Simulation dynamic modeling method for machining deformation of large complex structural member

Abstract

The invention provides a large-scale complex structural member machining deformation simulation dynamic modeling method, and relates to the technical field of machining and manufacturing, and the method comprises the steps: building an intelligent model, and carrying out the dynamic optimization: building a three-dimensional model, marking an easy-to-deform region, embedding tolerance information, and carrying out the self-adaptive grid division; carrying out multi-source data driven physical modeling, collecting data and correcting material attributes, and constructing an intelligent constitutive model; multi-scale coupling dynamic simulation is carried out, and macro-micro model coupling and machining force dynamic prediction are achieved; and digital twinborn verification and closed-loop control are carried out, deformation deviation is quantified, and machining parameters are optimized. According to the method, the machining mechanical behavior of the large complex structural part can be more accurately simulated by accurately marking the easily-deformed area, embedding tolerance information, performing adaptive grid division, driving and correcting material attributes through multi-source data and constructing the intelligent constitutive model, and the simulation precision is improved.

Description

Technical Field

[0001] The present invention relates to the technical field of machining and manufacturing, and particularly to a dynamic modeling method for machining deformation simulation of large complex structural parts. Background Art

[0002] In the machining and manufacturing industry, large complex structural parts are widely used in important fields such as aerospace, shipbuilding, and the automotive industry. Such structural parts usually have the characteristics of complex shapes and high precision requirements. Their machining processes are extremely susceptible to various factors, such as material properties, machining process parameters, cutting forces, and thermal deformation, which can lead to machining deformation problems. Machining deformation not only reduces the dimensional accuracy and surface quality of products, affects the performance and reliability of products, but may also cause product scrapping, increasing production costs and production cycles.

[0003] Traditional machining deformation simulation methods mainly rely on finite element analysis (FEA) for static or quasi-static simulations, and it is difficult to accurately predict the deformation behavior during the dynamic machining process; traditional methods have obvious deficiencies in aspects such as model construction, data processing, multi-scale coupling, real-time control, and data management. For example, it is difficult to accurately mark the easily deformed areas during model construction, and the mesh division lacks adaptability, resulting in low simulation accuracy; in data processing, the correction of material properties is inaccurate, the fusion of multi-source data is insufficient, and the constitutive model cannot reflect the actual mechanical behavior of materials; multi-scale coupling is difficult, and it is difficult to achieve precise coordination of macro and micro models; real-time monitoring and closed-loop control are lacking, and machining parameters cannot be dynamically adjusted according to deformation data; data management is decentralized and insecure, lacking a unified data channel and a reliable evidence storage mechanism. Therefore, it is necessary to provide a dynamic modeling method for machining deformation simulation of large complex structural parts to solve the above technical problems. Summary of the Invention

[0004] The purpose of the present invention is to provide a dynamic modeling method for machining deformation simulation of large complex structural parts. By constructing an intelligent model, multi-source data-driven physical modeling, multi-scale coupling dynamic simulation, digital twin verification and closed-loop control, and full-process data management, it realizes the accurate prediction and effective control of machining deformation of large complex structural parts, improves machining accuracy and production efficiency, and reduces production costs.

[0005] To solve the above technical problems, a dynamic modeling method for machining deformation simulation of large complex structural parts provided by the present invention includes the following steps:

[0006] Step 1, construct an intelligent model and perform dynamic optimization, including establishing a three-dimensional model, marking the easily deformed areas, embedding tolerance information, and adaptive mesh division;

[0007] Step 2, perform multi-source data-driven physical modeling, collect data and correct material properties, and construct an intelligent constitutive model;

[0008] Step 3: Conduct multi-scale coupled dynamic simulation to achieve the coupling of macro and micro models and the dynamic prediction of machining forces;

[0009] Step 4: Conduct digital twin verification and closed-loop control to quantify deformation deviation and optimize machining parameters;

[0010] Step 5: Implement full-process data management, establish a real-time data channel and conduct blockchain evidence storage.

