A method for detecting forming dimensions based on feature point recognition of cladding channels in additive manufacturing using full-topography point clouds.

Abstract

This invention proposes a method for identifying feature points and detecting forming dimensions of cladding passes in arc additive manufacturing based on 3D visual full-morphology point cloud information. The method includes the following steps: using a high-precision 3D scanner to perform omnidirectional scanning of each cladding pass layer, and preprocessing the point cloud data such as filtering, 3D surface reconstruction, point cloud refinement, and feature fitting alignment to obtain effective point cloud data; performing dimensionality reduction on the effective point cloud, and using the DW-GCT method to bidirectionally approximate and search for weld toe feature points of the first cladding pass layer; employing a heterogeneous region adaptive interpolation method to search for edge convex and non-edge convex regions and obtain cladding pass interpolation contours that retain small surface features; and combining this with a layer intersection point dynamic positioning method to bidirectionally search for bite points in the incomplete fusion zone between adjacent layers. The method proposed in this invention can effectively solve the problems of lack of cladding pass side point cloud information, depth of field error, and breakpoint error in existing technologies, and provides a method for extracting the top-layer cladding pass dimension data using the full-morphology 3D point cloud contour information of the cladding pass.

Description

Technical Field

[0001] This invention relates to the field of metal additive manufacturing technology, and in particular to a method for identifying feature points of cladding channels and detecting forming dimensions in additive manufacturing based on 3D visual full-shape point cloud information. Background Technology

[0002] Wire Arc Additive Manufacturing (WAAM) is a technology based on the discrete-stacking principle. It utilizes an electric arc heat source to melt the tip of a metal wire at high temperatures, continuously feeding molten metal droplets into the molten pool via free transition or liquid bridge transition, and depositing them layer by layer along a pre-planned path to form the final shape. Compared to traditional subtractive manufacturing, WAAM technology can rapidly manufacture large-sized, complex, irregularly shaped structural parts, offering advantages such as rapid near-net-shape forming, extremely high material utilization, high design flexibility, and high production efficiency. It has been widely applied in aerospace, automotive, and shipbuilding industries.

[0003] However, due to the large heat input of the electric arc, the flow behavior and shape of the molten pool become more complex and difficult to control due to the influence of heat input. The heat accumulation effect changes the thermal boundary conditions of the molten pool, affecting the morphology and dimensional consistency of the deposited layer. This makes it difficult to maintain a consistent height and width of the cladding channel, resulting in lower robustness of the formed dimensions and poor process stability. Automatic forming control technology, by continuously correcting the deviation between the actual measured dimensions and the expected dimensions, is the core approach to improving the stability of the electric arc additive manufacturing process and the accuracy of the cladding channel's formed dimensions. Furthermore, the accuracy of detecting the formed dimensions of the cladding channel in the additive manufacturing process is a crucial prerequisite for precise control of the formed dimensions and the judgment of formed quality.

[0004] Currently, there are methods in the field of additive manufacturing that use visual sensors to measure the size of the cladding track. For example, Deng Junhao et al. [Deng Junhao. Research on morphology detection and prediction modeling of arc additive forming based on visual sensing [D]. Shanghai Jiaotong University, 2019.] used a two-dimensional line laser sensor to vertically scan the cladding track to construct three-dimensional point cloud information, and used the first-order and second-order height difference methods to identify the weld toe and calculate the forming width size. This method can identify the weld toe feature points from the three-dimensional point cloud information containing the substrate and the deposited layer. However, this feature point search method is only based on the height difference between adjacent points, which has poor robustness. Moreover, the detection accuracy is limited by the accuracy and scanning range of the line laser sensor. Therefore, the final detection result has a large error compared with the actual result.

