An additive area sequential optimization method and apparatus

Abstract

The application discloses an additive area sequence optimization method and device, comprising the following steps: S1, a plurality of sub-area deposition sequences are taken as a total data set, and the total data set is divided into a training set, a prediction set and a verification set; S2, finite element simulation is performed on different area deposition sequences in the training set, and according to a simulation result, a group of nodes on a substrate surface of a three-dimensional model, which are greatly affected by the sub-area deposition sequence and are selected as key points, are stress and deformation; S3, physical characteristics related to the area sequence of the training set are taken as an input of an LSTM model, simulation stress and deformation data of the key points are taken as a result, and the LSTM model is trained; and S4, stress and displacement under a residual area sequence are predicted on the test set, and the optimal area deposition sequence is selected based on stress and displacement size and uniformity. By adopting the technical scheme of the application, the calculation efficiency is greatly improved, and the neural network model stability and generalization are improved by integrating physical characteristics.

Description

Technical Field

[0001] This invention belongs to the field of additive manufacturing technology, and particularly relates to a method and apparatus for optimizing the sequence of additive manufacturing regions. Background Technology

[0002] In arc additive manufacturing, the thermal gradient differences caused by different deposition paths have a significant impact on component forming and mechanical properties. Optimizing the deposition path is one of the most important control methods. Florent Michel et al. proposed a modular deposition path planning scheme, which integrates modular feature design into the traditional layering strategy, dividing each layer into separate deposition parts. This method allows users to adjust the path planning to the target geometry, enabling the construction of various complex geometries. By optimizing the deposition path for each region through partitioning, defects such as voids in the finished component are ensured. This idea has significant implications for additive manufacturing applications of large and complex components.

[0003] In the optimization of deposition regions in the electric arc additive manufacturing process, the corresponding stress and strain fields can be obtained through finite element calculations under different paths, and the optimal scheme can be selected based on the simulation results. However, due to the large number of regional sequence combinations, finite element calculations are very time-consuming. Currently, the neural network models applied to the problem of sub-region deposition sequence optimization lack physical interpretability, require a large amount of data for training, and are unstable with poor generalization. Summary of the Invention

[0004] The technical problem to be solved by this invention is to provide an additive region sequence optimization method and apparatus. By combining the physical characteristics of the sub-regions such as the center node and heat input under different region deposition sequences with the stress-strain field simulation results, an LSTM model is established, which greatly improves the computational efficiency and incorporates physical characteristics to improve the stability and generalization of the neural network model.

[0005] To achieve the above objectives, the present invention adopts the following technical solution:

[0006] An additive manufacturing region sequence optimization method includes:

[0007] Step S1: Divide the overall region of the two-dimensional slice of the three-dimensional model into multiple sub-regions, and at the same time, preset the deposition order of multiple sub-regions as the total dataset, and divide the total dataset into training set, prediction set and validation set.

[0008] Step S2: Perform finite element simulation on the deposition sequence of different regions in the training set. Based on the simulation results, select a set of nodes on the substrate surface of the three-dimensional model whose stress and deformation are greatly affected by the deposition sequence of the sub-regions as key points.

[0009] Step S3: Use the physical features of the training set related to the region order as input to the LSTM model, and the simulation stress and deformation data of the key nodes as the result to train the LSTM model.

[0010] Step S4: Predict the stress and displacement of the remaining regions in the test set, and select the optimal region deposition sequence based on the magnitude and uniformity of stress and displacement.

[0011] As a preferred option, in step S1, the three-dimensional model is subjected to layered path planning, and the overall area is divided into multiple sub-regions layer by layer. According to the order of the deposition process in the sub-regions, the stress and strain are minimized and evenly distributed.

[0012] Preferably, in step S3, the area and heat input of each sub-region are calculated based on the regional order and the relative position of the center node of each sub-region to the key point in the training set. The relative position, the area of ​​each sub-region and the heat input are used as feature sequences to input into the Long Short-Term Memory (LSTM) model, and the output is the stress and strain of the key point. The LSTM model is trained, and the hyperparameters of the LSTM model are adjusted on the validation set.

