Method and system for three-dimensional modeling and collaborative labeling of existing equipment

Abstract

The application discloses a kind of existing equipment three-dimensional modeling and collaborative labeling method and system, it is related to computer aided design field.The method acquires the three-dimensional point cloud of existing equipment, two-dimensional engineering drawing element and the size data obtained from field measurement, three belong to independent acquisition coordinate system;With rigid body transformation, the multi-source heterogeneous data is unified to the same model coordinate system and extracts geometric features, constructs parameterized feature set;To the same size parameter, the inverse variance weight is used to weight and fuse multiple source observation values, to drive three-dimensional computer aided design system to generate parameterized three-dimensional model;Process route, tolerance and detection point information are collaboratively labeled on the model, and the cumulative transfer quantity of labeled tolerance is calculated according to size sensitivity chain;The model and its manufacturing information are output to manufacturing execution end, and the size deviation fed back by manufacturing execution end is used for re-fusion.The standard deviation of reconstructed size is lower than any single source.

Description

Technical Field

[0001] This invention relates to the field of computer-aided design and digital manufacturing technology, specifically to a method and system for 3D modeling and collaborative annotation of existing equipment. Background Technology

[0002] Among the equipment assets of various manufacturing and processing enterprises, existing equipment often undergoes multiple repairs, partial modifications, and component replacements. Original design drawings are often missing, outdated, or inconsistent with the actual condition, and interface constraints and critical dimensions lack standardized records. When upgrading or modifying such equipment, designers need to recreate editable 3D models and supplement manufacturing, process, and testing information. Traditional methods of drawing reconstruction involve manually measuring key dimensions with a rangefinder and calipers, then sketching and reconstructing in modeling software. This process is time-consuming, labor-intensive, and makes it impossible to obtain dimensions for areas inaccessible to humans, such as those at heights or deep cavities.

[0003] To replace manual measurement, 3D laser scanning has been widely adopted in this field, and the point cloud obtained from the scan is used in conjunction with existing drawings or design models to reconstruct the model. CN113051652A discloses an engineering renovation and expansion method based on BIM and 3D scanning reverse modeling technology. It initially establishes a design model based on historical building structure drawings and converts it into a design BIM model. A 3D laser scanner is used to scan the internal solid targets of the building structure to obtain a point cloud model. After stitching, denoising, and coloring, a complete point cloud model is generated and lightweighted. Then, a reverse BIM model is created using reverse modeling software and compared with the design BIM model. The system automatically analyzes unqualified points and deviations and dynamically calibrates the design BIM model. Finally, the renovation design is based on the calibrated design BIM model, and blueprints are exported. This approach uses the design model as the skeleton and calibrates the skeleton using point cloud deviations, transforming the drawing reconstruction problem into a model comparison problem. It has been widely adopted in building and structural renovation and expansion scenarios.

[0004] CN116051724A discloses a parameterized integrated construction method for component models based on point clouds. This method imports the component's design drawings, performs structural analysis on the pre-modeled component to determine its shape regularity and selects the scanning method accordingly, uses 3D laser scanning technology to obtain an initial point cloud model and generate a new point cloud model, aligns the component's design drawings and point cloud model, uses the design drawings as a framework to denoise the point cloud model to obtain a new point cloud model, converts the new point cloud model into a TXT file and extracts the X, Y, and Z information to establish the geometric model of the pre-modeled component, then extracts the RGB information and stores it in a dataset, creates a neural network based on deep learning to identify materials and assign attributes to the pre-modeled component, completing the component information modeling. The alignment of the drawings and point clouds is performed in two steps: first, principal component analysis is used to align the centers and principal axes of both using the feature values ​​of key points; then, the boundary curves of both are further aligned using an iterative nearest-point algorithm. During denoising, a smaller penalty factor is applied to point clouds within the constraints of the design drawings, and a larger penalty factor is applied to point clouds outside the constraints, thereby distinguishing between noise-free and noisy point clouds.

[0005] The aforementioned approach, which uses point clouds as the source of dimensions and drawings as geometric constraints, ultimately determines the model dimensions point-by-point from the coordinates of the denoised point cloud. The drawings serve as the constraint boundary during denoising, rather than an independent source of dimensional observation. Therefore, the reliability of the dimensional values ​​depends on the measurement uncertainty of the point cloud source itself. When the point cloud is sparse or missing in areas such as obstructed surfaces, deep hole walls, or reflective surfaces, the dimensions at those locations lack a second source for cross-checking and can only be filled by extrapolation from neighboring point clouds. This extrapolation error then enters the model and cannot be evaluated. Furthermore, the attribute information assigned to components is material information such as material type. The model's geometric elements do not carry manufacturing and inspection information such as process routes, tolerances, and inspection points. There is no quantitative basis for how the tolerances of derived features accumulate from the fluctuations of the individual component dimensions. Consequently, there is a lack of quantifiable criteria for assessing the assembly interchangeability and inspection consistency of modified parts. Furthermore, the process ends at component information modeling; the resulting geometric model and attribute information are not sent to CNC machining equipment, manufacturing execution systems, or on-site inspection terminals. Actual dimensions generated during machining and inspection are not fed back for model correction. Once the model is delivered, it is disconnected from subsequent manufacturing processes. How to obtain an existing equipment model with reliable dimensions, complete manufacturing information, and convergence with manufacturing feedback, under conditions of incomplete drawings and inconsistent accuracy from various sources, remains a practical problem that needs to be solved in the digital transformation of equipment. Summary of the Invention

[0006] The technical problem to be solved by this invention is: how to reconstruct a three-dimensional model with reliable dimensions that can be directly used for manufacturing and inspection, given that the original drawings of existing equipment are incomplete, the sources of reference information are scattered and the accuracy of each source is different.

