Workpiece correction process quality detection and analysis method and device
Through three-dimensional surface scanning, convolution processing and finite element analysis, combined with adaptive interactive multi-model filtering technology, intelligent orthopedic strategies are generated, which solves the problem of insufficient accuracy and reliability in traditional orthopedic methods, and achieves efficient and accurate workpiece orthopedic processes and stability improvement.
Patent Information
- Application Number
- CN202411175441.5
- Authority / Receiving Office
- CN · China
- Patent Type
- Patents(China)
- Current Assignee / Owner
- Filing Date
- 2024-08-26
- Publication Date
- 2024-11-26
- Estimated Expiration
- 2044-08-26
Smart Images

Figure CN118690617B_ABST
Abstract
Description
Technical Field
[0001] The present application relates to the technical field of workpiece correction, and in particular to a method and device for detecting and analyzing the quality of a workpiece correction process. Background Art
[0002] Workpiece correction is a key process to ensure product quality in modern manufacturing, and is of great significance to improving product accuracy and performance. Traditional workpiece correction methods mainly rely on manual experience and simple measurement methods, which are difficult to meet the production needs of high precision and high efficiency. At the same time, workpieces are prone to various deformations and defects during processing and use. The complexity and uncertainty of these problems have brought great challenges to the correction process.
[0003] Existing orthopedic quality inspection methods are often limited to a single inspection method, which is difficult to fully reflect the actual state of the workpiece. In addition, most orthopedic analysis methods lack the comprehensive use of multi-source information and cannot effectively deal with the randomness and nonlinear characteristics of workpiece deformation. These problems make it difficult to ensure the accuracy and reliability of the orthopedic process, affecting the overall quality and production efficiency of the product. Summary of the invention
[0004] The present application provides a method and device for detecting and analyzing the quality of a workpiece correction process, which are used to improve the accuracy of detecting and analyzing the quality of a workpiece correction process.
[0005] In a first aspect, the present application provides a method for detecting and analyzing the quality of a workpiece correction process, the method comprising:
[0006] Perform three-dimensional scanning on the surface of the target workpiece to obtain a three-dimensional data model of the workpiece;
[0007] Performing convolution processing on the three-dimensional data model of the workpiece to obtain surface defect information of the workpiece;
[0008] Performing finite element analysis on the three-dimensional data model of the workpiece and the surface defect information of the workpiece to obtain multi-model deformation state information;
[0009] Performing adaptive interactive multi-model filtering on the multi-model deformation state information to obtain a real-time state estimation result;
[0010] Performing multi-objective optimization calculation on the real-time state estimation result to obtain a target correction strategy;
[0011] According to the target orthopedic strategy, an orthopedic operation is performed on the target workpiece and a multi-index evaluation is performed to generate an orthopedic quality evaluation report.
[0012] In a second aspect, the present application provides a workpiece correction process quality detection and analysis device, the workpiece correction process quality detection and analysis device comprising:
[0013] A scanning module is used to perform three-dimensional scanning on the surface of a target workpiece to obtain a three-dimensional data model of the workpiece;
[0014] A processing module, used for performing convolution processing on the three-dimensional data model of the workpiece to obtain surface defect information of the workpiece;
[0015] An analysis module, used for performing finite element analysis on the three-dimensional data model of the workpiece and the surface defect information of the workpiece to obtain multi-model deformation state information;
[0016] A filtering module, used for performing adaptive interactive multi-model filtering on the multi-model deformation state information to obtain a real-time state estimation result;
[0017] A calculation module, used for performing multi-objective optimization calculation on the real-time state estimation result to obtain a target correction strategy;
[0018] An evaluation module is used to perform an orthopedic operation on the target workpiece according to the target orthopedic strategy and conduct a multi-index evaluation to generate an orthopedic quality evaluation report.
[0019] The third aspect of the present application provides a workpiece correction process quality detection and analysis device, comprising: a memory and at least one processor, wherein the memory stores instructions; the at least one processor calls the instructions in the memory so that the workpiece correction process quality detection and analysis device performs the above-mentioned workpiece correction process quality detection and analysis method.
[0020] A fourth aspect of the present application provides a computer-readable storage medium, wherein the computer-readable storage medium stores instructions, which, when executed on a computer, enable the computer to execute the above-mentioned workpiece correction process quality detection and analysis method.
[0021] In the technical solution provided by the present application, by performing three-dimensional scanning and convolution processing on the surface of the workpiece, combined with finite element analysis, a full-scale analysis of the surface morphology, internal structure and mechanical properties is achieved. This multi-source information fusion method can more comprehensively and accurately reflect the actual state of the workpiece, and the adaptive interactive multi-model filtering technology can effectively handle the randomness and nonlinear characteristics of the workpiece deformation. By dynamically adjusting the model weights and transition probabilities, accurate estimation of complex deformation states is achieved, and the adaptability and robustness of the correction process are improved. Based on the real-time state estimation results, a multi-objective optimization model including shape deviation, internal stress distribution and surface quality is constructed. Through the improved particle swarm algorithm and Pareto optimal set analysis, the intelligent generation of correction strategies is achieved, multiple correction targets are balanced, and the comprehensive performance of the correction effect is improved. During the correction execution process, the dynamic adjustment of correction parameters is achieved through real-time state monitoring and deviation analysis. This closed-loop control mechanism can effectively deal with uncertain factors in the correction process and improve the accuracy and stability of correction. Through comprehensive three-dimensional scanning, surface defect analysis and internal structure integrity detection of the workpiece after correction, a multi-index comprehensive evaluation system is established. From data collection, state estimation to strategy generation and execution, a high degree of intelligence and automation is achieved, which greatly reduces manual intervention, improves orthopedic efficiency, reduces human errors, and ensures the consistency of orthopedic quality. BRIEF DESCRIPTION OF THE DRAWINGS
[0022] In order to more clearly illustrate the technical solutions of the embodiments of the present invention, the accompanying drawings required for use in the description of the embodiments will be briefly introduced below. Obviously, the accompanying drawings described below are some embodiments of the present invention. For ordinary technicians in this field, other accompanying drawings can be obtained based on these accompanying drawings without paying any creative work.
[0023] Figure 1 A schematic diagram of an embodiment of a method for detecting and analyzing the quality of a workpiece correction process in an embodiment of the present application;
[0024] Figure 2 It is a schematic diagram of an embodiment of a device for detecting and analyzing the quality of a workpiece correction process in an embodiment of the present application. DETAILED DESCRIPTION
[0025] The embodiments of the present application provide a method and device for detecting and analyzing the quality of a workpiece correction process. The terms "first", "second", "third", "fourth", etc. (if any) in the specification and claims of the present application and the above-mentioned drawings are used to distinguish similar objects, and are not necessarily used to describe a specific order or sequence. It should be understood that the data used in this way can be interchanged where appropriate, so that the embodiments described herein can be implemented in an order other than that illustrated or described herein. In addition, the terms "including" or "having" and any variations thereof are intended to cover non-exclusive inclusions, for example, a process, method, system, product or device that includes a series of steps or units is not necessarily limited to those steps or units that are clearly listed, but may include other steps or units that are not clearly listed or inherent to these processes, methods, products or devices.
[0026] For ease of understanding, the specific process of the embodiment of the present application is described below. Figure 1 , an embodiment of the workpiece correction process quality detection and analysis method in the embodiment of the present application includes:
[0027] Step S101, performing a three-dimensional surface scan on a target workpiece to obtain a three-dimensional data model of the workpiece;
[0028] It is understandable that the execution subject of the present application may be a workpiece correction process quality detection and analysis device, or a terminal or a server, which is not limited here. The present application embodiment is described by taking the server as the execution subject as an example.
