A method for simulating deep drawing of anisotropic materials

Abstract

The present invention provides a method for simulating the deep drawing of anisotropic materials, which belongs to the technical field of deep drawing simulation, including first determining the microscopic crystal structure of the anisotropic material and constructing a constitutive model of the dynamic adaptive material; importing the constitutive model into the finite element analysis software to form a finite element model of the interactive behavior of multiple physical fields; importing the finite element model into the adaptive meshing algorithm to form a finite element simulation result; using a deep learning algorithm to analyze the finite element simulation result in real time, and outputting the analysis report in the form of a visual chart; the mold fillet radius, deep drawing gap, blank holding force, deep drawing speed, friction coefficient and electromagnetic control parameters are dynamically optimized in coordination, and the lug rate of the cup-shaped part is obtained according to the deep drawing simulation. The present invention provides a method for simulating the deep drawing of anisotropic materials, which obtains the lug rate of the cup-shaped part by constructing a constitutive model and forming the finite element results of the finite element model, so as to predict the lug problem of the deep drawing of the cup-shaped part.

Description

Technical Field

[0001] The invention belongs to the technical field of drawing simulation, and more specifically, relates to a drawing simulation method for anisotropic materials. Background Art

[0002] In the complex process system of contemporary metal sheet forming, anisotropic materials have unique and complex crystal microstructure arrangements, which causes their mechanical properties to show extremely significant differences in different directions.

[0003] When this type of material is used in the key process of cup-shaped parts drawing, the deformation process of each part of the plate is complex, changeable and extremely uneven under the driving force of the drawing, which results in the regular generation of lugs on the edge of the cup. Traditional strategies for dealing with lug problems mostly rely on technicians who have accumulated long-term on-site practical experience to adjust mold parameters manually, while also testing large-scale and coarse-grained process parameters.

[0004] However, this traditional model has many drawbacks that are difficult to overcome: First, due to the lack of in-depth analysis of the intrinsic close relationship between the microstructure and macroscopic deformation behavior of the material, it is like a blind man touching an elephant when exploring the root cause of the lug formation, and the mold debugging process is full of blindness; second, the repeated mold trial process requires a huge amount of manpower, material resources and precious time costs, and the final product quality is not stable, and the scrap rate has remained high for a long time. It has been unable to adapt to the trend of the times where modern manufacturing is developing rapidly towards intelligence and precision, and has seriously hindered the pace of industrial upgrading. Summary of the invention

[0005] The purpose of the present invention is to provide a drawing simulation method for anisotropic materials, by constructing a constitutive model and forming finite element results of the corresponding finite element model, obtaining the lug rate of the cup-shaped part according to the drawing simulation, so as to predict the lug problem of the cup-shaped part during deep drawing.

[0006] To achieve the above object, the technical solution adopted by the present invention is: to provide a method for simulating the deep drawing of anisotropic materials, comprising the following steps:

[0007] S1: Determine the microscopic crystal structure of anisotropic materials, obtain the crystal orientation distribution function, dislocation density tensor and lattice distortion parameters, and construct a constitutive model of dynamic adaptive materials based on the combination of dislocation theory and crystal plastic constitutive law;

[0008] S2: Import the constitutive model into the finite element analysis software, set corresponding boundary conditions based on the temperature change, strain rate mutation, electromagnetic induction effect between the die and the sheet, and fluid dynamics behavior of the lubricating medium caused by plastic deformation during the drawing process, and form a finite element model of multi-physical field interaction behavior;

[0009] S3: Import the finite element model into the adaptive meshing algorithm, adjust the density of the mesh according to the model stress, strain gradient and physical field gradient, and form the finite element simulation results;

[0010] S4: Use deep learning algorithms to analyze finite element simulation results in real time, obtain sensitive areas of lug formation, potential triggering factors, and the relationship between multiple physical fields, and output analysis reports in the form of visual charts;

[0011] S5: The die corner radius, drawing gap, blank holder force, drawing speed, friction coefficient and electromagnetic control parameters are dynamically optimized. The drawing simulation is performed by setting a multi-objective optimization function, and the lug rate of the cup-shaped part is obtained according to the drawing simulation.

[0012] In one possible implementation, in step S1, electron backscatter diffraction technology, atomic force microscopy and transmission electron microscopy are used to work together to comprehensively analyze the crystal orientation information obtained by the electron backscatter diffraction technology, the surface information obtained by the atomic force microscope and the internal microstructure information obtained by the transmission electron microscope to determine the microscopic crystal structure of the anisotropic material.

[0013] In a possible implementation, when setting the corresponding boundary conditions in step S2,

[0014] 3.1. Set thermal boundary conditions based on the heat conduction generated by the material itself and the heat conduction between the surrounding environment and the material;

[0015] 3.2. Based on the drawing speed, die geometry and material flow characteristics, an adaptive time step algorithm is used to adjust the parameters in the constitutive model according to the material strain rate sensitivity index combined with the dislocation dynamics theory to set the strain rate boundary conditions;

[0016] 3.3. Calculate the electromagnetic force based on the electromagnetic field distribution corresponding to the electromagnetic induction effect, and load the electromagnetic force into the constitutive model as a body force to set the electromagnetic induction boundary conditions;

[0017] 3.4. Based on lubrication theory and actual lubrication method, set the pressure boundary conditions at the inlet and outlet, and set the flow rate boundary conditions in combination with the flow rate change of the lubrication medium.

[0018] In one possible implementation, in step S2, the finite element simulation software has a built-in virtual material library, which covers basic parameters of a variety of anisotropic materials. The corresponding basic parameters can be retrieved according to the material composition and organizational structure, and the finite element model can be automatically improved through a machine learning algorithm in combination with measured data.