[0007] To address its technical problems, this invention provides a method for 3D modeling and collaborative annotation of existing equipment. In an equipment modeling system with multi-source data perception, parametric reconstruction, and manufacturing closed-loop output capabilities, the method includes the following steps: collecting multi-source heterogeneous data of the existing equipment; acquiring surface point clouds of the existing equipment through a 3D scanning device; acquiring exterior images of the existing equipment through an image acquisition device and identifying 2D engineering drawing elements from them; and receiving dimensional data obtained from on-site measurements of key parts of the existing equipment directly measured by measuring instruments. The multi-source heterogeneous data includes at least 3D point cloud data, 2D engineering drawing data, and on-site measured dimensional data, with the three types of data belonging to mutually independent first, second, and third acquisition coordinate systems, respectively. Heterogeneous source data are uniformly transformed to the same model coordinate system to obtain a registered unified feature dataset. Geometric features are extracted from the unified feature dataset and a parametric feature set is constructed. Based on the observations of various data sources for the same dimensional parameter in the multi-source heterogeneous data, a complete dimensional parameter vector is obtained through weighted fusion and used to drive the 3D computer-aided design system to generate a parametric 3D model with design variables. Manufacturing information is collaboratively labeled on the parametric 3D model. The manufacturing information includes at least process route information, tolerance information, and inspection point information. The cumulative transfer of the labeled tolerance on the feature chain is calculated based on the dimensional sensitivity chain. The labeled parametric 3D model and its manufacturing information are output to the manufacturing execution end, forming a closed loop from data acquisition to manufacturing execution.

[0008] In the above method, in the step of uniformly transforming multi-source heterogeneous data to the same model coordinate system, a rigid body transformation is performed to align the feature point set belonging to the first acquisition coordinate system with the feature point set belonging to the model coordinate system. The rotation matrix and translation vector of the rigid body transformation are obtained by performing singular value decomposition on the cross covariance matrix of the two feature point sets, and the translation vector is determined by the difference between the centroids of the two feature point sets.

[0009] In the above method, during the step of obtaining the complete size parameter vector through weighted fusion, the first... Each dimension parameter is weighted by the inverse variance of the square of the standard deviation of the observations. Observations from no fewer than two data sources are weighted and fused to obtain the fused dimension parameters and their fused standard deviation. The fused dimension parameter vector is then used to drive the 3D computer-aided design system to reconstruct the parametric 3D model.

[0010] In the above method, in the step of collaboratively annotating manufacturing information on a parametric 3D model, for a derived feature determined by a function based on several dimensional parameters, the cumulative tolerance of the derived feature is calculated based on the first-order sensitivity of the function to each component dimension. The sum of the products of the annotation tolerance of each component dimension and the absolute value of its sensitivity is taken as the cumulative tolerance in the worst case. The cumulative tolerance is written into the manufacturing information as the annotation tolerance of the derived feature.

[0011] The above method, in the step of outputting the annotated parametric 3D model and its manufacturing information to the manufacturing execution terminal, outputs the model and manufacturing information to at least one of CNC machining equipment, manufacturing execution system and on-site inspection terminal, and transmits the dimensional deviation fed back by the manufacturing execution terminal as a new observation value to participate in the weighted fusion to trigger model parameter correction, thereby forming a closed loop.

[0012] To address its technical problem, this invention provides a 3D modeling and collaborative annotation system for existing equipment, used to execute the steps of any of the methods described above, comprising: a data acquisition unit, whose data output is connected to the input of a coordinate registration unit, for acquiring multi-source heterogeneous data of the existing equipment, wherein the multi-source heterogeneous data includes at least 3D point cloud data, 2D engineering drawing data, and dimensional data obtained from on-site measurements, and the three types of data belong to mutually independent first acquisition coordinate systems, second acquisition coordinate systems, and third acquisition coordinate systems, respectively; a coordinate registration unit, for uniformly transforming the multi-source heterogeneous data to the same model coordinate system to obtain a registered unified feature dataset and sending it to a parametric reconstruction unit; and a parametric reconstruction unit, for extracting geometric features from the unified feature dataset and constructing a parametric feature set based on the multi-source heterogeneous data for the same size. The observations from various data sources of the dimensional parameters are weighted and fused to obtain a complete dimensional parameter vector, which drives the 3D computer-aided design system to generate a parametric 3D model with design variables. The manufacturing information annotation unit, whose input end is connected to the output end of the parametric reconstruction unit, is used to collaboratively annotate manufacturing information on the parametric 3D model. The manufacturing information includes at least process route information, tolerance information, and inspection point information, and calculates the cumulative transfer amount of the annotation tolerance on the feature chain based on the dimensional sensitivity chain. The manufacturing end output unit, whose input end is connected to the output end of the manufacturing information annotation unit and whose output end is connected to the manufacturing execution end, is used to output the annotated parametric 3D model and its manufacturing information to the manufacturing execution end, and to transmit the dimensional deviation fed back by the manufacturing execution end back to the parametric reconstruction unit, forming a closed loop from data acquisition to manufacturing execution.