[0013] Preferably, in step S4, the stress and strain of each region are predicted by the prediction set, an evaluation function is constructed, and the optimal region order is selected based on the minimum function value.

[0014] The present invention also provides an additive region sequence optimization apparatus, comprising:

[0015] The partitioning module is used to divide the overall region of the 2D slice of the 3D model into multiple sub-regions. At the same time, the deposition order of multiple sub-regions is preset as the total dataset, and the total dataset is divided into training set, prediction set and validation set.

[0016] A module is selected to perform finite element simulations on different regions of the training set. Based on the simulation results, a set of nodes whose stress and deformation are significantly affected by the sub-region deposition sequence are selected as key points on the substrate surface of the three-dimensional model.

[0017] The training module is used to train the LSTM model by taking the physical features of the training set related to the region order as input and the simulated stress and deformation data of the key nodes as the result.

[0018] The optimization module is used to predict the stress and displacement of the remaining regions on the test set, and select the optimal region deposition sequence based on the magnitude and uniformity of stress and displacement.

[0019] As a preferred approach, the partitioning module performs hierarchical path planning on the 3D model, dividing the overall region into multiple sub-regions layer by layer. Based on the order of deposition processes in the sub-regions, the stress and strain are minimized and evenly distributed.

[0020] As a preferred approach, the training module calculates the area and thermal input of each sub-region based on the regional order and the relative position of the sub-region center node to the key point in the training set. The relative position, the area of ​​each sub-region, and the thermal input are used as feature sequences to input into the Long Short-Term Memory (LSTM) model, and the output is the stress and strain of the key point. The LSTM model is then trained, and the hyperparameters of the LSTM model are adjusted on the validation set.

[0021] As a preferred approach, the optimization module predicts the stress and strain of each region sequence using the prediction set, constructs an evaluation function, and selects the optimal region sequence based on the minimum function value.

[0022] By constructing an LSTM neural network surrogate model that combines finite element simulation data with physical characteristics, this invention can reasonably express the complex nonlinear relationship between region order, path order, and nodal stress and strain. Compared with the traditional multiphysics finite element calculation path selection method, this invention greatly saves costs. Compared with the pure data-driven neural network model, this model improves the stability and generalization of the neural network model and reduces the required dataset, providing theoretical guidance and technical implementation for optimizing the deposition sequence of complex structural regions. Attached Figure Description

[0023] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on the provided drawings without creative effort.

[0024] Figure 1 This is a flowchart of the additive manufacturing region sequence optimization method according to an embodiment of the present invention;

[0025] Figure 2 A schematic diagram of sub-region division and regional sequence planning;

[0026] Figure 3 This is a schematic diagram of the Length Short-Term Memory (LSTM) neural network framework. Detailed Implementation

[0027] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0028] To make the above-mentioned objects, features and advantages of the present invention more apparent and understandable, the present invention will be further described in detail below with reference to the accompanying drawings and specific embodiments.

[0029] Example 1:

[0030] like Figure 1 As shown, this embodiment of the invention provides a method for optimizing the sequence of additive manufacturing regions, used to control the residual stress of the component and the deformation of the substrate, including:

[0031] Step S1: Divide the overall region of the two-dimensional slice of the three-dimensional model into multiple sub-regions, and at the same time, preset the deposition order of multiple sub-regions as the total dataset, and divide the total dataset into training set, prediction set and validation set.

[0032] Step S2: Perform finite element simulation on different deposition sequences in the training set. Based on the simulation results, select a set of nodes on the substrate surface of the three-dimensional model whose stress and deformation are greatly affected by the deposition sequence of the sub-region as key points; use them to calculate their stress and displacement change history and determine the influence of different deposition sequences on the forming and mechanical properties of the component.

[0033] Step S3: Use the physical features of the training set related to the region order as input to the LSTM model, and the simulation stress and deformation data of the key nodes as the result to train the LSTM model.

[0034] Step S4: Predict the stress and displacement of the remaining regions in the test set, and select the optimal region deposition sequence based on the magnitude and uniformity of stress and displacement.

