Parameter prediction method and system in laser bed powder melting and solidification process

Abstract

The invention relates to the technical field of additive manufacturing and simulation, and provides a parameter prediction method and system in the powder melting and solidification process of a laser bed. According to the method, based on an established mesoscopic-scale molten pool kinetic model in the powder melting process, temperature field data of a mesoscopic molten pool and unit grid point thermodynamic data of the edge of the mesoscopic molten pool are extracted; coupling the temperature field data with a pre-established heat flow coupling model of the molded part in a macroscopic scale; establishing a dendritic crystal growth model of a microscopic scale in a solidification process based on a phase field method, and coupling the dendritic crystal growth model with thermodynamic data of unit grid points; based on the molten pool kinetic model, the heat flow coupling model and the dendritic crystal growth model, a simulation data set of a formed part corresponding to powder melting is obtained; and finally, according to the simulation data set, training, optimizing and evaluating a parameter prediction model constructed based on the MLP neural network so as to predict the residual stress and yield strength of the formed part under different process parameter combinations.

Description

Technical Field

[0001] The present application relates to the field of additive manufacturing and simulation technology, and in particular to a method and system for predicting parameters in a laser bed powder melting and solidification process. Background Art

[0002] Laser powder bed fusion technology is based on the idea of ​​computer model slicing and layer-by-layer printing. It realizes the manufacture of precision components after stacking layer by layer. It has great advantages in the design of component structures and is widely used in the fields of automobiles, biomedicine, aerospace, etc. However, in the laser powder bed fusion manufacturing process, the material undergoes repeated alternations of hot and cold, resulting in complex anisotropic properties in the microstructure. The heat transfer, mass transfer, melting, solidification and other physical phenomena in the mesoscopic molten pool will directly affect the nucleation, growth, solute segregation, phase change of microscopic grains, and the temperature field distribution and stress-strain changes of macroscopic parts, thereby causing fluctuations in the quality and mechanical properties of the formed parts. Therefore, it is particularly important to master and regulate the residual stress distribution of the molded parts, the dynamic flow behavior of the molten pool and the microstructure growth morphology, and at the same time, it can eliminate defects such as cracks, spheroidization, erosion, and pores in the formed products. Summary of the invention

[0003] The purpose of the present application is to provide a method and system for predicting parameters in a laser bed powder melting and solidification process, so as to solve or alleviate the problems existing in the above-mentioned prior art.

[0004] In order to achieve the above objectives, this application provides the following technical solutions:

[0005] The present application provides a parameter prediction method in the process of laser bed powder melting and solidification, which predicts the residual stress and yield strength of a molded part obtained by melting and solidifying powder material by laser under different process parameter combinations. The prediction method includes:

[0006] Step S101: based on the established mesoscopic-scale molten pool kinetics model during the melting process of the powder material, extract the temperature field data of the mesoscopic molten pool and the thermodynamic data of the unit grid points at the edge of the mesoscopic molten pool; step S102: couple the temperature field data with the pre-established macroscopic-scale heat-flow coupling model of the molded part; step S103: establish a microscopic-scale dendrite growth model during the solidification process based on the phase field method, and couple it with the extracted thermodynamic data of the unit grid points at the edge of the mesoscopic molten pool; step S104: based on the molten pool kinetics model, the heat-flow coupling model and the dendrite growth model, obtain a simulation data set of the molded part corresponding to the melting of the powder material; wherein the simulation data set includes a plurality of sets of residual simulated stresses, yield simulated strengths and their corresponding process parameters; step S105: according to the simulation data set, train and optimize the parameter prediction model constructed based on the MLP neural network to predict the residual stress and yield strength of the molded part under different process parameter combinations.

[0007] Preferably, in step S101, the molten pool dynamics model is:

[0008]

[0009] Among them, q l is the laser combined heat source, β is the absorption rate of the powder in the powder bed to the laser, δ is the heat source proportional coefficient, P is the laser power, ω is the laser beam action radius, r is the distance from the point within the laser beam range to the laser combined heat source q l h is the height of the laser combined heat source; e is the left and right swing frequency of the laser combined heat source; P recoil is the steam recoil pressure at the mesoscopic scale, P0 is the standard atmospheric pressure; ΔH is the enthalpy of evaporation; R is the ideal gas constant, T lv is the boiling point temperature of the powder material, T is the current temperature of the powder material during the melting process; γ(T) is the surface tension of the powder material at temperature T, γ m The powder material is at the melting temperature T m The surface tension under m is the melting temperature of the powder material.

[0010] Preferably, in step S102, the heat-fluid coupling model with the same size and grid as the molten pool dynamics model is established, and the laser combined heat source is added to the heat-fluid coupling model; the temperature field data is added to the heat-fluid coupling model in the form of a predefined field.

