Intelligent manufacturing method for rotary workpiece

Abstract

The present invention discloses an intelligent manufacturing method for rotary workpieces, belonging to the technical field of rotary workpiece processing and used for manufacturing rotary workpieces, including automated process design, solid topology analysis, process optimization closed-loop, and adaptive correction of NC programs; the automated process design includes geometric feature extraction, abnormal topology processing, and parametric modeling; the solid topology analysis includes solid topology decomposition, surface type discrimination, geometric feature parameterization, and abnormal processing; the process optimization closed-loop includes multi-modal in-machine detection, real-time data processing, intelligent compensation modeling, and closed-loop control; the adaptive correction of NC programs includes data parsing and preprocessing, intelligent path matching, path planning optimization, and closed-loop verification. The present invention greatly reduces manual intervention in mechanical design and processing, reduces work mistakes, and has greatly improved quality and efficiency; it greatly reduces the workload of repetitive design, programming, and detection, and greatly improves the design and development efficiency.

Description

Technical Field

[0001] The present invention discloses an intelligent manufacturing method for rotary parts, belonging to the technical field of rotary part processing. Background Art

[0002] With the rapid development of global manufacturing, intelligent manufacturing has become an important direction for the transformation and upgrading of the manufacturing industry. As basic parts in mechanical manufacturing, the processing quality of rotary parts is directly related to the performance, reliability, and lifespan of products. With the continuous improvement of the requirements for product quality and efficiency in modern industry, the traditional processing technology for rotary parts can no longer meet the development needs of intelligent manufacturing. Therefore, it is of great significance to study the processing technology for rotary parts based on intelligent manufacturing. The challenges faced by the processing technology for rotary parts are as follows: high requirements for processing accuracy, with the improvement of industrial automation, the requirements for the processing accuracy of rotary parts are getting higher and higher; large demand for processing efficiency, the demand for the processing efficiency of rotary parts in modern industry is increasing day by day to meet the production rhythm; control of processing costs, how to reduce production costs has become the focus of attention of enterprises on the premise of ensuring processing quality; low level of intelligence and automation, traditional processing methods rely on manual operations, and the level of intelligence and automation is relatively low. Summary of the Invention

[0003] The purpose of the present invention is to provide an intelligent manufacturing method for rotary parts to solve the problem in the prior art that the intelligent manufacturing method for rotary parts relies on manual labor, resulting in low efficiency.

[0004] The intelligent manufacturing method for rotary parts includes automated process design, entity topology analysis, process optimization closed-loop, and adaptive correction of NC programs;

[0005] The automated process design includes geometric feature extraction, abnormal topology processing, and parametric modeling;

[0006] The geometric feature extraction includes dimensional topology correlation analysis and contour boundary reconstruction;

[0007] The dimensional topology correlation analysis includes obtaining the primitive elements of the rotary part, calling the API of the 2D drawing software to traverse the primitive element database, establishing a dimension-primitive element mapping relationship, and determining the main contour through extreme value screening;

[0008] The contour boundary reconstruction includes performing polar coordinate sorting on the main contour to determine the starting edge, and constructing a closed main contour using depth-first search (DFS);

[0009] The abnormal topology processing includes duplicate curve disambiguation and geometric discontinuity repair;

[0010] The parametric modeling includes importing the repaired main contour into 3D modeling software, ensuring the correct parsing of the curve through geometric verification, validating the curve by combining geometric tolerance detection methods, and issuing a warning and performing automatic repair in a timely manner if any deviation is found;

[0011] The solid topology analysis includes solid topology decomposition, surface type discrimination, geometric feature parameterization, and exception handling;

[0012] The solid topology decomposition includes generating feature-sensitive sampling points for the reconstructed 3D features using the adaptive chord difference method and establishing a curvature weight sampling model;

[0013] The surface type discrimination includes plane feature detection, rotational surface analysis, and torus feature parsing;

[0014] The process optimization closed loop includes multi-modal in-machine inspection, real-time data processing, intelligent compensation modeling, and closed-loop control;

[0015] The adaptive correction of the NC program includes data parsing and preprocessing, intelligent path matching, path planning optimization, and closed-loop verification.

[0016] Duplicate curve disambiguation includes constructing an adjacency matrix to detect topological conflicts for the extracted geometric features, performing redundant path pruning, and defining a curve set , the curve is represented by continuous points Calculate the start and end distances of any two curves and . When the distance is less than the threshold , it is determined that and overlap and one of them is deleted. The start and end distance is the Euclidean distance :

[0017] ;

[0018] ;

[0019] In the formula, is the starting point of the point of , is the starting point of the point of , is the ending point of the point of , is the ending point of the point of , is and The coordinates of the two starting points, are and the coordinates of the two ending points;

[0020] Geometric discontinuity repair includes quickly locating breakpoints for the curves after duplicate curve disambiguation based on the R-tree spatial index, processing the breakpoints using the minimum distance bridging algorithm, defining an open point set There are points in , calculate the Euclidean distance between any two open points. If it is less than the threshold, the curve is restored by interpolation :

[0021] ;

[0022] In the formula, is the Bernstein polynomial, is the th point, represents the total number of interpolation points, and fitting is performed by the least squares method to optimize the interpolation process and ensure the smooth continuity of the restored curve.

