A discrete element modeling method and system for compaction behavior of printed materials

Abstract

The present invention belongs to the technical field related to the study of printing material mechanics. It discloses a discrete element modeling method and system for the compaction behavior of printing materials. The method comprises: S1, establishing a discrete element model of the compaction behavior of printing materials, determining the parameters to be optimized in the model and their ranges; establishing an objective function based on the deviation between the numerical calculation results obtained from the discrete element model simulation and the experimental measurement results; S2, establishing a distributed reinforcement learning model: designing the state space based on the objective function value and each parameter, setting a reward structure and the action space of each parameter to be optimized, wherein the reward structure is the inverse of the objective function value; S3, performing parameter inversion using the distributed reinforcement learning model, obtaining the optimized parameters, and completing the discrete element modeling. The present invention proposes optimizing and determining model parameters through parameter inversion in the discrete element model based on a distributed reinforcement learning algorithm, thereby improving the speed and efficiency of parameter inversion and making the parameter inversion more accurate.

Description

Technical Field

[0001] The present invention belongs to the technical field related to the study of printing material mechanics, and more specifically, relates to a discrete element modeling method and system for the compaction behavior of printing materials. Background Art

[0002] Sintering and melting lunar regolith powder to create lunar regolith bricks is a key technology for preparing lunar base construction materials. This method can reduce transportation costs between Earth and the Moon, maximize the utilization of extraterrestrial materials, and enable in-situ manufacturing. To ensure the smooth implementation of this technology, relevant research experiments have been carried out on Earth, including printing lunar regolith bricks using simulated lunar regolith powder and high-energy lasers. However, due to the limitations of printing technology and the unknown properties of the printed lunar regolith material, research on the mechanical properties of lunar regolith bricks is still in the preliminary stage.

[0003] Discrete element method (DEM) is a numerical simulation technique based on the modeling of discrete granular materials. It can analyze the mechanical behavior of the entire system by simulating interactions such as collision, friction, and adhesion between particles. This method has been widely used in civil engineering and material forming. The primary purpose of DEM modeling is to accurately reflect the mechanical behavior of materials during compression and failure. In order for DEM simulation models to more realistically and accurately reflect the mechanical behavior of materials, it is necessary to first determine the particle contact properties and determine the model parameters. Model parameter calibration is an important part of DEM modeling.

[0004] There are currently two methods for parameter calibration: experimental-manual calibration and automatic calibration based on optimization algorithms. The method of manually comparing experimental results to find patterns requires a lot of time and manpower costs, and the interrelated parameters of molten printing materials are usually difficult to decouple, making debugging difficult and the accuracy poor. Automatic calibration based on optimization algorithms often relies on heuristic algorithms, such as particle swarm optimization, genetic algorithm, and simulated annealing methods. These methods usually solve problems through guesswork and search, have a large search space, and are prone to falling into local optimality. Summary of the Invention

[0005] In response to the above-mentioned defects or improvement needs of the prior art, the present invention provides a discrete element modeling method and system for the compaction behavior of printed materials, which is used to solve the problem of low efficiency and accuracy of the parameter calibration algorithm in the current discrete element modeling.

[0006] To achieve the above objectives, according to one aspect of the present invention, a discrete element modeling method for compaction behavior of printed materials is provided, comprising:

[0007] S1, establishing a discrete element model of the compaction behavior of the printed material using the discrete element method, determining the parameters to be optimized in the discrete element model and their ranges; obtaining experimental measurement results of the compaction behavior of the printed material sample, and establishing an objective function with the goal of minimizing the deviation between the numerical calculation results obtained by simulating the compaction behavior of the printed material using the discrete element model and the experimental measurement results;

[0008] S2, establish a distributed reinforcement learning model: use the objective function value and the parameters to be optimized as the state space elements of the distributed reinforcement learning model, set the reward structure and the action space of the parameters to be optimized, where the reward structure is the inverse of the objective function value;

[0009] S3, solve the distributed reinforcement learning model, obtain the optimized parameters, and complete the discrete element modeling of the printing material compaction behavior; wherein, in each iteration of the distributed reinforcement learning model solution process, the action combination of the parameters to be optimized is substituted into the discrete element model to simulate the printing material compaction behavior to obtain the objective function value of each action combination.

