Finned tube welding seam feature extraction method and system based on three-dimensional point cloud

Abstract

The invention discloses a three-dimensional point cloud-based finned tube welding seam feature extraction method and system, and the method comprises the following steps: correspondingly moving a visual sensor to different regions, and correspondingly obtaining the three-dimensional point cloud data of the different regions; splicing the three-dimensional point cloud data to obtain an original point cloud model of the finned tube heat exchange unit; an initial main canal body point cloud is separated from the original point cloud model, and the axis direction of a main canal body is determined; performing axial filtering on the initial main canal body point cloud to obtain an axial main canal body point cloud, and obtaining a complete main canal body geometric model; and determining a virtual welding seam track. According to the method, three-dimensional point cloud data processing and a mathematical modeling method are combined, a complete and accurate finned tube model and a virtual welding seam track can be automatically obtained, an operator can conveniently optimize welding seam characteristic parameters according to the virtual welding seam track so as to form an optimized welding seam track, the conditions of insufficient welding, undercut and the like are effectively avoided, and the working efficiency is improved. Therefore, the welding quality is improved.

Description

Technical Field

[0001] The invention belongs to the technical field of laser processing, and in particular to a method and system for extracting a fin tube weld based on three-dimensional point cloud. Background Art

[0002] like Figure 1 As shown, the core heat exchange device of the air-temperature vaporizer is a finned tube heat exchange unit, and its main structure includes: a finned tube made of aluminum alloy and a main pipe body 2, wherein the finned tube includes a finned tube body 1' and fins 1 connected to the outer peripheral surface of the finned tube body 1', the finned tube body 1' is sleeved on the main pipe body 2, and then welded by laser welding or the like to form a weld 3 at the end of the finned tube body 1'.

[0003] When the fin tube body 1 ′ and the main tube body 2 are welded by laser welding, how to accurately identify the weld and extract the weld features is the key to ensuring the welding quality.

[0004] However, since the number and shape of the fins 1 connected to the outer peripheral surface of the fin tube body 1' can be adjusted, the overall structure of the fin tube is complicated, and when the sensor collects image data of the weld 3 along a certain direction, the fin tube body 1' and the main body 2 (such as Figure 1 As shown, the main pipe 2 can be a U-shaped structure with a fin tube respectively connected at both ends. The volume and shape of the main pipe 2 will block the sensor's line of sight, causing the sensor to only capture the morphology of the side facing the sensor's line of sight, and unable to capture the back side of the weld 3 (i.e., the area between the two free ends of the U-shaped main pipe 2), and further unable to extract the complete weld features. Summary of the invention

[0005] To solve the above problems, the present invention proposes a fin tube weld feature extraction method and system based on three-dimensional point cloud, which combines three-dimensional point cloud data processing and mathematical modeling methods based on the structural characteristics of the fin tube, and can automatically obtain a complete and accurate fin tube model and a virtual weld trajectory, so that the operator can optimize the weld feature parameters according to the virtual weld trajectory to form an optimized weld trajectory, so as to effectively avoid the occurrence of false welds, undercuts and the like, thereby improving the welding quality.

[0006] To achieve the above object, the technical solution adopted by the present invention is:

[0007] On the one hand, a fin tube weld extraction method based on three-dimensional point cloud is provided, which comprises the following steps:

[0008] The visual sensor is moved to the upper area, the outer area and the lower area of ​​the weld respectively, and the upper three-dimensional point cloud data, the outer three-dimensional point cloud data and the lower three-dimensional point cloud data are obtained respectively through the visual sensor;

[0009] The upper three-dimensional point cloud data, the outer three-dimensional point cloud data, and the lower three-dimensional point cloud data are all preprocessed, and all the preprocessed three-dimensional point cloud data are spliced ​​to obtain the original point cloud model of the fin-tube heat exchange unit;

[0010] Separating an initial main body point cloud from the original point cloud model, and determining the axis direction of the main body according to the initial main body point cloud;

[0011] Perform axial filtering on the initial main body point cloud according to the axial direction of the main body to obtain an axial main body point cloud, and perform fitting according to the axial main body point cloud to obtain a complete main body geometric model;

[0012] The virtual weld plane is intersected with the complete main pipe model, and the intersection is determined as the virtual weld trajectory.

