High-throughput prediction method for temperature field of powder feeding type additive manufacturing based on matrix operator

Abstract

The application discloses a powder feeding type additive manufacturing temperature field high-throughput prediction method based on a matrix operator, relates to the technical field of high-throughput model calculation, and comprises the following steps: a unit volume net heat input expression is discretized in space-time through a finite difference method, a three-dimensional space discrete format and an equivalent time step are constructed, and a finite difference temperature prediction model is formed; based on the process characteristics of layer-by-layer forming of additive manufacturing, grid nodes of a current manufacturing layer and a formed layer in the finite difference temperature prediction model are set as an active calculation domain, and the grid nodes of the remaining layers are shielded; a matrix operator is constructed based on the heat conduction rules in the finite difference temperature prediction model, a matrix calculation process is formed, and the temperature field distribution is updated. The calculation method for constructing the matrix operator greatly improves the calculation efficiency of the prediction model; meanwhile, aiming at the constructed model, a prediction model super parameter trade-off method is proposed, so that the trade-off between the calculation efficiency and the prediction accuracy of the prediction model is realized.

Description

Technical Field

[0001] This invention relates to the field of high-throughput model calculation technology, and in particular to a high-throughput prediction method for temperature field in powder-feed additive manufacturing based on matrix operators. Background Technology

[0002] Powder-fed metal additive manufacturing is widely used for defect repair, surface coating enhancement, and near-net-shape forming. It has significant advantages in high-value-added fields such as aerospace, energy equipment, molds, and medical implants. It can realize on-demand material addition for complex curved surfaces and support the manufacturing of complex configuration components such as internal flow channels, lightweight topologies, and porous structures. Due to the continuous melting of metal powder by the high-energy laser heat source, huge thermal gradients and massive temperature accumulation are unavoidable during the manufacturing process. The resulting residual stress and deformation, porosity and defect formation, and anisotropy of microstructure and properties are bottlenecks restricting powder-fed metal additive manufacturing. Reasoning and analysis of the thermal history of the manufacturing process is of great significance for improving the performance and forming quality of powder-fed metal additive components.

[0003] Currently, the thermal processes of powder-fed metal additive manufacturing have evolved into two main approaches after long-term research. The first approach is based on numerical methods, which analyze the thermal field evolution through theoretical formulas, such as commercial simulation software like Ansys and COMSOL. The second approach is based on machine learning or deep learning, which uses data-driven methods to predict the evolution of the temperature field. The former mainly relies on the heat conduction equation and the energy conservation framework of phase change heat, combined with moving heat sources, convection and radiation boundary conditions, and layer-by-layer deposition geometry updates to obtain the temperature field distribution of the component during the manufacturing process. However, the computational load is huge, making it difficult to achieve engineering applications such as online simulation and offline optimization. The latter approach is data-driven, which learns the mapping relationship between process parameters, geometric paths, and temperature field evolution to achieve rapid prediction and online analysis of the thermal field. Data-driven thermal history prediction methods rely on a large amount of data to train machine learning prediction models. When the amount of data is insufficient, data-driven thermal history prediction models face a cold start problem. Summary of the Invention

[0004] In view of the aforementioned existing problems, the present invention is proposed.

[0005] Therefore, this invention provides a high-throughput prediction method for temperature field in powder-feed additive manufacturing based on matrix operators to solve the problem of insufficient computational efficiency in node-by-node indexing and difference updates of traditional finite difference models.

[0006] To solve the above-mentioned technical problems, the present invention provides the following technical solution:

[0007] This invention provides a high-throughput prediction method for temperature field in powder-feed additive manufacturing based on matrix operators. The method includes: constructing a theoretical model for predicting the thermal field in powder-feed metal additive manufacturing; setting laser Gaussian heat source terms, convective heat transfer terms, and radiative heat dissipation terms according to the heat conduction control equation to obtain a net heat input expression per unit volume; spatially and temporally discretizing the net heat input expression per unit volume using the finite difference method to construct a three-dimensional spatial discretization format and equivalent time step, forming a finite difference temperature prediction model; based on the layer-by-layer forming process characteristics of additive manufacturing, setting the mesh nodes of the current manufacturing layer and the formed layers in the finite difference temperature prediction model as the active computational domain, and shielding the mesh nodes of other layers; constructing matrix operators based on the heat conduction rules in the finite difference temperature prediction model to form a matrix-based computational flow and updating the temperature field distribution; adjusting the spatial resolution and the number of active layers of network nodes in the finite difference temperature prediction model to establish an accuracy-efficiency tradeoff objective function, and minimizing it through a parameter optimization algorithm to obtain the optimal tradeoff configuration.