[0010] According to the discrete element modeling method for the compaction behavior of printing materials provided by the present invention, establishing the discrete element model of the compaction behavior of printing materials in S1 specifically includes:

[0011] According to the printed material sample, a sample model in the discrete element model is established, and the particle type and particle contact properties in the sample model are determined;

[0012] Construct parameter adjustment model in discrete element model based on Hertz contact model;

[0013] The loading model in the discrete element model is established based on the compaction behavior test of the printed material sample;

[0014] The parameters in the parameter adjustment model are the parameters to be optimized in the discrete element model.

[0015] According to the discrete element modeling method for the compaction behavior of printed materials provided by the present invention, a sample model in a discrete element model is established based on a printed material sample, and the particle type and particle contact properties in the sample model are determined, specifically including:

[0016] Build a three-dimensional sample model based on the appearance and morphology of the printed material sample; the appearance and morphology of the printed material sample include: sample length, width, height, number of printing layers, melt channel height, melt channel width, and printing route;

[0017] The sample model was determined to include three different types of particles, namely, melt channel interlayer particles, melt channel interlayer particles, and melt channel particles;

[0018] Six types of particle contact properties are created, namely: melt channel particle-melt channel particle contact, inter-melt channel particle-inter-melt channel particle contact, inter-melt channel layer particle-inter-melt channel layer particle contact, melt channel particle-inter-melt channel particle contact, melt channel particle-inter-melt channel layer particle contact, and inter-melt channel layer particle-inter-melt channel particle contact.

[0019] According to the discrete element modeling method for the compaction behavior of printed materials provided by the present invention, establishing a sample model in the discrete element model based on the printed material sample further includes:

[0020] According to the CT scanning results of the printed material sample, a three-dimensional model of the cracks in the sample is created, and the three-dimensional model of the cracks is imported into the sample model by deleting particles.

[0021] According to the discrete element modeling method for the compaction behavior of printed materials provided by the present invention, the parameter adjustment model in the discrete element model is constructed based on the Hertz contact model, specifically comprising:

[0022] A nonlinear contact parameter adjustment model is constructed based on the Hertz contact model. The particle stiffness calculation formula in the nonlinear contact parameter adjustment model is:

[0023]

[0024] Among them, K n is the contact stiffness between particles, E * is the equivalent elastic modulus of the contact particles to be optimized, R * is the equivalent radius of the contact particles, δ is the overlap displacement between the contact particles, and α is the exponential parameter to be optimized; δ = βε + c, where ε is the compressive strain, and β and c are the coefficients to be optimized related to the overlap amount.

[0025] The elastic modulus E used in the calculation update is converted using the following formula:

[0026] E=K n L / A;

[0027] Where A = πr 2 , L=R (1) +R (2) , r=min(R (1) ,R (2) ), R (1) With R (2) To contact the diameters of two particles, the average radius of the particles in the group is taken in the calculation.

[0028] According to the discrete element modeling method for the compaction behavior of printed materials provided by the present invention, establishing a loading model in the discrete element model based on the compaction behavior test of the printed material sample specifically includes: establishing a loading plate in the loading model based on the test parameters and determining the loading speed of the loading plate;

[0029] Correspondingly, the compaction behavior of the printed materials in S1 and S3 is simulated by the discrete element model as follows:

[0030] The sample model is loaded and simulated according to the loading model. The real-time stress-strain relationship of the sample model is calculated during the simulation process, and the contact fracture process is recorded. At the same time, the real-time strain is taken at each calculation step of the loading simulation process, and the model is adjusted according to the real-time strain and nonlinear contact parameters to update and record the elastic modulus of the particles in the sample model, thereby realizing nonlinear update of the elastic modulus of the particles.