[0005] In existing methods, most studies use two-dimensional line laser sensors to vertically illuminate the cladding channel to obtain point cloud information. However, because two-dimensional line laser sensors cannot scan the side contours of the cladding wall and cannot identify the true point cloud information of the incompletely fused areas between deposition layers, and because the vertically illuminating line laser causes discontinuities on both sides of the cladding channel, most studies directly use these discontinuities as feature points for calculating the width. However, due to the depth-of-field problem of two-dimensional line lasers, the location of the discontinuities is related to the laser height and the included angle α. Furthermore, when there are edge convex regions on both sides of the cladding channel caused by the liquid metal not spreading evenly before solidification, the discontinuities may not be feature points reflecting the true width (e.g., ...). Figure 1 As shown). Therefore, the two-dimensional line laser method for measuring the width of each cladding channel inevitably has certain depth-of-field errors [1] and breakpoint errors, and the detection accuracy is limited by the accuracy and scanning range of the line laser sensor. Ultimately, the detected width may have a large error with the actual width, resulting in distorted width detection data. Therefore, it is not suitable for cases with a high number of layers. In addition, there is currently no method to directly use the three-dimensional point cloud contour information of the full morphology of the cladding channel to extract the size data of the top cladding channel. Summary of the Invention

[0006] In view of this, the purpose of this invention is to overcome the problems of depth-of-field error and breakpoint error that always exist in existing two-dimensional line laser sensors. It proposes a method to extract the top layer cladding channel size data by directly using the three-dimensional point cloud contour information of the full morphology of the cladding channel, and provides a method for identifying feature points and detecting forming dimensions of cladding channels in arc additive manufacturing based on 3D visual full morphology point cloud information. This is to solve the problems that existing methods are not suitable for high-layer applications, cannot identify the real point cloud information of the biting points on both sides of the incomplete fusion zone between deposition layers, and have depth-of-field error and breakpoint error.

[0007] To achieve the above objectives, the following technical solution is proposed:

[0008] A method for detecting forming dimensions in additive manufacturing cladding channels based on feature point recognition of full-topography point clouds, comprising the following steps:

[0009] S1, collect the original 3D point cloud data of each sedimentary layer;

[0010] S2, use a statistical filtering algorithm to filter out noise points from the original three-dimensional point cloud data of each deposition layer described in S1 to obtain filtered point cloud data;

[0011] S3, perform surface 3D reconstruction, uniform refinement of point cloud data, and fitting and alignment of point cloud features in adjacent layers on the filtered point cloud data described in S2 to obtain effective point cloud data;

[0012] S4, for the effective point cloud data described in S3, identify the weld toe feature points and height feature points in the first layer point cloud data, and calculate the width and height dimensions;

[0013] S5, for the effective point cloud data described in S3, identify the biting points and height feature points representing the incomplete fusion area between layers in the nth layer of point cloud data, where n>1, and calculate the biting width, maximum width and height dimensions;

[0014] In a preferred embodiment, the effective point cloud data of S4 and S5 are subjected to dimensionality reduction processing, wherein the first layer weld toe feature points and the nth layer interlayer bite points in S4 and S5 are extracted from cross sections along the additive direction, and each cross section is 1 mm apart along the additive direction.

[0015] S6, obtain the forming size data of the cladding channel.

[0016] In a preferred embodiment, step S4 specifically includes:

[0017] S4-1, Determine two weld toe feature points in the cross section, and the current cross section cladding width is the distance between the two weld toe feature points;

[0018] S4-2, the point with the maximum height in the search section is taken as the vertex of the cladding channel in the current layer of this section. The line connecting the vertices of each section is the center line of the cladding channel in the current layer, and the distance from the vertex to the line connecting the two weld toe feature points is taken as the height of the first layer of cladding channel.

[0019] S4-3 records the cladding channel height, maximum cladding channel width, undercut width, and vertex coordinate data for each section in the first layer.