[0008] As a preferred embodiment of the high-throughput prediction method for temperature field of powder-feeding additive manufacturing based on matrix operators described in this invention, the theoretical model for predicting the thermal field of powder-feeding metal additive manufacturing includes temperature conduction inside the component, laser Gaussian heat source term, thermal convection between the component and the environment, and thermal radiation of the component itself.

[0009] As a preferred embodiment of the high-throughput prediction method for temperature field of powder-feeding additive manufacturing based on matrix operators described in this invention, the net heat input expression per unit volume is composed of a laser Gaussian heat source term, a convective heat transfer term between the component and the environment, and a component thermal radiation term, wherein the laser heat source term is a positive term, and the convective heat transfer term and the radiative heat transfer term are negative terms.

[0010] As a preferred embodiment of the high-throughput prediction method for temperature field of powder-feeding additive manufacturing based on matrix operators described in this invention, wherein: the construction of a three-dimensional spatial discrete format and an equivalent time step to form a finite difference temperature prediction model refers to dividing the component in space by gridded nodes to construct a three-dimensional spatial discrete format, and constructing an equivalent time step in time by equivalent heating times per unit volume to form a finite difference temperature prediction model.

[0011] As a preferred embodiment of the high-throughput prediction method for temperature field in powder-feeding additive manufacturing based on matrix operators described in this invention, the steps of setting the mesh nodes of the current manufacturing layer and the formed layer in the finite difference temperature prediction model as the active computational domain, and masking the mesh nodes of the remaining layers, are as follows:

[0012] The finite difference temperature prediction model is divided into a manufacturing layer, an intermediate layer, and a base layer. The nodes in the forming region of the manufacturing layer are activated, while the nodes in the unforming region are shielded. As the molten pool moves, the corresponding nodes in the unforming region are gradually activated.

[0013] Set all nodes in the intermediate layer to be active, and set all nodes in the base layer to be hidden.

[0014] As a preferred embodiment of the high-throughput prediction method for temperature field of powder-feeding additive manufacturing based on matrix operators described in this invention, the matrix calculation process includes matrix operator construction and matrix process operation. The matrix operator construction includes temperature difference operation operators for calculating inter-layer and intra-layer temperature conduction, and boundary condition operation operators for calculating two boundary conditions: thermal convection and thermal radiation.

[0015] As a preferred embodiment of the high-throughput prediction method for temperature field of powder-feeding additive manufacturing based on matrix operators described in this invention, wherein: the temperature difference operation operator for intra-layer temperature conduction is performed by dividing network nodes in the same manufacturing layer into nodes in the direction of conduction tail, nodes in the direction of conduction middle and nodes in the direction of conduction head according to the conduction direction, and performing matrix-based difference operation on the temperature matrix based on the filling transformation matrix, the copying transformation matrix and the reduction matrix, and calculating the temperature difference between adjacent network nodes in the same layer;

[0016] The temperature difference operation operator for interlayer temperature conduction performs matrix-based difference operations on the temperature matrices of adjacent manufacturing layers to facilitate temperature transfer between different layers.

[0017] As a preferred embodiment of the high-throughput prediction method for temperature field of powder-feeding additive manufacturing based on matrix operators described in this invention, the boundary condition operation operator is to construct a screening matrix to screen network nodes located on the outer surface of the component, and to perform thermal convection and thermal radiation calculations on the network nodes based on the screening matrix, and to participate in the temperature field update calculation in the form of a matrix of thermal convection and thermal radiation terms.

[0018] As a preferred embodiment of the high-throughput prediction method for temperature field of powder-feeding additive manufacturing based on matrix operators described in this invention, the method of adjusting the spatial resolution and the number of activated layers of the network nodes of the finite difference temperature prediction model includes setting the resolution of the finite difference temperature prediction model and the number of activated layers of the network nodes as adjustable parameters, obtaining the model prediction accuracy evaluation index based on the comparison results between the predicted temperature field and the actual manufacturing temperature field, and simultaneously determining the time required for the theoretical model of powder-feeding metal additive manufacturing thermal field prediction to infer a complete manufacturing process.

