Method for predicting lapping spacing of asymmetric sedimentary channels in transverse direct energy deposition

Abstract

The invention provides a transverse direct energy deposition asymmetric deposition channel lap joint distance prediction method. The method comprises the following steps: preparing an asymmetric deposition channel through a transverse direct energy deposition experiment; adopting a three-dimensional point cloud scanning device to obtain point cloud data of the deposition channel; carrying out cutting, correction and cross section slicing on the point cloud data; fitting the cross section morphology of the sedimentary channel by adopting a sedimentary channel contour self-adaptive fitting algorithm so as to obtain a fitting function expression; according to the fitting function, constructing a flat-top overlapping lap joint model, and iteratively solving a lap joint interval by adopting a dichotomy method; constructing an XGBoost regression prediction model by taking a material supply speed, a heat source moving speed and a deposition rate as input features and taking a lap joint distance as an output feature; and verifying the model precision through the root mean square error RMSE and the determination coefficient R2. According to the method, the key problem in asymmetric deposition channel lap joint distance prediction is effectively solved, and the flatness and the forming quality of a deposition layer in the transverse direct energy deposition process can be improved.

Description

Technical Field

[0001] The present invention belongs to the technical field of additive manufacturing, and in particular relates to a method for predicting the overlap spacing of asymmetric deposition paths of lateral direct energy deposition. Background Art

[0002] Metal additive manufacturing (AM) has been widely used in aerospace, marine engineering, and defense industries due to its high efficiency, low cost, and advantages in processing large-scale metal components. Metal direct energy deposition (DED) uses a laser beam or arc as a heat source and metal wire or powder as a filler material. The filler material is melted and deposited at different locations along a pre-defined deposition path to form a component.

[0003] In a typical direct energy deposition process, components are usually constructed layer by layer using a multi-pass deposition method. First, a "2.5D" structure deposition layer of a certain thickness is formed through multiple parallel depositions, and then it is stacked layer by layer in the vertical direction to finally complete the manufacturing of the component. This process places high demands on the overlap quality of the deposition paths and the surface flatness of the deposition layers. If the surface of the deposition layer is uneven, it will seriously affect the continuity of the manufacturing process and the forming quality. Therefore, accurately fitting the deposition path profile and establishing a reasonable overlap model have become the key to ensuring the high-quality forming of direct energy deposition components.

[0004] Among them, transverse direct energy deposition, as a special process, does not rely on positioners or additional support structures. Simply by adjusting the angle of the heat source, it can achieve efficient manufacturing of complex structures such as inclined and overhanging structures. However, because the molten metal in the molten pool is affected by gravity before solidification, the molten metal will flow in the direction of gravity, resulting in an asymmetric cross-section of the deposition path. This phenomenon is significantly different from the symmetrical deposition path morphology formed by traditional horizontal direct energy deposition.

[0005] Traditional overlap models for transverse direct energy deposition (DED) typically assume a symmetrical cross-section and accurately calculate the overlap spacing using an idealized mathematical relationship based on symmetric functions. However, during transverse DED, due to the influence of gravity, the molten metal in the molten pool flows along the gravity direction before fully solidifying, resulting in asymmetric cross-sections of the DED, even deviating from the functional mapping relationship. This asymmetry makes it difficult to directly apply traditional overlap models based on symmetry assumptions to the asymmetric DED DED process, and it is impossible to establish an effective mathematical relationship to calculate the overlap spacing. The lack of reliable overlap spacing guidance makes it difficult to stably form deposited layers with high surface flatness during the DED manufacturing process, which in turn affects the overall forming quality and manufacturing stability. Therefore, it is urgent to develop a method that can accurately fit the geometric characteristics of asymmetric DED cross-sections and establish a corresponding asymmetric DED overlap model to accurately predict the overlap spacing, thereby ensuring the continuity and quality stability of the DED process. Summary of the Invention

[0006] The present invention addresses the problem that the traditional overlap model based on the symmetry assumption in lateral direct energy deposition is difficult to apply to asymmetric deposition paths, and thus cannot provide a reasonable overlap spacing. A method for predicting the overlap spacing of asymmetric deposition paths in lateral direct energy deposition is proposed. This method can accurately fit asymmetric deposition paths and predict a reasonable overlap spacing, thereby improving the surface flatness and overall forming quality of the deposited layer. To achieve the above-mentioned purpose of the invention, the present invention provides a method for predicting the overlap spacing of asymmetric deposition paths in lateral direct energy deposition, which specifically includes the following steps:

[0007] Step 1: Prepare different material feeding speeds V based on lateral direct energy deposition experiments f and heat source moving speed V h Assemblage of asymmetric sedimentary tracks;

[0008] Step 2: Use a 3D point cloud scanning device to acquire 3D point cloud data for each asymmetric deposition track, crop and correct it, and then slice it along the deposition direction to generate a cross-sectional slice sequence. Subsequently, a random sampling method is used to select n cross-sectional slice sequences and merge them to characterize the cross-sectional morphology of each asymmetric deposition track.