[0013] In the aforementioned system, the coordinate registration unit registers the feature point set belonging to the first acquisition coordinate system. and the feature point set belonging to the model coordinate system Perform rigid body transformation alignment; the points in this feature point set, after transformation, satisfy... , where the rotation matrix With translation vector The cross covariance matrix of two feature point sets Obtained by singular value decomposition. and Let be the centroids of the two feature point sets respectively, and let the translation vector satisfy . .

[0014] In the above system, the parameterized reconstruction unit is for the first... Each size parameter is used to obtain observations from at least two data sources. and its uncertainty Weighted by inverse variance The observations are weighted and fused to obtain the fused size parameters. and its fusion standard deviation The fused dimension parameter vector is used to drive the 3D computer-aided design system to reconstruct the parametric 3D model.

[0015] In the aforementioned system, the manufacturing information annotation unit is configured based on several dimensional parameters. via function Given a defined derived feature, the cumulative tolerance of that derived feature is calculated based on the first-order sensitivity of the function to each component dimension. ,in For the first The tolerances for each dimensional parameter are specified, and the cumulative tolerances are calculated. The tolerance specified as a derived feature is written into the manufacturing information.

[0016] The beneficial effects of this invention are:

[0017] (1) Multi-source observations of the same size parameter are weighted by inverse variance and the fused size standard deviation is lower than any single source standard deviation. In the local missing parts of the point cloud, the drawing elements and the measurement values ​​of the measuring tools can still complement each other, so that the reconstructed size is still reliable under the condition of incomplete drawing data.

[0018] (2) Coordinate registration and weighted fusion are connected before and after. Registration unifies the observations belonging to three independent acquisition coordinate systems into the same model coordinate system, making multi-source observations of the same size comparable. Fusion compresses the comparable multi-source observations into a single optimal estimate. The coupling of the two makes the reconstructed model have both the properties of benchmark unification and size optimization.

[0019] (3) Manufacturing, process and inspection information are collaboratively bound on the geometric elements of the parametric model. The annotation tolerance of the derived features is obtained by accumulating the tolerance of the component dimensions according to the sensitivity chain, so that the annotation tolerance covers the fluctuation of all component dimensions on the feature chain, and the assembly, interchangeability and inspection of the modified parts have a verifiable basis.

[0020] (4) The model and manufacturing information are sent to the manufacturing execution end. The actual dimensions obtained from processing and inspection are returned as new observations and participate in the re-fusion. The model converges and locks the version with the manufacturing feedback. The subsequent modification of similar equipment in the same enterprise can quickly obtain a special model with a small number of supplementary observations on the converged model examples. Detailed Implementation

[0021] This invention is implemented in an equipment modeling system with multi-source data sensing, parametric reconstruction, and manufacturing closed-loop output functions. The system is functionally divided into five parts: a data acquisition unit, a coordinate registration unit, a parametric reconstruction unit, a manufacturing information annotation unit, and a manufacturing end output unit. The data acquisition unit acquires and preprocesses multi-source heterogeneous data and sends the results to the coordinate registration unit. The coordinate registration unit, after establishing a coordinate system, sends a unified feature dataset to the parametric reconstruction unit. The parametric reconstruction unit extracts geometric features and solves for dimensional parameters, driving a 3D computer-aided design system to generate a parametric 3D model. The manufacturing information annotation unit writes three types of information—process route, tolerance, and inspection points—onto the model. The manufacturing end output unit sends the model and manufacturing information to the manufacturing execution end and retrieves the actual dimensions, sending them back to the parametric reconstruction unit. These five parts are sequentially connected to form a closed loop from data acquisition to manufacturing execution. The specific implementation process of this embodiment is described below in this order.

