Digital-twin-based laser processing parameter self-generating system and method

Abstract

The application discloses a laser processing parameter autonomous generation system and method based on digital twinning, relates to the field of digital twinning, and comprises the following steps: collecting and preprocessing processing data in real time through a multi-source sensor; using natural language processing to analyze task instructions, extract constraint conditions and form structured requirements; generating a processing parameter candidate set through a Transformer model combined with historical data migration learning, and performing virtual processing with the aid of multi-physics field simulation; dynamically optimizing parameter weights by using an improved non-dominated sorting genetic algorithm, and obtaining a globally optimal parameter combination through iterative verification of a digital twin; after comparison between the virtual processing verification and the task requirements, adaptively correcting the parameters, and continuously optimizing the model parameters according to the deviation between the physical and simulation data after actual processing. The application has the advantages that the digital twin is taken as the core, multi-source real-time data, AI algorithms and multi-physics field simulation are fused, and the autonomous generation, multi-objective optimization and virtual-real closed-loop iteration of laser processing parameters are realized.

Description

TECHNICAL FIELD

[0001] The present application relates to the field of digital twinning, in particular to a laser processing parameter autonomous generation system and method based on digital twinning. BACKGROUND

[0002] At present, with the rapid development of advanced laser technologies such as high-power and ultrafast pulse, and the urgent demand for high-quality processing of complex components in the fields of aerospace and precision electronics, the traditional trial-and-error parameter debugging method relying on the experience of master craftsmen has been difficult to meet the production requirements of high efficiency, high consistency and intelligentization. Under this background, laser processing parameter autonomous generation technology emerged as the times require and became a research hotspot.

[0003] The core disadvantages of the laser processing parameter generation methods on the market at present are concentrated in low degree of autonomy, lack of virtual-real collaboration and single optimization dimension. Most methods still rely on manual experience or a large number of physical trial-and-error to set parameters, which has a long debugging cycle and high cost when facing new materials and complex working conditions, and is affected by the experience differences of operators, resulting in poor processing quality stability. Some methods that use model prediction or simulation are either limited to single physical field simulation, ignoring the complex influence of multi-field coupling, or rely on idealized assumptions and simplification, making it difficult to accurately reflect the actual processing dynamics. Even if AI algorithms are introduced, they often lack deep integration with real-time data, have weak generalization ability and lack a closed-loop iteration mechanism. In addition, existing methods mostly focus on a single optimization target, making it difficult to balance the multi-objective conflicts of processing precision, efficiency, energy consumption and defect control, and lack intelligent analysis capability for task instructions, requiring manual conversion of constraint conditions, which has insufficient adaptability when facing small-batch and multi-variety production, and cannot realize the full-process intelligentization and dynamic optimization of parameter generation. SUMMARY

[0004] In order to improve the existing system and method, a laser processing parameter autonomous generation system and method based on digital twinning is provided. This method takes a full-element digital twin as the core, integrates multi-source real-time data, AI algorithms and multi-physics field simulation, realizes the autonomous generation, multi-objective optimization and virtual-real closed-loop iteration of laser processing parameters, reduces the dependence on manual work and trial-and-error costs, and ensures processing accuracy and stability.

[0005] To achieve the above purposes, the technical solution adopted by the present application is as follows:

[0006] The laser processing parameter autonomous generation method based on digital twinning comprises:

[0007] A full-element digital twin of the laser processing system is constructed, which includes a physical entity mapping layer, a multi-physics field coupling simulation layer, a data-driven decision layer and a twin data platform.

[0008] Real-time acquisition of multi-dimensional data in the processing process through multi-source sensors deployed on the laser processing equipment, real-time preprocessing, transmission of preprocessed data to the twin data center, alignment of digital twin and laser processing system based on timestamp;

[0009] Obtaining processing task instructions, analyzing key information in the task instructions through natural language processing algorithm, extracting laser processing constraint conditions, forming structured task requirements and constraint condition set;

[0010] Input task requirements and constraint conditions into data-driven decision layer, generate laser processing parameter candidate set through Transformer parameter prediction model based on historical processing data in twin data center, obtain simulation result data through virtual processing simulation of multi-physical field coupling simulation layer;

[0011] Optimize and solve laser processing parameter candidate set and simulation result based on improved non-dominated sorting genetic algorithm, iteratively verify through real-time calling of multi-physical field coupling simulation model of digital twin, dynamically adjust weight coefficients of each optimization objective, and obtain global optimal laser processing parameter combination;

[0012] Optimal laser processing parameter combination is verified through multi-physical field coupling simulation layer for full-process virtual processing, virtual processing workpiece data is obtained, parameter adaptive correction is performed through comparison of virtual processing workpiece data and task requirement data, until virtual processing result meets all task requirements and constraint conditions;

[0013] After the completion of the processing task, the physical processing workpiece data is collected, and based on the deviation analysis result of the physical processing workpiece data and the virtual processing workpiece data, the model parameters of the multi-physical field coupling simulation layer are adjusted to realize continuous model iteration optimization.