[0011] Preferably, in step S103,

[0012] In the dendrite growth model, the thermodynamic data of the unit grid points at the edge of the mesoscopic molten pool are applied to the microscopic temperature field in the form of a temperature gradient; wherein the microscopic temperature field control equation is:

[0013]

[0014] Wherein, T is the temperature of the melt obtained by melting the powder material during the solidification process; z is the direction perpendicular to the solid-liquid interface in the dendrite growth model; t is the solidification time of the melt obtained by melting the powder material during the solidification process; T0 is the initial temperature of the bottom area of ​​the dendrite growth model; V p is the solidification rate of the melt obtained after the powder material is melted; v is the laser scanning speed, α is the angle between the laser scanning direction and the solid-liquid interface; G(t) is the temperature gradient, T B is the temperature of the intersection of the center of the mesoscopic molten pool surface and the normal line of the mesoscopic molten pool surface, T m is the melting temperature of the powder material; L f is the melting latent heat of the powder material per unit volume; c p is the specific heat at constant pressure.

[0015] Preferably, the control equation of the dendrite growth model is:

[0016]

[0017] in,

[0018]

[0019] In the formula, is the normal vector of the diffusion interface of the melt obtained by melting the powder material during the solidification process, k is the solute distribution coefficient of the dendrite during the solidification process; z is the coordinate perpendicular to the direction of the solid-liquid interface; x and y are two directions perpendicular to each other in a plane parallel to the solid-liquid interface; V p is the solidification rate of the melt obtained after the powder material is melted; T is the thermal capillary length; is the phase field variable between the solid and liquid states in the dendrite growth model, t is the solidification time of the melt obtained after the powder material is melted; λ is the solute of the dendrite during the solidification process and the phase field variable The coupling coefficient of is the phase field variable In the x-direction, is the phase field variable Component in the y direction;

[0020] τ0 is the relaxation time, which represents the time it takes for the current phase field distribution state to reach equilibrium. W0 is the interface thickness of the solid-liquid interface. is the anisotropic factor of the powder material; is the antisolute retention term; is the interpolation function for controlling the diffusion coefficient change of the solid-liquid interface, U is the dimensionless supersaturation, c is the actual solute solubility distribution, c ∞ is the equilibrium solute solubility in the liquid phase at the boundary of the simulation region, D s is the solid phase diffusion coefficient, D l is the liquid phase diffusion coefficient;

[0021] d0 is the chemical capillary length during the solidification process of the melt obtained after the powder material is melted, and a1 and a2 are both constants; D is the diffusion rate of the melt obtained after the powder material is melted; ΔT is the temperature of the melt obtained after the powder material is melted at the current moment; Γ is the Gibbs-Thomson coefficient, m is the liquidus slope of the melt obtained after the powder material is melted, and ε6 is the anisotropic strength of the powder material; G(t) is the temperature gradient of the unit grid point thermodynamic data extracted at the edge of the mesoscopic molten pool;

[0022] is the angle between the thermal gradient direction and the preferred growth direction of the dendrite; is the kinetic coefficient of the solid-liquid interface, is the gradient energy coefficient; is the phase field variable The gradient of is the gradient of the dimensionless supersaturation U.

[0023] Preferably, in step S104, the molten pool dynamics model is calculated according to different combinations of process parameters to obtain the temperature field of the molded part at a mesoscopic scale; the heat-flux coupling model is calculated based on different scanning strategies according to the temperature field of the molded part at a mesoscopic scale to obtain the residual simulated stress of the molded part; the dendrite growth model is calculated according to the thermodynamic data of the unit grid points at the edge of the mesoscopic molten pool to obtain the yield simulated strength of the molded part.

[0024] Preferably, step S105 includes: normalizing multiple groups of residual simulated stresses, yield simulated strengths and their corresponding process parameters in the simulation data set to construct a normalized data set for optimizing the parameter prediction model; adjusting multiple hyperparameters of the parameter prediction model by orthogonal traversal method on the normalized data set until the optimal hyperparameter combination is obtained, and then optimizing the parameter prediction model based on the determination coefficient R. 2 Evaluate the optimization results.

[0025] The embodiment of the present application also provides a parameter prediction system in the process of laser bed powder melting and solidification, which predicts the residual stress and yield strength of the molded part obtained by melting and solidifying the powder material by laser under different process parameter combinations. The prediction system includes: a mesoscale unit, configured to extract the temperature field data of the mesoscale molten pool and the thermodynamic data of the unit grid points at the edge of the mesoscale molten pool based on the established mesoscale molten pool dynamics model in the process of melting the powder material; a macro coupling unit, configured to couple the temperature field data with the pre-established macroscale heat flow coupling model of the molded part; a micro coupling unit, configured to establish a solidification process based on the phase field method. A dendrite growth model at the meso-micro scale is coupled with the extracted thermodynamic data of the unit grid points at the edge of the mesoscopic molten pool; a data simulation unit is configured to obtain a simulation data set of the molded part corresponding to the melting of the powder material based on the molten pool dynamics model, the heat flow coupling model and the dendrite growth model; wherein the simulation data set includes a plurality of sets of residual simulated stresses, yield simulated strengths and their corresponding process parameters; a parameter prediction unit is configured to train and optimize the evaluation of the parameter prediction model constructed based on the MLP neural network according to the simulation data set, so as to predict the residual stress and yield strength of the molded part under different process parameter combinations.