[0009] Step 3: Process the merged cross-sectional slice data using a deposition track profile adaptive fitting algorithm to obtain a fitting function expression for the cross-sectional morphology of each asymmetric deposition track. The deposition track profile adaptive fitting algorithm is specifically as follows:

[0010] Calculate the deposition rate R of each deposition channel, R = V f / V h , where V f is the material feeding speed, V h is the heat source moving speed, and is compared with the threshold m of the deposition rate;

[0011] (1) If R<m, a differential logic function is used to fit the merged cross-sectional slice data. The differential logic function is formed by subtracting two logic functions. The specific expression is:

[0012] f(x)=g(x)-h(x)

[0013] Where:

[0014]

[0015] Where f(x) is a differential logic function, g(x) and h(x) are two logic functions with different coefficients, L1 is the maximum amplitude of g(x), k1 is the slope factor of g(x), x1 is the abscissa of the inflection point of g(x), L2 is the maximum amplitude of h(x), k2 is the slope factor of h(x), x2 is the abscissa of the inflection point of h(x), x represents the abscissa position of the input, and L1=L2, indicating that the maximum amplitudes of g(x) and h(x) are the same;

[0016] (2) If R ≥ m, first determine the segmentation point and select the point with the largest horizontal coordinate among all data points as the segmentation point, that is, (x max ,y(x max )); and divide the merged cross-sectional slice data into two subsets based on the segmentation points, specifically:

[0017] G1={(x i ,y i )|x i >x mean ,y i <y(x max )}

[0018] G2={(x i ,y i )|x i ≤x mean}∪{(x i ,y i )|x i >x mean ,y i ≥y(x max )}

[0019] Among them, G1 is subset one, G2 is subset two, x i is the horizontal coordinate value of the data point, y i is the vertical coordinate value of the data point, x mean is the mean of the horizontal coordinates of all data points, x max is the maximum value of the horizontal coordinate among all data points, y(x max ) is x max The corresponding vertical coordinate value;

[0020] (a) G1 is fitted using a differential logistic function, specifically:

[0021] f(x)=g(x)-h(x)

[0022] Where:

[0023]

[0024] Where f(x) is a differential logic function, g(x) and h(x) are two logic functions with different coefficients, L1 is the maximum amplitude of g(x), k1 is the slope factor of g(x), x1 is the abscissa of the inflection point of g(x), L2 is the maximum amplitude of h(x), k2 is the slope factor of h(x), x2 is the abscissa of the inflection point of h(x), x represents the abscissa position of the input, and L1=L2, indicating that the maximum amplitudes of g(x) and h(x) are the same;

[0025] (b) Fit G2 with a linear function, specifically:

[0026] u(x)=ax+b

[0027] Where u(x) is a linear function, a is the slope of u(x), and b is the ordinate intercept of u(x);

[0028] In order to ensure that f(x) and u(x) are continuously connected at the segmentation points, the continuity loss term is introduced:

[0029] continuity loss = (f(x max )-u(x max )) 2

[0030] The continuity loss term aims to minimize the difference between the two functions at the segmentation points, thereby ensuring that the values ​​output by f(x) and u(x) are consistent, and ultimately forming a continuous fitting function;

[0031] Step 4: Based on the fitting function expression of the cross-sectional morphology of each asymmetric deposition channel, a flat-top overlapping model is established. On this basis, the bisection method is used to solve the optimal solution of the objective function f(d), that is, the overlapping spacing d. The expression of the objective function is:

[0032] f(d)=argmin|S1(d)-S2(d)|

[0033] Where f(d) is the objective function, S1(d) and S2(d) represent the overlapping area function and the valley area function, respectively. The overlapping area function and the valley area function are obtained by integrating the functions of the corresponding areas.

