinforcement height, continuity, and uniformity of formation.

[0102] The specific implementation method of this step includes:

[0103] A multiphysics solver is determined, and then the initial welding parameters are deployed into the digital twin model. The multiphysics solver is called to perform multiphysics coupling simulation to obtain the welding prediction results.

[0104] The multiphysics solver includes at least a thermal solver, a fluid solver, and a structural solver. The thermal solver establishes the heat conduction equation and calculates the temperature field evolution in the welding area; its output temperature field serves as the basis for correcting the physical property parameters of the fluid solver. The fluid solver describes the flow behavior of the metal in the molten pool under the influence of surface tension, buoyancy, and evaporation backpressure; its obtained molten pool velocity field is fed back to the structural solver to update the local load distribution. The structural solver analyzes the stress, strain, and deformation distribution of the workpiece under thermal load; its calculated deformation results correct the subsequent heat source application location, forming a closed-loop coupling relationship. Through these interconnected relationships, the welding prediction results include molten pool morphology, temperature distribution, stress distribution, and weld formation effect, significantly improving the completeness and reliability of the simulation results.

[0105] In this application, the solver can be deployed within the same simulation platform and coupled and transmitted through a unified data interface. In practical applications, other models of the solver platform can also be selected, and this application embodiment does not limit this.

[0106] After determining the solver, the initial welding parameters can be deployed into the digital twin model. In this step, they are typically converted into simulation boundary conditions, initial conditions, or load inputs and loaded into the mesh model corresponding to the welded workpiece, allowing the heat input, molten pool dynamics, and structural response to be solved collaboratively on the same time axis. During the multiphysics coupled simulation, the temperature field output by the thermal solver can be used as a basis for correcting the physical property parameters of the fluid solver. The molten pool velocity field obtained by the fluid solver can be fed back to the structural solver to update the local load distribution. The deformation results calculated by the structural solver are further used to correct the subsequent heat source application positions, thus forming a closed-loop coupling relationship. The final output includes welding simulation results such as weld formation, penetration depth, temperature distribution, and residual stress. This method allows for joint verification of parameter combinations before welding, reducing the bias caused by single-physics analysis and improving the completeness and reliability of welding prediction results.

[0107] For example, in specific implementation, the system first corrects the intensity, range, and speed of the moving heat source based on the initial welding parameters. For instance, laser power directly determines the total input energy of the heat source; welding speed determines the displacement of the heat source along the weld path at each discrete time step; and defocusing affects the focal spot size and peak energy density, thus changing the width and depth of the heat source distribution function. If a Gaussian heat source model is used, the heat flux density can be expressed as a distribution function that decays with radial distance, with laser power determining the total integral value and defocusing determining the standard deviation or equivalent spot radius. If a double-ellipsoidal heat source model is used, the energy distribution parameters and depth parameters of the front and rear ellipsoids can be set according to the workpiece thickness and welding mode to more accurately describe the energy deposition behavior of deep-penetration welding. After writing the above parameters into the model, the system performs time-domain discrete solution according to the welding path sequence, updating the current heat source position, local heat input, boundary heat transfer, and material state at each time step.

[0108] During the thermal field solution process, the temperature response of each mesh element of the workpiece can be calculated based on the transient heat conduction equation. The phase transition behavior of the material from solid to liquid and then to solidification is handled by combining the material's specific heat capacity, thermal conductivity, and latent heat parameters. For determining the molten pool morphology, regions with temperatures above the liquidus temperature can be identified as liquid metal regions, and the molten pool contour can be obtained through isothermal surface extraction. For some alloy materials, a mushy region can be set between the solidus and liquidus to reflect a semi-solid, semi-liquid state. Subsequently, the thermal field results are input as loads into the structural field solution module. Based on the temperature dependence of the coefficient of thermal expansion, elastic modulus, and yield strength, the distribution of thermal strain, plastic strain, and residual stress during welding is solved. For scenarios where clamping conditions significantly affect deformation, the clamping constraints and release sequence can be loaded into the model to obtain stress evolution results that more closely resemble actual working conditions.

[0109] In terms of weld formation effect extraction, the system can set multiple cross-sections along the welding path, analyze the maximum width of the molten pool, penetration depth, continuity of the fusion boundary, and upper surface formation trend for each cross-section, and comprehensively generate weld formation evaluation data. For thin plate welding, the weld formation can focus on evaluating burn-through risk, fluctuation width, and surface reinforcement height; for thick plate or busbar welding, the effective penetration depth, root fusion condition, and heat-affected zone width can be evaluated. The resulting welding prediction results can include a three-dimensional molten pool cloud map, temperature field time series, residual stress distribution map, and weld cross-section prediction map, or can be converted into structured index data, such as maximum penetration depth, peak temperature, maximum tensile stress, and average weld width.