[0011] Preferably, in the construction of the intelligent model and dynamic optimization, specifically:

[0012] Obtain the design drawings of large complex structural parts, and use 3D modeling software to create their 3D geometric models according to the design drawings; calculate the wall thickness and curvature radius of each surface in the 3D geometric model;

[0013] Use a feature recognition algorithm to mark the easily deformed areas of each surface in the 3D geometric model, and set the marking conditions for the easily deformed areas; the marking conditions include the surface areas where both the wall thickness and the curvature radius are less than the corresponding preset thresholds; use the Kriging interpolation method to embed the wall thickness fluctuation tolerance information;

[0014] The adaptive mesh division adopts octree coarse division, and the voxel edge is set. Combine the Zienkiewicz-Zhu error estimation formula and the particle swarm optimization algorithm for mesh encryption. The particle swarm velocity update formula in the particle swarm optimization algorithm is:

[0015] ;

[0016] where represents the velocity of the i-th particle at the (k + 1)-th iteration, w is the inertia weight, and its value range is usually [0.4, 0.9], represents the learning factors, which respectively adjust the step sizes of the particles flying towards their own historical optimal positions and the global optimal position, represents a random number uniformly distributed in the interval [0, 1]; represents the historical optimal position of the i-th particle, represents the global optimal position, represents the position of the i-th particle at the k-th iteration.

[0017] Preferably, when conducting multi-source data-driven physical modeling, use a dynamometer and a laser displacement sensor to collect cutting data and displacement data; the dynamometer uses Kistler9257B, its sampling rate is 10 kHz, and the laser displacement sensor uses Keyence, with an accuracy of ±1 μm; use the extended Kalman filter algorithm to fuse the data to correct the material properties. The state transition matrix of the extended Kalman filter algorithm is:

[0018] ; where represents the sampling time of the collected data;

[0019] Construct a non - linear constitutive model based on LSTM, with the input being cutting data and temperature, and the output being the stress - strain curve; where the cutting data includes cutting speed and feed rate;

[0020] The training process of the non - linear constitutive model based on LSTM uses the mean - square error loss function, and the formula is: ; where L represents the value of the loss function, u represents the number of training samples, b represents the number of the training sample, represents the true stress - strain value of the b - th sample, is the model - predicted stress - strain value of the b - th sample, and the model parameters are optimized by minimizing the loss function.

[0021] Preferably, in the multi - scale coupling dynamic simulation, the macro - micro model includes a macro model and a micro - crystal plasticity model. The macro model uses Abaqus / Explicit, and the micro - crystal plasticity model uses VPSC. They are coupled through the homogenization theory, and the coupling formula is: , ; where, represents the macro stress, represents the micro stress, V is the volume of the material, represents the macro strain, represents the micro strain, represents the local strain;

[0022] Use the alternating iteration method to transfer stress - strain and preset the convergence accuracy; the LSTM model based on the attention mechanism predicts the cutting force fluctuation by setting a time interval in advance;

[0023] The calculation formula for the LSTM model based on the attention mechanism to predict the cutting force fluctuation is: ; where represents the predicted cutting force, represents the attention weight of the output of the s - th LSTM hidden layer to the predicted cutting force, , and hs represents the output of the LSTM hidden layer.

[0024] Preferably, in the digital twin verification and closed - loop control, use industrial CT scanning to obtain point cloud data, and use the ICP algorithm to align the point cloud data with the simulation model. The IPC algorithm formula is: , ; where, R represents the rotation matrix, t represents the translation vector, represents the j-th point in the point cloud data, which is the corresponding point in the simulation model, represents the centroid of the point cloud data, represents the centroid of the corresponding point set in the simulation model;

[0025] Then, identify the out-of-tolerance area based on the normal vector consistency;

[0026] Design a multi-objective reward function: ; where r represents the reward value, represents the deformation deviation, represents the machining time deviation, represents the energy consumption deviation; online adjust the machining parameters through the Actor-Critic network.

[0027] Preferably, in the Actor-Critic network, the Actor-Critic network includes an Actor network and a Critic network; the output of the Actor network is the adjustment amount of the machining parameters, the input of the Critic network is the current state and the output of the Actor network, and the output is the value evaluation of the current state and action. Through continuous iterative training, the Actor network can output the optimal adjustment amount of the machining parameters according to different states; where the current state includes the deformation deviation, the machining time deviation, and the energy consumption deviation.