[0026] Beneficial effects:

[0027] The parameter prediction method in the laser bed powder melting process provided in the embodiment of the present application is used to predict the residual stress and yield strength of the molded part obtained by laser powder melting and solidification under different process parameter combinations. First, based on the established mesoscopic-scale molten pool dynamics model in the powder melting process, the temperature field data of the mesoscopic molten pool and the thermodynamic data of the unit grid points at the edge of the mesoscopic molten pool are extracted; then, the temperature field data is coupled with the pre-established macroscopic-scale heat-flow coupling model of the molded part; at the same time, a microscopic-scale dendrite growth model in the solidification process is established based on the phase field method, and coupled with the extracted thermodynamic data of the unit grid points at the edge of the mesoscopic molten pool; then, based on the molten pool dynamics model, the heat-flow coupling model and the dendrite growth model, a simulation data set of the molded part corresponding to the powder melting is obtained; wherein the simulation data set includes a plurality of sets of residual simulated stresses, yield simulated strengths and their corresponding process parameters; finally, according to the simulation data set, the parameter prediction model constructed based on the MLP neural network is trained and optimized and evaluated to predict the residual stress and yield strength of the molded part under different process parameter combinations.

[0028] In this way, the problems of difficult observation of heat transfer, mass transfer, microstructure evolution, easy loss of internal information, and low matching degree between quality stability and process in the process of laser powder bed melting and solidification can be effectively solved. By establishing a multi-scale model and coupling machine learning, a comprehensive analysis and prediction of "manufacturing process-microstructure-mechanical properties" is carried out to optimize process parameters and achieve the purpose of dual control of form and properties. BRIEF DESCRIPTION OF THE DRAWINGS

[0029] The drawings constituting part of the present application are used to provide a further understanding of the present application. The exemplary embodiments and descriptions of the present application are used to explain the present application and do not constitute an improper limitation on the present application. Among them:

[0030] Figure 1 A schematic flow chart of a parameter prediction method in a laser bed powder melting and solidification process provided according to some embodiments of the present application;

[0031] Figure 2 A schematic diagram of an MLP neural network architecture provided according to some embodiments of the present application;

[0032] Figure 3 A schematic diagram of a simulation result provided according to an embodiment of the present application;

[0033] Figure 4 A schematic diagram of the structure of a parameter prediction system in a laser bed powder melting and solidification process provided according to some embodiments of the present application. DETAILED DESCRIPTION

[0034] The present application will be described in detail below with reference to the accompanying drawings and in conjunction with embodiments. Each example is provided by way of explanation of the present application and does not limit the present application. In fact, it will be clear to those skilled in the art that modifications and variations may be made in the present application without departing from the scope or spirit of the present application. For example, a feature shown or described as a part of an embodiment may be used in another embodiment to produce yet another embodiment. All other embodiments obtained by those of ordinary skill in the art based on the embodiments in the embodiments of the present invention should belong to the scope of protection of the embodiments of the present invention.

[0035] Since the laser powder bed fusion manufacturing process involves multi-scale, multi-physical field, and cross-time coupling problems, the traditional implementation has problems such as low efficiency, long cycle, difficulty in observing the evolving organization, easy loss of information, and low quality stability and process matching. Therefore, it is necessary to use numerical simulation methods, establish a multi-scale model and couple machine learning methods to conduct a comprehensive analysis and prediction of the "manufacturing process-microstructure-mechanical properties" to optimize the process parameters and achieve the purpose of dual control of shape and properties.

[0036] like Figure 1 , Figure 2As shown, the parameter prediction method in the laser bed powder melting and solidification process for predicting the residual stress and yield strength of the molded part obtained by laser powder melting and solidification under different process parameter combinations includes:

[0037] Step S101 : based on the established mesoscopic molten pool dynamics model during powder melting, extract the temperature field data of the mesoscopic molten pool and the thermodynamic data of the unit grid points at the edge of the mesoscopic molten pool.

[0038] In this application, a mesoscopic melt pool dynamics model is established in the powder melting process based on the constructed parameter database of process parameters and material parameters. In the parameter database, process parameters include: laser power, laser scanning speed, spot radius, laser scanning spacing, laser scanning direction, low temperature remelting, powder bed density, powder layer thickness, powder radius, etc.; material parameters include: density, viscosity, surface tension, thermal conductivity, melting point, etc. of powder materials.

[0039] In a specific example, Flow-3D can be used to establish a mesoscopic melt pool dynamics model during powder melting. Specifically, the powder bed can be established using EDEM, the powder particles are considered as standard spheres, and gravity and the interaction force between powders are combined. The powder diameter and position are randomly distributed after a given range, and the substrate and powder bed are coupled using the same coordinate system to build a melt pool dynamics model in Flow-3D.