[0031] According to the discrete element modeling method for the compaction behavior of printed materials provided by the present invention, establishing the objective function in S1 specifically includes:

[0032] Obtaining a stress-strain curve of a printed material sample measured under a uniaxial compression test, extracting multiple strain characteristic points, and obtaining the multiple strain characteristic points measured in the test and the corresponding multiple stress values;

[0033] Obtaining a stress-strain curve obtained by numerical calculation when the printed material is subjected to uniaxial compression simulation using a discrete element model, and obtaining a plurality of numerically calculated stress values ​​corresponding to a plurality of strain characteristic points;

[0034] An objective function is established with the goal of minimizing the deviation between multiple stress values ​​calculated numerically and multiple stress values ​​measured experimentally.

[0035] According to the discrete element modeling method for the compaction behavior of printed materials provided by the present invention, the objective function F is specifically:

[0036]

[0037] Among them, α, β, E * , ... are the parameters to be optimized in the discrete element model, λ k To optimize the weight, σ k ′ is the stress value of the characteristic point calculated numerically, σ k is the stress value of the characteristic point measured in the test.

[0038] According to the discrete element modeling method of the printing material compaction behavior provided by the present invention, the distributed reinforcement learning model in S2 is specifically: the state space S t =[F,α,β,E * ,...], where F is the objective function value; the parameter value range to be optimized follows 1.5<α<9.5, 1<β<9, E * min <E * <E * max ; The action space of each parameter to be optimized is specifically:

[0039]

[0040] Among them, n is the action update step size; the reward structure R is set to the inverse of the objective function, that is,

[0041] According to another aspect of the present invention, a discrete element modeling system for the compaction behavior of printed materials is provided, the system comprising a memory and a processor, the memory storing a computer program, and the processor executing any one of the discrete element modeling methods for the compaction behavior of printed materials described above when executing the computer program.

[0042] In general, compared with the prior art, the above technical solutions conceived by the present invention provide a discrete element modeling method and system for the compaction behavior of printed materials:

[0043] 1. A distributed reinforcement learning algorithm is proposed to optimize and determine model parameters through parameter inversion in a discrete element model. The distributed reinforcement learning algorithm defines the action space of each parameter and sets a reward structure. On the one hand, it can save time and reduce manual intervention, significantly improving the speed and efficiency of parameter inversion. On the other hand, the reinforcement learning method can continuously optimize the decision-making strategy through interaction with the environment, making parameter inversion more accurate, which is conducive to improving the accuracy of parameter determination in the discrete element model and enhancing the model's predictive ability.

[0044] 2. The distributed reinforcement learning algorithm (D4PG) used collects experience in a distributed environment, efficiently utilizing multiple agents to sample in parallel, accelerating the learning process and improving the stability and robustness of the strategy.

[0045] 3. Based on the Hertz contact theory, a nonlinear Hertz-like simulation material modulus adjustment scheme, namely the nonlinear contact parameter adjustment model, which can be used in discrete element programs, is proposed. This makes the update of the particle elastic modulus during the compression behavior simulation more close to reality. This scheme can not only more reasonably show the compression behavior of printed materials such as lunar soil printing samples before uniaxial compression failure, but also more accurately simulate the deformation and failure characteristics of materials under different stress conditions, thereby achieving a high degree of consistency with actual experimental conditions; at the same time, it also provides a more reliable theoretical basis for the design and application of lunar soil printing materials, which is conducive to the development of building materials with better performance. BRIEF DESCRIPTION OF THE DRAWINGS

[0046] Figure 1 This is the overall process of discrete element modeling and parameter inversion of printing material compaction behavior provided by the present invention;

[0047] Figure 2 This is a flow chart of discrete element modeling of compaction behavior of printing materials provided by the present invention;

[0048] Figure 3 This is a flow chart of discrete element parameter inversion based on distributed reinforcement learning provided by the present invention. DETAILED DESCRIPTION

[0049] In order to make the objectives, technical solutions and advantages of the present invention more clearly understood, the present invention is further described in detail below with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are merely for the purpose of explaining the present invention and are not intended to limit the present invention. In addition, the technical features involved in the various embodiments of the present invention described below may be combined with each other as long as they do not conflict with each other.