[0023] The parametric modeling includes, after importing the curve, sorting according to the th endpoint and generating a solid of revolution in order. The sorting algorithm is based on heuristic search. First, sort according to the coordinates, and then sort according to the coordinates to ensure the smooth connection of all curves; obtain the three-dimensional solid of revolution through a rotation transformation The mathematical formula of is:

[0024] ;

[0025] In the formula, is the rotation matrix for rotation around the axis, is the rotation angle, and the generated solid of revolution is visually displayed through computer-aided design software;

[0026] After the three-dimensional modeling is completed, use the point cloud data to analyze the details of the model, and adopt the plane fitting technology based on the RANSAC algorithm to automatically identify the feature planes from the point cloud:

[0027] ;

[0028] In the formula, is the error function, is the normal vector, is the offset, represents the points represents the total number of planar points. The feature surface and boundary are extracted through plane fitting, providing a basis for the subsequent generation of machining paths.

[0029] The planar feature detection includes performing principal component analysis (PCA) on the reconstructed three-dimensional features to verify the consistency of the normal vectors and establishing a flatness error model.

[0030] The rotational surface analysis includes cylinder surface recognition, cone surface verification, and spherical surface determination of the reconstructed three-dimensional features.

[0031] The torus feature analysis includes constructing a bi-curvature feature space for the reconstructed three-dimensional features and detecting the sweeping trajectory using the Hough transform.

[0032] The geometric feature parameterization includes the reconstruction of the reference axis system and the inverse solution of feature parameters.

[0033] The reconstruction of the reference axis system includes determining the rotation axis through least squares fitting and establishing the tolerance zone of the axis system.

[0034] The inverse solution of feature parameters includes respectively obtaining the radius, semi-angle, major radius, and center of the cylinder surface, cone surface, torus, and spherical surface of the three-dimensional features.

[0035] The exception handling includes the discrimination of mixed surfaces and the processing of transition features.

[0036] The discrimination of mixed surfaces includes constructing a Mahalanobis distance classifier and setting a type confidence threshold for the discrimination of mixed surfaces.

[0037] The processing of transition features includes detecting chamfers and fillets using curvature derivatives, classifying and feature-recognizing point cloud data using PointNet++, and automatically identifying key features in complex geometric shapes in combination with deep learning. The input of the deep learning network is a point cloud data set with points , and the output is the machining feature category. Each layer of the network learns local features through a multi-layer perceptron (MLP) and extracts global features through a global pooling operation:

[0038] ;

[0039] In the formula, is the feature of the i-th point, represents the -th point of is the weight, is a constant term. The global features are extracted through global pooling , and finally the machining feature category is output, helping the system to automatically identify the key machining features on the workpiece;

[0040] Based on the recognized key machining features, through the tool path interpolation algorithm, the machining path for finish milling is automatically generated to ensure that the tool moves along the optimal path, avoiding machining errors. For different machining features, the system automatically adjusts the cutting parameters according to the material and geometric features of the workpiece to optimize the machining efficiency;

[0041] After completing the exception handling, the features of the rotary workpiece are recognized.

[0042] The multi-modal in-machine detection includes constructing a high-precision tactile sensing unit and unifying the machine tool coordinate system;

[0043] The construction of the high-precision tactile sensing unit includes constructing a probe dynamic compensation model for the features of the rotary workpiece and using the quaternion method vector solution algorithm to solve the probe dynamic compensation model;

[0044] The unification of the machine tool coordinate system includes establishing a machine tool coordinate transformation matrix, planning the path based on the curvature adaptive spiral scanning path, using a global optimization algorithm to find the optimal closed path, iteratively adjusting the link order based on the endpoint set, and minimizing the target parameters :

[0045] ;

[0046] In the formula, is the curve closing function, which avoids the local optimal solution problem through global optimization;

[0047] The real-time data processing includes intelligent analysis of measurement data and machining allowance analysis;

[0048] The intelligent analysis of measurement data includes constructing an abnormal point filtering model for the features of the rotary workpiece, using the ICP algorithm with curvature constraint to perform point cloud matching;

[0049] The machining allowance analysis includes calculating the normal allowance and tangential deviation.

[0050] The intelligent compensation modeling includes reverse surface reconstruction algorithm and dynamic adjustment of process parameters;

[0051] The reverse surface reconstruction algorithm includes using NURBS surface for adaptive interpolation and performing gradient descent optimization;

[0052] The dynamic adjustment of process parameters includes constructing a cutting parameter optimization model and calculating the tool compensation amount.

[0053] The closed-loop control includes in-machine detection, data verification for the features of the rotary workpiece. If the verification is qualified, point cloud registration, allowance analysis, compensation modeling, tool path regeneration, secondary machining, and return to in-machine detection are performed. If the verification is unqualified, an alarm is given and the machine stops.