[0022] The data acquisition unit acquires multi-source heterogeneous data from existing equipment through physical acquisition devices. A 3D scanning device performs a non-contact scan of the existing equipment surface, outputting a 3D point cloud reflecting its true shape; an image acquisition device acquires images of the equipment's appearance from several perspectives, identifying graphic elements and key annotations from early 2D engineering drawings; simultaneously, engineering technicians directly measure key parts of the equipment using calipers, micrometers, or coordinate measuring machines, obtaining dimensional data from on-site measurements. These three types of data belong to independent acquisition coordinate systems: the point cloud is in the scanner coordinate system (first acquisition coordinate system); the drawing elements obtained from image recognition are in the image coordinate system (second acquisition coordinate system); and the dimensions obtained from measuring instruments are in the world coordinate system of the measuring instruments (third acquisition coordinate system). The acquired point cloud undergoes denoising and downsampling preprocessing before execution to remove outlier noise points and reduce the computational scale of subsequent registration; the 2D drawing elements are extracted into geometric primitives that can participate in registration through image binarization and line vectorization; and the dimensions obtained from on-site measurements are directly entered into the parametric reconstruction stage in scalar form. Before acquisition, the image acquisition equipment undergoes checkerboard calibration to determine its internal parameters and lens distortion coefficients. The perspective image is then back-projected onto a normalized plane to eliminate wide-angle distortion. Edge detection and non-maximum suppression are then applied to the back-projected lines to obtain single-pixel lines. Subsequently, an accumulator voting process is used to extract line segment endpoints for straight line elements, and circle and arc elements are estimated to extract center coordinates. This transforms early 2D engineering drawings into a set of geometric primitives that can be registered with point clouds. For dimensions only marked on the drawings but not corresponding to existing structures, the parametric reconstruction unit marks them as items to be measured. These are then added to the fusion process after on-site measurement or secondary scanning, avoiding the introduction of systematic biases from direct determination of values ​​from a single source (drawings).

[0023] To achieve coordinate unification of multi-source heterogeneous data, the coordinate registration unit aligns feature points from different acquisition coordinate systems to the same model coordinate system. Let the set of feature points belonging to the first acquisition coordinate system be the invention. The feature point set belonging to the model coordinate system is the invention Points in the two sets are paired one-to-one according to geometric correspondence. Coordinate registration is based on the fundamental definition of rigid body motion in three-dimensional space: rigid body transformation preserves the Euclidean distance between any two points, and for any feature point in the first acquisition coordinate system... The transformed coordinates of the invention Invention satisfies In the formula, Invention is an invention Invent a rotation matrix, The invention is a translation vector. The invention represents the determinant of the invention. The invention utilizes an orthogonal rotation group. The registration criterion is to make the source point set fit the target point set as closely as possible after transformation, taking the error function... In the formula, The invention is the number of corresponding point pairs. The invention represents the Euclidean norm.

[0024] To eliminate the translation, we introduce the centroids of the two sets. In the formula, Inventions and Inventions The inventions are the centroids of two feature point sets, and the formula is given by the definition of the arithmetic mean.

[0025] Invention , Substitute into equation (2) and utilize the invention The error function can be decomposed into In the formula, Inventions and Inventions The invention refers to the points obtained after removing the centroids from two feature point sets. The invention is the number of corresponding point pairs. Its source is the expansion of the sum of squares after substituting equation (3) into equation (2).

[0026] The second term on the right-hand side of the equation is independent of the coordinates inside the source point set if and only if the invention Since the value was zero at the time of invention, the translation is uniquely determined by the difference between the two centroids: In the formula, The invention is a translation vector. Its source is the term after taking the second term on the right side of equation (4) and shifting it to zero.

[0027] Substituting equation (5) back, the registration problem is reduced to a minimization problem involving only rotations. Substituting equation (5) back into equation (2) and eliminating the translation term yields equation (6). In the equation... , Expanding the squared terms yields In the formula, The invention represents the Euclidean norm. From the expansion of the square of the vector norm, we obtain equation (7), and the resulting cross term is the invention. .

[0028] The first two items are related to invention. Invention is irrelevant, therefore minimize invention. Invention is equivalent to maximizing invention Using the cyclic property of the matrix trace, it can be written as: In the formula, The invention represents the trace of a matrix. Equation (8) can be obtained from the cyclic property of the trace of a matrix.

[0029] Equation (9) can be obtained by summing the outer products of the two centroid-free point sets. In the equation, The invention is the cross-covariance matrix of two feature point sets. Inventions and Inventions The invention is defined by equation (6), which is derived from the sum of the outer products of two centroid-free sets. Regarding the invention... Invention for Singular Value Decomposition and construct In the formula, Inventions and Inventions Invention is an invention Inventing orthogonal factors obtained from singular value decomposition. The invention is a singular value diagonal matrix. The invention represents a determinant.

[0030] Among them, the invention of taking the last digit of the diagonal correction term To ensure the invention Illegal solutions involving mirror flips are eliminated. This leads to the invention of rotation matrices. Invention and Translation Vector Invention The invention is fully finalized; multi-source heterogeneous data is unified into the model coordinate system, resulting in a registered unified feature dataset. The invention will be implemented in practice. The invention first calculates the centroid according to formula (3) and the cross covariance matrix according to formula (9) for stable corresponding feature points, such as the center of the shaft end, the intersection of the box edge, and the center of the mounting hole. Then, it obtains the rigid body transformation parameters through singular value decomposition, and unifies the scanned point cloud and the drawing elements to the same reference.