[0034] The dichotomy solution process includes the following steps:

[0035] (1) Set the overlap spacing search interval [d min , d max ], where the lower bound of the interval d min = 0 indicates the critical spacing of adjacent asymmetric sedimentary tracks that completely overlap, and the upper limit of the interval d max=w represents the critical spacing between adjacent asymmetric deposition tracks without overlap, and w is the fitting width of the asymmetric deposition track;

[0036] (2) Define the convergence threshold parameter, the objective function accuracy threshold ε = 1 × 10 -3 , interval threshold δ = 1 × 10 -3 ;

[0037] (3) Calculate the midpoint value d of the current interval mid , specifically:

[0038]

[0039] (4) The calculation of the overlap spacing follows the solution steps of the bisection method, as follows:

[0040] (a) When the function convergence condition is met, that is, |f(d mid )|<ε, then the output overlap spacing d=d mid And terminate the iteration, where f(d mid ) represents the midpoint value d of the current interval mid The corresponding objective function value;

[0041] (b) When the function convergence condition is not met, the search interval is updated, specifically:

[0042] When f(d mid )·f(d min )<0, then update the search interval to [d min , d mid ]; when f(d mid )·f(d max )<0, then update the search interval to [d mid , d max ], repeat the iteration until |d is satisfied max -d min |<δ or |f(d mid )|<ε, then the output overlap spacing d=d mid , where f(d min ) represents the lower bound of the current interval d min The corresponding objective function value, f(d max ) represents the upper bound of the current interval d max The corresponding objective function value;

[0043] Step 5: Construct a material supply speed V f , heat source moving speed V h A data set with deposition rate R as input feature and overlap spacing d as output feature; the data set is normalized and 80% of it is divided into a training set and 20% is divided into a test set;

[0044] Step 6: Build an XGBoost regression model to predict the overlap distance d, and use hyperparameter optimization method to adjust the model parameters on the training set;

[0045] Step 7: Use the test set to verify the performance of the trained XGBoost regression model by calculating the root mean square error RMSE and the coefficient of determination R 2 Two key indicators are used to evaluate the model's prediction accuracy for the overlap spacing of asymmetric sedimentary tracks.

[0046] As a preferred method, the lateral direct energy deposition experiment in step 1 specifically includes:

[0047] Clamp and fix the substrate at the center of the positioner platform, then flip the positioner platform to a position perpendicular to the ground; adjust the angle of the heat source so that it is perpendicular to the substrate, and set the distance between the heat source and the substrate to 12mm (this value is the preferred setting in the experiment and can be adjusted according to specific experimental conditions in actual use); then set the material supply speed V f and heat source moving speed V h and set the heat source moving direction parallel to the ground; after starting the heat source, the energy beam acts stably on the substrate, and the material supply mechanism starts to continuously supply material, and the experimental process begins.

[0048] As a preferred embodiment, the heat source used in the lateral direct energy deposition experiment in step 1 is a laser heat source or an arc heat source.

[0049] As a preferred method, the material supply speed V in step 1 f The range is 2m / min to 7m / min, and the heat source moving speed V h The range is 0.06m / min to 0.9m / min.

[0050] As a preferred embodiment, in step 2, n is 10.

[0051] As a preferred embodiment, the threshold value m of the deposition rate in step three is 30.

[0052] As a preferred method, the parameters and their ranges for hyperparameter optimization in step 6 specifically include:

[0053] The value range of learning rate η is η∈{0.01, 0.05, 0.1, 0.2}; the value range of maximum tree depth max_depth is max_depth∈{3, 5, 7, 9}; the value range of column sampling ratio colsample_bytree is colsample_bytree∈{0.6, 0.8, 1.0}; the value range of L1 regularization parameter α is α∈{0.1, 1, 10}; the value range of L2 regularization parameter λ is λ∈{0, 0.1, 1}; the value range of maximum number of leaf nodes max_leaves is max_leaves∈{20, 30, 50}.

[0054] As a preferred method, the evaluation goal in step 7 is to minimize the root mean square error RMSE and make the coefficient of determination R 2 Close to 1; RMSE and R 2 The specific formula is as follows:

[0055]

[0056] Among them, d i is the target value of the overlap spacing, is the average value of the overlap spacing target value, is the predicted value of the overlap spacing, i=1, 2,…N, N is the total number of samples.

[0057] The significant advantages of the method of the present invention are:

[0058] 1. Aiming at the asymmetric flow characteristics of the molten pool caused by gravity, the present invention proposes an adaptive fitting algorithm for the deposition track profile. Through the dynamic combination of differential logic function and linear function, high-precision fitting of the asymmetric deposition track cross-sectional profile is achieved.

