A rotary table type automated three-dimensional scanning measurement system

Abstract

The present invention discloses a turntable type automatic 3D scanning and measuring system, which relates to the technical field of 3D scanning, and comprises a coordinate reference module, an instruction generation module, a conversion control module, a scanning execution module, a point cloud restoration module and a comparison and analysis module; the coordinate reference module is used to establish a reference coordinate system; the instruction generation module is used to generate control instructions or generate scanning instructions based on a partition parameter calibration method; the conversion control module generates control instructions by adopting a posture conversion method combined with reinforcement learning; the scanning execution module performs pre-scanning or scanning region by region based on the control instructions; the point cloud restoration module acquires a scanning point cloud set by conversion using a reverse restoration method, and generates a pre-scanned 3D model or a scanning 3D model by a triangulation method; the comparison and analysis module determines whether the size of the inspected workpiece is qualified based on Boolean operations, and if it is unqualified, a defect recognition network is used to generate a defect label, thereby realizing adaptive and efficient scanning of the inspected workpiece and accurate comparison result analysis.

Description

Technical Field

[0001] The invention relates to the technical field of three-dimensional scanning, and in particular to a rotary table type automated three-dimensional scanning measurement system. Background Art

[0002] Due to the lack of detection means in the production process, some workpieces can only be verified whether the specifications are within the tolerance range during assembly. Only then can the workpieces be trimmed to meet the design standards. In order to solve the above problems, the laser flight time measurement method is used in the prior art. Although this method can draw the measurement model, it has low efficiency and poor accuracy. Subsequently, the structured light measurement method was adopted. This method solves the problem of low efficiency, but the accuracy is still low and it is not suitable for measuring larger workpieces. The above measurement methods all require multiple disassembly and repair, resulting in increased material costs and waste of time, affecting work efficiency, thereby increasing production costs.

[0003] The existing Chinese patent application with authorization announcement number CN118424152B proposes a three-dimensional scanning method and a three-dimensional scanning device. The three-dimensional scanning method includes: projecting structured light onto an area to be measured and obtaining multiple scanning images, wherein the multiple scanning images respectively record structured light of different brightness projected onto the area to be measured; comparing the multiple scanning images and selecting a scanning image with better quality of the recorded structured light; constructing three-dimensional point cloud data of the surface of the area to be measured based on the selected scanning image; wherein the structured light includes at least two sub-structured lights of different frequencies, and the step of constructing three-dimensional point cloud data of the surface of the area to be measured based on the selected scanning image includes: evaluating the quality of each sub-structured light, and selecting a sub-structured light that meets preset requirements to construct the three-dimensional point cloud data.

[0004] However, in order to achieve better scanning effects, the three-dimensional laser scanners in the prior art are either set up based on manual experience or general scanning parameters, or selected through multiple acquisitions to select better scanning images. The first method cannot adaptively generate targeted scanning parameters for workpieces of different shapes, structures and sizes, resulting in poor scanning effects. The second method reduces the scanning efficiency. In addition, the prior art generally generates dimensional data differences through simple comparison during measurement, and lacks advanced technical means to automatically analyze the comparison results. Summary of the invention

[0005] In view of the deficiencies in the prior art, the present invention proposes a rotary table type automated three-dimensional scanning measurement system to achieve adaptive scanning parameter determination and scanning path planning for different inspected workpieces, and provide accurate comparison result analysis.

[0006] The technical solutions adopted to achieve the purpose of the present invention are:

[0007] A rotary table type automated three-dimensional scanning and measuring system, comprising a coordinate reference module, an instruction generation module, a conversion control module, a scanning execution module, a point cloud restoration module and a comparison and analysis module;

[0008] The coordinate reference module establishes the reference coordinate system and makes it public in the system;

[0009] The instruction generation module receives the activation instruction, generates the control instruction and sends it directly to the scanning execution module, receives the pre-scanned 3D model of the inspected workpiece, and constructs the scanning parameter set using the partition parameter calibration method. , set the task label , generate scan instructions And sent to the conversion control module;

[0010] The conversion control module obtains the scanning instruction , using the pose conversion method combined with reinforcement learning to generate the planning scanning pose matrix set and planning scan rotation angle set , with the task label Packaged together as control instructions And send it to the scanning execution module;

[0011] Scan execution module receives control instructions And identify the task tag , based on the task label, select pre-scan or scan area by area, and send the pre-scan point cloud set Or send a sequence of scanned point cloud subsets and planning scan rotation angle set To point cloud restoration module;

[0012] The point cloud restoration module is based on the reverse restoration method combined with the planned scanning rotation angle set Scan point cloud subset sequence Convert to scan point cloud set , using triangulation to process the pre-scanned point cloud set Or scan point cloud collection , feeding back the pre-scanned 3D model to the instruction generation module or sending the scanned 3D model to the comparison and analysis module;

[0013] The comparative analysis module compares the scanned 3D model and the standard 3D model based on Boolean operations to determine whether the size of the inspected workpiece is qualified. If it is unqualified, a defect label is generated based on the defect recognition network.

[0014] Further, the instruction generation module includes an activation control unit, a partition calibration unit and a scanning instruction unit;

[0015] The activation control unit receives the activation command and retrieves the pre-scan pose matrix , set the pre-scan rotation angle and task tags 0, package and generate control instructions And directly sent to the scanning execution module;

[0016] The partition calibration unit receives the pre-scanned 3D model, divides the pre-scanned 3D model into multiple regions using the partition parameter calibration method, obtains the scanning parameters of a single region, and constructs a scanning parameter set. and deliver the scanning instruction unit;

[0017] The scan command unit receives the scan parameter set , set the task label 1, package and generate scanning instructions And sent to the conversion control module.

[0018] Furthermore, the partition calibration unit uses the partition parameter calibration method to construct the scanning parameter set The specific steps include:

[0019] Use principal component analysis to obtain the minimum directed bounding box and the coordinates of the first vertex at the bottom of the minimum directed bounding box. , coordinates of the second vertex and the coordinates of the third vertex , determine the bottom plane equation;

[0020] Calculate the bottom of the bottom 2D coordinate system Axis direction vector and bottom Axis direction vector ;

[0021] Project the points in the pre-scanned 3D model onto the bottom plane based on the bottom plane equation, define the long side and wide side, and calculate the long side length and width length , according to the length of the field of view The length of the long side Divide into long side segments, according to the width of the field of view The width of the side Divide into wide edge segments, generating Regions, and are the total number of long side segments and the total number of wide side segments, is the total number of regions;

[0022] Taking the first vertex as the starting point, give the Mark the area number of each area, project the vertices of all triangular facets in the pre-scanned 3D model onto the bottom plane to obtain the corresponding projection point coordinates, and mark a single triangular facet based on the belonging area of ​​the three projection point coordinates of the single triangular facet;

[0023] Get the area number marked The normal vectors of all triangles and calculate the average normal vector , based on the average normal vector Length of mold The average normal vector Normalize and flip to get The initial scanning direction vector of the region ;

[0024] Based on The broadside segment number corresponding to the area and long side segment number Determine the coordinates of the center point of the region ;

[0025] Based on the best scanning distance , combined with the initial scanning direction vector and the coordinates of the center point of the region Determine Initial scan position of the area ;

[0026] Based on the minimum rotation principle on the positive side, the The scanning direction vector of the region , Scan location and scan rotation angle , with the bottom Axis direction vector and bottom Axis direction vector Generate scanning parameter sequence together ;

[0027] statistics The scanning parameter sequence of each region is constructed to construct a scanning parameter set. .