[0035] As one embodiment of the present invention, in step S1, the three-dimensional model is subjected to layered path planning, and the overall area is divided into multiple sub-regions layer by layer. According to the order of the deposition process of the sub-regions, the stress and strain are minimized and evenly distributed.

[0036] Furthermore, the 3D model is sliced ​​layer by layer, dividing the overall region into multiple sub-regions, such as... Figure 2 As shown, the square region is divided into 10 smaller square sub-regions, each containing 16 structured grids, and each sub-region is numbered.

[0037] As one embodiment of the present invention, in step S2, according to as follows Figure 2Finite element simulations were performed on 10 preset region sequences. The deposition process of a sub-region was calculated by simultaneously activating all meshes of that sub-region using the birth-death element method, ignoring the path order. The influence of different region sequences on the substrate stress and strain was considered. Based on the simulation results of the 10 different region sequences, four nodes (kp1-kp4) were selected on the substrate surface as a key point set kp, with coordinates represented as (kp_x, kp_y). The influence of different region sequences on the stress and strain of the key nodes was studied. The stress corresponding to key point i is σ. i , strain is u i Design evaluation function The magnitude and uniformity of stress and strain at key points are considered to evaluate the quality of the region sequence.

[0038] As one embodiment of the present invention, in step S3, the area of ​​each sub-region under each sub-region order is calculated according to the region order on the training set, and a region area sequence S is generated. Considering that the region area is linearly related to the heat input under constant heat input, the region area is used to represent the influence of heat input on stress and displacement. At the same time, the coordinate difference (Δx, Δy) between the geometric center node cp and the key node kp of each sub-region under the region order is calculated, where Δx = cp_x - kp_x, Δy = cp_y - kp_y, representing the correlation between the region deposition order and the key node. The sub-region area sequence and the coordinate difference sequence between the region geometric center point and the key point are merged as an array input to the Long Short-Term Memory (LSTM) model, and the output is the stress and strain of the key point. The LSTM model is trained, and the hyperparameters of the LSTM model are adjusted on the validation set.

[0039] Furthermore, the center points cp of each sub-region are numbered cp1-cp16, and an array is defined to store the coordinates (cp_x, cp_y) of the center nodes and the area S of the sub-region in numerical order. The relative positions (Δx, Δy) of each key point (kp_x, kp_y) with p1-p16 are calculated (where Δx = cp_x - kp_x, Δy = cp_y - kp_y). An array is defined to store the coordinate information of the key points. According to the regional order in the training set, the physical characteristics of the sub-regions are calculated, including the coordinate difference (Δx, Δy) between the geometric center node and the key point of each sub-region, and the area S of each sub-region. Then, the physical characteristics of each sub-region are arranged according to the deposition order to form a feature sequence (Δx, Δy, S), which represents the influence of the heat input movement order and the amount of input heat on the key points under the regional order.

[0040] The set of physical feature sequences from multiple regions in the training set is input into the LSTM model, such as... Figure 3 As shown, the model's forward propagation consists of multiple LSTM units with randomly initialized weights, and the loss function is... An LSTM cell comprises the following components:

[0041] f t =σ(W f ·[x t ,h t-1 ]+b f )

[0042] i t =σ(W i ·[x t ,h t-1 ]+b i )

[0043] o t =σ(W o ·[x t ,h t-1 ]+b o )

[0044] c t =f t oc t-1 +i t otanh(W c ·[x t ,h t-1 ]+b c )

[0045] h t =o t otanh(c t-1 )

[0046] Where σ is the sigmoid activation function, expressed as σ(x)=1 / (1+e -x ), tanh activation function, parameter W f W i W o W c These are the weight matrices of the hidden layers connecting the forget gate, input gate, output gate, and computation memory units of the LSTM unit at time state t-1 and time state t, respectively. Parameter b f ,b i ,b o ,b c These are the deviation matrices of the hidden layers connecting the forget gate, input gate, output gate, and computational memory units of the LSTM unit at time state t-1 and time state t, respectively. t This refers to the input of the t-th unit, h. t-1 This refers to the hidden layer vector output by the (t-1)th unit.