[0019] As a preferred embodiment of the high-throughput prediction method for temperature field of powder-feeding additive manufacturing based on matrix operators described in this invention, the step of establishing an accuracy-efficiency trade-off objective function and minimizing it through a parameter optimization algorithm to obtain the optimal trade-off configuration refers to establishing an accuracy-efficiency trade-off objective function composed of prediction error terms and computational overhead terms based on the model prediction accuracy evaluation index and the theoretical model for predicting the thermal field of powder-feeding metal additive manufacturing, and minimizing it through a parameter optimization algorithm to obtain the optimal trade-off configuration.

[0020] The beneficial effects of this invention are as follows: by constructing a matrix operator, a matrix calculation process for the temperature field distribution of powder-feeding metal additive components is proposed, which improves the calculation efficiency of the finite difference temperature prediction model while ensuring the physical rules of the model; at the same time, for the finite difference temperature prediction model, an intelligent prediction model hyperparameter trade-off method is proposed to achieve a trade-off between the calculation efficiency and prediction accuracy of the finite difference temperature prediction model. Attached Figure Description

[0021] To more clearly illustrate the technical solutions of the embodiments of the present invention, the drawings used in the following description of the embodiments will be briefly introduced. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0022] Figure 1 This is a flowchart of a high-throughput prediction method for the temperature field in powder-feed additive manufacturing based on matrix operators.

[0023] Figure 2 A schematic diagram showing the classification of mesh nodes in the direction of temperature conduction.

[0024] Figure 3 A schematic diagram of the matrix operator construction for calculating intralayer temperature difference.

[0025] Figure 4 A flowchart for building and calibrating the parameters of a finite difference temperature prediction model under a high-throughput matrix calculation process. Detailed Implementation

[0026] To make the above-mentioned objects, features and advantages of the present invention more apparent and understandable, the specific embodiments of the present invention will be described in detail below with reference to the accompanying drawings.

[0027] Many specific details are set forth in the following description in order to provide a full understanding of the invention. However, the invention may also be practiced in other ways different from those described herein, and those skilled in the art can make similar extensions without departing from the spirit of the invention. Therefore, the invention is not limited to the specific embodiments disclosed below.

[0028] Secondly, the term "one embodiment" or "embodiment" as used herein refers to a specific feature, structure, or characteristic that may be included in at least one implementation of the present invention. The phrase "in one embodiment" appearing in different places in this specification does not necessarily refer to the same embodiment, nor is it a single or selective embodiment that is mutually exclusive with other embodiments.

[0029] Reference Figures 1-4 As one embodiment of the present invention, this embodiment provides a high-throughput prediction method for temperature field in powder-feed additive manufacturing based on matrix operators, comprising the following steps:

[0030] S1. Construct a theoretical model for predicting the thermal field in powder-feeding metal additive manufacturing, and set up laser Gaussian heat source terms, convective heat transfer terms, and radiative heat dissipation terms according to the heat conduction control equation to obtain the expression for net heat input per unit volume.

[0031] Specifically, the theoretical model for predicting the thermal field in powder-feeding metal additive manufacturing includes temperature conduction inside the component, laser Gaussian heat source term, thermal convection between the component and the environment, and thermal radiation of the component itself.

[0032] The expression for net heat input per unit volume consists of a laser Gaussian heat source term, a convective heat transfer term between the component and the environment, and a component thermal radiation term, where the laser heat source term is positive, and the convective heat transfer term and the radiative heat transfer term are negative.

[0033] Furthermore, a numerical model for predicting the thermal field is constructed for the powder-feeding metal additive manufacturing process, and a formula for component temperature diffusion is derived based on the thermal diffusion model.

[0034] The expression for component temperature diffusion is:

[0035] ;

[0036] in, Indicates the temperature of the thermal field. This represents the net heat input per unit volume, where P is the equivalent laser heat source input power for powder-feeding metal additive manufacturing. Thermal conductivity, For thermal diffusivity, , , For physical world coordinates, The symbol represents the partial derivative.