[0028] Preferably, in the multi-scale coupled dynamic simulation, every time a set number of iterations are performed, the crystal orientation distribution in the microscopic model is updated once.

[0029] Preferably, the Zienkiewicz-Zhu error estimation formula is: ; where represents the overall error estimation value, represents the number of grid cells, represents the error estimation value of the e-th grid cell; when the overall error estimation value is greater than its preset overall error tolerance threshold, then trigger the use of the particle swarm optimization algorithm for grid encryption.

[0030] Compared with the related technologies, a large complex structural part machining deformation simulation dynamic modeling method provided by the present invention has the following beneficial effects:

[0031] 1. By accurately marking the easily deformed areas, embedding tolerance information, adaptive mesh division, and multi-source data-driven correction of material properties and construction of an intelligent constitutive model, the present invention can more accurately simulate the machining mechanical behavior of large complex structural parts and improve the simulation accuracy.

[0032] 2. The present invention realizes the effective combination of macro and micro models through multi-scale coupled dynamic simulation, considers the influence of the micro-structure on the macro-deformation, predicts the cutting force fluctuation in advance, provides comprehensive information for the machining process control; and uses digital twin verification and closed-loop control for real-time monitoring and feedback, quantifies the deformation deviation and optimizes the machining parameters, optimizes the machining process, and improves the machining accuracy and product quality. Brief Description of the Drawings

[0033] Figure 1 It is a flow block diagram of a method for dynamic modeling of machining deformation simulation of a large and complex structural part provided by the present invention. Detailed Embodiments

[0034] Next, the technical solutions in the embodiments of the present invention will be clearly and completely described in conjunction with the accompanying drawings in the embodiments of the present invention. Obviously, the described embodiments are only a part of the embodiments of the present invention, rather than all the embodiments. All other embodiments obtained by those of ordinary skill in the art based on the embodiments of the present invention without creative efforts shall fall within the protection scope of the present invention.

[0035] The terms used in this disclosure are only for the purpose of describing specific embodiments and are not intended to limit this disclosure. The singular forms of "group", "class" and "the" used in this disclosure and the appended claims are also intended to include the plural forms unless the context clearly indicates otherwise. It should also be understood that the term "and / or" as used herein refers to and includes any and all possible combinations of one or more of the associated listed items.

[0036] It should be understood that although the terms first, second, third, etc. may be used in this disclosure to describe various information, such information should not be limited to these terms. These terms are only used to distinguish the same type of information from each other. For example, without departing from the scope of this disclosure, the first information may also be referred to as the second information, and similarly, the second information may also be referred to as the first information. Depending on the context, the word "if" as used herein may be interpreted as "when" or "while" or "in response to a determination".

[0037] Please refer to Figure 1 A method for dynamic modeling of machining deformation simulation of a large and complex structural part, comprising the following steps:

[0038] Step 1, constructing an intelligent model and performing dynamic optimization, including establishing a three-dimensional model, marking the easily deformed areas, embedding tolerance information, and adaptive mesh generation;

[0039] Step 2, performing physics-based modeling driven by multi-source data, collecting data and correcting material properties, and constructing an intelligent constitutive model;

[0040] Step 3: Conduct multi-scale coupled dynamic simulation to achieve the coupling of macro and micro models and the dynamic prediction of machining forces;

[0041] Step 4: Perform digital twin verification and closed-loop control to quantify deformation deviations and optimize machining parameters;

[0042] Step 5: Implement full-process data management, establish a real-time data channel and conduct blockchain evidence storage, specifically:

[0043] Establish a real-time data channel based on the OPCUA protocol to achieve data interaction between systems such as CAD, CAE, and CNC; install OPCUA servers and clients in each system, configure the corresponding communication parameters to ensure low data interaction latency; use Hyperledger Fabric blockchain technology to store evidence of key process parameters (such as cutting speed, feed rate, spindle speed, machining time); after encrypting the key process parameters, package them into transaction data and send them to the blockchain network; nodes in the blockchain network verify and record the transaction data to form an immutable distributed ledger.