[0019] In a possible implementation, when the finite element model is imported into the adaptive meshing algorithm in step S3,

[0020] 5.1. Convert the data of the finite element model into a format that can be recognized by ABAQUS or ANSYS software;

[0021] 5.2. Use the Python scripting interface provided by ABAQUS software or the APDL language programming environment of ANSYS software to write adaptation code;

[0022] 5.3. Specify to enable the adaptive meshing algorithm and bind the written adaptation code to it;

[0023] 5.4. Perform preliminary debugging, set the drawing speed and mold geometry, observe whether the algorithm module can correctly read the model data, obtain physical quantities in real time, and make preliminary grid adjustments based on the gradient information. Analyze and troubleshoot possible data transmission interruptions and algorithm errors through debugging log analysis, and optimize the association settings between the adaptation code and the algorithm module.

[0024] In a possible implementation, in step S4, a convolutional neural network or a long short-term memory network is selected as the basic architecture of the deep learning model, the preprocessed data set is divided into a training set, a validation set, and a test set in a ratio of 70:15:15, the training set is used to train the deep learning model, the simulated physical quantity data is used as input, the characteristics of the drawn lug are used as output, and the weight of the deep learning model is adjusted.

[0025] In one possible implementation, in step S4, stress-strain cloud maps at different times are input into the deep learning model to obtain the sensitive area of ​​lug formation; different process parameters of deep drawing are input into the deep learning model to obtain the potential triggering factors of the lug; and the correlation between the temperature field, electromagnetic intensity and fluid velocity multi-physical fields and the lug is analyzed.

[0026] In a possible implementation, in step S4, the visualization chart includes at least a stress-strain cloud diagram, a temperature field distribution diagram, an electromagnetic intensity vector diagram, and a fluid velocity vector diagram.

[0027] In one possible implementation, in step S5, a hybrid optimization strategy consisting of a genetic algorithm, a particle swarm optimization algorithm, and a simulated annealing algorithm is used to drive the coordinated dynamic optimization of the mold fillet radius, the drawing gap, the blank holding force, the drawing speed, the friction coefficient, and the electromagnetic control parameters. By setting a multi-objective optimization function, the comprehensive molding quality indicators of minimizing the lug height, maximizing the wall thickness uniformity, minimizing the residual stress, and optimizing the surface quality are obtained.

[0028] In a possible implementation, after step S5, scanning technology is used to scan the entire process of trial production cup-shaped parts from blank cutting to forming, and the collected scanning data is transmitted to the finite element model in real time to dynamically correct the finite element model.

[0029] The beneficial effect of the deep drawing simulation method of anisotropic materials provided by the present invention is that: compared with the prior art, the microscopic crystal structure of the anisotropic material is first determined, the crystal orientation distribution function, the dislocation density tensor and the lattice distortion parameter are obtained, and a constitutive model of a dynamic adaptive material based on the combination of dislocation theory and crystal plastic constitutive law is constructed; the constitutive model is imported into the finite element analysis software, and corresponding boundary conditions are set based on the temperature change, strain rate mutation, electromagnetic induction effect between the die and the plate, and the fluid dynamics behavior of the lubricating medium caused by plastic deformation during the deep drawing process, so as to form a multi-physical field interactive behavior. Finite element model; import the finite element model into the adaptive meshing algorithm, adjust the density of the mesh according to the model stress, strain gradient and physical field gradient, and form the finite element simulation results; use the deep learning algorithm to analyze the finite element simulation results in real time, obtain the sensitive areas of lug formation, potential triggering factors and the relationship between multiple physical fields, and output the analysis report in the form of visual charts; the mold fillet radius, drawing gap, blank holding force, drawing speed, friction coefficient and electromagnetic control parameters are dynamically optimized, and the drawing simulation is performed by setting a multi-objective optimization function, and the lug rate of the cup-shaped part is obtained according to the drawing simulation. A drawing simulation method for anisotropic materials provided by the present invention constructs a constitutive model and forms the finite element results of the corresponding finite element model, and obtains the lug rate of the cup-shaped part according to the drawing simulation to predict the lug problem of the cup-shaped part. BRIEF DESCRIPTION OF THE DRAWINGS

[0030] In order to more clearly illustrate the technical solutions in the embodiments of the present invention, the drawings required for use in the embodiments or the description of the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying creative work.

[0031] Figure 1 A flow chart of a method for simulating deep drawing of anisotropic materials provided in an embodiment of the present invention. DETAILED DESCRIPTION

[0032] In order to make the technical problems, technical solutions and beneficial effects to be solved by the present invention more clearly understood, the present invention is further described in detail below in conjunction with the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are only used to explain the present invention and are not intended to limit the present invention.

[0033] Unless explicitly defined otherwise, the use of terms such as "first," "second," or "third," etc., are intended to distinguish different objects rather than to describe a specific order.

[0034] Unless otherwise expressly defined, directional words such as the terms "center", "lateral", "longitudinal", "horizontal", "vertical", "top", "bottom", "inside", "outside", "up", "down", "front", "back", "left", "right", "clockwise", "counterclockwise", "high", "low", etc., indicating directions or positional relationships are based on the directions and positional relationships shown in the drawings, and are only for the convenience of describing the present invention and simplifying the description, and do not indicate or imply that the device or element referred to must have a specific direction or be constructed and operated in a specific direction, so they cannot be understood as limiting the specific scope of protection of the present invention.

