On-line vision measurement method and system based on vehicle body aluminum alloy die casting production

Abstract

The invention provides an online vision measurement method and system based on vehicle body aluminum alloy die casting production, and relates to the technical field of industrial automation, and the method comprises the steps: building a three-dimensional measurement model of an aluminum alloy die casting; acquiring high-density point cloud data of the surface of the aluminum alloy die casting; calculating critical dimension parameters of the aluminum alloy die casting based on the registered point cloud data; and feature information of the three-dimensional CAD model of the to-be-stamped aluminum alloy die casting is obtained, the robot arm is controlled to grab the randomly-placed to-be-stamped aluminum alloy die casting according to the optimal movement track, the to-be-stamped aluminum alloy die casting is accurately placed to the target positioning position of the stamping die, and automatic feeding and accurate positioning of the to-be-stamped aluminum alloy stamping part are completed.

Description

Technical Field

[0001] The invention relates to industrial automation technology, and in particular to an online visual measurement method and system based on the production of aluminum alloy die castings for vehicle bodies. Background Art

[0002] The production process of aluminum alloy die castings is complex, which easily introduces quality problems such as dimensional deviation and surface defects, which seriously affect the assembly accuracy of the body and the performance of the whole vehicle. Traditional die casting quality inspection mainly relies on manual sampling and offline measurement, which has the disadvantages of low efficiency and strong subjectivity, and it is difficult to meet the high quality and high efficiency requirements of the modern automobile manufacturing industry.

[0003] Existing technical solutions still have limitations in practical applications. The surface of die-castings is complex and the reflection is uneven. The image quality is easily affected by lighting conditions and environmental factors, making it difficult to ensure measurement accuracy and stability. Most technical solutions use two-dimensional imaging from a single perspective, which makes it difficult to fully obtain the three-dimensional morphology information of die-castings, and the measurement capability of complex structural parts is limited. Summary of the invention

[0004] The embodiments of the present invention provide an online visual measurement method and system based on the production of aluminum alloy die castings for vehicle bodies, which can solve the problems in the prior art.

[0005] According to a first aspect of the embodiments of the present invention,

[0006] Provide online visual measurement methods based on the production of aluminum alloy die castings for car bodies, including:

[0007] Acquire the feature information of the 3D CAD model of the aluminum alloy die-casting to be measured, determine the key measurement feature points of the aluminum alloy die-casting, and establish a 3D measurement model of the aluminum alloy die-casting; use a binocular stereo vision camera to shoot the aluminum alloy die-casting in multiple perspectives, with a large depth of field and high resolution to obtain high-density point cloud data on the surface of the aluminum alloy die-casting; input the acquired high-density point cloud data on the surface of the aluminum alloy die-casting into the point cloud data intelligent preprocessing module, automatically optimize the high-density point cloud data on the surface of the aluminum alloy die-casting through downsampling, denoising and smoothing processing, and obtain high-quality streamlined point cloud data on the surface of the aluminum alloy die-casting; The simplified point cloud data is input into the intelligent registration module based on the iterative closest point algorithm. Through coarse registration and fine registration, the high-precision coordinate transformation relationship between the high-quality simplified point cloud data and the three-dimensional measurement model is obtained, and the accurate registration of the high-quality simplified point cloud data and the three-dimensional CAD model is realized; based on the registered point cloud data, the edge extraction algorithm and the regional growth algorithm are used to accurately extract the three-dimensional coordinate information of the key measurement feature points, and the extracted three-dimensional coordinate information is input into the intelligent size calculation module. By automatically selecting the measurement benchmark and using the point-line distance and point-surface distance calculation methods, the key size parameters of the aluminum alloy die-casting are calculated;

[0008] Acquire feature information of a three-dimensional CAD model of an aluminum alloy die-casting to be stamped, determine the optimal positioning reference surface of the aluminum alloy die-casting to be stamped in a stamping die, and establish a stamping positioning measurement model of the aluminum alloy die-casting to be stamped; based on the high-precision coordinate transformation relationship between the key dimensional parameters of the aluminum alloy die-casting and the stamping positioning measurement model of the aluminum alloy die-casting to be stamped, achieve accurate registration between randomly placed aluminum alloy die-castings to be stamped and the optimal positioning state, and accurately extract the three-dimensional spatial position information of the optimal positioning reference surface of the aluminum alloy die-casting to be stamped;

[0009] The extracted three-dimensional spatial position information is input into the robot motion path planning module to automatically generate the optimal motion trajectory and target placement position coordinates for the robot arm to grab the aluminum alloy die-castings to be stamped; the robot arm is controlled to grab the randomly placed aluminum alloy die-castings to be stamped according to the optimal motion trajectory, and accurately place them at the target positioning position of the stamping die, completing the automated loading and precise positioning of the aluminum alloy stampings to be stamped.

[0010] According to a second aspect of the embodiments of the present invention,

[0011] Provide online visual measurement system based on the production of aluminum alloy die castings for car bodies, including:

[0012] The first unit is used to obtain the feature information of the three-dimensional CAD model of the aluminum alloy die-casting to be measured, determine the key measurement feature points of the aluminum alloy die-casting, and establish a three-dimensional measurement model of the aluminum alloy die-casting; use a binocular stereo vision camera to shoot the aluminum alloy die-casting in multiple perspectives, with a large depth of field and high resolution to obtain high-density point cloud data on the surface of the aluminum alloy die-casting; input the obtained high-density point cloud data on the surface of the aluminum alloy die-casting into the point cloud data intelligent preprocessing module, and automatically optimize the high-density point cloud data on the surface of the aluminum alloy die-casting through downsampling, denoising and smoothing processing to obtain high-quality streamlined point cloud data on the surface of the aluminum alloy die-casting ; Input high-quality streamlined point cloud data into the intelligent registration module based on the iterative closest point algorithm, and obtain the high-precision coordinate transformation relationship between the high-quality streamlined point cloud data and the three-dimensional measurement model through coarse and fine registration, so as to achieve accurate registration of high-quality streamlined point cloud data with the three-dimensional CAD model; Based on the registered point cloud data, the edge extraction algorithm and the regional growth algorithm are used to accurately extract the three-dimensional coordinate information of the key measurement feature points, and the extracted three-dimensional coordinate information is input into the intelligent size calculation module, and the key size parameters of the aluminum alloy die-casting are calculated by automatically selecting the measurement benchmark and using the point-line distance and point-surface distance calculation methods;

[0013] The second unit is used to obtain feature information of a three-dimensional CAD model of an aluminum alloy die-casting to be stamped, determine the optimal positioning reference surface of the aluminum alloy die-casting to be stamped in a stamping die, and establish a stamping positioning measurement model of the aluminum alloy die-casting to be stamped; based on the high-precision coordinate transformation relationship between the key dimensional parameters of the aluminum alloy die-casting and the stamping positioning measurement model of the aluminum alloy die-casting to be stamped, realize accurate registration between randomly placed aluminum alloy die-castings to be stamped and the optimal positioning state, and accurately extract the three-dimensional spatial position information of the optimal positioning reference surface of the aluminum alloy die-casting to be stamped;

[0014] The third unit is used to input the extracted three-dimensional spatial position information into the robot motion path planning module, automatically generate the optimal motion trajectory and target placement position coordinates for the robot arm to grab the aluminum alloy die-castings to be stamped; control the robot arm to grab the randomly placed aluminum alloy die-castings to be stamped according to the optimal motion trajectory, and accurately place them at the target positioning position of the stamping die, thereby completing the automatic loading and precise positioning of the aluminum alloy stampings to be stamped.

[0015] According to a third aspect of the embodiments of the present invention,

[0016] An electronic device is provided, comprising:

[0017] processor;

[0018] a memory for storing processor-executable instructions;

[0019] The processor is configured to call the instructions stored in the memory to execute the aforementioned method.

[0020] A fourth aspect of the embodiments of the present invention is:

[0021] A computer-readable storage medium is provided, on which computer program instructions are stored. When the computer program instructions are executed by a processor, the aforementioned method is implemented.

[0022] This paper elaborates on the technical scheme for accurate registration of high-quality streamlined point cloud data with 3D CAD models. The scheme first converts the two heterogeneous data into a unified point set representation, performs preprocessing to improve the data quality, then uses the principal component analysis method for coarse registration, and then uses the iterative closest point algorithm for fine registration, and finally obtains the rigid body transformation matrix that accurately registers the point cloud data to the CAD model coordinate system. The entire registration process makes full use of the geometric feature information of the point cloud data and the CAD model, and finally achieves a high-precision registration effect. Based on the registered point cloud data, subsequent applications such as reverse engineering modeling and digital detection can be further carried out.

[0023] This paper is based on the offline establishment of a high-precision correspondence between the theoretical model of die-castings and the measurement model, combined with real-time visual positioning on the stamping equipment, and uses coordinate transformation to align the measurement model to the actual workpiece posture, thereby guiding the robot to achieve precise grasping and optimal positioning of the workpiece, comprehensively improving the automation and intelligence level of the stamping production line. BRIEF DESCRIPTION OF THE DRAWINGS

[0024] Figure 1 It is a schematic flow chart of an online visual measurement method for the production of aluminum alloy die castings for vehicle bodies according to an embodiment of the present invention;

[0025] Figure 2 It is a schematic structural diagram of an online visual measurement system for the production of aluminum alloy die castings for vehicle bodies according to an embodiment of the present invention. DETAILED DESCRIPTION

[0026] In order to make the purpose, technical solution and advantages of the embodiments of the present invention clearer, the technical solution in the embodiments of the present invention will be clearly and completely described below in conjunction with the drawings in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0027] The technical solution of the present invention is described in detail with specific embodiments below. The following specific embodiments can be combined with each other, and the same or similar concepts or processes may not be described in detail in some embodiments.

