Additive manufacturing intelligent path planning method based on machine learning

Abstract

The invention discloses an additive manufacturing intelligent path planning method based on machine learning, and the method comprises the steps: carrying out the polygonal convex decomposition of hierarchical contour information, forming a plurality of sub-regions, predicting an optimal scanning mode according to a machine learning model, generating the sub-paths of the sub-regions, and finally connecting the sub-paths. The method mainly aims at reducing deformation of printed parts in the additive manufacturing process and improving the surface quality of the parts. According to the invention, a scanning path mode can be dynamically planned according to geometrical characteristics and printing parameters of the contour of the to-be-filled object slice. The method can better adapt to printing objects with different shapes and complexities, can generate formed parts with high strength and high surface quality, and has a wide application prospect. According to the method, the contour of each layer of slice is subjected to multi-layer equidistant contour offset, the shape and size precision of a printed piece can be better controlled, the surface quality of the final printed piece is ensured, and the overall strength and durability are improved.

Description

Technical Field

[0001] The present invention relates to the technical field of additive manufacturing, and in particular to an intelligent path planning method for additive manufacturing based on machine learning. Background Art

[0002] At present, additive manufacturing (3D printing) technology has been widely used in industrial production. Additive manufacturing technology completes the molding of a single part in a very short time through modeling, layering, path planning and printing stacking. Among them, path planning is a key step in the above process. Different scanning methods have a great influence on the molding accuracy, mechanical properties of the molded parts and molding efficiency. During the molding process, the shape of the workpiece grows dynamically, and the molding temperature field and the state of the material change dynamically with the scanning path. This change will cause the workpiece to deform and residual stress, thereby affecting the accuracy, surface quality and performance of the molded parts. In addition, different scanning paths will also cause different molding times, which will affect the molding efficiency.

[0003] At present, most of the path planning technologies have relatively simple filling modes, such as zigzag filling, equidistant contour offset filling, partition filling, mixed filling, etc. However, using a single scanning method is difficult to apply to all printing parameters and the geometric features of the area to be filled, resulting in unsatisfactory printing accuracy, strength and printing efficiency. For complex geometric shapes, multiple filling modes can be used in the same layer. For example, efficient zigzag filling is used in large flat areas, and equidistant contour offset filling is used in edges or detail areas to ensure accuracy and strength. Therefore, it is necessary to use a suitable scanning method to plan the scanning path based on the printing parameters and the geometric feature information of each slice contour to ensure the final forming quality of the part.

[0004] With the popularity of artificial intelligence, intelligent path planning algorithms in 3D printing have attracted more and more attention. As a subset of artificial intelligence (AI), machine learning (ML) has the potential to improve product quality, optimize manufacturing processes and reduce costs. Machine learning models learn knowledge and make inferences from reliable training data sets. They can then predict the optimal process parameters more effectively. Therefore, ML has the potential to provide more time- and cost-effective solutions, automatically optimizing path planning based on a large amount of historical printing data to improve the strength and surface accuracy of additively manufactured printed parts. Summary of the invention

[0005] In order to overcome the shortcomings of the prior art, the present invention relates to an intelligent path planning method for additive manufacturing based on machine learning, which forms multiple sub-regions by polygonal convex decomposition of hierarchical contour information, and then generates sub-paths of the sub-regions according to the optimal scanning method predicted by the machine learning model, and finally connects the sub-paths. The main purpose is to reduce the deformation of printed parts in the additive manufacturing process and improve the surface quality of the parts.

[0006] The technical solution adopted by the present invention is as follows:

[0007] Step 1: Slice the three-dimensional model to be processed to obtain the two-dimensional cross-sectional profile information of each slice layer.

[0008] Step 2: According to the contour information of each slice layer, visit each layer in turn starting from the first layer, perform equidistant contour offset filling according to the filling path interval, obtain the contour offset filling curve, connect the obtained contour offset curves, and save the innermost layer offset contour path.