[0028] Furthermore, generating the scanning direction vector, scanning position and scanning rotation angle of a single area based on the minimum rotation principle on the positive side includes the following specific steps:

[0029] Judgement Initial scan position of the area Initial Axis coordinates Is it greater than or equal to 0?

[0030] If the initial Axis coordinates is greater than or equal to 0, then The scanning direction vector of the region and scan location The initial scanning direction vector and initial scan position , scan rotation angle ;

[0031] If the initial Axis coordinates Less than 0, the initial scan position Projected to the reference coordinate system Plane, get the initial scan projection position ;

[0032] Set the origin of the reference coordinate system to the initial scan projection position The vector and the base Axis direction vector The angle between the two is taken as the initial scanning rotation angle , compared with the initial scan rotation angle and The size of the rotation direction is selected to obtain the The scanning rotation angle of the area ;

[0033] Based on the scan rotation angle Determine The scanning direction vector after the region is rotated and scan location .

[0034] Further, the conversion control module includes an instruction conversion unit and an instruction planning unit;

[0035] The instruction conversion unit extracts the scan instruction The scan parameter set in , the scan parameter set Split into scan device parameter sets and scan rotation angle set , the scanning device parameter set is converted into Convert to scan pose matrix set , and the scan rotation angle set and task tags Pack together to generate initial control instructions and transmitted to the instruction planning unit;

[0036] The command planning unit receives the initial control command And extract the scan pose matrix set and scan rotation angle set , using reinforcement learning to reorder the scan pose matrix set and scan rotation angle set , generate the planning scan pose matrix set and planning scan rotation angle set , with the task label Packaged together as control instructions And sent to the scan execution module.

[0037] Furthermore, the instruction conversion unit adopts a posture conversion method to obtain a scanning posture matrix of a single area, including the following specific steps:

[0038] Get the Scanning device parameter sequence for a region , is the total number of regions;

[0039] Based on the bottom Axis direction vector ,bottom Axis direction vector and The scanning direction vector of the region Generate Areas of the region Axis direction vector ,area Axis direction vector and Region Axis direction vector ;

[0040] Get the The scan pose matrix of the region , the specific formula is as follows:

[0041] ;

[0042] in, and Scanning positions of Axis coordinates, Axis coordinates and Axis coordinates, , and The datum of the reference coordinate system Axis direction vector, datum Axis direction vector and datum Axis direction vector, Represents the angle between two direction vectors.

[0043] Furthermore, the instruction planning unit uses reinforcement learning to generate the planning scanning pose matrix set and planning scan rotation angle set , including the following specific steps:

[0044] Defining the time step , after the scanning execution module performs a scan of an area, the time step Add 1;

[0045] Defining the time step Action The scanning execution module is about to scan the area;

[0046] Defining the time step Status ,state The scanning direction vector of the scanning device and scan location And the scanning rotation angle of the lifting device ;

[0047] Defining the time step Rewards For scanning execution module in state Execute an action Instant rewards that can be obtained after The specific formula is as follows:

[0048] ;

[0049] in, The scanning device is in time step Execute an action After reaching the scanning position Fixed reward, , and They are direction weight coefficient, distance weight coefficient and rotation weight coefficient respectively. Indicates that the connected device performs an action The rotation angle in the direction of Indicates that the connected device performs an action The translation distance, Indicates that the lifting device performs an action The actual rotation angle, is the reward weight coefficient;

[0050] Defining the time step The value of ,value To scan the execution module The cumulative return after the action of time steps is as follows:

[0051] ;

[0052] in, is the learning rate, is the time step Rewards, is the discount factor, indicating the importance of future value, To complete the action After, action In Status The time steps available based on The maximum value of

[0053] The optimal region selection order of the scanning execution module is obtained through repeated iterations, and the scanning pose matrix set is rearranged according to the optimal region selection order. and scan rotation angle set , generate the planning scan pose matrix set and planning scan rotation angle set .

[0054] Furthermore, the scanning execution module includes a parsing control unit, a scanning device, a connecting device and a lifting device;

[0055] The analysis control unit receives the control command And identify the task tag , if the task tag , the scanning device is started to perform pre-scanning and obtain the pre-scan point cloud set And sent to the point cloud restoration module, if the task label , start the scanning device and the lifting device to scan each area and obtain the scanning point cloud subset sequence , together with the planned scan rotation angle set All of them are sent to the point cloud restoration module;

[0056] The scanning device is used to perform pre-scanning and area-by-area scanning;

[0057] The connecting device is used to connect the scanning device and the lifting device, and enables the scanning device to reach the scanning position and scan in the scanning direction through movement and adjustment;

[0058] The lifting device is used to place the workpiece to be inspected and can be located around the reference coordinate system. The axis rotates in the specified direction and at the specified angle.

[0059] Furthermore, the point cloud restoration module includes a pre-scanning restoration unit and an integrated reconstruction unit;

[0060] The pre-scan restoration unit receives the pre-scan point cloud set , based on the triangulation method, a pre-scanned 3D model of the inspected workpiece is restored and generated and fed back to the instruction generation module;

[0061] The integrated reconstruction unit receives a sequence of scanned point cloud subsets and planning scan rotation angle set , based on the reverse restoration method to generate the scanning point cloud set of the inspected workpiece , based on the triangulation method, the scanned 3D model of the inspected workpiece is restored and generated and sent to the comparative analysis module.

[0062] Furthermore, the integrated reconstruction unit generates a scanning point cloud set of the inspected workpiece based on the reverse restoration method. The specific steps include:

[0063] Extract The point cloud subset corresponding to the region and scan rotation angle ;

[0064] Get a subset of the point cloud The The rotation point cloud coordinates of the point cloud ;

[0065] Combined scanning rotation angle And rotate the point cloud coordinates Inverse point cloud coordinates ;

[0066] Subset the point cloud sequence All the rotated point cloud coordinates in the image are converted into corresponding point cloud coordinates and merged to generate a scanned point cloud set. .