[0044] It should be noted that through the complementarity of each step, together they achieve the accurate prediction and effective control of machining deformation of large and complex structural parts, as well as the comprehensive management and optimization of the machining process, which plays an important role in improving machining quality, reducing costs, enhancing data security and traceability, etc.;

[0045] Construct an intelligent model and dynamically optimize: Establish a 3D model to provide a basic framework for subsequent analysis, accurately mark the deformation-prone areas, and make subsequent analysis focus on these key parts. Embed tolerance information to make the model more in line with the actual machining situation. Adaptive mesh generation can automatically adjust the mesh density according to the complexity and deformation characteristics of the structural part, improving the calculation efficiency while ensuring calculation accuracy, laying a foundation for accurately simulating machining deformation;

[0046] Physics-based modeling driven by multi-source data: Collect multi-source data (such as cutting data, displacement data, and temperature data) to truly reflect various physical phenomena during the machining process. Modify the material properties to enable the model to more accurately describe the mechanical behavior of the material during actual machining. Construct an intelligent constitutive model that can establish an accurate relationship between input parameters (cutting parameters and temperature) and the output stress-strain curve, providing a reliable material model for subsequent simulation analysis;

[0047] Multi-scale coupled dynamic simulation: The coupling of macro and micro models can comprehensively analyze the deformation of the structural part from both macro and micro levels, considering the influence of the micro-structure on macro-deformation and improving the accuracy of the simulation. The dynamic prediction of machining forces can obtain the force changes during the machining process in advance, providing a basis for timely adjustment of machining parameters and avoiding machining deformation caused by excessive or fluctuating machining forces;

[0048] Digital twin verification and closed-loop control: Quantify the deformation deviation, compare the actual machining results with the simulation model, and accurately evaluate the machining quality; By optimizing the machining parameters and adjusting the machining process in real time according to the deformation deviation, realize the closed-loop control of the machining process, continuously improve the machining accuracy, and reduce the scrap rate;

[0049] Full-process data management: Establish a real-time data channel based on the OPCUA protocol to achieve efficient data interaction between systems such as CAD, CAE, and CNC, ensure real-time data sharing and collaborative work in all links, and improve production efficiency; Use Hyperledger Fabric blockchain technology to deposit and prove the key process parameters, ensure the security, immutability, and traceability of the data, and provide strong support for quality traceability, process improvement, and production management.

[0050] In this application, in the construction of the intelligent model and dynamic optimization, specifically:

[0051] Obtain the design drawings of large complex structural parts, and use 3D modeling software to create their 3D geometric models according to the design drawings, building a basic framework for the entire simulation process; Calculate the wall thickness and curvature radius of each surface in the 3D geometric model, quantify the geometric features of the structural parts, and provide specific data basis for marking the easily deformed areas later;

[0052] Use feature recognition algorithms to mark the easily deformed areas on each surface in the 3D geometric model and set the marking conditions for the easily deformed areas; The marking conditions include surface areas where both the wall thickness and curvature radius are less than the corresponding preset thresholds; Embed the wall thickness fluctuation tolerance information using Kriging interpolation method;

[0053] It should be further noted that for the complex structures in the 3D geometric model, use a pre-trained convolutional neural network for feature recognition. This convolutional neural network is trained using a large number of structural part models containing easily deformed areas, and can automatically identify the easily deformed areas in the model and mark them, providing a basis for subsequent analysis and processing; Using feature recognition algorithms and convolutional neural networks to mark the easily deformed areas can accurately locate the parts of the structural parts that are prone to deformation, enabling subsequent analysis and processing to focus on these key areas, improving the pertinence and accuracy of the simulation;

