Vision-based workpiece machining precision control method

Abstract

The invention discloses a workpiece machining precision control method based on vision, and relates to the technical field of vision recognition. The control method comprises the following steps: 1, deploying a multi-view visual system, collecting a multi-view original image of a target workpiece, and executing a preprocessing action to generate a three-dimensional point cloud model; 2, extracting key geometric features based on the three-dimensional point cloud model; according to the technical scheme, a function application model fusing geometric accuracy, point cloud regularity and machining parameter constraint is constructed, measurement robustness and process feasibility are comprehensively considered, accuracy is guaranteed, cutter service life and energy consumption control are also considered, and the reasonability of compensation decision is remarkably improved; a proxy model based on historical data is introduced, the weight of an error model is updated after compensation each time, the maximum number of iterations is limited to prevent oscillation, dynamic optimization of a machining strategy is achieved, and it is ensured that the effects of high precision and high stability are continuously kept in long-term operation.

Description

TECHNICAL FIELD

[0001] The present application relates to the field of visual recognition technology, in particular to a workpiece machining precision control method based on vision. BACKGROUND

[0002] Visual recognition is a technology that computers acquire image or video data of target objects through image acquisition devices such as cameras, automatically analyze and understand the visual information in the data by using image processing, pattern recognition and machine learning, and realize detection, positioning, classification, measurement or state judgment of objects. In the industrial field, existing visual recognition technology is usually based on 2D images or 3D point clouds, and combines edge detection, template matching, feature extraction or deep learning models to complete tasks. For example, in workpiece quality inspection, the system can identify whether the hole position is offset, whether the edge is burr, and whether the size is out of tolerance, etc. Traditional methods rely on high-contrast lighting and fixed background, and have poor adaptability to reflection, occlusion or complex curved surfaces. In recent years, structured light, laser scanning and multi-view fusion technology have improved the accuracy of three-dimensional reconstruction, enabling visual recognition to gradually move from qualitative judgment to quantitative measurement. However, most existing systems still remain in the stage of offline sampling inspection or one-way detection, and lack deep closed-loop linkage with machining control systems, limiting their application in high-precision adaptive manufacturing.

[0003] In the current industrial field of workpiece machining, traditional visual detection systems mostly exist as independent quality inspection units, lacking real-time data interaction with numerical control machining equipment. For example, when a certain parts factory is machining engine cylinder bodies, although it is equipped with high-precision cameras for hole detection, the detection results are only used for final inspection and sorting, and cannot be dynamically corrected in the tool path during part of the machining process, resulting in some workpieces having to be scrapped or reworked. At the same time, existing error judgment mostly relies on single size comparison, such as only comparing the difference between the measured diameter and the CAD value, without considering the noise interference of point clouds or the instantaneous deformation caused by sudden changes in cutting force, which can easily lead to misjudgment or excessive compensation, thereby exacerbating tool wear. In addition, when switching between machining materials or when the tool has been in use for a long time, the system cannot automatically adjust the recognition threshold and compensation strategy, and still requires manual parameter resetting by workers, which seriously affects the machining efficiency of the workpiece production line and restricts the good development of high-precision intelligent manufacturing. SUMMARY

[0004] To achieve the above purpose, the present application is implemented by the following technical solutions: The workpiece machining precision control method based on vision includes the following steps: Step one: deploy a multi-view visual system, acquire multi-view original images of the target workpiece, and perform a pre-processing action to generate a three-dimensional point cloud model; Step two: based on the three-dimensional point cloud model, extract key geometric features and compare them with the CAD design model, introduce a pre-constructed function application model, and generate an error scalar; Step three: synchronously trigger the optimization hybrid strategy when the error scalar is detected, solve new machining parameters, and generate a compensation vector according to the new machining parameters to complete the compensation machining action, and issue it to the numerical control system through the preset target protocol; Step four: after each compensation machining action is completed, the iteration cycle is realized again after step one is triggered; if the error scalar in the newly generated three-dimensional point cloud model is reduced to below the standard threshold, it is determined to be qualified; otherwise, the compensation machining action is continued until it is determined to be qualified.

