Electric arc additive manufacturing weld bead morphology prediction method based on three-dimensional scanning point cloud

Abstract

The invention provides an electric arc additive manufacturing weld bead morphology prediction method based on three-dimensional scanning point cloud. The method comprises the following steps that firstly, a weld bead of an electric arc additive manufacturing stainless steel weldment is obtained through an electric arc additive manufacturing stainless steel experiment; 2, a three-dimensional point cloud model of the stainless steel welding beads manufactured through the electric arc additive is obtained through a handheld three-dimensional scanner, and each welding bead is cut and straightened; thirdly, the straightened uniform arc additive manufacturing stainless steel weld bead is subjected to slice projection in the Y direction, and the average shape of the weld bead is obtained and serves as a data set; 4, determining input, output and the number of hidden neurons, and establishing a neural network model suitable for a weld bead morphology data set; 5, training the neural network, and predicting the weld bead morphology of the target process parameters; and 6, comparing the weld bead morphology predicted by the neural network with the verified weld bead morphology, and calculating a prediction error and precision. The method has the characteristics of high speed and high precision.

Description

Technical Field

[0001] The present invention relates to the field of arc additive manufacturing, and more specifically, to a technical field of using a neural network method to predict arc additive manufacturing weld bead morphology. Background Art

[0002] Since the arc additive manufacturing process involves multiple process variables, and there is a complex nonlinear relationship between each process parameter and the weld bead morphology variable, it is difficult for traditional mathematical modeling methods to accurately realize the mapping between input variables and output variables. In the current research on arc additive manufacturing weld bead morphology prediction, the general method is to compare the weld width and residual height of the formed part measured by a vernier caliper with the weld width and residual height predicted by the mathematical model, and it is impossible to predict the complete morphology of the entire weld bead. Usually, circular arcs, cosine functions, and parabolas are used to fit the weld bead contour. Some scholars control the geometric dimensions of the formed part by establishing mathematical prediction weld bead models and droplet size models, thereby improving the overall forming quality of the part.

[0003] When predicting the morphology of a single-layer, single-pass weld of stainless steel in arc additive manufacturing, if only one of the process parameters, welding current, welding speed, or wire feeding speed, is changed, the prediction can be made by linear interpolation. In practice, the morphology of the weld bead when only one process parameter is changed is not a uniform change predicted by the interpolation method. Neural networks can learn the relationship between complex process parameters and weld bead morphology from a large amount of arc additive manufacturing process data, thereby realizing the prediction of unknown samples. Therefore, a neural network model is used to predict the morphology of a single-layer, single-pass stainless steel formed by arc additive manufacturing under unknown process parameters. Summary of the invention

[0004] The purpose of the present invention is to provide a method for predicting the morphology of arc additively manufactured welds based on three-dimensional scanning point clouds. The present invention uses a three-dimensional scanner to obtain three-dimensional point cloud data of welds, and slices and projects the weld point cloud data through MATLAB to obtain the average morphology of welds. A forming morphology data set of arc additively manufactured stainless steel welds is established through control experiments, and a suitable neural network model structure is established to predict the morphology of arc additively manufactured stainless steel welds with unknown process parameters, so that the morphology of welds can be obtained without experiments, reducing experimental costs and time costs. The weld morphology prediction method based on three-dimensional scanning point clouds of the present invention has the characteristics of fast speed and high accuracy.

[0005] In order to achieve the above technical features, the purpose of the present invention is to achieve: a method for predicting arc additive manufacturing weld bead morphology based on three-dimensional scanning point cloud, comprising the following steps: Step 1: Obtain the weld of arc additively manufactured stainless steel weldment through arc additively manufactured stainless steel experiment; Step 2: Obtain a 3D point cloud model of the arc additively manufactured stainless steel weld bead using a handheld 3D scanner, and cut and align each weld bead; Step 3: Slice and project the aligned uniform arc additively manufactured stainless steel weld along the Y direction to obtain the average morphology of the weld as a data set; Step 4: Determine the number of inputs, outputs and hidden neurons, and establish a neural network model suitable for the weld bead morphology data set; Step 5: Train the neural network to predict the weld bead morphology of the target process parameters; Step 6: Compare the weld bead morphology predicted by the neural network with the verified weld bead morphology, and calculate the prediction error and accuracy.