[0037] From the perspective of temperature field evolution mechanism, the component temperature diffusion formula , and This can be viewed as the thermal diffusion term of the component, reflecting the spatial diffusion process of the thermal field. It is the net heat exchange term between the component and the environment, reflecting the thermal energy interaction between the component and the environment per unit time. It is the component thermal field change term, which reflects the instantaneous increment of the thermal field.

[0038] In the powder-feeding metal additive manufacturing process, the main heat evolution mechanisms include the heat energy input of the laser heat source, the temperature transfer caused by the internal temperature gradient of the component, the heat convection between the component and the environment, and the heat radiation of the component. The net heat exchange term per unit volume of the component is obtained through the component temperature diffusion formula.

[0039] The expression for net heat input per unit volume is:

[0040] ;

[0041] in, The equivalent laser heat source input has a Gaussian distribution shape. The width of the melt channel, The distance from the current node to the center of the equivalent heat source. For heat convection dissipation between component medium and ambient medium, The convective heat transfer coefficient is... For the ambient medium temperature, Heat dissipation due to component thermal radiation. For surface emissivity, The Stefan-Boltzmann constant. denoted as the laser energy absorption coefficient.

[0042] Since the equivalent laser heat source input is the input energy into the component, its sign is positive. Conversely, the other two terms are energy dissipation processes, and their signs are negative.

[0043] S2. The net heat input expression per unit volume is spatially and temporally discretized using the finite difference method to construct a three-dimensional spatial discretization format and an equivalent time step, thus forming a finite difference temperature prediction model.

[0044] Specifically, the components are divided into three-dimensional spatial discrete formats by using gridded nodes in space, and an equivalent time step is constructed by using the equivalent number of heating times per unit volume in time, thus forming a finite difference temperature prediction model.

[0045] Furthermore, the component temperature diffusion formula and the net heat input per unit volume expression are discretized using the finite difference method to derive the spatial heat conduction-diffusion term and the net heat input term, which are expressed as follows:

[0046] ;

[0047] in, for Second-order central difference in direction, for Second-order central difference in direction, for Second-order central difference of direction, using nodes , , With the central node A six-neighborhood difference template is constructed to characterize the heat conduction transfer between this node and its neighboring nodes. for Grid spacing in the direction, for Grid spacing in the direction, for Grid spacing in the direction, Indicates equivalent heat source input, The heat term per unit volume representing convective heat transfer. This represents the heat term per unit volume of radiative heat transfer, expressed through the thermal conductivity. After normalization, it becomes heat flux density. Let be the first-order time-difference term of the heat flux density, representing the rate of change of the node temperature within adjacent time steps, where For thermal diffusivity, For discrete time steps, express The index number of the grid node in the direction. express The index number of the grid node in the direction. express The index number of the grid node in the direction. The index of the discrete time step representing the time direction.

[0048] The spatial discretization method is uniform discretization, meaning the network spacing is the same. The formula for uniform spatial discretization can be expressed as:

[0049] ;

[0050] ;

[0051] .

[0052] In the finite difference thermal field solution, the equivalent time discretization method is used to describe the scanning process of the moving heat source, defining... This represents the number of time steps required for a heat source to move from the current grid node to an adjacent grid node; in continuous... Within each time step, the equivalent heat source input is applied to the same grid node to complete the process. After the update, the heat source location is moved to the next grid node. The time-space mapping relationship between the equivalent number of heating cycles and the scanning speed is as follows:

[0053] ;

[0054] in, For discrete time steps, This represents the scanning speed during the actual manufacturing process. For the computational domain in Discrete spatial step size of the direction.

[0055] S3. Based on the process characteristics of additive manufacturing layer-by-layer forming, the mesh nodes of the current manufacturing layer and the formed layer in the finite difference temperature prediction model are set as the active computation domain, and the mesh nodes of the remaining layers are shielded.

[0056] The finite difference temperature prediction model is divided into a manufacturing layer, an intermediate layer, and a base layer. The nodes in the forming region of the manufacturing layer are activated, while the nodes in the unforming region are shielded. As the molten pool moves, the corresponding nodes in the unforming region are gradually activated.