[0067] Furthermore, restoring and generating a pre-scanned 3D model or a scanned 3D model based on the triangulation method includes the following specific steps:

[0068] Use statistical filters to remove pre-scan point cloud sets Or scan point cloud collection Generate a pre-scanned surface point cloud set based on the outlier points in Or scan surface point cloud collection ;

[0069] The normal vector is estimated based on the local surface fitting method, and the pre-scanned surface point cloud is collected. Or scan surface point cloud collection The single point cloud in and the 6 nearest point clouds construct a local plane, and the normal vector of the local plane is used as the normal vector of the single point cloud;

[0070] Point cloud collection for pre-scanned surface Or scan surface point cloud collection The normal vector direction is determined by one point cloud in the image, and the normal vectors of the remaining point clouds are traversed for redirection based on the principle of adjacent parallelism;

[0071] Projecting a single point cloud to the reference coordinate system based on its normal vector Plane, generate pre-scanned point cloud point set Or scan point cloud point set ;

[0072] For pre-scanned point cloud point set Or scan point cloud point set A single point cloud point in the image is used to perform a neighborhood search using a KD tree to obtain a neighborhood point cloud point set of the single point cloud point, and a triangular facet sequence of the single point cloud point is generated based on the empty circle property to construct a pre-scanned point cloud point set. Or scan point cloud point set The corresponding triangle patch set;

[0073] Pre-scanned surface point cloud set based on corresponding connection of triangular facet set Or scan surface point cloud collection All point clouds in the image can be restored to generate a pre-scanned 3D model or a scanned 3D model.

[0074] Furthermore, generating a triangle patch sequence of a single point cloud point based on the empty circle property includes the following specific steps:

[0075] Connect all the neighborhood point cloud points in the neighborhood point cloud point set with the single point cloud point to generate a center line sequence of the single point cloud point;

[0076] Determine the sorting index based on the center line sequence, and mark a single neighborhood point cloud point according to the sorting index;

[0077] Connect the neighboring point cloud points in sequence according to the marking order to generate the neighboring lines, delete the neighboring lines that intersect with the center line, and generate the initial triangle patch sequence of a single point cloud point;

[0078] Delete neighboring point cloud points to generate a triangle patch sequence of a single point cloud point.

[0079] Further, the comparison and analysis module includes a comparison detection unit and a defect analysis unit;

[0080] The comparison detection unit obtains the scanned 3D model, calls the standard 3D model of the inspected workpiece, and uses Boolean operations to determine whether the size of the inspected workpiece is qualified. If it is unqualified, the scanned 3D model and the standard 3D model are sent to the defect analysis unit;

[0081] The defect analysis unit generates defect labels based on the defect recognition network.

[0082] Furthermore, the defect analysis unit generates defect labels through the defect recognition network including:

[0083] Convert the scanned 3D model and the standard 3D model into a first image and a second image of a specified pixel size;

[0084] The first image and the second image are synchronously input into the downsampling layer, the downsampling layer uses a large convolution kernel and a small convolution kernel to mine in sequence, and generates the first feature and the second feature through maximum pooling processing;

[0085] The first feature and the second feature are input into the reconstruction coding layer, the reconstruction coding layer generates the first reconstruction feature and the second reconstruction feature through the self-attention mechanism, and the first feature and the second feature are superimposed through the residual structure to generate the first code and the second code;

[0086] The first code and the second code are input into the KAN layer. The KAN layer concatenates the first code and the second code into a combined code, processes each element in the combined code through different activation functions, and accumulates and generates an aggregated code.

[0087] The aggregated code is input into the upsampling layer, and the upsampling layer generates the expanded dimension features through two deconvolutions. The convolution kernels used in the two deconvolutions are exactly the same as those in the downsampling layer.

[0088] The expanded features are input into the logistic regression layer, and a probability vector is generated through multiple linear modulation and softmax function processing, and the defect label with the highest probability in the probability vector is output.

[0089] Compared with the prior art, the present invention has the following significant advantages:

[0090] 1. The instruction generation module analyzes the size, structure and other characteristics of the pre-scanned 3D model based on the partition parameter calibration method, and autonomously divides the area based on the field of view length and field of view width of the small field of view scanning device. It also adaptively obtains the scanning parameter sequence of a single area by analyzing the regional center point and average normal vector of a single area, and statistically constructs a scanning parameter set, thereby realizing the generation of adaptive scanning parameters that combine the size and structural characteristics of different inspected workpieces, effectively improving the scanning accuracy;

[0091] 2. The conversion control module uses the posture conversion method to convert the scanning device parameter set into a scanning posture matrix set that can be recognized by the scanning execution module, and further considers the scanning order of the area through reinforcement learning to generate a planned scanning posture matrix set and a planned scanning rotation angle set, which effectively reduces the movement and rotation of the scanning execution module during area-by-area scanning and improves the efficiency of area-by-area scanning;

[0092] 3. The comparative analysis module uses Boolean operations instead of traditional numerical comparisons to accurately compare the scanned 3D model with the standard 3D model to determine whether the size of the inspected workpiece is qualified. If the inspected workpiece is unqualified, a defect label indicating the type of unqualified workpiece is generated through the defect recognition network, replacing traditional manual judgment behavior. BRIEF DESCRIPTION OF THE DRAWINGS

[0093] Figure 1 A schematic diagram of a rotary table type automated three-dimensional scanning and measuring system of the present invention;

[0094] Figure 2 A flow chart for generating a scanning direction vector, scanning position and scanning rotation angle of a single area in the present invention;

[0095] Figure 3 A schematic diagram of a triangular patch sequence for generating a single point cloud point based on the empty circle property in the present invention;

[0096] Figure 4 This is a diagram of the defect identification network model in the present invention. DETAILED DESCRIPTION

[0097] The present invention is further described in detail below in conjunction with the accompanying drawings and embodiments.

[0098] Example 1

[0099] like Figure 1 As shown, a specific embodiment of the present invention discloses a rotary table type automated three-dimensional scanning and measuring system, including a coordinate reference module, an instruction generation module, a conversion control module, a scanning execution module, a point cloud restoration module and a comparison analysis module;

[0100] The coordinate reference module takes the plane where the lifting device is located as plane, establish a reference coordinate system with the center of the lifting device as the origin and make it public in the system, where the reference coordinate system The axis direction vector is , benchmark The axis direction vector is , benchmark Axis direction vector ;

[0101] The instruction generation module receives the activation instruction, generates the control instruction and sends it directly to the scanning execution module, receives the pre-scanned 3D model of the inspected workpiece, divides the pre-scanned 3D model into several areas using the partition parameter calibration method, and determines the scanning parameters of each area to construct a scanning parameter set. , set the task label To generate scan instructions , send a scan command to the conversion control module;

[0102] The conversion control module extracts the scanning instructions The scan parameter set in And split into scanning device parameter sets and scan rotation angle set , the scanning device parameter set is converted into Convert to scan pose matrix set , and use reinforcement learning to rearrange the scan pose matrix set and scan rotation angle set The order of scanning pose matrices and scanning rotation angles of the same area in the , generates a set of planning scanning pose matrices and planning scan rotation angle set , with the task label Packaged together as control instructions And send it to the scanning execution module;