[0028] Figure 1 FIG. 1 is a flow chart of an online visual measurement method for producing aluminum alloy die castings for vehicle bodies according to an embodiment of the present invention. Figure 1 As shown, the method includes:

[0029] S1. Obtain feature information of the three-dimensional CAD model of the aluminum alloy die-casting to be measured, determine the key measurement feature points of the aluminum alloy die-casting, and establish a three-dimensional measurement model of the aluminum alloy die-casting; use a binocular stereo vision camera to shoot the aluminum alloy die-casting in multiple perspectives, with a large depth of field and high resolution to obtain high-density point cloud data on the surface of the aluminum alloy die-casting; input the obtained high-density point cloud data on the surface of the aluminum alloy die-casting into the point cloud data intelligent preprocessing module, automatically optimize the high-density point cloud data on the surface of the aluminum alloy die-casting through downsampling, denoising and smoothing processing, and obtain high-quality streamlined point cloud data on the surface of the aluminum alloy die-casting; The high-quality streamlined point cloud data is input into the intelligent registration module based on the iterative closest point algorithm. Through coarse and fine registration, the high-precision coordinate transformation relationship between the high-quality streamlined point cloud data and the three-dimensional measurement model is obtained, and the accurate registration of the high-quality streamlined point cloud data and the three-dimensional CAD model is realized; based on the registered point cloud data, the edge extraction algorithm and the regional growth algorithm are used to accurately extract the three-dimensional coordinate information of the key measurement feature points, and the extracted three-dimensional coordinate information is input into the intelligent size calculation module. By automatically selecting the measurement benchmark, the point-line distance and point-surface distance calculation methods are used to calculate the key size parameters of the aluminum alloy die-casting;

[0030] S2. Acquire the characteristic information of the three-dimensional CAD model of the aluminum alloy die-casting to be stamped, determine the optimal positioning reference surface of the aluminum alloy die-casting to be stamped in the stamping die, and establish a stamping positioning measurement model of the aluminum alloy die-casting to be stamped; based on the high-precision coordinate transformation relationship between the key dimensional parameters of the aluminum alloy die-casting and the stamping positioning measurement model of the aluminum alloy die-casting to be stamped, realize the precise alignment between the randomly placed aluminum alloy die-casting to be stamped and the optimal positioning state, and accurately extract the three-dimensional spatial position information of the optimal positioning reference surface of the aluminum alloy die-casting to be stamped;

[0031] S3. Input the extracted three-dimensional spatial position information into the robot motion path planning module, automatically generate the optimal motion trajectory and target placement position coordinates for the robot arm to grab the aluminum alloy die-castings to be stamped; control the robot arm to grab the randomly placed aluminum alloy die-castings to be stamped according to the optimal motion trajectory, and accurately place them at the target positioning position of the stamping die, completing the automatic loading and precise positioning of the aluminum alloy stampings to be stamped.

[0032] In an optional embodiment,

[0033] The steps of obtaining feature information of a three-dimensional CAD model of an aluminum alloy die casting to be measured, determining key measurement feature points of the aluminum alloy die casting, and establishing a three-dimensional measurement model of the aluminum alloy die casting include:

[0034] Obtain the three-dimensional CAD model file of the aluminum alloy die-casting to be measured through the product data management system, and import the obtained three-dimensional CAD model file into the computer-aided design software; use the function modules provided by the computer-aided design software to identify the key geometric features of the aluminum alloy die-casting through a feature tree or a hierarchical structure, including a plane, a cylindrical surface, a conical surface, a spherical surface and a free-form surface, and extract the type, position, size, and normal vector of the key geometric features; extract the key dimensional parameters and the corresponding tolerance grade of the aluminum alloy die-casting through dimensioning and geometric tolerance information; extract the material grade and mechanical property parameters of the aluminum alloy die-casting through material attribute information; encapsulate the extracted feature information of the aluminum alloy die-casting according to a preset data format, and the feature information includes the key geometric feature information, key dimensional parameters, and the corresponding tolerance grade, material grade, and mechanical property parameters of the aluminum alloy die-casting to be measured;

[0035] According to the functional requirements and assembly requirements of the aluminum alloy die casting, combined with the extracted geometric feature information, key geometric features that affect the performance and quality of the die casting are determined, measurement sampling points are arranged on the key geometric features using a measurement tool, and for complex surface features, a discretization method is used to convert the surface into a series of discrete measurement sampling points. For small-sized or thin-walled features, the accessibility and measurability of the measurement sampling points are considered and optimized and adjusted, and key measurement feature point information is determined based on the measurement sampling point information, and the key measurement feature point information is extracted and packaged according to a preset data structure;

[0036] The feature information and key measurement feature point information of the acquired 3D CAD model of the aluminum alloy die-casting are imported into the measurement software, and the corresponding measuring probe stylus ball is generated according to the coordinate position information of the key measurement feature points. The 3D CAD model of the aluminum alloy die-casting is converted into a triangular face mesh model using the surface meshing tool provided by the measurement software. The generated measuring probe stylus ball is virtually assembled with the triangular face mesh model and the virtual calibration of the measuring probe is completed. The corresponding measurement strategy and measurement path are set according to the type and attribute of the key measurement feature points to generate the 3D measurement model of the aluminum alloy die-casting.

[0037] Exemplarily, the detailed technical contents of obtaining the feature information of the three-dimensional CAD model of the aluminum alloy die casting to be measured, determining the key measurement feature points of the aluminum alloy die casting, and establishing the three-dimensional measurement model of the aluminum alloy die casting are as follows:

[0038] The three-dimensional CAD model file of the aluminum alloy die-casting to be measured is obtained through the product data management system, and the obtained three-dimensional CAD model file is imported into the computer-aided design software.

[0039] Using the product data management (PDM) system, retrieve and download the 3D CAD model file of the aluminum alloy die casting to be measured.

[0040] Import the downloaded CAD model file into mainstream computer-aided design software. During the import process, select the appropriate conversion options as needed to ensure the integrity and accuracy of the model data.

[0041] Using the functional modules provided by the computer-aided design software, the key geometric features of aluminum alloy die castings, including planes, cylinders, cones, spheres and free-form surfaces, are identified through feature trees or hierarchical structures, and the types, positions, sizes and normal vectors of the key geometric features are extracted.

[0042] In the CAD software, by browsing the feature tree or the hierarchical structure, the key geometric features of the aluminum alloy die casting are identified, and the key geometric features include: plane features, cylindrical surface features, conical surface features, spherical surface features, and free-form surface features.

[0043] For each identified key geometric feature, extract its position information, such as the reference coordinate system and origin coordinates of the feature; extract its size information, such as the length and width of the plane, the diameter and height of the cylinder, the base diameter and top angle of the cone, etc.; extract its normal vector information, such as the normal vector of the plane, the normal vector of the surface, etc.

[0044] The extracted feature information is encapsulated according to the preset data structure.

[0045] Through dimensioning and geometric tolerance information, the key dimensional parameters and corresponding tolerance grades of aluminum alloy die castings are extracted.

[0046] In the CAD model, dimensioning and geometric tolerance information are identified. Dimensioning gives the key size parameters of the feature, such as length, width, height, diameter, angle, etc.; geometric tolerance information gives the shape and position accuracy requirements of the feature, such as flatness, roundness, verticality, position, etc.

[0047] For each dimension, extract its name, value and tolerance grade.

[0048] For each geometric tolerance, extract its type, value and reference datum.

[0049] The extracted dimension parameters and tolerance information are packaged according to the preset data structure.

[0050] The material grade and mechanical property parameters of aluminum alloy die castings are extracted through material property information.

[0051] The material grade of the aluminum alloy die casting is extracted from the material attribute information of the CAD model. Based on the material grade, the mechanical properties of the material, such as yield strength, tensile strength, elongation, elastic modulus, etc., are queried and extracted from the material database.

[0052] The extracted material information is packaged according to the preset data structure.

[0053] According to the functional requirements and assembly requirements of aluminum alloy die-castings, combined with the extracted geometric feature information, the key geometric features that affect the performance and quality of die-castings are determined, and measurement sampling points are arranged on the key geometric features using measurement tools.

[0054] Analyze the functional and assembly requirements of aluminum alloy die castings to determine the key geometric features that affect their performance and quality. For example, for a cylinder head die casting, its key geometric features may include the diameter and coaxiality of the cylinder bore, the flatness and position of the assembly surface, etc.

[0055] On the identified key geometric features, use the measurement tool to arrange measurement sampling points. For plane features, 3 to 5 measurement points can be arranged; for cylindrical features, 4 to 8 measurement points can be arranged; for conical features, 6 to 12 measurement points can be arranged; for spherical features, 8 to 16 measurement points can be arranged.

[0056] For complex surface features, the discretization method is used to convert the surface into a series of discrete measurement sampling points. For example, the surface can be divided into several grid cells, and a measurement point is arranged at the center of each grid cell. The size of the grid cell can be reasonably selected according to the size and complexity of the surface, generally 1 to 5 mm.

[0057] For small or thin-walled features, consider the accessibility and measurability of the measurement sampling points and make optimization adjustments. For example, for holes with a diameter less than 5 mm, the number of measurement points can be reduced; for features with a wall thickness less than 2 mm, the measurement point position can be adjusted to avoid deformation or damage to the workpiece by the probe contact.

[0058] The feature information and key measurement feature point information of the acquired 3D CAD model of the aluminum alloy die casting are imported into the measurement software, and the corresponding measurement probe stylus ball is generated according to the coordinate position information of the key measurement feature points.

[0059] The acquired feature information and key measurement feature point information of the aluminum alloy die-casting 3D CAD model are imported into the measurement software according to the agreed data format.

[0060] In the measurement software, the corresponding measuring probe ball is generated according to the coordinate position information of each key measurement feature point. The diameter of the probe ball is set according to the selected probe specification, generally 1 to 8 mm.

[0061] The position and direction of the probe ball are set according to the position and normal vector information of the measurement feature point to ensure that the probe ball can properly contact the measured surface and obtain accurate measurement data.

[0062] The surface meshing tool provided by the measurement software is used to convert the 3D CAD model of the aluminum alloy die casting into a triangular face mesh model.

[0063] In the measurement software, the surface meshing tool is used to discretize the imported aluminum alloy die-casting 3D CAD model and convert it into a mesh model composed of a large number of triangular facets.

[0064] The density of the triangular facet mesh can be reasonably selected according to the measurement accuracy requirements and calculation efficiency. For general die castings, the size of the triangular facet can be 1 to 5 mm; for precision die castings, the size of the triangular facet can be 0.5 to 2 mm.

[0065] For surface features, the mesh density in the local area should be appropriately increased to ensure that the discretized mesh model can accurately represent the shape characteristics of the original surface.

[0066] The generated measuring probe stylus ball and the triangular patch mesh model are virtually assembled and the virtual calibration of the measuring probe is completed.

[0067] In the measurement software, the measuring probe ball generated in step 6 and the triangular facet mesh model generated in step 7 are virtually assembled so that the probe ball comes into contact with the measured surface.

[0068] Through virtual assembly, check whether the measuring probe can properly contact the measured surface at different measuring positions, and whether there are problems such as interference or collision. If there are problems, it is necessary to adjust the position and direction of the measuring probe, or optimize the measuring path.

[0069] After completing the virtual assembly, the measuring probe is virtually calibrated. The purpose of virtual calibration is to establish the transformation relationship between the measuring probe coordinate system and the workpiece coordinate system to ensure that the measured data can accurately represent the position and direction in the workpiece coordinate system. The virtual calibration process usually includes steps such as collecting calibration point data and fitting calibration parameters.

[0070] According to the type and attributes of key measurement feature points, the corresponding measurement strategy and measurement path are set to generate a 3D measurement model of the aluminum alloy die casting.

[0071] According to the type (such as plane point, cylindrical point, conical point, etc.) and attributes (such as size, tolerance, etc.) of each key measurement feature point, set the corresponding measurement strategy, including measurement method, measurement speed, measurement force and other parameters. For example, for plane features, three-point or four-point measurement method can be used; for cylindrical features, circumference measurement method can be used; for features with tight dimensional tolerance, the measurement speed can be appropriately reduced and the measurement force can be increased.

[0072] Plan the measurement path according to the position distribution of key measurement feature points and the measurement sequence requirements. The measurement path should minimize the probe's idle travel and switching times to improve measurement efficiency. At the same time, the measurement path should also consider avoiding collision interference between the probe and the workpiece or fixture.