[0020] In a preferred embodiment, in step S4-1, since the point cloud data contains substrate point cloud information, a bidirectional approximation search is performed based on a dynamic sliding window using a Savitzky-Golay filter and combined with a gradient curvature threshold method to separate feature points representing solder toes from the substrate point cloud. The method is as follows:

[0021]

[0022] Where N represents the set of all original data points in the current section, and i represents the index of each point; x[c n +j] represents the center point c in the nth window. n The data points are shifted forward by j units; c j f[c] represents the coefficients for polynomial fitting of each point within the window, used to smooth the data points; m represents the width of the window, whose value changes dynamically and is related to the total number of point clouds in the current section; ... n ] represents the center index c of each filtered window. nThe values ​​of data points can reduce noise interference and more accurately capture the trend of data changes; G(c n ,j) and C(c n j) represent the central index c n Gradient and curvature values ​​in this window; G tolerence and C tolerance These represent the tolerances for gradient and curvature, respectively. When a point that meets the gradient and curvature tolerances is found by bidirectionally approaching the center from both sides of the weld cladding path, these two points are recorded as weld toe feature points.

[0023] In a preferred embodiment, step S5 specifically includes:

[0024] S5-1, extract the point with the maximum height in the cross section as the vertex of the cladding channel in the current layer of this cross section, where the height of this layer is the difference between the height of the vertex in the current cross section and the height of the vertex in the previous layer, and the line connecting the vertices of each cross section is the center line of the cladding channel in the current layer.

[0025] S5-2, Identify edge convex and non-edge convex regions, and perform heterogeneous region adaptive Hermite interpolation on the (n-1)th layer point cloud data;

[0026] S5-3, the two points where the interpolation contours of the nth layer and the (n-1th)th layer intersect are identified by the layer intersection point dynamic positioning method, and all point clouds between the two intersection points are extracted and stored in an array. The width of the cladding channel of the nth layer in the current section is the maximum distance between any two points in the array, and the width of the bite edge on both sides of the cladding channel of the nth layer in the current section is the distance between the two intersection points.

[0027] S5-4 records the cladding run height, cladding run width, undercut width, and vertex coordinate data for each cross section in the (n-1)th layer;

[0028] S5-5, The current layer intersection identification and size detection are completed, and we are ready to process the next layer of point cloud data.

[0029] In a preferred embodiment, in step S5-2, since the surface of the cladding channel is not smooth due to oxidized welding slag, and the fitting method ignores the small defect features in the point cloud data contour surface, the interpolation method is used to completely preserve the defect features of the cladding channel surface.

[0030] In a preferred embodiment, in step S5-2, since small molten droplets may not spread evenly and solidify at the edge of the cladding channel contour (i.e., the edge convex region), this special region has multiple x values ​​in certain specific z values, making conventional interpolation impossible. Therefore, a heterogeneous region adaptive interpolation method is used to search for edge convex and non-edge convex regions, and inverse Hermite interpolation and forward Hermite interpolation are performed respectively, as shown in the following formulas:

[0031]

[0032] Wherein, N1 and N2 represent the point sets belonging to the edge convex and non-edge convex regions, respectively; u a and v a U represents the x and z values ​​of the a-th point in the N1-point set. b and v b g represents the x and z values ​​of the b-th point in the N2 point set; a and g b They represent respectively in v a and u b The gradient value at the given point; h(t) and lk(z) represent the shape function and Lagrange basis function of the Hermite interpolation formula, respectively.

[0033] In a preferred embodiment, in step S5-3, the layer intersection dynamic positioning method utilizes the heterogeneous region adaptive interpolation method and combines it with the idea of ​​bisection to bidirectionally search and compare the interpolation and coordinate values ​​of each current point and the next point. When a point that meets the conditions is found bidirectionally from the middle of the cladding channel to both sides, the two points are recorded as bite feature points.

[0034] Compared to the shortcomings and deficiencies of existing technologies, the beneficial effects of this invention are:

[0035] 1. This invention uses a high-precision 3D vision scanner to collect three-dimensional point cloud data of the full morphology of each deposition layer, and proposes a method to extract the top layer size data from the contour information of the full morphology three-dimensional point cloud. This avoids the problem of distortion in the calculation of the cladding width caused by the inability of two-dimensional line laser sensors to collect the side contour of the cladding channel and insufficient acquisition accuracy.