[0006] The step 2 specifically includes: First, the substrate containing three welds was scanned with a handheld 3D scanner to obtain a point cloud model of the welds, and the cropped and aligned weld point clouds were saved in STL format.

[0007] The step three specifically includes: Program the point cloud reading and slicing program in MATLAB to read the aligned weld point cloud data, and calculate the number of slice sections according to the difference between the maximum and minimum y coordinate values ​​of the point cloud data and the set interval distance. The calculation formula for the cross-section slicing of the point cloud is as follows: ; ; Where: and The minimum and maximum values ​​of the weld point cloud in the y direction; h is the slice interval; n is the number of slice planes; Ceiling " is the rounding up function; i The slice number.

[0008] The step 4 specifically includes: A neural network model is established to predict the forming morphology of single-layer single-pass stainless steel in arc additive manufacturing. The welding current, welding speed, wire feeding speed and weld width and excess height of the weld are set as input, and the height coordinate value of the weld morphology is output.

[0009] The step five specifically includes: The welding current, welding speed, and wire feeding speed were set to conduct arc additive manufacturing stainless steel weld verification experiments. A 3D scanner was used and sliced ​​to obtain the average morphology of the verification weld. The morphology of a given single-layer single-pass weld predicted by the neural network was compared with the actual morphology of the weld.

[0010] The step six specifically includes: Select RMSE, MAE, RMSE is the average of the square root of the difference between all individual observations and the true value; MAE is used to evaluate the difference between the actual value and the predicted value; R 2 It is used to evaluate the quality of the regression model and measure the correlation between the neural network prediction results and the target value. 2 The closer the value and R value are to 1, the better the model performance is. The specific formula is as follows: ; ; ; ; Where: is the prediction data of the neural network model; is the average value of the predicted data; is the actual value of the formed size, m is the sample size.

[0011] The present invention has the following beneficial effects: 1. Traditional experimental methods require a lot of time and experimental costs to study the forming size and morphology of welds. Compared with the traditional arc additive manufacturing weld morphology prediction method, the present invention uses three-dimensional scanning point cloud and neural network for prediction to obtain a complete weld morphology, and accurately predict the weld forming morphology in a shorter time, saving experimental resources and time costs, effectively reducing experimental costs and time, and providing a more economical and efficient solution for arc additive manufacturing stainless steel forming size and morphology prediction.

[0012] 2. The present invention can accurately and non-destructively obtain the three-dimensional point cloud information of arc additively manufactured stainless steel welds by scanning the substrate with a handheld three-dimensional scanner. The point cloud data contains the spatial position information of each point on the weld surface, and can capture the detailed changes on the weld surface. Compared with the traditional method of fitting by cutting cross sections, the three-dimensional point cloud data can obtain the information of any point on the entire arc additively manufactured weld and the morphology at any position.

[0013] 3. The present invention uses MATLAB to read the three-dimensional point cloud of the stainless steel weld, calculates the number of slice sections based on the difference between the maximum and minimum y-coordinate values ​​of the point cloud data and the set interval distance, and calculates the average value of the points on the slice sections to obtain the three-dimensional average morphology of the weld.

[0014] 4. The present invention establishes a neural network to predict the arc additive manufacturing stainless steel weld model. It can predict the weld morphology of unknown process parameters without conducting experiments by learning the relationship between welding current, welding speed, wire feeding speed, forming size and other parameters and weld morphology, thereby providing a morphology prediction method with fast speed, high accuracy and strong adaptability.

[0015] 5. The weld bead morphology prediction method based on three-dimensional scanning point cloud of the present invention can be applied to the morphology prediction of formed parts made of different materials in the field of additive manufacturing, and has broad application prospects. BRIEF DESCRIPTION OF THE DRAWINGS

[0016] The present invention will be further described below in conjunction with the accompanying drawings and embodiments.

[0017] Figure 1 This is a flow chart for predicting the morphology of stainless steel welds manufactured by arc additive manufacturing according to the present invention.

[0018] Figure 2 The three-dimensional point cloud model of the stainless steel weld manufactured by arc additive manufacturing of the present invention.