[0073] Apply the set measurement strategy and measurement path to the virtual assembly model generated in step 8 to form a complete 3D measurement model of the aluminum alloy die casting. The 3D measurement model contains the geometry information, key dimensions and tolerance information, material property information, measurement point distribution information, measurement strategy and measurement path of the die casting to be measured, providing the necessary input data for the subsequent measurement program generation and measurement process execution.

[0074] Through the above steps, all technical contents of obtaining the feature information of the 3D CAD model of aluminum alloy die castings, determining the key measurement feature points, and establishing the 3D measurement model have been completed. This solution makes full use of modern tools such as product data management system, CAD software, and measurement software. Starting from the product design model, the key geometry and quality features of the die castings are extracted, and on this basis, the measurement sampling point arrangement, probe virtual calibration, measurement strategy and path planning are completed, and finally a complete 3D measurement model of the die castings is established. This solution can effectively guide the development and application of online measurement systems for aluminum alloy die castings, improve measurement efficiency and accuracy, and is of great significance to the guarantee of production quality and performance of die castings.

[0075] In an optional embodiment,

[0076] The high-quality simplified point cloud data is input into the intelligent registration module based on the iterative closest point algorithm. Through coarse registration and fine registration, the high-precision coordinate transformation relationship between the high-quality simplified point cloud data and the 3D measurement model is obtained. The steps to achieve accurate registration of the high-quality simplified point cloud data and the 3D CAD model include:

[0077] Inputting high-quality simplified point cloud data and a three-dimensional measurement model into an intelligent registration module, wherein the high-quality simplified point cloud data and the measurement model are both represented in the form of a three-dimensional coordinate point set;

[0078] Preprocess the input high-quality streamlined point cloud data and 3D measurement models, including data format conversion, noise filtering and outlier removal, to improve data quality and registration efficiency;

[0079] The high-quality streamlined point cloud data and the 3D measurement model are roughly aligned using a principal component analysis-based method. The geometric center points of the two data sets are calculated as the centroid, the covariance matrix is ​​calculated, and the eigenvalue decomposition is performed. The high-quality streamlined point cloud data and the 3D measurement model are rotated along their eigenvector directions to align the main directions of the two data sets, and the rotated high-quality streamlined point cloud data is translated so that its centroid coincides with the centroid of the 3D measurement model.

[0080] The iterative closest point algorithm is used to perform fine registration on the high-quality streamlined point cloud data and the 3D measurement model after rough registration. Under the current transformation matrix, for each point in the high-quality streamlined point cloud data, the nearest point is found in the 3D measurement model to form corresponding point pairs. The optimal rigid body transformation matrix is ​​calculated using the least squares method to minimize the sum of squared distance errors of all corresponding point pairs. The high-quality streamlined point cloud data is transformed using the solved transformation matrix to obtain new high-quality streamlined point cloud data. It is determined whether the iteration termination condition is met. If not, the updated high-quality streamlined point cloud data is used as the new input to continue the iteration. If it is met, the final transformation matrix is ​​output.

[0081] The transformation matrix obtained by precise registration is used to transform the high-quality simplified point cloud data into the coordinate system of the 3D measurement model to obtain the registered high-quality simplified point cloud data.

[0082] Exemplarily, high-quality simplified point cloud data is input into an intelligent registration module based on an iterative closest point algorithm, and through coarse registration and fine registration, a high-precision coordinate transformation relationship between the high-quality simplified point cloud data and the three-dimensional measurement model is obtained, and the detailed technical content of realizing accurate registration of high-quality simplified point cloud data with the three-dimensional CAD model is as follows:

[0083] The high-quality simplified point cloud data and the three-dimensional measurement model are input into the intelligent registration module. The high-quality simplified point cloud data and the measurement model are both represented in the form of a three-dimensional coordinate point set.

[0084] High-quality simplified point cloud data is collected by the online measurement system and obtained through preprocessing such as downsampling and filtering. Its data format is usually a common point cloud format such as PCD, PLY, XYZ, etc. For example, a typical PCD format point cloud data contains a text header of information such as the point cloud version number, coordinate field, size type, number of points, viewpoint parameters, etc., and ASCII point data of X-coordinate, Y-coordinate, and Z-coordinate values ​​separated by spaces.

[0085] The three-dimensional measurement model is generated through step nine, and its data format is usually a common three-dimensional model format such as STL, OBJ, VRML, etc. For example, a typical STL format measurement model contains multiple triangular facets, each facet is composed of three vertices and a normal vector, and the vertices and normal vectors are represented by three floating point coordinate values.

[0086] The input high-quality streamlined point cloud data and 3D measurement model are preprocessed, including data format conversion, noise filtering and outlier removal, to improve data quality and alignment efficiency.

[0087] Convert high-quality streamlined point cloud data and 3D measurement models into a unified data format, such as a 3D point set. For example, read the PCD point cloud data as a matrix of N rows and 3 columns, with each row corresponding to the XYZ coordinates of a point; read the STL measurement model as a matrix of M rows and 3 columns, with each row corresponding to the vertex coordinates of a triangular face. It should be noted that the same vertex may be shared by multiple faces.

[0088] The converted point cloud matrix and model matrix are subjected to noise filtering and outlier removal. Commonly used filtering algorithms include statistical filtering and radius filtering. Taking statistical filtering as an example, the distance from each point to its nearest neighbor is first calculated, and then the mean and standard deviation of all distances are calculated. If the nearest neighbor distance of a point is greater than the mean plus 1.5 times the standard deviation, the point is marked as an outlier and removed from the data.

[0089] A principal component analysis-based method is used to perform coarse alignment on high-quality streamlined point cloud data and three-dimensional measurement models. By calculating the geometric center points of the two data sets as the centroid, the covariance matrix is ​​calculated and eigenvalue decomposition is performed. The high-quality streamlined point cloud data and the three-dimensional measurement model are rotated along their eigenvector directions to align the main directions of the two data sets. The rotated high-quality streamlined point cloud data is translated so that its centroid coincides with the centroid of the three-dimensional measurement model.

[0090] Calculate the geometric center points of the point cloud matrix P and the model matrix Q as their respective centroids. Taking the point cloud matrix P as an example, assuming it contains N points, the X coordinate of the centroid Cp is the sum of the X coordinates of all points divided by the number of points N, and the same applies to the Y and Z coordinates.

[0091] Calculate the de-centroided point cloud matrix P' and model matrix Q', that is, subtract the corresponding centroid coordinates from the XYZ coordinates of each point in P and Q. Use the matrices P' and Q' to construct a 3x3 covariance matrix, perform eigenvalue decomposition on the covariance matrix, and obtain three eigenvalues ​​and corresponding three eigenvectors arranged in descending order of eigenvalues.

[0092] Multiply the matrices P' and Q' by their eigenvector matrices respectively to obtain the rotated new matrices P" and Q". At this time, the main directions of P" and Q" are aligned. Calculate the change in the center of mass coordinates before and after the rotation as the translation matrix. Add the translation matrix to the matrix P" so that its center of mass coincides with the center of mass of the matrix Q".

[0093] The iterative nearest point algorithm is used to perform fine alignment on the high-quality streamlined point cloud data and the three-dimensional measurement model after rough alignment. Under the current transformation matrix, for each point in the high-quality streamlined point cloud data, the nearest point is found in the three-dimensional measurement model to form corresponding point pairs. The optimal rigid body transformation matrix is ​​calculated using the least squares method to minimize the sum of squared distance errors of all corresponding point pairs. The high-quality streamlined point cloud data is transformed using the solved transformation matrix to obtain new high-quality streamlined point cloud data. It is determined whether the iteration termination conditions are met. If not, the updated high-quality streamlined point cloud data is used as the new input to continue the iteration. If it is met, the final transformation matrix is ​​output.

[0094] Let the current transformation matrix be T, and the initial transformation matrix be the transformation matrix obtained by rough registration. Under the current transformation matrix T, transform each point in the point cloud matrix P to the coordinate system of the model matrix Q. For each transformed point, find the closest point in Q. If the distance between two points is less than a given threshold (such as 1 mm), they constitute a corresponding point pair. The search for the nearest point can be achieved by brute force enumeration.

[0095] Using all corresponding point pairs, the least squares objective function is constructed to solve the optimal rigid body transformation matrix so that the sum of the squares of the distances of all point pairs is minimized. This least squares problem can be solved by singular value decomposition. First, the coordinates of the point cloud points and the model points are subtracted from their centroids to obtain the centroid-free coordinates. Then, the centroid-free coordinate matrix of the point cloud points is multiplied by the transpose of the centroid-free coordinate matrix of the model points to obtain a 3x3 matrix W. Singular value decomposition is performed on W. Let W = U*S*V', then the optimal rotation matrix R = U*V', and the optimal translation vector t is the centroid of the model point minus the centroid of the point cloud point rotated by R.

[0096] Using the obtained rigid body transformation matrix (R, t), the point cloud matrix P is transformed to obtain the new point cloud matrix P new . Determine whether the iteration termination conditions are met. Common termination conditions include: the number of iterations reaches the preset value, the difference between the new and old transformation matrices is less than the threshold, the squared drop value of the distance between the corresponding point pairs is less than the threshold, etc. If it is met, the current transformation matrix is ​​output, otherwise P new As the input of the next iteration, the above process is repeated until the termination condition is met.

[0097] Using the transformation matrix obtained by precise registration, the high-quality simplified point cloud data is transformed into the coordinate system of the three-dimensional measurement model to obtain the registered high-quality simplified point cloud data.

[0098] Assume that the final transformation matrix obtained by precise registration is (R final ,t final ), multiply each point in the original point cloud data matrix P by the rotation matrix R final , plus the translation vector t final , get the new point cloud coordinate matrix after registration.

[0099] Through the above steps, the technical scheme for accurately registering high-quality streamlined point cloud data with three-dimensional CAD models is elaborated in detail. The scheme first converts the two heterogeneous data into a unified point set representation, and performs preprocessing to improve the data quality. Then, the principal component analysis method is used for coarse registration, and then the iterative closest point algorithm is used for fine registration. Finally, the rigid body transformation matrix that accurately registers the point cloud data to the CAD model coordinate system is obtained. The entire registration process makes full use of the geometric feature information of the point cloud data and the CAD model, and finally achieves a high-precision registration effect. Based on the registered point cloud data, subsequent applications such as reverse engineering modeling and digital detection can be further carried out.

[0100] In an optional embodiment,

[0101] Based on the registered point cloud data, the steps of accurately extracting the 3D coordinate information of key measurement feature points using edge extraction algorithm and region growing algorithm include:

[0102] Preprocessing the registered point cloud data, including noise filtering and outlier removal, to improve data quality and feature extraction accuracy, wherein the noise filtering adopts one or more methods of statistical filtering, Gaussian filtering or bilateral filtering;

[0103] The edge extraction algorithm uses an edge extraction algorithm based on local surface fitting to extract edge features in point cloud data. For each point in the point cloud data, a certain number of points are selected in its neighborhood, and the local surface is fitted using the least squares method to obtain the local surface equation at the point, and the partial derivatives of the local surface equation at the point in three coordinate directions are calculated to obtain the gradient vector, and the modulus of the gradient vector is calculated. Points with a gradient modulus greater than a set threshold are marked as potential edge points, and morphological operations and connectivity analysis are performed on the potential edge points to eliminate isolated points and short edges, extract continuous and significant edge features, and obtain an edge point set.