[0013] In another aspect, a fin tube weld extraction system is provided, comprising:

[0014] Vision sensors;

[0015] A mobile device connected to the visual sensor, used to drive the visual sensor to move to different positions and obtain three-dimensional point cloud data of different areas through the visual sensor;

[0016] An original point cloud acquisition unit, which is used to preprocess the upper three-dimensional point cloud data, the outer three-dimensional point cloud data, and the lower three-dimensional point cloud data, and to splice all the preprocessed three-dimensional point cloud data to obtain an original point cloud model of the fin-tube heat exchange unit;

[0017] A point cloud separation unit, connected to the original point cloud acquisition unit, for separating an initial main body point cloud from the original point cloud model;

[0018] An axial determination unit, used to determine the axial direction of the main body according to the initial main body point cloud;

[0019] A main body point cloud acquisition unit, which is used to perform axial filtering on the initial main body point cloud according to the axial direction of the main body to obtain an axial main body point cloud;

[0020] A main body model acquisition unit, which is used to perform fitting according to the axial main body point cloud to obtain a complete main body geometric model;

[0021] A virtual weld trajectory determination unit, which is used to determine the virtual weld trajectory according to the center point c of the complete main pipe geometric model;

[0022] And a display unit, which is used to display a virtual weld trajectory and to form an optimized weld trajectory after optimizing weld characteristic parameters according to the virtual weld trajectory.

[0023] The beneficial effects of using the present invention are:

[0024] The present invention firstly obtains the three-dimensional point cloud data of the fin tube through a visual sensor, and further combines the three-dimensional point cloud data processing with the mathematical modeling method, so as to automatically obtain a complete and accurate fin tube model and a virtual weld trajectory, and obtain the weld characteristics, thereby optimizing the weld extraction accuracy, and visually displaying the virtual weld trajectory, so as to facilitate the operator to optimize the weld characteristic parameters according to the virtual weld trajectory to form an optimized weld trajectory, making it closer to the actual fin tube assembly situation, and adjusting the laser welding parameters according to the optimized weld trajectory to effectively avoid the occurrence of false welding, undercutting and the like, thereby improving the welding quality. BRIEF DESCRIPTION OF THE DRAWINGS

[0025] Figure 1 It is a structural schematic diagram of a fin tube heat exchange unit in the prior art;

[0026] Figure 2 It is a schematic diagram of the steps of the fin tube weld extraction method of the present invention;

[0027] Figure 3 A schematic diagram of the state of the visual sensor in the present invention acquiring upper three-dimensional point cloud data in the area above the weld;

[0028] Figure 4 A schematic diagram of the state of the visual sensor in the present invention acquiring the outer three-dimensional point cloud data in the outer area of ​​the weld;

[0029] Figure 5 It is a schematic diagram of the state of the visual sensor in the present invention acquiring the lower three-dimensional point cloud data in the area below the weld;

[0030] Figure 6 is the original point cloud model of the fin-tube heat exchange unit in the present invention;

[0031] Figure 7 The initial main tube point cloud and the initial fin tube point cloud separated from the original point cloud model in the present invention;

[0032] Figure 8 It is a schematic diagram of the state of determining the axial direction according to the initial main pipe point cloud in the present invention;

[0033] Fig. 9 is the point cloud of the axial main pipe in the present invention;

[0034] Fig.10 It is a complete geometric model of the main pipe in the present invention;

[0035] Fig.11 A schematic diagram of a virtual weld trajectory and an optimized weld trajectory in the present invention;

[0036] Fig.12 It is a structural schematic diagram of the fin tube weld extraction system in the present invention. DETAILED DESCRIPTION

[0037] In order to make the purpose, technical solution and advantages of the technical solution more clear, the technical solution is further described in detail below in conjunction with specific implementation methods. It should be understood that these descriptions are only exemplary and are not intended to limit the scope of the technical solution.

[0038] Embodiment 1:

[0039] The present embodiment provides a fin tube weld extraction method based on three-dimensional point cloud, which is applicable to a fin tube heat exchange unit of an air-temperature vaporizer, and the fin tube heat exchange unit comprises: a fin tube and a main pipe 2, wherein the fin tube comprises a fin tube body 1' and fins 1 connected to the outer peripheral surface of the fin tube body 1', the fin tube body 1' is sleeved on the main pipe 2, and then welded by laser welding or the like to form a weld 3 at the end of the fin tube body 1', and the main pipe 2 is a U-shaped structure as a whole, and a fin tube is sleeved at each of the two ends.