[0036] 2. This invention proposes a method for rapid identification of weld toe feature points of cladding traces based on 3D visual full-shape point cloud information. By performing dimensionality reduction processing on high-precision three-dimensional point cloud data, the method effectively realizes the segmentation and extraction of the first layer of cladding trace point cloud data from the substrate.

[0037] 3. This invention proposes a rapid identification method for biting points in the incomplete fusion zone between cladding layers based on 3D visual full-topography point cloud information. By using heterogeneous region adaptive interpolation on the point clouds of edge-convex and non-edge-convex regions, the complete information of the cladding surface contour can be effectively preserved. Furthermore, the biting points in the incomplete fusion zone between layers can be effectively identified through the layer intersection point dynamic positioning method. Finally, continuous change data of the morphology of each cladding layer is obtained without manual inspection. This solves the problems of existing methods, such as the inability to identify the morphology data of the incomplete fusion zone between layers, unsuitability for high-layer cases, and the existence of depth-of-field errors and breakpoint errors. This provides data support for achieving precise control of forming dimensions and forming quality judgment in arc additive manufacturing. Attached Figure Description

[0038] Figure 1 This is a schematic diagram of depth-of-field error and breakpoint error in a preferred embodiment of the present invention;

[0039] Figure 2 This is a flowchart of a preferred embodiment of the present invention for identifying feature points and detecting forming dimensions of an arc additive manufacturing cladding track based on 3D visual full-shape point cloud information;

[0040] Figure 3 This is an example diagram of point cloud preprocessing according to a preferred embodiment of the present invention;

[0041] Figure 4 This is an example diagram of the heterogeneous region adaptive interpolation method according to a preferred embodiment of the present invention;

[0042] Figure 5 This is an example diagram showing the detection results of the width of the first cladding layer according to a preferred embodiment of the present invention;

[0043] Figure 6 This is an example diagram showing the detection results of the width dimension of the second cladding channel in a preferred embodiment of the present invention;

[0044] Figure 7 This is an example diagram showing the detection results of the width of the third cladding layer according to a preferred embodiment of the present invention;

[0045] Figure 8 This is an example diagram showing the detection results of the width of the fourth cladding layer according to a preferred embodiment of the present invention;

[0046] Figure 9 This is an example diagram showing the detection results of the width of the fifth cladding layer according to a preferred embodiment of the present invention;

[0047] Figure 10 This is an example diagram of the detection results of the forming size of the cladding channel according to a preferred embodiment of the present invention. Detailed Implementation

[0048] 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.

[0049] like Figure 2 As shown, this embodiment proposes a method for identifying feature points and detecting forming dimensions of the cladding track in arc additive manufacturing based on 3D visual full-topography point cloud information, including the following steps:

[0050] S1, a high-precision 3D vision mesh structured light scanner is installed on the robotic arm and scans all sides of the cladding channel to collect the original three-dimensional point cloud data of each deposition layer and store it in PLY format.

[0051] S2, use a statistical filtering algorithm to filter out noise points from the original three-dimensional point cloud data of each deposition layer described in S1 to obtain filtered point cloud data;

[0052] S3, perform surface 3D reconstruction, uniform refinement of the point cloud data, and feature fitting and alignment of adjacent layer point cloud data on the filtered point cloud data described in S2 to obtain effective point cloud data, such as... Figure 3 As shown;

[0053] S4, perform dimensionality reduction processing on the effective point cloud data described in S3, extract point cloud data from each cross-section of the effective point cloud along a section perpendicular to the additive direction, with each cross-section spaced 1mm apart along the additive direction; identify weld toe feature points and height feature points in the first layer point cloud data, and calculate the width and height dimensions, such as... Figure 5 and Figure 10 As shown;