[0019] Figure 3 This is the average morphology of the stainless steel weld after point cloud slicing of the present invention.

[0020] Figure 4 This is a diagram of the morphology structure of stainless steel welds produced by arc additive manufacturing predicted by the neural network of the present invention.

[0021] Figure 5 This is a comparison chart between the weld bead morphology predicted by the neural network of the present invention and the actual weld bead morphology. DETAILED DESCRIPTION

[0022] The embodiments of the present invention are further described below in conjunction with the accompanying drawings.

[0023] Embodiment 1: See also Figure 1-5 ,Step 1: Obtain the weld bead of arc additively manufactured stainless steel weldment through arc additively manufactured stainless steel experiment; Step 2: Obtain a 3D point cloud model of the arc additively manufactured stainless steel weld bead using a handheld 3D scanner, and cut and align each weld bead; Step 3: Slice and project the aligned uniform arc additively manufactured stainless steel weld along the Y direction to obtain the average morphology of the weld as a data set; Step 4: Determine the number of inputs, outputs and hidden neurons, and establish a neural network model suitable for the weld bead morphology data set; Step 5: Train the neural network to predict the weld bead morphology of the target process parameters; Step 6: Compare the weld bead morphology predicted by the neural network with the verified weld bead morphology, and calculate the prediction error and accuracy.

[0024] Embodiment 2: Step 1: The main process of the weld bead morphology prediction method based on 3D scanning point cloud of the present invention is divided into the following steps: "weld bead point cloud file acquisition → point cloud data slicing → weld bead average morphology → design and train neural network model → predict unknown parameter weld bead morphology → calculate prediction accuracy and error", the specific processing flow is as follows Figure 1 shown.

[0025] Step 2: In the weld bead morphology prediction method based on 3D scanning point cloud of the present invention, firstly, a substrate containing three welds is scanned with a handheld 3D scanner to obtain a point cloud model of the welds, and the cropped and aligned weld bead point cloud is saved in stl format, such as Figure 2 shown.

[0026] Step 3: Program the point cloud reading and slicing program in MATLAB to read the aligned weld point cloud data, and calculate the number of slice sections according to the difference between the maximum and minimum y-coordinate values ​​of the point cloud data and the set interval distance. The main calculation formula for the cross-section slicing of the point cloud is as follows: Program the point cloud reading and slicing program in MATLAB to read the aligned weld point cloud data, and calculate the number of slice sections according to the difference between the maximum and minimum y coordinate values ​​of the point cloud data and the set interval distance. The calculation formula for the cross-section slicing of the point cloud is as follows: ; ; Where: and The minimum and maximum values ​​of the weld point cloud in the y direction; h is the slice interval; n is the number of slice planes; Ceiling " is the rounding up function; i The slice number.

[0027] In order to obtain a more accurate average morphology of stainless steel welds, the present invention stipulates that the interval of longitudinal slicing along the weld is 1 mm, and multiple weld slicing sections are obtained. The average value of the weld cross sections obtained by longitudinally slicing each weld is calculated, and the average point coordinates of 24 groups of welds are plotted into a curve, which is the average morphology of the weld. Figure 3 shown.

[0028] Step 4: The present invention establishes a neural network model to predict the morphology of a single-layer single-pass stainless steel formed by arc additive manufacturing. The welding current, welding speed, wire feeding speed and weld width and excess height are set as inputs, and the height coordinate value of the weld morphology is output. The structure of the neural network is set as follows: Figure 4 shown.

[0029] Step 5: Set the welding current to 222A, the welding speed to 5.5mm / s, and the wire feeding speed to 201cm / min to conduct arc additive manufacturing stainless steel weld verification experiments. Use a 3D scanner and slice to obtain the average morphology of the verified weld, and compare the neural network prediction of a given single-layer single-pass weld morphology with the actual weld morphology. The comparison chart is as follows: Figure 5 shown.

[0030] Step 6: Calculate the accuracy and error of the neural network prediction of the weld bead morphology. The present invention selects the root mean square error RMSE , mean absolute error MAE . RMSE It is the average of the square root of the difference between all individual observations and the true value. MAE Used to evaluate the difference between actual and predicted values. R 2 It is generally used to evaluate the quality of regression models and measure the correlation between neural network prediction results and target values. R 2 Value and R The closer the value is to 1, the better the model performance is. The specific formula is as follows: ; ; ; ; Where: is the prediction data of the neural network model; is the average value of the predicted data; is the actual value of the formed size, m is the sample size.