[0110] This step is crucial for addressing the lack of intuitive predictive capabilities before welding. By realistically deploying initial welding parameters into a digital twin model and performing multiphysics coupled solutions, a heat conduction model calculates internal temperature diffusion within the workpiece, a molten pool dynamics model describes the flow and solidification evolution of the molten pool, and a stress-strain model analyzes local deformation and residual stress caused by thermal cycling. These three elements are linked through a unified solution framework, enabling simultaneous solutions for the thermal field, flow field, and mechanical response, thereby improving the accuracy of predicting weld penetration, forming quality, and deformation risk. This coupled simulation method can predict defects such as incomplete fusion, overheating, and deformation concentration in the pre-welding stage, reducing the number of trial welds and improving process debugging efficiency.

[0111] S104. Based on the preset defect quantification judgment standard and natural language processing technology, perform defect identification processing on the welding prediction results and construct a correlation map between welding process and defects.

[0112] Among them, the preset defect quantification judgment standard refers to the set of rules that are pre-established for common quality defects in laser welding and can transform simulation prediction results into defect judgment criteria. It can be determined by process specifications, industry standards, enterprise historical quality data and material welding characteristics. In this application, the preset defect quantification judgment standard includes, but is not limited to, the following: (1) Insufficient penetration: When the predicted penetration is less than 95% of the target plate thickness, it is judged as insufficient penetration risk; (2) Lack of fusion: When there is a continuous contact interface at the weld boundary that has not reached the melting temperature, it is judged as lack of fusion risk; (3) Excessive deformation: When the workpiece deformation exceeds 0.3 mm, it is judged as excessive deformation risk; (4) Hot cracking tendency: When the peak tensile stress in a specific area exceeds the critical strength of the material at the corresponding temperature, it is judged as hot cracking tendency risk.

[0113] Defect identification processing refers to extracting characteristic quantities representing defect risk from welding prediction results such as molten pool morphology, temperature distribution, stress distribution, and weld formation effect, and comparing them with corresponding quantification thresholds or discrimination models to output information such as defect category, defect location, defect degree, and risk level.

[0114] In this step, intelligent analysis and correlation modeling of simulation data are realized based on natural language processing technology. On the basis of traditional rule discrimination, it breaks through the technical bottleneck of manual interpretation of data and the lack of correlation between defects and process parameters. The specific implementation method includes three layers of logic: rule discrimination, NLP intelligent analysis, and correlation graph construction.

[0115] In one implementation, a defect quantification judgment library can be pre-built. This judgment library includes at least threshold rules for different materials, plate thicknesses, and joint types. For example, when the predicted penetration depth is less than the minimum penetration depth corresponding to the target plate thickness and the design penetration requirement, it is judged as a risk of insufficient penetration depth; when there are continuous contact interfaces at the weld boundary that have not reached the melting temperature, it is judged as a risk of incomplete fusion; when the local temperature on the lower surface of the workpiece is continuously higher than the burn-through threshold and the remaining unmelted thickness is lower than the set limit, it is judged as a risk of burn-through; when the temperature gradient in a specific area is too high and the peak tensile stress exceeds the critical strength of the material at the corresponding temperature, it can be judged as a tendency for hot cracking or a risk of high-stress cracking. For stress-related defects, the residual stress distribution after cooling can also be compared with the material yield strength and the allowable deformation value of the assembly to identify potential instability or cracking hazards in subsequent service stages.

[0116] In terms of algorithm implementation, defect identification can be performed using either rule-based judgment or a combination of rules and statistical models. When using rule-based judgment, the welding prediction results are first spatially partitioned and features extracted. For example, sections are extracted at fixed intervals along the weld centerline, and the weld width, weld depth, maximum temperature, cooling rate, and maximum principal stress of each section are calculated. These features are then compared one by one with preset thresholds to form a local defect judgment result. When using a combined model approach, historical test welding data and inspection results from the enterprise can be used to train a defect classifier. Simulated features are input into the classifier, which outputs the defect probability, and a preset confidence threshold is used as the final judgment criterion. Considering that the core of the claims lies in identifying defects based on preset standards, this embodiment can at least achieve this through interpretable rule-based judgment, ensuring that the system has clear judgment criteria in engineering applications.

[0117] Furthermore, to facilitate subsequent visualization and process decision-making, welding defect information is typically structured and encapsulated. This can include defect identification number, defect type, three-dimensional coordinate location, affected weld segment number, affected length, severity level, triggering criteria, and corresponding predicted feature values. For example, after identifying a non-fusion risk in a weld segment, it can be recorded that it is located 12 to 18 millimeters after the weld start, close to the joint root in the depth direction, with a corresponding penetration depth of only 82% of the design requirement, and a high risk level. This generated data can be directly accessed by the visualization module or used by process engineers for reverse analysis to identify the source of parameter setting problems.

[0118] In another implementation, the process of constructing a correlation map between welding processes and defects includes:

[0119] Based on natural language processing technology, defect identification processing is performed on welding prediction results to determine multidimensional defect features, which include at least penetration depth features, stress features, and temperature features.

[0120] Based on multidimensional defect characteristics and preset defect quantification judgment criteria, a correlation map between welding process and defects is constructed.

[0121] Among them, the correlation map between welding process and defects represents the relationship between welding parameters, operating conditions and defect types.