[0014] Preferably, the full-element digital twin of the laser processing system is constructed, and the digital twin includes a physical entity mapping layer, a multi-physical field coupling simulation layer, a data-driven decision layer, and a twin data center, specifically comprising:

[0015] The physical entity mapping layer is based on three-dimensional geometric models of laser processing equipment, workpieces, and tooling fixtures, and fuses material thermal physical properties, equipment kinematic parameters, and tooling constraint conditions to construct geometric twin models;

[0016] The multi-physical field coupling simulation layer realizes precise simulation of temperature field, stress field, and molten pool shape in the processing process through finite element method and smoothed particle hydrodynamics hybrid algorithm;

[0017] The data-driven decision layer integrates a deep learning framework;

[0018] The twin data platform is used for storing historical processing data, real-time sensing data, simulation data and optimization result data.

[0019] Preferably, the multi-source sensor deployed on the laser processing equipment collects multi-dimensional data in the processing process in real time, and performs real-time preprocessing. The preprocessed data is transmitted to the twin data platform, and the digital twin and the laser processing system are aligned based on the timestamp, which specifically includes:

[0020] An infrared temperature measurement sensor, a high-speed vision camera, a vibration sensor and a power sensor are installed on the laser processing equipment to collect laser power, focal point position of the light spot, workpiece temperature, equipment vibration, molten pool image and material removal rate data in real time.

[0021] The collected raw data is subjected to outlier rejection, data noise reduction and format conversion to obtain standardized data.

[0022] The preprocessed standardized data is transmitted to the twin data platform, and the transmission data is matched with the real-time state of the digital twin based on the timestamp alignment mechanism.

[0023] Preferably, the processing task instruction is obtained, the key information in the task instruction is analyzed by a natural language processing algorithm, the laser processing constraint conditions are extracted, and a structured task demand and constraint condition set is formed, which specifically includes:

[0024] The processing task instruction is obtained, the processing task instruction is subjected to semantic analysis by a natural language processing algorithm, the core elements related to processing are extracted, and the unstructured instruction is converted into identifiable key data points.

[0025] The material processing process knowledge base stored in the twin data platform is combined to match the process characteristics corresponding to the current task material, and the constraint conditions are determined.

[0026] The key data points and the constraint conditions are integrated to form a structured task demand and constraint condition set.

[0027] Preferably, the task demand and constraint condition input data drive the decision layer, a parameter prediction model of Transformer is used for transfer learning based on the historical processing data in the twin data platform to generate a laser processing parameter candidate set, and a simulation result data is obtained through virtual processing simulation of the multi-physical field coupling simulation layer.

[0028] The obtained structured task demand and constraint condition are transmitted to the data-driven decision layer of the digital twin, a parameter prediction model based on Transformer is used for transfer learning based on the historical processing data in the twin data platform.

[0029] A candidate set of laser processing parameters satisfying basic constraints is initially generated, including laser power, scanning speed, spot diameter, defocusing amount, pulse frequency, pulse width, auxiliary gas type and flow rate;

[0030] Each candidate parameter set is input into a multi-physics field coupling simulation layer for virtual processing simulation to obtain simulation result data.

[0031] Preferably, the improved non-dominated sorting genetic algorithm optimizes and solves the candidate set of laser processing parameters and the simulation results, iteratively verifies the multi-physics field coupling simulation model in real time through the digital twin, dynamically adjusts the weight coefficients of each optimization target, and obtains the globally optimal laser processing parameter combination, which specifically includes:

[0032] With the highest processing precision, the fastest processing efficiency, the lowest energy consumption, and the smallest defect risk as the core optimization targets, a multi-objective optimization function is constructed in combination with the simulation result data;

[0033] The generated parameter candidate set and corresponding simulation results are input into the algorithm through the improved non-dominated sorting genetic algorithm, and the parameter combination meeting the multi-objective requirements is selected;

[0034] The multi-physics field coupling simulation model is called in real time through the digital twin to iteratively verify the virtual processing of the parameter combination selected by the algorithm, and to determine whether each parameter combination meets the basic constraint conditions;

[0035] The weight of each target is assigned based on the priority of the optimization target in the task requirement, and the entropy weight method is used to calculate the weight when there is no clear priority;

[0036] The parameter combination that does not meet the constraint conditions in the iterative verification is removed, and the globally optimal laser processing parameter combination is obtained.

[0037] Preferably, the optimal laser processing parameter combination is verified through the multi-physics field coupling simulation layer for full-process virtual processing, virtual processing workpiece data is obtained, parameter adaptive correction is performed by comparing the virtual processing workpiece data with the task requirement data, and the virtual processing result meets all task requirements and constraint conditions, which specifically includes:

[0038] The selected globally optimal laser processing parameter combination is imported into the multi-physics field coupling simulation layer of the digital twin to simulate the complete processing process, and the workpiece data after virtual processing is collected;

[0039] The virtual processing result is compared with the task requirement data through a deviation analysis algorithm to determine whether the deviation value exceeds a preset threshold, and if the deviation exceeds the threshold, the parameters are fine-tuned using the gradient descent algorithm according to the defect position and type shown in the simulation;

[0040] The revised parameters are re-input into the simulation layer for virtual machining, and the deviation analysis and correction process is repeated until the virtual machining result meets all task requirements and constraint conditions.