[0054] The Kriging interpolation method is used to embed the wall thickness fluctuation tolerance information. Specifically, interpolation points are selected at the key parts of the structural member model (such as the boundaries of easily deformed areas and stress concentration areas) and evenly distributed on the model surface. The Gaussian semi-variogram is selected as the basic function for Kriging interpolation, and parameters such as its range and nugget value are determined through the cross-validation method. According to the selected interpolation points and the determined semi-variogram parameters, the wall thickness fluctuation values of each point of the model are calculated and embedded into the three-dimensional geometric model, making the model more realistically reflect the dimensional changes in actual processing. By using the Kriging interpolation method to embed the wall thickness fluctuation tolerance information, the dimensional fluctuations in actual processing are considered. By selecting interpolation points at key parts and on the model surface, combining the Gaussian semi-variogram and cross-validation to determine parameters, calculating and embedding the wall thickness fluctuation values, the model more realistically reflects the dimensional changes in actual processing, improves the accuracy and reliability of the model, and considering that wall thickness fluctuation is an inevitable factor in actual processing, incorporating it into the model can better simulate the uncertainties in the processing process and provide more realistic boundary conditions for subsequent simulation analysis.

[0055] Adaptive mesh generation uses octree coarse meshing and sets the voxel edge. Combining the Zienkiewicz-Zhu error estimation formula and the particle swarm optimization algorithm for mesh refinement, the particle swarm velocity update formula in the particle swarm optimization algorithm is:

[0056] ;

[0057] where represents the velocity of the i-th particle at the (k + 1)-th iteration. w is the inertia weight, and its value range is usually [0.4, 0.9], which is used to balance the global search and local search capabilities of the particle. represents the learning factor, and generally takes the value of to adjust the step lengths of the particle flying towards its own historical optimal position and the global optimal position respectively. represents a random number uniformly distributed in the interval [0, 1], which is used to increase the randomness of the search; represents the historical optimal position of the i-th particle, represents the global optimal position, represents the position of the i-th particle at the k-th iteration. By using octree coarse meshing and combining the Zienkiewicz-Zhu error estimation formula and the particle swarm optimization algorithm for mesh refinement, the mesh density can be adaptively adjusted according to the complexity of the model and the error distribution, avoiding unnecessary waste of computing resources while ensuring the calculation accuracy and improving the calculation efficiency.

[0058] In this application, when performing multi-source data-driven physical modeling, a dynamometer and a laser displacement sensor are used to collect cutting data and displacement data; the Extended Kalman Filter algorithm is used to fuse the data to correct the material properties, and the state transition matrix of the Extended Kalman Filter algorithm is:

[0059] ; where represents the sampling time of the collected data;

[0060] A non-linear constitutive model based on LSTM is constructed, with the cutting data and temperature as inputs and the stress-strain curve as the output; the cutting data includes cutting speed and feed rate;

[0061] By using the Extended Kalman Filter algorithm to fuse the data to correct the material properties, in actual machining, the properties of the material will change due to the influence of various factors, and traditional static material properties are difficult to accurately describe this dynamic change. The Extended Kalman Filter algorithm can comprehensively consider the information of different data sources and dynamically update and correct the material properties through the state transition matrix, making the material properties more in line with the actual machining situation; the state transition matrix contains the sampling time, which means that the algorithm can take into account the influence of time factors on the material properties, so as to better simulate the dynamic mechanical behavior of the material during the machining process;

[0062] The training process of the non-linear constitutive model based on LSTM uses the mean square error loss function, and the formula is: ; where L represents the value of the loss function, u represents the number of training samples, b represents the number of the training sample, represents the true stress-strain value of the b-th sample, is the model-predicted stress-strain value of the b-th sample, and the model parameters are optimized by minimizing the loss function;

[0063] By constructing a non-linear constitutive model based on LSTM (Long Short-Term Memory network), the complex non-linear relationship between cutting data, temperature and stress-strain curve can be processed; during the machining process of large and complex structural parts, the mechanical behavior of the material is comprehensively affected by various factors and shows highly non-linear characteristics; the LSTM network has the ability to memorize and process sequence data, can capture the long-term dependence relationship between these factors, and thus more accurately predict the stress-strain response of the material.