[0122] Specifically, in this embodiment, the core application logic of natural language processing technology is as follows: massive structured data (melt depth values, temperature field data, stress peak values, weld formation parameters), unstructured cloud map description data, and working condition log data output from multi-physics simulations are subjected to unified semantic parsing and feature extraction. Through semantic word segmentation, feature clustering, and semantic matching algorithms, core features strongly correlated with various defects are automatically selected. Multi-dimensional defect features such as melt depth deviation, stress concentration, temperature change, and forming fluctuation are accurately extracted, discarding invalid and redundant data. This achieves intelligent and precise extraction of defect features, solving the problems of low efficiency, strong subjectivity, and feature omissions associated with traditional manual feature selection.

[0123] After multidimensional defect feature extraction, the system establishes a four-dimensional correlation mapping relationship between process parameters, operating conditions, defect features, and defect types, using preset defect quantification judgment criteria as constraints. This leads to the construction of a visualized and quantifiable welding process-defect correlation map. The map uses laser power, welding speed, defocusing amount, ambient temperature and humidity, material batch, and equipment status as core process nodes, and insufficient penetration, lack of fusion, burn-through, hot cracking, and stress exceeding limits as defect nodes. Correlation weight values ​​characterize the degree of influence of different process parameters on various defects, accurately locating the core causes of defect generation.

[0124] For example, natural language processing technology can automatically analyze implicit correlations such as "excessive welding speed leads to insufficient unit heat input, resulting in insufficient weld root penetration and incomplete fusion defects" and "excessive laser power and rapid cooling rate lead to local stress concentration, causing a tendency for hot cracking." This semantic pattern is then transformed into quantified weights in a graph, enabling the datafication and modeling of empirical process rules. Simultaneously, the graph can be updated and iterated in real time; each set of simulation data and each defect identification result can be added to the graph database, continuously optimizing the correlation accuracy between processes and defects.

[0125] For stress-related defects, the residual stress distribution after cooling can be compared with the material yield strength and the allowable deformation value of the assembly to identify potential instability or cracking hazards in subsequent service stages, and updated to the correlation map to improve the correlation between long-term defect risks and process parameters.

[0126] To facilitate subsequent demonstration and process decision-making, the obtained welding defect information and related graph data are structured and encapsulated. Specifically, this includes defect identifier number, defect type, three-dimensional coordinate location, affected weld segment number, affected length, severity level, triggering criteria, corresponding predicted feature values, and corresponding associated process parameter weights. This step transforms the raw results of multiphysics simulation into quality assessment information that can be directly used for engineering decisions, and is a crucial step in achieving pre-identification of welding risks. By establishing preset quantitative defect judgment standards and systematically identifying the molten pool, temperature, stress, and forming results, subjective differences arising from relying solely on manual interpretation of complex simulation cloud maps can be avoided. This allows different process engineers to arrive at consistent defect conclusions when faced with the same predicted results, thereby improving the standardization and efficiency of welding quality assessment.

[0127] S105. Based on the correlation map between welding process and defects, generate visual risk markers at the corresponding positions in the three-dimensional space of the digital twin model to complete the laser welding simulation.

[0128] In this step, the visual risk marker refers to the graphical prompts overlaid on the 3D scene of the digital twin model, which are used to establish an intuitive association between the defect identification results and the actual spatial location of the workpiece.

[0129] The corresponding position in three-dimensional space refers to the geometric mapping point, region, or volume range of defect information in the digital twin model, which can be determined based on the coordinates of mesh nodes, weld path parameter coordinates, cross-section index, or solid surface position.

[0130] In addition, completing laser welding simulation is not limited to completing the calculation task itself, but also includes outputting the simulation results in an interactive, locatable, and comparable form, so that users can identify and analyze potential risks based on a three-dimensional view before actual welding.

[0131] In practice, based on the defect type, location coordinates, and severity in the constructed welding process and defect correlation map, different forms of markers are generated at corresponding locations in the digital twin model. For example, point risks can be represented by flashing dots, spherical envelopes, or anchor point icons; risks continuously distributed along the weld can be represented by colored line segments, highlighted bands, or semi-transparent volumetric overlays; and volumetric risk areas can be displayed using isosurfaces or color-coded cloud maps. To enhance visibility, different defect types can be color-coded, such as red for incomplete fusion risk, orange for insufficient penetration risk, purple for burn-through risk, and blue for excessive stress risk. Severity can be further differentiated by color intensity, marker size, or flashing frequency. When a user clicks or hovers over a risk marker in the 3D interface, a details window pops up, displaying information such as the defect name, coordinate location, corresponding welding parameter group, trigger criteria, key predicted values, and suggested areas of concern.

[0132] To achieve accurate positioning, a mapping relationship needs to be established between the simulation mesh coordinates and the coordinates displayed in the 3D model. If the welding defect information originates from finite element nodes or elements, the corresponding surface or internal spatial location is retrieved by looking up the node number and the model topology. If the defect information originates from the analysis results along the weld section, it is mapped to the 3D centerline based on the arc length parameter of the welding path, and then the specific location is determined by combining the weld penetration direction coordinates. For complex workpieces, such as multi-point welding scenarios for battery pack busbars, the risk status of each weld point or weld segment can also be listed synchronously in the model tree, allowing users to view the risk status in the 3D scene and quickly filter high-risk locations through a list.