[0035] See also Figure 1 Now, a drawing simulation method for anisotropic materials provided by the present invention is described. The drawing simulation method for anisotropic materials comprises the following steps:

[0036] S1: Determine the microscopic crystal structure of anisotropic materials, obtain the crystal orientation distribution function, dislocation density tensor and lattice distortion parameters, and construct a constitutive model of dynamic adaptive materials based on the combination of dislocation theory and crystal plastic constitutive law;

[0037] S2: Import the constitutive model into the finite element analysis software, set corresponding boundary conditions based on the temperature change, strain rate mutation, electromagnetic induction effect between the die and the sheet, and fluid dynamics behavior of the lubricating medium caused by plastic deformation during the drawing process, and form a finite element model of multi-physical field interaction behavior;

[0038] S3: Import the finite element model into the adaptive meshing algorithm, adjust the density of the mesh according to the model stress, strain gradient and physical field gradient, and form the finite element simulation results;

[0039] S4: Use deep learning algorithms to analyze finite element simulation results in real time, obtain sensitive areas of lug formation, potential triggering factors, and the relationship between multiple physical fields, and output analysis reports in the form of visual charts;

[0040] S5: The die corner radius, drawing gap, blank holder force, drawing speed, friction coefficient and electromagnetic control parameters are dynamically optimized. The drawing simulation is performed by setting a multi-objective optimization function, and the lug rate of the cup-shaped part is obtained according to the drawing simulation.

[0041] The present invention provides a method for simulating the deep drawing of anisotropic materials. Compared with the prior art, the present invention first determines the microscopic crystal structure of the anisotropic material, obtains the crystal orientation distribution function, the dislocation density tensor and the lattice distortion parameter, and constructs a constitutive model of the dynamic adaptive material based on the combination of dislocation theory and crystal plastic constitutive law; the constitutive model is imported into the finite element analysis software, and the corresponding boundary conditions are set based on the temperature change, the strain rate mutation, the electromagnetic induction effect between the die and the plate, and the fluid dynamics behavior of the lubricating medium caused by the plastic deformation during the deep drawing process, so as to form a finite element model of the interactive behavior of multiple physical fields. Model; import the finite element model into the adaptive meshing algorithm, adjust the density of the mesh according to the model stress, strain gradient and physical field gradient, and form the finite element simulation results; use the deep learning algorithm to analyze the finite element simulation results in real time, obtain the sensitive areas of lug formation, potential triggering factors and the relationship between multiple physical fields, and output the analysis report in the form of visual charts; the mold fillet radius, drawing gap, blank holding force, drawing speed, friction coefficient and electromagnetic control parameters are dynamically optimized, and the drawing simulation is performed by setting a multi-objective optimization function, and the lug rate of the cup-shaped part is obtained according to the drawing simulation. A drawing simulation method for anisotropic materials provided by the present invention constructs a constitutive model and forms the finite element results of the corresponding finite element model, and obtains the lug rate of the cup-shaped part according to the drawing simulation to predict the lug problem of the cup-shaped part.

[0042] For electron backscatter diffraction technology, the anisotropic material needs to be prepared into a flat surface, and mechanical polishing and ion beam polishing are used to obtain a high-quality surface to ensure that the electron beam can produce a clear backscattered Kikuchi diffraction pattern. The sample is scanned point by point in a scanning electron microscope, and the backscattered Kikuchi diffraction pattern of each point is collected. These patterns are analyzed by software to obtain crystal orientation information.

[0043] The atomic force microscope has relatively loose requirements for samples, but it is also necessary to ensure that the sample surface is relatively clean and flat to avoid excessive contaminants or large particles on the surface that affect the measurement results. The atomic force microscope is used to scan the sample surface, and the surface morphology information is recorded in real time based on the interaction between the probe and the sample surface to generate a surface information image.

[0044] The sample preparation for transmission electron microscopy is the most complicated, and the sample needs to be thinned to a thickness that the electron beam can penetrate, usually between tens of nanometers and hundreds of nanometers. Commonly used methods include mechanical thinning, ion thinning, and double-spray thinning. When observing samples using a transmission electron microscope, by adjusting the acceleration voltage, focus and other parameters of the electron beam, a clear internal microstructure image is obtained, and various structural features inside the sample are recorded.

[0045] The crystal orientation information obtained by electron backscatter diffraction technology, the surface information obtained by atomic force microscopy, and the internal microstructure information obtained by transmission electron microscopy are integrated together. Through specialized data analysis software or methods, combined with relevant theories and knowledge of materials science, this information is comprehensively analyzed. For example, the crystal orientation is linked to the dislocation distribution in the internal microstructure to study the effect of crystal orientation on material properties; or the surface information is combined with the internal microstructure to analyze the relationship between the surface state and the internal structure of the material. Through this comprehensive analysis, the microscopic crystal structure of anisotropic materials can be comprehensively and deeply determined, providing a solid foundation for subsequent material performance research and process optimization.

[0046] During the deep drawing process of anisotropic materials, the phenomenon of heat conduction cannot be ignored. On the one hand, the material itself generates heat due to plastic deformation, which originates from microscopic mechanisms such as dislocation movement and grain boundary sliding, causing the internal energy accumulation of the material to be converted into thermal energy. On the other hand, heat is constantly exchanged between the surrounding environment and the material. For example, in a high temperature environment, the environment transfers heat to the material; while in a low temperature environment or when there are cooling measures, the heat of the material is dissipated outward.

[0047] Drawing speed, die geometry, and material flow characteristics are intertwined and jointly affect the strain rate distribution of the material. The drawing speed directly determines the degree of deformation of the material per unit time, and the die geometry, such as the fillet radius and the die depth, limits the flow path of the material. The flow of materials in different regions is blocked or accelerated in different conditions, which leads to a sudden change in strain rate. The use of an adaptive time step algorithm is the key. In the initial stage of the simulation, a smaller initial time step is given. As the simulation progresses, the time step is adjusted based on the material strain rate sensitivity index (this index reflects the sensitivity of the material strain rate to stress, and different anisotropic materials vary greatly and need to be experimentally determined) combined with the theory of dislocation dynamics. When the strain rate in a certain area changes dramatically, it indicates that the deformation state of the material is unstable. At this time, the time step needs to be reduced to capture a more detailed deformation process; on the contrary, when the strain rate is stable, the time step should be appropriately increased to improve the calculation efficiency.