[0054] S4-1, based on the DW-GCT (Dynamic Sliding Window-Gradient Curvature Threshold) method, a bidirectional approximation search is performed on two weld toe feature points in the cross section, and the current cross section cladding width is the distance between the two weld toe feature points;

[0055] S4-2, search for the point with the maximum height in the cross section as the vertex of the cladding channel in the current layer of this cross section, the line connecting the vertices of each cross section is the center line of the cladding channel in the current layer, and the distance from the vertex to the line connecting the two weld toe feature points is taken as the height of the first layer of cladding channel;

[0056] S4-3, record the cladding channel height, maximum cladding channel width, undercut width, and vertex coordinate data for each section in the first layer;

[0057] S4-4, The identification and size detection of the weld toe feature points in the first layer are completed, and we are ready to process the point cloud data in the next layer.

[0058] According to the above-mentioned method for identifying feature points and detecting forming dimensions of weld toes in arc additive manufacturing based on 3D visual full-shape point cloud information, in step S4-1, since the point cloud data includes substrate point cloud information, the feature points representing weld toes are separated from the substrate point cloud by using a dynamic sliding window based on the Savitzky-Golay filter and combined with the bidirectional approximation search of the GCT method. The DW-GCT method formula is as follows:

[0059]

[0060] Where N represents the set of all original data points in the current section, and i represents the index of each point; x[c n +j] represents the center point c in the nth window. n The data points are shifted forward by j units; c j f[c] represents the coefficients for polynomial fitting of each point within the window, used to smooth the data points; m represents the width of the window, whose value changes dynamically and is related to the total number of point clouds in the current section; ... n ] represents the center index c of each filtered window. n The values ​​of data points can reduce noise interference and more accurately capture the trend of data changes; G(c n ,j) and C(c n j) represent the central index c n Gradient and curvature values ​​in this window; G tolerence and C tolerance These represent the tolerances for gradient and curvature, respectively. When a point that meets the gradient and curvature tolerances is found by bidirectionally approaching the center from both sides of the weld cladding path, these two points are recorded as weld toe feature points.

[0061] S5, perform feature point recognition on the cross-sectional point cloud after dimensionality reduction processing described in S4, identify the biting points and height feature points representing the incomplete fusion area between layers in the nth layer (n>1) point cloud data, and calculate the biting width, maximum width and height dimensions;

[0062] S5-1, extract the point with the maximum height in the cross section as the vertex of the cladding channel in the current layer of this cross section, where the height of this layer is the difference between the height of the vertex in the current cross section and the height of the vertex in the previous layer, and the line connecting the vertices of each cross section is the center line of the cladding channel in the current layer.

[0063] S5-2, Identify edge convex and non-edge convex regions, and perform heterogeneous region adaptive Hermite interpolation on the (n-1)th layer (n>1) point cloud data;

[0064] S5-3, the two points where the interpolation functions of the nth layer and the (n-1th)th layer intersect are identified by the layer intersection point dynamic positioning method, and all point clouds between the two intersection points are extracted and stored in an array. The width of the cladding channel of the nth layer in the current section is the maximum distance between any two points in the array, and the width of the bite edge on both sides of the cladding channel of the nth layer in the current section is the distance between the two intersection points.

[0065] S5-4 records the cladding run height, cladding run width, undercut width, and vertex coordinates of each cross-section in the nth (n>1) layer, such as... Figure 6 , Figure 7 , Figure 8 , Figure 9 , Figure 10 As shown;

[0066] S5-5, The current layer intersection identification and size detection are completed, and we are ready to process the next layer of point cloud data.

[0067] According to the above-mentioned method for identifying feature points and detecting forming dimensions of arc additive manufacturing cladding channels based on 3D visual full-shape point cloud information, in step S5-2, since the surface of the cladding channel is not smooth due to oxidized welding slag, and the fitting method will ignore the small defect features in the surface of the point cloud data contour, the interpolation method is used to completely preserve the defect features of the cladding channel surface.