[0050] See also Figures 1 to 3 The present invention provides a discrete element modeling method for compaction behavior of printing materials, the discrete element modeling method comprising:

[0051] S1, establishing a discrete element model of the compaction behavior of the printed material using the discrete element method, determining the parameters to be optimized in the discrete element model and their ranges; obtaining experimental measurement results of the compaction behavior of the printed material sample, and establishing an objective function with the goal of minimizing the deviation between the numerical calculation results obtained by simulating the compaction behavior of the printed material using the discrete element model and the experimental measurement results;

[0052] S2, establish a distributed reinforcement learning model: use the objective function value and the parameters to be optimized as the state space elements of the distributed reinforcement learning model, set the reward structure and the action space of the parameters to be optimized, where the reward structure is the inverse of the objective function value;

[0053] S3, solve the distributed reinforcement learning model, obtain the optimized parameters, and complete the discrete element modeling of the printing material compaction behavior; wherein, in each iteration of the distributed reinforcement learning model solution process, the action combination of the parameters to be optimized is substituted into the discrete element model to simulate the printing material compaction behavior to obtain the objective function value of each action combination.

[0054] In this embodiment, an objective function is established based on the compression behavior test data of the printed material sample, and the parameters of the discrete element model are then optimized based on the objective function. This enables the discrete element simulation model to more realistically and accurately reflect the mechanical behavior of the material. The discrete element model constructed after the parameters are determined can simulate the compression behavior of the printed material for mechanical research on the printed material.

[0055] This modeling method is based on the distributed reinforcement learning algorithm to determine the model parameters by performing parameter inversion in the discrete element model, that is, the distributed reinforcement learning algorithm generates multiple action combinations of each parameter to be optimized in each iteration, that is, generates multiple specific value combinations of each parameter, and the multiple specific value combinations of each parameter can be brought into the discrete element model to perform compression behavior simulation consistent with the experiment, and the numerical calculation results corresponding to the multiple specific value combinations of each parameter are obtained. By calculating the deviation with the experimental measurement results, the objective function values ​​corresponding to the multiple specific value combinations of each parameter can be obtained, and then the better specific value combination of each parameter is selected according to the objective function value, and then the next iterative optimization is performed based on the better specific value combination of each parameter, and finally the optimized specific value combination of each parameter is output when the end condition of the distributed reinforcement learning algorithm is met, which is the determined model parameters.

[0056] Specifically, the discrete element model of the printing material compaction behavior established in S1 includes:

[0057] According to the printed material sample, a sample model in the discrete element model is established, and the particle type and particle contact properties in the sample model are determined;

[0058] Construct parameter adjustment model in discrete element model based on Hertz contact model;

[0059] The loading model in the discrete element model is established based on the compaction behavior test of the printed material sample;

[0060] The parameters in the parameter adjustment model are the parameters to be optimized in the discrete element model.

[0061] refer to Figure 1 and Figure 2 The printing material can be lunar soil bricks or other printing materials, and there is no specific limitation. In the discrete element modeling method for the compaction behavior of printed materials provided in this embodiment, a sample model in the discrete element model is established based on the printed material sample, and the particle type and particle contact properties in the sample model are determined, specifically including:

[0062] First, the appearance dimensions and morphological characteristics of the printed material sample and the printed material product are obtained, where the appearance dimensions and morphological characteristics of the printed material sample include: sample length, width, height, number of printing layers, melt channel height, melt channel width, and printing route; based on the appearance dimensions and morphological characteristics of the printed material sample, a three-dimensional sample model is established; the sample model can be created in a discrete element particle flow program (PFC).