[0031] Not all corresponding point pairs are reliable. Before obtaining the exact rigid body transformation, principal component analysis is used to obtain the principal axis directions of the two feature point sets respectively. The source point set is rotated around the centroid until it is roughly aligned with the principal axis of the target point set as the initial value for coarse registration, thereby narrowing the convergence domain of the subsequent solution and improving the initial accuracy of the corresponding point pairs. The rotation obtained from the coarse registration is substituted into the initial value for solving Equation (10). The parameterized reconstruction unit calculates the distance residual of each point pair after each round of solution. Any residual greater than the threshold is considered as a result. The inventors identified these as incorrect correspondences and discarded them. The process was repeated until the residual variation in two consecutive rounds fell below a set tolerance. The resulting rotation matrix and translation vector were then used as the final registration parameters. (Threshold invention) The invention takes three times the standard deviation of the residual distribution of repeated registration of the same type of equipment, which is obtained by statistical analysis of the repeated registration records of the batch of equipment. The tolerance is set to half of the fusion uncertainty, and its value varies with the equipment specifications. It must be determined for specific equipment according to the above method. This is only used to illustrate the process of determining the threshold and is not a limitation of the invention.

[0032] After obtaining the registered unified feature dataset, the parametric reconstruction unit extracts geometric features and constructs a parametric feature set. Geometric feature extraction is based on identifying basic geometric primitives from point clouds and drawing elements: cylindrical features are defined by axial direction and radius to determine shaft segment dimensions; planar features are defined by normal direction and boundary to determine the mounting panel; circular hole features are defined by center and diameter to determine the mounting hole position; and transition fillets are defined by radius of curvature to determine the connection form. Each geometric primitive is converted into a set of design variables, including shaft segment length, hole diameter, wall thickness, mounting center distance, and transition fillet radius, forming the parametric feature set. A bidirectional relationship is established between the design variables and the sketch dimensions and feature parameters in the model: modifying the design variable values ​​drives sketch and feature reconstruction; conversely, dragging dimensions directly on the model will write back the design variables, thus ensuring the editability of the parametric model and its subsequent serial reuse. For adjacent components with assembly and mating relationships, the parametric reconstruction unit establishes geometric constraints between its parametric feature sets, including coaxiality, equidistance, and symmetry. The constraints participate in the subsequent fusion solution in the form of equality or inequality, so that the multi-component models maintain consistency under a unified benchmark and avoid the contradiction of reasonable single component dimensions but assembly interference.

[0033] Since the same dimension is often observed from different angles by multiple data sources, and the observations from each data source are obtained from the aforementioned feature extraction of the multi-source heterogeneous data, and each source carries different measurement uncertainties, the parameterized reconstruction unit uses weighted fusion to solve for the optimal estimate of each dimension parameter. The weighted fusion is based on the Gauss-Markov theorem: when observations are independent and the variances of each observation are known, the inverse variance weighted estimator is the optimal linear unbiased estimator in the sense of mean square error. For the first invention... Invent a dimension parameter, let its first invention Invent a data source observation value Invent Uncertainty is expressed as standard deviation in the invention. The weights are defined as inverse variance. In the formula, The invention provides the number of data sources that provide observations for this size parameter.

[0034] The fusion estimate is the weighted average of the observations. In the formula, The invention is the first invention The fused estimate of the size parameter is invented. Its variance is calculated by the variance propagation law of independent observations, and the fused standard deviation is obtained by substituting it into equation (11). In the formula, The invention is the standard deviation of the fusion estimator.

[0035] As can be seen from equation (13), Inventions less than any single-source standard deviation In other words, multi-source fusion reduces dimensional uncertainty compared to any single source. The following values ​​are illustrative and are only used to illustrate the calculation steps of weighted fusion; they are not actual measured values: Assume the three source observations of the support shaft diameter are taken as illustrative values. (Uncertainty Invention) ), ( ), ( The three sources of weight are invention, invention, and invention. , , The sum of the weights is the invention From equation (12), the fusion value is obtained. From equation (13), we obtain the fusion standard deviation invention. Lower than the most reliable single-source invention For the axial length of this part, let its schematic observation value be the point cloud source invention. (Uncertainty Invention) ), supplementary source invention ( If the weight is invention, then the weight is invention. Inventions and Inventions The sum of weights invention Fusion value invention Invention of fused standard deviation Similarly, it is lower than the most reliable single-source invention. The complete dimensional parameter vector is constructed by fusing all dimensional parameters. Based on this, the parametric reconstruction unit calls the application programming interface of the 3D computer-aided design system to create sketches, features, and assembly constraints according to design variables, generating a parametric 3D model with editable design variables.

[0036] After modeling is completed, the parameterized reconstruction unit uses the reconstruction residuals as a closed-loop validation metric. This metric is derived from the sum of squared residuals of the weighted least squares estimation: when the observation noise is independent and follows a zero-mean Gaussian distribution, the standardized sum of squared residuals weighted by inverse variance follows a chi-square distribution, thus its numerical value can be used to measure the overall agreement between the model and the observations. Let the invention... Invention of size parameters, first invention The noise standard deviation of the invention's observations is the invention The standardized residual of this observation is In the formula, The invention is the first invention The standardized residuals of the observed values ​​are invented. Taking the weights as shown in equation (11), the weighted sum of squared residuals can be written as: In the formula, The invention is a weighted sum of squared residuals.