[0068] According to the above-mentioned method for identifying feature points and detecting forming dimensions of cladding channels in arc additive manufacturing based on 3D visual full-shape point cloud information, in step S5-2, because small molten droplets may not be evenly spread and solidify at the edge of the cladding channel contour, i.e., the edge convex region, this special region has multiple x values ​​in certain specific z values, so conventional interpolation cannot be performed, such as... Figure 4 As shown. Therefore, the heterogeneous region adaptive interpolation method is used to search for edge convexity and non-edge convexity regions, and parametric inverse Hermite interpolation and forward Hermite interpolation are performed respectively, as shown in the following formulas:

[0069]

[0070] Wherein, N1 and N2 represent the point sets belonging to the edge convex and non-edge convex regions, respectively; u a and v a U represents the x and z values ​​of the a-th point in the N1-point set. b and v b g represents the x and z values ​​of the b-th point in the N2 point set; a and g b They represent respectively in v a and u b The gradient value at the given point; h(t) and lk(z) represent the shape function and Lagrange basis function of the Hermite interpolation formula, respectively.

[0071] According to the above-mentioned method for identifying feature points and detecting forming dimensions of cladding channels in arc additive manufacturing based on 3D visual full-shape point cloud information, in step S5-3, the dynamic positioning method of the layer intersection point utilizes the heterogeneous region adaptive interpolation method and combines the idea of ​​bisection to search and compare the interpolation and coordinate values ​​of each current point and the next point in both directions. When a point that meets the conditions is found in both directions from the middle of the cladding channel to both sides, the two points are recorded as edge bite feature points.

[0072] S6, obtain the forming dimension data of the first layer of cladding channel and the top layer of cladding channel.

[0073] The above embodiments are only used to illustrate the technical solutions of the present invention and are not intended to limit it. The scope of protection of the present invention should not be limited to the details of the above embodiments. Although the present invention has been described in detail with reference to preferred embodiments, those skilled in the art should understand that any modifications, equivalent substitutions and improvements made to the technical solutions of the present invention without departing from the spirit and scope of the technical solutions of the present invention should be covered within the scope of the claims of the present invention.

Claims

1. A method for detecting forming dimensions based on feature point recognition of cladding channels in additive manufacturing using full-topography point clouds, the method comprising the following steps: S1, collect the original 3D point cloud data of each sedimentary layer; S2, use a statistical filtering algorithm to filter out noise points from the original three-dimensional point cloud data of each deposition layer described in S1 to obtain filtered point cloud data; S3, perform surface 3D reconstruction, uniform refinement of point cloud data, and fitting and alignment of point cloud features in adjacent layers on the filtered point cloud data described in S2 to obtain effective point cloud data; S4, for the effective point cloud data described in S3, identify the weld toe feature points and height feature points in the first layer point cloud data, and calculate the width and height dimensions; S5, for the effective point cloud data described in S3, identify the biting points and height feature points representing the incomplete fusion area between layers in the nth layer of point cloud data, where n > 1, and calculate the biting width, maximum width and height dimensions; S5-1, extract the point with the maximum height in the cross section as the vertex of the cladding channel in the current layer of this cross section, where the height of this layer is the difference between the height of the vertex in the current cross section and the height of the vertex in the previous layer, and the line connecting the vertices of each cross section is the center line of the cladding channel in the current layer. S5-2 identifies edge convex and non-edge convex regions and performs heterogeneous region adaptive Hermite interpolation on the (n-1)th layer point cloud data; S5-3, the two points where the interpolation contours of the nth layer and the (n-1th)th layer intersect are identified by the layer intersection point dynamic positioning method, and all point clouds between the two intersection points are extracted and stored in an array. The width of the cladding channel of the nth layer in the current section is the maximum distance between any two points in the array, and the width of the bite edge on both sides of the cladding channel of the nth layer in the current section is the distance between the two intersection points. S5-4 records the cladding run height, cladding run width, undercut width, and vertex coordinate data for each cross section in the (n-1)th layer; S5-5, The current layer's intersection point identification and size detection are completed, and preparation is made for the next layer of point cloud data processing; S6, obtain the forming dimension data of the cladding channel; In step S5-2, the heterogeneous region adaptive interpolation method is used to search for edge convexity and non-edge convexity regions, and inverse Hermite interpolation and forward Hermite interpolation are performed respectively, as shown in the following formulas: ； where N1 and N2 represent the point sets belonging to the edge convex and non-edge convex regions, respectively; u a and v a represent the x and z values of the a-th point in the point set N1, u b and v b represent the x and z values of the b-th point in the point set N2; g a and g b represent the gradient values at v a and u b , respectively; h(t) represents the shape function of the Hermite interpolation formula.