[0103] Scan execution module receives control instructions And identify the task tag , if the task tag , start the large field of view scanning device for pre-scanning and obtain the pre-scan point cloud set And sent to the point cloud restoration module, if the task label , start the small field of view scanning device and the lifting device to scan each area and obtain the scanning point cloud subset sequence , together with the planned scan rotation angle set Send it to the point cloud restoration module together;

[0104] The point cloud restoration module receives the pre-scanned point cloud set Or scan a sequence of point cloud subsets and planning scan rotation angle set , based on the reverse restoration method, the scanned point cloud subset sequence Convert to scan point cloud set , using triangulation to process the pre-scanned point cloud set Or scan point cloud collection , generate a pre-scanned 3D model and feed it back to the instruction generation module or generate a scanned 3D model and send it to the comparison and analysis module;

[0105] The comparative analysis module uses Boolean operations to compare the scanned 3D model and the standard 3D model, and determines whether the size of the inspected workpiece is qualified based on the Boolean operation results. If it is judged to be unqualified, a defect label is generated based on the defect recognition network. The defect label is used to indicate the type of unqualified workpiece being inspected, which includes but is not limited to size discrepancy, surface depression and structure missing.

[0106] Further, the instruction generation module includes an activation control unit, a partition calibration unit and a scanning instruction unit;

[0107] The activation control unit receives the activation command and retrieves the pre-scan pose matrix , rotate the pre-scan by and task tags All are set to 0, and control instructions are generated by packaging And directly sent to the scanning execution module, the activation instruction is automatically generated when the system receives the scanning task of the inspected workpiece;

[0108] The partition calibration unit receives the pre-scanned 3D model of the inspected workpiece, divides the pre-scanned 3D model into several regions using the partition parameter calibration method, and obtains the scanning parameters of a single region based on the center point coordinates and average normal vector of a single region to construct a scanning parameter set. and deliver the scanning instruction unit;

[0109] The scan command unit receives the scan parameter set , change the task label Set to 1, indicating that the task type is area-by-area scanning, and the scanning parameter set is packaged and task tags Generate scan instructions And sent to the conversion control module.

[0110] Furthermore, the partition calibration unit uses the partition parameter calibration method to construct the scanning parameter set The specific steps include:

[0111] The principal component analysis method is used to find the main direction of the pre-scanned 3D model, and the minimum directed bounding box aligned with the main direction is obtained. The minimum directed bounding box is a cuboid that can tightly surround the pre-scanned 3D model.

[0112] Get the vertex coordinates of the first vertex, the second vertex, and the third vertex at the bottom of the minimum directional bounding box in the reference coordinate system, and record them as the first vertex coordinates , coordinates of the second vertex and the coordinates of the third vertex ,in, and The coordinates of the first vertex are In the base coordinate system Axis value, Axis value and Axis value, and The coordinates of the second vertex are In the base coordinate system Axis value, Axis value and Axis value, and The coordinates of the third vertex are In the base coordinate system Axis value, Axis value and Axis value, based on the first vertex coordinates , coordinates of the second vertex and the coordinates of the third vertex Determine the bottom plane equation , the specific calculation formula is as follows:

[0113] ;

[0114] ;

[0115] ;

[0116] ;

[0117] in, , and are the coefficients of the bottom plane equation, is the constant of the bottom plane equation;

[0118] Calculate the bottom of the bottom 2D coordinate system Axis direction vector and bottom Axis direction vector , the specific formula is as follows:

[0119] ;

[0120] ;

[0121] The bottom two-dimensional coordinate system reflects the bottom of the minimum oriented bounding box with respect to the reference coordinate system Plane transformation relations;

[0122] Based on the bottom plane equation Project the points in the pre-scanned 3D model to the bottom plane. Since the projection of the pre-scanned 3D model on the bottom plane coincides with the bottom of the minimum directed bounding box, define the line segment formed by the first vertex and the second vertex as the long side, and the line segment formed by the first vertex and the third vertex as the wide side. Calculate the length of the long side. and width length , get the field of view length of the small field of view scanning device and field of view width , according to the length of the field of view The length of the long side Divide into equal intervals long side segments, according to the width of the field of view The width of the side Divide into equal intervals Wide side segments, and are the total number of long side segments and the total number of wide side segments, respectively. The specific calculation formula is as follows:

[0123] ;

[0124] ;

[0125] in, Indicates a round-up operation, where the bottom is divided into Regions;

[0126] Starting from the first vertex, the regions are labeled and numbered in order according to the principle of length first and width later. Area code for the area ,in, and Respectively The wide side segment number and long side segment number corresponding to each area, is the total number of regions;

[0127] Project the vertices of all triangular facets in the pre-scanned 3D model onto the bottom plane, using vertex coordinates For example, and The vertex coordinates are In the base coordinate system Axis value, Axis value and Axis value, after projection, becomes the bottom projection point coordinate on the bottom plane , , and The coordinates of the bottom projection point are In the base coordinate system Axis value, Axis value and Axis value, the specific formula of the projection process is as follows:

[0128] ;

[0129] ;

[0130] ;

[0131] ;

[0132] in, is the offset of the vertex;

[0133] Traverse a single triangular face in the pre-scanned 3D model, count the coordinates of the projection points corresponding to the three vertices of the single triangular face, determine the respective regions to which the three projection point coordinates belong on the bottom plane, and mark the single triangular face with the region number corresponding to the region to which it belongs. Therefore, a single triangular face is marked with at least one region number and at most three different region numbers.

[0134] For Regions, get the region number The normal vectors of all triangles of and calculate the mean to obtain the The average normal vector of the region , calculate the average normal vector Length of mold , will The average normal vector of the region Divide by the corresponding modulus Normalize and flip the direction to get The initial scanning direction vector of the region ;

[0135] For Regions, determine the coordinates of the region center , the specific calculation formula is as follows:

[0136] ;

[0137] Get the best scanning distance for small field of view scanning devices , optimal scanning distance It is usually determined by the model of the small field of view scanning device, which can be obtained through preliminary measurement and combined with the first The initial scanning direction vector of the region and the coordinates of the center point of the region Determine Initial scan position of the area ;

[0138] Based on the minimum rotation principle on the positive side, further analysis and adjustment The initial scanning direction vector of the region and initial scan position , generating The scanning direction vector of the region , Scan location and scan rotation angle , with the bottom Axis direction vector and bottom Axis direction vector Generate scanning parameter sequence together ,in, For the A sequence of scanning device parameters for each area;

[0139] statistics The scanning parameter sequence of each region is constructed to construct a scanning parameter set. ,in, For the The scanning parameter sequence of the region, is the total number of regions.