[0063] The sample model was determined to include three different types of particles, namely, melt layer particles, inter-melt particles, and melt particles. Specifically, based on the characteristics of laser melting printing, the printed material sample was divided into three regions according to the remelting method. The unremelted single-channel printing area was divided into the melt channel area, and the melt channel particles were used for modeling, that is, the particles in this part of the area were melt channel particles. The two melt channel remelting areas in the same plane were divided into the inter-melt channel area, and the inter-melt channel particles were used for modeling, that is, the particles in this part of the area were inter-melt channel particles. The remelting areas in the upper and lower layers of different planes were divided into the melt channel interlayer area, and the melt channel interlayer particles were used for modeling, that is, the particles in this part of the area were melt channel interlayer particles.

[0064] Next, based on the properties of different regions of the molten laser-printed material sample, six types of particle contact properties were created for the particles in the sample model: melt channel particle-melt channel particle contact, inter-melt channel particle-inter-melt channel particle contact, melt channel particle-melt channel particle contact, melt channel particle-melt channel particle contact, and inter-melt channel particle-melt channel particle contact. All contacts were initially assumed to be parallel bond models, and the calculation method was incremental. When the bond failed, the model was allowed to degenerate to a linear contact model.

[0065] Furthermore, establishing a sample model in the discrete element model based on the printed material sample also includes:

[0066] According to the CT scanning results of the printed material sample, a three-dimensional model of the cracks in the sample can be created as a crack stl file, and the three-dimensional model of the cracks is imported into the sample model by deleting particles.

[0067] Furthermore, the parameter adjustment model in the discrete element model constructed based on the Hertz contact model specifically includes:

[0068] A nonlinear contact parameter adjustment model, namely a Hertz-like model, is constructed based on the Hertz contact model. The nonlinear contact parameter adjustment model is specifically as follows:

[0069]

[0070] The particle stiffness update formula can be derived from the above formula. The particle stiffness calculation formula in the nonlinear contact parameter adjustment model is:

[0071]

[0072] Where F is the contact force between particles, K n is the contact stiffness between particles, E * is the equivalent elastic modulus of the contact particles to be optimized, R *is the equivalent radius of the contact particles, δ is the overlap displacement between the contact particles, and α is the exponential parameter to be optimized; since the particle overlap displacement in the particle flow program is difficult to calculate, it is assumed that δ is linearly related to the strain of the material compression process, and δ = βε + c, where ε is the compression strain, β and c are the coefficients to be optimized related to the overlap; c can be taken as a constant when determining the parameter optimization, so the parameters to be optimized in the discrete element model are α, β and E * c can also be used as a parameter to be optimized and added to the state space of the distributed reinforcement learning model to define the action space and value range. There is no specific limitation.

[0073] When simulating with a discrete element model in a particle flow program (PFC), the elastic modulus E used in the calculation update is converted using the following formula:

[0074] E=K n L / A;

[0075] Where A = πr 2 , L=R (1) +R (2) , r=min(R (1) ,R (2) ), R (1) With R (2) The average radius of the particles in the group is used to represent the diameter of the two contacting particles. This means that during the simulation, the model is adjusted based on the nonlinear contact parameters, and the elastic modulus of the particles is continuously updated according to the calculation step settings, making the simulation more realistic and the results more accurate and reliable.

[0076] Furthermore, establishing a loading model in the discrete element model according to the compaction behavior test of the printing material sample specifically includes: establishing a loading plate in the loading model according to the test parameters and determining a loading speed of the loading plate;

[0077] Correspondingly, the compaction behavior of the printed materials in S1 and S3 is simulated by the discrete element model as follows:

[0078] The sample model is loaded and simulated according to the loading model. The real-time stress-strain relationship of the sample model is calculated during the simulation process, and the contact fracture process is recorded. At the same time, the real-time strain is taken at each calculation step of the loading simulation process, and the model is adjusted according to the real-time strain and nonlinear contact parameters to update and record the elastic modulus of the particles in the sample model, thereby realizing nonlinear update of the elastic modulus of the particles.

