Silicon carbide non-pressure cylinder surface finishing and grinding equipment and method

Abstract

The invention discloses silicon carbide non-pressure barrel surface finishing and grinding equipment and method, and belongs to the technical field of precision ceramic machining. The method comprises the following steps: cooperatively obtaining a high-precision three-dimensional point cloud on the surface of a cylinder; macroscopic geometric deviation identification based on three-dimensional registration and microdefect intelligent identification based on a PointNet + + deep learning model are executed in parallel, and a structured defect feature set is generated; according to the defect features, a layered contour grinding instruction for macroscopic protrusions and a fixed-point spiral grinding instruction for microscopic pore cracks are generated through self-adaptive planning; and when the machining instruction is executed, the posture of the barrel body is monitored in real time through a high-speed vision sensing system, the movement track of the grinding head is dynamically adjusted in a closed-loop mode, and system errors are compensated. By constructing an intelligent closed loop of perception-decision-execution-compensation, integrated accurate identification and self-adaptive high-precision grinding of macro and micro composite defects on the surface of the silicon carbide cylinder are achieved, the technical problems that a traditional method is not accurate in identification, the strategy lacks adaptability, and errors cannot be compensated in real time are effectively solved, and the method is suitable for large-scale popularization and application. And the machining precision and the automation level are obviously improved.

Description

Technical Field

[0001] This application relates to the field of silicon carbide pressureless cylinder technology, and in particular to a silicon carbide pressureless cylinder surface finishing grinding equipment and method. Background Technology

[0002] Pressureless sintered silicon carbide ceramics are widely used in key components of high-end equipment due to their excellent properties such as high hardness, high temperature resistance, and corrosion resistance. However, once formed, their surfaces are prone to both macroscopic geometric protrusions and depressions and microscopic material defects such as pores and cracks. Furthermore, these materials have extremely high hardness (HRA90-95), making it difficult for traditional finishing grinding techniques to balance defect identification accuracy with processing adaptability. Current technologies lack an integrated solution that can simultaneously and accurately capture macroscopic and microscopic composite defects and dynamically adapt processing strategies. Most methods employ grinding with fixed process parameters or rely on manual defect identification, failing to adjust the grinding path and process parameters according to the specific type, location, and geometric parameters of the defects. This results in incomplete correction of macroscopic protrusions and inadequate grinding of microscopic defects. Moreover, systemic errors such as clamping deviations and vibrations during processing cannot be compensated for in real time. Ultimately, the surface machining accuracy of the cylinder is difficult to exceed 0.01mm, failing to meet the stringent requirements of high-end equipment for silicon carbide cylinder surface quality and severely limiting its application in high-precision scenarios.

[0003] To address the problems mentioned above, a silicon carbide pressureless cylinder surface finishing grinding equipment and method are invented. Summary of the Invention

[0004] To address the problems of inaccurate identification of macro- and micro-complex defects, lack of adaptability in processing strategies, and inability to compensate for system errors in real time in existing silicon carbide pressureless cylinder surface finishing grinding technology, which leads to insufficient processing accuracy, this invention aims to provide a silicon carbide pressureless cylinder surface finishing grinding device and method. By constructing a closed-loop system of "perception-decision-execution-compensation", it achieves collaborative and accurate identification of macro- and micro-defects, adaptive processing planning driven by defect features, and real-time error compensation during the processing, thereby improving the accuracy, automation level, and adaptability of silicon carbide pressureless cylinder surface finishing and meeting the high-precision surface quality requirements of high-hardness silicon carbide cylinders.

[0005] The silicon carbide pressureless cylinder surface finishing grinding method provided in this application adopts the following technical solution, including the following steps: S1: Collaborative data acquisition steps: Acquire high-precision three-dimensional point cloud data of the outer surface of the cylinder blank; S2: Collaborative identification and parameterized characterization steps: Based on the three-dimensional point cloud data, macroscopic geometric deviation identification and microscopic material defect intelligent identification are performed in parallel, and a defect feature set containing location, type and geometric parameters is output; S3: Adaptive planning step: Based on the defect feature set, generate composite machining instruction sets with different grinding strategies and process parameters for macroscopic protrusion defects and microscopic pores or cracks, respectively. S4: Closed-loop execution steps: Control the grinding head to execute the composite machining instruction set, and at the same time monitor the cylinder position and dynamically adjust the grinding head movement trajectory in real time through the vision sensing system to compensate for system errors in a closed-loop manner.