[0037] Invention Invent the sum of squares of mutually independent standard normal variables, which, according to the definition of a chi-square distribution, follow a distribution with approximately [number] degrees of freedom. The invented chi-square distribution, The invention takes the difference between the total number of observations and the number of dimensional parameters to be estimated. Taking the aforementioned support shaft diameter as an example, the number of observations is three, the number of parameters to be estimated is one, and the degrees of freedom are two. If the invention... If the numerical value of the invention significantly exceeds the conventional quantile of the chi-square distribution under that degree of freedom, it indicates that there is a deviation between the model and the observation that is not explained by uncertainty. Based on this, the parameterized reconstruction unit prompts a review of the geometric primitive identification results and corresponding relationships of the feature.

[0038] After the parametric 3D model is generated, the manufacturing information annotation unit collaboratively annotates three types of manufacturing information on the model's geometric elements: process route information describes the machining or assembly sequence of the component, tolerance information describes the allowable deviations of each dimension, and inspection point information describes the measurement locations and datums during quality inspection. The process route information is automatically arranged based on feature type and material; for example, the outer circle of a shaft segment progresses sequentially from rough turning, semi-finish turning to finish turning, mounting holes are formed from drilling to reaming, and assembly processes are arranged according to the order of positioning on the datum surface followed by tightening. Inspection point information is selected based on tolerance-sensitive features, setting up measurement points and specifying measurement datums for mating dimensions and areas related to geometric tolerances. The selection of inspection points follows the principle of prioritizing sensitivity while considering measurability: measurement points are prioritized for geometric parts corresponding to component dimensions with large absolute sensitivity values, as their small fluctuations, amplified by the feature chain, significantly affect derived features; for parts related to form and position tolerances, such as the reference surfaces for coaxiality and parallelism, separate measurement references and sampling directions are specified; for dimensions obtained only from a single source with high fusion uncertainty, additional re-measurement points are added to form cross-validation at the manufacturing execution end. The sampling ratio of process route information is correlated with the tolerance level of the inspection procedures, with stricter tolerance levels correspondingly increasing the frequency of first-piece inspection and in-process sampling. The above three types of information—process, tolerance, and inspection points—are bound together in a relational relationship on the model's geometric elements. When the model parameters for any dimension are modified, the associated inspection points and process procedures are automatically updated synchronously.

[0039] Determining tolerance information requires quantifying its cumulative transmission along the feature chain, based on the principle of linear cumulative dimensional tolerance analysis, i.e., the first-order sensitivity method corresponding to the linear tolerance superposition rule in geometric product specifications. Suppose a derived feature dimension is invented based on a function of the constituent dimensions. Invention confirmed, The invention is a component size vector, in nominal value invention Perform a first-order Taylor expansion near the invention: In the formula, The invention is the deviation of the derived feature size from its nominal value. The invention is the first invention The invention relates to the deviation of the component dimensions from their nominal values. The invention relates to the number of components in the dimension.

[0040] Composition Dimension Invention The invention fluctuates within its tolerance zone, where the tolerance is the invention. Invention Derivative feature tolerance is for invention To satisfy the invention Taking the worst-case scenario using the triangle inequality: From the triangle inequality, we can obtain equation (17). The derivation process involves taking the absolute value of each term on the right-hand side of equation (16) and applying it term by term to the invention. The invention is enlarged. In the formula, the meanings of each symbol are the same as in formula (16), and the upper bound on the right is obtained by the triangle inequality.

[0041] Multiplying both sides by two yields the worst-case cumulative tolerance of the derived feature. The derivation process involves multiplying the rightmost term of equation (17) by two and rearranging the terms. In the equation, The worst-case cumulative tolerance for the invention as a derived feature. The invention is the first invention Invent a specification tolerance for the dimensions of each component.

[0042] If the errors of each component dimension are independent of each other, and each tolerance zone is invented at the same magnification... The invention covers its standard deviation, i.e., the invention Then, according to the variance propagation law of independent observations, the standard deviation of the derived features satisfies In the formula, The invention is the standard deviation of the derived feature size.

[0043] Square root on both sides and invention Invention Substitution, and Multiplication Invention That is, the cumulative tolerance of the root mean square method is obtained. In the formula, the meaning of each symbol is the same as in formula (18), which is derived from the variance propagation law of independent observation.

[0044] For the derived feature of the center distance between two shaft segments, its function is that the center distance is equal to the sum of the lengths of the two shaft segments, and the sensitivity is the same as that of the invention. If the length tolerances of both shaft segments are the invention Then, from equation (18), we can obtain the worst-case cumulative tolerance invention. From equation (20), the root mean square cumulative tolerance is obtained. For the coaxiality-related derived feature of the difference in the depth of the two mounting holes, its function is that the depth difference equals the difference in the depths of the two holes, and the sensitivities are respectively the invention Inventions and Inventions If the depth tolerance of both holes is the same as the invention Then the worst-case cumulative tolerance invention Root mean square cumulative tolerance invention The manufacturing information annotation unit calculates the annotation tolerance of each derived feature according to the worst-case formula (18) and writes it into the manufacturing information, so that the annotation tolerance covers the fluctuation of all component dimensions on the feature chain, thereby ensuring the assembly interchangeability and inspection consistency of the modified parts.