[0057] Specifically, the finite difference temperature prediction model is divided into a manufacturing layer, an intermediate layer (several layers close to the manufacturing layer), and a base layer (several layers far from the manufacturing layer). The manufacturing layer is the top layer of the component. In the manufacturing layer, the laser heat source melts the powder and moves the molten pool. As the manufacturing process progresses, the solidified material continuously increases. In the manufacturing layer, grid nodes are continuously generated along with the moving molten pool. The material that has already been manufactured also serves as an active node.

[0058] Set all nodes in the intermediate layer to be active, and set all nodes in the base layer to be hidden.

[0059] Specifically, the intermediate layer consists of several layers below the manufacturing layer. Since the intermediate layer does not involve the addition of material, the number and position of the mesh nodes are fixed, and all mesh nodes are in an active state. The base layer is located below the intermediate layer. Because it is far from the heat source of the top layer, the internal temperature gradient is relatively low and the temperature distribution is more uniform. No active network nodes are set for the base layer, and all network nodes have the same temperature. The base layer is treated as a whole. Its overall temperature is summarized based on the heat flow that interacts with it, and the average value is used for temperature updates.

[0060] S4. Construct matrix operators based on the heat conduction rules in the finite difference temperature prediction model to form a matrix-based calculation process and update the temperature field distribution.

[0061] The matrix-based computation process includes matrix operator construction and matrix process operation. Matrix operator construction includes temperature difference operation operators for calculating inter-layer and intra-layer temperature conduction, as well as boundary condition operation operators for calculating two types of boundary conditions: thermal convection and thermal radiation.

[0062] Specifically, for the manufacturing layer and intermediate layer, due to the existence of boundary conditions in the network nodes of the component, the construction of the temperature difference matrix needs to comply with the corresponding physical temperature transfer rules. The positional relationship between the network nodes is divided into six conduction directions: upward, downward, leftward, rightward, forward, and backward. Among them, the leftward, rightward, forward, and backward directions are the same-layer temperature conduction, while the upward and downward directions are the cross-layer temperature conduction. This operator involves three cases: intra-layer temperature transfer, inter-layer temperature transfer, and matrix operation of boundary conditions.

[0063] The temperature difference calculation operator for intralayer temperature conduction divides network nodes within the same manufacturing layer into nodes in the conduction tail direction, nodes in the conduction middle section, and nodes in the conduction head direction according to the conduction direction. Based on the filling transformation matrix, copying transformation matrix, and reduction matrix, the operator performs matrix-based difference calculation on the temperature matrix to calculate the temperature difference between adjacent network nodes in the same layer.

[0064] Specifically, when calculating the temperature difference within the same layer, for nodes that partially perform temperature transfer due to boundary condition interference, the nodes are first divided into three types: tail-direction nodes, mid-direction nodes, and head-direction nodes. Simultaneously, the four temperature transfer directions in the x and y directions are decomposed, and calculations are performed for each direction. In the defined transfer directions, tail-direction nodes only transfer heat to adjacent nodes along the direction of transfer, without transferring heat themselves. Head-direction nodes only receive heat from adjacent nodes and do not transfer heat themselves. Mid-direction nodes, according to the temperature transfer direction, receive upstream temperature transfer while simultaneously transferring temperature to downstream nodes. The specific process can be seen in [the diagram]. Figure 2 Four types of temperature conduction normal vectors ( The corresponding node classification and matrix encoding are based on the direction of conduction of the head and tail nodes according to the direction of the normal vector, and the internal region is defined as the middle node, which is used to construct the same-layer temperature difference matrix operator.

[0065] Temperature difference in this direction is achieved through matrix operations. A filling transformation matrix is ​​constructed, and multiplication is performed between it and the temperature distribution matrix. This inserts a vector with arbitrary values ​​at the beginning of the temperature distribution matrix along the conduction direction. This transformed matrix is ​​named the filling matrix. Next, a copy transformation matrix is ​​constructed. Multiplication is performed between this copy transformation matrix and the temperature distribution matrix, copying the tail vector of the temperature matrix to the tail of the matrix. This transformed matrix is ​​named the copy matrix. Subtraction is performed between the filling matrix and the copy matrix. Since the copy matrix is ​​in the last column, it naturally achieves temperature gradient difference between adjacent nodes. Finally, a subtraction matrix is ​​constructed to eliminate the head vector of the matrix, achieving difference between adjacent network nodes along a conduction direction while ensuring it is unaffected by boundary conditions. The specific process can be found in [link to documentation]. Figure 3 A schematic diagram of the matrix operator construction for intra-layer temperature difference calculation: (a) represents (b) represents the temperature difference matrix operator in the same layer along the axial conduction direction; Temperature difference matrix operator in the same layer along the axial conduction direction.