[0040] On this basis, if Figure 2-5 As shown, the fin tube weld extraction method comprises the following steps:

[0041] S1. Connect the visual sensor 100 to a mobile device (such as a robot), so that the visual sensor 100 can be moved by the mobile device; in this embodiment, the visual sensor 100 includes a line laser visual sensor, and can generate a fan-shaped laser light plane 101;

[0042] like Figure 3 As shown, the visual sensor 100 is moved to the upper area of ​​the weld 3 by a mobile device, and is simultaneously located in the upper area of ​​the fin tube, and the upper area of ​​the fin tube heat exchange unit is scanned by the visual sensor 100 to obtain upper three-dimensional point cloud data;

[0043] S2, such as Figure 4 As shown, the visual sensor 100 is moved to the outer area of ​​the weld 3 by a mobile device, and is located at the outer area of ​​the fin tube at the same time, and the outer area of ​​the fin tube heat exchange unit is scanned by the visual sensor 100 to obtain the outer three-dimensional point cloud data;

[0044] In this embodiment, the area between the two free ends of the U-shaped main tube 2 is the "inside", and the direction opposite thereto is the "outside";

[0045] S3, such as Figure 5As shown, the visual sensor 100 is moved to the area below the weld 3 by a mobile device, and is simultaneously located in the area below the fin tube, and the visual sensor 100 is used to scan the lower area of ​​the fin tube heat exchange unit to obtain the lower three-dimensional point cloud data;

[0046] It should be noted that, in any scanning process, the laser light plane 101 is perpendicular to the central axis of the weld 3, and the upper three-dimensional point cloud data, the outer three-dimensional point cloud data, and the lower three-dimensional point cloud data all include three-dimensional point cloud data of a part of the fin tube body 1', a part of the fin 1, a part of the main body 2, and a part of the weld 3;

[0047] S4, preprocessing the upper three-dimensional point cloud data, the outer three-dimensional point cloud data, and the lower three-dimensional point cloud data, and splicing all the preprocessed three-dimensional point cloud data to obtain the original point cloud model of the fin tube heat exchange unit (such as Figure 6 As shown), wherein the data preprocessing includes one or more of denoising, under-sampling, etc.; thus, the amount of data can be reduced by data preprocessing, the efficiency of subsequent data processing can be improved, and sufficient geometric information can be retained;

[0048] S5, separating the initial main tube point cloud and the initial fin tube point cloud from the original point cloud model, specifically comprising the following steps:

[0049] Construct the point cloud density histogram in the Z direction and calculate the point cloud density gradient according to formula (1);

[0050]

[0051] Among them, ▽H(z i ) is the point cloud density gradient in the Z direction; H(z i ) is the point cloud density histogram in the Z direction; Δz is the width of the Z coordinate interval; thus, the original point cloud model can be filtered in the Z direction by calculating the point cloud density gradient to reduce point cloud interference; in this embodiment, the Z direction is the axial direction of the assumed main body 2, but due to the different placement positions of the tube body, the Z direction may not be consistent with the actual axial direction of the main body 2, so the axial direction of the main body 2 needs to be clearly determined later;

[0052] And, the demarcation threshold (i.e., the maximum gradient point) between the fin tube and the main body 2 is determined according to the point cloud density gradient, and the initial main body point cloud (i.e., Figure 7 (a) in the figure) and the initial finned tube point cloud (i.e. Figure 7 (b) of the Regulations);

[0053] Furthermore, the method for constructing the point cloud density histogram includes the following steps: setting the Z coordinate interval width Δz, for example, in this embodiment, setting Δz=0.1; counting the i-th interval [z min +(i-1)Δz,z min +iΔz]; repeat the above steps until all points are completed. By counting the point cloud data of each interval, we can get the point cloud density histogram H(z);

[0054] S6, determining the axial direction (i.e., axial direction) of the main body 2 according to the initial main body point cloud, which specifically includes the following steps:

[0055] The initial main body point cloud is filtered, comprising the following steps:

[0056] Constructing a point cloud density histogram of the initial main body point cloud in the X direction and the Y direction, and calculating the point cloud density gradient in the X direction and the Y direction respectively, so as to complete the X direction filtering and the Y direction filtering, so as to minimize the interference of the fin tube part on the main body point cloud, wherein the construction method of the point cloud density histogram and the calculation method of the point cloud density gradient are the same as step S5, only the filtering direction is different, and no further description is given; and the X direction and the Y direction are both perpendicular to the Z direction;