[0029] Specifically, the target workpiece is scanned three-dimensionally from multiple angles and multiple light sources to obtain the original point cloud data, ensuring the comprehensive capture and accurate reflection of the workpiece surface features. The original point cloud data is Gaussian filtered to reduce the impact of noise, and the abnormal data points are removed by statistical outlier filtering method to obtain the point cloud data with preliminary noise reduction. The normal vector and curvature are calculated based on the preliminary noise reduction point cloud data to generate the point cloud feature descriptor. The feature descriptor provides basic information for describing the local geometric shape of the point cloud. Using the point cloud feature descriptor, multiple point clouds from different scanning angles are iteratively aligned to the nearest point, so that these point clouds are fused in a unified coordinate system to obtain a complete fused point cloud model. The fused point cloud is voxelized and downsampled to form a uniformly distributed sparse point cloud. By reducing the amount of data, subsequent calculations are made more efficient while retaining key geometric information. Based on the sparse point cloud data, a kd tree index structure is constructed, and the structure is used to perform a k nearest neighbor search for each point to obtain local neighborhood information. Based on the local neighborhood information, each point is reconstructed by moving least squares to obtain a continuous and smooth surface model. Moving least squares reconstruction technology can effectively smooth surface noise and restore the original geometric features. The generated surface model is adaptively meshed to obtain a more uniform and adaptable triangular mesh model. Topology optimization and boundary repair are performed on the triangular mesh model to optimize the mesh quality and the accuracy of the model boundary to obtain a three-dimensional data model of the workpiece.
[0030] Step S102, performing convolution processing on the three-dimensional data model of the workpiece to obtain surface defect information of the workpiece;
[0031] Specifically, the three-dimensional data model of the workpiece is normalized and standardized to eliminate the scale differences between different input data and obtain the preprocessed input data. The preprocessed input data is imported into the ResNet-50 backbone network in the preset Double-HeadRegion-RCNN network for feature extraction to obtain the initial feature map. As a backbone network, ResNet-50 has strong feature extraction capabilities and can capture subtle features on the surface of the workpiece. The initial feature map is enhanced with context features through the Transformer encoder module to obtain richer global context information. The attention mechanism of Transformer can capture long-distance dependencies between features and enhance the expression ability of the feature map. Based on the enhanced global context features, the convolution kernel is dynamically adjusted and a deformable convolution operation is performed to more flexibly adapt to the complex geometric morphology of the workpiece surface and obtain adaptive local features. The feature pyramid network with content-aware feature recombination is used to perform multi-scale feature fusion on the adaptive local features to obtain multi-scale fusion features. Multi-scale feature fusion can handle surface defects of different scales and effectively capture defect features of various sizes. Based on the multi-scale fusion features, candidate regions for the scale distribution of surface defects of the workpiece are generated in the region proposal network. These candidate regions identify the parts where defects may exist. The RoI Align operation is performed on the candidate regions to obtain a feature map of a fixed size. The candidate regions of different sizes are processed uniformly to make the subsequent classification and positioning more accurate. The fixed-size feature map is input into the classification head and regression head for processing, and the category probability and bounding box coordinates of the defect are calculated respectively. This information provides the category, location and boundary information of each defect. The category probability and bounding box coordinates of the defect are non-maximum suppressed to remove overlapping candidate boxes and retain the optimal detection results to obtain the surface defect information of the workpiece. The final output defect information includes the location, size and type of the defect, which is used for subsequent orthopedic process planning and quality assessment.
[0032] Step S103, performing finite element analysis on the workpiece three-dimensional data model and the workpiece surface defect information to obtain multi-model deformation state information;
[0033] Specifically, the three-dimensional data model of the workpiece is meshed to generate an initial finite element mesh model, thereby establishing the geometric and physical basis of the workpiece in the analysis. According to the surface defect information of the workpiece, the defect area in the initial mesh model is locally refined to obtain a target finite element mesh model with higher resolution. The material properties and boundary conditions of the target finite element mesh model are defined. The material properties include the elastic modulus, yield strength and plastic deformation characteristics of the material, while the boundary conditions involve the setting of external forces and constraints. Based on these parameters, the mechanical model of the target workpiece is generated. According to the mechanical model, an elastic deformation model, a plastic deformation model and a composite deformation model are constructed to describe the deformation behavior of the workpiece under different stress conditions respectively. Through these models, the complex deformation characteristics of the workpiece under the action of external forces are fully captured. A random field theory analysis is performed on multiple deformation models to consider the inhomogeneity of the material and the uncertainty of the defects, and a random deformation model is obtained. The establishment of the random deformation model aims to evaluate the deformation possibility of the workpiece under different conditions by means of probability distribution. According to the random deformation model, simulation calculations are performed to obtain the probability distribution of the workpiece deformation. These probability distributions reflect the possibility and frequency of occurrence of different deformation modes. The probability distribution of workpiece deformation is analyzed by principal component analysis to extract the main deformation modes, which represent the main deformation forms of the workpiece under different stress states. According to the target deformation mode, the displacement field of each grid node is calculated to obtain the deformation field distribution of the target workpiece. The calculation of the displacement field distribution is to quantify the displacement size and direction of the workpiece at each point. The stress-strain analysis of the deformation field distribution is performed to obtain the internal stress distribution of the target workpiece. The internal stress distribution reveals the stress state inside the workpiece, which helps to understand the possible failure points and weaknesses of the material. According to the deformation field distribution and internal stress distribution, the weights of each deformation model are assigned to obtain the multi-model deformation state information.
[0034] Step S104, performing adaptive interactive multi-model filtering on the multi-model deformation state information to obtain a real-time state estimation result;
[0035] Specifically, the state space modeling is performed on the deformation state information of multiple models, and the state equation and observation equation of each model are established to describe the dynamic behavior and observation characteristics of the model. According to the defined state equation and observation equation, the volumetric Kalman filter is initialized for each model to obtain the initial state estimate and the first error covariance matrix. Multiple fading factors are introduced into the first error covariance matrix to obtain the second prediction error covariance. The fading factor can adjust the covariance matrix according to the historical state of the system, thereby optimizing the response capability of the filter. According to the second prediction error covariance, the filter gain is calculated to obtain the adaptive filter gain matrix. The adaptive filter gain matrix determines the response strength of the filter to the new observation data. The adaptive filter gain matrix is subjected to singular value decomposition to obtain a numerically stable matrix decomposition result. Based on these results, state update and covariance update are performed to obtain the filter estimation value of each model. The filter estimation value is the optimal state estimation of each model at the current moment, reflecting the latest dynamics of the system. The likelihood function value is calculated for the filter estimation value of each model to quantify the goodness of fit of each model. Based on the calculated likelihood function values, the model weight factors are determined, which reflect the importance of each model in the current situation. Based on the model weight factors, the Markov transition matrix is dynamically adjusted to obtain the adaptive model transition probability. The transition probability reflects the dynamic switching trend of the system between different models, so that the filtering process can more flexibly adapt to changes in the actual situation. Model interaction is performed based on the adaptive model transition probability to obtain a mixed state estimate. The estimated values of multiple models are integrated to obtain a more accurate system state. The mixed state estimate is weighted fused to obtain a real-time state estimation result.