[0064] In this application, in the multi-scale coupled dynamic simulation, the macro-micro model includes a macro model and a microcrystal plasticity model. The macro model uses Abaqus / Explicit, and the microcrystal plasticity model uses VPSC. They are coupled through the homogenization theory, and the coupling formula is: , ; among them, represents the macroscopic stress, represents the microscopic stress, V is the volume of the material, represents the macroscopic strain, represents the microscopic strain, represents the local strain;

[0065] Abaqus / Explicit is used for the macroscopic model, and VPSC is used for the microscopic crystal plasticity model. The mechanical behavior of the structural component is analyzed from the macroscopic overall structure and the microscopic crystal level respectively. The macroscopic model can describe the overall deformation trend and mechanical response of the structural component, while the microscopic model delves into the crystal scale to reveal the plastic deformation mechanism of the material. The two are coupled through the homogenization theory, and the relationship between the macroscopic stress-strain and the microscopic stress-strain is established using the coupling formula, enabling the deformation of the microscopic crystal to affect the mechanical properties of the macroscopic structure, and the macroscopic mechanical conditions can also be fed back into the deformation analysis of the microscopic crystal, truly reflecting the influence of the microscopic structure on the macroscopic deformation;

[0066] The stress-strain is transferred using the alternating iteration method to enable continuous information exchange between the macroscopic model and the microscopic model, gradually approaching the real mechanical state. During the iteration process, the stress-strain results of the macroscopic model are transferred to the microscopic model. The microscopic model calculates the microscopic stress-strain based on this information and then feeds the results back to the macroscopic model. This cycle continues, enabling the model to more accurately simulate the mechanical changes during the processing. And a convergence accuracy of 1e-5 is preset to ensure that the iteration process stops after reaching a certain accuracy requirement. The LSTM model based on the attention mechanism sets a time interval (50 ms) in advance to predict the cutting force fluctuation;

[0067] The calculation formula for the LSTM model based on the attention mechanism to predict the cutting force fluctuation is: ; among them represents the predicted cutting force, represents the attention weight of the output of the s-th LSTM hidden layer to the predicted cutting force, , hs represents the output of the LSTM hidden layer. Through the attention mechanism, the LSTM model can automatically learn the importance of the outputs of different hidden layers at different times to the predicted cutting force, assign corresponding attention weights to each hidden layer output, enabling the model to focus on key information when processing complex cutting force data, and thus more accurately predict the cutting force fluctuation.

[0068] It should be noted that its attention weight is obtained by calculating the similarity between the hidden layer output hs and a learnable vector u and passing through the Softmax function, that is ; The input of the model is the previous cutting force data. After being processed by the LSTM layer, the attention weights at each moment are calculated through the attention mechanism, and the weighted sum of the hidden layer outputs is used to obtain the predicted cutting force. ; During model training, the learning rate is set to 0.001, and the number of training epochs is set to 500. By continuously optimizing the model parameters, the accuracy of cutting force prediction is improved, providing a basis for adjusting the machining process.

[0069] In this application, during digital twin verification and closed-loop control, industrial CT scanning (accuracy ±5μm) is used to obtain point cloud data, and the ICP algorithm is adopted to align the point cloud data with the simulation model. The ICP algorithm formula is: , ; where, R represents the rotation matrix, t represents the translation vector, represents the j-th point in the point cloud data, is the corresponding point in the simulation model, represents the centroid of the point cloud data, represents the centroid of the corresponding point set in the simulation model; by calculating the rotation matrix R and the translation vector t, the two are accurately corresponding in space, providing a basis for subsequent precise comparison;

[0070] Then, the out-of-tolerance area is identified based on the normal vector consistency. Specifically: calculate the normal vectors of each point in the point cloud data and the normal vectors of the corresponding points in the simulation model. By comparing the included angle θ of the normal vectors, when cosθ < 0.95, it indicates that there is an out-of-tolerance in this area, and this area is marked as the out-of-tolerance area;

[0071] Design a multi-objective reward function: ; where, r represents the reward value, represents the deformation deviation, represents the machining time deviation, represents the energy consumption deviation; the machining parameters are adjusted online through the Actor-Critic network. The feed speed range is 500 - 2000 mm / min, and the spindle speed range is 8000 - 15000 rpm; by using the Actor - Critic network to adjust the machining parameters online, the Actor network outputs the adjustment amount of the machining parameters according to the current state (deformation deviation, machining time deviation, energy consumption deviation), and the Critic network evaluates the value of the current state and action. Through continuous iterative training, the Actor network can learn to output the optimal machining parameter adjustment amount in different states, realizing the real-time optimization of the machining process; for example, when the deformation deviation is large, adjust the feed speed and spindle speed to reduce the deformation; when the machining time is too long, appropriately increase the feed speed to shorten the machining time.