[0041] Preferably, the physical machining workpiece data is collected after the machining task is completed, and the model parameters of the multi-physical field coupling simulation layer are adjusted based on the deviation analysis result of the physical machining workpiece data and the virtual machining workpiece data, so as to realize continuous model iteration optimization, which specifically includes:

[0042] The optimal laser machining parameter combination that passes the virtual verification is transmitted to the laser machining equipment control system, and machining data is collected in real time during actual machining of the equipment;

[0043] The real-time collected physical machining data is compared with the simulation data, the deviation degree is quantified by constructing a data deviation model, and a deviation value is obtained;

[0044] Based on the stability of the machining process, a dynamic threshold is calculated in real time to determine whether the deviation value exceeds the threshold, and if it exceeds, an adjustment mechanism is triggered;

[0045] Based on the deviation model, the causes of the deviation are analyzed, and the laser power and scanning speed machining parameters are corrected accordingly. Specifically, if the power attenuation is caused by laser aging, the laser power parameter is fine-tuned according to the attenuation ratio, and if the temperature deviation is caused by the difference in material thermal conductivity, the scanning speed is adjusted.

[0046] Further, a laser machining parameter self-generation system based on digital twinning is proposed, which includes:

[0047] Digital twin construction module: integrates geometric, physical and behavioral multi-dimensional models to construct a virtual mapping of the laser machining system;

[0048] Multi-source perception and data access module: real-time collection of multi-modal data in the machining site through various sensors, and completion of preprocessing and time synchronization;

[0049] Task intelligent analysis module: automatically identifies machining task instructions using natural language processing technology, and converts them into structured requirements and constraint conditions;

[0050] Parameter preliminary screening decision module: based on the Transformer model and data-driven transfer learning, quickly generates an initial set of machining parameter candidates from historical data;

[0051] Multi-physical field simulation module: performs high-precision virtual machining process simulation to predict temperature field, stress field physical effects and machining results;

[0052] Multi-objective optimization solving module: applies an improved genetic algorithm to iteratively optimize parameters to balance the conflict between machining quality and efficiency;

[0053] Virtual-actual ratio comparison and feedback correction module: compare the difference between virtual and actual processing results, and dynamically adjust the simulation model and processing parameters according to the deviation;

[0054] Processor: the processor is used for processing the calculation process of each formula and the construction calculation process of each model.

[0055] Compared with the prior art, the advantages of the present application are:

[0056] By fusing physical entity mapping, multi-physical field coupling simulation, data-driven decision and twin data center, a closed-loop system of virtual-real deep interaction is constructed, relying on multi-source sensors to collect and synchronize processing data in real time, ensuring the high consistency of the twin and the physical system. With the help of natural language processing, intelligent analysis of task instructions can be realized, and structured constraint conditions can be generated without manual conversion. With the help of Transformer model and transfer learning, a parameter candidate set with strong adaptability can be quickly generated based on historical data, greatly reducing the dependence on human experience. The combination of multi-physical field simulation and improved non-dominated sorting genetic algorithm not only realizes the global optimization of multi-objective of processing precision, efficiency, energy consumption and defect risk, but also reduces the cost of physical trial and error through virtual simulation; and the virtual-real data comparison and continuous iteration optimization mechanism can dynamically correct the simulation model and processing parameters, continuously improve the accuracy and stability of parameter generation, and finally realize efficient, low-cost and high-quality output of the processing process. BRIEF DESCRIPTION OF DRAWINGS

[0057] Figure 1 The method proposed by the present application is shown in the figure;

[0058] Figure 2 The construction of digital twin proposed by the present application is shown in the figure;

[0059] Figure 3 The data acquisition diagram proposed by the present application is shown in the figure;

[0060] Figure 4 The formation of task demand and constraint condition set proposed by the present application is shown in the figure;

[0061] Figure 5 The acquisition of simulation result data proposed by the present application is shown in the figure;

[0062] Figure 6 The acquisition of global optimal laser processing parameter combination proposed by the present application is shown in the figure;

[0063] Figure 7 The virtual processing verification proposed by the present application is shown in the figure;

[0064] Figure 8 The model iteration optimization proposed by the present application is shown in the figure. DETAILED DESCRIPTION

[0065] The following description is used to disclose the present application to enable a person skilled in the art to implement the present application. The preferred embodiments in the following description are only as examples, and other obvious modifications can be thought of by those skilled in the art.

[0066] The laser processing parameter autonomous generation system based on digital twinning includes:

[0067] The digital twinning body construction module integrates geometric, physical, and behavior multidimensional models to construct a virtual mapping of the laser processing system.

[0068] The multi-source perception and data access module collects multi-modal data of the processing site in real time through various sensors, and completes preprocessing and time sequence synchronization.

[0069] The task intelligent analysis module automatically identifies processing task instructions using natural language processing technology and converts them into structured requirements and constraint conditions.

[0070] The parameter preliminary screening decision module generates an initial set of processing parameter candidates from historical data based on a Transformer model and data-driven transfer learning.