[0045] The manufacturing output unit outputs the annotated parametric 3D model and its manufacturing information to the manufacturing execution unit, forming a closed loop from data acquisition to manufacturing execution. Output targets include at least one of CNC machining equipment, the manufacturing execution system, and a field inspection terminal: the CNC machining equipment generates machining programs based on the model and process route information and executes material cutting and machining; the manufacturing execution system schedules inspection processes and generates inspection work orders based on tolerance and inspection point information; and the field inspection terminal presents the modified parts, inspection benchmarks, and annotated tolerances to maintenance personnel. Actual dimensional deviations generated during machining and inspection at the manufacturing execution unit are fed back to the parametric reconstruction unit as new observations in the aforementioned weighted fusion, triggering model parameter correction and causing the parametric 3D model to continuously converge with manufacturing feedback. If the returned deviation exceeds three times the fusion uncertainty, the manufacturing output unit determines it as an abnormal condition, suspends the automatic correction of this feature, and prompts manual verification. After verification, the dimensional measurement value is incorporated into the fusion as a high-weight observation to avoid single-point anomalies contaminating the overall parameter estimation. The on-site inspection terminal uses augmented reality to overlay the modified parts, inspection benchmarks, and marked tolerances onto the physical equipment. Maintenance personnel complete the measurements according to the presented benchmarks and send back the obtained dimensional values. Deviation recovery is triggered at a fixed pace. After each batch of modifications is completed, all the measured dimensions of that batch are sent back. The parametric reconstruction unit refreshes the fusion estimate and releases a new version of the model based on this.

[0046] As feedback from the manufacturing execution end is continuously incorporated, the dimensional parameters of the parametric 3D model gradually converge. The parametric reconstruction unit uses the difference between two consecutive fusion estimates being less than the fusion uncertainty as the convergence criterion. After convergence, the model version is locked and the source weight composition of each dimensional parameter is recorded for easy traceability of subsequent changes. When multiple existing similar equipment in the same enterprise undergo upgrades, the previously converged models can be used as parametric model examples. New equipment only needs to supplement a small amount of measurement data for its different parts to quickly obtain a unique model for that equipment based on the existing model examples through weighted fusion, without having to reconstruct all features from scratch, thus significantly reducing the design workload for a single upgrade. Model version management and model example reuse are driven by the same set of coordinate registration, weighted fusion, and tolerance transfer logic, ensuring that the results of upgrades across equipment and batches remain consistent in geometric benchmarks and annotation standards.

[0047] Once the model version is locked, the manufacturing output unit exports it in a data exchange format. The exported model carries three types of related information: process route, tolerances, and inspection points. Downstream CNC machining equipment uses this information to generate machining programs, the manufacturing execution system uses it to generate inspection work orders, and the on-site inspection terminal uses it to generate augmented reality operation guidelines. All three types of outputs share the same geometric datum and the same set of annotation tolerances, eliminating information inconsistencies at the source. During model sample reuse, the design variable names, tolerance levels, and inspection point layouts of existing model samples are used as initial assumptions and introduced into the new equipment. The measurement data obtained from the new equipment participates in the fusion with high weight, enabling the model sample to quickly converge into a model specific to this equipment. If the new equipment has structural features not covered by the model sample, corresponding design variables are dynamically added to the parametric feature set and independently fused, thus expanding the model sample and realizing the continuous accumulation of transformation knowledge and cross-project reuse.

[0048] The data acquisition unit, coordinate registration unit, parametric reconstruction unit, manufacturing information annotation unit, and manufacturing output unit can be implemented by computer program instructions and run on a 3D computer-aided design system equipped with parametric modeling and 3D annotation functions. The parametric reconstruction unit creates sketch features and assembly constraints in batches using design variables through the system's application programming interface. The manufacturing information annotation unit writes process, tolerance, and inspection point information onto the model's geometric elements through the system's 3D annotation interface. The manufacturing output unit exports its data exchange format to downstream CNC and manufacturing execution systems. The functional division of the above units is for descriptive purposes only; any equivalent combination of functional modules capable of achieving multi-source heterogeneous data acquisition, coordinate registration, parametric reconstruction, collaborative manufacturing information annotation, and closed-loop manufacturing output is within the scope of this invention.

Claims

1. A method for 3D modeling and collaborative annotation of existing equipment, characterized in that, In a device modeling system with multi-source data sensing, parametric reconstruction, and manufacturing closed-loop output capabilities, the following steps are included: The system collects multi-source heterogeneous data from existing equipment, acquires surface point clouds of existing equipment through a 3D scanning device, acquires external images of existing equipment through an image acquisition device and identifies 2D engineering drawing elements from them, and receives dimensional data obtained from on-site measurements of key parts of existing equipment by measuring tools. The multi-source heterogeneous data includes at least 3D point cloud data, 2D engineering drawing data, and on-site measured dimensional data, and the three types of data belong to mutually independent first acquisition coordinate system, second acquisition coordinate system, and third acquisition coordinate system, respectively. Multi-source heterogeneous data are uniformly transformed to the same model coordinate system to obtain a registered unified feature dataset. Geometric features are extracted from a unified feature dataset and a parameterized feature set is constructed. Based on the observations of each data source for the same size parameter in the multi-source heterogeneous data, a complete size parameter vector is obtained through weighted fusion. This vector is then used to drive a 3D computer-aided design system to generate a parameterized 3D model with design variables. Manufacturing information is collaboratively annotated on a parametric 3D model. The manufacturing information includes at least process route information, tolerance information, and inspection point information. The cumulative transfer amount of the annotation tolerance on the feature chain is calculated based on the size sensitivity chain. The annotated parametric 3D model and its manufacturing information are output to the manufacturing execution end, forming a closed loop from data acquisition to manufacturing execution.