[0072] In this application, in the Actor-Critic network, the Actor-Critic network includes an Actor network and a Critic network; the output of the Actor network is the adjustment amount of the processing parameters, the input of the Critic network is the current state and the output of the Actor network, and the output is the value evaluation of the current state and action. Through continuous iterative training, the Actor network can output the optimal processing parameter adjustment amount according to different states; where the current state includes deformation deviation, processing time deviation, and energy consumption deviation; it should be noted that the Actor-Critic network is mainly used in this application to intelligently optimize the processing parameters and achieve multi-objective dynamic adjustment of the processing process; according to the real-time deformation deviation, processing time deviation, and energy consumption deviation, the Actor network outputs an appropriate processing parameter adjustment amount, and the Critic network evaluates the effects of these adjustments. Through continuous iterative training, the Actor network can find the optimal adjustment scheme in various states, thereby improving the processing quality, efficiency, and reducing energy consumption.

[0073] In this application, in the multi-scale coupled dynamic simulation, every 10 iterations are performed, and the crystal orientation distribution in the microscopic model is updated once.

[0074] In this application, the Zienkiewicz-Zhu error estimation formula is: ; where represents the overall error estimation value, represents the number of grid cells, represents the error estimation value of the e-th grid cell, which is used to evaluate the accuracy of the current grid division; when the overall error estimation value is greater than its preset overall error tolerance threshold, the particle swarm optimization algorithm is triggered to perform grid encryption.

[0075] Those skilled in the art will readily conceive of other embodiments of the present invention after considering the specification and practicing the invention disclosed herein. This application is intended to cover any variations, uses, or adaptations of the present invention, which follow the general principles of the present invention and include known common knowledge or conventional technical means in the technical field not disclosed in this disclosure. The specification and examples are only considered exemplary, and the true scope and spirit of the present invention are pointed out by the following claims.

[0076] It should be understood that the present invention is not limited to the exact structures described above and shown in the drawings, and various modifications and changes can be made without departing from its scope. The scope of the present invention is only limited by the appended claims.

Claims

1. A dynamic modeling method for simulating the machining deformation of large complex structural parts, characterized in that, It includes the following steps: Step 1, construct an intelligent model and perform dynamic optimization, including establishing a 3D model, marking the deformable regions, embedding tolerance information, and adaptive mesh generation; Step 2, perform physics-based modeling driven by multi-source data, collect data and correct material properties, and construct an intelligent constitutive model; Step 3, conduct multi-scale coupled dynamic simulation to achieve macro-micro model coupling and dynamic prediction of machining forces; Step 4, perform digital twin verification and closed-loop control to quantify deformation deviation and optimize machining parameters; Step 5, implement full-process data management, establish a real-time data channel, and perform blockchain evidence storage.

2. A dynamic modeling method for machining deformation simulation of large complex structural parts according to claim 1, characterized in that, In the construction of the intelligent model and dynamic optimization, specifically: Obtain the design drawings of large complex structural parts, and use 3D modeling software to create a 3D geometric model according to the design drawings; calculate the wall thickness and curvature radius of each surface in the 3D geometric model; Use a feature recognition algorithm to mark the deformable regions of each surface in the 3D geometric model and set the marking conditions for the deformable regions; the marking conditions include surface regions where both the wall thickness and curvature radius are less than the corresponding preset thresholds; use the Kriging interpolation method to embed wall thickness fluctuation tolerance information; The adaptive mesh generation uses octree coarse meshing and sets the voxel edge, and combines the Zienkiewicz-Zhu error estimation formula and the particle swarm optimization algorithm for mesh refinement. The particle swarm velocity update formula in the particle swarm optimization algorithm is: ； Among them represents the velocity of the i-th particle at the (k + 1)-th iteration, w is the inertia weight, are the learning factors, which respectively adjust the step lengths for the particle to fly towards its own historical best position and the global best position, represents a random number uniformly distributed in the interval [0, 1]; represents the historical best position of the i-th particle, represents the global best position, represents the position of the i-th particle at the k-th iteration.