[0071] The multi-physical field simulation module performs high-precision virtual processing process simulation to predict temperature field, stress field physical effects, and processing results.

[0072] The multi-objective optimization solving module applies an improved genetic algorithm to iteratively optimize parameters to balance the conflict between processing quality and efficiency.

[0073] The virtual-real comparison and feedback correction module compares the differences between virtual and actual processing results and dynamically adjusts the simulation model and processing parameters according to the deviation.

[0074] The processor is used to process the calculation process of each formula and the construction and calculation process of each model.

[0075] Referring to Figure 1 The laser processing parameter autonomous generation method based on digital twinning includes:

[0076] Step one: Construct a full-factor digital twinning body of the laser processing system, which includes a physical entity mapping layer, a multi-physical field coupling simulation layer, a data-driven decision layer, and a twinning data hub.

[0077] Step two: Collect multi-dimensional data during the processing process in real time through multi-source sensors deployed on the laser processing equipment, and perform real-time preprocessing. Transmit the preprocessed data to the twinning data hub, and align the digital twinning body and the laser processing system based on the timestamp.

[0078] Step three: Obtain the processing task instruction, analyze the key information in the task instruction through natural language processing algorithm, extract the laser processing constraint conditions, and form the structured task demand and constraint condition set;

[0079] Step four: Input the task demand and constraint condition into the data-driven decision layer, generate the laser processing parameter candidate set through the parameter prediction model of the Transformer based on the historical processing data in the twin data center, and obtain the simulation result data through the virtual processing simulation of the multi-physical field coupling simulation layer;

[0080] Step five: Optimize and solve the laser processing parameter candidate set and simulation result based on the improved non-dominated sorting genetic algorithm, iteratively verify through the real-time calling of the multi-physical field coupling simulation model of the digital twin, dynamically adjust the weight coefficients of each optimization objective, and obtain the global optimal laser processing parameter combination;

[0081] Step six: Perform full-process virtual processing verification on the optimal laser processing parameter combination through the multi-physical field coupling simulation layer, obtain the virtual processing workpiece data, and perform parameter adaptive correction by comparing the virtual processing workpiece data with the task demand data until the virtual processing result meets all task demands and constraint conditions;

[0082] Step seven: After the completion of the processing task, collect the physical processing workpiece data, adjust the model parameters of the multi-physical field coupling simulation layer based on the deviation analysis result of the physical processing workpiece data and the virtual processing workpiece data, and realize continuous model iteration optimization.

[0083] Referring to Figure 2 As shown in the figure, a full-element digital twin of the laser processing system is constructed, which includes a physical entity mapping layer, a multi-physical field coupling simulation layer, a data-driven decision layer, and a twin data center, which specifically includes:

[0084] The physical entity mapping layer is based on the three-dimensional geometric model of the laser processing equipment, workpiece, and tooling fixture, and fuses the material thermal physical properties, equipment kinematics parameters, and tooling constraint conditions to construct a geometric twin model;

[0085] The multi-physical field coupling simulation layer realizes precise simulation of temperature field, stress field, and molten pool shape during the processing process through the finite element method and the smoothed particle hydrodynamics hybrid algorithm;

[0086] The data-driven decision layer integrates a deep learning framework;

[0087] The twin data center is used to store historical processing data, real-time sensing data, simulation data, and optimization result data.

[0088] Referring to Figure 3As shown, the multi-source sensor deployed on the laser processing equipment collects multi-dimensional data in real time during the processing process, and performs real-time preprocessing. The preprocessed data is transmitted to the twin data center, and the digital twin and the laser processing system are aligned based on the timestamp, which specifically includes:

[0089] An infrared temperature measurement sensor, a high-speed vision camera, a vibration sensor, and a power sensor are installed on the laser processing equipment to collect laser power, focal point position of the spot, workpiece temperature, equipment vibration, molten pool image, and material removal rate data in real time.

[0090] The collected raw data is subjected to outlier rejection, data noise reduction, and format conversion to obtain standardized data.

[0091] The preprocessed standardized data is transmitted to the twin data center, and the transmission data is matched with the real-time state of the digital twin based on the timestamp alignment mechanism.

[0092] Specifically, after receiving the data, the twin data center matches the sensor data according to the collection time and the simulation time axis of the digital twin based on the timestamp alignment mechanism. The state update of the twin is triggered synchronously, and the real-time collected power, temperature, vibration, and other data are substituted into the twin model to adjust the physical entity state parameters in the model. After synchronization is completed, the current state of the twin is automatically compared with the real-time data of the physical processing system. If the deviation is less than 1%, the synchronization is determined to be valid. If the deviation exceeds the threshold, data synchronization is immediately triggered again to ensure that the state of the twin is consistent with the physical entity.

[0093] Referring to Figure 4 As shown, the processing task instruction is obtained, the key information in the task instruction is analyzed by a natural language processing algorithm, the laser processing constraint conditions are extracted, and a structured task requirement and constraint condition set is formed. Specifically, it includes:

[0094] The processing task instruction is obtained, the processing task instruction is analyzed by a natural language processing algorithm, the core elements related to processing are extracted, and the unstructured instruction is converted into identifiable key data points;

[0095] The material processing technology knowledge base stored in the twin data center is combined to match the process characteristics corresponding to the current task material, and the constraint conditions are determined.