[0031] The present invention predicts and verifies the weld R The value is 0.9674, R 2 Value 0.9385; RMSE is 0.1159, MAEis 0.0935. The results show that the neural network model can quickly and accurately predict the morphology of the weld bead. The method of the present invention has broad application prospects in the prediction of the forming morphology of arc additive manufacturing stainless steel, and can obtain a more accurate weld bead morphology of arc additive manufacturing stainless steel weld bead without experiments. At the same time, it can be used for the weld bead morphology of other additive manufacturing materials. This method also provides a useful reference for the prediction of additive manufacturing morphology of other metal materials.

Claims

1. A method for predicting arc additive manufacturing weld bead morphology based on three-dimensional scanning point cloud, characterized in that: The following steps are involved: Step 1: Obtain the weld of arc additively manufactured stainless steel weldment through arc additively manufactured stainless steel experiment; Step 2: Obtain a 3D point cloud model of the arc additively manufactured stainless steel weld bead using a handheld 3D scanner, and cut and align each weld bead; Step 3: Slice and project the aligned uniform arc additively manufactured stainless steel weld along the Y direction to obtain the average morphology of the weld as a data set; Step 4: Determine the number of inputs, outputs and hidden neurons, and establish a neural network model suitable for the weld bead morphology data set; Step 5: Train the neural network to predict the weld bead morphology of the target process parameters; Step 6: Compare the weld bead morphology predicted by the neural network with the verified weld bead morphology, and calculate the prediction error and accuracy.

2. The arc additive manufacturing weld bead morphology prediction method based on three-dimensional scanning point cloud according to claim 1, characterized in that: The step 2 specifically includes: First, the substrate containing three welds was scanned with a handheld 3D scanner to obtain a point cloud model of the welds, and the cropped and aligned weld point clouds were saved in STL format.

3. The arc additive manufacturing weld bead morphology prediction method based on three-dimensional scanning point cloud according to claim 1, characterized in that: The step three specifically includes: Program the point cloud reading and slicing program in MATLAB to read the aligned weld point cloud data, and calculate the number of slice sections according to the difference between the maximum and minimum y coordinate values ​​of the point cloud data and the set interval distance. The calculation formula for the cross-section slicing of the point cloud is as follows: ； ； Where: and The minimum and maximum values ​​of the weld point cloud in the y direction; h is the slice interval; n is the number of slice planes;" Ceiling " is the rounding up function; i The slice number.

4. The arc additive manufacturing weld bead morphology prediction method based on three-dimensional scanning point cloud according to claim 1, characterized in that: The step 4 specifically includes: A neural network model is established to predict the forming morphology of single-layer single-pass stainless steel in arc additive manufacturing. The welding current, welding speed, wire feeding speed and weld width and excess height of the weld are set as input, and the height coordinate value of the weld morphology is output.

5. The arc additive manufacturing weld bead morphology prediction method based on three-dimensional scanning point cloud according to claim 1, characterized in that: The step five specifically includes: The welding current, welding speed, and wire feeding speed were set to conduct arc additive manufacturing stainless steel weld verification experiments. A 3D scanner was used and sliced ​​to obtain the average morphology of the verification weld. The morphology of a given single-layer single-pass weld predicted by the neural network was compared with the actual morphology of the weld.

6. The arc additive manufacturing weld bead morphology prediction method based on three-dimensional scanning point cloud according to claim 1, characterized in that: The step six specifically includes: Select RMSE, MAE, RMSE is the average of the square root of the difference between all individual observations and the true value; MAE is used to evaluate the difference between the actual value and the predicted value; R 2 It is used to evaluate the quality of the regression model and measure the correlation between the neural network prediction results and the target value. 2 The closer the value and R value are to 1, the better the model performance is. The specific formula is as follows: ； ； ； ； Where: is the prediction data of the neural network model; is the average value of the predicted data; is the actual value of the formed size, m is the sample size.

Images