[0009] Step 3: Determine the scanning direction of each layer based on the innermost filling curve information, and divide it into multiple sub-areas through polygonal convex decomposition.

[0010] Step 4: According to the geometric feature information and printing parameter information of each sub-area, different scanning paths are planned for each sub-area through pre-trained machine learning, and the scanning paths of each sub-area are generated in turn.

[0011] Step 5: Connect the path planning information in each sub-area with the contour offset filling information of each slice layer generated in step 3 to obtain the laser scanning path of the slice layer.

[0012] Furthermore, in step 2, the number of offsets for contour offset filling is 2-5 times.

[0013] Furthermore, in step three, the scanning direction is designed to differ by 67 degrees for each layer, and the scanning direction angle of the first layer is 0 degree.

[0014] Furthermore, the specific steps of the partitioning method based on polygon convex decomposition in step 3 include:

[0015] (1) Input the innermost filling curve information obtained. The direction of each curve must comply with the "outer reverse and inner straight" rule;

[0016] (2) Traverse the vertices of the polygon and store them in the point set to search for all concave points. Take vertices from the polygon point set one by one, determine whether they are concave points, and if they are concave points, store them in the concave point set;

[0017] (3) Traverse all concave points to find the concave peak point;

[0018] (4) Traverse all concave peaks in turn and construct a segmentation line passing through them;

[0019] (5) Divide the polygon into several simply connected sub-regions along the dividing line segment;

[0020] (6) Output the obtained sequence of simply connected sub-regions.

[0021] Furthermore, in step 4, a machine learning method is used to plan the path for each sub-area. The specific steps are as follows:

[0022] (1) Data acquisition and preprocessing: Prepare a training data set containing geometric features and printing parameters. The data should cover a variety of 3D models of different complexity and shapes to ensure the generalization ability of the model. Standardize the acquired training sample data to form a training data set;

[0023] (2) Data annotation: annotate the collected print data and determine the optimal scanning method corresponding to each training sample;

[0024] (3) Machine learning model training: Based on the standardized data, a machine learning method is selected, the samples are divided into several groups for rotation training, and the machine learning parameters are determined based on the training results;

[0025] (4) Model validation: Use validation data sets to evaluate the accuracy and generalization ability of the model. Adjust model hyperparameters to optimize model performance. Use cross-validation methods to prevent overfitting and ensure that the model has good predictive ability for new data.

[0026] (5) Model deployment: Apply the trained machine learning model, combined with the slice layer contour data and printing parameters of the printed object, to generate the optimal filling method for each sub-area.

[0027] Furthermore, in step 4, path planning is performed for each sub-area using a machine learning method, and the machine learning methods that can be used include XGBoost, a support vector machine method, or a neural network.

[0028] Furthermore, in step four, path planning is performed on each sub-region through machine learning methods, and the characteristic information of the sub-region includes the area of ​​the sub-region, the proportion of sharp corners, the aspect ratio of the sub-region, the perimeter of the sub-region, the shape complexity of the sub-region (the number of polygon vertices of the shape), the maximum and minimum inner angles of the sub-region, the boundary length and area ratio of the sub-region, etc.

[0029] Furthermore, in step 4, a machine learning method is used to plan the path of each sub-area, and the printing parameter information includes printing speed, filling density, material properties, etc. The selection of printing parameters depends on the specific printing material and equipment performance.

[0030] Furthermore, in step 4, the path planning is performed for each sub-area through machine learning methods, and the scanning methods predicted by machine learning include zig-zag scanning, equidistant contour offset scanning, partition scanning, spiral scanning, checkerboard scanning and random scanning. Different scanning methods are suitable for different geometric shapes and printing requirements. For example, zig-zag scanning is suitable for more regular areas, while spiral scanning is suitable for circular or elliptical areas.