[0040] Then, the powder bed model is imported from EDME into Flow-3D, and the powder bed model is divided into a gradient grid in Flow-3D. An inert gas is set above the powder layer, and material parameters, thermophysical parameters, substrate preheating temperature and boundary conditions are set. At the same time, the laser heat source, the recoil pressure generated by steam and the surface tension are coupled to the powder bed model to form a molten pool dynamics model. Among them, the laser heat source is a combined heat source, such as a combination of a Gaussian surface heat source and a Gaussian body heat source. The actual heat source shape is achieved by adjusting the heat source proportional coefficient k. Finally, the established molten pool dynamics model is:

[0041]

[0042] In the formula, q l is the laser combined heat source, β is the absorption rate of the powder in the powder bed to the laser, δ is the heat source proportional coefficient, P is the laser power, ω is the laser beam action radius, r is the distance from the point within the laser beam range to the laser combined heat source q l The distance from the center, h is the height of the laser combined heat source, and e is the left and right swing frequency of the heat source; P recoil is the steam recoil pressure at the mesoscopic scale, P0 is the standard atmospheric pressure, ΔH is the evaporation enthalpy, R is the ideal gas constant, T is the current temperature of the powder material during melting, and Tlv is the boiling point temperature of the powder material; γ(T) is the surface tension of the powder material at temperature T, γ m The melting temperature of the powder material is T m The surface tension under m is the melting temperature of the powder material. It should be noted that when the powder material melts, a keyhole phenomenon will occur when the melt reaches the evaporation point. For this, the steam recoil pressure P recoil Perform definition simulation.

[0043] Finally, the temperature field data of the mesoscopic molten pool and the thermodynamic data of the unit grid points at the edge of the mesoscopic molten pool are extracted. For example, the temperature field data of the mesoscopic molten pool and the thermodynamic data of the unit grid points at the edge of the mesoscopic molten pool are extracted through the post-processing software FlowSight provided by Flow-3D.

[0044] Step S102: coupling the temperature field data with a pre-established macro-scale thermal-fluid coupling model of the molded part.

[0045] In this application, a heat-fluid coupling model of a molded part at a macroscopic scale is established by finite element software (e.g., Abaqus), and coupled with the temperature field data at a mesoscopic scale (i.e., the temperature field data of a mesoscopic molten pool). Specifically, a heat-fluid coupling model of a molded part at a macroscopic scale with the same size and mesh as the mesoscopic molten pool dynamics model is established in the finite element software, and material parameters, thermophysical property parameters, and boundary conditions are set. A laser combined heat source is added to the established heat-fluid coupling model, and the temperature field data is added to the heat-fluid coupling model in the form of a predefined field.

[0046] Among them, since the powder bed and the substrate are both continuous media, the USDFLD subroutine can be used to identify and switch materials; the laser combined heat source can be added through the DFLUX subroutine, and the extracted temperature field data of the mesoscopic scale grid points can be added in the form of a predefined field, and the structural unit is selected as the unit category.

[0047] Step S103: establishing a microscopic dendrite growth model during solidification based on the phase field method, and coupling it with the extracted thermodynamic data of the unit grid points at the edge of the mesoscopic molten pool.

[0048] In this application, a microscopic phase field dendrite growth model is established in the solidification process based on the Ginzburg-Landau theory, and the extracted thermodynamic data of the unit grid points at the edge of the mesoscopic molten pool are applied to the microscopic dendrite growth model in the form of a temperature gradient to couple the mesoscopic molten pool dynamics model with the microscopic dendrite growth model. That is to say, in the dendrite growth model, the thermodynamic data of the unit grid points at the edge of the mesoscopic molten pool are applied to the microscopic temperature field in the form of a temperature gradient. Among them, the microscopic temperature field control equation is:

[0049]

[0050] Wherein, T is the temperature of the melt obtained by melting the powder material during the solidification process; z is the direction perpendicular to the solid-liquid interface in the dendrite growth model, t is the time of the melt obtained by melting the powder material during the solidification process, and T0 is the initial temperature of the bottom area of ​​the dendrite growth model; V p is the solidification rate of the melt obtained after the powder material is melted, v is the laser scanning speed, and α is the angle between the laser scanning direction and the solid-liquid interface;

[0051] G(t) is the temperature gradient of the thermodynamic data of the unit grid point at the edge of the mesoscopic melt pool, T B is the temperature at the intersection of the center of the mesoscopic molten pool surface and the normal line of the mesoscopic molten pool surface, T m is the melting temperature of the powder material, L f is the melting latent heat per unit volume of powder material, c p is the specific heat at constant pressure.

[0052] In this application, by introducing a phase field variable that changes continuously between 0 and 1 Represents the solid-liquid interface through the phase field variable (i.e., order parameter) characterizes whether a solid phase ( Close to 1) or liquid phase ( close to 0), or a transition state between the two. With the evolution of time and space, we can understand the growth morphology, growth rate and other characteristics of dendrites during solidification.

[0053] The diffusion of substances within the diffusion interface provides the material basis for the phase change of the solid-liquid interface, and the movement of the solid-liquid interface is restricted by the transport of substances within the diffusion interface; the solid-liquid interface is the boundary that clearly distinguishes the solid phase and the liquid phase; the diffusion interface requires that physical quantities such as the concentration of the substance or the phase field variables show a continuous change transition.