[0054] The data parsing and preprocessing includes constructing an efficient coordinate extraction engine and constructing a spatial index;

[0055] The construction of the efficient coordinate extraction engine includes accelerating file reading by using a buffer stream, characterizing a rotating workpiece through closed-loop control, and constructing an abnormal data filtering model for data filtering;

[0056] The construction of the spatial index includes performing fast point cloud retrieval based on Octree;

[0057] The intelligent path matching includes invoking a dynamic distance analysis model and optimizing a projection algorithm;

[0058] The invocation of the dynamic distance analysis model includes invoking a hybrid distance calculation strategy and setting an adaptive threshold for dynamic distance analysis;

[0059] The optimization of the projection algorithm includes accelerating surface projection by using the LM algorithm and introducing normal constraint projection correction;

[0060] The path planning optimization includes three types of corrections, namely linear movement, circular arc path, and free surface. The minimum energy interpolation method, curvature matching correction, and parametric remapping are respectively used for adjustment, and acceleration continuity, continuity preservation, and a bow height error not exceeding 0.001 mm are respectively imposed as constraint conditions;

[0061] In the path planning optimization, a doubly linked list is used for program segment management, and an incremental program update strategy is invoked.

[0062] The closed-loop verification includes performing point cloud matching analysis on the characteristics of the obtained rotating workpiece and the manufactured rotating workpiece. If the deviation does not exceed the threshold, it is directly output. If the deviation exceeds the threshold, surface projection optimization, path parameter recalculation, tool compensation update, generation of a verification path, and virtual machining verification are performed, and then the point cloud matching analysis is returned.

[0063] Compared with the prior art, the present invention has the following beneficial effects: The present invention greatly reduces manual intervention in mechanical design and processing, reduces work mistakes, and greatly improves quality and efficiency; it greatly reduces the workload of repeated design, programming, and detection; it realizes the automation of product from scratch and greatly improves the design and development efficiency; through the inheritance of two-dimensional curve repair, three-dimensional conversion, and deep learning feature recognition, it solves the problems of efficiency and accuracy in the processing of complex geometric shapes by traditional methods. In practical applications, the repair and modeling time is shortened to 70% of the traditional method, effectively controlling the machining error, and is applicable to the high-precision machining requirements in the fields of aerospace, automotive molds, etc. Description of the Drawings

[0064] Figure 1It is a closed-loop control flowchart;

[0065] Figure 2 It is a closed-loop verification flowchart;

[0066] Figure 3 It is a schematic diagram of a rotating body workpiece to be processed;

[0067] Figure 4 For Figure 3 It is a schematic diagram for feature recognition of the first side;

[0068] Figure 5 For Figure 3 It is a schematic diagram for feature recognition of the second side;

[0069] Figure 6 It is a schematic diagram of spline interpolation fitting;

[0070] The reference numerals include 1 - large end, 2 - small end. Detailed implementation manners

[0071] To make the objectives, technical solutions and advantages of the present invention clearer, the technical solutions in the present invention will be clearly and completely described below. Obviously, the described embodiments are some but not all of the embodiments of the present invention. All other embodiments obtained by those of ordinary skill in the art based on the embodiments in the present invention without any creative efforts fall within the scope of protection of the present invention.

[0072] An intelligent manufacturing method for a rotating body workpiece includes automated process design, entity topology analysis, process optimization closed-loop, and adaptive correction of a numerical control program;

[0073] The automated process design includes geometric feature extraction, abnormal topology processing, and parametric modeling;

[0074] The geometric feature extraction includes dimensional topology correlation analysis and contour boundary reconstruction;

[0075] The dimensional topology correlation analysis includes obtaining rotating body workpiece primitives, traversing the primitive database by calling the API of 2D drawing software, establishing a dimension - primitive mapping relationship, and determining the main contour through extreme value screening;

[0076] The contour boundary reconstruction includes performing polar coordinate sorting on the main contour to determine the starting edge, and constructing a closed main contour using depth - first search (DFS);

[0077] The abnormal topology processing includes duplicate curve disambiguation and geometric discontinuity repair;

[0078] The parametric modeling includes importing the repaired main contour into 3D modeling software, ensuring the correct parsing of the curve through geometric verification, verifying the curve in combination with geometric tolerance detection methods, and if any deviation is found, issuing a warning in a timely manner and performing automatic repair;

[0079] The entity topology analysis includes entity topology decomposition, surface type discrimination, geometric feature parameterization, and exception handling;

[0080] The entity topology decomposition includes generating feature-sensitive sampling points for the reconstructed 3D features using the adaptive chordal deviation method and establishing a curvature-weighted sampling model;

[0081] The surface type discrimination includes plane feature detection, rotational surface analysis, and torus feature parsing;

[0082] The process optimization closed-loop includes multi-modal in-machine detection, real-time data processing, intelligent compensation modeling, and closed-loop control;

[0083] The adaptive correction of the NC program includes data parsing and preprocessing, intelligent path matching, path planning optimization, and closed-loop verification.