[0031] Compared with the existing scanning path planning technology, the present invention has the following obvious outstanding substantial features:

[0032] 1. The present invention adopts an intelligent path planning method for additive manufacturing based on machine learning, which can dynamically plan the scanning path according to the geometric characteristics of the slice contour of the object to be filled and the printing parameters. Compared with the traditional method, the present invention can better adapt to printing objects of different shapes and complexities, generate molded parts with high strength and high surface quality, and has broad application prospects.

[0033] 2. The present invention sets different scanning angles between adjacent scanning layers, and the scanning path planning is planned along the scanning angle, which can evenly distribute heat and reduce stress concentration, thereby reducing warping and deformation of the printed part; enhancing the isotropy of the material can avoid structural weakness caused by interlayer weaknesses and improve the overall strength and durability; increasing the bonding area and bonding strength between layers, reducing the risk of interlayer separation, and ensuring the stability and integrity of the printed part.

[0034] 3. The present invention performs multi-level equidistant contour offset on each layer of slice contour, which can better control the shape and dimensional accuracy of the printed part and ensure the surface quality of the final printed part; the superposition of multiple layers of contours can increase the structural strength of the printed part, especially when printing complex shapes and thin-walled structures, which can effectively prevent interlayer separation and fracture and improve overall strength and durability. BRIEF DESCRIPTION OF THE DRAWINGS

[0035] Figure 1 This is a flow chart of the operating steps of the intelligent path planning method for additive manufacturing based on machine learning of the present invention.

[0036] Figure 2 This is a flow chart of the partitioning steps of the intelligent path planning method for additive manufacturing based on machine learning of the present invention.

[0037] Figure 3(a) to (d) are schematic diagrams of the results of the equidistant contour offset and partitioning of the present invention, wherein (a) is a schematic diagram of the cross-sectional contour of a certain layer of a 3D model slice, (b) is a schematic diagram of the equidistant offset contour of the cross-sectional contour of the slice layer, and the number of offsets is 3 times. (c) is a schematic diagram of the innermost offset contour of the equidistant offset contour of the cross-sectional contour. (d) is a schematic diagram of the partitioning result of the cross-sectional contour of the slice layer according to the present invention, and the scanning tilt angle is 0 degrees.

[0038] Figure 4 This is a flow chart of the prediction of sub-region scanning mode based on machine learning in the present invention. DETAILED DESCRIPTION

[0039] The present invention will be further described below in conjunction with the accompanying drawings.

[0040] like Figure 1 As shown, the present invention provides an intelligent path planning method for additive manufacturing based on machine learning, and the steps are as follows:

[0041] Step 1: Set the layer height according to the part accuracy requirements, and use the slicing software to slice the 3D model to be processed to obtain the 2D cross-sectional profile information of each slice layer.

[0042] Step 2: Determine the filling spacing and the number of equidistant offset fillings (2-5 times) and obtain the contour information of each slice layer, visit each layer in turn starting from the first layer, perform multiple equidistant contour offset fillings according to the filling path interval, obtain the contour offset filling curve, connect the obtained contour offset curves, and save the innermost layer offset contour path. Figure 3 (b) is a schematic diagram of the equidistant contour offset for the slice contour, with the number of offsets being 3. (c) is a schematic diagram of the innermost biased contour path.

[0043] Step 3: Determine the scanning direction inclination of each layer, set it to 0 degrees for the first layer, and then 67 degrees for each layer. Obtain the innermost offset filling curve information of each layer after equidistant offset, and divide it into multiple sub-areas through polygonal convex decomposition. Figure 2 As shown, the specific processing process of the partitioning method based on polygon convex decomposition of the present invention is as follows:

[0044] (1) Input the innermost filling curve information obtained. The direction of each curve must comply with the "outer reverse and inner straight" rule;

[0045] (2) Traverse the vertices of the polygon and store them in the point set to search for all concave points. Take vertices from the polygon point set one by one to determine whether they are concave points. If they are concave points, store them in the concave point set. If there are no concave points after traversal, proceed directly to the next step of path filling;

[0046] (3) Traverse all concave points to find the concave peak point;

[0047] (4) Traverse all concave peak points in turn and construct a segmentation line that passes through them and is along the scanning direction. It is only necessary to find the two points on both sides of the concave peak point and all the intersection points of the contour that are closest to each other to construct the segmentation line segment.