[0066] It should be noted that the temperature conduction normal vector is When the direction is specified, its temperature difference matrix operator can be expressed as:

[0067] ;

[0068] in, For matrix operators, To fill the transformation matrix, To copy the transformation matrix, To reduce the matrix, The temperature distribution is in matrix form.

[0069] When the direction of the normal vector is In the positive direction of the axis, , , The matrix is:

[0070] ;

[0071] ;

[0072] .

[0073] in, Represents the real number field. Represents a set of real numbers 3D matrix set for The number of grid nodes along the discrete axis. for An identity matrix of dimension 1.

[0074] When the direction of the normal vector is In the negative direction of the axis, , , The matrix is:

[0075] ;

[0076] ;

[0077] .

[0078] The temperature conduction normal vector is When the direction is specified, its temperature difference matrix operator can be expressed as:

[0079] .

[0080] When the direction of the normal vector is In the positive direction of the axis, , , The matrix is:

[0081] ;

[0082] ;

[0083] ;

[0084] in, Represents the real number field. Represents a set of real numbers 3D matrix set for The number of grid nodes along the discrete axis. for An identity matrix of dimension 1.

[0085] When the direction of the normal vector is In the negative direction of the axis, , , The matrix is:

[0086] ;

[0087] ;

[0088] .

[0089] The temperature difference operation operator for interlayer temperature conduction realizes the heat transfer between the current manufacturing layer and the adjacent formed layer below at the same plane position by performing matrix difference operation on the temperature matrix of adjacent manufacturing layers.

[0090] Specifically, apart from the manufacturing layer, the influence of boundary conditions does not need to be considered for interlayer temperature transfer. Therefore, the temperature transfer can be solved simply by using the temperature matrix difference in matrix form.

[0091] Since the temperature matrices have the same size, they can be directly expressed using matrix calculation methods, and the expression is as follows:

[0092] .

[0093] Because the top manufacturing layer only transfers heat downwards. The transfer of temperature along the axial direction is represented as:

[0094] .

[0095] The boundary condition operation operator is to construct a screening matrix to screen network nodes located on the outer surface of the component, and perform thermal convection and thermal radiation calculations on the network nodes based on the screening matrix, and participate the thermal convection and thermal radiation terms in the temperature field update calculation in matrix form.

[0096] Specifically, matrix operations for boundary conditions include thermal radiation calculations and thermal convection calculations. Both can be viewed as interactions between the outermost nodes of the component and the environment. Therefore, it is necessary to construct a matrix operator capable of filtering node position coordinates. For the contour nodes of the intermediate layer, the filtering matrix... Constructed as:

[0097] .

[0098] Thermal radiation and thermal interaction heat flux density term The expression for the joint calculation is:

[0099] .

[0100] Input energy of the laser beam It can also be converted to matrix form, as shown in the expression:

[0101] ;

[0102] in, This is a binary indicator matrix used to indicate the position of the laser beam. The nodes covered by the laser beam have a value of 1, and the remaining nodes have a value of 0. To represent the spatial position vector corresponding to each node in the finite difference grid, a set of nodes containing the moving trajectory of the powder-feeding metal additive manufacturing process is included, and a shape parameter matrix Σ is introduced to characterize the scale of the laser beam energy distribution. The heat source amplitude coefficient represents the maximum volumetric heat input intensity at the center of the laser beam. Its value is calibrated based on laser power, material absorptivity, and energy distribution scale parameters.

[0103] S5. Adjust the spatial resolution and the number of activated layers of the network nodes of the finite difference temperature prediction model to establish an accuracy-efficiency tradeoff objective function, and minimize it through a parameter optimization algorithm to obtain the optimal tradeoff configuration.