[0048] When the drawing die and the anisotropic material sheet move relative to each other, the electromagnetic induction effect occurs immediately. According to Faraday's law of electromagnetic induction, the changing magnetic field will induce an electromotive force in the material, which will then form a current and generate electromagnetic force. First, the electromagnetic field distribution must be accurately calculated. This requires consideration of the electromagnetic properties of the die (such as magnetic permeability, electrical conductivity, etc.), the drawing speed, and the electromagnetic parameters of the sheet. Solving the electromagnetic field distribution through the Maxwell equations is generally achieved with the help of professional electromagnetic field simulation software or numerical algorithms. For example, the finite difference method is used to discretize the solution area, and the electromagnetic field components are iteratively solved in combination with boundary conditions. Once the electromagnetic field distribution is obtained, the electromagnetic force can be calculated according to the Lorentz force formula. The calculated electromagnetic force is loaded into the constitutive model as a body force, that is, the electromagnetic force term is added to the mechanical equilibrium equation of the finite element model, so that the model can fully consider the influence of the electromagnetic induction effect on the material deformation and accurately set the electromagnetic induction boundary conditions.

[0049] Lubrication plays a vital role in the deep drawing process of anisotropic materials, which is related to the friction between the die and the sheet and the smoothness of the material flow. Based on the lubrication theory, the lubricating medium forms an oil film in the gap between the die and the sheet, and its pressure distribution follows the Reynolds equation. According to the actual lubrication method, such as oil lubrication, grease lubrication or solid lubrication, the initial pressure conditions of the inlet and outlet are determined. For oil lubrication, the inlet pressure is usually provided by the oil supply pump, and the outlet pressure may be affected by the external environment or the oil return system. Combined with the flow rate change of the lubricating medium, this requires consideration of the material drawing speed, the die gap and the viscosity characteristics of the lubricating medium. For example, in the case of high-speed deep drawing and a small die gap, the flow rate of the lubricating medium is accelerated. According to the continuity equation and momentum equation of fluid mechanics, the flow rate value at the boundary is adjusted. In the finite element model, the pressure boundary conditions and flow rate boundary conditions of the inlet and outlet are reasonably set, so that the model can accurately simulate the fluid dynamics behavior of the lubricating medium, reduce the friction between the die and the sheet, ensure the smooth progress of the deep drawing process, and optimize the deep drawing process effect.

[0050] The virtual material library built into the finite element simulation software contains the basic parameters of various anisotropic materials. These basic parameters cover a wide range, from the basic physical properties of materials, such as thermal conductivity, electrical conductivity, magnetic permeability, specific heat capacity, to mechanical properties, such as elastic modulus, Poisson's ratio, yield strength, hardening index, etc. Its construction is based on a large amount of experimental data accumulation and in-depth research results on the microstructure of various materials. For anisotropic materials with different crystal structures, such as metals and their alloys with face-centered cubic, body-centered cubic, and close-packed hexagonal structures, their commonalities and characteristic parameters are sorted out respectively. Taking aluminum alloy as an example, according to the different ratios of alloying elements contained in it, the corresponding basic parameter sets of 2000 series, 6000 series, 7000 series, etc. are refined. These parameter sets reflect the differences in material properties of different series of aluminum alloys under changes in crystal structure and phase composition.

[0051] When conducting a specific anisotropic material drawing simulation, the software can intelligently match and retrieve the material composition and organizational structure accurately. If a new type of titanium alloy is input, and its composition contains titanium, aluminum, vanadium and other elements in a specific proportion, the software will quickly locate the titanium alloy material parameter template with similar composition and organizational structure in the virtual material library. In terms of organizational structure, if the material has undergone a special heat treatment process to form a unique grain size distribution and dislocation configuration, the software can also identify these microstructural feature information. Through the built-in intelligent algorithm, the material composition and organizational structure information are converted into retrieval codes, and the most suitable basic parameters are quickly screened out, providing reliable data support for the initial construction of the finite element model, greatly saving the time cost of manual search and input of parameters, and improving modeling efficiency.

[0052] After retrieving the basic parameters, the virtual material library combines the measured data with the machine learning algorithm to automatically improve the finite element model. Collect the real performance data of the material from the experimental test, such as the stress-strain curve obtained by the tensile test, the hardness value distribution obtained by the hardness test, etc. The machine learning algorithm uses these measured data and the initial parameters retrieved from the virtual material library as input for model training. The neural network simulates the learning process of the human brain by constructing a multi-layer neuron structure, and automatically finds the potential rules between the measured data and the initial parameters. For example, if the measured material yield strength is higher than the initial parameter prediction value, the algorithm will automatically adjust the model parameters related to the strength, such as the crystal slip resistance coefficient, dislocation interaction parameters, etc. After multiple rounds of iterative training, the finite element model can accurately reflect the real performance of the material, provide a highly accurate model basis for subsequent simulations, and greatly improve the consistency between the simulation results and the actual situation.

[0053] When the finite element model is imported into the adaptive meshing algorithm in step S3, the following steps are included:

[0054] 5.1. Convert the data of the finite element model into a format recognizable by ABAQUS or ANSYS. ABAQUS usually uses the .inp file format to describe the model information, including detailed data such as node coordinates, unit connection relationships, material properties, boundary conditions, etc. ANSYS commonly uses .cdb or .iges formats.