[0104] The region growing algorithm uses a region growing algorithm based on graph segmentation to extract regional features in point cloud data, regards point cloud data as an undirected graph, defines a similarity measurement function between points, calculates the similarity weight between each pair of points in the point cloud data, constructs an edge weight set of the graph, and uses a graph segmentation algorithm to divide the undirected graph into several non-intersecting subgraphs. The subgraphs obtained by the division are post-processed, small regions are merged, and noise regions are eliminated to obtain the regional segmentation result of the point cloud data;

[0105] The edge point set obtained by edge extraction is merged with the region segmentation result obtained by region growing, and the points located at the region boundary and belonging to the edge point set are extracted as candidate key measurement feature points. According to the geometric properties of the feature points and the measurement requirements, the most representative and reliable feature point subset is selected from the candidate key measurement feature points as the final key measurement feature points.

[0106] Evaluate the accuracy and completeness of the extracted key measurement feature points, calculate the position error and omission rate of the key measurement feature points by comparing with the reference data, adjust the algorithm parameters according to the evaluation results, optimize the neighborhood size of local surface fitting and the similarity measurement function of graph segmentation, and improve the extraction quality of key measurement feature points.

[0107] For example, based on the high-quality streamlined point cloud data after registration, the detailed technical contents of accurately extracting the three-dimensional coordinate information of key measurement feature points by using edge extraction algorithm and region growing algorithm are as follows:

[0108] The high-quality streamlined point cloud data after registration is preprocessed, including noise filtering and outlier removal, to improve data quality and feature extraction accuracy.

[0109] Noise filtering can be performed using one or more of the following methods: statistical filtering, Gaussian filtering, or bilateral filtering. Taking statistical filtering as an example, the average distance from each point to its nearest neighbor is first calculated, and then the mean and standard deviation of all average distances are calculated. If the average distance of the nearest neighbor of a point is greater than the mean distance plus 1.5 times the standard deviation of the distance, the point is marked as a noise point and removed from the point cloud data.

[0110] Outlier removal can be achieved by setting a density threshold. A local spherical neighborhood is constructed with a given radius (such as 5 mm) with a certain point as the center, and the number of other points falling within the neighborhood is counted. If the number of points is less than a given threshold (such as 10), the point is marked as an outlier and removed.

[0111] After filtering and outlier removal, the quality of point cloud data is improved, laying a good foundation for subsequent feature extraction. For example, the original point cloud data of a mechanical part surface contains 1 million points, of which noise points account for 5% and outliers account for 1%. After preprocessing, the amount of valid point cloud data is reduced to 950,000, and the data quality is significantly improved.

[0112] The edge features in point cloud data are extracted using an edge extraction algorithm based on local surface fitting.

[0113] For each point in the point cloud data, a certain number (e.g., 50) of nearest neighboring points are selected within a certain neighborhood, and these nearest neighboring points are used to fit the local surface. The local surface equation at the point can be obtained by the least squares method. The coefficients of the three coordinates x, y, and z in the surface equation (a i ,b i ,c i ) can be solved by minimizing the sum of squares of distances from neighborhood points to the fitting plane.

[0114] Based on the local surface equation, the partial derivatives (a i ,b i ,-1), get the normal vector at the point, and then normalize the normal vector to the unit normal vector.

[0115] Calculate the curvature of the local surface at each point as an indicator to measure the local shape mutation. Let the distance from each point in the neighborhood of a point to the fitting surface be d j , the curvature at this point is defined as the standard deviation of the distances from all neighboring points to the fitted surface.

[0116] Points with curvature greater than a set threshold (such as 1 mm) are marked as potential edge points to obtain a set of potential edge points.

[0117] Perform morphological operations and connectivity analysis on the potential edge point set. Use a spherical structural element with a radius of 3 mm to open the potential edge point set to eliminate isolated noise points; then use a spherical structural element with a radius of 1 mm to close the potential edge point set to fill the narrow and long holes on the edge. Then extract the connected domain of the potential edge point set after morphological processing, and eliminate the tiny connected domains with less than a threshold (such as 20 points). Finally, a significant and continuous edge point set is obtained.

[0118] A region growing algorithm based on graph segmentation is used to extract regional features from point cloud data.

[0119] The point cloud data is considered as an undirected graph, where the vertices are all points and the edges are the connections between any two points. A similarity measurement function between points is defined to measure the similarity of the local geometric properties of two points. Commonly used similarity measures include distance similarity (the closer the Euclidean distance between two points, the higher the similarity), normal vector similarity (the smaller the angle between the normal vectors of the two points, the higher the similarity), and curvature similarity (the smaller the absolute value of the curvature difference between the two points, the higher the similarity).

[0120] Calculate the similarity weights between each pair of vertices in the undirected graph and obtain the edge weight set of the graph.i and p j ,The similarity weight can comprehensively consider the three similarity metrics of distance, normal vector and curvature, and they are multiplied by different weight coefficients and then superimposed.

[0121] Based on the edge weights of the graph, the undirected graph is divided into several non-intersecting connected subgraphs using the graph partitioning algorithm. Commonly used graph partitioning algorithms include spectral clustering and graph cuts. Taking spectral clustering as an example, by solving the eigenvalues ​​and eigenvectors of the Laplacian matrix of the graph, the vertices are embedded in a low-dimensional space for clustering.

[0122] The subgraphs obtained by graph segmentation are merged and removed. For each subgraph, the number of vertices and the sum of edge weights are counted. When the number of vertices or the sum of edge weights is less than the set threshold, it is marked as a noise area and removed; when the sum of edge weights of two subgraphs is greater than the threshold, they are considered to be merged into the same area.

[0123] After region growing, the point cloud data is divided into different regions. These regions semantically correspond to different parts or faces of the object. For example, for the point cloud data of a car engine compartment, 37 initial sub-graphs are obtained through graph segmentation. After merging and eliminating, 12 main region faces such as the engine hood, bumper, and fender are obtained.

[0124] The edge point set obtained by edge extraction is fused with the segmentation result obtained by region growing to extract potential key measurement feature points.

[0125] For each region, extract its boundary point set. A boundary point is defined as: for a point in the region, if there is a point in its neighborhood that does not belong to the region, then it is a boundary point.

[0126] The intersection of the edge point set and the boundary point set of each region is calculated to obtain the candidate key measurement feature point set. The candidate feature point set contains points on the edges of objects and the boundaries of regions, and contains rich geometric feature information.

[0127] According to the characteristics of the measured object and the actual measurement requirements, the candidate feature point set is screened to obtain the most representative key measurement feature point set. The screening criteria include feature significance (prioritize the selection of points with large edge curvature and obvious normal vector changes), feature distribution uniformity (on the premise of meeting feature significance, select feature points that are evenly distributed and cover all major areas as much as possible), and neighborhood validity (the local neighborhood of the feature point should contain enough valid points to improve the accuracy of subsequent coordinate measurement).

[0128] Evaluate the accuracy and completeness of key measurement feature point extraction and optimize the feature extraction algorithm.

[0129] A set of reference key points is manually selected as real data and manually annotated using professional point cloud processing software. The reference points should include typical features of the object, such as edge endpoints, hole centers, and area inflection points.

[0130] Calculate the deviation between the extracted key feature points and the reference feature points respectively. For each extracted feature point, find the nearest point in the reference point set and calculate the Euclidean distance between the two as the position deviation; for each reference feature point, find the nearest point in the extracted point set. If the distance is greater than a given threshold (such as 2 mm), it is considered that the reference point has not been successfully extracted and is recorded as a missed point.

[0131] Define the evaluation indicators: mean and standard deviation of feature point position deviation, ratio of missing reference points to total reference points. Evaluate multiple different types of measurement objects and get the average value of each indicator.

[0132] According to the evaluation results, analyze the causes of errors and omissions, and optimize the feature extraction algorithm in a targeted manner. If the position deviation is large, consider optimizing the size of the local surface fitting neighborhood in edge extraction. If the value is too small, the fitting accuracy is insufficient, and if the value is too large, it is easy to introduce points from other regions; optimize the similarity weight in regional growth, and increase the distance weight to improve the adaptability of regional segmentation to the edge. If there are many missed points, consider lowering the curvature threshold of edge extraction, increasing the number of graph segmentation clusters, and appropriately relaxing various rules when screening candidate feature points. If omissions and mis-extractions coexist, it means that the overall generalization of the algorithm is insufficient. Consider uniformly optimizing the algorithm parameters on different types of prior data, or introducing an adaptive strategy.

[0133] Through quantitative evaluation and iterative optimization, the accuracy and completeness of key measurement feature point extraction are continuously improved. Taking the impeller point cloud data obtained by industrial CT scanning as an example, 1572 feature points were initially extracted and 532 reference points were manually annotated, with a coincidence of 76%. By adjusting parameters such as thresholds and weights and introducing adaptive neighborhood fitting, the coincidence was increased to 91% and the average position deviation was reduced to 0.15 mm, meeting the accuracy requirements of subsequent reverse modeling and dimensional detection.

[0134] The above steps elaborate on the technical solution of accurately extracting the three-dimensional coordinates of key measurement feature points from high-quality point cloud data based on edge extraction and region growing. This solution makes full use of the local geometric features of point cloud data and realizes cross-scale extraction from low-level original point cloud to high-level measurement features. By fusing edge features with regional features and introducing dynamic optimization of prior knowledge, the accuracy of feature extraction is greatly improved, ensuring reliability in subsequent reverse engineering applications. Compared with traditional manual calibration methods, this solution has the advantages of full automation, high efficiency, and high precision. Compared with the model matching-based method, this solution does not rely on CAD models and has a wider range of applications.

[0135] In an optional embodiment,

[0136] According to the high-precision coordinate transformation relationship between the key dimensional parameters of the aluminum alloy die-casting and the stamping positioning measurement model of the aluminum alloy die-casting to be stamped, the steps of realizing accurate registration between the randomly placed aluminum alloy die-casting to be stamped and the optimal positioning state and accurately extracting the three-dimensional spatial position information of the optimal positioning reference surface of the aluminum alloy die-casting to be stamped include:

[0137] In the CAD model or theoretical model of the aluminum alloy die casting, the theoretical coordinate values ​​of the key dimensional parameters are determined, the geometric features extracted from the stamping positioning measurement model of the aluminum alloy die casting to be stamped are aligned with the corresponding features in the theoretical model, and a corresponding relationship between the two is established. The optimal coordinate transformation matrix between the stamping positioning measurement model of the aluminum alloy die casting to be stamped and the theoretical model is calculated using the least squares method, including a translation vector and a rotation matrix, and the calculated optimal coordinate transformation matrix is ​​used as a high-precision coordinate transformation relationship between the key dimensional parameters and the stamping positioning measurement model of the aluminum alloy die casting to be stamped;

[0138] On the stamping equipment, a visual sensor or a laser sensor is used to collect the three-dimensional spatial position information of randomly placed aluminum alloy die-castings to be stamped, and the collected three-dimensional spatial position information is initially aligned with the stamping positioning measurement model of the aluminum alloy die-castings to be stamped to obtain the initial posture of the aluminum alloy die-castings to be stamped, and the initial posture of the aluminum alloy die-castings to be stamped is converted to the posture in the optimal positioning state by using a high-precision coordinate transformation relationship, and the converted posture is accurately aligned, and the posture of the aluminum alloy die-castings to be stamped in the optimal positioning state after the accurate alignment is obtained through an iterative optimization algorithm.