[0006] Optionally, in the collaborative identification and parameterized characterization steps, the macroscopic geometric deviation identification is achieved by performing three-dimensional registration between the three-dimensional point cloud and the design CAD model, and outputting the normal deviation distribution; the microscopic material defect intelligent identification is achieved by directly inputting the three-dimensional point cloud into the trained PointNet++ deep learning model, and outputting the defect category, three-dimensional coordinates and point cloud segmentation results.

[0007] Optionally, in the adaptive planning step, the specific method for generating the composite processing instruction set is as follows: For macroscopic protrusion defects, a layered contour grinding command is generated, and the grinding depth of a single layer is adaptively set based on the normal deviation distribution. For microscopic pores or cracks, a fixed-point spiral grinding command is generated, and the radius and depth of the spiral trajectory are adaptively determined based on the three-dimensional coordinates of the defect and the point cloud segmentation results.

[0008] Optionally, the single-layer grinding depth of the layer contour grinding command is set between 0.02 mm and 0.05 mm, and the overlap rate of adjacent grinding paths is not less than 30%.

[0009] Optionally, the single grinding depth increment of the fixed-point spiral grinding command is no more than 0.005 mm.

[0010] Optionally, in the closed-loop execution step, the visual sensing system samples at a frequency of not less than 100Hz, the dynamically adjusted closed-loop control cycle is not greater than 20ms, and the trajectory tracking accuracy is better than 0.01mm.

[0011] Optionally, the single-point measurement accuracy of the three-dimensional point cloud data is better than 0.005 mm.

[0012] Optionally, the defect feature set is organized in a structured data format to drive the adaptive planning step.

[0013] The silicon carbide pressureless cylinder surface finishing grinding equipment includes: a high-precision three-dimensional scanning module, used to acquire high-precision three-dimensional point cloud data of the outer surface of the cylinder blank; The data processing and recognition module is communicatively connected to the 3D scanning module and is configured to perform the collaborative recognition and parameterized characterization steps. The intelligent planning module, which is communicatively connected to the data processing and recognition module, is configured to execute the adaptive planning steps. Multi-axis motion actuators are used to carry and drive the movement of the grinding head; The real-time vision servo module is communicatively connected to the control terminal of the multi-axis motion actuator and is configured to perform the vision monitoring and trajectory dynamic adjustment functions in the closed-loop execution steps.

[0014] Optionally, the data processing and recognition module includes a macroscopic comparison unit and a microscopic intelligent recognition unit arranged in parallel; the macroscopic comparison unit is configured to perform three-dimensional registration and deviation analysis; the microscopic intelligent recognition unit integrates a trained PointNet++ deep learning model for directly processing three-dimensional point cloud data and identifying microscopic defects.

[0015] In summary, this application includes the following beneficial technical effects: 1. Accurate and efficient defect identification: By working in parallel with the PointNet++ deep learning model through 3D registration technology, it achieves collaborative identification of macroscopic geometric deviations and microscopic defects. The single-point measurement accuracy of 3D point cloud is better than 0.005mm, and the average intersection-union ratio (mIoU) of microscopic defect identification exceeds 90%. It can accurately output the defect type, location and geometric parameters. 2. Adaptive processing strategy: Layered contour grinding and fixed-point spiral grinding commands are generated for macroscopic protrusions and microscopic pores and cracks, respectively. The process parameters are dynamically adjusted according to the defect characteristics to ensure the repair effect of different types of defects. 3. Significantly improved machining accuracy: The real-time vision servo module samples at a frequency of no less than 100Hz, with a closed-loop control cycle of ≤20ms and a trajectory tracking accuracy better than 0.01mm, effectively compensating for system errors such as clamping and vibration; 4. High automation and adaptability: It forms a closed-loop automated processing system without human intervention. It can adapt to cylinders of different sizes with diameters of 300-2500mm and lengths of 500-6000mm. It can also adapt to various working conditions such as high surface quality requirements and deep and narrow defects through parameter adjustment. Attached Figure Description