[0055] 5.2. Use the Python scripting interface provided by ABAQUS software or the APDL language programming environment of ANSYS software to write adaptation code;

[0056] 5.3. Specify to enable the adaptive meshing algorithm and bind the written adaptation code to it. In ABAQUS software, by selecting to enable the adaptive meshing function in the mesh module setting options of the model, and specifying the previously written Python script as the control script, the software can dynamically adjust the mesh according to the rules set by the script during the meshing process. The same is true for ANSYS software. During the solution setting process, the adaptive mesh control commands written in the APDL language are associated with the solution process to ensure that the algorithm can optimize the mesh according to the code instructions during the solution. This binding process is crucial, as it enables the adaptation code to intervene in the meshing process in real time, fine-tune the mesh layout according to the model characteristics and simulation requirements, and improve the simulation accuracy.

[0057] 5.4. Carry out preliminary debugging, set the drawing speed and mold geometry, observe whether the algorithm module can correctly read the model data, obtain physical quantities in real time, and make preliminary mesh adjustments based on the gradient information. Through debugging log analysis, check possible data transmission interruptions and algorithm error problems, and optimize the associated settings of the adaptation code and algorithm module. If there is a syntax error, correct the code and rebind it; if the parameter setting is unreasonable, adjust the parameters such as the convergence criterion of adaptive meshing, the initial mesh size, etc., and re-debug until the algorithm module runs stably, can efficiently and accurately realize adaptive meshing, and provide a high-quality mesh model foundation for finite element simulation.

[0058] In step S4, the convolutional neural network (CNN) is good at processing data with spatial structure. For two-dimensional image information such as stress-strain cloud maps generated during the drawing process, CNN can automatically extract features in the image through the convolution layer, such as the texture of the stress concentration area, the strain gradient change pattern, etc. Its convolution kernel slides on the image to extract features of the local area. After multiple layers of convolution and pooling operations, the original image is gradually converted into a high-level abstract representation containing key features, providing a strong basis for subsequent model decisions. The long short-term memory network (LSTM) focuses on processing time series data. The drawing process is a process that changes dynamically over time. The evolution of physical quantities such as stress, strain, and temperature over time is crucial to understanding the mechanism of lug formation. Through the memory unit structure, LSTM can remember the key information in the long time series and forget the unimportant information, thereby capturing the changing trend of physical quantities in different drawing stages, and providing deep insights based on the time dimension for predicting the formation of lugs.

[0059] The preprocessed data set is divided into training set, validation set and test set in a ratio of 70:15:15. First, a large amount of physical quantity data obtained by simulation is preprocessed, which includes data cleaning to remove outliers caused by simulation errors or numerical instability; data normalization to unify the numerical range of different physical quantities into a reasonable interval, such as normalizing the stress value to the interval [0,1], so as to facilitate faster convergence during model training. When dividing the data set, the training set accounts for 70%, which is used to allow the deep learning model to learn the potential patterns and laws in the data. The model repeatedly learns the physical quantity input and the corresponding lug feature output under different drawing conditions in the training set, and gradually adjusts its internal parameters to build an accurate prediction model. The validation set accounts for 15%. During the model training process, the validation set data is regularly used to evaluate the model to observe the performance of the model on unseen data, such as accuracy, loss value and other indicators, to monitor whether the model is overfitting or underfitting, and to adjust the model's hyperparameters, such as learning rate, number of network layers, etc. The test set also accounts for 15%. After the model training is completed, the test set data is used to objectively evaluate the final performance of the model to ensure that the model has good generalization ability and can accurately predict the lug characteristics under different anisotropic materials and drawing processes.

[0060] The physical quantity data obtained by simulation are used as input. These physical quantities cover various information such as stress, strain, temperature, electromagnetic intensity, fluid velocity, etc. in the drawing process, which fully reflects the physical state of the material during deformation. Based on the selected architecture (CNN or LSTM), the model calculates the predicted output through forward propagation, that is, the characteristics of the drawn lug, including the position, height, shape and other information of the lug. Then, the predicted output is compared with the actual drawn lug characteristics to calculate the loss value, usually using loss functions such as mean square error (MSE) and cross entropy. Based on the loss value, the error information is transmitted layer by layer from the output layer to the input layer through the back propagation algorithm to adjust the weights of each neuron in the model. During multiple iterative training processes, the model continuously optimizes the weights, so that the gap between the predicted output and the actual output is gradually narrowed, thereby improving the model's prediction accuracy for the characteristics of the drawn lug.

[0061] In step S4, in the application process of the deep learning model (such as the convolutional neural network CNN), the input of stress-strain cloud maps at different times can accurately mine the sensitive areas of lug formation. The stress-strain cloud map is a visual presentation of the internal mechanical state of the material during the drawing process and contains rich information. The CNN model uses its powerful image feature extraction ability to parse the pixel-level information in the cloud map. In the model training stage, a large number of stress-strain cloud maps of different drawing conditions and different times are used as input samples, and the corresponding labels are the information of the sensitive areas of lug formation predetermined by experimental measurement or high-precision simulation. The model learns the potential correlation patterns between stress concentration areas, strain gradient mutation areas and lug formation areas through layer-by-layer processing of convolution layers and pooling layers. For example, when a local area in the cloud map continues to have high stress concentration and the strain gradient changes sharply, the model will identify the area as a highly sensitive area for lug formation after learning. In actual application, for the stress-strain cloud map generated by the new deep drawing simulation, the model can quickly locate the sensitive areas that may induce lugs, provide key directions for process optimization, and take measures in advance to suppress the formation of lugs, such as targeted adjustment of the local shape of the mold or change of the material flow path.

[0062] Different drawing process parameters are input into the deep learning model to obtain the potential triggering factors of the lugs. The drawing process parameters include drawing speed, blank holder force, die corner radius, drawing gap, friction coefficient, and electromagnetic control parameters. The dynamic changes of these parameters during the drawing process have a complex impact on the formation of lugs.