[0084] Duplicate curve disambiguation includes constructing an adjacency matrix to detect topological conflicts for the extracted geometric features, performing redundant path pruning, and defining a curve set , the curve is represented by continuous points Calculate the start and end distances of any two curves and . When the distance is less than the threshold , it is determined that and overlap and one of them is deleted. The start and end distance is the Euclidean distance :

[0085] ;

[0086] ;

[0087] In the formula, is the starting point of the point of , is the starting point of the point of , is the ending point of the point of , is the ending point of the point of , is and The coordinates of the two starting points, are and the coordinates of the two ending points;

[0088] Geometric discontinuity repair includes quickly locating breakpoints based on the R-tree spatial index for the curve after duplicate curve disambiguation, processing the breakpoints using the minimum distance bridging algorithm, defining an open point set There are points in , calculate the Euclidean distance between any two open points. If it is less than the threshold, the curve is restored by interpolation :

[0089] ;

[0090] In the formula, is the Bernstein polynomial, is the th point, represents the total number of interpolation points, and fitting is performed by the least squares method to optimize the interpolation process. As Figure 6 shown, ensure that the repaired curve is smooth and continuous.

[0091] The parametric modeling includes, after the curve is imported, sorting according to the th endpoint and generating a solid of revolution in order. The sorting algorithm is based on heuristic search. First, sort according to the coordinates, and then sort according to the coordinates to ensure smooth connection of all curves; obtain a three-dimensional solid of revolution through rotation transformation The mathematical formula is:

[0092] ;

[0093] In the formula, is the rotation matrix for rotation around the axis, is the rotation angle, and the generated solid of revolution is visually displayed through computer-aided design software;

[0094] After the three-dimensional modeling is completed, use the point cloud data to analyze the details of the model, and adopt the plane fitting technology based on the RANSAC algorithm to automatically identify the feature planes from the point cloud:

[0095] ;

[0096] In the formula, is the error function, is the normal vector, is the offset, Indicates the th point, represents the total number of planar points. The feature surface and boundary are extracted through plane fitting to provide a basis for subsequent machining path generation.

[0097] The planar feature detection includes performing principal component analysis PCA on the reconstructed three-dimensional features to verify the consistency of the normal vectors and establishing a flatness error model;

[0098] The rotational surface analysis includes cylindrical surface recognition, conical surface verification, and spherical surface determination on the reconstructed three-dimensional features;

[0099] The torus feature analysis includes constructing a bi-curvature feature space for the reconstructed three-dimensional features and detecting the sweeping trajectory using the Hough transform.

[0100] The geometric feature parameterization includes the reconstruction of the reference axis system and the inverse solution of feature parameters;

[0101] The reconstruction of the reference axis system includes determining the rotation axis through least squares fitting and establishing the tolerance zone of the axis system;

[0102] The inverse solution of feature parameters includes obtaining the radius, half-angle, principal radius, and center for the cylindrical surface, conical surface, torus, and spherical surface of the three-dimensional features respectively;

[0103] The exception handling includes the discrimination of hybrid surfaces and the processing of transition features;

[0104] The discrimination of hybrid surfaces includes constructing a Mahalanobis distance classifier and setting a type confidence threshold for hybrid surface discrimination;

[0105] The processing of transition features includes detecting chamfers and fillets using curvature derivatives, classifying and feature-recognizing point cloud data using PointNet++, and automatically identifying key features in complex geometric shapes in combination with deep learning. The input of the deep learning network is a point cloud data set with points , and the output is the machining feature category. Each layer of the network learns local features through a multi-layer perceptron MLP and extracts global features through a global pooling operation:

[0106] ;

[0107] In the formula, is the feature of the i-th point, represents the th point of, is the weight, is the constant term. Global features are extracted through global pooling and the machining feature category is finally output to help the system automatically identify the key machining features on the workpiece;

[0108] Based on the identified key machining features, through the tool path interpolation algorithm, the machining path for finish milling is automatically generated to ensure that the tool moves along the optimal path, avoiding machining errors. For different machining features, the system automatically adjusts the cutting parameters and optimizes the machining efficiency according to the material and geometric features of the workpiece;

[0109] After completing the exception handling, the features of the rotational workpiece are identified. In the present invention, the rotational workpiece to be processed is as Figure 3 shown, and feature recognition is performed on both sides of Figure 3 , as shown in Figure 4 and Figure 5 shown.