[0079] In some specific embodiments, the compaction behavior of the printed material sample can be uniaxial compression, and a displacement servo compression simulation process model is established based on the uniaxial compression test process. The main simulation program includes four steps: loading plate creation, servo loading speed determination, stress-strain calculation, and contact fracture recording. The sub-program includes strain retrieval for each calculation step of the loading process, parameter update under a Hertzian-like model, and parameter recording. The sub-program is embedded in the main program using the Python interface and callback function of the particle flow program to achieve nonlinear updates of the particle modulus, thereby realizing nonlinear compaction characteristics during the compression phase.

[0080] refer to Figure 3 This embodiment provides a discrete element parameter inversion method based on distributed reinforcement learning. First, the objective function is determined. Establishing the objective function in S1 specifically includes:

[0081] Obtaining a stress-strain curve of a printed material sample measured under a uniaxial compression test, extracting multiple strain characteristic points, and obtaining the multiple strain characteristic points measured in the test and the corresponding multiple stress values;

[0082] Obtaining a stress-strain curve obtained by numerical calculation when the printed material is subjected to uniaxial compression simulation using a discrete element model, and obtaining a plurality of numerically calculated stress values ​​corresponding to a plurality of strain characteristic points;

[0083] An objective function is established with the goal of minimizing the deviation between multiple stress values ​​calculated numerically and multiple stress values ​​measured experimentally.

[0084] Specifically, based on the stress-strain curve obtained from the uniaxial compression test, multiple strain characteristic points can be extracted by first determining the strain value corresponding to the peak stress in the stress-strain curve obtained from the test measurement as the reference strain value. Strain values ​​of 20%, 40%, 60%, 80%, and 100% of the reference strain value can be taken as strain characteristic points. A strain characteristic point after sample failure is added, that is, a strain value after sample failure in the stress-strain curve is selected as the strain characteristic point. The strain value after sample failure can be the strain value corresponding to the stress after sample failure being 60% of the peak stress.

[0085] In some specific embodiments, there are 6 strain characteristic points, and the objective function F is specifically:

[0086]

[0087] Among them, α, β, E * , ... are the parameters to be optimized in the discrete element model, λ k is the optimized weight of the kth group of stress values, σ k ′ is the numerically calculated stress value of the kth characteristic point, σ k is the stress value of the kth characteristic point measured in the test.

[0088] Furthermore, the state space, action space and reward structure of the distributed reinforcement learning algorithm are defined. The distributed reinforcement learning model in S2 is specifically as follows: select the objective function value and the parameters to be optimized as the state design, that is, the state space S t =[F,α,β,E * ,...], where F is the objective function value; the parameter value range to be optimized follows 1.5<α<9.5, 1<β<9, E * min <E * <E * max The action space of each parameter to be optimized is the three different update forms designed for each parameter:

[0089]

[0090] Among them, n is the action update step size; the reward structure R is set to the inverse of the objective function, that is, i is the number of iterations.

[0091] Furthermore, the distributed reinforcement learning (D4PG) model parameters are initialized, including the training batch size, trace length, number of actor networks, replay size, exploration coefficient, learning rate, and weights of the actor and critic networks. These parameters are standard settings for distributed reinforcement learning algorithms and can be selected based on empirical values. The specific settings are not limited. Following the D4PG algorithm procedure, an initial set of parameter update strategies is generated.