[0063] In building the deep learning model architecture, we can integrate the advantages of CNN and LSTM, use CNN to process the spatial distribution visualization data of temperature field, electromagnetic intensity field, and fluid velocity field (such as presenting the distribution of each physical field on the plate plane in the form of a two-dimensional image), and use LSTM to process the sequence data of these physical fields changing over time.

[0064] In step S4, the visualization chart includes at least a stress-strain cloud diagram, a temperature field distribution diagram, an electromagnetic intensity vector diagram, and a fluid velocity vector diagram.

[0065] The stress-strain cloud diagram visually displays the distribution of stress and strain by color-coding the stress and strain values ​​at various positions inside the material during the drawing process; the temperature field distribution diagram reflects the temperature changes caused by comprehensive factors such as heat generated by plastic deformation of the material during drawing, heat conduction between the mold and the material, and heat exchange with the environment; the electromagnetic intensity vector diagram is used to display the electromagnetic field intensity and direction generated by the electromagnetic induction effect during the drawing process. The software calculates the electromagnetic field components based on electromagnetic theory, and represents the electromagnetic intensity with vector arrows. The length of the arrow corresponds to the intensity, and the direction indicates the direction of the magnetic field; the fluid velocity vector diagram focuses on the flow state of the lubricating medium in the gap between the mold and the sheet. The flow velocity of the lubricating medium is solved according to the fluid dynamics equation and drawn in the simulation area in vector form. The above visualization charts do not exist in isolation, but work together in the process optimization process.

[0066] In step S5, a hybrid optimization strategy consisting of a genetic algorithm, a particle swarm optimization algorithm, and a simulated annealing algorithm is used to drive the coordinated dynamic optimization of the mold fillet radius, drawing gap, blank holding force, drawing speed, friction coefficient, and electromagnetic control parameters. By setting a multi-objective optimization function, the comprehensive molding quality indicators of minimizing the lug height, maximizing the wall thickness uniformity, minimizing the residual stress, and optimizing the surface quality are obtained.

[0067] In the collaborative dynamic optimization process, the value range of each parameter is first initialized, such as the radius of the mold corner and the drawing gap. Then, the three algorithms are started in parallel, and each generates a batch of candidate solutions based on the initial parameters. The genetic algorithm screens, crosses, and mutates according to its evolutionary rules; the particle swarm optimization algorithm allows particles to fly in the search space, update speed and position; the simulated annealing algorithm uses the annealing temperature to control the acceptance probability of parameter changes. Every certain number of iterations, the three algorithms communicate with each other and share the information of the currently found better solutions. For example, the parameter combination found by the particle swarm optimization algorithm that significantly improves the uniformity of wall thickness, the genetic algorithm can refer to some of its "genes" for optimization, and the simulated annealing algorithm also adjusts the acceptance probability strategy accordingly to avoid repeated searches. As the iteration proceeds, the parameter combinations found by each algorithm are continuously evaluated according to the multi-objective optimization function. When the function value change is less than the set threshold after multiple consecutive iterations, it indicates that the optimization tends to converge. At this time, the parameter combination obtained is the optimal solution to minimize the lug height, maximize the uniformity of wall thickness, minimize the residual stress, and optimize the surface quality, providing accurate process parameter guidance for the actual anisotropic material drawing production.

[0068] After step S5, scanning technology is used to scan the entire process of trial production cup-shaped parts from blank cutting to forming. Different scanning technologies can be selected according to the accuracy requirements, size and material properties of the cup-shaped parts. For high-precision, small cup-shaped parts, optical three-dimensional scanning technology is used. It uses the principle of laser triangulation or structured light encoding technology to quickly obtain three-dimensional coordinate information on the surface of the object with an accuracy of sub-millimeter or even micron level. For larger cup-shaped parts with strong surface reflection, industrial CT scanning technology is used. It penetrates the object through X-rays, collects projection data from multiple angles, and then uses reconstruction algorithms to generate a three-dimensional model of the interior and surface of the object. It can not only capture the external contour, but also provide insight into internal structural defects, such as pores and inclusions inside the material.

[0069] In order to transmit the collected scanning data to the finite element model in real time, it is necessary to build a stable and efficient data transmission link. First, on the scanning device side, it is necessary to configure the data acquisition software, which must have the function of real-time data output and encapsulate the scanned point cloud data or volume data in a predetermined format. At the same time, the data transmission protocol must be set up to ensure the reliability of the data in network transmission. On the computing side where the finite element model is located, it is necessary to develop a corresponding data receiving interface program that can identify and parse the data format transmitted from the scanning device and convert it into a data structure that can be directly read by the finite element model. To ensure real-time performance, multi-threading technology can be used. On the one hand, it is responsible for data reception, and on the other hand, it is responsible for real-time updating of the geometric information and material properties in the finite element model to avoid model correction lags due to data transmission delays.

[0070] When dynamically correcting the finite element model, the first step is to correct the geometric model. The actual shape of the cup-shaped part obtained by scanning is compared with the initial geometric shape of the finite element model. The geometric shape of the finite element model is accurately matched with the real object through means such as node displacement adjustment and unit deformation correction. The second step is material property correction. According to the information such as internal defects and thickness changes of the material obtained by scanning, the material parameters in the finite element model are adjusted. If it is found that the hardness of a certain area increases and the elastic modulus changes due to work hardening, the material constitutive relationship parameters in the model are updated accordingly so that the model can reflect the real changes in the mechanical properties of the material. Through continuous dynamic correction, the finite element model can closely follow the actual situation of trial production, continuously optimize the prediction accuracy, and provide an increasingly reliable basis for subsequent process improvements.