[0005] Further, the deployed multi-view vision system is: a plurality of industrial cameras, and a structured light projector; wherein the layout of the plurality of industrial cameras completely covers all key feature surfaces of the target workpiece, and joint internal and external parameter calibration is performed through a calibration board, and the key feature surfaces at least include: hole positions, edges and curved surfaces.

[0006] Further, the pre-processing action content includes: denoising, distortion correction, image registration and point cloud fusion to realize the final generation of the three-dimensional point cloud model , wherein p i ∈R 3 ; R is a real set, representing a sampling point in space, i=1, 2,..,N, N is a positive integer, and i represents the number of the sampling point.

[0007] Further, the extracted key geometric features at least include: diameter, flatness, profile and curvature.

[0008] Further, the application process of the function application model is as follows: S201, according to the feature extraction operator and the ideal geometric feature set in the CAD design model, a geometric distance metric sub-function D(Fr(P), Gr) is established; wherein the geometric distance metric sub-function adopts any one of Hausdorff distance, ICP residual and least squares fitting error, Fr(P) is the feature extraction operator; Gr is the ideal geometric feature set in the CAD design model; S202, extract the curvature in the key geometric features, and based on the smoothness constraint of the curvature change, configure the point cloud regularity penalty term Rt(P); S203, construct a constraint function based on the machining parameters: Cj(x); wherein Cj(x) is a proxy model for predicting tool wear based on machining parameters x, x∈R m : machining parameter vector, at least including: spindle speed, feed speed and tool path offset; S204, according to the results corresponding to the geometric distance metric sub-function, the point cloud regularity penalty term and the constraint function based on the machining parameters, weighted calculation is performed to obtain the error scalar Et(x).

[0009] Furthermore, the triggered optimization hybrid strategy adopts a hybrid strategy of gradient-based iterative optimization combined with historical data priors; The process of optimizing the hybrid strategy is as follows: S301. Calculate the error scalar ∇Et(x) using the finite difference method, that is, the sensitivity of the error to each processing parameter; S302. Introduce a machining knowledge base and construct a lightweight surrogate model to predict the error scalar Et(x) under different machining parameters x. The machining knowledge base includes at least: historical successful cases, material and tool matching tables, and machine tool dynamics models. The lightweight surrogate model adopts Gaussian process regression.

[0010] Furthermore, after the optimization of the hybrid strategy is completed, the new processing parameter x_0 is solved: x_0 = argmin x Et(x); In the formula, argmin x Let x represent the machining parameter that minimizes the error scalar E(x).

[0011] Furthermore, a compensation vector Δx is generated based on the new processing parameter x_0: Δx = x_0 - x_cu; where x_cu represents the processing parameter vector currently in use; the target protocol adopts either OPC UA or MTConnect protocol.

[0012] Furthermore, a maximum of 3 iterations are allowed in the compensation processing, and the standard threshold is ±5.

[0013] A vision-based workpiece machining accuracy control system, comprising: Point cloud construction module: Deploy a multi-view vision system to acquire multi-view original images of the target workpiece and perform preprocessing actions to generate a 3D point cloud model; Error generation module: Based on the 3D point cloud model, extract key geometric features and compare them with the CAD design model. Introduce a pre-built function application model to generate an error scalar. Compensation and maintenance module: When an error scalar is detected, an optimized hybrid strategy is triggered synchronously to solve for new machining parameters and generate a compensation vector based on the new machining parameters to complete the compensation machining action. The vector is then sent to the CNC system through a preset target protocol. Iterative update module: After each compensation processing action is completed, step one is triggered again to implement the iterative loop; if the error scalar in the newly generated 3D point cloud model drops below the standard threshold, it is judged as qualified; otherwise, the compensation processing action continues until it is judged as qualified.