[0133] At the output level, this embodiment can generate a simulation report and an interactive result file containing three-dimensional risk markers. The simulation report may include basic workpiece information, input parameters, key prediction indicators, a defect list, and risk screenshots; the interactive result file can be reloaded later to continue comparing the changes in risk markers under different parameter combinations. Thus, process engineers can directly observe the risk location distribution caused by a certain set of parameters in the same digital twin scenario, and adjust the laser power, welding speed, or defocusing amount accordingly before resimulating, achieving virtual process iteration. In this way, laser welding simulation no longer remains at the level of background numerical calculation, but forms a visual verification closed loop oriented towards actual process decisions.

[0134] This step ultimately transforms the abstract defect identification results into a spatialized and perceptible risk representation, enabling process problems to be directly located within the workpiece's three-dimensional structure. Based on the above analysis, generating visual risk markers at corresponding locations in the three-dimensional space of the digital twin model can solve the problem of disconnect between welding test results, inspection conclusions, and workpiece spatial location in existing technologies. This allows process engineers to intuitively see potential defect areas caused by different parameter combinations before welding, thereby improving risk location efficiency, parameter comparison efficiency, and the targeted nature of subsequent quality control.

[0135] This application provides a laser welding simulation method based on digital twins. It introduces a multi-physics gradient feedback adaptive mesh optimization mechanism and natural language processing technology for defect identification and constructs a process defect correlation map. This overcomes the technical bottlenecks of traditional laser welding simulation, such as static modeling, insufficient accuracy, subjective defect identification, lack of correlation between process defects, and reliance on experience for optimization. By constructing a high-precision dynamic digital twin model that couples geometry, process, materials, and environment in multiple dimensions, and combining adaptive mesh iterative optimization to achieve a balance between simulation accuracy and computing power, and relying on natural language processing technology to intelligently analyze simulation data and construct a process defect correlation map, it achieves a full-process intelligent upgrade of high-precision simulation of the welding process, intelligent defect identification, risk visualization and location, and defect source analysis. Virtual verification of parameter combinations, accurate prediction of potential defects, and process cause analysis can be completed before actual welding, significantly reducing the time, material, and labor costs associated with repeated trial welding. This comprehensively improves the predictive, visualization, traceability, and intelligent optimization capabilities of laser welding processes in high-end manufacturing fields, ensuring the stability and consistency of welding quality for high-precision products such as new energy vehicle batteries and aerospace precision components.

[0136] Figure 2 This is a flowchart illustrating Embodiment 2 of the laser welding simulation method based on digital twins provided in this application. Figure 2 As shown, in Figure 1 Based on the previous embodiment, before deploying the initial welding parameters into the digital twin model and performing multiphysics coupling simulation of the welding process to obtain the welding prediction results, the method further includes:

[0137] S201. Real-time acquisition of molten pool, stress, and temperature data during the welding process.

[0138] In this step, molten pool data can be acquired through a high-speed camera, industrial vision sensor, or coaxial imaging assembly to capture the edge position of the molten pool, surface reflective features, and changes in molten pool size; stress data can be acquired through strain gauges, force sensors, displacement sensors, or load detection units arranged on the fixture to reflect the deformation trend of the workpiece during the heating and cooling stages; temperature data can be acquired through an infrared thermal imager, thermocouples, or temperature array to obtain the temperature distribution of the welding area and its neighborhood.

[0139] The aforementioned sensing unit is electrically connected to the data acquisition controller, which is used to buffer, filter, and synchronize data at different sampling frequencies. The processed data is then written into the spatial grid nodes or surface feature points corresponding to the digital twin model.

[0140] S202. Map the molten pool, stress, and temperature data into the digital twin model to generate real-time monitoring data.

[0141] During the mapping process, sensor data is converted into corresponding state variables in the model based on the current position of the welding torch, the workpiece coordinate system, and the pre-established geometric registration relationship. A timestamp alignment mechanism is then used to achieve joint updates of multi-source data at the same time. For example, the molten pool, stress, and temperature data collected by the sensors are mapped to the three-dimensional mesh nodes or surface feature points of the digital twin model using the current position of the welding torch, the workpiece coordinate system, and the pre-established geometric registration relationship.

[0142] When the collected data on the molten pool, stress, and temperature change, the corresponding areas in the digital twin model are synchronously updated with real-time monitoring data, enabling the model to reflect the dynamic state of the welding site. This real-time monitoring data can be used as a basis for boundary correction in subsequent multiphysics coupled simulations, and can also be used to compare deviations with simulation prediction results.

[0143] By employing the above method, continuous, synchronous, and spatially correlated real-time monitoring data can be obtained before entering the welding prediction calculation. This mapping process enables the digital twin model to be synchronously updated to reflect the dynamic state of the welding site, providing boundary condition support for subsequent simulation calibration and improving the consistency between the model and the physical welding process. Since the molten pool, stress, and temperature information can be uniformly presented in the model, process engineers can promptly identify abnormal heat input, local deformation, and potential defect signs, thereby improving the monitorability, traceability, and accuracy of subsequent simulation calibration of the welding process.