[0057] According to formula (2), in the initial main body point cloud after X-direction filtering and Y-direction filtering, each point p i The local neighborhood covariance matrix of :

[0058]

[0059] Where C is the covariance matrix; N(p i ) is the initial main body point cloud after X-direction filtering and Y-direction filtering. i The neighborhood point set of ; is the neighborhood centroid; p j is a point in the neighborhood point set;

[0060] Point p i The covariance matrix of is decomposed into eigenvalues, and the eigenvector corresponding to the minimum eigenvalue is used as the normal vector N of the point i , thereby obtaining the normal vector of each point and constructing the normal vector matrix N;

[0061] According to formula (3), the normal vector matrix N is subjected to singular value decomposition (SVD), and a vector orthogonal to all normal vectors is determined based on the decomposition result, and the vector is determined as the axial unit direction vector d of the main body 2.axis , and the axial unit direction vector d axis The vector direction is the axis direction of the main tube 2:

[0062] N=UΣV T (3)

[0063] Among them, U and V are orthogonal matrices; ∑ is a diagonal matrix, the elements on its diagonal are singular values, and the right singular vector V[:,end] corresponding to the minimum singular value is the vector that is orthogonal to all normal vectors;

[0064] Since SVD can decompose the normal vector matrix into orthogonal basis vectors, where the basis vector direction data corresponding to the minimum singular value has the smallest change, this vector is considered to be a vector orthogonal to all normal vectors, that is, the unit direction vector d of the axis of the main body 2 axis ;

[0065] S7, according to the axial direction of the main body 2, axial filtering is performed on the initial main body point cloud after the X-direction filtering and the Y-direction filtering to obtain the axial main body point cloud, which specifically includes the following steps:

[0066] According to formula (4), the projection length Length of each point cloud data P in the axis direction of the main body 2 in the initial main body point cloud after X-direction filtering and Y-direction filtering is obtained. proj ;

[0067] Length proj =P·d axis (4)

[0068] Construct projection length Length proj Point cloud density histogram and calculate the point cloud density gradient;

[0069] And, the demarcation threshold (i.e., the maximum gradient point) between the fin tube and the main body 2 is determined according to the point cloud density gradient, and the axial fin tube point cloud (e.g., Fig. 9 As shown); wherein, the method for constructing the point cloud density histogram and the method for calculating the point cloud density gradient are the same as those in step S5, only the filtering direction is different, and will not be described in detail;

[0070] from Fig. 9 It can be seen that after axial filtering, the point cloud of the finned tube part is significantly reduced compared with the initial main tube point cloud, which can further eliminate the point cloud interference of the finned tube part, making the subsequent weld extraction results more accurate;

[0071] S8. Fitting is performed according to the axial main pipe point cloud to obtain a complete main pipe geometric model, which specifically includes the following steps:

[0072] S81, solving the position of the center point c based on the axis direction of the main tube 2, which specifically includes the following steps:

[0073] The median value in the X direction and the median value in the Y direction of the point cloud data of the axial main body point cloud are used as the initial guess point c0, and there are;

[0074]

[0075] Among them, x max 、x min are the maximum and minimum values ​​of the point cloud data of the axial main body point cloud in the X direction; max ,y min They are the maximum and minimum values ​​of the point cloud data of the axial main tube point cloud in the Y direction respectively;

[0076] The distance variance from each point in the axial main body point cloud to the initial guess point c0 is defined as the objective function;

[0077] The objective function is minimized using the Nelder-Mead simplex algorithm until the simplex algorithm converges, and the output of the objective function when the algorithm converges is the position coordinate of the center point c;

[0078] In addition, the position of the center point c can be iteratively optimized by the loss function L, where and Among them, ||p i -c|| is the point p in the point cloud of the axial main tube i Euclidean distance to the center point c;

[0079] S82, such as Fig.10 As shown, the axis passing through the center point c and in the direction of the axis of the main body 2 is determined as the fitting axis of the complete main body;

[0080] S83, obtaining the radius r of the complete main body model. Specifically, the radius r of the complete main body is determined by the following formula:

[0081]

[0082] Where v is the vector of the fitting axis;

[0083] And S84, construct a complete main pipe model according to the fitting axis and radius r (such as Fig.10 shown);

[0084] S9, determine the plane passing through the center point c and with the normal vector v as the virtual weld plane, and intersect the virtual weld plane with the complete main body model, and the intersection is the virtual weld trajectory (i.e. Fig.11The geometric features of the virtual weld track (including center point coordinates, radius, etc.) are the weld features.