[0139] Exemplarily, according to the high-precision coordinate transformation relationship between the key dimensional parameters of the aluminum alloy die-casting and the stamping positioning measurement model of the aluminum alloy die-casting to be stamped, the precise registration between the randomly placed aluminum alloy die-casting to be stamped and the optimal positioning state is achieved, and the detailed technical content of accurately extracting the three-dimensional spatial position information of the optimal positioning reference surface of the aluminum alloy die-casting to be stamped is as follows:

[0140] In the CAD model or theoretical model of the aluminum alloy die-casting, the theoretical coordinate values ​​of the key dimensional parameters are determined, the geometric features extracted from the stamping positioning measurement model of the aluminum alloy die-casting to be stamped are aligned with the corresponding features in the theoretical model, and the corresponding relationship between the two is established.

[0141] The theoretical model of aluminum alloy die castings was obtained by modeling with 3D CAD software, and key dimensional parameters such as length, width, height, aperture, and wall thickness were defined in the feature management tree of the model. The theoretical values ​​of each key dimensional parameter in the model coordinate system were annotated using the dimension annotation tool. The die casting was scanned to obtain 3D volume data with a voxel size of 0.05 mm. The point cloud processing software was used to perform grayscale threshold segmentation on the 3D volume data and extract the surface point cloud of the die casting. Typical geometric features such as planes, holes, and slots were extracted from the point cloud using the curvature-based region growing algorithm and the edge-based feature extraction algorithm. Plane features are represented by their centroid coordinates and normal vectors, hole features are represented by their axis and diameter, and slot features are represented by their skeleton line and width.

[0142] The correspondence between the point cloud features and the CAD model features is manually selected. For example, six main plane features are extracted from the point cloud, corresponding to the six faces of the CAD model: front, back, left, right, top, and bottom. The inner cavity surface of the cuboid is extracted from the point cloud, corresponding to the inner cavity size in the CAD model. The eight straight holes extracted from the point cloud correspond to the bolt holes in the CAD model.

[0143] The least squares method is used to calculate the optimal coordinate transformation matrix between the stamping positioning measurement model of the aluminum alloy die-casting to be stamped and the theoretical model, including the translation vector and the rotation matrix. The calculated optimal coordinate transformation matrix is ​​used as the high-precision coordinate transformation relationship between the key dimensional parameters and the stamping positioning measurement model of the aluminum alloy die-casting to be stamped.

[0144] According to the correspondence between the point cloud features and CAD features obtained in step 1, the least squares objective function is established. The independent variable of the objective function is the coordinate transformation parameter to be determined, that is, the translation vector (t x ,t y ,t z ) and the rotation matrix (r 11 ,r 12 ,...,r 33 ), a total of twelve degrees of freedom.

[0145] For each pair of corresponding features, an error term is added to the objective function. For surface features, the error term is the distance between the centroid coordinates of the point cloud patch after transformation by the transformation matrix and the centroid coordinates of the CAD patch; for hole features and slot features, the error term is the distance between several sampling points on the axis or skeleton line of the point cloud feature after transformation by the transformation matrix and the axis or skeleton line of the CAD feature.

[0146] The least squares equations are constructed and solved iteratively using the Levenberg-Marquardt (LM) algorithm. The LM algorithm combines the advantages of gradient descent and Gauss-Newton method, and has high efficiency and accuracy in solving nonlinear least squares problems. The optimal solution of the twelve transformation parameters is obtained, and the third-order transformation matrix is ​​reconstructed. This matrix is ​​denoted as T pc , as the theoretical transformation relationship from the point cloud coordinate system to the CAD coordinate system.

[0147] On the stamping equipment, visual sensors or laser sensors are used to collect the three-dimensional spatial position information of randomly placed aluminum alloy die-castings to be stamped, and the collected three-dimensional spatial position information is initially aligned with the stamping positioning measurement model of the aluminum alloy die-castings to be stamped to obtain the initial posture of the aluminum alloy die-castings to be stamped.

[0148] Select four or more non-coplanar feature points on the die casting and paste visual positioning marks. Place the die casting randomly on the work surface of the stamping equipment, and use an industrial camera to collect multiple RGB images with markers from different angles. Through the principle of multi-eye vision measurement, perform feature extraction and stereo matching on the image to obtain the three-dimensional coordinates of each marker point in the camera coordinate system.

[0149] The coordinates of the marker points in the camera coordinate system are converted to the stamping equipment base coordinate system as the initial coordinates of the four reference points on the die casting surface. The corresponding four reference point coordinates are extracted from the measurement model point cloud. The two sets of corresponding point coordinates constitute a 3D-3D point pair.

[0150] Apply absolute orientation to calculate the coordinate transformation matrix T from the point cloud of the measurement model to the initial placement state of the workpiece wm Multiply the matrix on the left by the homogeneous coordinates of all points in the measurement model to obtain the initial pose point cloud of the workpiece in the equipment base coordinate system. wm The solution can be completed in real time within 0.1 second.

[0151] The initial posture of the aluminum alloy die-casting to be stamped is converted to the posture in the optimal positioning state by using high-precision coordinate transformation relationship, and the converted posture is accurately aligned. The posture of the aluminum alloy die-casting to be stamped in the optimal positioning state is obtained through an iterative optimization algorithm.

[0152] The theoretical transformation matrix T obtained above pc And the initial transformation matrix T obtained in step 3 wm Cascade to obtain the three-dimensional spatial transformation relationship T of the workpiece to be stamped from the random initial state to the optimal positioning state wc ;

[0153] T wcThe homogeneous coordinates of the initial pose point cloud of the die casting to be punched are multiplied on the left to obtain the approximate pose point cloud of the die casting in the optimal positioning state. The pose of this point cloud is very close to that of the measured model point cloud, but there is still a certain pose deviation.

[0154] In order to further improve the positioning accuracy, the ICP (Iterative Closest Point) registration algorithm is used to accurately register the two point clouds. The ICP algorithm minimizes the distance between the two point clouds through iterative optimization and solves the rigid body transformation matrix to make the two coincide.

[0155] In each round of ICP iteration, for each point in the pose point cloud, the nearest corresponding point is found in the reference point cloud to construct a point pair set; based on the point pair set, the optimal rigid body transformation matrix in the least squares sense is solved, and the pose point cloud is updated with the matrix; the mean and variance of the distance between the two point clouds after registration are calculated, and the iteration is terminated if it is less than the set threshold or the maximum number of iterations is reached.

[0156] After the iteration is terminated, the transformation matrix obtained from the last iteration is used as the result of precise alignment, and is left-multiplied by the pose point cloud to obtain the three-dimensional morphology and pose of the workpiece to be stamped in the optimal positioning state after precise alignment.

[0157] Taking the efficiency and accuracy of the best positioning state of the die casting to be stamped as the evaluation index, the parameters of the ICP algorithm (such as the corresponding point search radius, the maximum number of iterations, the convergence threshold, etc.) are experimentally optimized to obtain the optimal parameter combination that is robust, efficient and accurate. The algorithm speed is improved by parallel acceleration and local search.

[0158] The three-dimensional morphology data of the aluminum alloy die-casting to be stamped in the optimal positioning state obtained by precise registration is applied to the stamping automatic positioning system, and the spatial position of the optimal positioning reference plane of the die-casting is extracted as the target posture for grasping and handling by the stamping robot.

[0159] The point cloud data of the workpiece in the optimal positioning state is plane fitted, and the largest and relatively horizontal plane is extracted as the optimal positioning reference plane for the die casting to be placed on the stamping fixture. A spatial plane can be determined by the coordinates of the three points on the reference plane, and the spatial position and normal vector of the reference plane are sent to the robot controller.

[0160] The robot controller converts the position of the reference surface in the robot base coordinate system into the target motion angle of each axis joint, plans the motion trajectory from the current position to the target grasping position, controls the robot to accurately move to the optimal grasping position and open the suction cup. The die-casting is sucked up by vacuum adsorption and moved above the stamping fixture to align the reference surface of the die-casting with the positioning surface of the stamping fixture accurately. The die-casting is slowly lowered to place it on the fixture and the suction cup is released, completing the automatic transfer and positioning of the die-casting from the random state to the optimal stamping positioning state.

[0161] After positioning is completed, the robot quickly withdraws, the signal triggers the stamping mechanism to start, and the predetermined load is applied to the die casting to complete the stamping forming. After that, the robot grabs the finished die casting from the lower die and transfers it to the next process. The introduction of the visual-guided stamping automatic positioning system has greatly improved the automation level and production efficiency of the die casting stamping process.

[0162] The above steps elaborate on the high-precision coordinate transformation relationship between the key dimensional parameters of aluminum alloy die-castings and the stamping positioning measurement model to achieve accurate alignment from the random placement state of the die-castings to the optimal stamping positioning state, and apply it to the complete technical solution of the stamping automatic positioning system.

[0163] This solution is based on the offline establishment of a high-precision correspondence between the theoretical model of the die-casting and the measurement model, combined with real-time visual positioning on the stamping equipment. The measurement model is aligned to the actual workpiece posture through coordinate transformation, and then the robot is guided to achieve precise grasping and optimal positioning of the workpiece, which comprehensively improves the automation and intelligence level of the stamping production line.