[0144] In another aspect of the embodiments of this application, after generating real-time monitoring, the following steps are also included:

[0145] The welding prediction results are compared with real-time monitoring data to obtain simulation deviation data. Based on the particle swarm optimization algorithm and the simulation deviation data, the initial welding parameters are iteratively optimized until the simulation deviation data returns to the preset deviation threshold range.

[0146] Among them, the simulation deviation data is used to characterize the difference between the predicted results and the actual welding state, and can reflect the degree of deviation of the molten pool profile, temperature peak, penetration depth or residual stress.

[0147] Particle swarm optimization (PSO) is used to optimize initial welding parameters. Specifically, laser power, welding speed, and defocusing amount can be set as particle position variables, and fitness values ​​can be constructed based on simulation deviation data. Furthermore, the algorithm continuously updates the particle velocity and position within the parameter space, gradually bringing the welding prediction results closer to the actual state corresponding to the real-time monitoring data. When the deviation data converges to a preset deviation threshold range, the current welding parameters are determined to meet the process consistency requirements, and this parameter combination is retained as the optimized welding parameters. In practical applications, PSO can also employ different inertia weights and learning factors; this embodiment does not limit this.

[0148] This step involves a closed-loop comparison between simulation results and real-time monitoring data, followed by iterative correction of welding parameters using a particle swarm optimization algorithm. This allows the digital twin model to continuously closely approximate the actual welding process, thereby improving parameter optimization efficiency and model prediction accuracy.

[0149] After the above steps, the deviation between the initial welding parameters and the actual working conditions can be reduced, the number of repeated trial welds can be reduced, and the stability of the welding process and the consistency of weld quality can be improved. At the same time, it is beneficial to complete the parameter correction in time before the deviation exceeds the limit, and enhance the controllability of the welding process.

[0150] Figure 3 This is a flowchart illustrating Embodiment 3 of the laser welding simulation method based on digital twins provided in this application. Figure 3 As shown, in this embodiment... Figure 1 and Figure 2 Based on the implementation examples, the method iteratively optimizes the initial welding parameters using a particle swarm optimization algorithm and simulation deviation data until the simulation deviation data returns to a preset deviation threshold range. The method includes:

[0151] S301. Based on the welding task, the initial welding parameters, real-time monitoring data, and simulation deviation data are stored in a structured manner to generate a task ID.

[0152] The task ID is used to uniquely identify a single welding task, and the structured storage is used to collect initial welding parameters, real-time monitoring data and simulation deviation data into the database according to the same data model, so that data from different sources can form an associated index around the same welding task.

[0153] In this step, the initial welding parameters include at least laser power, welding speed, and defocusing amount. Real-time monitoring data includes molten pool temperature, stress response, and weld formation state. Simulation deviation data is used to characterize the difference between the predicted results and the monitoring results.

[0154] In the specific implementation process, the structured storage can be established as a collaborative method between relational database and time-series storage. Fields can be created according to workpiece number, batch number, equipment number, process version and timestamp, and the deviation data corresponding to each parameter adjustment can be written into the same task record to form a complete welding archive.

[0155] S302. Based on the task ID, determine the cause of welding defects through timeline panoramic tracing and algorithmic attribution analysis.

[0156] Among them, the timeline panoramic traceability is used to replay the parameter settings, real-time monitoring changes, deviation generation and parameter iteration process according to the welding occurrence sequence, while the algorithm attribution analysis is used to combine deviation curves, process parameter trajectories and monitoring features to identify the key factors that lead to defects.

[0157] Algorithm-based attribution analysis can be implemented based on a preset process knowledge base, rule matching, and correlation calculation. By comparing the correspondence between the time of deviation exceeding the limit and the time of abnormality in real-time monitoring, welding defects caused by power fluctuations, speed mismatch, defocusing, insufficient material heat input, or abnormal sensor sampling can be identified. If a certain model of database server or edge computing terminal is used, its high-concurrency writing and fast retrieval capabilities are beneficial to improving traceability efficiency. In practical applications, other models of this component can also be selected, and this application embodiment does not limit this.

[0158] On the other hand, during data tracing based on task IDs, infrared thermal imagery, high-speed camera video, force sensor waveforms, and structured parameters (such as laser power and welding speed) can be uniformly mapped to welding task IDs and stored as structured data in a database. During defect feature extraction, convolutional neural networks (CNNs) can be used to detect defects in weld images, marking abnormal areas such as porosity and lack of fusion, and associating the feature data with the task ID. In attribution analysis, Bayesian networks or graph neural networks (GNNs) can be used to analyze the causal relationship between data anomalies (such as power fluctuations) and defect types (such as lack of fusion) through time-axis correlation analysis, outputting an attribution report of the defect root cause.

[0159] This application retains all task-level data based on the optimized welding parameters, enabling subsequent cross-time and cross-data source tracing by task ID. It also allows for the correlation analysis between defect phenomena and process changes, thereby pinpointing the causes of defects. By uniformly recording the optimization process, real-time monitoring process, and deviation feedback process, the accuracy and traceability of welding defect cause identification can be improved, providing a basis for subsequent process correction and quality closed-loop control.