[0048] (5) Divide the polygon into several simply connected sub-regions along the dividing line segment;

[0049] (6) Output the obtained sequence of simply connected sub-regions.

[0050] For a point P on the contour k , use the vector cross multiplication method to determine whether it is a concave point. The steps are as follows:

[0051] (1) Calculate the vector v k 、v k+1 So that:

[0052] v k =P k -P k-1

[0053] v k+1 =P k+1 -P k

[0054] (2) If P k satisfy:

[0055] (v k ×v k+1 ).dz<0

[0056] Then P k For concave points.

[0057] By scanning the line direction vector v s With v k and v k+1 Find the peak point on the contour. If the local contour is at P k The point reaches its peak value, then the vector v k 、v k+1 Will point to v s The opposite side of , satisfies:

[0058] (v k ×v s ).dz·(v k+1 ×v s ).dz≤0

[0059] (d) is a schematic diagram of partitioning the slice contour along the scanning direction of 0 degrees using the above partitioning steps.

[0060] Step 4: Based on the feature information and printing parameter information of each sub-area obtained in step 3, different scanning paths are planned for each sub-area through pre-trained machine learning, and the scanning paths of each sub-area are generated in turn.

[0061] like Figure 4 As shown, the specific processing process of the method flow of the present invention based on machine learning prediction of the optimal scanning mode of the sub-region is as follows:

[0062] (1) Data acquisition and preprocessing: Prepare a training data set containing geometric features and printing parameters. The data should cover a variety of 3D models of different complexity and shapes to ensure the generalization ability of the model. Standardize the acquired training sample data to form a training data set; Geometric feature data:

[0063] Collect models of different complexity and shapes from the existing 3D printing model library. Use slicing software to slice these models and obtain the 2D cross-sectional profile information of each layer. Print parameter data: record the parameters of each printing task, including printing speed, filling density, material properties, etc.

[0064] (2) Data labeling: Determine the optimal scanning method for each training sample and label it by combining geometric features and printing parameter data.

[0065] (3) Machine learning model training: Standardize the geometric features and printing parameter data to eliminate the impact of different data magnitudes. Common methods include z-score standardization, Min-Max normalization, etc. Select a machine learning method and divide the standardized data set into training set, validation set and test set, with a division ratio of 70% training set, 20% validation set and 10% test set. Select appropriate hyperparameters, such as learning rate, regularization parameters, etc., divide the samples into several groups for rotation training, and determine the machine learning parameters to initialize the model parameters based on the training results. Improve model performance by adjusting hyperparameters (such as grid search or random search);

[0066] (4) Model validation: Use validation data sets to evaluate the accuracy and generalization ability of the model. Adjust model hyperparameters to optimize model performance. Use cross-validation methods to prevent overfitting and ensure that the model has good predictive ability for new data.

[0067] (5) Model deployment: Apply the trained machine learning model, combined with the slice layer contour data and printing parameters of the printed object, to generate the optimal filling method for each sub-area.

[0068] Step 5: Connect the path planning information in each sub-area generated in step 4 with the contour offset filling information of each slice layer generated in step 3 to obtain the laser scanning path of the slice layer.

Claims

1. A machine learning-based intelligent path planning method for additive manufacturing, characterized in that: The steps include: Step 1: Slice the constructed three-dimensional model to be processed to obtain the two-dimensional cross-sectional profile information of each slice layer; Step 2: According to the contour information of each slice layer, visit each layer in turn starting from the first layer, perform equidistant contour offset filling according to the filling path interval, obtain the contour offset filling curve, connect the obtained contour offset curves, and save the innermost layer offset contour path; Step 3: Determine the scanning direction of each layer according to the innermost filling curve information, and divide it into multiple sub-areas by polygon convex decomposition; Step 4: According to the geometric feature information and printing parameter information of each sub-area, different scanning paths are planned for each sub-area through pre-trained machine learning, and the scanning paths of each sub-area are generated in turn; Step 5: Connect the path planning information in each sub-area with the contour offset filling information of each slice layer generated in step 3 to obtain the laser scanning path of the slice layer.