[0085] Furthermore, the weld characteristic parameters can be optimized according to the virtual weld trajectory to form an optimized weld trajectory (i.e. Fig.11 The green track in the figure is used to make it closer to the actual fin tube assembly situation, and the laser welding parameters are adjusted according to the optimized weld trajectory to effectively avoid the occurrence of false welds, undercuts, etc., thereby improving the welding quality.

[0086] Embodiment 2:

[0087] The difference between this embodiment and embodiment 1 is that the objective function is minimized using the Nelder-Mead simplex algorithm, which specifically includes the following steps:

[0088] Calculate the objective function value at each vertex in the initial two-dimensional simplex, and sort the vertices according to the objective function value. The initial two-dimensional simplex consists of an initial guess point c0, an x-direction offset point, and a y-direction offset point, a total of 3 vertices, wherein the initial guess point c0 and the x-direction offset point have the same y-direction coordinates and axial coordinates, and the difference in the x-direction coordinates is less than or equal to the x-direction offset threshold dx; the initial guess point c0 and the y-direction offset point have the same x-direction coordinates and axial coordinates, and the difference in the y-direction coordinates is less than or equal to the y-direction offset threshold dy. For example, in this embodiment, dx=[max(x max -x min )] / 100,dy=[max(y max -y min )] / 100;

[0089] Calculate the centroid G of the other two vertices (i.e., vertex v1 and vertex v2) in the current initial two-dimensional simplex except the vertex H with the largest objective function value, and

[0090] Reflect the vertex H with the largest objective function value through the center of gravity G to obtain the reflection point R, and R = G + α (GH), α ranges from [0.5, 1];

[0091] Get the objective function value of the reflection point R. If the objective function value of the reflection point R is the minimum value among all the current objective function values, then get the expansion point E, and E = G + γ (RG), and the value range of γ is [1, 2]. If the objective function value of the reflection point R is the maximum value among all the current objective function values, then get the contraction point C, and C = G + ρ (HG), and the value range of ρ is [0.5, 1]. If the objective function values ​​of the reflection point R and the contraction point C are both greater than or equal to the minimum value of all the current objective function values, then according to v i=L+σ(v i -L) for overall contraction, where L is the vertex with the smallest objective function value in the initial two-dimensional simplex (i.e., the vertex with the smallest objective function value among vertex H, vertex v1, and vertex v2), and the value range of σ is [0.5, 1];

[0092] Repeat the above steps to adjust the position and shape of the simplex until the Euclidean distance between any two vertices in the simplex meets the preset conditions (such as the maximum value of all Euclidean distances is less than 0.01), which means that the simplex algorithm converges. At this time, the output result of the objective function is the position coordinates of the center point c.

[0093] Embodiment 3:

[0094] This embodiment provides a fin tube weld extraction system for implementing the fin tube weld extraction method described in Embodiment 1 or 2, such as Fig.12 As shown, it includes:

[0095] Visual sensor 100;

[0096] A mobile device 200, which is connected to the visual sensor 100, and is used to drive the visual sensor 100 to move to different positions, and obtain three-dimensional point cloud data of different areas through the visual sensor 100, and the process is the same as steps S1-S3;

[0097] The original point cloud acquisition unit 300 is used to preprocess the upper three-dimensional point cloud data, the outer three-dimensional point cloud data, and the lower three-dimensional point cloud data, and to splice all the preprocessed three-dimensional point cloud data to obtain the original point cloud model of the fin-tube heat exchange unit, and the process is the same as step S4;

[0098] A point cloud separation unit 400, which is connected to the original point cloud acquisition unit 300 and is used to separate the initial main tube point cloud and the initial fin tube point cloud from the original point cloud model, and the process is the same as step S5;

[0099] An axial determination unit 500 is used to determine the axial direction (i.e., axial direction) of the main body 2 according to the initial main body point cloud, and the process is the same as step S6;

[0100] The main body point cloud acquisition unit 600 is used to perform axial filtering on the initial main body point cloud according to the axial direction of the main body 2 to obtain the axial main body point cloud, and the process is the same as step S7;

[0101] The main body model acquisition unit 700 is used to perform fitting according to the axial main body point cloud to obtain a complete main body geometric model, and the process is the same as step S8;

[0102] A virtual weld trajectory determining unit 800 is used to determine a virtual weld trajectory according to the center point c of the complete main pipe geometric model, and the process is the same as step S9;

[0103] And a display unit 900, which is used to display a virtual weld trajectory, and to optimize weld characteristic parameters according to the virtual weld trajectory to form an optimized weld trajectory.