[0104] Specifically, the resolution and the number of active grid nodes of the finite difference discretization model are set as adjustable parameters. The model prediction accuracy evaluation index is obtained based on the comparison between the predicted temperature field and the actual manufacturing temperature field. At the same time, the time required for the theoretical model of powder-feed metal additive manufacturing thermal field prediction to infer a complete manufacturing process is also calculated. Based on the model prediction accuracy evaluation index and the theoretical model of powder-feed metal additive manufacturing thermal field prediction, a weight measurement function is added to construct a global error function, and the global error function is minimized through a parameter optimization adjustment algorithm.

[0105] Furthermore, the hyperparameters are the number of intermediate layers and the resolution size. The selection of model hyperparameters is achieved by comparing with the temperature field in the actual manufacturing process. In the actual manufacturing process, the thermal imager can only obtain the temperature distribution information of the top surface of the component and it is difficult to directly observe the temperature field inside the component. In order to ensure that the output is consistent with the measurable data, the top surface temperature field of the powder feeding metal additive manufacturing thermal field prediction theoretical model is used as the model prediction object in parameter adjustment and process optimization.

[0106] The model hyperparameters are used as an evaluation index for the model's prediction accuracy by minimizing the average relative error between the predicted top surface temperature field and the measured top surface temperature field during the actual manufacturing process. The expression is as follows:

[0107] ;

[0108] in, To measure the top surface temperature field during the actual manufacturing process, on the [number]th [year]... Temperature values ​​of each sampling unit In hyperparameters and The corresponding predicted temperature value output by the model below. The average relative error percentage between the prediction and the actual measurement is used to evaluate the prediction accuracy of the finite difference temperature prediction model. The smaller the value, the more accurate the prediction.

[0109] To balance the prediction accuracy and computational efficiency of the finite difference temperature prediction model, an accuracy-efficiency tradeoff objective function is constructed, the expression of which is:

[0110] ;

[0111] in, It's the model weight measurement function; increasing it... This can be seen as an evaluation that prioritizes model accuracy, while conversely, it prioritizes model efficiency. The physical world computation time required for predicting the complete manufacturing process. To calculate the time required to manufacture a component in the real physical world, a log function is used to compress the order of magnitude difference between the two terms on the right-hand side of the equation, by minimizing... Achieving a balance between prediction accuracy and computational efficiency in finite difference temperature prediction models. It is the objective function that balances accuracy and efficiency.

[0112] In summary, this invention proposes a matrix calculation process for the temperature field distribution of powder-feeding metal additive components by constructing a matrix operator, which significantly improves the computational efficiency of the prediction model while ensuring the physical regularity of the model. At the same time, for the constructed model, an intelligent prediction model hyperparameter trade-off method is proposed to achieve a trade-off between the computational efficiency and prediction accuracy of the prediction model.

[0113] It should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and are not intended to limit it. Although the present invention has been described in detail with reference to preferred embodiments, those skilled in the art should understand that modifications or equivalent substitutions can be made to the technical solutions of the present invention without departing from the spirit and scope of the technical solutions of the present invention, and all such modifications or substitutions should be covered within the scope of the claims of the present invention.

Claims

1. A high-throughput prediction method for temperature field in powder-feed additive manufacturing based on matrix operators, characterized in that: include, A theoretical model for predicting the thermal field in powder-feeding metal additive manufacturing was constructed, and the laser Gaussian heat source term, convective heat transfer term, and radiative heat dissipation term were set according to the heat conduction control equation to obtain the expression for net heat input per unit volume. The net heat input expression per unit volume is spatially and temporally discretized using the finite difference method, and a three-dimensional spatial discretization format and equivalent time step are constructed to form a finite difference temperature prediction model. Based on the process characteristics of additive manufacturing layer-by-layer forming, the mesh nodes of the current manufacturing layer and the formed layer in the finite difference temperature prediction model are set as the active computation domain, and the mesh nodes of the other layers are shielded. A matrix operator is constructed based on the heat conduction rules in the finite difference temperature prediction model to form a matrix-based calculation process and update the temperature field distribution. The matrix-based calculation process includes matrix operator construction and matrix process operation. The matrix operator construction includes temperature difference operation operators for calculating inter-layer and intra-layer temperature conduction, as well as boundary condition operation operators for calculating two boundary conditions: thermal convection and thermal radiation. The temperature difference calculation operator for intra-layer temperature conduction divides network nodes within the same manufacturing layer into nodes in the conduction tail direction, nodes in the conduction middle section, and nodes in the conduction head direction according to the conduction direction. Based on the filling transformation matrix, the copying transformation matrix, and the reduction matrix, the temperature matrix is ​​matrix-based and differentially calculated to calculate the temperature difference between adjacent network nodes in the same layer. The temperature difference operation operator for interlayer temperature conduction performs matrix-based difference operations on the temperature matrices of adjacent manufacturing layers to facilitate temperature transfer between different layers. By adjusting the spatial resolution and the number of activation layers of the network nodes in the finite difference temperature prediction model, an accuracy-efficiency tradeoff objective function is established and minimized using a parameter optimization algorithm to obtain the optimal tradeoff configuration.