[0160] Figure 4 A schematic diagram of the structure of the laser welding simulation device based on digital twin provided in this application. Figure 4 As shown, the laser welding simulation device 40 based on digital twins provided in this embodiment is used to verify process parameters and predict defects before trial welding, including:

[0161] The model building module 401 is used to build a digital twin model based on an adaptive mesh optimization mechanism. The digital twin model is generated based on the three-dimensional model of the welded workpiece and welding data. The welding data includes laser welding parameters, material properties and environmental parameters. The adaptive mesh optimization mechanism uses the multiphysics solution results as feedback to adaptively and dynamically adjust the mesh.

[0162] The parameter receiving module 402 is used to receive the initial welding parameters input by the user. The initial welding parameters include at least the laser power, welding speed and defocusing amount.

[0163] The multiphysics coupling simulation module 403 is used to deploy the initial welding parameters into the digital twin model, perform multiphysics coupling simulation on the welding process, and obtain welding prediction results. The welding prediction results include at least the molten pool morphology, temperature distribution, stress distribution, and weld formation effect.

[0164] The welding defect determination module 404 is used to perform defect identification processing on the welding prediction results based on the preset defect quantification judgment standard and natural language processing technology, and to construct a correlation map between welding process and defects.

[0165] The defect marking module 405 is used to generate visual risk markers at corresponding positions in the three-dimensional space of the digital twin model based on the constructed correlation map between welding process and defects, so that users can adjust the initial welding parameters based on the visual risk markers to complete the laser welding simulation.

[0166] Furthermore, the model building module 401 is specifically used for:

[0167] Based on geometric replication and mesh generation, an initial digital twin model is generated from the 3D model of the welded workpiece.

[0168] Welding data is input into the initial digital twin model to obtain an updated digital twin model;

[0169] By using multiphysics coupling modeling, the temperature gradient, molten pool velocity gradient, and stress gradient are solved simultaneously to obtain gradient feedback data;

[0170] A digital twin model is constructed based on an adaptive grid optimization mechanism and gradient feedback data.

[0171] Furthermore, the model building module 401 is specifically used for:

[0172] Based on the adaptive grid optimization mechanism, a preset gradient threshold is determined;

[0173] The gradient feedback data is compared with the preset gradient threshold to obtain the comparison result;

[0174] If any of the temperature gradient, melt pool velocity gradient, and stress gradient represented by the comparison results does not meet the preset gradient threshold, then the mesh will be updated.

[0175] The grid density is dynamically adjusted based on the updated grid division to construct a digital twin model.

[0176] Furthermore, the model building module 401 is specifically used for:

[0177] In the offline state before welding begins, the digital twin model is calibrated based on historical test welding data and error compensation algorithm to obtain a digital twin model that meets the preset simulation welding requirements.

[0178] During the welding process, welding data is collected in real time based on multiple sensors;

[0179] Based on the timestamp synchronization algorithm, welding data is timestamped using a unified clock reference, and communication link delay is compensated to obtain updated welding data;

[0180] The updated welding data is mapped to the digital twin model in real time, completing the real-time calibration of the digital twin model.

[0181] Furthermore, the multiphysics coupling simulation module 403 is also specifically used for:

[0182] The multiphysics solver is determined, and the multiphysics solver includes at least a thermal solver, a fluid solver, and a structural solver;

[0183] The initial welding parameters are deployed into the digital twin model, and the multiphysics solver is called to perform multiphysics coupling simulation to obtain the welding prediction results;

[0184] The multiphysics coupling simulation includes: the temperature field output by the thermal solver serves as the basis for correcting the physical property parameters of the fluid solver; the velocity field of the molten pool obtained by the fluid solver is fed back to the structural solver to update the local load distribution; and the deformation results calculated by the structural solver correct the position of the subsequent heat source, forming a closed-loop coupling relationship.

[0185] Furthermore, the welding defect determination module 405 is also specifically used for:

[0186] This tool is used to identify defects in welding prediction results based on preset defect quantification criteria and natural language processing technology, and to construct a correlation map between welding processes and defects.

[0187] Furthermore, the multiphysics coupling simulation module 403 is also specifically used for:

[0188] Real-time data collection of molten pool, stress, and temperature during welding;

[0189] Based on the current position of the welding torch, the workpiece coordinate system, and the pre-established geometric registration relationship, the molten pool, stress, and temperature data are mapped to the digital twin model to generate real-time monitoring data.

[0190] The laser welding simulation device based on digital twin provided in this embodiment can execute the method provided in the above method embodiment. Its implementation principle and technical effect are similar, and will not be described in detail here.

[0191] Figure 5 A schematic diagram of the structure of the laser welding simulation equipment based on digital twins provided in this application. Figure 5As shown, the laser welding simulation device 50 based on digital twins provided in this embodiment includes at least one processor 501 and a memory 502. Optionally, the device 50 also includes a communication component 503. The processor 501, memory 502, and communication component 503 are connected via a bus 504.