[0059] 2. The present invention creatively proposes a method that combines the flat-top overlapping lap model with bisection iterative optimization, breaking through the limitations of the traditional symmetric function model and achieving accurate solution of the overlapping spacing of asymmetric sedimentary tracks.

[0060] 3. This invention is a complete prediction system for path planning of lateral direct energy deposition, which can provide technical support for production process optimization. It also has potential application value in the manufacture of complex and overhanging structural parts, and will help further expand the practical application of additive manufacturing. BRIEF DESCRIPTION OF THE DRAWINGS

[0061] Figure 1 Flowchart of the method for predicting the overlap spacing of asymmetric deposition paths for lateral direct energy deposition;

[0062] Figure 2 This is a diagram of the experimental setup for lateral direct energy deposition;

[0063] Figure 3 is the image of differential logic function;

[0064] Figure 4 It is the 3D point cloud data outline of the actual asymmetric sedimentary track;

[0065] Figure 5 The fitting results of the sedimentary track profile adaptive fitting algorithm for the asymmetric sedimentary track cross-section data: (a) fitting results when R<30; (b) fitting results when R≥30;

[0066] Figure 6 This is a schematic diagram of the asymmetric sedimentary track flat-top overlap model;

[0067] Figure 7 This is a schematic diagram of the XGBoost regression model.

[0068] 1-ABB robot control cabinet, 2-ABB robot, 3-heat source, 4-Q235 base plate, 5-positioner, 6-material supply mechanism, 7-protective gas cylinder, 8-heat source power supply. DETAILED DESCRIPTION

[0069] The following describes the embodiments of the present invention through specific examples. Those skilled in the art will readily understand the other advantages and benefits of the present invention from the disclosure herein. The present invention may also be implemented or applied through various other specific embodiments, and the details in this specification may be modified or altered based on different viewpoints and applications without departing from the spirit of the present invention.

[0070] The test platform of the present invention is as follows Figure 2 As shown, the robot 2 is connected to the ABB robot control cabinet 1 and is connected to the material supply mechanism 6 through the heat source 3. The material supply mechanism 6 supplies ER70S-6 welding wire and sends it to the deposition area through the heat source 3; the heat source 3 is connected to the shielding gas supply system, and the shielding gas cylinder 7 provides shielding gas, and the gas is transported to the deposition area through the heat source; the positioner 5 is used to fix and adjust the posture of the Q235 substrate 4 to meet the posture requirements during the lateral direct energy deposition process.

[0071] like Figure 1 As shown, a method for predicting the overlap spacing of asymmetric deposition paths of lateral direct energy deposition includes the following steps:

[0072] Step 1: Prepare different material feeding speeds V based on lateral direct energy deposition experiments f and heat source moving speed V h Assembled asymmetric sedimentary track samples;

[0073] Step 2: Use 3D point cloud scanning equipment to obtain 3D point cloud data of each asymmetric sedimentation track, such as Figure 4 As shown in the figure, it is cropped and corrected, and then sliced ​​along the deposition direction to generate a cross-sectional slice sequence. Subsequently, n cross-sectional slice sequences are selected by random sampling and merged to characterize the cross-sectional morphology of each asymmetric deposition track.

[0074] Step 3: The merged cross-sectional slice data are processed using a deposition track profile adaptive fitting algorithm to obtain a fitting function expression for the cross-sectional morphology of each asymmetric deposition track; the deposition track profile adaptive fitting algorithm is specifically as follows:

[0075] Calculate the deposition rate R of each deposition channel, R = V f / V h , where V f is the material feeding speed, V h is the heat source moving speed, and is compared with the threshold m of the deposition rate;

[0076] (1) If R<m, a differential logic function is used to fit the merged cross-sectional slice data. The differential logic function is formed by subtracting two logic functions. The specific expression is:

[0077] f(x)=g(x)-h(x)

[0078] Where:

[0079]

[0080] Where f(x) is a differential logic function, g(x) and h(x) are two logic functions with different coefficients, L1 is the maximum amplitude of g(x), k1 is the slope factor of g(x), x1 is the abscissa of the inflection point of g(x), L2 is the maximum amplitude of h(x), k2 is the slope factor of h(x), x2 is the abscissa of the inflection point of h(x), x represents the abscissa position of the input, and L1=L2, indicating that the maximum amplitudes of g(x) and h(x) are the same;

[0081] (2) If R ≥ m, first determine the segmentation point and select the point with the largest horizontal coordinate among all data points as the segmentation point, that is, (x max ,y(x max )); and divide the merged cross-sectional slice data into two subsets based on the segmentation points, specifically:

[0082] G1={(x i ,y i )|x i >x mean ,y i <y(x max )}

[0083] G2={(xi ,y i )|x i ≤x mean}∪{(x i ,y i )|x i >x mean ,y i ≥y(x max )}

[0084] Among them, G1 is subset one, G2 is subset two, x i is the horizontal coordinate value of the data point, y i is the vertical coordinate value of the data point, x mean is the mean of the horizontal coordinates of all data points, x max is the maximum value of the horizontal coordinate among all data points, y(x max ) is x max The corresponding vertical coordinate value;

[0085] (a) G1 is fitted using a differential logistic function, specifically:

[0086] f(x)=g(x)-h(x)

[0087] Where:

[0088]

[0089] Where f(x) is a differential logic function, g(x) and h(x) are two logic functions with different coefficients, L1 is the maximum amplitude of g(x), k1 is the slope factor of g(x), x1 is the abscissa of the inflection point of g(x), L2 is the maximum amplitude of h(x), k2 is the slope factor of h(x), x2 is the abscissa of the inflection point of h(x), x represents the abscissa position of the input, and L1=L2, indicating that the maximum amplitudes of g(x) and h(x) are the same;

[0090] (b) Fit G2 with a linear function, specifically:

[0091] u(x)=ax+b

[0092] Where u(x) is a linear function, a is the slope of u(x), and b is the ordinate intercept of u(x);

[0093] In order to ensure that f(x) and u(x) are continuously connected at the segmentation points, the continuity loss term is introduced:

[0094] continuity loss = (f(x max )-u(x max )) 2

[0095] The continuity loss term aims to minimize the difference between the two functions at the segmentation points, thereby ensuring that the values ​​output by f(x) and u(x) are consistent, and ultimately forming a continuous fitting function. The fitting result is as follows: Figure 5 As shown;

[0096] Step 4: Based on the fitting function expression of the cross-sectional morphology of each asymmetric deposition channel, a flat-top overlapping model is established, such as Figure 6 As shown, on this basis, the bisection method is used to solve the optimal solution of the objective function f(d), that is, the overlap spacing d, where the expression of the objective function is:

[0097] f(d)=argmin|S1(d)-S2(d)|

[0098] Where f(d) is the objective function, S1(d) and S2(d) represent the overlapping area function and the valley area function, respectively. The overlapping area function and the valley area function are obtained by integrating the functions of the corresponding areas.

[0099] The dichotomy solution process includes the following steps:

[0100] (1) Set the overlap spacing search interval [d min , d max ], where the lower bound of the interval d min = 0 indicates the critical spacing of adjacent asymmetric sedimentary tracks that completely overlap, and the upper limit of the interval d max =w represents the critical spacing between adjacent asymmetric deposition tracks without overlap, and w is the fitting width of the asymmetric deposition track;

[0101] (2) Define the convergence threshold parameter, the objective function accuracy threshold ε = 1 × 10 -3 , interval threshold δ = 1 × 10 -3 ;

[0102] (3) Calculate the midpoint value d of the current interval mid , specifically:

[0103]

[0104] (4) The calculation of the overlap spacing follows the solution steps of the bisection method, as follows:

[0105] (a) When the function convergence condition is met, that is, |f(d mid )|<ε, then the output overlap spacing d=d mid And terminate the iteration, where f(d mid ) represents the midpoint value d of the current interval mid The corresponding objective function value;

[0106] (b) When the function convergence condition is not met, the search interval is updated, specifically:

[0107] When f(d mid )·f(d min )<0, then update the search interval to [d min , d mid ]; when f(d mid )·f(d max )<0, then update the search interval to [d mid , d max ], repeat the iteration until |d is satisfied max -d min |<δ or |f(d mid )|<ε, then the output overlap spacing d=d mid , where f(d min ) represents the lower bound of the current interval d min The corresponding objective function value, f(d max ) represents the upper bound of the current interval d max The corresponding objective function value;

[0108] Step 5: Construct a material supply speed V f , heat source moving speed V h A data set with deposition rate R as input feature and overlap spacing d as output feature; the data set is normalized and 80% of it is divided into a training set and 20% is divided into a test set;

[0109] Step 6: Establish an XGBoost regression model to predict the overlap distance d. The principle is as follows Figure 7 As shown, the hyperparameter optimization method is used to adjust the parameters of the model on the training set;

[0110] Step 7: Use the test set to verify the trained XGBoost regression model and use the root mean square error RMSE and determination coefficient R 2 Its prediction performance for asymmetric sedimentary track overlap spacing is evaluated.