[0096] The key data points and the constraint conditions are integrated to form a structured task requirement and constraint condition set.

[0097] Specifically, based on natural language processing technology, information extraction is performed on the standardized task instructions. In the material information module, the material categories are distinguished and the subcategories are refined, such as analyzing the specific materials of aluminum alloy, stainless steel, titanium alloy, etc. for the metal category, and specifying the matrix and reinforcing phase components for the composite material; in the workpiece parameter module, the three-dimensional size, machining area position, and clamping requirements of the workpiece are extracted; in the quality requirement module, the dimensional tolerance, surface roughness, appearance defect limitation, and internal quality requirement are disassembled; in the efficiency target module, the time requirement or relative efficiency demand is extracted and specified.

[0098] A preset material processing knowledge base is called from the twin data center, which includes processing characteristic data of different materials, such as melting point and thermal conductivity of aluminum alloy, laser absorption rate of non-metallic materials, historical processing cases of similar workpieces, and equipment adaptation process range; the material type and workpiece structure are associated with the knowledge base data: for example, if the material is TC4 titanium alloy, the laser processing taboo and conventional processing parameter interval corresponding to TC4 titanium alloy in the knowledge base are automatically matched; if the workpiece is a thin-walled piece, the special constraints of thin-walled piece processing are associated to provide process basis for subsequent constraint extraction.

[0099] Referring to Figure 5 The task requirements and constraint conditions are input into the data-driven decision layer, and through the parameter prediction model of Transformer, the historical processing data in the twin data center is used for transfer learning to generate a candidate set of laser processing parameters. Through virtual processing simulation of the multi-physical field coupling simulation layer, simulation result data is obtained, which specifically includes:

[0100] The obtained structured task requirements and constraint conditions are transmitted to the data-driven decision layer of the digital twin, and through the parameter prediction model based on Transformer, transfer learning is performed based on the historical processing data in the twin data center;

[0101] A candidate set of laser processing parameters that meet the basic constraints is initially generated, and the parameters include laser power, scanning speed, spot diameter, defocusing amount, pulse frequency, pulse width, auxiliary gas type, and flow rate;

[0102] Each candidate parameter set is input into the multi-physical field coupling simulation layer for virtual processing simulation, and simulation result data is obtained.

[0103] Specifically, the Transformer parameter prediction model of the data-driven decision layer is started. First, based on the material type and machining precision requirement in the task, historical high-quality machining data is filtered from the twin data to form a model training data set; the verified parameter generation logic in the historical data is reused through the transfer learning algorithm, such as the power-speed matching relationship corresponding to the aluminum alloy thin-walled part, the model weight parameters are adjusted to avoid efficiency loss caused by training from zero; after training, the constraint conditions are input, and the model automatically outputs 3-5 sets of preliminary parameter candidate sets, each set containing 8 core parameters such as laser power, scanning speed, spot diameter, and defocusing amount, and the parameter values are within the basic constraint range of the equipment and materials;

[0104] The parameter candidate set after compliance is input into the multi-physics field coupling simulation layer one by one to start batch virtual machining simulation: for each set of parameters, the simulation layer simulates the whole stage of laser emission, spot focusing, and material interaction, and synchronously calculates the temperature field distribution, stress change, and molten pool dynamics during the machining process; after simulation, the quantitative results corresponding to each set of parameters are output, including machining precision, surface roughness, machining time, total energy consumption, and defect risk assessment.

[0105] Referring to Figure 6 The laser machining parameter candidate set and the simulation results are optimized and solved based on the improved non-dominated sorting genetic algorithm, the multi-physics field coupling simulation model is called in real time by the digital twin for iterative verification, the weight coefficients of each optimization objective are dynamically adjusted, and the globally optimal laser machining parameter combination is obtained, which specifically includes:

[0106] Taking the highest machining precision, the fastest machining efficiency, the lowest energy consumption, and the smallest defect risk as the core optimization objectives, a multi-objective optimization function is constructed combined with the simulation result data;

[0107] The generated parameter candidate set and the corresponding simulation results are input into the algorithm through the improved non-dominated sorting genetic algorithm, and the parameter combination that meets the multi-objective requirements is selected;

[0108] The multi-physics field coupling simulation model is called in real time by the digital twin to iteratively verify the virtual machining of the parameter combination selected by the algorithm, and to determine whether each parameter combination meets the basic constraint conditions;

[0109] Based on the priority of the optimization objectives in the task requirements, the weights of each objective are allocated; when there is no clear priority, the entropy weight method is used to calculate the weights;

[0110] The parameter combinations that do not meet the constraint conditions in the iterative verification are eliminated to obtain the globally optimal laser machining parameter combination.