[0192] In a specific implementation, at least one processor 501 executes computer execution instructions stored in memory 502, causing at least one processor 501 to perform the above-described method.

[0193] The specific implementation process of processor 501 can be found in the above method embodiments, and its implementation principle and technical effect are similar. It will not be repeated here.

[0194] In the above embodiments, it should be understood that the processor can be a Central Processing Unit (CPU), or other general-purpose processors, digital signal processors (DSPs), application-specific integrated circuits (ASICs), etc. The general-purpose processor can be a microprocessor or any conventional processor. The steps of the method disclosed in this invention can be directly implemented by a hardware processor, or implemented by a combination of hardware and software modules within the processor.

[0195] The memory may include random access memory (RAM) and may also include non-volatile memory (NVM), such as at least one disk storage device.

[0196] The bus can be an Industry Standard Architecture (ISA) bus, a Peripheral Component Interconnect (PCI) bus, or an Extended Industry Standard Architecture (EISA) bus, etc. Buses can be categorized as address buses, data buses, control buses, etc. For ease of illustration, the buses shown in the accompanying drawings are not limited to a single bus or a single type of bus.

[0197] This application also provides a computer program product, including a computer program that, when executed by a processor, implements the above-described method.

[0198] This application also provides a computer-readable storage medium storing computer-executable instructions, which, when executed by a processor, implement the above-described method.

[0199] The aforementioned readable storage medium can be implemented by any type of volatile or non-volatile storage device or a combination thereof, such as static random access memory (SRAM), electrically erasable programmable read-only memory (EEPROM), erasable programmable read-only memory (EPROM), programmable read-only memory (PROM), read-only memory (ROM), magnetic storage, flash memory, magnetic disk, or optical disk. The readable storage medium can be any available medium accessible to a general-purpose or special-purpose computer.

[0200] An exemplary readable storage medium is coupled to a processor, enabling the processor to read information from and write information to the readable storage medium. Of course, the readable storage medium can also be a component of the processor. The processor and the readable storage medium can reside in an Application Specific Integrated Circuit (ASIC). Alternatively, the processor and the readable storage medium can exist as discrete components in the device.

[0201] The division of units is merely a logical functional division; in actual implementation, there may be other division methods. For example, multiple units or components may be combined or integrated into another system, or some features may be ignored or not executed. Furthermore, the coupling or direct coupling or communication connection shown or discussed may be indirect coupling or communication connection through some interfaces, devices, or units, and may be electrical, mechanical, or other forms.

[0202] The units described as separate components may or may not be physically separate. The components shown as units may or may not be physical units; that is, they may be located in one place or distributed across multiple network units. Some or all of the units can be selected to achieve the purpose of this embodiment according to actual needs.

[0203] In addition, the functional units in the various embodiments of the present invention can be integrated into one processing unit, or each unit can exist physically separately, or two or more units can be integrated into one unit.

[0204] If a function is implemented as a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of this invention, or the part that contributes to the prior art, or a part of the technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute all or part of the steps of the methods of the various embodiments of this invention. The aforementioned storage medium includes various media capable of storing program code, such as USB flash drives, portable hard drives, read-only memory (ROM), random access memory (RAM), magnetic disks, or optical disks.

[0205] Those skilled in the art will understand that all or part of the steps of the above-described method embodiments can be implemented by hardware related to program instructions. The aforementioned program can be stored in a computer-readable storage medium. When executed, the program performs the steps of the above-described method embodiments; and the aforementioned storage medium includes various media capable of storing program code, such as ROM, RAM, magnetic disks, or optical disks.

[0206] Finally, it should be noted that other embodiments of the invention will readily occur to those skilled in the art upon consideration of the specification and practice of the invention disclosed herein. This invention is intended to cover any variations, uses, or adaptations of the invention that follow the general principles of the invention and include common knowledge or customary techniques in the art not disclosed herein, and is not limited to the precise structures described above and shown in the accompanying drawings, and various modifications and changes can be made without departing from its scope. The scope of the invention is limited only by the appended claims.

Claims

1. A laser welding simulation method based on digital twins, characterized in that, Used for verifying process parameters and predicting defects before trial welding, including: A digital twin model is constructed based on an adaptive mesh optimization mechanism. The digital twin model is generated based on a three-dimensional model of the welded workpiece and welding data. The welding data includes laser welding parameters, material properties, and environmental parameters. The adaptive mesh optimization mechanism uses the multiphysics solution results as feedback to adaptively and dynamically adjust the mesh. Receive initial welding parameters input by the user, which include at least laser power, welding speed, and defocusing amount; The initial welding parameters are deployed into the digital twin model, and the welding process is simulated using multiphysics coupling to obtain welding prediction results. The welding prediction results include at least the molten pool morphology, temperature distribution, stress distribution, and weld formation effect. Based on the preset defect quantification judgment standard and natural language processing technology, the welding prediction results are processed for defect identification to construct a correlation map between welding process and defects. Based on the correlation map between the welding process and defects, a visual risk marker is generated at the corresponding position in the three-dimensional space of the digital twin model, so that the user can adjust the initial welding parameters based on the visual risk marker to complete the laser welding simulation.