[0104] In summary, based on the structural characteristics of the fin tube, the present invention first obtains the three-dimensional point cloud data of the fin tube through a visual sensor, and further combines the three-dimensional point cloud data processing and the mathematical modeling method, so as to automatically obtain a complete and accurate fin tube model and a virtual weld trajectory, and obtain the weld characteristics, thereby optimizing the weld extraction accuracy, and visually displaying the virtual weld trajectory, so that the operator can optimize the weld characteristic parameters according to the virtual weld trajectory to form an optimized weld trajectory, so that it is closer to the actual fin tube assembly situation, and adjust the laser welding parameters according to the optimized weld trajectory to effectively avoid the occurrence of false welding, undercutting and the like, thereby improving the welding quality.

[0105] The above contents are only preferred embodiments of the present invention. For ordinary technicians in this field, many changes can be made in the specific implementation methods and application scopes based on the ideas of the present technical content. As long as these changes do not deviate from the concept of the present invention, they all fall within the scope of protection of this patent.

Claims

1. A fin tube weld extraction method based on three-dimensional point cloud, characterized in that: The steps include: The visual sensor is moved to the upper area, the outer area and the lower area of ​​the weld respectively, and the upper three-dimensional point cloud data, the outer three-dimensional point cloud data and the lower three-dimensional point cloud data are obtained respectively through the visual sensor; The upper three-dimensional point cloud data, the outer three-dimensional point cloud data, and the lower three-dimensional point cloud data are all preprocessed, and all the preprocessed three-dimensional point cloud data are spliced ​​to obtain the original point cloud model of the fin-tube heat exchange unit; Separating an initial main body point cloud from the original point cloud model, and determining the axis direction of the main body according to the initial main body point cloud; Perform axial filtering on the initial main body point cloud according to the axial direction of the main body to obtain an axial main body point cloud, and perform fitting according to the axial main body point cloud to obtain a complete main body geometric model; The virtual weld plane is intersected with the complete main pipe model, and the intersection is determined as the virtual weld trajectory.

2. The fin tube weld extraction method according to claim 1, characterized in that: The vision sensor includes a line laser vision sensor and can generate a fan-shaped laser light plane.

3. The fin tube weld extraction method according to claim 1, characterized in that: Separating the initial main body point cloud from the original point cloud model includes the following steps: Construct the point cloud density histogram in the Z direction and calculate the point cloud density gradient; Furthermore, a demarcation threshold between the fin tube and the main tube is determined according to the point cloud density gradient, and an initial main tube point cloud is separated according to the demarcation threshold.

4. The fin tube weld extraction method according to claim 1, characterized in that: Determining the axis direction of the main body according to the initial main body point cloud comprises the following steps: Performing X-direction filtering and Y-direction filtering processing on the initial main body point cloud; According to the following formula, the initial main body point cloud after X-direction filtering and Y-direction filtering is constructed, and each point p i The local neighborhood covariance matrix of : Where C is the covariance matrix; N(p i ) is the initial main body point cloud after X-direction filtering and Y-direction filtering. i The neighborhood point set of ; is the neighborhood centroid; p j is a point in the neighborhood point set; For point p i The covariance matrix of is decomposed into eigenvalues, and the eigenvector corresponding to the minimum eigenvalue is used as the normal vector N of the point i , to obtain the normal vector of each point and construct the normal vector matrix N; The normal vector matrix N is subjected to singular value decomposition according to the following formula, and a vector orthogonal to all normal vectors is determined based on the decomposition result, and the vector is determined as the axial unit direction vector d of the main body. axis , and the axial unit direction vector d axis The vector direction is the axis direction of the main body: N=UΣV T Among them, U and V are orthogonal matrices; ∑ is a diagonal matrix, the elements on its diagonal are singular values, and the right singular vector V[:,end] corresponding to the minimum singular value is a vector that is orthogonal to all normal vectors.