[0014] This invention provides a vision-based method for controlling workpiece machining accuracy, which has the following advantages: (1) This solution deploys a multi-view vision system and precisely calibrates it with the machine tool coordinate system to realize online three-dimensional reconstruction of the target workpiece after processing, solves the problem of disconnect between perception and execution, realizes the real-time perception of processing status, and achieves the effect of obtaining complete geometric error information within a single processing cycle, providing a reliable and effective data basis for the compensation action proposed in subsequent steps; (2) This scheme constructs a function application model that integrates geometric accuracy, point cloud regularity and machining parameter constraints. It comprehensively considers measurement robustness and process feasibility, ensuring accuracy while also taking into account tool life and energy consumption control, and significantly improving the rationality of compensation decision. (3) This scheme introduces a proxy model based on historical data, updates the error model weights after each compensation, and limits the maximum number of iterations to prevent oscillations, thereby realizing dynamic optimization of the processing strategy and ensuring that it maintains high precision and high stability in long-term operation. Attached Figure Description

[0015] Figure 1 This is a schematic diagram of the overall process of the vision-based workpiece machining accuracy control method in this invention. Detailed Implementation

[0016] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those of ordinary skill in the art without creative effort are within the scope of protection of the present invention.

[0017] Example 1: Please see Figure 1 This embodiment provides a vision-based workpiece machining accuracy control method. The specific steps of the method are as follows: Step 1: Deploy a multi-view vision system, acquire multi-view original images of the target workpiece, and perform preprocessing actions to generate a three-dimensional point cloud model. Step 1 is the starting point of the entire control process, and its core objective is to obtain high-quality, high-fidelity three-dimensional geometric information of the workpiece surface to provide reliable input for subsequent error analysis.

[0018] The specific explanation for step one is as follows: The deployed multi-view vision system consists of several industrial-grade high-resolution cameras, using 5-megapixel camera components, and a structured light projector or laser contour scanning module to form the required multi-view vision system. The camera layout must cover all key feature surfaces of the target workpiece, and joint intrinsic and extrinsic parameter calibration is performed using a calibration board to ensure that the coordinate systems of each viewpoint are consistently aligned to the world coordinate system of the machine tool where the target workpiece is located. Key feature surfaces include at least: holes, edges, and curved surfaces. Additionally, positioning fixtures can be configured during the acquisition of original multi-view images to prevent image misalignment caused by target workpiece vibration or machine tool movement. For potentially reflective or dark surfaces, polarizing filters or multi-frequency phase-shifting structured light technology are introduced as needed to suppress specular reflection interference and improve the signal-to-noise ratio.

[0019] The preprocessing actions performed include: Denoising, distortion correction, image registration, and point cloud fusion are performed to ultimately generate a high-density, low-noise 3D point cloud model. , where p i ∈R 3 R is a set of real numbers, representing a sampling point in space; in this embodiment, nonlocal mean filtering is used for denoising, Brown-Conrady model-based distortion correction is used for distortion correction, ICP or SIFT feature matching can be used for image registration, and Poisson reconstruction or TSDF voxel fusion is used for point cloud fusion; i = 1, 2, ..., N, where N is a positive integer and i represents the number of the sampling point; Extended Explanation: The key to Step 1 is that data quality determines the upper limit of all subsequent steps. If the point cloud has holes, excessive noise, or registration errors, it will lead to distorted feature extraction, which in turn will trigger error compensation. Therefore, this step introduces an adaptive exposure and gain control algorithm to automatically adjust the light source intensity and camera parameters according to the material of the target workpiece, ensuring that target workpieces of different batches and materials can obtain consistent imaging quality. In addition, through a GPU-accelerated parallel processing pipeline, the time for generating a point cloud in a single acquisition is compressed to less than 200ms, thereby meeting the requirements for online real-time performance.

[0020] In traditional machining processes, visual inspection and CNC machining systems are often independent of each other, resulting in the inability to promptly feed back detected errors for correction of machining actions. This solution, however, deploys a multi-view vision system and precisely calibrates it with the machine tool coordinate system, enabling online 3D reconstruction of the target workpiece after machining. This solves the problem of the disconnect between perception and execution, achieving real-time perception of the machining status. Consequently, it achieves the effect of obtaining complete geometric error information within a single machining cycle, providing a reliable and effective data foundation for the compensation actions proposed in subsequent steps.