[0110] The multi-modal in-machine inspection includes constructing a high-precision tactile sensing unit and normalizing the machine tool coordinate system;

[0111] The construction of the high-precision tactile sensing unit includes constructing a probe dynamic compensation model for the features of the rotational workpiece and using the quaternion method vector solution algorithm to solve the probe dynamic compensation model;

[0112] The normalization of the machine tool coordinate system includes establishing a machine tool coordinate transformation matrix, performing path planning based on the curvature adaptive spiral scanning path, using a global optimization algorithm to find the optimal closed path, iteratively adjusting the link order based on the set of end points, and minimizing the target parameter :

[0113] ;

[0114] In the formula, is the curve closing function, which avoids the local optimal solution problem through global optimization;

[0115] The real-time data processing includes intelligent analysis of measurement data and machining allowance analysis;

[0116] The intelligent analysis of measurement data includes constructing an abnormal point filtering model for the features of the rotational workpiece, adding curvature constraints using the ICP algorithm, and performing point cloud matching;

[0117] The machining allowance analysis includes calculating the normal allowance and tangential deviation.

[0118] The intelligent compensation modeling includes an inverse surface reconstruction algorithm and dynamic adjustment of process parameters;

[0119] The inverse surface reconstruction algorithm includes using NURBS surfaces for adaptive interpolation and performing gradient descent optimization;

[0120] The dynamic adjustment of process parameters includes constructing a cutting parameter optimization model and calculating the tool compensation amount.

[0121] The closed-loop control is as follows Figure 1 shown in the figure, which includes in-machine detection, data verification for the characteristics of the rotary workpiece, and if the verification is qualified, point cloud registration, allowance analysis, compensation modeling, tool path regeneration, secondary machining, and return to in-machine detection. If the verification is unqualified, an alarm will be issued and the machine will stop.

[0122] The data parsing and preprocessing include constructing an efficient coordinate extraction engine and constructing a spatial index;

[0123] The construction of the efficient coordinate extraction engine includes accelerating file reading by using a buffer stream for the characteristics of the rotary workpiece through closed-loop control, and constructing an abnormal data filtering model for data filtering;

[0124] The construction of the spatial index includes rapid point cloud retrieval based on Octree;

[0125] The intelligent path matching includes calling a dynamic distance analysis model and optimizing the projection algorithm;

[0126] The calling of the dynamic distance analysis model includes calling a hybrid distance calculation strategy and setting an adaptive threshold for dynamic distance analysis;

[0127] The optimization of the projection algorithm includes accelerating surface projection by using the LM algorithm and introducing normal constraint projection correction;

[0128] The path planning optimization includes three types of corrections, namely linear movement, circular arc path, and free surface. The minimum energy interpolation method, curvature matching correction, and parametric remapping are used for adjustment respectively, and acceleration continuity, continuity preservation, and a bow height error not exceeding 0.001 mm are applied as constraint conditions respectively;

[0129] In the path planning optimization, the program segments are managed based on a doubly linked list, and an incremental program update strategy is called.

[0130] The closed-loop verification is as follows Figure 2 shown in the figure, which includes point cloud matching analysis for the characteristics of the obtained rotary workpiece and the manufactured rotary workpiece. If the deviation does not exceed the threshold, it is directly output. If the deviation exceeds the threshold, surface projection optimization, path parameter recalculation, tool compensation update, generation of a verification path, virtual machining verification are carried out, and then return to point cloud matching analysis.

[0131] Starting from the background of intelligent manufacturing, this invention analyzes the challenges faced by the machining technology of rotary workpieces, explores the current development status of the machining technology of rotary workpieces based on intelligent manufacturing, and proposes corresponding application strategies, providing a reference for the development of the intelligent manufacturing industry. In specific implementation, this invention is carried out by using three parts: intelligent geometric reconstruction, feature-driven process planning, and adaptive machining compensation system.

[0132] In intelligent geometric reconstruction, a model-driven automated modeling method is adopted, including geometric feature analysis and midpoint reconstruction, a parametric modeling engine, topology optimization and verification.

[0133] Geometric feature analysis and midpoint reconstruction include parsing the geometric topological relationship of the input engineering drawing through a constraint solver (D-Cubed component), extracting the cross-sectional feature contours of rotational parts (including primitives such as lines, arcs, and splines), and performing midpoint model reconstruction based on the tolerance zone.

[0134] The parametric modeling engine includes converting 2D contours into 3D solids based on the ACIS geometric kernel, axially stretching to generate a basic rotational body, realizing circumferential uniform distribution of feature arrays through polar coordinate transformation, and performing union / difference set operations on geometric bodies through Boolean operations to construct complex cavity structures.

[0135] Topology optimization and verification include applying B-rep-based geometric validity checks, ensuring the manifold characteristics of the solid model, performing minimum wall thickness detection (Ray-casting algorithm) and machining accessibility analysis.

[0136] In feature-driven process planning, an intelligent programming method based on model feature recognition is adopted, including feature extraction and characterization, similarity-driven strategy matching, and incremental feature library optimization.

[0137] The feature extraction and characterization include geometric topological analysis based on the midpoint model (MID), extracting machining features using multi-scale feature descriptors (including Fourier descriptors and shape context features), constructing a machining feature knowledge graph, and defining feature-process mapping relationships.

[0138] The similarity-driven strategy matching includes applying the k-nearest neighbor search algorithm (k-NN) to perform similarity matching in the machining feature library. When the similarity is greater than or equal to the preset threshold of 0.85, the associated machining strategy and parameter combination are automatically loaded.