[0071] Embodiment 1:

[0072] Taking the high-strength titanium alloy anisotropic sheet commonly used in the aerospace field as an example, the sheet thickness is 2.0 mm.

[0073] Firstly, an emission scanning electron microscope equipped with an EBSD probe, combined with an atomic force microscope and a transmission electron microscope, was used to perform a detailed scanning analysis of the microscopic crystal structure of the plate, and the crystal orientation distribution function (ODF), dislocation density tensor and lattice distortion parameters were obtained. The micromechanical parameters were measured using a Hysitron TI 950 nanoindenter, and a dynamic adaptive crystal plasticity constitutive model was constructed and imported into the ABAQUS software.

[0074] With the help of Siemens NX software, a three-dimensional model including thermal-mechanical-structural-electromagnetic-fluid multi-physics field coupling was created according to the cup-shaped part design drawings. After being imported into ABAQUS software, the adaptive meshing algorithm refined the mesh of the key deformation area of ​​the plate to 50μm accuracy based on the initial simulated stress, strain and temperature gradient, set the thermal-mechanical-electromagnetic-fluid coupling boundary conditions, and set the friction coefficient between the mold and the plate to 0.12. Considering the deep drawing thermal effect and electromagnetic induction, stamping oil was used as the lubricating medium with a viscosity of 30-100cSt (30℃).

[0075] The simulation was run, and the deep learning algorithm tracked the evolution of stress, strain, temperature, electromagnetic intensity, and fluid velocity in real time to lock the potential area of ​​the lug. A hybrid optimization strategy consisting of genetic algorithm, particle swarm optimization algorithm, simulated annealing algorithm, and ant colony algorithm was used to drive the collaborative optimization of the mold fillet radius in the range of 2 to 5 mm, the drawing gap in the range of 0.1 to 0.3 mm, the blank holding force in the range of 50 to 150 kN, the drawing speed in the range of 10 to 30 mm / s, the electromagnetic control parameters in the range of 0.5 to 2.0 T, and the viscosity of the lubricating medium in the range of 30-100 cSt (30°C). The optimal parameters were determined after 50 rounds of iterative optimization.

[0076] Based on this parameter, 100 cup-shaped parts were trial-produced in a small batch. The GOM ATOS 5 series 3D optical measurement system combined with FARO Focus 3D lidar scanning technology and Zeiss Xradia 520 Versa X-ray tomography imaging system were used for real-time monitoring. The deviation data was fed back to the ABAQUS software correction model in real time. Ultimately, the lug height was reduced from the initial average of 3.5mm to below 0.8mm, the wall thickness uniformity was improved by 30%, the residual stress was reduced by 40%, and the product molding quality was excellent, meeting aerospace standards.

[0077] Embodiment 2:

[0078] Taking the anisotropic aluminum alloy sheet commonly used in automobile engine manufacturing as an example, the sheet thickness is 1.5 mm.

[0079] Initially, an emission scanning electron microscope equipped with an EBSD probe was used, combined with an atomic force microscope and a transmission electron microscope, to comprehensively analyze the microscopic crystal structure of the plate, accurately determine the parameters related to the crystal orientation, the details of the dislocation distribution, and the microscopic distortion state of the lattice, and use the Nano Indenter G200 nanoindenter to accurately obtain the microscopic mechanical properties, and then build a dynamic constitutive model that matches the characteristics of the plate, and successfully implant it into the ANSYS software platform.

[0080] With the help of CATIA software, a sophisticated three-dimensional model covering the multi-physics interaction of heat, force, structure, electromagnetic and fluid is constructed according to the specific design blueprint of special-shaped parts. After being imported into ANSYS software, the adaptive meshing algorithm refines the mesh of the key deformation area of ​​the plate to 30μm accuracy based on the initial simulated stress, strain and temperature gradient, sets the thermal-mechanical-electromagnetic-fluid collaborative boundary conditions, and sets the friction coefficient between the mold and the plate to 0.1. Taking full consideration of the heat accumulation and electromagnetic influence in the stamping process, the lubricating medium is selected as stamping oil, and its viscosity is in the range of 20-80cSt (35℃).

[0081] Running simulation, deep learning algorithm tracks the evolution of stress, strain, temperature, electromagnetic intensity, fluid velocity in real time, and locks the potential area of ​​lug. A hybrid optimization strategy consisting of genetic algorithm, particle swarm optimization algorithm, simulated annealing algorithm, and ant colony algorithm is used to drive the collaborative optimization of die fillet radius in the range of 3-6mm, drawing gap in the range of 0.08-0.25mm, blank holding force in the range of 40-120kN, drawing speed in the range of 8-25mm / s, electromagnetic control parameters in the range of 0.3-1.5T, and lubricating medium viscosity in the range of 20-80cSt (35℃). After 60 rounds of iterative optimization, the optimal parameters are determined.

[0082] Based on the optimized parameters, a small batch trial was carried out to produce 80 special-shaped parts. The quality was controlled in real time with the help of Leica T-Scan 5 3D laser scanning system, Creaform HandySCAN 3D handheld scanner and Nikon XT H 225 ST micro-focus X-ray inspection equipment. The collected deviation information was immediately fed back to ANSYS software to calibrate the model. Ultimately, the dimensional deviation of key parts of the product was reduced from the initial average of 2.5mm to within 0.6mm, the uniformity of cross-sectional thickness was enhanced by 25%, the residual stress was reduced by 35%, and the molding quality was excellent, which fully met the manufacturing standards of automobile engines.

[0083] The above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions and improvements made within the spirit and principles of the present invention should be included in the protection scope of the present invention.