[0111] Specifically, the improved non-dominated sorting genetic algorithm is started, and the parameter candidate set is iteratively optimized: in the first iteration, the algorithm takes the parameter candidate set as the initial population, calculates the comprehensive score of each group of parameters according to the weight, and selects the parameter combinations with the top 60% comprehensive scores; then the multi-physical field coupling simulation layer of the digital twin is called to re-simulate the virtual machining of the selected parameter combinations, and more accurate target function values are obtained; according to the new simulation results, the algorithm performs non-dominated sorting again, eliminates the dominated parameter combinations, and retains the non-dominated parameters for the next iteration;

[0112] During the iteration process, if a target conflict occurs, such as a parameter with high efficiency but defect risk exceeding the constraint threshold, the target weight dynamic adjustment is triggered: if the defect risk exceeds the threshold, the defect risk weight is automatically increased by 20%, while the efficiency weight is reduced by 10% and the energy consumption weight is reduced by 10%, and the comprehensive score is recalculated; if a parameter meets all constraints but the machining precision is close to the lower limit of the threshold, temporarily increase the precision weight, and perform fine-tuning simulation on the laser power and defocusing amount of the parameter to verify whether the fine-tuning can optimize the precision; after each iteration, the number of selected non-dominated parameter combinations is controlled at 2-3 groups to ensure that the subsequent optimization focuses on the core candidates.

[0113] Referring to Figure 7 As shown, the optimal laser processing parameter combination is verified through the multi-physical field coupling simulation layer for full-process virtual machining, and virtual machining workpiece data is obtained. Through comparison of the virtual machining workpiece data and the task requirement data, parameter adaptive correction is performed until the virtual machining result meets all task requirements and constraint conditions, specifically including:

[0114] The selected global optimal laser processing parameter combination is imported into the multi-physical field coupling simulation layer of the digital twin to simulate the complete machining process and collect workpiece data after virtual machining;

[0115] The virtual machining result is compared with the task requirement data through deviation analysis algorithm to determine whether the deviation value exceeds the preset threshold. If the deviation exceeds the threshold, the parameters are fine-tuned using the gradient descent algorithm based on the defect location and type displayed by the simulation;

[0116] The corrected parameters are re-input into the simulation layer for virtual machining, and the deviation analysis and correction process is repeated until the virtual machining result meets all task requirements and constraint conditions.

[0117] Specifically, the collected virtual machining data is compared and analyzed with the task requirements: the three-dimensional size comparison algorithm is adopted to check the virtual workpiece size and the required tolerance one by one, and the size deviation value is calculated; through the surface roughness analysis module, the Ra value of the virtual workpiece surface is extracted and compared with the required threshold value; for internal stress and defect data, it is judged whether there is stress concentration exceeding the safe range or defects not meeting the requirements; if all indicators meet the requirements, the virtual verification is passed; if the deviation of any indicator exceeds the threshold value, the parameter self-adaptive correction mechanism is triggered;

[0118] According to the deviation analysis result, the defect type and cause are located: if edge ablation occurs, it is determined that the laser energy is excessively concentrated, and the laser power is adjusted or the defocusing amount is increased; if the size is out of tolerance, it is determined that the material removal amount is insufficient, and the scanning speed is increased or the laser power is increased; if internal stress cracks occur, it is determined that the stress release is uneven during the cooling stage, and the cooling rate is adjusted or the laser pulse parameters are optimized; the adjustment range is dynamically determined according to the deviation size to avoid excessive adjustment leading to new defects.

[0119] Referring to Figure 8 After the machining task is completed, the physical machining workpiece data is collected, and based on the deviation analysis result of the physical machining workpiece data and the virtual machining workpiece data, the model parameters of the multi-physical field coupling simulation layer are adjusted to realize continuous model iteration optimization, which specifically includes:

[0120] The optimal laser machining parameter combination that passes the virtual verification is transmitted to the laser machining equipment control system, and in the actual machining process of the equipment, the machining data is collected in real time;

[0121] The real-time collected physical machining data is compared with the simulation data, the deviation degree is quantified by constructing a data deviation model, and the deviation value is obtained;

[0122] Based on the stability of the machining process, the dynamic threshold value is calculated in real time to judge whether the deviation value exceeds the threshold value, and if it exceeds, the adjustment mechanism is triggered;

[0123] Based on the deviation model, the deviation causes are analyzed and the laser power and scanning speed machining parameters are corrected accordingly.

[0124] Specifically, the digital twin is based on the stability of the machining process to calculate the dynamic deviation threshold value in real time: if the deviation value of a certain indicator does not exceed the threshold value, the current machining parameters are maintained; if the deviation exceeds the threshold value, the deviation reason analysis mechanism is triggered immediately, which calls the historical running data of the equipment and the material batch attribute data in the twin data center, locates the reason combined with the deviation model, and generates a reason analysis report. The dynamic deviation threshold value formula is:

[0125]

[0126] Among them, a dynamic threshold value of the kth index at the tth moment, a historical average deviation value of the kth index, a real-time standard deviation of the kth index at the tth moment;

[0127] According to the deviation cause analysis result, the digital twin automatically generates a parameter correction scheme: if the power attenuation is caused by laser aging, the laser power parameter is fine-tuned according to the attenuation ratio; if the temperature deviation is caused by the difference in material thermal conductivity, the scanning speed is appropriately adjusted; the corrected parameter is quickly verified through virtual simulation, and then is transmitted to the equipment control system in real time through the industrial bus; after receiving the corrected parameter, the equipment completes the parameter switching in the non-key section of the machining path, avoids the influence of parameter mutation on the machining quality, and records the parameter correction record to form a closed-loop adjustment account.