[0139] Incremental feature library optimization includes establishing a feature library extension mechanism based on online learning. For new features with similarity less than the preset threshold, start the process parameter migration process, and realize knowledge reuse across manufacturing scenarios through transfer learning to support the adaptive generalization of process parameters.

[0140] In the adaptive machining compensation system, a closed-loop machining control system based on digital twin is constructed, including multi-modal data fusion acquisition, intelligent diagnosis of machining deviation, and dynamic path optimization engine.

[0141] The multi-modal data fusion acquisition includes an on-machine measurement (OMM) system that uses a contact probe (Renishaw MP700) for measurement to generate three-dimensional point cloud data in real time, and a Kalman filter is used to suppress measurement noise (signal-to-noise ratio improvement ≥ 35 dB).

[0142] The intelligent diagnosis of machining deviation includes establishing a three-dimensional registration model based on the ICP algorithm and calculating the feature dimension error vector.

[0143] The dynamic path optimization engine includes using the PID control algorithm to correct the machining trajectory in real time. The control parameters are adaptively adjusted through the particle swarm optimization (PSO) algorithm, supporting multi-axis linkage compensation (RTCP function), and the maximum correction speed reaches 120 mm / s.

[0144] The intelligent manufacturing of rotary workpieces is divided into three parts: automatic modeling conversion, feature programming, and in-machine detection. Automatic modeling is to identify and read the external contour of the rotary body in the customer's incoming drawing through software, adjust the middle tolerance, and then use software development to realize the conversion from 2D to 3D drawings. Then, the final machining model is generated through operations such as stretching, arraying, and Boolean operations. Feature programming is to identify the features in the middle tolerance model, then match them in the machining feature library, identify the same or similar features, and automatically match the machining strategies and machining parameters. During the user's use of this system, the feature library will be automatically filled. When the feature matching degree is lower than the preset threshold, it will automatically learn the machining methods and machining parameters used by the user to continuously improve the feature library. The detection system obtains real-time machining data through on-machine measurement, compares it with the theoretical data, and automatically adjusts the machining path.

[0145] The present invention proposes an intelligent machining process optimization technology based on two-dimensional drawings and three-dimensional modeling, aiming to extract the contour curves required for machining from two-dimensional drawings by automated means and generate accurate three-dimensional models and machining paths. The core technologies in the solution include automatic contour curve recognition and repair, three-dimensional modeling and topological analysis, deep learning-assisted feature recognition, machining path optimization and adaptive control, etc. By combining point cloud technology, deep learning technology with a real-time feedback system, this solution can not only effectively improve machining accuracy, but also reduce manual intervention, adapt to complex machining tasks, and improve production efficiency. During the machining process, the accurate extraction and repair of contour curves are the key links to ensure machining accuracy. Especially in complex geometric shapes, the repair and extraction of contour curves are crucial.

[0146] The present invention obtains the graphics and dimension information exported by the 2D software by reading text files, and in the automatic modeling mode, offsets the surface that needs to adjust the medium tolerance by calling the method of offsetting the surface in the 3D software, and adjusts the corresponding dimension to the medium tolerance state. Match the corresponding processing process flow, processing strategy, processing parameters, processing tools, processing machine tools, and process equipment for the identified surface features from the formed feature database. The general process of matching is to identify the processing features starting from the surface with the largest X value, identify the feature pictures through machine learning, and then match the processing parameters from the feature library.

[0147] Taking the blank size of Ø152×78 and the material of #45 stainless steel as an example, the processing process flow includes 4 processing processes, namely rough turning of the large end 1, finish turning of the large end 1, rough turning of the small end 2, and finish turning of the small end 2. The processing machine tool uses a Mazak QTE-300L horizontal lathe, and the clamping method uses three-jaw chuck positioning and clamping. The processing parameters and processing strategies are automatically matched from the SQL SERVER parameter library. The content of the match is to match the most suitable rotational speed and feed parameters from the parameter library according to the processing material, machine tool type, processing process (rough turning, semi-finish turning, finish turning), and roughness requirements. If there are no parameters suitable for the current working conditions in the parameter library, the software will automatically record the currently set parameters and update the parameter library automatically.

[0148] Use the function of discretizing the curve into points to discretize the cross-section inspection line into a large number of points. The method of obtaining the vector of a specified point on the surface can obtain the UV direction and normal vector of the specified point on the surface. Finally, generate a curve based on the obtained points and vector directions. By adding the curve to the processing strategy in the 3D software, the generated curve can be added to the processing strategy to automatically generate a program. During the process of generating the program, due to the continuous change of the processing working conditions, it may occur that the processing parameter library cannot meet the current situation and the program cannot be generated. To solve this problem, the system adds a process iteration and learning function. When encountering an unmatched working condition, the system will prompt the user to perform manual operations on unknown processing features. When generating the program, the system will record the process and processing parameters provided by the user and automatically recommend them the next time a similar working condition is encountered.