[0140] like Figure 2 As shown, further, generating the scanning direction vector, scanning position and scanning rotation angle of a single area based on the minimum rotation principle on the positive side includes the following specific steps:

[0141] Judgement Initial scan position of the area Initial Axis coordinates Is it greater than or equal to 0, where and Respectively Initial scan position of the area In the base coordinate system Axis value, Axis value and Axis value;

[0142] If the initial Axis coordinates is greater than or equal to 0, then Initial scan position of the area lie in On the positive side of the shaft, the lifting device does not need to rotate. The scanning direction vector of the region and scan location That is equivalent to the initial scanning direction vector and initial scan position , and the scanning rotation angle ;

[0143] If the initial Axis coordinates is less than 0, then Initial scan position of the area lie in The negative side of the axis needs to be around the reference coordinate system. The shaft rotates the lifting device so that the Scan location of the area of Axis coordinates ;

[0144] Since the rotation of the lifting device is horizontal, it only affects the initial scanning position. of Axis coordinates and axis coordinates, the initial scanning position Projected to the reference coordinate system Plane, get the initial scan projection position ;

[0145] Calculate the projection position from the origin of the reference coordinate system to the initial scan The vector and the benchmark The axis direction vector is The initial scan rotation angle , i.e. the initial scanning position Rotate the initial scan in counterclockwise direction After that, the scanning position of Axis coordinates ;

[0146] Since the lifting device has two rotation directions, clockwise and counterclockwise, except for the initial scanning rotation angle rotated in the counterclockwise direction , the horizontal complementary angle of the initial scanning rotation angle rotated clockwise It can also make the scanning position of Axis coordinates , it is necessary to select the rotation direction so that the actual rotation angle of the lifting device is the smallest. The scanning rotation angle of the area The specific formula is as follows:

[0147] ;

[0148] The scanning rotation angle When it is less than 0, it indicates the absolute value of the scanning rotation angle. Rotate clockwise and scan the rotation angle When it is greater than or equal to 0, it indicates the scanning rotation angle Rotate counterclockwise;

[0149] Determine The scanning direction vector after the region is rotated and scan location , the specific formula is as follows:

[0150] ;

[0151] ;

[0152] in, Represents the transpose of a vector.

[0153] Further, the conversion control module includes an instruction conversion unit and an instruction planning unit;

[0154] The instruction conversion unit extracts the scan instruction The scan parameter set in , the scan parameter set Split into scan device parameter sets and scan rotation angle set ,in, and Respectively The scanning device parameter sequence and scanning rotation angle of each area are converted into scanning device parameter set by using posture conversion method. The scanning device parameter sequences of all regions in the scan are converted into corresponding scanning pose matrices. The scanning pose matrix is ​​a data type that can be recognized by the scanning execution module. The scanning pose matrix set is constructed. , and the scan rotation angle set and scan instructions Tasks tab in Pack together to generate initial control instructions , transmit initial control instructions to the instruction planning unit;

[0155] The command planning unit receives the initial control command And extract the scan pose matrix set and scan rotation angle set , using reinforcement learning to rearrange the scan pose matrix set and scan rotation angle set The scanning pose matrix and scanning rotation angle sequence in the program are used to generate the planning scanning pose matrix set and planning scan rotation angle set , with the task label Packaged together as control instructions And sent to the scan execution module, .

[0156] Furthermore, the instruction conversion unit adopts a posture conversion method to obtain a scanning posture matrix of a single area, including the following specific steps:

[0157] Get the Scanning device parameter sequence for a region is the total number of regions;

[0158] Based on the bottom Axis direction vector ,bottom Axis direction vector and The scanning direction vector of the region Generate Areas of the region Axis direction vector ,area Axis direction vector and Region Axis direction vector , the specific calculation formula is as follows:

[0159] ;

[0160] ;

[0161] ;

[0162] No. Areas of the region Axis direction vector ,area Axis direction vector and Region Axis direction vector Together they constitute the The regional coordinate system of the region shows the transformation relationship of the region with respect to the bottom of the minimum oriented bounding box;

[0163] Get the The scan pose matrix of the region , the specific formula is as follows:

[0164] ;

[0165] in, and Scanning positions of Axis coordinates, Axis coordinates and Axis coordinates, , and The datum of the reference coordinate system Axis direction vector, datum Axis direction vector and datum Axis direction vector, Represents the angle between two direction vectors.

[0166] Furthermore, the instruction planning unit uses reinforcement learning to generate the planning scanning pose matrix set and planning scan rotation angle set , including the following specific steps:

[0167] Defining the time step , after the scanning execution module performs a scan of an area, the time step Add 1;

[0168] Defining the time step Action ,action is the time step The area that the scanning execution module is about to scan is specifically the area number corresponding to the area;

[0169] Defining the time step Status ,state Including time steps The scanning direction vector of the scanning device in the scanning execution module and scan location And the scanning rotation angle of the lifting device , the specific value is determined by the time step Action Decision, i.e. state Status Complete the action Status;

[0170] Defining the time step Rewards ,award For scanning execution module in state Execute an action Instant rewards that can be obtained after The specific formula is as follows:

[0171] ;

[0172] in, The scanning device is in time step Execute an action After arrival time step Scanning location Fixed reward, , and They are direction weight coefficient, distance weight coefficient and rotation weight coefficient respectively. Indicates the connection device from the scanning direction vector To scan direction vector The rotation angle in the direction of Indicates the connected device from the scanning position To scanning position The translation distance, Indicates the rotation angle of the lifting device from the scanning To scan rotation angle The actual rotation angle, is a reward weight coefficient between 0 and 1, at time step Execute an action Make the state Status The larger the adjustment of the scanning pose matrix and the scanning rotation angle, the faster the action The lower the instant reward, in this embodiment, , ;

[0173] Defining the time step The value of ,value To scan the execution module The cumulative return after the action of time steps is as follows:

[0174] ;

[0175] in, is the learning rate, is the time step Rewards, is the discount factor, indicating the importance of future value, To complete the action After, action In Status The time steps available based on This is because the scanned area does not need to be scanned again. After the action of each time step is executed, the area available for the action of the next time step will be reduced. Therefore, the value Consider the cumulative return in combination with rewards, rather than focusing on the reward of a single time step;

[0176] By repeatedly selecting different actions at each time step to update the value , For the total number of regions, get the value The optimal region selection order of the execution module is maximized, and the scanning pose matrix set is rearranged according to the optimal region selection order. and scan rotation angle set The order of scanning pose matrix and scanning rotation angle in the program is used to generate the planning scanning pose matrix set and planning scan rotation angle set .