3. A dynamic modeling method for machining deformation simulation of large complex structural parts according to claim 1, characterized in that, When performing multi-source data-driven physical modeling, a dynamometer and a laser displacement sensor are used to collect cutting data and displacement data; the extended Kalman filter algorithm is used to fuse the data to correct the material properties, and the state transition matrix of the extended Kalman filter algorithm is as follows: ; where represents the sampling time of the collected data; Construct a non-linear constitutive model based on LSTM, with the input being cutting data and temperature, and the output being the stress-strain curve; Among them, the cutting data includes cutting speed and feed rate; The training process of the LSTM-based non-linear constitutive model uses the mean square error loss function, and the formula is: ; where L represents the value of the loss function, u represents the number of training samples, b represents the number of the training sample, represents the true stress-strain value of the b-th sample, is the model-predicted stress-strain value of the b-th sample, and the model parameters are optimized by minimizing the loss function.

4. A dynamic modeling method for machining deformation simulation of a large complex structural part according to claim 1, characterized in that, In the multi-scale coupled dynamic simulation, the macro-micro model includes a macroscopic model and a microscopic crystal plasticity model. The macroscopic model uses Abaqus / Explicit, and the microscopic crystal plasticity model uses VPSC. They are coupled through the homogenization theory, and the coupling formula is: , ; where represents the macroscopic stress, represents the microscopic stress, V is the volume of the material, represents the macroscopic strain, represents the microscopic strain, represents the local strain. Use the alternating iteration method to transfer stress-strain and preset the convergence accuracy; the LSTM model based on the attention mechanism pre-sets a time interval to predict cutting force fluctuations; The calculation formula for predicting the cutting force fluctuation by the LSTM model based on the attention mechanism is as follows: ; where represents the predicted cutting force, represents the attention weight of the output of the s-th LSTM hidden layer to the predicted cutting force, , and hs represents the output of the LSTM hidden layer.

5. A dynamic modeling method for machining deformation simulation of large complex structural parts according to claim 1, characterized in that In the digital twin verification and closed-loop control, industrial CT scanning is used to obtain point cloud data, and the ICP algorithm is adopted to align the point cloud data with the simulation model. The ICP algorithm formula is as follows: , ; where R represents the rotation matrix, t represents the translation vector, represents the j-th point in the point cloud data, is the corresponding point in the simulation model, represents the centroid of the point cloud data, represents the centroid of the corresponding point set in the simulation model; Then identify the out-of-tolerance regions based on the normal vector consistency; Design a multi-objective reward function: ; where r represents the reward value, represents the deformation deviation, represents the processing time deviation, represents the energy consumption deviation; online adjust the processing parameters through the Actor-Critic network.

6. A dynamic modeling method for machining deformation simulation of large complex structural parts according to claim 5, characterized in that, In the Actor-Critic network, the Actor-Critic network includes an Actor network and a Critic network; the output of the Actor network is the adjustment amount of machining parameters, the input of the Critic network is the current state and the output of the Actor network, and the output is the value evaluation of the current state and action. Through continuous iterative training, the Actor network can output the optimal adjustment amount of machining parameters according to different states; Among them, the current state includes deformation deviation, machining time deviation, and energy consumption deviation.

7. A dynamic modeling method for machining deformation simulation of large complex structural parts according to claim 1, characterized in that In the multi-scale coupled dynamic simulation, every time a set number of iterations are performed, the crystal orientation distribution in the microscopic model is updated once.

8. A dynamic modeling method for machining deformation simulation of large complex structural parts according to claim 2, characterized in that The Zienkiewicz-Zhu error estimation formula is as follows: ; where represents the overall error estimation value, represents the number of grid cells, represents the error estimation value of the e-th grid cell; when the overall error estimation value is greater than its preset overall error tolerance threshold, the particle swarm optimization algorithm is triggered to perform grid encryption.

Images