[0149] The present invention uses a PC in-machine measurement software, controls the numerical control system through the measurement NC system, the numerical control system controls the machine tool servo system and then controls the turntable. At the same time, it also controls the probe through the spindle. The probe measures the equipment on the turntable and then transmits the data to the numerical control system, and the numerical control system then feeds back the measurement results to the PC in-machine measurement software.

[0150] The above embodiments are only used to illustrate the technical solutions of the present invention, rather than limiting it. Although the present invention has been described in detail with reference to the foregoing embodiments, those of ordinary skill in the art should understand that they can still modify the technical solutions described in the foregoing embodiments, or perform equivalent replacements on some or all of the technical features, and 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. Intelligent manufacturing method for a revolving workpiece, characterized in that, It includes automated process design, solid topology analysis, process optimization closed-loop, and adaptive correction of NC programs; The automated process design includes geometric feature extraction, abnormal topology processing, and parametric modeling; The geometric feature extraction includes dimensional topology correlation analysis and contour boundary reconstruction; The dimensional topology correlation analysis includes obtaining primitive elements of a rotary workpiece, invoking the API of 2D drawing software to traverse the primitive database, establishing a dimension-primitive mapping relationship, and determining the main contour through extreme value screening; The contour boundary reconstruction includes performing polar coordinate sorting on the main contour to determine the starting edge and constructing a closed main contour using depth-first search (DFS); The abnormal topology processing includes duplicate curve disambiguation and geometric discontinuity repair; The parametric modeling includes importing the repaired main contour into 3D modeling software, ensuring correct curve parsing through geometric verification, verifying the curve using a geometric tolerance detection method, and issuing a warning and performing automatic repair in a timely manner if any deviation is found; The solid topology analysis includes solid topology decomposition, surface type discrimination, geometric feature parameterization, and abnormal handling; The solid topology decomposition includes using the adaptive chord error method to generate feature-sensitive sampling points for the reconstructed 3D features and establishing a curvature weight sampling model; The surface type discrimination includes plane feature detection, surface of revolution analysis, and torus feature analysis; The process optimization closed-loop includes multi-modal in-machine detection, real-time data processing, intelligent compensation modeling, and closed-loop control; The adaptive correction of NC programs includes data parsing and preprocessing, intelligent path matching, path planning optimization, and closed-loop verification; The geometric feature parameterization includes reference axis system reconstruction and feature parameter inverse solution; The reference axis system reconstruction includes determining the rotation axis through least squares fitting and establishing an axis system tolerance zone; The feature parameter inverse solution includes respectively obtaining the radius, half angle, major radius, and center for the cylindrical surface, conical surface, torus surface, and spherical surface of the 3D feature; The abnormal handling includes hybrid surface discrimination and transition feature processing; The hybrid surface discrimination includes constructing a Mahalanobis distance classifier and setting a type confidence threshold for hybrid surface discrimination; The transitional feature processing includes using curvature derivatives to detect chamfers and fillets, using PointNet++ to classify and identify point cloud data, and automatically identifying key features in complex set shapes in combination with deep learning. The input of the deep learning network is a point cloud data set with points , and the output is the machining feature category. Each layer of the network learns local features through a multi-layer perceptron MLP and extracts global features through a global pooling operation: ； Wherein, is the feature of the i-th point, represents the th point, is the weight, is the constant term. The global features are extracted through global pooling to finally output the processed feature category, which helps the system automatically identify the key processing features on the workpiece; Based on the identified key machining features, through the tool path interpolation algorithm, automatically generate the machining path for finish milling, ensure that the tool moves along the optimal path, avoid machining errors, and for different machining features, the system automatically adjusts the cutting parameters according to the material and geometric features of the workpiece and optimizes the machining efficiency; After completing the abnormal handling, identify the features of the rotary workpiece.

2. The intelligent manufacturing method for a rotary workpiece according to claim 1, characterized in that Duplicate curve disambiguation includes constructing an adjacency matrix to detect topological conflicts for the extracted geometric features, performing redundant path pruning, and defining a curve set , a curve is represented by consecutive points , calculating the start and end distances of any two curves and . When the distance is less than a threshold , it is determined that and overlap and one of them is deleted. The start and end distance is the Euclidean distance :[[]]END]] ； ； In the formula, is the point of the starting point, is the point of the starting point, is the point of the ending point, is the point of the ending point, is and the coordinates of the two starting points, is and the coordinates of the two ending points; Geometric discontinuity repair includes quickly locating breakpoints for the curves after disambiguating repeated curves based on the R-tree spatial index, processing the breakpoints using the minimum distance bridging algorithm, and defining an open point set There are points in , calculate the Euclidean distance between any two open points, and if it is less than the threshold, restore the curve by interpolation : ； wherein, is the Bernstein polynomial, is the th point, represents the total number of interpolation points. The fitting is performed by the least squares method to optimize the interpolation process and ensure that the repaired curve is smooth and continuous.