2. The high-throughput prediction method for temperature field in powder-feed additive manufacturing based on matrix operators as described in claim 1, characterized in that: The theoretical model for predicting the thermal field in powder-feeding metal additive manufacturing includes temperature conduction inside the component, laser Gaussian heat source term, thermal convection between the component and the environment, and thermal radiation of the component itself.

3. The high-throughput prediction method for temperature field in powder-feed additive manufacturing based on matrix operators as described in claim 1, characterized in that: The expression for net heat input per unit volume consists of a laser Gaussian heat source term, a convective heat transfer term between the component and the environment, and a component thermal radiation term, wherein the laser heat source term is a positive term, and the convective heat transfer term and the radiative heat transfer term are negative terms.

4. The high-throughput prediction method for temperature field in powder-feed additive manufacturing based on matrix operators as described in claim 1, characterized in that: The construction of a three-dimensional spatial discrete format and an equivalent time step to form a finite difference temperature prediction model refers to dividing the components in space by gridded nodes to construct a three-dimensional spatial discrete format, and constructing an equivalent time step in time by equivalent heating times per unit volume to form a finite difference temperature prediction model.

5. The high-throughput prediction method for temperature field in powder-feed additive manufacturing based on matrix operators as described in claim 1, characterized in that: The steps for setting the mesh nodes of the current manufacturing layer and the formed layer in the finite difference temperature prediction model as the active computational domain, and masking the mesh nodes of the remaining layers, are as follows. The finite difference temperature prediction model is divided into a manufacturing layer, an intermediate layer, and a base layer. The nodes in the forming region of the manufacturing layer are activated, while the nodes in the unforming region are shielded. As the molten pool moves, the corresponding nodes in the unforming region are gradually activated. Set all nodes in the intermediate layer to be active, and set all nodes in the base layer to be hidden.

6. The high-throughput prediction method for temperature field in powder-feed additive manufacturing based on matrix operators as described in claim 1, characterized in that: The boundary condition operation operator is to construct a screening matrix to screen network nodes located on the outer surface of the component, and perform thermal convection and thermal radiation calculations on the network nodes based on the screening matrix, so that the thermal convection and thermal radiation terms participate in the temperature field update calculation in matrix form.

7. The high-throughput prediction method for temperature field in powder-feed additive manufacturing based on matrix operators as described in claim 1, characterized in that: The adjustment of the spatial resolution and the number of active layers of the network nodes in the finite difference temperature prediction model includes setting the resolution and the number of active layers of the network nodes as adjustable parameters, obtaining the model prediction accuracy evaluation index based on the comparison results between the predicted temperature field and the actual manufacturing temperature field, and determining the time required for the powder-feeding metal additive manufacturing thermal field prediction theoretical model to infer a complete manufacturing process.

8. The high-throughput prediction method for temperature field in powder-feed additive manufacturing based on matrix operators as described in claim 1, characterized in that: The establishment of the accuracy-efficiency trade-off objective function and its minimization through a parameter optimization algorithm to obtain the optimal trade-off configuration refers to establishing an accuracy-efficiency trade-off objective function composed of prediction error terms and computational overhead terms based on the model prediction accuracy evaluation index and the theoretical model for predicting the thermal field of powder-feeding metal additive manufacturing, and minimizing it through a parameter optimization algorithm to obtain the optimal trade-off configuration.

Images