2. The method according to claim 1, characterized in that, In the step of uniformly transforming multi-source heterogeneous data to the same model coordinate system, a rigid body transformation is performed to align the feature point set belonging to the first acquisition coordinate system with the feature point set belonging to the model coordinate system. The rotation matrix and translation vector of the rigid body transformation are obtained by performing singular value decomposition on the cross covariance matrix of the two feature point sets, and the translation vector is determined by the difference between the centroids of the two feature point sets.

3. The method according to claim 1, characterized in that, In the step of obtaining the complete size parameter vector through weighted fusion, for the first... Each dimension parameter is weighted by the inverse variance of the square of the standard deviation of the observations. Observations from no fewer than two data sources are weighted and fused to obtain the fused dimension parameters and their fused standard deviation. The fused dimension parameter vector is then used to drive the 3D computer-aided design system to reconstruct the parametric 3D model.

4. The method according to claim 1, characterized in that, In the step of collaboratively annotating manufacturing information on a parametric 3D model, for a derived feature determined by a function based on several dimensional parameters, the cumulative tolerance of the derived feature is calculated based on the first-order sensitivity of the function to each component dimension. The sum of the products of the annotation tolerance of each component dimension and the absolute value of its sensitivity is taken as the cumulative tolerance in the worst case. The cumulative tolerance is written into the manufacturing information as the annotation tolerance of the derived feature.

5. The method according to claim 1, characterized in that, In the step of outputting the annotated parametric 3D model and its manufacturing information to the manufacturing execution terminal, the model and manufacturing information are output to at least one of the CNC machining equipment, the manufacturing execution system, and the on-site inspection terminal. The dimensional deviations fed back by the manufacturing execution terminal are used as new observations and fed back to participate in the weighted fusion to trigger model parameter correction, thereby forming a closed loop.

6. A 3D modeling and collaborative annotation system for existing equipment, used to perform the steps of the method according to any one of claims 1 to 5, characterized in that, include: The data acquisition unit has its data output end connected to the input end of the coordinate registration unit, and is used to acquire multi-source heterogeneous data from existing equipment. The multi-source heterogeneous data includes at least three-dimensional point cloud data, two-dimensional engineering drawing data, and dimensional data obtained from on-site measurements. The three types of data belong to the first acquisition coordinate system, the second acquisition coordinate system, and the third acquisition coordinate system, which are independent of each other. The coordinate registration unit is used to uniformly transform multi-source heterogeneous data to the same model coordinate system, obtain the registered unified feature dataset, and send it to the parametric reconstruction unit. The parametric reconstruction unit is used to extract geometric features from a unified feature dataset and construct a parametric feature set. Based on the observations of each data source for the same size parameter in the multi-source heterogeneous data, a complete size parameter vector is obtained through weighted fusion, which drives the three-dimensional computer-aided design system to generate a parametric three-dimensional model with design variables. The manufacturing information annotation unit, whose input end is connected to the output end of the parametric reconstruction unit, is used to collaboratively annotate manufacturing information on the parametric 3D model. The manufacturing information includes at least process route information, tolerance information, and inspection point information, and calculates the cumulative transfer amount of the annotation tolerance on the feature chain based on the size sensitivity chain. The manufacturing end output unit has its input end connected to the output end of the manufacturing information annotation unit and its output end connected to the manufacturing execution end. It is used to output the annotated parametric 3D model and its manufacturing information to the manufacturing execution end, and to transmit the dimensional deviations fed back by the manufacturing execution end back to the parametric reconstruction unit, forming a closed loop from data acquisition to manufacturing execution.

7. The system according to claim 6, characterized in that, The coordinate registration unit performs a set of feature points belonging to the first acquisition coordinate system. and the feature point set belonging to the model coordinate system Perform rigid body transformation alignment; the points in this feature point set, after transformation, satisfy... , where the rotation matrix With translation vector The cross covariance matrix of two feature point sets Obtained by singular value decomposition and Let be the centroids of the two feature point sets respectively, and let the translation vector satisfy . .

8. The system according to claim 6, characterized in that, The parameterized reconstruction unit for the first Each size parameter is used to obtain observations from at least two data sources. and its uncertainty Weighted by inverse variance The observations are weighted and fused to obtain the fused size parameters. and its fusion standard deviation The fused dimension parameter vector is used to drive the 3D computer-aided design system to reconstruct the parametric 3D model.

9. The system according to claim 6, characterized in that, The manufacturing information labeling unit is based on several dimensional parameters. via function Given a defined derived feature, the cumulative tolerance of that derived feature is calculated based on the first-order sensitivity of the function to each component dimension. ,in For the first The tolerances for each dimensional parameter are specified, and the cumulative tolerances are calculated. The tolerance specified as a derived feature is written into the manufacturing information.

Images