Claims

1. A method for simulating deep drawing of anisotropic materials, characterized in that: The following steps are involved: S1: Determine the microscopic crystal structure of anisotropic materials, obtain the crystal orientation distribution function, dislocation density tensor and lattice distortion parameters, and construct a constitutive model of dynamic adaptive materials based on the combination of dislocation theory and crystal plastic constitutive law; S2: Import the constitutive model into the finite element analysis software, set corresponding boundary conditions based on the temperature change, strain rate mutation, electromagnetic induction effect between the die and the sheet, and fluid dynamics behavior of the lubricating medium caused by plastic deformation during the drawing process, and form a finite element model of multi-physical field interaction behavior; S3: Import the finite element model into the adaptive meshing algorithm, adjust the density of the mesh according to the model stress, strain gradient and physical field gradient, and form the finite element simulation results; S4: Use deep learning algorithms to analyze finite element simulation results in real time, obtain sensitive areas of lug formation, potential triggering factors, and the relationship between multiple physical fields, and output analysis reports in the form of visual charts; S5: The die corner radius, drawing gap, blank holder force, drawing speed, friction coefficient and electromagnetic control parameters are dynamically optimized. The drawing simulation is performed by setting a multi-objective optimization function, and the lug rate of the cup-shaped part is obtained according to the drawing simulation.

2. A method for simulating deep drawing of anisotropic materials according to claim 1, characterized in that: In step S1, electron backscatter diffraction technology, atomic force microscopy and transmission electron microscopy are used to work together to comprehensively analyze the crystal orientation information obtained by electron backscatter diffraction technology, the surface information obtained by atomic force microscopy and the internal microstructure information obtained by transmission electron microscopy to determine the microscopic crystal structure of the anisotropic material.

3. A method for simulating deep drawing of anisotropic materials according to claim 1, characterized in that: When setting the corresponding boundary conditions in step S2, 3.

1. Set thermal boundary conditions based on the heat conduction generated by the material itself and the heat conduction between the surrounding environment and the material; 3.

2. Based on the drawing speed, die geometry and material flow characteristics, an adaptive time step algorithm is used to adjust the parameters in the constitutive model according to the material strain rate sensitivity index combined with the dislocation dynamics theory to set the strain rate boundary conditions; 3.

3. Calculate the electromagnetic force based on the electromagnetic field distribution corresponding to the electromagnetic induction effect, and load the electromagnetic force into the constitutive model as a body force to set the electromagnetic induction boundary conditions; 3.

4. Based on lubrication theory and actual lubrication method, set the pressure boundary conditions at the inlet and outlet, and set the flow rate boundary conditions in combination with the flow rate change of the lubrication medium.

4. A method for simulating deep drawing of anisotropic materials according to claim 1, characterized in that: In step S2, the finite element simulation software has a built-in virtual material library, which covers the basic parameters of various anisotropic materials. The corresponding basic parameters can be retrieved according to the material composition and organizational structure, and the finite element model can be automatically improved through a machine learning algorithm in combination with the measured data.

5. A method for simulating deep drawing of anisotropic materials according to claim 1, characterized in that: When the finite element model is imported into the adaptive meshing algorithm in step S3, 5.

1. Convert the data of the finite element model into a format recognizable by ABAQUS or ANSYS software; 5.

2. Use the Python scripting interface provided by ABAQUS software or the APDL language programming environment of ANSYS software to write adaptation code; 5.

3. Specify to enable the adaptive meshing algorithm and bind the written adaptation code to it; 5.

4. Perform preliminary debugging, set the drawing speed and mold geometry, observe whether the algorithm module can correctly read the model data, obtain physical quantities in real time, and make preliminary grid adjustments based on the gradient information. Analyze and troubleshoot possible data transmission interruptions and algorithm errors through debugging log analysis, and optimize the association settings between the adaptation code and the algorithm module.

6. A method for simulating deep drawing of anisotropic materials according to claim 1, characterized in that: In step S4, a convolutional neural network or a long short-term memory network is selected as the basic architecture of the deep learning model, and the preprocessed data set is divided into a training set, a validation set, and a test set in a ratio of 70:15:

15. The deep learning model is trained using the training set, and the simulated physical quantity data is used as input. The characteristics of the drawn lug are used as output, and the weight of the deep learning model is adjusted at the same time.

7. A method for simulating deep drawing of anisotropic materials according to claim 1, characterized in that: In step S4, the stress-strain cloud diagrams at different times are input into the deep learning model to obtain the sensitive area formed by the lug; Different process parameters of deep drawing were input into the deep learning model to obtain the potential triggering factors of the lugs; Analyze the temperature field, electromagnetic intensity, and fluid velocity multiphysics fields associated with the lug.

8. A method for simulating deep drawing of anisotropic materials according to claim 7, characterized in that: In step S4, the visualization chart at least includes a stress-strain cloud diagram, a temperature field distribution diagram, an electromagnetic intensity vector diagram, and a fluid velocity vector diagram.

9. A method for simulating deep drawing of anisotropic materials according to claim 1, characterized in that: In step S5, a hybrid optimization strategy consisting of a genetic algorithm, a particle swarm optimization algorithm, and a simulated annealing algorithm is used to drive the coordinated dynamic optimization of the mold fillet radius, drawing gap, blank holding force, drawing speed, friction coefficient, and electromagnetic control parameters. By setting a multi-objective optimization function, the comprehensive molding quality indicators of minimizing the lug height, maximizing the wall thickness uniformity, minimizing the residual stress, and optimizing the surface quality are obtained.

10. The method for simulating deep drawing of anisotropic materials according to claim 1, characterized in that: After step S5, scanning technology is used to scan the entire process of trial production cup-shaped parts from blank cutting to forming, and the collected scanning data is transmitted to the finite element model in real time to dynamically correct the finite element model.

Images