[0036] Step S105, performing multi-objective optimization calculation on the real-time state estimation result to obtain a target correction strategy;
[0037] Specifically, the orthopedic objective function is constructed based on the real-time state estimation results. The function comprehensively considers factors such as shape deviation, internal stress distribution and surface quality to form a multi-objective optimization model. These objectives jointly define the quality standard of the orthopedic process, and the best orthopedic effect is sought by optimizing the combination of each objective. Process constraints are set for the multi-objective optimization model, such as the physical properties of the workpiece material, the operating capacity of the equipment and the limitations of the production process, to form a constrained optimization problem. These constraints ensure that the solution in the optimization process is not only theoretically feasible, but also operational in practical applications. The particle swarm algorithm is used to initialize the constrained optimization problem and generate an initial particle swarm. The particle swarm algorithm searches for the global optimal solution by simulating the clustering behavior of swarm organisms, where each particle represents a possible solution. In order to improve the search efficiency and accuracy of the algorithm, an adaptive inertia weight is introduced to dynamically adjust the speed and position of each particle. The introduction of the adaptive inertia weight factor enables the algorithm to better balance global exploration and local development during the search process and avoid falling into the local optimal solution. Based on the dynamic inertia weight factor, the speed and position of the particle are updated to obtain a new generation of particle swarms, and the position of the particles is continuously updated iteratively, so that the swarm gradually approaches the optimal solution. The new generation of particle swarms are non-dominated sorted to generate the Pareto optimal set. The Pareto optimal set contains a set of non-inferior solutions, that is, no other solution can completely outperform these solutions in all objectives. The Pareto optimal set is processed by a chaotic perturbation mechanism to expand the search space. The chaotic perturbation mechanism introduces random perturbations to prevent the algorithm from converging to the local optimal solution too early and maintain the diversity of the search. In the expanded search space, local search is performed to refine the solution and further approach the global optimal solution. The optimized Pareto front is comprehensively evaluated and the quality of each solution is measured by the comprehensive score. The calculation of the comprehensive score is based on the pre-set evaluation criteria and weights to ensure that the final selected solution can best meet the comprehensive requirements of multi-objective optimization. According to the comprehensive scores of each solution, the best solution is selected as the output target orthopedic strategy. The strategy includes detailed information such as the specific force position, magnitude, direction and loading sequence.
[0038] Step S106: Perform an orthopedic operation on the target workpiece according to the target orthopedic strategy and conduct a multi-index evaluation to generate an orthopedic quality evaluation report.
[0039] Specifically, the operation instruction sequence of the CNC orthopedic equipment is generated according to the target orthopedic strategy. These sequences describe the various steps of performing the orthopedic operation and form an orthopedic execution plan. The plan provides precise control parameters for the CNC equipment, including the position, magnitude, direction and loading sequence of the force. The orthopedic execution plan is executed step by step, and the execution results of each step are recorded to form the staged orthopedic results. The staged results help monitor the progress and effect of the orthopedic process. During the orthopedic operation, the staged orthopedic results are monitored in real time and real-time state monitoring data are collected. These data include the real-time deformation information, stress distribution and other key parameters of the workpiece. Deviation analysis is performed based on the real-time state monitoring data and the predicted results to obtain the orthopedic parameter adjustment amount. Deviation analysis can reveal the gap between the actual execution results and the expected goals. According to the deviation analysis results, the orthopedic parameters are appropriately adjusted to form an optimized orthopedic execution plan to ensure that the orthopedic operation is more accurate. After the orthopedic operation is completed, the target workpiece is fully three-dimensionally scanned to obtain the surface data of the workpiece after correction. These data reflect the actual geometric shape and surface characteristics of the workpiece. Convolution processing is performed on the surface data of the workpiece after correction to extract the updated surface defect information. This information includes the details of small surface cracks and uneven defects. Based on the updated surface defect information, the shape accuracy score is calculated. Acoustic emission testing is performed on the target workpiece. Acoustic emission testing can capture small cracks, defects and other internal structural integrity issues inside the workpiece. The shape accuracy score, updated surface defect information and internal structural integrity data are integrated together for a multi-index comprehensive evaluation. An orthopedic quality assessment report is generated to record every step of the orthopedic operation and its results.
[0040] In the embodiment of the present application, by performing three-dimensional scanning and convolution processing on the surface of the workpiece, combined with finite element analysis, a full range of analysis of surface morphology, internal structure and mechanical properties is achieved. This multi-source information fusion method can more comprehensively and accurately reflect the actual state of the workpiece, and the adaptive interactive multi-model filtering technology can effectively handle the randomness and nonlinear characteristics of workpiece deformation. By dynamically adjusting the model weights and transition probabilities, accurate estimation of complex deformation states is achieved, and the adaptability and robustness of the orthopedic process are improved. Based on the real-time state estimation results, a multi-objective optimization model including shape deviation, internal stress distribution and surface quality is constructed. Through the improved particle swarm algorithm and Pareto optimal set analysis, the intelligent generation of orthopedic strategies is achieved, multiple orthopedic goals are balanced, and the comprehensive performance of orthopedic effects is improved. During the orthopedic execution process, dynamic adjustment of orthopedic parameters is achieved through real-time state monitoring and deviation analysis. This closed-loop control mechanism can effectively deal with uncertain factors in the orthopedic process and improve the accuracy and stability of orthopedics. A multi-index comprehensive evaluation system is established through comprehensive three-dimensional scanning, surface defect analysis and internal structure integrity detection of the workpiece after orthopedics. From data collection, state estimation to strategy generation and execution, a high degree of intelligence and automation is achieved, which greatly reduces manual intervention, improves orthopedic efficiency, reduces human errors, and ensures the consistency of orthopedic quality.
[0041] In a specific embodiment, the process of executing step S101 may specifically include the following steps:
[0042] (1) Perform 3D scanning of the target workpiece from multiple angles and multiple light sources to obtain raw point cloud data, and then perform Gaussian filtering and statistical outlier filtering on the raw point cloud data to obtain preliminary denoised point cloud data;
[0043] (2) Calculate the normal vector and curvature based on the initially denoised point cloud data to obtain the point cloud feature descriptor, and perform iterative closest point registration on multiple scanning viewpoint point clouds with point cloud feature descriptors to obtain a fused point cloud in a unified coordinate system;
[0044] (3) Perform voxel downsampling on the fused point cloud to obtain a uniformly distributed sparse point cloud, and construct a kd-tree index structure based on the sparse point cloud to perform a k-nearest neighbor search on each point to obtain local neighborhood information;
[0045] (4) Perform moving least square reconstruction on each point based on the local neighborhood information to obtain a continuous and smooth surface model, and then perform adaptive meshing on the surface model to obtain a triangular mesh model;
[0046] (5) Perform topology optimization and boundary repair on the triangular mesh model to obtain a three-dimensional data model of the workpiece.