2. The forming dimension detection method based on feature point recognition of additive manufacturing cladding channels using full-topography point clouds as described in claim 1, characterized in that, The effective point cloud data of S4 and S5 are further subjected to dimensionality reduction processing. The first layer weld toe feature points and the nth layer interlayer bite points in S4 and S5 are extracted from the cross sections along the direction perpendicular to the additive direction, and each cross section is 1 mm apart along the additive direction.

3. The forming dimension detection method based on feature point recognition of additive manufacturing cladding channels using full-topography point clouds as described in claim 1, characterized in that, Step S4 specifically includes: S4-1, Determine two weld toe feature points in the cross section. The current cross section cladding width is the distance between the two weld toe feature points. S4-2, the point with the maximum height in the search section is taken as the vertex of the cladding channel in the current layer of this section. The line connecting the vertices of each section is the center line of the cladding channel in the current layer, and the distance from the vertex to the line connecting the two weld toe feature points is taken as the height of the first layer of cladding channel. S4-3 records the cladding channel height, maximum cladding channel width, undercut width, and vertex coordinate data for each section in the first layer.

4. The forming dimension detection method based on feature point recognition of additive manufacturing cladding channels using full-topography point clouds as described in claim 1, characterized in that, In step S5-2, since the surface of the cladding channel is not smooth due to the oxidation of welding slag, and the fitting method ignores the small defect features in the surface of the point cloud data contour, the interpolation method is used to completely preserve the defect features of the cladding channel surface.

5. The forming dimension detection method based on feature point recognition of additive manufacturing cladding channels using full-topography point clouds as described in claim 1, characterized in that, In step S5-3, the layer intersection dynamic positioning method uses the heterogeneous region adaptive interpolation method and combines the idea of ​​bisection to search and compare the interpolation and coordinate values ​​of each current point and the next point in both directions. When a point that meets the conditions is found in both directions from the middle of the cladding channel to both sides, the two points are recorded as edge bite feature points.

6. The forming dimension detection method based on feature point recognition of additive manufacturing cladding channels using full-topography point clouds as described in claim 3, characterized in that, In step S4-1, since the point cloud data contains substrate point cloud information, a bidirectional approximation search is performed based on a dynamic sliding window using a Savitzky-Golay filter and combined with a gradient curvature threshold method to separate feature points representing solder toes from the substrate point cloud. The method is as follows: ； ； ； Where N represents the set of all original data points in the current section, and i represents the index of each point; x[c n +j] represents the center point c in the nth window. n The data points are shifted forward by j units; c j This represents the coefficients used for polynomial fitting of each point within the window, used to smooth the data points; m represents the width of the window, whose value changes dynamically and is related to the total number of point clouds in the current section. c represents the center index of each filtered window n The value of the data point; G(c n ,j) and C(c n j) represent the central index c n Gradient and curvature values ​​in this window; G tolerence and C tolerance These represent the tolerances for gradient and curvature, respectively. When a point that meets the gradient and curvature tolerances is found by bidirectionally approaching the center from both sides of the weld cladding path, these two points are recorded as weld toe feature points.

Images