[0047] Train an LSTM model and output the stress and displacement under different regional sequences. On the validation set, use finite element analysis to verify the consistency between the stress and strain magnitudes at key points under the regional sequence and the predicted stress and strain, and adjust the hyperparameters.

[0048] In one embodiment of the present invention, in step S4, the stress and strain of m key points under a total of n regional deposition sequences are predicted using a prediction set, such as predicting the stress σ of key node j under the i-th regional sequence. ij and strain u ij Construct the evaluation function This represents the priority of the i-th region order, and selects the region order with the minimum evaluation function f. i_min The corresponding regional order is the optimal sedimentary region order.

[0049] In this embodiment of the invention, the overall area of ​​a two-dimensional slice of a three-dimensional model is defined as a square. This overall area is divided into multiple sub-regions, and multiple sets of region orders are preset. Finite element simulations are performed under different region orders to find a set of nodes whose stress and strain are significantly affected by the region order. These nodes are designated as the key point set. The influence of different region orders and sub-region paths on the key points is studied. An evaluation function is constructed, and the optimal region order is selected based on the stress and strain evaluation function.

[0050] Example 2:

[0051] This invention also provides an additive manufacturing region sequence optimization device, comprising:

[0052] The partitioning module is used to divide the overall region of the 2D slice of the 3D model into multiple sub-regions. At the same time, the deposition order of multiple sub-regions is preset as the total dataset, and the total dataset is divided into training set, prediction set and validation set.

[0053] A module is selected to perform finite element simulations on different regions of the training set. Based on the simulation results, a set of nodes whose stress and deformation are significantly affected by the sub-region deposition sequence are selected as key points on the substrate surface of the three-dimensional model.

[0054] The training module is used to train the LSTM model by taking the physical features of the training set related to the region order as input and the simulated stress and deformation data of the key nodes as the result.

[0055] The optimization module is used to predict the stress and displacement of the remaining regions on the test set, and select the optimal region deposition sequence based on the magnitude and uniformity of stress and displacement.

[0056] As one embodiment of the present invention, the partitioning module performs hierarchical path planning on the three-dimensional model, dividing the overall region into multiple sub-regions layer by layer, and according to the order of the deposition process in the sub-regions, the stress and strain are minimized and evenly distributed.

[0057] As one embodiment of the present invention, the training module calculates the area and heat input of each sub-region according to the regional order and the relative position of the center node of each sub-region and the key point on the training set. The relative position, the area of ​​each sub-region and the heat input are used as feature sequences to input into the Long Short-Term Memory (LSTM) model, and the output is the stress and strain of the key point. The LSTM model is trained, and the hyperparameters of the LSTM model are adjusted on the validation set.

[0058] As one embodiment of the present invention, the optimization module predicts the stress and strain of each region sequence through the prediction set, constructs an evaluation function, and selects the optimal region sequence based on the minimum function value.

[0059] The embodiments described above are merely preferred embodiments of the present invention and are not intended to limit the scope of the present invention. Various modifications and improvements made to the technical solutions of the present invention by those skilled in the art without departing from the spirit of the present invention should fall within the protection scope defined by the claims of the present invention.