[0128] It should be noted that the above-mentioned sequence of the embodiments of the present application is only for description, and does not represent the advantages and disadvantages of the embodiments. The above describes a specific embodiment of the present application. In addition, the processes depicted in the drawings do not necessarily require the specific order or continuous order shown to achieve the desired results. In some embodiments, multi-task processing and parallel processing are possible or may be advantageous.

[0129] Each embodiment in the specification is described in a progressive manner, and the same and similar parts between each embodiment can be referred to each other. Each embodiment focuses on the difference from other embodiments.

[0130] The above only describes the preferred embodiments of the present application, and does not limit the present application. Any modification, equivalent replacement, improvement, etc. within the principles of the present application shall be included in the protection scope of the present application.

Claims

1. A method for self-generating laser processing parameters based on digital twinning, characterized in that, The application relates to a full-element digital twin of a laser processing system, which comprises a physical entity mapping layer, a multi-physical field coupling simulation layer, a data-driven decision layer and a twin data center. Real-time multi-dimensional data in a processing process are collected by deploying multi-source sensors on a laser processing device, and real-time preprocessing is performed, the preprocessed data are transmitted to the twin data center, and the digital twin and the laser processing system are timestamp-aligned; Task instruction data are obtained, key information in the task instruction is analyzed by using a natural language processing algorithm, laser processing constraint conditions are extracted, and a structured task requirement and constraint condition set is formed; The task requirement and constraint condition are input into the data-driven decision layer, a parameter prediction model of a Transformer is used for migration learning based on historical processing data in the twin data center, a laser processing parameter candidate set is generated, and simulation result data are obtained by virtual processing simulation of the multi-physical field coupling simulation layer; The laser processing parameter candidate set and the simulation result are optimized and solved based on an improved non-dominated sorting genetic algorithm, the multi-physical field coupling simulation model is called in real time by the digital twin for iterative verification, the weight coefficients of various optimization objectives are dynamically adjusted, and a global optimal laser processing parameter combination is obtained; The optimal laser processing parameter combination is subjected to full-process virtual processing verification by the multi-physical field coupling simulation layer, virtual processing workpiece data are obtained, parameter adaptive correction is performed by comparing the virtual processing workpiece data with the task requirement data, and the virtual processing result meets all the task requirements and constraint conditions; After the processing task is completed, physical processing workpiece data are collected, model parameters of the multi-physical field coupling simulation layer are adjusted based on deviation analysis results of the physical processing workpiece data and the virtual processing workpiece data, and continuous model iterative optimization is realized. The full-element digital twin of the laser processing system comprises a physical entity mapping layer, a multi-physical field coupling simulation layer, a data-driven decision layer and a twin data center.

2. The digital-twin-based method for autonomous generation of laser machining parameters according to claim 1, characterized in that, The multi-physical field coupling simulation layer realizes accurate simulation of a temperature field, a stress field and a molten pool shape in a processing process by using a finite element method and a smoothed particle hydrodynamics hybrid algorithm; The data-driven decision layer integrates a deep learning framework; The twin data center is used for storing historical processing data, real-time sensing data, simulation data and optimization result data. The multi-source sensors are installed on the laser processing device to collect laser power, focal point position of a light spot, workpiece temperature, equipment vibration, molten pool image and material removal rate data in real time. ​ 3.The digital-twin-based method for self-generating laser processing parameters according to claim 1, wherein, ​ ​ The collected raw data is subjected to outlier rejection, data noise reduction and format conversion to obtain standardized data; The pre-processed standardized data is transmitted to the twin data platform, and the transmission data is matched with the real-time state of the digital twin based on a timestamp alignment mechanism. 4.The digital-twin-based method for autonomous generation of laser processing parameters according to claim 1, wherein, The task instruction is obtained, the key information in the task instruction is analyzed by a natural language processing algorithm, the laser processing constraint conditions are extracted, and a structured task requirement and constraint condition set is formed, specifically including: The task instruction is obtained, the task instruction is analyzed by a natural language processing algorithm, the core elements related to the task are extracted, and the unstructured instruction is converted into identifiable key data points; The material processing knowledge base stored in the twin data platform is combined to match the process characteristics corresponding to the current task material, and the constraint conditions are determined; The key data points and constraint conditions are integrated to form a structured task requirement and constraint condition set. 5.The digital-twin-based autonomous generation of laser processing parameters method according to claim 1, characterized in that, The task requirement and constraint condition are input into the data-driven decision layer, a parameter prediction model based on Transformer is used for transfer learning based on historical processing data in the twin data platform, a laser processing parameter candidate set is generated, and a virtual processing simulation is performed through a multi-physical field coupling simulation layer to obtain simulation result data, specifically including: The structured task requirement and constraint condition are transmitted to the data-driven decision layer of the digital twin, and a parameter prediction model based on Transformer is used for transfer learning based on historical processing data in the twin data platform; A laser processing parameter candidate set that meets the basic constraints is generated, and the parameters include laser power, scanning speed, spot diameter, defocusing amount, pulse frequency, pulse width, auxiliary gas type and flow rate; Each candidate parameter set is input into the multi-physical field coupling simulation layer for virtual processing simulation to obtain simulation result data.