[0047] Specifically, a multi-angle, multi-light source scanning system is built. The system consists of multiple laser scanners and cameras, and multiple light sources are configured to evenly illuminate the surface of the workpiece, thereby reducing the impact of shadows and reflections. By scanning at different angles, the three-dimensional morphology of the workpiece is fully captured. The original point cloud data is generated by the scanning system, including the coordinates of each point on the surface of the workpiece and its color information. Due to noise and environmental influences, the original point cloud may contain some error data, which is smoothed using a Gaussian filter. The Gaussian filter formula is:
[0048] ;
[0049] in, is the filtered point cloud data, and are the coordinates of the midpoint of the point cloud, is the standard deviation, which determines the smoothness of the filter. Gaussian filtering can effectively reduce noise while retaining the main features of the workpiece surface. Outliers are filtered out by statistical methods. These outliers are usually abnormal data caused by sensor errors or environmental factors. The average distance and standard deviation of each point in the point cloud are calculated, and points exceeding a certain threshold are regarded as outliers and are eliminated. Through these processing steps, preliminary denoised point cloud data is obtained. The normal vector and curvature are calculated based on the preliminary denoised point cloud data. These features help describe the geometric characteristics of each point in the point cloud. The normal vector represents the normal direction of the point cloud surface, and the curvature describes the degree of curvature of the point cloud surface at this point. Normal vector The calculation of can be achieved by principal component analysis method, the formula is:
[0050] ;
[0051] in, and are the eigenvalues and eigenvectors of the point cloud data covariance matrix, is the number of neighborhood points. The curvature is calculated based on the rate of change of the normal vector. Iterative closest point registration is performed on multiple scanned viewpoint point clouds with point cloud feature descriptors so that these point cloud data are fused in a unified coordinate system. The iterative closest point registration algorithm aligns point clouds from different viewpoints by minimizing the distance difference between the two point clouds. Voxelization downsampling is performed on the fused point cloud to reduce the amount of data and evenly distribute the point cloud. Voxelization downsampling divides the point cloud into a fixed-size voxel grid, and only one representative point is retained in each voxel, which effectively reduces the computational complexity. A kd-tree index structure is constructed based on the sparse point cloud to accelerate the neighborhood search process. The kd-tree is a binary search tree structure for multidimensional spatial data, which helps to quickly find the nearest neighbor points. Based on the local neighborhood information, a moving least squares reconstruction is performed on each point to generate a continuous and smooth surface model. Moving least squares reconstruction smoothes the point cloud surface by fitting a local quadratic surface, eliminating noise and restoring the true geometric shape. The surface model is adaptively meshed, and the mesh density is dynamically adjusted according to the surface complexity to generate an adaptable triangular mesh model. The triangular mesh model is topologically optimized and boundary repaired to ensure the integrity and accuracy of the model. Topological optimization involves eliminating redundant vertices and edges, simplifying the mesh structure, and keeping the surface geometry unchanged. Boundary repair deals with boundary discontinuities caused by noise or missing data, making the model closer to the actual workpiece geometry.
[0052] In a specific embodiment, the process of executing step S102 may specifically include the following steps:
[0053] (1) Normalizing and standardizing the workpiece 3D data model to obtain preprocessed input data;
[0054] (2) Input the preprocessed input data into the ResNet-50 backbone network in the preset Double-Head Region-RCNN network for feature extraction to obtain the initial feature map;
[0055] (3) The context feature of the initial feature map is enhanced through the Transformer encoder module to obtain enhanced global context features;
[0056] (4) According to the enhanced global context features, the convolution kernel is dynamically adjusted to perform deformable convolution operations to obtain adaptive local features;
[0057] (5) Based on the feature pyramid network of content-aware feature reorganization, multi-scale feature fusion is performed on the adaptive local features to obtain multi-scale fusion features;
[0058] (6) Based on the multi-scale fusion features, candidate regions for the scale distribution of surface defects of the workpiece are generated in the region proposal network;
[0059] (7) Perform the RoI Align operation on the candidate region to obtain a fixed-size feature map, and input the fixed-size feature map into the classification head and regression head for processing to obtain the defect category probability and bounding box coordinates;
[0060] (8) Non-maximum suppression processing is performed on the defect category probability and bounding box coordinates to obtain the workpiece surface defect information, which includes the location, size and type of the defect.
[0061] Specifically, the geometric data of the workpiece is adjusted to a uniform scale and range. Each coordinate value in the point cloud data is subtracted from its corresponding mean and divided by the standard deviation. Standardization helps to eliminate scale differences between different data sets and improve the stability and prediction accuracy of the model. Normalization is to scale the data to a specific range, such as [0, 1] or [-1, 1]. The preprocessed input data is input into the ResNet-50 backbone network in the preset Double-Head Region-RCNN network for feature extraction. ResNet-50 is a deep convolutional neural network that can better capture complex image features by introducing residual connections to alleviate the problem of gradient disappearance in deep networks. The Transformer encoder module is used to enhance the contextual features of the initial feature map. The Transformer encoder captures the dependencies between distant points in the feature map through the self-attention mechanism and generates enhanced global context features. According to the enhanced global context features, the convolution kernel is dynamically adjusted to perform deformable convolution operations. Deformable convolution allows the convolution kernel to deform freely in space, adapt to local features of different shapes and sizes, and extract more accurate adaptive local features. The feature pyramid network based on content-aware feature recombination performs multi-scale feature fusion on adaptive local features. The feature pyramid network extracts information from feature maps of different scales and fuses them, so that the network can process defects of different sizes and scales, and obtains feature maps containing multi-level information. According to the multi-scale fusion features, candidate regions for the scale distribution of defects on the workpiece surface are generated in the region proposal network. The region proposal network scans the feature map in a sliding window manner to generate a series of candidate regions, each of which may contain a defect. For each candidate region, the RoI Align operation is performed to unify it to a fixed-size feature map to ensure that candidate regions of different scales can be processed consistently. The fixed-size feature map is input into the classification head and regression head for processing. The classification head predicts the category probability of the defect for each candidate region, while the regression head predicts the corresponding bounding box coordinates. The outputs of the classification head and regression head represent the type of defect (such as cracks, scratches, etc.) and its position and size on the workpiece surface, respectively. The defect category probability and bounding box coordinates are processed by non-maximum suppression to obtain the surface defect information of the workpiece. Non-maximum suppression can remove redundant candidate boxes and only retain the boxes with the highest confidence to ensure the accuracy of the detection results. The basic formula of non-maximum suppression is:
[0062] IoU ;
[0063] Among them, IoU It is the intersection-over-union ratio of two candidate regions A and B, indicating the proportion of the overlapping area of A and B. When the IoU of two candidate boxes exceeds the set threshold, the box with lower confidence will be suppressed.
[0064] In a specific embodiment, the process of executing step S103 may specifically include the following steps:
[0065] (1) Meshing the three-dimensional data model of the workpiece to obtain an initial finite element mesh model, and locally refining the mesh in the defect area based on the surface defect information of the workpiece to obtain a target finite element mesh model;
[0066] (2) Define the material properties and boundary conditions of the target finite element mesh model and generate a mechanical model of the target workpiece;
[0067] (3) Based on the mechanical model, an elastic deformation model, a plastic deformation model and a composite deformation model are constructed to obtain a multi-model deformation description, and a random field theory analysis is performed on the multi-model deformation description to obtain a random deformation model;
[0068] (4) Perform simulation calculations based on the random deformation model to obtain the probability distribution of workpiece deformation, and perform principal component analysis on the probability distribution of workpiece deformation to obtain the target deformation mode;
[0069] (5) Calculate the displacement field of each grid node according to the target deformation mode to obtain the deformation field distribution of the target workpiece, and perform stress-strain analysis on the deformation field distribution to obtain the internal stress distribution of the target workpiece;
[0070] (6) Deformation model weights are assigned according to the deformation field distribution and internal stress distribution to obtain the deformation state information of multiple models.