[0111] The lateral direct energy deposition experiment in step 1 specifically includes:

[0112] Clamp and fix the substrate at the center of the positioner platform, then flip the positioner platform to a position perpendicular to the ground; adjust the angle of the heat source so that it is perpendicular to the substrate, and set the distance between the heat source and the substrate to 12mm (this value is the preferred setting in the experiment and can be adjusted according to specific experimental conditions in actual use); then set the material supply speed V f and heat source moving speed V hand set the heat source moving direction parallel to the ground; after starting the heat source, the energy beam acts stably on the substrate, and the material supply mechanism starts to continuously supply material, and the experimental process begins.

[0113] The heat source used in the lateral direct energy deposition experiment in step 1 is a laser heat source or an arc heat source.

[0114] Material supply speed V in step 1 f The range is 2m / min to 7m / min, and the heat source moving speed V h The range is 0.06m / min to 0.9m / min.

[0115] In step 2, n is set to 10.

[0116] The threshold m of the deposition rate in step 3 is set to 30.

[0117] The parameters and ranges for hyperparameter optimization in step 6 include:

[0118] The value range of learning rate η is η∈{0.01, 0.05, 0.1, 0.2}; the value range of maximum tree depth max_depth is max_depth∈{3, 5, 7, 9}; the value range of column sampling ratio colsample_bytree is colsample_bytree∈{0.6, 0.8, 1.0}; the value range of L1 regularization parameter α is α∈{0.1, 1, 10}; the value range of L2 regularization parameter λ is λ∈{0, 0.1, 1}; the value range of maximum number of leaf nodes max_leaves is max_leaves∈{20, 30, 50}.

[0119] The evaluation goal in step 7 is to minimize the root mean square error RMSE and make the coefficient of determination R 2 Close to 1; RMSE and R 2 The specific formula is as follows:

[0120]

[0121] Among them, d i is the target value of the overlap spacing, is the average value of the overlap spacing target value, is the predicted value of the overlap spacing, i=1, 2,…N, N is the total number of samples.

[0122] Figure 6 Schematic diagram of the asymmetric sedimentary track flat-top overlap model; Figure 6As shown, A1 and C1 are the two intersection points of the previous deposition track on the substrate, and B1 is the highest point of the previous deposition track; A2 and C2 are the two intersection points of the next deposition track on the substrate, and B2 is the highest point of the next deposition track; E is the intersection point of the profile fitting function of the previous deposition track and the profile fitting function of the next deposition track; line segment B1B2 is the surface of the sedimentary layer under ideal conditions; the area surrounded by B2EB1 is the valley area; the area surrounded by A1EC2 is the overlapping area; d0 is the offset distance of the highest point of the deposition track affected by gravity, d is the overlap spacing, h is the deposition track height, and w is the deposition track width;

[0123] The model is based on the following assumptions:

[0124] (a) Although gravity can cause the sedimentary track profile to shift, thus forming an asymmetric sedimentary morphology, under the same sedimentation rate conditions, the offset of each sedimentary track is consistent, denoted as d0;

[0125] (b) Under the same deposition rate conditions, the profile of each deposition channel remains constant during the deposition process, and the cross-sectional profile of each asymmetric deposition channel is consistent with that of the first deposition channel and can be represented by the same fitting function;

[0126] The asymmetric deposition path profile A1B1C1 is translated along the substrate plane A1C1. The translated asymmetric deposition path profile is represented by A2B2C2. When the translation distance makes the area of ​​the valley region B2EB1 equal to the area of ​​the overlap region A1EC2, the translation distance at this time is the overlap spacing d.

[0127] Figure 7 This is the schematic diagram of the XGBoost regression model; Figure 7 As shown, first, a data set D containing multiple samples is constructed. i , V fi , V hi , d i )}, where R i represents the deposition rate of the i-th sample, V fi represents the material feeding rate of the i-th sample, V hi represents the heat source moving speed of the i-th sample, d i represents the true value of the overlap spacing of the i-th sample;

[0128] The XGBoost algorithm is used to model the above dataset. The XGBoost algorithm performs fitting prediction by integrating multiple regression tree models. First, the first regression tree f1(x) is trained, and the initial sample features are used as input to fit the sample label d iThe residual between the predicted value of the current model and the residual; then, based on the residual of the previous round, the second regression tree f2(x) is trained, and so on, the 1st to the mth decision trees are iteratively trained in turn, and finally a set of regression tree models {f1(x), f2(x), ..., f m (x)};

[0129] The model calculates the prediction results of each regression tree model for the input sample and performs weighted summation of all prediction results to obtain the overlap spacing prediction value of the i-th sample. Right now:

[0130]

[0131] Where x=(R i , V fi , V hi ) is the input feature vector of the i-th sample, f k (x) represents the output value of the kth decision tree model.