[0021] Step 2: Based on the 3D point cloud model, extract key geometric features and compare them with the CAD design model. Introduce a pre-built function application model to generate an error scalar, which is used to quantify the actual processing deviation. The specific details of step two are as follows: Regarding the extraction of key geometric features, these features should include at least: diameter, flatness, contour, and curvature. The application process of the function application model is as follows: S201. Based on the feature extraction operator and the ideal geometric feature set in the CAD design model, establish a geometric distance metric function: D(Fr(P), Gr); where the geometric distance metric function adopts any one of Hausdorff distance, ICP residual, and least squares fitting error; Fr(P) is the feature extraction operator, i.e., the key geometric feature set; Gr is the ideal geometric feature set in the CAD design model, which corresponds one-to-one with each key geometric feature in the key geometric feature set; P is the actual target workpiece point cloud generated in step one, i.e., the three-dimensional point cloud model. S202. Extract the curvature from the key geometric features and configure the point cloud regularity penalty term Rt(P) based on the smoothness constraint of the curvature change; wherein the point cloud regularity penalty term is used to suppress outliers or noise interference. S203. Construct a constraint function based on machining parameters: Cj(x); where Cj(x) is a surrogate model that predicts cutting force, tool wear, or energy consumption based on machining parameter x. In this embodiment, either a regression function or a neural network is used to constrain the rationality of the process during optimization, preventing excessive tool wear or excessive energy consumption in pursuit of accuracy; x∈R m The machining parameter vector should include at least the following: spindle speed, feed rate, and toolpath offset. Therefore, the value of m at this point is 3; S204. Based on the results corresponding to the geometric distance metric function, the point cloud regularity penalty term, and the constraint function based on the processing parameters, a weighted calculation is performed to obtain the error scalar Et(x). The weights used in the weighted calculation in S204 are used to balance accuracy, robustness, and processing efficiency. They can be dynamically adjusted through online learning, and the value range can be set between 0 and 1. It should be noted that the application model of this function does not simply calculate the deviation between the actual and the design, but constructs a multi-objective optimization framework. Among them, the geometric distance metric sub-function serves as the core accuracy term, directly reflecting the machining error; the point cloud regularity penalty term serves as a regularization term to prevent overfitting compensation due to point cloud noise; and the constraint function based on machining parameters introduces process constraints to avoid sacrificing tool life or machining efficiency in pursuit of ultimate accuracy. Specifically, step two relies on the high-quality point cloud from step one and provides a basis for compensation decisions in the subsequent step three. More importantly, the function application model used in step two is scalable and universal. Even if the processing object changes, the ideal geometric feature set Gr and weight coefficients in the CAD design model can be updated according to actual needs, without having to rebuild the overall control scheme or system as in the traditional method.

[0022] Existing error assessment methods typically focus only on the simple deviation between actual dimensions and design values, easily ignoring factors such as point cloud noise, process constraints, and machining costs, thus leading to overcompensation or ineffective adjustments. However, this solution constructs a function application model that integrates geometric accuracy, point cloud regularity, and machining parameter constraints, comprehensively considering measurement robustness and process feasibility. It ensures accuracy while also taking into account tool life and energy consumption control, significantly improving the rationality of compensation decisions.

[0023] Step 3: When an error scalar is detected, the optimization hybrid strategy is triggered synchronously to solve for new machining parameters and generate a compensation vector based on the new machining parameters to complete the compensation machining action. The vector is then sent to the CNC system through a preset target protocol to complete the machining of the target workpiece. The specific details of step three are as follows: In this embodiment, the triggered optimization hybrid strategy is a hybrid strategy combining gradient-based iterative optimization with prior historical data; the specific process of optimizing the hybrid strategy is as follows: S301. Calculate the error scalar ∇Et(x) using the finite difference method, that is, the sensitivity of the error to each processing parameter; For example, if a small change in feed rate is found to have an impact on the aperture error, it will be given a higher weight in the optimization; a small change means that the change in feed rate is within the preset standard threshold range; S302. Introduce a machining knowledge base and construct a lightweight surrogate model to predict the error scalar Et(x) under different machining parameters x, thereby accelerating convergence. The machining knowledge base includes at least: historical successful cases, material and tool matching tables, and machine tool dynamics models. In this embodiment, the lightweight surrogate model adopts Gaussian process regression.

[0024] Once the optimized hybrid strategy has run, the optimization operation is complete; solve for the new processing parameter x_0: x_0=argmin x Et(x); where argmin x Let x be the x that makes E(x) reach its minimum value.