3. The intelligent manufacturing method of the rotary workpiece according to claim 2, characterized in that The parametric modeling includes, after importing the curves, sorting according to the th end point and generating a solid of revolution in sequence. The sorting algorithm is based on heuristic search. First, sort according to the coordinates, and then sort according to the coordinates to ensure smooth connection of all curves; obtain a three-dimensional solid of revolution through rotation transformation. The mathematical formula is: ； In the formula, is the rotation matrix rotating around axis, is the rotation angle, and the generated solid of revolution is visually displayed through computer-aided design software; After completing the 3D modeling, use point cloud data to perform detailed analysis on the model, and adopt the plane fitting technology based on the RANSAC algorithm to automatically identify the feature plane from the point cloud: ； wherein, is the error function, is the normal vector, is the offset, represents the -th point, represents the total number of plane points. The feature surface and the boundary are extracted by plane fitting to provide a basis for the subsequent generation of the machining path.

4. The intelligent manufacturing method for a rotary workpiece according to claim 3, characterized in that The plane feature detection includes performing principal component analysis (PCA) on the reconstructed 3D features to verify the normal vector consistency and establishing a flatness error model; The surface of revolution analysis includes cylindrical surface identification, conical surface verification, and spherical surface determination for the reconstructed 3D features; The torus feature analysis includes constructing a bi-curvature feature space for the reconstructed 3D features and using the Hough transform to detect the swept trajectory; 5. The intelligent manufacturing method of the rotary workpiece according to claim 4, characterized in that, The multi-modal in-machine detection includes constructing a high-precision tactile sensing unit and unifying the machine tool coordinate system; The construction of the high-precision tactile sensing unit includes constructing a probe dynamic compensation model for the characteristics of the rotating body workpiece, and using the quaternion normal vector calculation algorithm to solve the probe dynamic compensation model; The normalization of the machine tool coordinate system includes establishing a machine tool coordinate transformation matrix, performing path planning based on a curvature-adaptive spiral scanning path, using a global optimization algorithm to find the optimal closed path, iteratively adjusting the connection order based on the set of endpoints, and minimizing the target parameters : ； In the formula, is a curve closing function, which avoids the problem of local optimal solutions through global optimization; The real-time data processing includes intelligent analysis of measurement data and machining allowance analysis; The intelligent analysis of measurement data includes constructing an abnormal point filtering model for the characteristics of the rotating body workpiece, using the ICP algorithm with curvature constraint to perform point cloud matching; The machining allowance analysis includes calculating the normal allowance and tangential deviation.

6. The intelligent manufacturing method of the rotary workpiece according to claim 5, wherein, The intelligent compensation modeling includes the reverse surface reconstruction algorithm and dynamic adjustment of process parameters; The reverse surface reconstruction algorithm includes using NURBS surface for adaptive interpolation and performing gradient descent optimization; The dynamic adjustment of process parameters includes constructing a cutting parameter optimization model and calculating the tool compensation amount.

7. The intelligent manufacturing method of the rotary workpiece according to claim 6, characterized in that, The closed-loop control includes in-machine detection and data verification for the characteristics of the rotating body workpiece. If the verification is qualified, point cloud registration, allowance analysis, compensation modeling, tool path regeneration, secondary machining, and return to in-machine detection are performed. If the verification is unqualified, an alarm is given and the machine stops.

8. The intelligent manufacturing method for a rotary workpiece according to claim 7, characterized in that The data analysis and preprocessing include constructing an efficient coordinate extraction engine and constructing a spatial index; The construction of the efficient coordinate extraction engine includes accelerating file reading of the characteristics of the rotating body workpiece passing through the closed-loop control by using a buffer stream, and constructing an abnormal data filtering model for data filtering; The construction of the spatial index includes fast point cloud retrieval based on Octree; The intelligent path matching includes calling a dynamic distance analysis model and optimizing the projection algorithm; The calling of the dynamic distance analysis model includes calling a hybrid distance calculation strategy and setting an adaptive threshold for dynamic distance analysis; The optimization of the projection algorithm includes accelerating surface projection using the LM algorithm and introducing normal constraint projection correction; The path planning optimization includes three types of corrections, namely linear movement, circular arc path, and free surface. The minimum energy interpolation method, curvature matching correction, and parametric remapping are used for adjustment respectively, and acceleration continuity, continuity preservation, and a bow height error not exceeding 0.001 mm are applied as constraint conditions respectively; In the path planning optimization, a doubly linked list is used for program segment management, and an incremental program update strategy is called.

9. The intelligent manufacturing method for a rotating workpiece according to claim 8, characterized in that, The closed-loop verification includes performing point cloud matching analysis on the characteristics of the obtained rotating body workpiece and the manufactured rotating body workpiece. If the deviation does not exceed the threshold, it is directly output. If the deviation exceeds the threshold, surface projection optimization, path parameter recalculation, tool compensation update, generation of a verification path, and virtual machining verification are performed, and then return to point cloud matching analysis.

Images