[0132] The above embodiments are merely illustrative of the principles and effects of the present invention and are not intended to limit the present invention. Anyone skilled in the art may modify or alter the above embodiments without departing from the spirit and scope of the present invention. Therefore, all equivalent modifications or alterations made by one of ordinary skill in the art without departing from the spirit and technical principles disclosed herein are intended to be covered by the claims of the present invention.

Claims

1. A method for predicting the overlap spacing of asymmetric deposition paths using lateral direct energy deposition, characterized in that: It includes the following steps: Step 1: Prepare different material feeding speeds V based on lateral direct energy deposition experiments f and heat source moving speed V h Assemblage of asymmetric sedimentary tracks; Step 2: Use a three-dimensional point cloud scanning device to obtain the three-dimensional point cloud data of each asymmetric deposition channel, crop and correct it, and then slice it along the deposition direction to generate a cross-sectional slice sequence; Subsequently, use the random sampling method to select and merge n cross-sectional slice sequences to characterize the cross-sectional morphology of each asymmetric deposition channel; Step 3: Process the merged cross-sectional slice data using the deposition channel contour adaptive fitting algorithm to obtain the fitting function expression of the cross-sectional morphology of each asymmetric deposition channel; The specific process of the deposition channel contour adaptive fitting algorithm is as follows: Calculate the deposition rate R of each deposition channel, R = V f / V h , where V f is the material feeding speed, V h is the heat source moving speed, and is compared with the threshold m of the deposition rate; (1) If R < m, use the differential logic function to fit the merged cross-sectional slice data. The differential logic function is composed of subtracting two logic functions, and the specific expression is: f(x) = g(x) - h(x) In the formula: where, f(x) is the differential logic function, g(x) and h(x) are two logic functions with different coefficients, L1 is the maximum amplitude of g(x), k1 is the slope factor of g(x), x1 is the abscissa of the inflection point of g(x), L2 is the maximum amplitude of h(x), k2 is the slope factor of h(x), x2 is the abscissa of the inflection point of h(x), x represents the input abscissa position, and L1 = L2, indicating that the maximum amplitudes of g(x) and h(x) are the same; (2) If R ≥ m, first determine the segmentation point and select the point with the largest horizontal coordinate among all data points as the segmentation point, that is, (x max ,y(x max )); and divide the merged cross-sectional slice data into two subsets based on the segmentation points, specifically: G1={(x i ,y i )|x i >x mean ,y i <y(x max )} G2={(x i ,y i )|x i ≤x mean }∪{(x i ,y i )|x i >x mean ,y i ≥y(x max )} Among them, G1 is subset one, G2 is subset two, x i is the horizontal coordinate value of the data point, y i is the vertical coordinate value of the data point, x mean is the mean of the horizontal coordinates of all data points, x max is the maximum value of the horizontal coordinate among all data points, y(x max ) is x max The corresponding vertical coordinate value; (a) Fit G1 using the differential logic function, specifically: f(x) = g(x) - h(x) In the formula: where, f(x) is the differential logic function, g(x) and h(x) are two logic functions with different coefficients, L1 is the maximum amplitude of g(x), k1 is the slope factor of g(x), x1 is the abscissa of the inflection point of g(x), L2 is the maximum amplitude of h(x), k2 is the slope factor of h(x), x2 is the abscissa of the inflection point of h(x), x represents the input abscissa position, and L1 = L2, indicating that the maximum amplitudes of g(x) and h(x) are the same; (b) Fit G2 using a linear function of the first order, specifically: u(x) = ax + b where, u(x) is a linear function of the first order, a is the slope of u(x), and b is the vertical intercept of u(x); To ensure the continuous connection of f(x) and u(x) at the segmentation point, introduce a continuity loss term: continuity loss=(f(x max )-u(x max )) 2 This continuity loss term aims to minimize the difference between the two functions at the segmentation point, thereby ensuring that the values output by f(x) and u(x) are consistent, and finally forming a continuous fitting function; Step 4: Based on the fitting function expression of the cross-sectional morphology of each asymmetric deposition channel, establish a flat-top overlapping lap model, and on this basis, use the bisection method to solve the optimal solution of the objective function f(d), that is, the lap spacing d, where the expression of the objective function is: f(d) = argmin|S1(d) - S2(d)| where, f(d) is the objective function, S1(d) and S2(d) respectively represent the overlapping area function and the valley area function, and the overlapping area function and the valley area function are obtained by integrating the functions of the corresponding regions; The bisection method solving process includes the following steps: (1) Set the overlap spacing search interval [d min , d max ], where the lower bound of the interval d min = 0 indicates the critical spacing of adjacent asymmetric sedimentary tracks that completely overlap, and the upper limit of the interval d max =w represents the critical spacing between adjacent asymmetric deposition tracks without overlap, and w is the fitting width of the asymmetric deposition track; (2) Define the convergence threshold parameter, the objective function accuracy threshold ε = 1 × 10 -3 , interval threshold δ = 1 × 10 -3 ; (3) Calculate the midpoint value d of the current interval mid , specifically: (4) The calculation of the overlap spacing follows the solution steps of the bisection method, as follows: (a) When the function convergence condition is met, that is, |f(d mid )|<ε, then the output overlap spacing d=d mid And terminate the iteration, where f(d mid ) represents the midpoint value d of the current interval mid The corresponding objective function value; (b) When the function convergence condition is not met, the search interval is updated, specifically: When f(d mid )·f(d min )<0, then update the search interval to [d min , d mid ]; when f(d mid )·f(d max )<0, then update the search interval to [d mid , d max ], repeat the iteration until |d is satisfied max -d min |<δ or |f(d mid )|<ε, then the output overlap spacing d=d mid , where f(d min ) represents the lower bound of the current interval d min The corresponding objective function value, f(d max ) represents the upper bound of the current interval d max The corresponding objective function value; Step 5: Construct a material supply speed V f , heat source moving speed V h A data set with deposition rate R as input feature and overlap spacing d as output feature; the data set is normalized and 80% of it is divided into a training set and 20% is divided into a test set; Step 6: Build an XGBoost regression model to predict the overlap distance d, and use hyperparameter optimization method to adjust the model parameters on the training set; Step 7: Use the test set to verify the performance of the trained XGBoost regression model by calculating the root mean square error RMSE and the coefficient of determination R 2 Two key indicators are used to evaluate the model's prediction accuracy for the overlap spacing of asymmetric sedimentary tracks.