[0092] Furthermore, relying on the discrete element program and Python interface, the parameter update values ​​are input into the established discrete element simulation model, the simulation curve characteristic point values ​​are calculated, and the updated objective function value is calculated. The above steps are repeated: parameter update - discrete element calculation - reinforcement learning model calculation of new parameters until the program converges or reaches the maximum iteration limit. The optimal parameter update strategy and final parameter update value are obtained. This value is applied to the discrete element model parameters. The determined discrete element model can be used in the simulation modeling of printed parts' stretching, bending, and service performance, and can more realistically simulate material behavior.

[0093] This embodiment also provides a discrete element modeling system for the compaction behavior of printed materials, the system comprising a memory and a processor, the memory storing a computer program, and the processor executing the discrete element modeling method for the compaction behavior of printed materials described in any one of the above embodiments when executing the computer program.

[0094] In some specific embodiments, the printed material can be printed lunar soil bricks. This embodiment provides a discrete element modeling method for the compaction behavior of lunar soil powder printed materials and a reinforcement learning method for parameter inversion. The method primarily includes the following steps: establishing a discrete element model based on the size, morphology, and pore characteristics of simulated lunar soil printed samples; extracting key features based on the uniaxial compression stress-strain curve of the lunar soil printed samples; adjusting inter-particle contact properties based on a Hertzian-like nonlinear contact model; establishing a parameter inversion model based on distributed reinforcement learning (D4PG); and establishing an interactive program between particle flow program (PFC) simulation and parameter inversion model to obtain optimal parameters and ensure consistency between the simulation model and experimental behavior. This embodiment lays the foundation for analyzing the microscopic contact characteristics of printed materials and the requirements of macroscopic failure simulation processes, and can also be extended to discrete element parameter calibration techniques for other materials with compaction behavior.

[0095] In response to the problems of unclear mechanical behavior of printed lunar soil bricks, difficult calibration of simulation material parameters, and poor accuracy, this embodiment provides a discrete element modeling method for the compaction behavior of lunar soil powder printed materials and a reinforcement learning method for parameter inversion. This method introduces nonlinear contact behavior between particles, explores the compaction process of lunar soil bricks during uniaxial compression, and inverts the nonlinear contact behavior parameters through a distributed reinforcement learning (D4PG) method, laying the foundation for analyzing the microscopic contact characteristics of printed materials and the requirements of macroscopic destruction simulation processes.

[0096] It will be easily understood by those skilled in the art that the above description is merely 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 scope of protection of the present invention.

Claims

1. A discrete element modeling method for compaction behavior of printed materials, characterized in that: include: S1, establishing a discrete element model of the compaction behavior of the printing material by using the discrete element method, and determining the parameters to be optimized in the discrete element model and their ranges; Obtaining experimental measurement results of the compaction behavior of the printed material sample, and establishing an objective function with the goal of minimizing the deviation between the numerical calculation results obtained by discrete element model simulation and the experimental measurement results of the compaction behavior of the printed material; S2, establish a distributed reinforcement learning model: use the objective function value and the parameters to be optimized as the state space elements of the distributed reinforcement learning model, set the reward structure and the action space of the parameters to be optimized, where the reward structure is the inverse of the objective function value; S3, solving the distributed reinforcement learning model, obtaining the optimized parameters, and completing the discrete element modeling of the printing material compaction behavior; wherein, in each iteration of the distributed reinforcement learning model, the action combination of the parameters to be optimized is substituted into the discrete element model to simulate the printing material compaction behavior to obtain the objective function value of each action combination; The discrete element model for the compaction behavior of printed materials established in S1 specifically includes: According to the printed material sample, a sample model in the discrete element model is established, and the particle type and particle contact properties in the sample model are determined; Construct parameter adjustment model in discrete element model based on Hertz contact model; The loading model in the discrete element model is established based on the compaction behavior test of the printed material sample; The parameters in the parameter adjustment model are the parameters to be optimized in the discrete element model; Based on the printed material sample, a sample model in the discrete element model is established, and the particle type and particle contact properties in the sample model are determined, including: Build a three-dimensional sample model based on the appearance and morphology of the printed material sample; the appearance and morphology of the printed material sample include: sample length, width, height, number of printing layers, melt channel height, melt channel width, and printing route; The sample model was determined to include three different types of particles, namely, melt channel interlayer particles, melt channel interlayer particles, and melt channel particles; Create six types of particle contact properties: melt channel particle-melt channel particle contact, inter-melt channel particle-inter-melt channel particle contact, inter-melt channel layer particle-inter-melt channel layer particle contact, melt channel particle-inter-melt channel particle contact, melt channel particle-inter-melt channel layer particle contact, and inter-melt channel layer particle-inter-melt channel particle contact; The parameter adjustment model in the discrete element model based on the Hertz contact model specifically includes: A nonlinear contact parameter adjustment model is constructed based on the Hertz contact model. The particle stiffness calculation formula in the nonlinear contact parameter adjustment model is: ； in, is the contact stiffness between particles, is the equivalent elastic modulus of the contact particles to be optimized, is the equivalent radius of the contact particle, is the overlapping displacement between contact particles, is the index parameter to be optimized; ,in is the compressive strain, and are the coefficients to be optimized related to the overlap amount; Calculate the elastic modulus used for the update Use the following formula for conversion: ； in, , , , and To determine the diameter of the two particles in contact, the average radius of the particles in the group is taken in the calculation; The distributed reinforcement learning model in S2 is specifically: state space , where F is the objective function value; the range of the parameter to be optimized obeys , , ; The action space of each parameter to be optimized is specifically: ， ， ； in, Update step size for actions; reward structure Set as the inverse of the objective function, that is .