[0164] In an optional embodiment,

[0165] The extracted three-dimensional spatial position information is input into the robot motion path planning module to automatically generate the optimal motion trajectory and target placement position coordinates for the robot arm to grab the aluminum alloy die-casting to be stamped; the robot arm is controlled to grab the randomly placed aluminum alloy die-casting to be stamped according to the optimal motion trajectory, and accurately place it at the target positioning position of the stamping die, and the steps of completing the automatic loading and precise positioning of the aluminum alloy stamping parts to be stamped include:

[0166] The posture information of the aluminum alloy die-casting to be stamped in the optimal positioning state obtained after accurate registration is transmitted to the robot motion path planning module. The posture information includes the position coordinates and posture angle of the aluminum alloy die-casting to be stamped in the equipment coordinate system. The posture angle uses Euler angles and quaternions to describe the rotation relationship of the aluminum alloy die-casting to be stamped relative to the equipment coordinate system. The posture information is transmitted to the robot motion path planning module in a standardized data format;

[0167] In the robot motion path planning module, the kinematic parameters, dynamic parameters and workspace constraints of the robot arm are comprehensively considered, and an optimization algorithm combining kinematics and dynamics is used to solve the optimal motion trajectory of the robot arm grabbing the die casting under the premise of satisfying the constraints;

[0168] According to the design parameters of the stamping die and the optimal positioning state of the aluminum alloy die-casting to be stamped, the optimal positioning state posture information of the die-casting is combined with the design parameters of the stamping die, and the target placement position coordinates of the aluminum alloy die-casting to be stamped in the stamping die coordinate system are calculated through coordinate transformation and space mapping. The target placement position coordinates are aligned with the positioning reference surface and key features of the stamping die;

[0169] The optimal motion trajectory and target placement position coordinates are transmitted to the robot control system. The robot control system realizes precise motion tracking and trajectory optimization of the robot arm through the motion control algorithm according to the received motion trajectory and target position coordinates, realizes precise contact and smooth operation between the robot arm and the aluminum alloy die-casting to be stamped through the end effector control algorithm, and controls the robot arm to perform grasping and placement operations;

[0170] Through the visual servo system and force sensing system of the robot arm, the visual sensor installed on the robot arm collects the position information of the aluminum alloy die-castings to be stamped and the stamping mold in real time. The force / torque sensor installed on the robot arm collects the contact force and torque information between the arm and the die-castings and the mold in real time. The visual servo and force feedback control instructions are combined with the motion control instructions. Through closed-loop control, real-time error compensation and trajectory fine-tuning are achieved during the robot arm's grasping and placement process, thereby improving positioning accuracy and operational stability.

[0171] Exemplarily, the extracted three-dimensional spatial position information is input into the robot motion path planning module, and the optimal motion trajectory and target placement position coordinates for the robot arm to grab the aluminum alloy die-casting to be stamped are automatically generated; the robot arm is controlled to grab the randomly placed aluminum alloy die-casting to be stamped according to the optimal motion trajectory, and accurately place it at the target positioning position of the stamping die, and the detailed technical content of completing the automatic loading and precise positioning of the aluminum alloy stamping parts to be stamped is as follows:

[0172] The posture information of the aluminum alloy die-casting to be stamped in the optimal positioning state obtained after precise alignment is passed to the robot motion path planning module. The posture information includes the position coordinates and posture angles of the aluminum alloy die-casting to be stamped in the equipment coordinate system. The posture angle uses Euler angles and quaternions to describe the rotation relationship of the aluminum alloy die-casting to be stamped relative to the equipment coordinate system. The posture information is passed to the robot motion path planning module in a standardized data format.

[0173] Through the registration calculation in the previous stage, the three-dimensional point cloud data of the die casting in the optimal positioning state is obtained, which reflects the accurate posture of the die casting relative to the equipment base coordinate system. In order to facilitate the robot path planning and control, the point cloud information needs to be converted into six-degree-of-freedom posture parameters.

[0174] First, the geometric center of the point cloud is obtained as the position coordinates (x, y, z) of the centroid of the die casting in the equipment base coordinate system. The principal component analysis of the point cloud is performed to calculate its main direction, and the eigenvectors corresponding to the three main directions are extracted as the orthogonal basis to describe the posture of the die casting. The Euler angles (α, β, γ) represent the rotation angles around the z-axis, y-axis, and x-axis respectively, which can be obtained by three elementary rotations of the rotation matrix corresponding to the three eigenvectors.

[0175] The position coordinates and attitude angles are combined into a posture information data packet (x, y, z, α, β, γ) in a fixed format and passed to the motion planning module through the TCP / IP protocol.

[0176] In the robot motion path planning module, the kinematic parameters, dynamic parameters and workspace constraints of the robot arm are comprehensively considered, and an optimization algorithm combining kinematics and dynamics is adopted to solve the optimal motion trajectory of the robot arm for grasping the die-casting while satisfying the constraints.

[0177] The robot arm is generally composed of several rigid rods connected by rotating or moving joints, forming an open-chain kinematic mechanism. The Denavit-Hartenberg (DH) method is a general method for describing the kinematic parameters of an open-chain mechanism. By measuring the size parameters of the robot arm and selecting the coordinate system, the DH parameter table of each joint rod of the arm is established. The kinematic forward solution takes the joint angle as input and solves the position and posture of the robot end through the cascade of coordinate transformation matrices; the kinematic inverse solution substitutes the desired end position and posture into the forward solution equation group and obtains the corresponding joint angle through numerical solution.

[0178] The dynamic model describes the force applied to the robot arm during its motion. The rigid body dynamics equation of the robot arm is established by the Lagrange equation method or the Newton-Euler method to solve the acceleration, velocity, and displacement of the robot joint and end under the action of the driving torque and external interference force. Considering the transmission characteristics of the motor and reducer, and supplementing factors such as friction and clearance, a complete dynamic model of the robot arm is established. The model parameters are calculated by the CAD model or measured by the identification experiment.

[0179] When solving the motion trajectory, it is necessary to combine the arm kinematics and dynamics model with the workspace constraints. Define the upper and lower limits of the robot's joint angle, angular velocity, and angular acceleration as the joint space constraints. Import the robot arm configuration into the simulation software, and establish the position and velocity constraints of the end Cartesian space based on the geometric information of the obstacles in the scene. According to the shape characteristics and quality parameters of the die casting, apply appropriate gripping force constraints to the robot end gripper.

[0180] The trajectory optimization model is constructed with the execution time, energy consumption, and smoothness of the robot joint trajectory as optimization targets, kinematic and dynamic equations as equality constraints, and joint space and Cartesian space restrictions as inequality constraints. The numerical optimization algorithm is used to search for the optimal motion trajectory of the robot arm from the initial posture to the target grasping posture under the premise of satisfying all constraints. The trajectory consists of a time series of several key posture points.

[0181] According to the design parameters of the stamping die and the optimal positioning state of the aluminum alloy die-casting to be stamped, the optimal positioning state posture information of the die-casting is combined with the design parameters of the stamping die. Through coordinate transformation and space mapping, the target placement position coordinates of the aluminum alloy die-casting to be stamped in the stamping die coordinate system are calculated. The target placement position coordinates are aligned with the positioning reference surface and key features of the stamping die.

[0182] The 3D CAD design data of stamping dies contains key information such as upper and lower die surfaces, guide mechanisms, and positioning reference surfaces. The position and posture of the center reference surface of the stamping die can be extracted as the origin and reference direction of the die coordinate system. The position and posture transformation matrix T of the die coordinate system relative to the equipment base coordinate system is determined by 3D measurement. md .

[0183] The optimal positioning state of the die casting to be stamped is determined by its position (x b ,y b ,z b ,α b ,β b ,γ b ) description, the position and posture of the die casting in the mold coordinate system can be obtained by coordinate transformation (x m ,y m ,z m ,α m ,β m ,γ m ).

[0184] To achieve precise positioning, the mold CAD data and die casting measurement data need to be spatially registered. Taking the minimum distance between the mold surface contour and the die casting contour as the optimization target, a constrained optimization problem is constructed as described in step 2 to solve the die casting coordinates that best match the two as the final target placement position of the die casting on the mold.

[0185] The optimal motion trajectory and target placement position coordinates are passed to the robot control system. The robot control system uses the motion control algorithm to achieve precise motion tracking and trajectory optimization of the robot arm based on the received motion trajectory and target position coordinates. The end effector control algorithm is used to achieve precise contact and smooth operation between the robot arm and the aluminum alloy die-casting to be stamped, and the robot arm is controlled to perform grasping and placement operations.

[0186] The robot control system hardware usually includes motion controller, servo drive, motor encoder, etc. According to the robot model, select the hardware model and specifications that meet the performance requirements, and reasonably set the controller parameters such as sampling period, servo gain, etc. Write the robot application, establish the communication interface with the motion planning module, and convert the planning results into executable control instructions.

[0187] The core goal of the motion control algorithm is to make the robot joint motion trajectory track the planned trajectory as accurately as possible. Common motion control algorithms include calculated torque method, PID (proportional-integral-differential) control, adaptive control, robust control, sliding mode variable structure control, etc. The calculated torque method is based on the robot dynamics model, substitutes the desired trajectory into the dynamics equation, pre-calculates the required torque of each joint, and then applies it to the driver. PID control uses the tracking error and its differential and integral as feedback, designs the proportional-integral-differential control law, and adjusts the control output in real time. Adaptive control adjusts the controller parameters online according to the robot parameter changes and external disturbances. Robust control and sliding mode variable structure control introduce nonlinear compensation to improve the robustness of trajectory tracking.

[0188] The gripper is installed at the end of the robot, which is driven by a linear translation cylinder and a rotary cylinder and is equipped with a pressure sensor. The gripper control algorithm needs to reasonably set the gripping force and gripping strategy according to the shape, size, and material properties of the die-casting. The grasping process is divided into four stages: approach, contact, force, and lift. In the approach stage, the cylinder extends quickly and the gripper moves at a constant speed; in the contact stage, the cylinder pressure rises, the gripper slows down and stops; in the force stage, the cylinder outputs the clamping torque, and the pressure peak is determined by the die-casting material. The peak force of aluminum alloy is generally 100-200N; in the lifting stage, the gripper maintains the torque to grasp the workpiece, and the robot arm lifts and translates. When the end of the robot contacts the die-casting, the contact force and torque need to increase slowly and steadily to avoid impact and damage caused by hard contact.

[0189] The placement process is similar to the grasping process. When the end of the robot moves to the target placement position, the pressure-controlled cylinder slowly descends to make the workpiece contact the mold positioning surface accurately. Then the cylinder is depressurized, the gripper opens, and the die casting is smoothly placed at the target position. Then the robot arm is raised to complete the placement process. The speed and strength of grasping and placing need to be repeatedly debugged and optimized to balance efficiency and stability.

[0190] Through the visual servo system and force sensing system of the robot arm, the visual sensor installed on the robot arm collects the position information of the aluminum alloy die-castings to be stamped and the stamping mold in real time. The force / torque sensor installed on the robot arm collects the contact force and torque information between the arm and the die-castings and the mold in real time. The visual servo and force feedback control instructions are combined with the motion control instructions. Through closed-loop control, real-time error compensation and trajectory fine-tuning of the robot arm during grasping and placing are achieved, thereby improving positioning accuracy and operational stability.

[0191] The robot end integrates visual sensors (usually industrial cameras) and force sensors. The industrial camera image is transmitted to the visual processing unit in real time, and the position information of the die-casting and the mold is obtained through algorithms such as image segmentation and edge extraction. It establishes communication with the robot controller to achieve visual servoing. Common visual servo control methods include position-based control (PBVS) and image-based control (IBVS). PBVS converts the three-dimensional position deviation reflected by the image information into robot joint motion instructions to control the robot to reduce spatial deviation. IBVS directly uses the image feature deviation as the control error to generate the robot joint speed instruction so that the image feature converges to the target value.