5. The fin tube weld extraction method according to claim 1, characterized in that: The initial main body point cloud is axially filtered according to the axial direction of the main body to obtain the axial main body point cloud, including the following steps: The projection length Length of each point cloud data P in the axis direction of the main body in the initial main body point cloud is obtained according to the following formula: proj ; Length proj =P·d axis Construct projection length Length proj Point cloud density histogram and calculate the point cloud density gradient; Furthermore, a demarcation threshold between the fin tube and the main tube is determined according to the point cloud density gradient, and the axial fin tube point cloud is separated according to the demarcation threshold.

6. The fin tube weld extraction method according to claim 1, characterized in that: Fitting is performed based on the axial main body point cloud to obtain a complete main body geometric model, including the following steps: Solve the position of the center point c based on the axis direction of the main body; Determine the axis passing through the center point c and in the direction of the axis of the main body as the fitting axis of the complete main body; Get the radius r of the complete main pipe model; And build a complete main pipe model according to the fitting axis and radius r.

7. The fin tube weld extraction method according to claim 6, characterized in that: Solving the position of the center point c based on the axis direction of the main tube includes the following steps: The median value in the X direction and the median value in the Y direction of the point cloud data of the axial main body point cloud are used as the initial guess point c0; The distance variance from each point in the axial main body point cloud to the initial guess point c0 is defined as the objective function; The objective function is minimized using the Nelder-Mead simplex algorithm, and the output of the objective function when the algorithm converges is used as the position coordinate of the center point c.

8. The fin tube weld extraction method according to claim 7, characterized in that: The objective function is minimized using the Nelder-Mead simplex algorithm, which specifically includes the following steps: Calculate the objective function value at each vertex in the initial two-dimensional simplex, and sort the vertices according to the objective function value, wherein the initial two-dimensional simplex consists of an initial guess point c0, an x-direction offset point, and a y-direction offset point, a total of three vertices, wherein the initial guess point c0 and the x-direction offset point have the same y-direction coordinate and axial coordinate, and the difference in the x-direction coordinate is less than or equal to the x-direction offset threshold dx, the initial guess point c0 and the y-direction offset point have the same x-direction coordinate and axial coordinate, and the difference in the y-direction coordinate is less than or equal to the y-direction offset threshold dy; Calculate the centroids G of the other two vertices in the current initial two-dimensional simplex except the vertex H with the largest objective function value; reflect the vertex H with the largest objective function value through the centroid G to obtain the reflection point R, and R = G + α (GH); Get the objective function value of the reflection point R. If the objective function value of the reflection point R is the minimum value among all the current objective function values, then get the expansion point E, and E = G + γ (RG); if the objective function value of the reflection point R is the maximum value among all the current objective function values, then get the contraction point C, and C = G + ρ (HG). If the objective function values ​​of the reflection point R and the contraction point C are both greater than or equal to the current optimal objective function value, then according to v i =L+σ(v i -L) for overall contraction, where L is the vertex with the smallest objective function value in the initial two-dimensional simplex; Repeat the above steps to adjust the position and shape of the simplex until the Euclidean distance between any two vertices in the simplex meets the preset conditions, and the simplex algorithm is considered to converge.

9. The fin tube weld extraction method according to claim 7, characterized in that: The radius r of the complete main pipe is determined by the following formula: Where v is the vector of the fitting axis; p i is a point in the initial main body point cloud after X-direction filtering and Y-direction filtering.

10. A fin tube weld extraction system, characterized in that: include: Vision sensors; A mobile device connected to the visual sensor, used to drive the visual sensor to move to different positions and obtain three-dimensional point cloud data of different areas through the visual sensor; An original point cloud acquisition unit, which is used to preprocess the upper three-dimensional point cloud data, the outer three-dimensional point cloud data, and the lower three-dimensional point cloud data, and to splice all the preprocessed three-dimensional point cloud data to obtain an original point cloud model of the fin-tube heat exchange unit; A point cloud separation unit, connected to the original point cloud acquisition unit, for separating an initial main body point cloud from the original point cloud model; An axial determination unit, used to determine the axial direction of the main body according to the initial main body point cloud; A main body point cloud acquisition unit, which is used to perform axial filtering on the initial main body point cloud according to the axial direction of the main body to obtain an axial main body point cloud; A main body model acquisition unit, which is used to perform fitting according to the axial main body point cloud to obtain a complete main body geometric model; A virtual weld trajectory determination unit, which is used to determine the virtual weld trajectory according to the center point c of the complete main pipe geometric model; And a display unit, which is used to display a virtual weld trajectory and to form an optimized weld trajectory after optimizing weld characteristic parameters according to the virtual weld trajectory.

Images