2. The discrete element modeling method for the compaction behavior of printed materials according to claim 1, characterized in that: Based on the printed material sample, the sample model in the discrete element model is also established: According to the CT scanning results of the printed material sample, a three-dimensional model of the cracks in the sample is created, and the three-dimensional model of the cracks is imported into the sample model by deleting particles.

3. The discrete element modeling method for the compaction behavior of printed materials according to claim 1, characterized in that: Establishing a loading model in a discrete element model according to a compression behavior test of a printing material sample specifically includes: establishing a loading plate in the loading model according to test parameters and determining a loading speed of the loading plate; Correspondingly, the compaction behavior of the printed materials in S1 and S3 is simulated by the discrete element model as follows: The sample model is loaded and simulated according to the loading model. The real-time stress-strain relationship of the sample model is calculated during the simulation process, and the contact fracture process is recorded. At the same time, the real-time strain is taken at each calculation step of the loading simulation process, and the model is adjusted according to the real-time strain and nonlinear contact parameters to update and record the elastic modulus of the particles in the sample model, thereby realizing nonlinear update of the elastic modulus of the particles.

4. The discrete element modeling method for printing material compaction behavior according to claim 1, wherein: The objective function established in S1 specifically includes: Obtaining a stress-strain curve of a printed material sample measured under a uniaxial compression test, extracting multiple strain characteristic points, and obtaining the multiple strain characteristic points measured in the test and the corresponding multiple stress values; Obtaining a stress-strain curve obtained by numerical calculation when the printed material is subjected to uniaxial compression simulation using a discrete element model, and obtaining a plurality of numerically calculated stress values ​​corresponding to a plurality of strain characteristic points; An objective function is established with the goal of minimizing the deviation between multiple stress values ​​calculated numerically and multiple stress values ​​measured experimentally.

5. The discrete element modeling method for the compaction behavior of printed materials according to claim 4, characterized in that: Objective function Specifically: ； in, is the parameter to be optimized in the discrete element model, To optimize the weights, is the numerically calculated stress value of the characteristic point, is the stress value of the characteristic point measured in the test.

6. A discrete element modeling system for compaction behavior of printed materials, characterized in that: The system includes a memory and a processor, wherein the memory stores a computer program, and when the processor executes the computer program, the discrete element modeling method for the compaction behavior of the printing material according to any one of claims 1 to 5 is executed.

Images