[0177] Furthermore, the scanning execution module includes a parsing control unit, a scanning device, a connecting device and a lifting device;

[0178] The analysis control unit receives the control command And identify the task tag , if the task tag , then start the large field of view scanning device separately, according to the pre-scanning posture matrix Control the movement of the connecting device to enable the large field of view scanning device to perform pre-scanning and obtain the pre-scan point cloud set And sent to the point cloud restoration module, if the task label , the small field of view scanning device and the lifting device are started, and the planned scanning pose matrix set and planning scan rotation angle set The planned scanning pose matrix and planned scanning rotation angle corresponding to the area are extracted in sequence, the movement of the connecting device is controlled, and the lifting device is controlled to rotate clockwise or counterclockwise, so that the small field of view scanning device scans the area, and the scanning point cloud subset of each area is obtained to construct a scanning point cloud subset sequence , send scan point cloud subset sequence and planning scan rotation angle set To the point cloud restoration module, where For the A subset of the scanned point cloud of a region, , is the total number of regions;

[0179] The scanning device includes a large field of view scanning device and a small field of view scanning device. The large field of view scanning device is used to perform a single pre-scan of the inspected workpiece in the lifting device, and the small field of view scanning device is used to perform region-by-region scanning of the inspected workpiece.

[0180] The connection device can be realized by an intelligent mechanical arm, which is used to connect the scanning device and the lifting device, and the scanning device can reach the scanning position and scan in the scanning direction by moving the mechanical arm and adjusting the angle of the mechanical arm;

[0181] The lifting device is a regular-shaped turntable used to place the inspected workpiece and can rotate around the reference coordinate system. The axis rotates in a specified direction and at a specified angle. Since the scanning device and the connecting device are installed on the right side of the lifting device and can only move on the right side, the moving trajectory is always located in the reference coordinate system. On the positive side of the axis, the inspected workpiece is located at The area on the negative side of the axis may not be scanned completely or cannot be scanned. By rotating the lifting device, the area on the negative side of the workpiece can be scanned. The area on the negative side of the axis is shifted to Positive side of the axis.

[0182] Furthermore, the point cloud restoration module includes a pre-scanning restoration unit and an integrated reconstruction unit;

[0183] The pre-scan restoration unit receives the pre-scan point cloud set , remove the pre-scan point cloud set based on triangulation The outliers in the image are detected and restored to generate a pre-scanned 3D model of the inspected workpiece, and the pre-scanned 3D model is fed back to the instruction generation module;

[0184] The integrated reconstruction unit receives a sequence of scanned point cloud subsets and planning scan rotation angle set , based on the reverse restoration method, the scanning point cloud set of the inspected workpiece is generated , remove the scan point cloud set based on triangulation The outliers in the image are detected and restored to generate a scanned 3D model of the inspected workpiece, and the scanned 3D model is sent to the comparison and analysis module.

[0185] Furthermore, the integrated reconstruction unit generates a scanning point cloud set of the inspected workpiece based on the reverse restoration method. The specific steps include:

[0186] From a sequence of scanned point cloud subsets and planning scan rotation angle set Extract the Point cloud subset corresponding to the region and scan rotation angle ;

[0187] Since the point cloud subset The middle is the first time after the lifting device rotates horizontally Surface point cloud of a region, get point cloud subset The The rotation point cloud coordinates of a point cloud in the base coordinate system ,in, , and The rotated point cloud coordinates are In the base coordinate system Axis value, Axis value and Axis value;

[0188] Combined scanning rotation angle And rotate the point cloud coordinates Inversely deduce the point cloud coordinates in the reference coordinate system before horizontal rotation ,in, , and Point cloud coordinates In the base coordinate system Axis value, Axis value and Axis value, the specific formula is as follows:

[0189] ;

[0190] ;

[0191] ;

[0192] Subset the point cloud sequence All the rotated point cloud coordinates of each point cloud subset are converted into corresponding point cloud coordinates and merged to generate the scanned point cloud set of the inspected workpiece. .

[0193] Furthermore, restoring and generating a pre-scanned 3D model or a scanned 3D model based on the triangulation method includes the following specific steps:

[0194] Get a pre-scan point cloud set Or scan point cloud collection , using statistical filters to count the pre-scanned point cloud set Or scan point cloud collection The neighborhood information of each point cloud in the image is used to identify and remove outliers, and obtain the pre-scanned surface point cloud set Or scan surface point cloud collection ;

[0195] Normal vector estimation is performed based on the local surface fitting method. Or scan surface point cloud collection For a single point cloud, search for the six point clouds closest to it to construct a local plane of the single point cloud, and use the normal vector corresponding to the local plane as the normal vector of the single point cloud;

[0196] Since the normal vector of a single point cloud obtained by normal vector estimation is ambiguous, that is, it only determines the line where the normal vector is located, but does not determine the direction of the normal vector, firstly, the surface point cloud set is pre-scanned. Or scan surface point cloud collection The normal vector direction of one point cloud is determined, and the normal vectors of the remaining point clouds are traversed for redirection based on the adjacent parallel principle. The adjacent parallel principle indicates that the normal vectors of two adjacent point clouds should have the same direction and be close to parallel. By calculating the inner product of the normal vectors of two adjacent point clouds, if the inner product is negative, the normal vector of the latter point cloud is flipped;

[0197] Projecting a single point cloud to the reference coordinate system based on its normal vector Plane, collect the pre-scanned surface point cloud Convert to pre-scanned point cloud point set Or scan the surface point cloud Convert to scanned point cloud point set ;

[0198] For pre-scanned point cloud point set Or scan point cloud point set A single point cloud point in the image is used to perform neighborhood search using a KD tree to obtain a neighborhood point cloud point set of a single point cloud point. A triangular facet sequence of a single point cloud point is generated based on the empty circle property to obtain a pre-scanned point cloud point set. Or scan point cloud point set The triangular facet sequence of all single point cloud points in the pre-scanned point cloud point set is constructed Or scan point cloud point set The corresponding triangular patch set indicates the pre-scanned point cloud point set Or scan point cloud point set The connection relationship between all point cloud points in ;

[0199] Based on the connection relationship of the triangular face set, the corresponding connection pre-scanned surface point cloud set Or scan surface point cloud collection All point clouds in the image are restored to generate a pre-scanned 3D model or a scanned 3D model of the inspected workpiece.

[0200] like Figure 3 As shown, further, generating a triangular face sequence of a single point cloud point based on the empty circle property includes the following specific steps:

[0201] Connect all the neighboring point cloud points in the neighborhood point cloud point set of a single point cloud point with the single point cloud point to generate a center line sequence of the single point cloud point;

[0202] Connect all center lines in the center line sequence to the base coordinate system Axis direction vector The angle between the center and the center is used as the sorting index, and the single neighborhood point cloud points corresponding to the single center line are marked in ascending order according to the sorting index. Figure 3 For example, the single neighborhood point cloud points are labeled 1-7 in sequence;

[0203] Connect the neighboring point cloud points in sequence to generate neighboring lines, and delete the neighboring lines that intersect with the center line to generate an initial triangle patch sequence for a single point cloud point;

[0204] Execute neighborhood point cloud point deletion. When the neighborhood point cloud point is located within the circumscribed circle of any triangle face in the initial triangle face sequence, remove the neighborhood point cloud point and the related center line and neighborhood line to generate a triangle face sequence of a single point cloud point.