[0192] The six-dimensional force / torque sensor can measure the force and torque in three orthogonal directions at the end of the robot. After the sensor signal is amplified, filtered, and AD sampled, it communicates with the robot controller through the bus. The controller superimposes the force feedback information in the position servo control loop to achieve force / position hybrid control. The force control part makes a difference between the torque sensor signal and the expected torque to generate a robot speed or torque correction, so that the actual contact force quickly tracks the expected force; the position control part coordinates and corrects the joint movement to ensure that the predetermined trajectory is not disturbed.

[0193] During the contact and lifting phase of grabbing the die-casting, the visual system determines the relative position of the die-casting and the gripper, and promptly corrects the movement of the gripper so that it contacts the die-casting in the correct position. At the same time, the force sensing system monitors the change in contact force. When the contact force exceeds the threshold, the gripper stops moving and tightens.

[0194] During the descending stage of placing the die-casting, the visual system determines the relative position of the die-casting and the mold reference surface, controls the mechanical arm to fine-tune the movement, and reduces the alignment deviation. At the same time, the force sensing system monitors the contact force in the vertical direction. When the contact force exceeds the threshold, the descending movement is stopped, the clamp is slowly opened, and the die-casting is placed smoothly to avoid collision.

[0195] For example, the robot uses a PBVS visual servo system and a six-dimensional force / torque sensor. The visual processing unit sampling period is 20ms, and the force sampling period is 0.5ms. When the normal distance between the bottom surface of the die casting and the mold positioning surface is less than 1mm, the visual system outputs a descending command; when the normal contact force exceeds 10N, the force system outputs a stop command.

[0196] Through visual and force feedback, the accumulated errors in the robot's hand-eye calibration and the positioning between the robot and the die-casting equipment are effectively overcome, so that the actual grasping / placement position of the die-casting parts is always controlled within the range of ±0.1mm of the mold positioning surface, achieving high-precision coordination between the robot's loading and unloading and the stamping equipment.

[0197] Figure 2 FIG. 1 is a schematic diagram of the structure of an online visual measurement system for the production of aluminum alloy die castings for vehicle bodies according to an embodiment of the present invention. Figure 2 As shown, the system comprises:

[0198] The first unit is used to obtain the feature information of the three-dimensional CAD model of the aluminum alloy die-casting to be measured, determine the key measurement feature points of the aluminum alloy die-casting, and establish a three-dimensional measurement model of the aluminum alloy die-casting; use a binocular stereo vision camera to shoot the aluminum alloy die-casting in multiple perspectives, with a large depth of field and high resolution to obtain high-density point cloud data on the surface of the aluminum alloy die-casting; input the obtained high-density point cloud data on the surface of the aluminum alloy die-casting into the point cloud data intelligent preprocessing module, and automatically optimize the high-density point cloud data on the surface of the aluminum alloy die-casting through downsampling, denoising and smoothing processing to obtain high-quality streamlined point cloud data on the surface of the aluminum alloy die-casting ; Input high-quality streamlined point cloud data into the intelligent registration module based on the iterative closest point algorithm, and obtain the high-precision coordinate transformation relationship between the high-quality streamlined point cloud data and the three-dimensional measurement model through coarse and fine registration, so as to achieve accurate registration of high-quality streamlined point cloud data with the three-dimensional CAD model; Based on the registered point cloud data, the edge extraction algorithm and the regional growth algorithm are used to accurately extract the three-dimensional coordinate information of the key measurement feature points, and the extracted three-dimensional coordinate information is input into the intelligent size calculation module, and the key size parameters of the aluminum alloy die-casting are calculated by automatically selecting the measurement benchmark and using the point-line distance and point-surface distance calculation methods;

[0199] The second unit is used to obtain feature information of a three-dimensional CAD model of an aluminum alloy die-casting to be stamped, determine the optimal positioning reference surface of the aluminum alloy die-casting to be stamped in a stamping die, and establish a stamping positioning measurement model of the aluminum alloy die-casting to be stamped; based on the high-precision coordinate transformation relationship between the key dimensional parameters of the aluminum alloy die-casting and the stamping positioning measurement model of the aluminum alloy die-casting to be stamped, realize accurate registration between randomly placed aluminum alloy die-castings to be stamped and the optimal positioning state, and accurately extract the three-dimensional spatial position information of the optimal positioning reference surface of the aluminum alloy die-casting to be stamped;

[0200] The third unit is used to input the extracted three-dimensional spatial position information into the robot motion path planning module, automatically generate the optimal motion trajectory and target placement position coordinates for the robot arm to grab the aluminum alloy die-castings to be stamped; control the robot arm to grab the randomly placed aluminum alloy die-castings to be stamped according to the optimal motion trajectory, and accurately place them at the target positioning position of the stamping die, thereby completing the automatic loading and precise positioning of the aluminum alloy stampings to be stamped.

[0201] According to a third aspect of the embodiments of the present invention,

[0202] An electronic device is provided, comprising:

[0203] processor;

[0204] a memory for storing processor-executable instructions;

[0205] The processor is configured to call the instructions stored in the memory to execute the aforementioned method.

[0206] According to a fourth aspect of the embodiments of the present invention,

[0207] A computer-readable storage medium is provided, on which computer program instructions are stored. When the computer program instructions are executed by a processor, the aforementioned method is implemented.

[0208] The present invention may be a method, an apparatus, a system and / or a computer program product. The computer program product may include a computer-readable storage medium carrying computer-readable program instructions for executing various aspects of the present invention.

[0209] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, rather than to limit it. Although the present invention has been described in detail with reference to the aforementioned embodiments, those skilled in the art should understand that they can still modify the technical solutions described in the aforementioned embodiments, or replace some or all of the technical features therein with equivalents. However, these modifications or replacements do not cause the essence of the corresponding technical solutions to deviate from the scope of the technical solutions of the embodiments of the present invention.

Claims

1. Based on the online visual measurement method of car body aluminum alloy die casting production, it is characterized by: include: Acquire the feature information of the 3D CAD model of the aluminum alloy die-casting to be measured, determine the key measurement feature points of the aluminum alloy die-casting, and establish a 3D measurement model of the aluminum alloy die-casting; use a binocular stereo vision camera to shoot the aluminum alloy die-casting in multiple perspectives, with a large depth of field and high resolution to obtain high-density point cloud data on the surface of the aluminum alloy die-casting; input the acquired high-density point cloud data on the surface of the aluminum alloy die-casting into the point cloud data intelligent preprocessing module, automatically optimize the high-density point cloud data on the surface of the aluminum alloy die-casting through downsampling, denoising and smoothing processing, and obtain high-quality streamlined point cloud data on the surface of the aluminum alloy die-casting; The simplified point cloud data is input into the intelligent registration module based on the iterative closest point algorithm. Through coarse registration and fine registration, the high-precision coordinate transformation relationship between the high-quality simplified point cloud data and the three-dimensional measurement model is obtained, and the accurate registration of the high-quality simplified point cloud data and the three-dimensional CAD model is realized; based on the registered point cloud data, the edge extraction algorithm and the regional growth algorithm are used to accurately extract the three-dimensional coordinate information of the key measurement feature points, and the extracted three-dimensional coordinate information is input into the intelligent size calculation module. By automatically selecting the measurement benchmark and using the point-line distance and point-surface distance calculation methods, the key size parameters of the aluminum alloy die-casting are calculated; Acquire feature information of a three-dimensional CAD model of an aluminum alloy die-casting to be stamped, determine the optimal positioning reference surface of the aluminum alloy die-casting to be stamped in a stamping die, and establish a stamping positioning measurement model of the aluminum alloy die-casting to be stamped; based on the high-precision coordinate transformation relationship between the key dimensional parameters of the aluminum alloy die-casting and the stamping positioning measurement model of the aluminum alloy die-casting to be stamped, achieve accurate registration between randomly placed aluminum alloy die-castings to be stamped and the optimal positioning state, and accurately extract the three-dimensional spatial position information of the optimal positioning reference surface of the aluminum alloy die-casting to be stamped; The extracted three-dimensional spatial position information is input into the robot motion path planning module to automatically generate the optimal motion trajectory and target placement position coordinates for the robot arm to grab the aluminum alloy die-castings to be stamped; the robot arm is controlled to grab the randomly placed aluminum alloy die-castings to be stamped according to the optimal motion trajectory, and accurately place them at the target positioning position of the stamping die, completing the automated loading and precise positioning of the aluminum alloy stampings to be stamped.

2. The method according to claim 1, characterized in that The steps of obtaining feature information of a three-dimensional CAD model of an aluminum alloy die casting to be measured, determining key measurement feature points of the aluminum alloy die casting, and establishing a three-dimensional measurement model of the aluminum alloy die casting include: Obtain the three-dimensional CAD model file of the aluminum alloy die-casting to be measured through the product data management system, and import the obtained three-dimensional CAD model file into the computer-aided design software; use the function modules provided by the computer-aided design software to identify the key geometric features of the aluminum alloy die-casting through a feature tree or a hierarchical structure, including a plane, a cylindrical surface, a conical surface, a spherical surface and a free-form surface, and extract the type, position, size, and normal vector of the key geometric features; extract the key dimensional parameters and the corresponding tolerance grade of the aluminum alloy die-casting through dimensioning and geometric tolerance information; extract the material grade and mechanical property parameters of the aluminum alloy die-casting through material attribute information; encapsulate the extracted feature information of the aluminum alloy die-casting according to a preset data format, and the feature information includes the key geometric feature information, key dimensional parameters, and the corresponding tolerance grade, material grade, and mechanical property parameters of the aluminum alloy die-casting to be measured; According to the functional requirements and assembly requirements of the aluminum alloy die casting, combined with the extracted geometric feature information, key geometric features that affect the performance and quality of the die casting are determined, measurement sampling points are arranged on the key geometric features using a measurement tool, and for complex surface features, a discretization method is used to convert the surface into a series of discrete measurement sampling points. For small-sized or thin-walled features, the accessibility and measurability of the measurement sampling points are considered and optimized and adjusted, and key measurement feature point information is determined based on the measurement sampling point information, and the key measurement feature point information is extracted and packaged according to a preset data structure; The feature information and key measurement feature point information of the acquired 3D CAD model of the aluminum alloy die-casting are imported into the measurement software, and the corresponding measuring probe stylus ball is generated according to the coordinate position information of the key measurement feature points. The 3D CAD model of the aluminum alloy die-casting is converted into a triangular face mesh model using the surface meshing tool provided by the measurement software. The generated measuring probe stylus ball is virtually assembled with the triangular face mesh model and the virtual calibration of the measuring probe is completed. The corresponding measurement strategy and measurement path are set according to the type and attribute of the key measurement feature points to generate the 3D measurement model of the aluminum alloy die-casting.