[0054] Therefore, the final control equation is the phase field control equation:

[0055]

[0056] Solute control equation:

[0057]

[0058] Antisolute interception equation:

[0059]

[0060] in,

[0061]

[0062] In the formula, is the normal vector of the diffusion interface of the melt obtained by melting the powder material during the solidification process, k is the solute distribution coefficient of the dendrite generated during the solidification process, is the order parameter (i.e., the phase field variable between the solid and liquid states in the dendrite growth model), z is the coordinate perpendicular to the solid-liquid interface, V p is the solidification rate of the melt obtained after the powder material is melted, t is the solidification time of the melt obtained after the powder material is melted, l T is the thermocapillary length; λ is the solute and phase field variables of the dendrite during solidification The coupling coefficient of

[0063] x and y are two directions perpendicular to each other in a plane parallel to the solid-liquid interface; is the phase field variable In the x-direction, is the phase field variable Component in the y direction; U is the dimensionless saturation; D is the diffusion rate of the melt obtained after the powder material is melted; is the antisolute retention term, D s is the solid phase diffusion coefficient, D l is the liquid phase diffusion coefficient, c is the solute concentration distribution; c ∞ is the solute concentration in the liquid phase at the boundary of the simulation area;

[0064] τ0 is the relaxation time, representing the phase field variable The time required to adjust from one state to another, that is, the time it takes for the phase field distribution state to reach equilibrium from the current phase field distribution state.

[0065] is the anisotropic factor of the powder material, W0 is the interface thickness of the solid-liquid interface, d0 is the chemical capillary length of the melt obtained after the powder material is melted during solidification, a1 and a2 are constants, with values ​​of 0.8839 and 0.6267 respectively, Γ is the Gibbs-Thomson coefficient, m is the liquidus slope of the melt obtained after the powder material is melted, ΔT is the temperature of the melt obtained after the powder material is melted at the current moment, G(t) is the temperature gradient of the unit grid point thermodynamic data extracted at the edge of the mesoscopic molten pool, and ε6 is the anisotropic strength of the powder material; is the angle between the temperature gradient direction and the preferred growth direction of the dendrite. Since the growth speed of the dendrite in different directions is different, the dendrite grows faster in some directions, thus forming a preferred growth direction.

[0066] To control the interpolation function of the diffusion coefficient variation at the solid-liquid interface; is the kinetic coefficient of the solid-liquid interface, is the gradient energy coefficient; is the phase field variable The gradient of is the gradient of the dimensionless supersaturation U.

[0067] At the same time, the intrinsic relationship between the yield strength of the molded part and the microscopic grain size inside the molded part is determined by coupling the Hall-Petch law:

[0068]

[0069] In the formula, σ d is the yield strength of the molded part, j d is the material constant of the powder alloy, and d is the average size of the microscopic grains inside the molded part.

[0070] Step S104: based on the molten pool dynamics model, the heat-flow coupling model and the dendrite growth model, a simulation data set of the molded part corresponding to the powder melting is obtained.

[0071] In this application, the material systems used for simulation include but are not limited to magnesium alloys, titanium alloys, stainless steel, copper alloys, zinc alloys, high entropy alloys, etc. The macroscopic and microscopic simulation results under different process parameters are used as the basis for input and output, and the process parameters (including process parameters, material parameters, scanning strategies, etc.) are used as input, and residual stress and yield strength are used as output to construct a simulation training data set.

[0072] First, according to different combinations of process parameters, the molten pool dynamics model is calculated to obtain the temperature field of the molded part at the mesoscopic scale. Specifically, different parameters such as laser power, scanning speed, spot radius, powder bed density, powder bed thickness, and powder radius are used to calculate the mesoscopic molten pool dynamics model.

[0073] Then, according to the temperature field of the molded part at the mesoscopic scale, based on different scanning strategies, the heat-fluid coupling model is calculated to obtain the residual simulated stress of the molded part. In other words, based on the temperature field data of the mesoscopic molten pool obtained by the molten pool dynamics model, the macroscopic heat-fluid coupling model is calculated using different scanning spacing, scanning direction, and low-temperature remelting parameters to analyze the residual stress (i.e., residual simulated stress) of the molded part.

[0074] Finally, based on the unit grid point thermodynamic data at the edge of the mesoscopic molten pool, the dendrite growth model is calculated to obtain the yield simulation strength of the molded part. That is, based on the unit grid point thermodynamic data at the edge of the mesoscopic molten pool obtained by the molten pool dynamics model, the micro-scale dendrite growth model is calculated and the yield strength of the molded part (i.e., yield simulation strength) is analyzed. Thus, the multi-scale model is coupled based on different process parameters using the same temperature field to make the simulation structure more accurate, so that the residual simulation stress, yield simulation strength and the corresponding process parameters are obtained as a simulation data set for training the MLP neural network model.

[0075] Step S105: According to the simulation data set, the parameter prediction model constructed based on the MLP neural network is trained and optimized and evaluated to predict the residual stress and yield strength of the molded part under different process parameter combinations.