2. The laser welding simulation method according to claim 1, characterized in that, A digital twin model is constructed based on an adaptive grid optimization mechanism, including: Based on geometric replication and mesh generation, an initial digital twin model is generated from the 3D model of the welded workpiece. Welding data is input into the initial digital twin model to obtain an updated digital twin model; temperature gradient, molten pool velocity gradient and stress gradient are solved in real time through multiphysics coupling modeling to obtain gradient feedback data; A digital twin model is constructed based on the adaptive grid optimization mechanism and the gradient feedback data.

3. The laser welding simulation method according to claim 2, characterized in that, Based on the adaptive grid optimization mechanism and the gradient feedback data, a digital twin model is constructed, including: Based on the adaptive grid optimization mechanism, a preset gradient threshold is determined; The gradient feedback data is compared with the preset gradient threshold to obtain the comparison result; If the comparison result indicates that any of the temperature gradient, the melt pool velocity gradient, and the stress gradient does not meet the preset gradient threshold, then mesh update and partitioning are performed. The grid density is dynamically adjusted based on the updated grid division to construct a digital twin model.

4. The laser welding simulation method according to any one of claims 1 to 3, characterized in that, The method for constructing a digital twin model further includes: In the offline state before welding begins, the digital twin model is calibrated based on historical test welding data and error compensation algorithm to obtain a digital twin model that meets the preset simulation welding requirements. During the welding process, welding data is collected in real time based on multiple sensors; According to the timestamp synchronization algorithm, the welding data is timestamped using a unified clock reference, and the communication link delay is compensated to obtain updated welding data; The updated welding data is mapped to the digital twin model in real time, thereby completing the real-time calibration of the digital twin model.

5. The laser welding simulation method according to any one of claims 1 to 3, characterized in that, The initial welding parameters are deployed into the digital twin model, and a multiphysics coupled simulation of the welding process is performed to obtain welding prediction results, including: A multiphysics solver is determined, which includes at least a thermal solver, a fluid solver, and a structural solver; The initial welding parameters are deployed into the digital twin model, and the multiphysics solver is called to perform multiphysics coupling simulation to obtain the welding prediction results. The multiphysics coupling simulation includes: the temperature field output by the thermal solver is used as the basis for correcting the physical property parameters of the fluid solver; the velocity field of the molten pool obtained by the fluid solver is fed back to the structural solver to update the local load distribution; the deformation results calculated by the structural solver correct the position of the subsequent heat source, forming a closed-loop coupling relationship.

6. The laser welding simulation method according to any one of claims 1 to 3, characterized in that, Based on preset defect quantification judgment criteria and natural language processing technology, the welding prediction results are processed for defect identification to construct a correlation map between welding processes and defects, including: Based on natural language processing technology, the welding prediction results are processed for defect identification to determine multidimensional defect features, which include at least penetration depth features, stress features and temperature features. Based on the multidimensional defect characteristics and the preset defect quantification judgment criteria, a correlation map between welding process and defects is constructed. The correlation diagram between welding process and defects represents the relationship between welding parameters, operating conditions and defect types.

7. The laser welding simulation method according to any one of claims 1 to 3, characterized in that, Before deploying the initial welding parameters into the digital twin model and performing multiphysics coupling simulation of the welding process to obtain the welding prediction results, the method further includes: Real-time data collection of molten pool, stress, and temperature during welding; Based on the current position of the welding torch, the workpiece coordinate system, and the pre-established geometric registration relationship, the data on the molten pool, stress, and temperature are mapped to the digital twin model to generate real-time monitoring data.

8. A laser welding simulation device based on digital twins, characterized in that, Used for verifying process parameters and predicting defects before trial welding, including: The model building module is used to build a digital twin model based on an adaptive mesh optimization mechanism. The digital twin model is generated based on a 3D model of the welded workpiece and welding data. The welding data includes laser welding parameters, material properties and environmental parameters. The adaptive mesh optimization mechanism uses the multiphysics solution results as feedback to adaptively and dynamically adjust the mesh. The parameter receiving module is used to receive the initial welding parameters input by the user, which include at least the laser power, welding speed and defocusing amount; The multiphysics coupling simulation module is used to deploy the initial welding parameters into the digital twin model, perform multiphysics coupling simulation on the welding process, and obtain welding prediction results. The welding prediction results include at least the molten pool morphology, temperature distribution, stress distribution, and weld formation effect. The welding defect determination module is used to perform defect identification processing on the welding prediction results based on preset defect quantification judgment standards and natural language processing technology, and to construct a correlation map between welding process and defects. The defect marking module is used to generate a visual risk mark at the corresponding position in the three-dimensional space of the digital twin model based on the correlation map between the welding process and the defect, so that the user can adjust the initial welding parameters based on the visual risk mark to complete the laser welding simulation.

9. A laser welding simulation device based on digital twins, characterized in that, include: Memory, processor; The memory stores computer-executed instructions; The processor executes computer execution instructions stored in the memory, causing the processor to perform the method as described in any one of claims 1-7.

10. A computer program product, characterized in that, Includes a computer program that, when executed by a processor, implements the method described in any one of claims 1-7.

Images