[0071] Specifically, the surface and internal space of the model are divided into multiple small units, which form the initial finite element mesh model. The initial mesh model will select a relatively uniform mesh size to ensure the integrity of the model and the feasibility of the calculation. According to the surface defect information of the workpiece, the mesh of the defect area is locally refined. Local refinement can be achieved by increasing the density of the mesh, that is, setting a smaller unit size in the defect area to more accurately describe the geometric shape of the defect and obtain the target finite element mesh model. Define the material properties and boundary conditions of the target finite element mesh model. Material properties include physical parameters such as elastic modulus, yield strength, Poisson's ratio, etc. These parameters describe the mechanical behavior of the material under external force. Boundary conditions are constraints on the model, such as the fixed end, free end, or the position and size of the applied external force. Based on these definitions, the mechanical model of the target workpiece is generated. The mechanical model is the basis of finite element analysis and describes the behavior of the workpiece in a mechanical environment. On the basis of the mechanical model, the elastic deformation model, plastic deformation model and composite deformation model are constructed. The elastic deformation model describes the reversible deformation behavior of the workpiece under the action of external force, that is, the workpiece can return to its original state after the external force is removed. The plastic deformation model describes the irreversible permanent deformation, which usually occurs after the external force exceeds the yield strength of the material. The composite deformation model combines elastic and plastic behaviors and can more accurately describe the actual deformation process of the workpiece. The random field theory is used to analyze the multi-model deformation description. The random field theory takes into account the spatial randomness of material properties and external forces. This randomness may come from the inhomogeneity of the material, the random distribution of defects, etc. Through random field analysis, a random deformation model is constructed. The random deformation model describes the deformation behavior of the workpiece under different conditions through probability distribution. The formula is:
[0072] ;
[0073] in, is the deformation vector The probability density function of is the mean vector, representing the average deformation, is a covariance matrix that describes the randomness of deformation and the correlation in different directions. According to the random deformation model, simulation calculation is performed to obtain the probability distribution of workpiece deformation. The probability distribution provides the probability of occurrence of different deformation modes, helping the system understand which deformation mode is most likely to occur. The probability distribution is subjected to principal component analysis. The original data is projected onto a set of unrelated principal components through linear transformation. These principal components are sorted by the size of data variance to obtain the target deformation mode, that is, several directions that describe the most important deformation characteristics of the workpiece. The displacement field of each grid node is calculated according to the target deformation mode. The displacement field describes the movement of each node under different stress conditions and is a specific manifestation of the deformation field distribution of the workpiece. The internal stress distribution of the target workpiece is obtained by performing stress-strain analysis on the displacement field distribution. The internal stress distribution reveals the stress state of different regions inside the workpiece, helping to identify possible stress concentration areas and potential points of material failure. According to the deformation field distribution and internal stress distribution, weights are assigned to different deformation models to generate multi-model deformation state information. The weight reflects the contribution of each deformation mode to the overall workpiece deformation. The weight calculation can be based on the projection size of each model in the principal component space or other correlation indicators.
[0074] In a specific embodiment, the process of executing step S104 may specifically include the following steps:
[0075] (1) Perform state space modeling on the deformation state information of multiple models to obtain the state equation and observation equation of each model;
[0076] (2) Based on the state equation and observation equation, the volumetric Kalman filter is initialized for each model to obtain the initial state estimate and the first error covariance matrix;
[0077] (3) introducing multiple fading factors into the first error covariance matrix to obtain a second prediction error covariance, and calculating the filter gain based on the second prediction error covariance to obtain an adaptive filter gain matrix;
[0078] (4) Perform singular value decomposition on the adaptive filter gain matrix to obtain a numerically stable matrix decomposition result, and perform state update and covariance update based on the matrix decomposition result to obtain the filter estimation value of each model;
[0079] (5) Calculate the likelihood function value for the filter estimate of each model to obtain the model weight factor, and dynamically adjust the Markov transfer matrix according to the model weight factor to obtain the adaptive model transfer probability;
[0080] (6) Model interaction is performed based on the adaptive model transition probability to obtain a mixed state estimate, and the mixed state estimate is weightedly fused to obtain a real-time state estimation result.
[0081] Specifically, define the state equation and observation equation of each model. The state equation describes the evolution of the system state over time, usually expressed as:
[0082] ;
[0083] in For the moment The state vector contains the displacement, speed and other information of the workpiece; is the control input; is the process noise, which is usually assumed to be Gaussian white noise. The observation equation is expressed as:
[0084] ;
[0085] in is the observation vector, which represents the quantity measured by the sensor; is the observation noise, which is also assumed to be Gaussian white noise. Through these equations, the dynamic behavior of the multi-model system under different states is described. According to the state equation and the observation equation, the cubature Kalman filter is initialized for each model. The initial estimate of the system state and the first error covariance matrix are calculated. Kalman filtering is a filter that processes nonlinear systems by performing a Gaussian approximation on the system state distribution. Initial state estimation Usually determined based on prior knowledge or historical data, the initial covariance matrix It reflects the estimation of the uncertainty of the initial state. In order to improve the response speed and stability of the filter, multiple fading factors are introduced into the first error covariance matrix. These factors adjust the elements of the covariance matrix to reflect the dynamic change characteristics of the system, and the second prediction error covariance matrix is obtained. The effect of the fading factor is similar to adjusting the learning rate of the filter, allowing it to adapt to new data more quickly. The filter gain is calculated based on the second prediction error covariance matrix , the formula is:
[0086] ;
[0087] in, is the state prediction covariance matrix, is the observation matrix, is the observation noise covariance matrix. Filter gain It determines the influence of the observed data on the state update. The adaptive filter gain matrix adapts to different dynamic environments by adjusting the filter gain. In order to ensure the numerical stability of the calculation, the adaptive filter gain matrix is subjected to singular value decomposition, and the gain matrix is decomposed into a set of singular values and singular vectors, making the subsequent matrix operations more stable and efficient. Based on the results of the singular value decomposition, the state update and covariance update are performed, and the formula is:
[0088] ;
[0089] These update steps provide the filter estimates of each model, reflecting the best estimate of the current state. Likelihood function values are calculated for the filter estimates of each model, which represent the credibility of the model estimates. The larger the likelihood function value, the closer the model is to the true state. The model weight factors are obtained through the likelihood function values, which reflect the importance of each model in the current situation. The Markov transition matrix is dynamically adjusted according to the model weight factors. The elements in the transfer matrix represent the transition probabilities between models. By adjusting these probabilities, the dynamic characteristics of the system and the dependencies between different models are reflected, and the adaptive model transition probabilities are obtained. Model interaction is performed based on the adaptive model transition probabilities to obtain a mixed state estimate. By weighted fusion of the estimates of different models, it is ensured that the final estimate can combine the advantages of different models. The mixed state estimate obtained by weighted fusion is the final result of real-time state estimation and provides the best estimate of the current state of the system.
[0090] In a specific embodiment, the process of executing step S105 may specifically include the following steps:
[0091] (1) The orthopedic objective function is constructed based on the real-time state estimation results to obtain a multi-objective optimization model that includes shape deviation, internal stress distribution, and surface quality;
[0092] (2) Setting process constraints for the multi-objective optimization model to obtain a constrained optimization problem, and initializing the constrained optimization problem through the particle swarm algorithm to obtain an initial particle swarm;
[0093] (3) Introducing adaptive inertia weight into the initial particle swarm to obtain dynamic inertia weight factor, and updating particle velocity and position according to the dynamic inertia weight factor to obtain a new generation of particle swarm;
[0094] (4) Perform non-dominated sorting on the new generation of particle swarms to obtain the Pareto optimal set, and then perform chaotic perturbation mechanism on the Pareto optimal set to obtain an expanded search space;
[0095] (5) Perform local search in the expanded search space to obtain the optimized Pareto frontier, and conduct a comprehensive evaluation on the optimized Pareto frontier to obtain the comprehensive score of each solution;
[0096] (6) The best solution is selected based on the comprehensive scores of each solution and the target correction strategy is output. The target correction strategy includes the position, magnitude, direction and loading sequence of the force.