2. The method for intelligent path planning for additive manufacturing based on machine learning according to claim 1, characterized in that: In step 2, the number of offsets for contour offset filling is 2-5 times.

3. The method for intelligent path planning for additive manufacturing based on machine learning according to claim 1, characterized in that: In step 3, the scanning direction is designed to differ by 67 degrees for each layer, and the scanning direction angle of the first layer is 0 degrees.

4. The method for intelligent path planning for additive manufacturing based on machine learning according to claim 1, characterized in that: The specific steps of the partitioning method based on polygon convex decomposition in step 3 include: (1) Input the innermost filling curve information obtained. The direction of each curve must comply with the "outer reverse and inner straight" rule; (2) traverse the vertices of the polygon and store them in a point set to search for all concave points; take vertices from the polygon point set in turn, determine whether they are concave points, and if so, store them in the concave point set; (3) Traverse all the concave points to find the concave peak point; (4) Traverse all concave peaks in turn and construct a segmentation line passing through them; (5) Divide the polygon into several simply connected sub-regions along the dividing line segment; (6) Output the obtained sequence of simply connected sub-regions.

5. The method for intelligent path planning for additive manufacturing based on machine learning according to claim 1, characterized in that: In step 4, the path planning for each sub-area is performed by machine learning method; the specific steps are as follows: (1) Data acquisition and preprocessing: Prepare a training data set containing geometric features and printing parameters; the data should cover a variety of 3D models of different complexity and shapes to ensure the generalization ability of the model; standardize the acquired training sample data to form a training data set; (2) Data annotation: annotate the collected print data and determine the optimal scanning method corresponding to each training sample; (3) Machine learning model training: Based on the standardized data, a machine learning method is selected. Divide the samples into several groups for rotation training, and determine the machine learning parameters based on the training results; (4) Model validation: Use validation data sets to evaluate the accuracy and generalization ability of the model; adjust model hyperparameters to optimize model performance; use cross-validation methods to prevent overfitting and ensure that the model has predictive power for new data; (5) Model deployment: Apply the trained machine learning model, combined with the slice layer contour data and printing parameters of the printed object, to generate the optimal filling method for each sub-area.

6. The method for intelligent path planning for additive manufacturing based on machine learning according to claim 1, characterized in that: In step 4, path planning is performed for each sub-area using machine learning methods, and the machine learning methods used include XGBoost, support vector machine method or neural network.

7. The method for intelligent path planning for additive manufacturing based on machine learning according to claim 1, characterized in that: In step 4, path planning is performed for each sub-region through machine learning methods. The characteristic information of the sub-region includes the area of ​​the sub-region, the proportion of sharp corners, the aspect ratio of the sub-region, the perimeter of the sub-region, the shape complexity of the sub-region, the maximum and minimum inner angles of the sub-region, and the boundary length and area ratio of the sub-region.

8. The method for intelligent path planning for additive manufacturing based on machine learning according to claim 1, characterized in that: In step 4, the path planning for each sub-area is performed through machine learning methods. The printing parameter information includes printing speed, filling density, and material properties. The selection of printing parameters depends on the specific printing material and equipment performance.

9. The method for intelligent path planning for additive manufacturing based on machine learning according to claim 1, characterized in that: In step 4, path planning is performed for each sub-area through machine learning methods. The scanning methods predicted by machine learning include zig-zag scanning, equidistant contour offset scanning, partition scanning, spiral scanning, checkerboard scanning and random scanning. Different scanning methods are suitable for different geometric shapes and printing requirements. The zig-zag scanning is suitable for more regular areas, and the spiral scanning is suitable for circular or elliptical areas.

Images