Claims

1. A method for optimizing the order of additive manufacturing regions, characterized in that, include: Step S1: Divide the overall region of the two-dimensional slice of the three-dimensional model into multiple sub-regions, and at the same time, preset the deposition order of multiple sub-regions as the total dataset, and divide the total dataset into training set, prediction set and validation set. Step S2: Perform finite element simulation on the different precipitation sequences in the training set. Based on the simulation results, select a set of nodes on the substrate surface of the three-dimensional model whose stress and deformation are greatly affected by the precipitation sequence of the sub-regions as key points. Step S3: Use the physical features of the training set related to the regional sedimentation order as input to the LSTM model, and the simulation stress and deformation data of key nodes as results to train the LSTM model. Step S4: Predict the stress and displacement under the sedimentation sequence of the remaining regions on the test set, and select the optimal sedimentation sequence based on the magnitude and uniformity of stress and displacement. In step S3, based on the region order in the training set, the area of ​​each sub-region under each sub-region order is calculated to generate a region area sequence S. Considering that under constant heat input, the region area is linearly related to the heat input, the region area is used to represent the influence of heat input on stress and displacement. At the same time, the coordinate difference between the geometric center node cp and the key node kp of each sub-region under the region order is calculated. ,in, This indicates the correlation between the regional depositional sequence and key nodes. The area sequence of each sub-region, the coordinate difference sequence between the geometric center point of the region and the key point are merged into an array and input into the Long Short-Term Memory (LSTM) model. The output is the stress and strain of the key points. The LSTM model is trained, and the hyperparameters of the LSTM model are adjusted on the validation set. The center points cp of each sub-region are numbered cp1-cp16, and an array is defined to store the coordinates of the center nodes in numerical order. and sub-region area Calculate each key point Relative position to p1-p16 ,in Define an array to store keypoint coordinate information. Following the region order in the training set, calculate the physical characteristics of each sub-region, including the coordinate difference between the geometric center node and the keypoint in each sub-region. And the area S of each sub-region, and then arrange the physical features of each sub-region in the depositional order to form a feature sequence. This indicates the impact of the heat input movement sequence and the magnitude of the input heat on the key points within this region. The set of physical feature sequences from multiple regions in the training set is input into the LSTM model. The model's forward propagation consists of multiple LSTM units with randomly initialized weights. The loss function is... An LSTM unit comprises the following components: in, The sigmoid activation function is expressed as follows: (x)=1 / (1+e -x ), tanh activation function, parameter W f W i W o W c These are the weight matrices of the LSTM unit's forget gate, input gate, output gate, and computational memory connection hidden layer at time state t-1 and time state t; parameter b f ,b i ,b o ,b c These are the deviation matrices of the hidden layers connecting the forget gate, input gate, output gate, and computational memory units of the LSTM unit at time state t-1 and time state t, respectively. t This refers to the input of the t-th unit, h. t-1 This refers to the hidden layer vector output by the (t-1)th unit; Train an LSTM model to output stress and displacement in different regions in sequence.

2. The additive manufacturing region sequence optimization method as described in claim 1, characterized in that, In step S1, a layered path planning is performed on the three-dimensional model, dividing the overall area into multiple sub-regions layer by layer. Based on the order of the deposition process in the sub-regions, the stress and strain are minimized and evenly distributed.

3. The additive manufacturing region sequence optimization method as described in claim 2, characterized in that, In step S4, the stress and strain under the precipitation sequence of each region are predicted by the prediction set, an evaluation function is constructed, and the optimal precipitation sequence of the region is selected based on the minimum function value.

4. An additive region sequence optimization apparatus for implementing the additive region sequence optimization method of claim 1, characterized in that, include: The partitioning module is used to divide the overall region of the 2D slice of the 3D model into multiple sub-regions. At the same time, the deposition order of multiple sub-regions is preset as the total dataset, and the total dataset is divided into training set, prediction set and validation set. A module is selected to perform finite element simulations on different regions of the training set. Based on the simulation results, a set of nodes whose stress and deformation are greatly affected by the sub-regional deposition sequence are selected as key points on the substrate surface of the three-dimensional model. The training module is used to train the LSTM model by taking the physical features of the training set related to the regional sedimentation order as input and the simulated stress and deformation data of key nodes as results. The optimization module is used to predict the stress and displacement under the sedimentation sequence of the remaining regions on the test set, and select the optimal sedimentation sequence based on the magnitude and uniformity of stress and displacement.

5. The additive manufacturing region sequence optimization device as described in claim 4, characterized in that, The partitioning module performs hierarchical path planning on the 3D model, dividing the overall area into multiple sub-regions layer by layer. Based on the order of deposition processes in the sub-regions, the stress and strain are minimized and evenly distributed.

6. The additive region sequence optimization apparatus as described in claim 5, characterized in that, The optimization module predicts the stress and strain under the sedimentation sequence of each region using the prediction set, constructs an evaluation function, and selects the optimal sedimentation sequence of the region based on the minimum function value.

Images