[0097] Specifically, the current shape characteristics and stress state of the workpiece are extracted from the real-time state data. The shape deviation can be quantified by measuring the distance between the actual surface of the workpiece and the ideal geometric model. The internal stress distribution reflects the stress state of different areas inside the workpiece and can be obtained through finite element analysis or other engineering mechanics methods. Surface quality involves the smoothness and defect characteristics of the surface, which can be evaluated through surface scanning data. The objective function of the multi-objective optimization model can be expressed as:
[0098] ;
[0099] in, represents the minimization of shape deviation, represents the optimization of internal stress distribution, Indicates an improvement in surface quality. It is a vector of optimization variables, which may include the position of the orthopedic tool, the magnitude and direction of the applied force, etc. Each objective function needs to be defined according to specific engineering requirements. For example, the goal of shape deviation can be defined as minimizing the average distance from the actual surface point to the ideal surface point, the goal of internal stress distribution can be defined as minimizing the stress concentration point, and the surface quality may involve minimizing the surface roughness. Set process constraints when building a multi-objective optimization model. These conditions may include the physical limits of the workpiece material, the operating range of the orthopedic equipment, the processing time limit, etc. These constraints can be expressed as:
[0100] ;
[0101] in, is the constraint function, is the total number of constraints. Solutions that satisfy these constraints are called feasible solutions. The constrained optimization problem is initialized by the particle swarm algorithm. The particle swarm algorithm is a bionic optimization algorithm that searches for the optimal solution by simulating the foraging behavior of bird flocks. In the particle swarm algorithm, each particle represents a possible solution, and the initial particle swarm is generated by randomly distributing particles in the solution space. Each particle has a position and velocity, the position represents the current solution, and the velocity represents the search direction and step size. The generation of the initial particle swarm can be done by random sampling in the solution space. In order to improve the search efficiency and avoid premature convergence, an adaptive inertia weight is introduced to the initial particle swarm. Inertia weight Control the update of particle velocity and adjust it according to the current velocity of the particle and the individual best position and the global best position. The adaptive adjustment of the inertia weight factor can be expressed as:
[0102] ;
[0103] in, and are the maximum and minimum values of the inertia weight, is the maximum number of iterations, is the current iteration number. Through adjustment, the algorithm can maintain a large inertia weight for global search in the early stage of the search, and gradually reduce the inertia weight for local search in the later stage. The new generation of particle swarms is non-dominated sorted to identify the Pareto optimal set. These solutions represent the best solutions to optimize a certain objective without being inferior to other solutions. The Pareto optimal set is processed by a chaotic perturbation mechanism to expand the search space and prevent the algorithm from falling into the local optimum. Chaotic perturbation expands the search range by introducing random changes in the neighborhood of the solution. In the expanded search space, a local search is performed to obtain the optimized Pareto front. By comprehensively evaluating these solutions, the comprehensive score of each solution is calculated. The score can be obtained by weighted summation based on the weight of the objective function or other multi-objective optimization evaluation methods. The best solution is selected according to the comprehensive score, and this best solution is the final target orthopedic strategy. The final output target orthopedic strategy includes detailed information such as the position, magnitude, direction and loading sequence of the force. For example, when correcting a part with a complex surface, the optimal solution may indicate that a force in a certain direction should be applied to a specific area of the part to correct its shape deviation while minimizing stress concentrations inside the part and ensuring that the surface quality meets the requirements.
[0104] In a specific embodiment, the process of executing step S106 may specifically include the following steps:
[0105] (1) Generate the operation instruction sequence of the CNC orthopedic equipment according to the target orthopedic strategy, obtain the orthopedic implementation plan, and execute the orthopedic implementation plan step by step to obtain the phased orthopedic results;
[0106] (2) Perform real-time status monitoring on the staged correction results to obtain real-time status monitoring data, and perform deviation analysis based on the real-time status monitoring data and the predicted results to obtain the correction parameter adjustment amount;
[0107] (3) Adjust the orthopedic parameter adjustment amount to obtain an optimized orthopedic execution plan, and after completing the orthopedic operation, perform a comprehensive three-dimensional scan of the target workpiece to obtain the surface data of the workpiece after correction;
[0108] (4) Performing convolution processing on the surface data of the workpiece after correction to obtain updated surface defect information, and calculating the shape accuracy score based on the updated surface defect information;
[0109] (5) Perform acoustic emission testing on the target workpiece to obtain internal structural integrity data, and perform a multi-index comprehensive evaluation based on the shape accuracy score, updated surface defect information, and internal structural integrity data to generate an orthopedic quality assessment report.
[0110] Specifically, the specific operating parameters in the orthopedic strategy are converted into instructions that can be understood by the CNC device. These instructions usually include positioning commands, movement commands, and force control commands, and the specific instruction format depends on the CNC system used. For example, in a three-axis CNC system, G code is used to define the movement path and applied force of the orthopedic tool. The generation of the instruction sequence takes into account the position, magnitude, direction, and loading sequence of the force specified in the target orthopedic strategy to ensure that the equipment can perform orthopedic operations on the workpiece according to the predetermined path and method. The generated operation instruction sequence is integrated into a complete orthopedic execution plan, and the plan is executed step by step. The execution results of each step are recorded to monitor the progress and effect of the orthopedic process. The staged orthopedic results reflect the shape changes and stress adjustments of the workpiece during the orthopedic process. The staged orthopedic results are monitored in real time. Real-time state monitoring usually obtains the current state data of the workpiece through sensors, which include position, stress, surface quality, etc. The real-time state monitoring data is compared with the predicted results to find any possible deviations. These deviations may be caused by model errors, equipment accuracy limitations, or external interference. The purpose of deviation analysis is to determine the amount of parameter adjustment during the correction process. The parameter adjustment will affect the subsequent operation instructions, and the correction execution plan will be modified accordingly to reflect these adjustments. Through the adjustment operation, the optimized correction execution plan is obtained. After all correction operations are completed, a comprehensive three-dimensional scan of the target workpiece is performed to obtain the surface data of the workpiece after correction. These data are used to evaluate the effect of the correction operation, including the shape deviation and surface features of the workpiece. The surface data of the workpiece after correction is convoluted to extract updated surface defect information. This information includes bumps, scratches or other minor defects on the surface of the workpiece, which can reflect the surface quality after correction. Based on the defect information, the shape accuracy score is calculated, which is used to quantify the geometric accuracy of the workpiece. For example, the shape accuracy score can be calculated using the following formula:
[0111] ;
[0112] in, Indicates The actual deviation value of the measurement point, is the target deviation value, is the maximum permissible deviation, is the total number of measurement points. A value close to 1 for the shape accuracy score indicates that the shape of the workpiece is close to the target shape. Perform acoustic emission testing on the target workpiece. Acoustic emission testing is a non-destructive testing method that can detect defects such as tiny cracks and delamination inside the workpiece. A multi-index comprehensive evaluation is performed by comprehensively considering the shape accuracy score, updated surface defect information, and internal structural integrity data. The evaluation process can use a weighted scoring method to combine different indicators according to their importance to obtain an overall score. Generate an orthopedic quality assessment report to record the various steps of the orthopedic process, real-time monitoring data, deviation analysis results, workpiece data after orthopedic surgery, and the final quality score.
[0113] The above describes the workpiece correction process quality detection and analysis method in the embodiment of the present application. The following describes the workpiece correction process quality detection and analysis device in the embodiment of the present application. Figure 2 In the embodiment of the present application, an embodiment of the workpiece correction process quality detection and analysis device includes:
[0114] The scanning module 201 is used to perform a three-dimensional surface scan on the target workpiece to obtain a three-dimensional data model of the workpiece;
[0115] The processing module 202 is used to perform convolution processing on the three-dimensional data model of the workpiece to obtain the surface defect information of the workpiece;
[0116] The analysis module 203 is used to perform finite element analysis on the three-dimensional data model of the workpiece and the surface defect information of the workpiece to obtain multi-model deformation state information;
[0117] A filtering module 204 is used to perform adaptive interactive multi-model filtering on the multi-model deformation state information to obtain a real-time state estimation result;
[0118] The calculation module 205 is used to perform multi-objective optimization calculation on the real-time state estimation result to obtain the target correction strategy;
[0119] The evaluation module 206 is used to perform an orthopedic operation on the target workpiece according to the target orthopedic strategy and perform a multi-index evaluation to generate an orthopedic quality evaluation report.