6. The digital-twin-based autonomous generation method of laser machining parameters according to claim 1, characterized in that, The laser processing parameter candidate set and simulation result are optimized and solved based on an improved non-dominated sorting genetic algorithm, the multi-physical field coupling simulation model is called by the digital twin for iterative verification, the weight coefficients of each optimization target are dynamically adjusted, and a globally optimal laser processing parameter combination is obtained, specifically including: The highest processing precision, the fastest processing efficiency, the lowest energy consumption and the smallest defect risk are taken as the core optimization targets, and a multi-objective optimization function is constructed based on the simulation result data; The generated parameter candidate set and corresponding simulation result are input into the algorithm through an improved non-dominated sorting genetic algorithm, and a parameter combination that meets the multi-objective requirements is selected; The parameter combination selected by the algorithm is subjected to virtual processing iterative verification by calling the multi-physical field coupling simulation model in real time by the digital twin, and it is determined whether the parameter combination meets the basic constraint conditions; The weight of each target is allocated based on the priority of the optimization target in the task requirement, and the weight is calculated by an entropy weight method when there is no clear priority; The parameter combinations that do not meet the constraint conditions in the iterative verification are rejected, and a globally optimal laser processing parameter combination is obtained.

7. The digital-twin-based autonomous generation method of laser machining parameters according to claim 1, characterized in that, The optimal laser processing parameter combination is verified through the multi-physical field coupling simulation layer for the whole process of virtual processing, virtual processing workpiece data is obtained, parameter adaptive correction is performed by comparing the virtual processing workpiece data with the task requirement data, and the virtual processing result meets all task requirements and constraint conditions, specifically including: The screened global optimal laser processing parameter combination is imported into the multi-physical field coupling simulation layer of the digital twin to simulate the complete processing process, and the workpiece data after virtual processing is collected; The virtual processing result is compared with the task requirement data through a deviation analysis algorithm to determine whether the deviation value exceeds a preset threshold value, and if the deviation exceeds the threshold value, the parameters are fine-tuned according to the defect position and type displayed by the simulation; The corrected parameters are re-input into the simulation layer for virtual processing, and the deviation analysis and correction process is repeated until the virtual processing result meets all task requirements and constraint conditions. 8.The digital-twin-based autonomous generation of laser machining parameters method of claim 1, wherein, After the processing task is completed, the physical processing workpiece data is collected, the model parameters of the multi-physical field coupling simulation layer are adjusted based on the deviation analysis result of the physical processing workpiece data and the virtual processing workpiece data, and continuous model iteration optimization is realized, specifically including: The optimal laser processing parameter combination that passes the virtual verification is transmitted to the laser processing equipment control system, and processing data is collected in real time during actual processing of the equipment; The real-time collected physical processing data is compared with the simulation data, the deviation degree is quantified by constructing a data deviation model, and the deviation value is obtained; A dynamic threshold value is calculated in real time based on the stability of the processing process to determine whether the deviation value exceeds the threshold value, and if it exceeds, an adjustment mechanism is triggered; The causes of the deviation are analyzed based on the deviation model, and the laser power and scanning speed processing parameters are corrected accordingly, specifically, if the laser power attenuation is caused by laser aging, the laser power parameter is fine-tuned according to the attenuation ratio, and if the temperature deviation is caused by the difference in material thermal conductivity, the scanning speed is adjusted.

9. A system for autonomous generation of laser processing parameters based on digital twinning for implementing the method for autonomous generation of laser processing parameters based on digital twinning according to any one of claims 1 to 8, characterized in that, It includes: Digital twin construction module: integrates geometric, physical and behavioral multi-dimensional models to construct a virtual mapping of the laser processing system; Multi-source perception and data access module: real-time collection of multi-modal data in the processing site through various sensors, and completion of preprocessing and time synchronization; Task intelligent analysis module: automatically identifies processing task instructions using natural language processing technology, and converts them into structured requirements and constraints; Parameter preliminary screening decision module: based on the Transformer model and data-driven transfer learning, quickly generates an initial set of processing parameter candidates from historical data; Multi-physical field simulation module: performs high-precision virtual processing process simulation to predict temperature field, stress field physical effects and processing results; Multi-objective optimization solving module: applies an improved genetic algorithm to iteratively optimize parameters to balance the conflict between processing quality and efficiency; Virtual-real comparison and feedback correction module: compares the differences between virtual and actual processing results, and dynamically adjusts the simulation model and processing parameters according to the deviation; Processor: the processor is used to process the calculation process of each formula and the construction and calculation process of each model.

Images