[0076] In the present application, an MLP neural network model is established, including an input layer, a hidden layer and an output layer. Among them, multiple sets of process parameters, residual stress (i.e., residual simulated stress), and yield strength (i.e., yield simulated strength) data are collected as a simulation data set, and the data in the simulation data set are normalized. That is, multiple sets of residual simulated stress, yield simulated strength and their corresponding process parameters in the simulation data set are normalized to construct a normalized data set for optimizing the parameter prediction model based on the MLP neural network.

[0077] Then, in the normalized data, the process parameters (including laser power, scanning direction, powder layer density, etc.) are used as the input of the MLP neural network model, and the residual stress (i.e., residual simulated stress) and yield strength (i.e., yield simulated strength) are used as the output of the MLP neural network model to constitute the training set.

[0078] In the MLP neural network model, the input layer sample features are used for input, and after all nodes are activated, the output value is obtained through forward propagation. If the output value does not match the expected value, the total deviation is calculated using the output value and the expected value, and the node weight is updated through back propagation. This cycle is repeated until the output value deviation reaches the expected convergence condition, and the training of the MLP neural network model is completed.

[0079] Among them, the output sample uses the activation function to perform nonlinear transformation in the forward propagation.

[0080]

[0081] In the formula, is the output of the jth node in the lth layer of the MLP neural network model, is the activation of the jth node in the l+1th layer of the MLP neural network model; is the weight of connecting the i-th neuron in the l-th layer to the j-th neuron in the l+1-th layer in the MLP neural network model; is the bias term connecting the lth layer to the jth neuron in the l+1th layer in the MLP neural network model, is the output value of the i-th neuron in the l-th layer of the MLP neural network model, x i is the specific value of the input data of the i-th neuron;

[0082] The activation function is:

[0083]

[0084] The error back propagation process is:

[0085]

[0086] In the formula, C(w,b) is the loss function, a i is the true value, y i is the output value (i.e., predicted value) of the MLP neural network model. At the same time, by optimizing the weights and biases to minimize the loss function, the output of the MLP neural network model converges to the true value. Specifically, according to the formula:

[0087]

[0088] Optimize weights and biases.

[0089] In the formula, represents the weight of the connection from the i-th neuron in the l-th layer to the j-th neuron in the l+1-th layer at the l+1-th iteration; δ is the learning rate; represents the bias term of the jth neuron in the l+1th layer at the l+1th iteration; is the output of the jth node in the l+1th layer of the MLP neural network model.

[0090] In the process of MLP neural network model training, 80% of the collected data, i.e. the data in the simulated data set, is used as training data, and 20% is used as test data. The orthogonal traversal method is used on the normalized data set to adjust multiple hyperparameters of the parameter prediction model until the optimal hyperparameter combination is obtained by comparing the calculation time and result accuracy of the model. 2 Evaluate the optimization results. That is, the orthogonal traversal method is used on the normalized data set to train the model, and the MLP neural network models containing 1-4 hidden layers are trained separately. That is to say, for each MLP neural network model with a different number of hidden layers (1 to 4 hidden layers), they are trained on the same normalized data set, but each MLP neural network model does not share data with other MLP neural network models during training. Each MLP neural network model independently reads data from the data set, performs forward propagation, calculates the loss, and back-propagates to update the parameters.

[0091] Using the coefficient of determination R 2 As an evaluation indicator for determining the optimal number of hidden layers, the range of 1-50 is used to determine the configuration of each node in the hidden layer, and the determination coefficient R 2 The highest one is the most effective. 2 for:

[0092]

[0093] In the formula, n is the number of samples, y is i is the true value of the i-th sample, is the predicted value of the MLP neural network model for the i-th sample, is the mean of the true values ​​of n samples.

[0094] In a specific example, Figure 3 As shown in the figure, (a) is the simulation of the melt pool flow process, (b) is the melt pool shape, (c) is the residual stress distribution of the molded part, (d) is the simulation of the microscopic dendrite morphology, and (e) is the prediction and experimental comparison of residual stress and yield strength. Through simulation, the dynamic evolution process of the melt pool morphology can be reproduced, and the temperature history in the melt pool, the change law of the velocity direction, and the influence of process parameters on the melt pool morphology and temperature distribution can be revealed; the residual stress distribution of the molded part can be mastered by the scanning strategy, the microscopic dendrite morphology at different positions in the melt pool can be simulated, and the residual stress and yield strength of the molded part under different process parameter combinations can be predicted, forming an analysis and prediction scheme integrating "manufacturing process-microstructure-mechanical properties".

[0095] In this way, the multi-scale model is coupled with the same temperature field based on different process parameters, making the simulated structure more accurate. The MLP neural network is used to establish a mapping relationship between the process parameters and the residual simulated stress and yield simulated strength, simulate a series of melting and solidification phenomena and anisotropic microstructures that occur in the additive manufacturing process, and predict the residual stress and yield strength of the molded parts under the same process parameter combination, and use the determination coefficient R 2 Determining the optimal number of hidden layers and node configuration greatly improves computational efficiency and accuracy.