3. The method according to claim 1, characterized in that The high-quality simplified point cloud data is input into the intelligent registration module based on the iterative closest point algorithm. Through coarse registration and fine registration, the high-precision coordinate transformation relationship between the high-quality simplified point cloud data and the 3D measurement model is obtained. The steps to achieve accurate registration of the high-quality simplified point cloud data and the 3D CAD model include: Inputting high-quality simplified point cloud data and a three-dimensional measurement model into an intelligent registration module, wherein the high-quality simplified point cloud data and the measurement model are both represented in the form of a three-dimensional coordinate point set; Preprocess the input high-quality streamlined point cloud data and 3D measurement models, including data format conversion, noise filtering and outlier removal, to improve data quality and registration efficiency; The high-quality streamlined point cloud data and the 3D measurement model are roughly aligned using a principal component analysis-based method. The geometric center points of the two data sets are calculated as the centroid, the covariance matrix is ​​calculated, and the eigenvalue decomposition is performed. The high-quality streamlined point cloud data and the 3D measurement model are rotated along their eigenvector directions to align the main directions of the two data sets, and the rotated high-quality streamlined point cloud data is translated so that its centroid coincides with the centroid of the 3D measurement model. The iterative closest point algorithm is used to perform fine registration on the high-quality streamlined point cloud data and the 3D measurement model after rough registration. Under the current transformation matrix, for each point in the high-quality streamlined point cloud data, the nearest point is found in the 3D measurement model to form corresponding point pairs. The optimal rigid body transformation matrix is ​​calculated using the least squares method to minimize the sum of squared distance errors of all corresponding point pairs. The high-quality streamlined point cloud data is transformed using the solved transformation matrix to obtain new high-quality streamlined point cloud data. It is determined whether the iteration termination condition is met. If not, the updated high-quality streamlined point cloud data is used as the new input to continue the iteration. If it is met, the final transformation matrix is ​​output. The transformation matrix obtained by precise registration is used to transform the high-quality simplified point cloud data into the coordinate system of the 3D measurement model to obtain the registered high-quality simplified point cloud data.

4. The method according to claim 1, characterized in that: Based on the registered point cloud data, the steps of accurately extracting the 3D coordinate information of key measurement feature points by using edge extraction algorithm and region growing algorithm include: Preprocessing the registered point cloud data, including noise filtering and outlier removal, to improve data quality and feature extraction accuracy, wherein the noise filtering adopts one or more methods of statistical filtering, Gaussian filtering or bilateral filtering; The edge extraction algorithm uses an edge extraction algorithm based on local surface fitting to extract edge features in point cloud data. For each point in the point cloud data, a certain number of points are selected in its neighborhood, and the local surface is fitted using the least squares method to obtain the local surface equation at the point, and the partial derivatives of the local surface equation at the point in three coordinate directions are calculated to obtain the gradient vector, and the modulus of the gradient vector is calculated. Points with a gradient modulus greater than a set threshold are marked as potential edge points, and morphological operations and connectivity analysis are performed on the potential edge points to eliminate isolated points and short edges, extract continuous and significant edge features, and obtain an edge point set. The region growing algorithm uses a region growing algorithm based on graph segmentation to extract regional features in point cloud data, regards point cloud data as an undirected graph, defines a similarity measurement function between points, calculates the similarity weight between each pair of points in the point cloud data, constructs an edge weight set of the graph, and uses a graph segmentation algorithm to divide the undirected graph into several non-intersecting subgraphs. The subgraphs obtained by the division are post-processed, small regions are merged, and noise regions are eliminated to obtain the regional segmentation result of the point cloud data; The edge point set obtained by edge extraction is merged with the region segmentation result obtained by region growing, and the points located at the region boundary and belonging to the edge point set are extracted as candidate key measurement feature points. According to the geometric properties of the feature points and the measurement requirements, the most representative and reliable feature point subset is selected from the candidate key measurement feature points as the final key measurement feature points. Evaluate the accuracy and completeness of the extracted key measurement feature points, calculate the position error and omission rate of the key measurement feature points by comparing with the reference data, adjust the algorithm parameters according to the evaluation results, optimize the neighborhood size of local surface fitting and the similarity measurement function of graph segmentation, and improve the extraction quality of key measurement feature points.

5. The method according to claim 1, characterized in that According to the high-precision coordinate transformation relationship between the key dimensional parameters of the aluminum alloy die-casting and the stamping positioning measurement model of the aluminum alloy die-casting to be stamped, the steps of realizing accurate registration between the randomly placed aluminum alloy die-casting to be stamped and the optimal positioning state and accurately extracting the three-dimensional spatial position information of the optimal positioning reference surface of the aluminum alloy die-casting to be stamped include: In the CAD model or theoretical model of the aluminum alloy die casting, the theoretical coordinate values ​​of the key dimensional parameters are determined, the geometric features extracted from the stamping positioning measurement model of the aluminum alloy die casting to be stamped are aligned with the corresponding features in the theoretical model, and a corresponding relationship between the two is established. The optimal coordinate transformation matrix between the stamping positioning measurement model of the aluminum alloy die casting to be stamped and the theoretical model is calculated using the least squares method, including a translation vector and a rotation matrix, and the calculated optimal coordinate transformation matrix is ​​used as a high-precision coordinate transformation relationship between the key dimensional parameters and the stamping positioning measurement model of the aluminum alloy die casting to be stamped; On the stamping equipment, a visual sensor or a laser sensor is used to collect the three-dimensional spatial position information of randomly placed aluminum alloy die-castings to be stamped, and the collected three-dimensional spatial position information is initially aligned with the stamping positioning measurement model of the aluminum alloy die-castings to be stamped to obtain the initial posture of the aluminum alloy die-castings to be stamped, and the initial posture of the aluminum alloy die-castings to be stamped is converted to the posture in the optimal positioning state by using a high-precision coordinate transformation relationship, and the converted posture is accurately aligned, and the posture of the aluminum alloy die-castings to be stamped in the optimal positioning state after the accurate alignment is obtained through an iterative optimization algorithm.

6. The method according to claim 1, characterized in that The extracted three-dimensional spatial position information is input into the robot motion path planning module to automatically generate the optimal motion trajectory and target placement position coordinates for the robot arm to grab the aluminum alloy die-casting to be stamped; the robot arm is controlled to grab the randomly placed aluminum alloy die-casting to be stamped according to the optimal motion trajectory, and accurately place it at the target positioning position of the stamping die, and the steps of completing the automatic loading and precise positioning of the aluminum alloy stamping parts to be stamped include: The posture information of the aluminum alloy die-casting to be stamped in the optimal positioning state obtained after accurate registration is transmitted to the robot motion path planning module. The posture information includes the position coordinates and posture angle of the aluminum alloy die-casting to be stamped in the equipment coordinate system. The posture angle uses Euler angles and quaternions to describe the rotation relationship of the aluminum alloy die-casting to be stamped relative to the equipment coordinate system. The posture information is transmitted to the robot motion path planning module in a standardized data format; In the robot motion path planning module, the kinematic parameters, dynamic parameters and workspace constraints of the robot arm are comprehensively considered, and an optimization algorithm combining kinematics and dynamics is used to solve the optimal motion trajectory of the robot arm grabbing the die casting under the premise of satisfying the constraints; According to the design parameters of the stamping die and the optimal positioning state of the aluminum alloy die-casting to be stamped, the optimal positioning state posture information of the die-casting is combined with the design parameters of the stamping die, and the target placement position coordinates of the aluminum alloy die-casting to be stamped in the stamping die coordinate system are calculated through coordinate transformation and space mapping. The target placement position coordinates are aligned with the positioning reference surface and key features of the stamping die; The optimal motion trajectory and target placement position coordinates are transmitted to the robot control system. The robot control system realizes precise motion tracking and trajectory optimization of the robot arm through the motion control algorithm according to the received motion trajectory and target position coordinates, realizes precise contact and smooth operation between the robot arm and the aluminum alloy die-casting to be stamped through the end effector control algorithm, and controls the robot arm to perform grasping and placement operations; Through the visual servo system and force sensing system of the robot arm, the visual sensor installed on the robot arm collects the position information of the aluminum alloy die-castings to be stamped and the stamping mold in real time. The force / torque sensor installed on the robot arm collects the contact force and torque information between the arm and the die-castings and the mold in real time. The visual servo and force feedback control instructions are combined with the motion control instructions. Through closed-loop control, real-time error compensation and trajectory fine-tuning are achieved during the robot arm's grasping and placement process, thereby improving positioning accuracy and operational stability.

7. An online visual measurement system for the production of aluminum alloy die castings for vehicle bodies, used to implement the method described in any one of claims 1 to 6, characterized in that: include: The first unit is used to obtain the feature information of the three-dimensional CAD model of the aluminum alloy die-casting to be measured, determine the key measurement feature points of the aluminum alloy die-casting, and establish a three-dimensional measurement model of the aluminum alloy die-casting; use a binocular stereo vision camera to shoot the aluminum alloy die-casting in multiple perspectives, with a large depth of field and high resolution to obtain high-density point cloud data on the surface of the aluminum alloy die-casting; input the obtained high-density point cloud data on the surface of the aluminum alloy die-casting into the point cloud data intelligent preprocessing module, and automatically optimize the high-density point cloud data on the surface of the aluminum alloy die-casting through downsampling, denoising and smoothing processing to obtain high-quality streamlined point cloud data on the surface of the aluminum alloy die-casting ; Input high-quality streamlined point cloud data into the intelligent registration module based on the iterative closest point algorithm, and obtain the high-precision coordinate transformation relationship between the high-quality streamlined point cloud data and the three-dimensional measurement model through coarse and fine registration, so as to achieve accurate registration of high-quality streamlined point cloud data with the three-dimensional CAD model; Based on the registered point cloud data, the edge extraction algorithm and the regional growth algorithm are used to accurately extract the three-dimensional coordinate information of the key measurement feature points, and the extracted three-dimensional coordinate information is input into the intelligent size calculation module, and the key size parameters of the aluminum alloy die-casting are calculated by automatically selecting the measurement benchmark and using the point-line distance and point-surface distance calculation methods; The second unit is used to obtain feature information of a three-dimensional CAD model of an aluminum alloy die-casting to be stamped, determine the optimal positioning reference surface of the aluminum alloy die-casting to be stamped in a stamping die, and establish a stamping positioning measurement model of the aluminum alloy die-casting to be stamped; based on the high-precision coordinate transformation relationship between the key dimensional parameters of the aluminum alloy die-casting and the stamping positioning measurement model of the aluminum alloy die-casting to be stamped, realize accurate registration between randomly placed aluminum alloy die-castings to be stamped and the optimal positioning state, and accurately extract the three-dimensional spatial position information of the optimal positioning reference surface of the aluminum alloy die-casting to be stamped; The third unit is used to input the extracted three-dimensional spatial position information into the robot motion path planning module, automatically generate the optimal motion trajectory and target placement position coordinates for the robot arm to grab the aluminum alloy die-castings to be stamped; control the robot arm to grab the randomly placed aluminum alloy die-castings to be stamped according to the optimal motion trajectory, and accurately place them at the target positioning position of the stamping die, thereby completing the automatic loading and precise positioning of the aluminum alloy stampings to be stamped.

8. An electronic device, characterized in that: include: processor; a memory for storing processor-executable instructions; The processor is configured to call the instructions stored in the memory to execute the method according to any one of claims 1 to 6.

9. A computer-readable storage medium having computer program instructions stored thereon, characterized in that: When the computer program instructions are executed by a processor, the method according to any one of claims 1 to 6 is implemented.

Images