[0025] A compensation vector Δx is generated based on the new machining parameters: Δx = x_0 - x_cu; where x_cu represents the current machining parameter vector, i.e., the parameters before compensation; the target protocol adopts either OPC UA or MTConnect protocol; the compensation machining actions include at least: tool path offset for position errors and spindle speed fine adjustment for surface roughness; to ensure safety, all compensation instructions must undergo feasibility verification; feasibility verification refers to verifying the generated compensation instructions against machine tool physical constraints to ensure that they do not exceed hardware safety limits such as travel range, maximum feed rate, spindle speed, and motor torque, to prevent equipment overload or collision.

[0026] In this embodiment, the CNC system is a processing device for machining the target workpiece.

[0027] Step 4: After each compensation processing action is completed, the visual acquisition in Step 1 is triggered again to achieve iterative loop; if the error scalar Et(x) in the newly generated 3D point cloud model drops below the standard threshold, it is considered qualified; otherwise, the compensation processing action continues until it is considered qualified. The compensation process is allowed to have a maximum of 3 iterations to prevent oscillations; the standard threshold is ±5.

[0028] In addition, this scheme can also introduce a self-evolution mechanism: the data pairs of input machining parameters x and output error scalar Et(x) are stored in the experience database, and the weights in the error model or the surrogate model are fine-tuned or updated with new data according to the set period. For example, when the target workpiece is changed to aluminum alloy, it is found after adopting the above scheme that the influence of the constraint function Cj(x) based on the machining parameters on tool wear is underestimated. Then, the corresponding weight is automatically increased according to the set step value, so that subsequent optimization pays more attention to tool protection. This online learning capability can adapt to dynamic factors such as material variation, tool aging, and environmental temperature drift.

[0029] Traditional vision-assisted machining systems often lack adaptive capabilities, requiring manual parameter readjustment when faced with material changes, tool wear, or environmental changes. This solution introduces a proxy model based on historical data and an online learning mechanism, updating the error model weights after each compensation and limiting the maximum number of iterations to prevent oscillations. This achieves dynamic optimization and autonomous evolution of the machining strategy, ensuring high precision and stability during long-term operation.

[0030] Example 2: Based on Example 1, this embodiment also provides a vision-based workpiece machining accuracy control system, which includes: Point cloud construction module: Deploy a multi-view vision system to acquire multi-view original images of the target workpiece and perform preprocessing actions to generate a 3D point cloud model; Error generation module: Based on the 3D point cloud model, extract key geometric features and compare them with the CAD design model. Introduce a pre-built function application model to generate an error scalar. Compensation and maintenance module: When an error scalar is detected, an optimized hybrid strategy is triggered synchronously to solve for new machining parameters and generate a compensation vector based on the new machining parameters to complete the compensation machining action. The vector is then sent to the CNC system through a preset target protocol. Iterative update module: After each compensation processing action is completed, step one is triggered again to implement the iterative loop; if the error scalar in the newly generated 3D point cloud model drops below the standard threshold, it is judged as qualified; otherwise, the compensation processing action continues until it is judged as qualified.

[0031] The above embodiments can be implemented, in whole or in part, by software, hardware, firmware, or any other combination thereof. When implemented in software, the above embodiments can be implemented, in whole or in part, as a computer program product. Those skilled in the art will recognize that the units and algorithm steps of the various examples described in conjunction with the embodiments disclosed herein can be implemented in electronic hardware, or a combination of computer software and electronic hardware. Whether these functions are implemented in hardware or software depends on the specific application and design constraints of the technical solution.

[0032] The units described as separate components may or may not be physically separate. The components shown as units may or may not be physical units; they may be located in one place or distributed across multiple network units. Some or all of the units can be selected to achieve the purpose of this embodiment, depending on actual needs.

[0033] The above description is merely a specific embodiment of this application, but the scope of protection of this application is not limited thereto. Any changes or substitutions that can be easily conceived by those skilled in the art within the scope of the technology disclosed in this application should be included within the scope of protection of this application.