[0096] The present application also provides a parameter prediction system for the laser bed powder melting and solidification process. Figure 4 As shown, it is used to predict the residual stress and yield strength of the molded parts obtained by laser powder melting and solidification under different process parameter combinations. The prediction system includes:

[0097] The mesoscale unit 401 is configured to extract the temperature field data of the mesoscale molten pool and the thermodynamic data of the unit grid points at the edge of the mesoscale molten pool based on the established mesoscale molten pool dynamics model during the melting process of the powder material;

[0098] A macroscopic coupling unit 402 is configured to couple the temperature field data with a pre-established thermal-fluid coupling model of the molded part at a macroscopic scale;

[0099] A microscopic coupling unit 403 is configured to establish a microscopic dendrite growth model during solidification based on a phase field method and couple it with the extracted thermodynamic data of the unit grid points at the edge of the mesoscopic molten pool;

[0100] The data simulation unit 404 is configured to obtain a simulation data set of a molded part corresponding to the melting of the powder material based on a melt pool dynamics model, a heat flow coupling model, and a dendrite growth model; wherein the simulation data set includes a combination of multiple sets of residual simulation stresses, yield simulation strengths, and corresponding process parameters;

[0101] The parameter prediction unit 405 is configured to train and optimize the parameter prediction model constructed based on the MLP neural network according to the simulation data set, so as to predict the residual stress and yield strength of the molded part under different process parameter combinations.

[0102] The parameter prediction system for the laser bed powder melting and solidification process provided in the embodiment of the present application can implement the steps and processes of the parameter prediction method for the laser bed powder melting and solidification process of any of the above embodiments, and achieve the same technical effects, which will not be repeated here one by one.

[0103] In the description of the present invention, the terms "one embodiment", "some embodiments", "examples", "specific examples", or "some examples" etc. mean that the specific features, structures, materials or characteristics described in conjunction with the embodiment or example are included in at least one embodiment or example of the present invention. In this specification, the schematic representations of the above terms do not necessarily refer to the same embodiment or example. Moreover, the specific features, structures, materials or characteristics described may be combined in any one or more embodiments or examples in a suitable manner.

[0104] The above description is only a preferred embodiment of the present application and is not intended to limit the present application. For those skilled in the art, the present application may have various modifications and variations. Any modification, equivalent replacement, improvement, etc. made within the spirit and principle of the present application shall be included in the protection scope of the present application.

Claims

1. A method for predicting parameters in a laser bed powder melting and solidification process, characterized in that: The residual stress and yield strength of a molded part obtained by melting and solidifying a powder material by laser are predicted under different process parameter combinations. The prediction method includes: Step S101, based on the established mesoscopic molten pool dynamics model during the melting process of the powder material, extracting the temperature field data of the mesoscopic molten pool and the thermodynamic data of the unit grid points at the edge of the mesoscopic molten pool; Step S102, coupling the temperature field data with a pre-established macro-scale thermal-fluid coupling model of the molded part; Step S103, establishing a microscopic dendrite growth model during solidification based on the phase field method, and coupling it with the extracted thermodynamic data of the unit grid points at the edge of the mesoscopic molten pool; Step S104, based on the molten pool dynamics model, the heat flow coupling model and the dendrite growth model, obtaining a simulation data set of a molded part corresponding to the melting of the powder material; wherein the simulation data set includes a set of multiple groups of residual simulation stresses, yield simulation strengths and their corresponding process parameters; Step S105: According to the simulation data set, the parameter prediction model constructed based on the MLP neural network is trained and optimized and evaluated to predict the residual stress and yield strength of the molded part under different process parameter combinations.

2. The parameter prediction method in the laser powder bed melting and solidification process according to claim 1 is characterized in that: In step S101, the molten pool dynamics model is: Among them, q l is the laser combined heat source, β is the absorption rate of the powder in the powder bed to the laser, δ is the heat source proportional coefficient, P is the laser power, ω is the laser beam action radius, r is the distance from the point within the laser beam range to the laser combined heat source q l The distance from the center; h is the height of the laser combined heat source; e is the left and right swing frequency of the laser combined heat source; P recoil is the steam recoil pressure at the mesoscopic scale, P0 is the standard atmospheric pressure; ΔH is the enthalpy of evaporation; R is the ideal gas constant, T lv is the boiling point temperature of the powder material, and T is the current temperature of the powder material during the melting process; γ(T) is the surface tension of the powder material at temperature T, γ m The powder material is at the melting temperature T m The surface tension under m is the melting temperature of the powder material.

3. The parameter prediction method in the laser powder bed melting and solidification process according to claim 2 is characterized in that: In step S102, The heat-fluid coupling model with the same size and grid as the molten pool dynamics model is established, and the laser combined heat source is added to the heat-fluid coupling model; the temperature field data is added to the heat-fluid coupling model in the form of a predefined field.