[0205] Further, the comparison and analysis module includes a comparison detection unit and a defect analysis unit;

[0206] The comparison detection unit obtains the scanned 3D model, calls the standard 3D model of the inspected workpiece, imports it into SolidWorks, and uses Boolean operation to automatically determine. If the Boolean operation result is 1, the size of the inspected workpiece is determined to be qualified. If the Boolean operation result is 0, the size of the inspected workpiece is determined to be unqualified. The scanned 3D model and the standard 3D model are sent to the defect analysis unit. Boolean operations include union operations, intersection operations, and difference operations, which can be used to compare the overlap between the scanned 3D model and the standard 3D model. SolidWorks is a commonly used 3D model analysis software;

[0207] The defect analysis unit generates defect labels based on the defect recognition network. The defect labels are used to indicate the unqualified types of the inspected workpieces. The unqualified types include but are not limited to size discrepancies, surface depressions, and structural deficiencies.

[0208] like Figure 4 As shown, further, the defect analysis unit generates defect labels through the defect recognition network including:

[0209] Convert the scanned 3D model and the standard 3D model into a first image and a second image of a specified pixel size;

[0210] The first image and the second image are synchronously input into the downsampling layer. The downsampling layer adopts The large convolution kernel and The first image and the second image are mined twice using a small convolution kernel, and the first feature and the second feature are generated by maximum pooling.

[0211] The first feature and the second feature are input into the reconstruction coding layer, and the reconstruction coding layer reconstructs the first feature and the second feature through the self-attention mechanism to generate the first reconstructed feature and the second reconstructed feature, and superimposes the first feature and the second feature through the residual structure to generate the first code and the second code;

[0212] The first code and the second code are input into the KAN layer. The KAN layer uses an activation function to replace the learnable linear weights in the linear fully connected layer, concatenates the first code and the second code into a combined code, processes each element in the combined code through different activation functions, and accumulates the processing results to generate an aggregate code. Different activation functions are Figure 4 The KAN layer includes activation function 1, activation function 2, etc. The number of activation functions is equal to the elements in the combined code;

[0213] The aggregated code is input into the upsampling layer, and the upsampling layer performs upsampling decoding through two deconvolutions to generate expanded-dimensional features. The convolution kernels used in the two deconvolutions are exactly the same as those used in the downsampling layer.

[0214] The expanded features are input into the logistic regression layer, the expanded features are linearly modulated twice, and the softmax function is used to generate a probability vector, and the defect label with the highest probability in the probability vector is output.

[0215] The present invention discloses a turntable type automatic 3D scanning measurement system, comprising a coordinate reference module, an instruction generation module, a conversion control module, a scanning execution module, a point cloud restoration module and a comparative analysis module; the coordinate reference module is used to establish a reference coordinate system and disclose it in the system; the instruction generation module generates a control instruction for pre-scanning based on an activation instruction, analyzes the characteristics of the pre-scanned 3D model based on a partition parameter calibration method, adaptively divides the area and constructs a scanning parameter set, and generates a scanning instruction with a collection of task labels; the conversion control module uses a posture conversion method combined with reinforcement learning to process the scanning instruction, obtains a planned scanning posture matrix set and a planned scanning rotation angle set with an optimal area selection order, and generates a control instruction in combination with the task label. ; The scanning execution module performs pre-scanning or area-by-area scanning based on control instructions, and sends a pre-scan point cloud set or a scanning point cloud subset sequence and a planned scanning rotation angle set; the point cloud restoration module uses the inverse restoration method combined with the planned scanning rotation angle set to transform the scanning point cloud subset sequence into a scanning point cloud set, and uses the triangulation method to process the pre-scan point cloud set or the scanning point cloud set to generate a pre-scanned three-dimensional model or a scanning three-dimensional model; the comparative analysis module compares the scanning three-dimensional model and the standard three-dimensional model based on Boolean operations to determine whether the size of the inspected workpiece is qualified. If it is unqualified, the defect recognition network is used to generate a defect label, thereby realizing adaptive scanning parameter determination of different inspected workpieces, efficient scanning path planning and accurate comparison result analysis.

[0216] The above is only a preferred embodiment of the present invention, and the protection scope of the present invention is not limited to the above embodiments. All technical solutions under the concept of the present invention belong to the protection scope of the present invention. It should be pointed out that for ordinary technicians in this technical field, some improvements and modifications without departing from the principle of the present invention should also be regarded as the protection scope of the present invention.

Claims

1. A rotary table type automated three-dimensional scanning measurement system, characterized in that: It includes a coordinate reference module, an instruction generation module, a conversion control module, a scanning execution module, a point cloud restoration module and a comparative analysis module; The coordinate reference module is used to establish a reference coordinate system; The instruction generation module generates a control instruction when receiving an activation instruction, and constructs a scanning parameter set by using a partition parameter calibration method when receiving a pre-scanned three-dimensional model, and generates a scanning instruction in combination with a task label; The conversion control module obtains scanning instructions, uses the posture conversion method combined with reinforcement learning to generate a planned scanning posture matrix set and a planned scanning rotation angle set, and generates control instructions in combination with the task label; The scanning execution module receives the control instruction, selects to perform a pre-scan or a region-by-region scan based on the task tag, and sends a pre-scan point cloud set or a scan point cloud subset sequence and a planned scan rotation angle set accordingly; The point cloud restoration module converts the scan point cloud subset sequence into a scan point cloud set based on the inverse restoration method combined with the planned scan rotation angle set, and uses the triangulation method to generate a pre-scanned 3D model or a scanned 3D model; The comparative analysis module receives the scanned three-dimensional model and determines whether the size of the inspected workpiece is qualified based on Boolean operations. If it is unqualified, a defect recognition network is used to generate a defect label; The partition parameter calibration method is based on the optimal scanning distance , initial scanning direction vector and the coordinates of the center point of the region Determine Initial scan position of the area ; Determine the initial scanning position based on the minimum rotation principle on the positive side Initial Axis coordinates Is it greater than or equal to 0? If it is greater than or equal to 0, The scanning direction vector of the region and scan location is the initial scanning direction vector and initial scan position , scan rotation angle ; If less than 0, the initial scanning position of the projection To the base coordinate system plane to obtain the initial scan projection position ; Compare the origin of the reference coordinate system to the initial scan projection position The vector and the base Axis direction vector The angle and The size of the lifting device is used to select the rotation direction and determine the The scanning rotation angle of the lifting device in each area , according to The scanning rotation angle of the lifting device in each area Determine the first The scanning direction vector of the region and scan location ; The reinforcement learning is used to generate a planning scanning pose matrix set and a planning scanning rotation angle set, including the following specific steps: Define the time step and update the time step after performing a scan of a single area; The action of defining a single time step is to select the area to be scanned; defining a state of a single time step, wherein the state includes a scanning direction vector, a scanning position, and a scanning rotation angle of the single time step; The reward of a single time step is defined as the immediate reward that can be obtained after executing a single time step action in a single time step state; The value of a single time step is defined as the cumulative reward after executing the actions of a single time step and all previous time steps, where the time step The value of , is the learning rate, is the time step Rewards, is the discount factor, indicating the importance of future value, To complete the action After, action In Status The time steps available based on The maximum value of The optimal region selection order is obtained through repeated iterations, and the scanning pose matrix set and the scanning rotation angle set are rearranged according to the optimal region selection order to generate the planned scanning pose matrix set and the planned scanning rotation angle set.