[0120] Through the cooperation of the above components, the surface morphology, internal structure and mechanical properties of the workpiece can be analyzed comprehensively by performing three-dimensional scanning and convolution processing on the workpiece surface, combined with finite element analysis. This multi-source information fusion method can more comprehensively and accurately reflect the actual state of the workpiece. The adaptive interactive multi-model filtering technology can effectively deal with the randomness and nonlinear characteristics of the workpiece deformation. By dynamically adjusting the model weights and transition probabilities, the complex deformation state can be accurately estimated, and the adaptability and robustness of the correction process can be improved. Based on the real-time state estimation results, a multi-objective optimization model including shape deviation, internal stress distribution and surface quality is constructed. Through the improved particle swarm algorithm and Pareto optimal set analysis, the intelligent generation of correction strategies is realized, multiple correction targets are balanced, and the comprehensive performance of the correction effect is improved. During the correction execution process, the correction parameters are dynamically adjusted through real-time state monitoring and deviation analysis. This closed-loop control mechanism can effectively deal with the uncertainties in the correction process and improve the accuracy and stability of the correction. A multi-index comprehensive evaluation system is established through comprehensive three-dimensional scanning, surface defect analysis and internal structure integrity detection of the corrected workpiece. From data collection, state estimation to strategy generation and execution, a high degree of intelligence and automation is achieved, which greatly reduces manual intervention, improves orthopedic efficiency, reduces human errors, and ensures the consistency of orthopedic quality.
[0121] The present application also provides a workpiece correction process quality detection and analysis device, the workpiece correction process quality detection and analysis device includes a memory and a processor, the memory stores computer-readable instructions, and when the computer-readable instructions are executed by the processor, the processor executes the steps of the workpiece correction process quality detection and analysis method in the above-mentioned embodiments.
[0122] The present application also provides a computer-readable storage medium, which may be a non-volatile computer-readable storage medium or a volatile computer-readable storage medium, wherein instructions are stored in the computer, and when the instructions are executed on a computer, the computer executes the steps of the workpiece correction process quality detection and analysis method.
[0123] Those skilled in the art can clearly understand that, for the convenience and brevity of description, the specific working processes of the above-described systems, systems and units can refer to the corresponding processes in the aforementioned method embodiments and will not be repeated here.
[0124] If the integrated unit is implemented in the form of a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present application is essentially or the part that contributes to the prior art or all or part of the technical solution can be embodied in the form of a software product. The computer software product is stored in a storage medium, including several instructions to enable a computer device (which can be a personal computer, server, or network device, etc.) to execute all or part of the steps of the method described in each embodiment of the present application. The aforementioned storage medium includes: U disk, mobile hard disk, read-only memory (ROM), random access memory (RAM), disk or optical disk and other media that can store program codes.
[0125] As described above, the above embodiments are only used to illustrate the technical solutions of the present application, rather than to limit them. Although the present application has been described in detail with reference to the aforementioned embodiments, those skilled in the art should understand that they can still modify the technical solutions described in the aforementioned embodiments, or make equivalent replacements for some of the technical features therein. However, these modifications or replacements do not deviate the essence of the corresponding technical solutions from the spirit and scope of the technical solutions of the embodiments of the present application.
Claims
1. A method for detecting and analyzing the quality of a workpiece correction process, characterized in that: The method comprises: Perform three-dimensional scanning on the surface of the target workpiece to obtain a three-dimensional data model of the workpiece; Performing convolution processing on the three-dimensional data model of the workpiece to obtain surface defect information of the workpiece; Finite element analysis is performed on the three-dimensional data model of the workpiece and the surface defect information of the workpiece to obtain multi-model deformation state information; specifically comprising: meshing the three-dimensional data model of the workpiece to obtain an initial finite element mesh model, and locally refining the mesh of the defect area according to the surface defect information of the workpiece to obtain a target finite element mesh model; defining material properties and boundary conditions of the target finite element mesh model, and generating a mechanical model of the target workpiece; constructing an elastic deformation model, a plastic deformation model and a composite deformation model according to the mechanical model to obtain a multi-model deformation description, wherein the composite deformation model combines elasticity and plasticity. Behavior is used to describe the actual deformation process of the workpiece, and perform random field theory analysis on the multi-model deformation description to obtain a random deformation model; perform simulation calculation based on the random deformation model to obtain the probability distribution of the workpiece deformation, and perform principal component analysis on the probability distribution of the workpiece deformation to obtain a target deformation mode; calculate the displacement field of each grid node based on the target deformation mode to obtain the deformation field distribution of the target workpiece, and perform stress-strain analysis on the deformation field distribution to obtain the internal stress distribution of the target workpiece; perform deformation model weight distribution based on the deformation field distribution and the internal stress distribution to obtain multi-model deformation state information; Adaptive interactive multi-model filtering is performed on the deformation state information of the multiple models to obtain a real-time state estimation result; specifically, the method includes: performing state space modeling on the deformation state information of the multiple models to obtain the state equation and observation equation of each model; initializing the cubature Kalman filter for each model according to the state equation and observation equation to obtain an initial state estimation and a first error covariance matrix; introducing multiple fading factors into the first error covariance matrix to obtain a second prediction error covariance, and calculating the filter gain according to the second prediction error covariance to obtain an adaptive filter gain matrix; performing singular value decomposition on the adaptive filter gain matrix to obtain a numerically stable matrix decomposition result, and performing state update and covariance update according to the matrix decomposition result to obtain a filter estimation value of each model; calculating the likelihood function value of the filter estimation value of each model to obtain a model weight factor, and dynamically adjusting the Markov transfer matrix according to the model weight factor to obtain an adaptive model transfer probability; performing model interaction based on the adaptive model transfer probability to obtain a mixed state estimation, and performing weighted fusion on the mixed state estimation to obtain a real-time state estimation result; Performing multi-objective optimization calculation on the real-time state estimation result to obtain a target correction strategy; According to the target orthopedic strategy, an orthopedic operation is performed on the target workpiece and a multi-index evaluation is performed to generate an orthopedic quality evaluation report.
2. The workpiece correction process quality detection and analysis method according to claim 1 is characterized in that: The three-dimensional scanning of the surface of the target workpiece to obtain a three-dimensional data model of the workpiece includes: Performing a three-dimensional scan of the target workpiece from multiple angles and multiple light sources to obtain original point cloud data, and performing Gaussian filtering and statistical outlier filtering on the original point cloud data to obtain preliminary noise-reduced point cloud data; Calculating a normal vector and a curvature according to the preliminary denoised point cloud data to obtain a point cloud feature descriptor, and performing iterative closest point registration on multiple scanning viewpoint point clouds having the point cloud feature descriptors to obtain a fused point cloud in a unified coordinate system; Perform voxel downsampling on the fused point cloud to obtain a uniformly distributed sparse point cloud, construct a kd-tree index structure based on the sparse point cloud, perform a k-nearest neighbor search on each point, and obtain local neighborhood information; Performing moving least square reconstruction on each point based on the local neighborhood information to obtain a continuous and smooth surface model, and performing adaptive meshing on the surface model to obtain a triangular mesh model; The triangular mesh model is topologically optimized and boundary repaired to obtain a three-dimensional data model of the workpiece.