Claims

1. A vision-based method for controlling workpiece machining accuracy, characterized in that: Includes the following steps: Step 1: Deploy a multi-view vision system to acquire original images of the target workpiece from multiple perspectives and perform preprocessing actions to generate a 3D point cloud model; Step 2: Based on the 3D point cloud model, extract key geometric features and compare them with the CAD design model. Introduce a pre-built function application model to generate an error scalar. Step 3: When an error scalar is detected, the optimization hybrid strategy is triggered synchronously to solve for new machining parameters and generate a compensation vector based on the new machining parameters to complete the compensation machining action. The vector is then sent to the CNC system through a preset target protocol. Step 4: After each compensation processing action is completed, Step 1 is triggered again to implement the iterative loop; if the error scalar in the newly generated 3D point cloud model drops below the standard threshold, it is considered qualified; otherwise, the compensation processing action continues until it is considered qualified.

2. The vision-based workpiece machining accuracy control method according to claim 1, characterized in that: The deployed multi-view vision system consists of several industrial-grade cameras, along with a structured light projector. The layout of the industrial-grade cameras completely covers all the key feature surfaces of the target workpiece, and the internal and external parameters are calibrated together through a calibration plate to be aligned to the world coordinate system of the machine tool where the target workpiece is located. The key feature surfaces include at least: holes, edges, and curved surfaces.

3. The vision-based workpiece machining accuracy control method according to claim 1, characterized in that: The preprocessing steps include denoising, distortion correction, image registration, and point cloud fusion to ultimately generate a 3D point cloud model. , where p i ∈R 3 R is the set of real numbers, representing a sampling point in space, i = 1, 2, ..., N, where N is a positive integer and i represents the number of the sampling point.

4. The vision-based workpiece machining accuracy control method according to claim 1, characterized in that: The key geometric features extracted include at least: diameter, flatness, contour, and curvature.

5. The vision-based workpiece machining accuracy control method according to claim 1, characterized in that: The application process of the function application model is as follows: S201. Based on the feature extraction operator and the ideal geometric feature set in the CAD design model, establish a geometric distance metric function: D(Fr(P), Gr); where the geometric distance metric function adopts any one of Hausdorff distance, ICP residual, and least squares fitting error, Fr(P) is the feature extraction operator, and Gr is the ideal geometric feature set in the CAD design model; S202. Extract the curvature from the key geometric features and configure the point cloud regularity penalty term Rt(P) based on the smoothness constraint of curvature changes; S203. Construct a constraint function based on machining parameters: Cj(x); where Cj(x) is a surrogate model for predicting tool wear based on machining parameter x, x∈R m The machining parameter vector includes at least the spindle speed, feed rate, and toolpath offset; S204, the error scalar Et(x) is obtained by weighting the results corresponding to the geometric distance metric function, point cloud regularity penalty term, and constraint function based on the machining parameters.

6. The vision-based workpiece machining accuracy control method according to claim 1, characterized in that: The triggered optimization hybrid strategy is a hybrid strategy that combines gradient-based iterative optimization with prior historical data. The process of optimizing the hybrid strategy is as follows: S301. Calculate the error scalar ∇Et(x) using the finite difference method, that is, the sensitivity of the error to each processing parameter; S302. Introduce a machining knowledge base and construct a lightweight surrogate model to predict the error scalar Et(x) under different machining parameters x. The machining knowledge base includes at least: historical successful cases, material and tool matching tables, and machine tool dynamics models. The lightweight surrogate model adopts Gaussian process regression.

7. The vision-based workpiece machining accuracy control method according to claim 6, characterized in that: After the optimized hybrid strategy has run, the new processing parameter x_0 is solved: x_0 = argmin x Et(x); In the formula, argmin x Let x represent the machining parameter that minimizes the error scalar E(x).

8. The vision-based workpiece machining accuracy control method according to claim 1, characterized in that: A compensation vector Δx is generated based on the new processing parameter x_0: Δx = x_0 - x_cu; where x_cu represents the processing parameter vector currently in use; the target protocol is either OPC UA or MTConnect protocol.

9. The vision-based workpiece machining accuracy control method according to claim 1, characterized in that: The compensation process is allowed a maximum of 3 iterations, with a standard threshold of ±5.

Images