2. A rotary table type automated three-dimensional scanning and measuring system as claimed in claim 1, characterized in that: The instruction generation module includes an activation control unit, a partition calibration unit and a scanning instruction unit; The activation control unit receives the activation instruction, retrieves the pre-scanning posture matrix, sets the pre-scanning rotation angle and task label to 0, packages and generates control instructions, and sends them to the scanning execution module; The partition calibration unit receives the pre-scanned three-dimensional model, divides the pre-scanned three-dimensional model into multiple regions using a partition parameter calibration method and obtains scanning parameters of a single region, constructs a scanning parameter set and delivers it to the scanning instruction unit; The scanning instruction unit receives a scanning parameter set, sets a task tag to 1, packages and generates a scanning instruction, and sends the scanning instruction to the conversion control module.

3. The rotary table type automated three-dimensional scanning and measuring system according to claim 2, characterized in that: The method of using the partition parameter calibration method to construct the scanning parameter set includes the following specific steps: A principal component analysis method is used to obtain a minimum directed bounding box, and a bottom plane equation is determined based on the first vertex coordinate, the second vertex coordinate, and the third vertex coordinate of the minimum directed bounding box; Calculate the bottom of the bottom 2D coordinate system Axis direction vector and bottom Axis direction vector; Projecting points in the pre-scanned three-dimensional model onto the bottom plane, defining a long side and a wide side and calculating the length of the long side and the length of the wide side, and dividing the long side and the wide side according to the field of view length and the field of view width to generate a plurality of regions; Mark each region with a region number in turn, and mark a single triangle according to the region to which the projection point coordinates of the single triangle belong; Obtaining normal vectors of all triangular facets marked with a single region number and calculating an average normal vector, normalizing and flipping the average normal vector, and obtaining an initial scanning direction vector of the single region; Determine the coordinates of the center point of the region based on the wide side segment number and the long side segment number corresponding to the single region, and determine the initial scanning position of the single region in combination with the optimal scanning distance and the initial scanning direction vector; Based on the minimum rotation principle on the positive side, the scanning direction vector, scanning position and scanning rotation angle of a single area are generated, and the bottom The axis direction vector and the bottom The axis direction vectors are packed to generate a scanning parameter sequence for a single area; The scanning parameter sequences of all regions are counted to construct a scanning parameter set.

4. The rotary table type automated three-dimensional scanning and measuring system according to claim 2, characterized in that: The conversion control module includes an instruction conversion unit and an instruction planning unit; The instruction conversion unit extracts the scanning parameter set in the scanning instruction, splits the scanning parameter set into a scanning device parameter set and a scanning rotation angle set, converts the scanning device parameter set into a scanning posture matrix set using a posture conversion method, and packages it with the scanning rotation angle set and the task label to generate an initial control instruction and transmit it to the instruction planning unit; The instruction planning unit extracts the scanning pose matrix set and the scanning rotation angle set from the initial control instructions, and uses reinforcement learning to reorder them to generate the planned scanning pose matrix set and the planned scanning rotation angle set, which are packaged together with the task label as control instructions and sent to the scanning execution module.

5. The rotary table type automated three-dimensional scanning and measuring system according to claim 1, characterized in that: The method of generating a pre-scanned 3D model or a scanned 3D model by using a triangulation method comprises: A statistical filter is used to remove outliers in a pre-scan point cloud set or a scan point cloud set to generate a pre-scan surface point cloud set or a scan surface point cloud set; Normal vector estimation is performed based on a local surface fitting method, a local plane is constructed by combining a single point cloud in a pre-scanned surface point cloud set or a scanned surface point cloud set with multiple nearest point clouds, and the normal vector of the local plane is used as the normal vector of the single point cloud; Determine the normal vector orientation for a point cloud in the pre-scanned surface point cloud set or the scanned surface point cloud set, and redirect the normal vectors of the remaining point clouds based on the principle of neighbor parallelism; Projecting a single point cloud to the reference coordinate system based on its normal vector Plane, generate pre-scanned point cloud point set or scanned point cloud point set; For a single point cloud point in the pre-scanned point cloud point set or the scanned point cloud point set, a KD tree is used to perform a neighborhood search to obtain the neighborhood point cloud point set of the single point cloud point, and a triangular facet sequence of the single point cloud point is generated based on the empty circle property to construct a triangular facet set corresponding to the pre-scanned point cloud point set or the scanned point cloud point set; Based on the corresponding connection of the triangular face set, all point clouds in the pre-scanned surface point cloud set or the scanned surface point cloud set are generated to generate a pre-scanned 3D model or a scanned 3D model.

6. The rotary table type automated three-dimensional scanning and measuring system according to claim 5, characterized in that: The method of generating a triangular facet sequence of a single point cloud point based on the empty circle property comprises the following specific steps: Connect all the neighborhood point cloud points in the neighborhood point cloud point set with the single point cloud point to generate a center line sequence of the single point cloud point; Determine the sorting index based on the center line sequence, and mark a single neighborhood point cloud point according to the sorting index; Connect the neighboring point cloud points in sequence according to the marking order to generate the neighboring lines, delete the neighboring lines that intersect with the center line, and generate the initial triangle patch sequence of a single point cloud point; Delete neighboring point cloud points to generate a triangle patch sequence of a single point cloud point.

7. The rotary table type automated three-dimensional scanning and measuring system according to claim 1, characterized in that: The comparison and analysis module includes a comparison detection unit and a defect analysis unit; The comparison detection unit obtains the scanned 3D model, calls the standard 3D model of the inspected workpiece, and uses Boolean operations to determine whether the size of the inspected workpiece is qualified. If it is unqualified, the scanned 3D model and the standard 3D model are sent to the defect analysis unit; The defect analysis unit generates a defect label based on a defect recognition network.

8. The rotary table type automated three-dimensional scanning and measuring system according to claim 7, characterized in that: The defect recognition network generates defect labels including: Converting the scanned three-dimensional model and the standard three-dimensional model into a first image and a second image; A large convolution kernel and a small convolution kernel are respectively used to mine the first image and the second image in turn, and a first feature and a second feature are generated by pooling; The first feature and the second feature are processed by combining the self-attention mechanism with the residual structure to generate the first code and the second code; The first code and the second code are concatenated into a combined code, each element in the combined code is processed by an activation function and accumulated to generate an aggregated code; The aggregate code is deconvolved twice to generate dimensional features, and a probability vector is generated through linear modulation and softmax function, and the defect label with the highest probability in the probability vector is output.

Images