2. The method according to claim 1, characterized in that The lateral direct energy deposition experiment in step 1 specifically includes: Clamp and fix the substrate at the center of the positioner platform, then flip the positioner platform to a position perpendicular to the ground; adjust the angle of the heat source to make it perpendicular to the substrate, and then set the material feeding speed V f and heat source moving speed V h and set the heat source moving direction parallel to the ground; after starting the heat source, the energy beam acts stably on the substrate, and the material supply mechanism starts to continuously supply material, and the experimental process begins.

3. The method according to claim 2, characterized in that The heat source is a laser heat source or an arc heat source.

4. The method according to claim 1, wherein The material supply speed V in step 1 f The range is 2m / min to 7m / min, and the heat source moving speed V h The range is 0.06m / min to 0.9m / min.

5. The method according to claim 1, characterized in that In the step 2, n is 10.

6. The method according to claim 1, characterized in that The threshold value m of the deposition rate in step 3 is 30.

7. The method according to claim 1, characterized in that In step 6, the parameters and ranges of the hyperparameter optimization specifically include: The value range of learning rate η is η∈{0.01, 0.05, 0.1, 0.2}; the value range of maximum tree depth max_depth is max_depth∈{3, 5, 7, 9}; the value range of column sampling ratio colsample_bytree is colsample_bytree∈{0.6, 0.8, 1.0}; the value range of L1 regularization parameter α is α∈{0.1, 1, 10}; the value range of L2 regularization parameter λ is λ∈{0, 0.1, 1}; the value range of maximum number of leaf nodes max_leaves is max_leaves∈{20, 30, 50}.

8. The method according to claim 1, characterized in that In step 7, the evaluation goal is to minimize the root mean square error RMSE and make the coefficient of determination R 2 Close to 1; RMSE and R 2 The specific formula is as follows: Among them, d i is the target value of overlap spacing, is the average value of the overlap spacing target value, is the predicted value of the overlap spacing, i=1, 2, ...N, N is the total number of samples.

Images