3. The workpiece correction process quality detection and analysis method according to claim 1 is characterized in that: The convolution process is performed on the three-dimensional data model of the workpiece to obtain the surface defect information of the workpiece, including: Normalizing and standardizing the three-dimensional data model of the workpiece to obtain preprocessed input data; Inputting the preprocessed input data into the ResNet-50 backbone network in the preset Double-Head Region-RCNN network for feature extraction to obtain an initial feature map; Performing context feature enhancement on the initial feature map through a Transformer encoder module to obtain enhanced global context features; According to the enhanced global context features, the convolution kernel is dynamically adjusted to perform a deformable convolution operation to obtain adaptive local features; Based on the feature pyramid network of content-aware feature recombination, multi-scale feature fusion is performed on the adaptive local features to obtain multi-scale fusion features; According to the multi-scale fusion features, generating candidate regions for the scale distribution of defects on the workpiece surface in a region proposal network; Performing a RoI Align operation on the candidate region to obtain a feature map of a fixed size, and inputting the feature map of the fixed size into a classification head and a regression head for processing to obtain a category probability and a bounding box coordinate of the defect; Non-maximum suppression processing is performed on the category probability and bounding box coordinates of the defect to obtain workpiece surface defect information, wherein the workpiece surface defect information includes the location, size and type of the defect.
4. The workpiece correction process quality detection and analysis method according to claim 1 is characterized in that: The performing of multi-objective optimization calculation on the real-time state estimation result to obtain a target correction strategy includes: Constructing an orthopedic objective function according to the real-time state estimation result to obtain a multi-objective optimization model including shape deviation, internal stress distribution and surface quality; Setting process constraints on the multi-objective optimization model to obtain a constrained optimization problem, and initializing the constrained optimization problem through a particle swarm algorithm to obtain an initial particle swarm; Introducing an adaptive inertia weight into the initial particle swarm to obtain a dynamic inertia weight factor, and updating the particle speed and position according to the dynamic inertia weight factor to obtain a new generation of particle swarm; Performing non-dominated sorting on the new generation particle swarm to obtain a Pareto optimal set, and performing a chaotic perturbation mechanism on the Pareto optimal set to obtain an expanded search space; Performing a local search in the expanded search space to obtain an optimized Pareto front, and performing a comprehensive evaluation on the optimized Pareto front to obtain a comprehensive score of each solution; The best solution is selected according to the comprehensive scores of the solutions and a target orthopedic strategy is output, wherein the target orthopedic strategy includes the position, magnitude, direction and loading sequence of the force.
5. The workpiece correction process quality detection and analysis method according to claim 1 is characterized in that: The performing of an orthopedic operation on the target workpiece according to the target orthopedic strategy and performing a multi-index evaluation to generate an orthopedic quality evaluation report includes: Generate an operation instruction sequence of a numerically controlled orthopedic device according to the target orthopedic strategy to obtain an orthopedic implementation plan, and execute the orthopedic implementation plan step by step to obtain a phased orthopedic result; Performing real-time status monitoring on the staged correction results to obtain real-time status monitoring data, and performing deviation analysis based on the real-time status monitoring data and the prediction results to obtain the correction parameter adjustment amount; An adjustment operation is performed on the orthopedic parameter adjustment amount to obtain an optimized orthopedic execution plan, and after the orthopedic operation is completed, a comprehensive three-dimensional scan is performed on the target workpiece to obtain surface data of the workpiece after the orthopedic operation; Performing convolution processing on the corrected workpiece surface data to obtain updated surface defect information, and calculating a shape accuracy score based on the updated surface defect information; The target workpiece is subjected to acoustic emission testing to obtain internal structure integrity data, and a multi-index comprehensive evaluation is performed based on the shape accuracy score, the updated surface defect information and the internal structure integrity data to generate an orthopedic quality assessment report.
6. A device for detecting and analyzing the quality of a workpiece correction process, characterized in that: The device is used to perform the workpiece correction process quality detection and analysis method as described in any one of claims 1 to 5, comprising: A scanning module is used to perform three-dimensional scanning on the surface of a target workpiece to obtain a three-dimensional data model of the workpiece; A processing module, used for performing convolution processing on the three-dimensional data model of the workpiece to obtain surface defect information of the workpiece; The analysis module is used to perform finite element analysis on the three-dimensional data model of the workpiece and the surface defect information of the workpiece to obtain multi-model deformation state information; specifically comprising: meshing the three-dimensional data model of the workpiece to obtain an initial finite element mesh model, and locally refining the mesh of the defect area according to the surface defect information of the workpiece to obtain a target finite element mesh model; defining the material properties and boundary conditions of the target finite element mesh model, and generating a mechanical model of the target workpiece; constructing an elastic deformation model, a plastic deformation model and a composite deformation model according to the mechanical model to obtain a multi-model deformation description, wherein the composite deformation model combines elastic deformation and plastic deformation. and plastic behavior, which are used to describe the actual deformation process of the workpiece, and perform random field theory analysis on the multi-model deformation description to obtain a random deformation model; perform simulation calculation based on the random deformation model to obtain the probability distribution of the workpiece deformation, and perform principal component analysis on the probability distribution of the workpiece deformation to obtain a target deformation mode; calculate the displacement field of each grid node based on the target deformation mode to obtain the deformation field distribution of the target workpiece, and perform stress-strain analysis on the deformation field distribution to obtain the internal stress distribution of the target workpiece; perform deformation model weight distribution based on the deformation field distribution and the internal stress distribution to obtain multi-model deformation state information; A filtering module is used to perform adaptive interactive multi-model filtering on the multi-model deformation state information to obtain a real-time state estimation result; specifically comprising: performing state space modeling on the multi-model deformation state information to obtain a state equation and an observation equation for each model; initializing a cubature Kalman filter for each model according to the state equation and the observation equation to obtain an initial state estimation and a first error covariance matrix; introducing multiple fading factors into the first error covariance matrix to obtain a second prediction error covariance, and calculating a filter gain according to the second prediction error covariance to obtain an adaptive filter gain matrix; performing singular value decomposition on the adaptive filter gain matrix to obtain a numerically stable matrix decomposition result, and performing state update and covariance update according to the matrix decomposition result to obtain a filter estimation value of each model; calculating a likelihood function value for the filter estimation value of each model to obtain a model weight factor, and dynamically adjusting the Markov transfer matrix according to the model weight factor to obtain an adaptive model transfer probability; performing model interaction based on the adaptive model transfer probability to obtain a mixed state estimation, and performing weighted fusion on the mixed state estimation to obtain a real-time state estimation result; A calculation module, used for performing multi-objective optimization calculation on the real-time state estimation result to obtain a target correction strategy; An evaluation module is used to perform an orthopedic operation on the target workpiece according to the target orthopedic strategy and conduct a multi-index evaluation to generate an orthopedic quality evaluation report.
7. A workpiece correction process quality detection and analysis equipment, characterized in that: The workpiece correction process quality detection and analysis device comprises: a memory and at least one processor, wherein the memory stores instructions; The at least one processor calls the instructions in the memory to enable the workpiece correction process quality detection and analysis device to perform the workpiece correction process quality detection and analysis method according to any one of claims 1 to 5.
8. A computer-readable storage medium having instructions stored thereon, characterized in that: When the instructions are executed by the processor, the workpiece correction process quality detection and analysis method according to any one of claims 1 to 5 is implemented.
Citation Information
Patent Citations
Workpiece deformation correction method based on support vector machine
CN104416017A
Annular part deformation correction method and system
CN114818172A
Visual life management method and system for processing part, terminal and storage medium
CN115292964A