4. The parameter prediction method in the laser powder bed melting and solidification process according to claim 1 is characterized in that: In step S103, In the dendrite growth model, the thermodynamic data of the unit grid points at the edge of the mesoscopic molten pool are applied to the microscopic temperature field in the form of a temperature gradient; wherein the microscopic temperature field control equation is: Wherein, T is the temperature of the melt obtained by melting the powder material during the solidification process; z is the direction perpendicular to the solid-liquid interface in the dendrite growth model; t is the solidification time of the melt obtained by melting the powder material during the solidification process; T0 is the initial temperature of the bottom area of ​​the dendrite growth model; V p is the solidification speed of the melt obtained after the powder material is melted; v is the laser scanning speed, and α is the angle between the laser scanning direction and the solid-liquid interface; G(t) is the temperature gradient, T B is the temperature of the intersection of the center of the mesoscopic molten pool surface and the normal line of the mesoscopic molten pool surface, T m is the melting temperature of the powder material; L f is the melting latent heat of the powder material per unit volume; c p is the specific heat at constant pressure.

5. The parameter prediction method in the laser bed powder melting and solidification process according to claim 4 is characterized in that: The governing equation of the dendrite growth model is: in, In the formula, is the normal vector of the diffusion interface of the melt obtained by melting the powder material during the solidification process, k is the solute distribution coefficient of the dendrite during the solidification process; z is the coordinate perpendicular to the direction of the solid-liquid interface; x and y are two directions perpendicular to each other in a plane parallel to the solid-liquid interface; V p is the solidification rate of the melt obtained after the powder material is melted; T is the thermal capillary length; is the phase field variable between the solid and liquid states in the dendrite growth model, t is the solidification time of the melt obtained after the powder material is melted; λ is the solute of the dendrite during the solidification process and the phase field variable The coupling coefficient of is the phase field variable In the x-direction, is the phase field variable Component in the y direction; τ0 is the relaxation time, which represents the time it takes for the current phase field distribution state to reach equilibrium. W0 is the interface thickness of the solid-liquid interface. are the anisotropic factors of the powder material; is the antisolute retention term; is the interpolation function for controlling the diffusion coefficient change of the solid-liquid interface, U is the dimensionless supersaturation, c is the actual solute solubility distribution, c ∞ is the equilibrium solute solubility in the liquid phase at the boundary of the simulation region, D s is the solid phase diffusion coefficient, D l is the liquid phase diffusion coefficient; d0 is the chemical capillary length during the solidification process of the melt obtained after the powder material is melted, and a1 and a2 are both constants; D is the diffusion rate of the melt obtained after the powder material is melted; ΔT is the temperature of the melt obtained after the powder material is melted at the current moment; Γ is the Gibbs-Thomson coefficient, m is the liquidus slope of the melt obtained after the powder material is melted, and ε6 is the anisotropic strength of the powder material; G(t) is the temperature gradient of the unit grid point thermodynamic data extracted at the edge of the mesoscopic molten pool; is the angle between the temperature gradient direction and the preferred growth direction of the dendrite; is the kinetic coefficient of the solid-liquid interface, is the gradient energy coefficient; is the phase field variable The gradient of is the gradient of the dimensionless supersaturation U.

6. The parameter prediction method in the laser bed powder melting and solidification process according to claim 1 is characterized in that: In step S104, According to different combinations of process parameters, the molten pool dynamics model is calculated to obtain the temperature field of the molded part at a mesoscopic scale; According to the temperature field of the molded part at the mesoscopic scale, based on different scanning strategies, the thermal-fluid coupling model is calculated to obtain the residual simulated stress of the molded part; The dendrite growth model is calculated based on the thermodynamic data of the unit grid points at the edge of the mesoscopic molten pool to obtain the yield simulation strength of the molded part.

7. The parameter prediction method in the laser bed powder melting and solidification process according to claim 1 is characterized in that: Step S105 includes: Normalizing multiple groups of residual simulated stresses, yield simulated strengths and corresponding process parameters in the simulation data set to construct a normalized data set for optimizing the parameter prediction model; The multiple hyperparameters of the parameter prediction model are adjusted on the normalized data set by the orthogonal traversal method until the optimal hyperparameter combination is obtained. 2 Evaluate the optimization results.

8. A parameter prediction system for laser bed powder melting and solidification process, characterized in that: The residual stress and yield strength of the molded parts obtained by melting and solidifying the powder material by laser under different process parameter combinations are predicted. The prediction system includes: A mesoscopic unit is configured to extract temperature field data of the mesoscopic molten pool and thermodynamic data of unit grid points at the edge of the mesoscopic molten pool based on an established mesoscopic molten pool dynamics model during the melting process of the powder material; A macroscopic coupling unit configured to couple the temperature field data with a pre-established thermal-fluid coupling model of the molded part at a macroscopic scale; A microscopic coupling unit is configured to establish a microscopic dendrite growth model during solidification based on a phase field method, and couple it with the extracted thermodynamic data of the unit grid points at the edge of the mesoscopic molten pool; A data simulation unit is configured to obtain a simulation data set of a molded part corresponding to the melting of the powder material based on the molten pool dynamics model, the heat flow coupling model and the dendrite growth model; wherein the simulation data set includes a set of multiple groups of residual simulation stresses, yield simulation strengths and corresponding process parameters; The parameter prediction unit is configured to train and optimize the parameter prediction model constructed based on the MLP neural network according to the simulation data set, so as to predict the residual stress and yield strength of the molded part under different process parameter combinations.

Images