[0016] Figure 1 This is a flowchart of the method steps for grinding and finishing the surface of a silicon carbide pressureless cylinder. Figure 2 This is a structural framework diagram of the silicon carbide pressureless cylinder surface finishing grinding equipment. Detailed Implementation

[0017] The following detailed description of specific embodiments of the present invention is intended to enable those skilled in the art to fully understand the technical solutions, technical effects and implementation methods of the present invention, and to implement them without creative effort.

[0018] I. Overview of the Overall Technical Solution This invention provides an intelligent finishing grinding method and equipment for the surface of pressureless silicon carbide cylindrical bodies, forming a complete closed-loop system of "perception-decision-execution-compensation". This method and equipment are particularly suitable for treating macro- and micro-complex defects on the surface of pressureless sintered silicon carbide ceramic cylindrical bodies (typical hardness HRA90-95, density 95-99%), aiming to achieve automated, high-precision, and adaptive surface finishing.

[0019] II. System Equipment Composition The intelligent dressing and grinding equipment mainly includes the following five functional modules, which interact and coordinate control through an industrial network: 1. High-precision 3D scanning module This module is used to acquire three-dimensional topographic data of the cylinder surface. Its core component is a laser line scanning sensor, which boasts a single-point measurement accuracy of no less than 0.005 mm. The sensor is mounted on a multi-degree-of-freedom moving mechanism, which drives the sensor to move axially and circumferentially along the cylinder, achieving full-coverage scanning of the outer surface of cylinders of different sizes (suitable for diameters of 300-2500 mm and lengths of 500-6000 mm). During scanning, the moving speed and sampling frequency are adjusted to ensure that the generated three-dimensional point cloud density is no less than 1000 points / cm². For ultra-long cylinders, segmented scanning is used, followed by point cloud stitching algorithms for fusion.

[0020] 2. Data Processing and Recognition Module This module is the intelligent core of the system, built upon an industrial computer or server with high-performance computing capabilities. It internally operates two processing units in parallel: Macroscopic alignment unit: This unit performs a 3D registration process based on the Iterative Closest Point (ICP) algorithm. It precisely aligns the scanned point cloud with the preset CAD design model and calculates the normal deviation of each point in the point cloud relative to the surface of the theoretical model, generating a normal deviation field to quantify macroscopic geometric deviations (such as bulges and depressions).

[0021] Microscopic Intelligent Recognition Unit: This unit integrates a specially trained deep learning point cloud segmentation and detection model for directly processing 3D point cloud data. The model is built on the PointNet++ network architecture and, addressing the highly imbalanced positive and negative samples in point cloud data in silicon carbide surface micro-defects (pores, cracks, etc.), employs the FocalLoss loss function during training to improve the detection capability for small target defects. This unit outputs the defect category, 3D spatial location, and a subset of the point cloud representing the defect region.

[0022] 3. Intelligent Planning Module This module receives structured defect information from the data processing and identification module, and automatically generates processing instructions for different defects based on a pre-set process knowledge base. Its core planning logic is as follows: For macroscopic raised areas, a layered contour (Z-level) machining strategy is adopted, and the number of grinding layers and the depth of a single layer are adaptively calculated based on the height of the raised area.

[0023] For defects such as microscopic pores and cracks, a fixed-point helical milling strategy is adopted, which adaptively determines the radius, depth and pitch of the helical trajectory based on the geometric dimensions of the defect (such as diameter and depth).

[0024] The planning module ultimately outputs a set of composite CNC machining instructions that can be parsed by the actuator.

[0025] 4. Multi-axis motion actuator This mechanism is a high-precision CNC machine tool or robot system, equipped with at least three linear axes (X, Y, Z) and two rotary axes (A, C) to machine complex curved surfaces of cylinders. A diamond grinding wheel is mounted at the end of the mechanism, and an integrated online wear monitoring unit (such as a laser rangefinder) is used to compensate for machining errors caused by wear in real time.

[0026] 5. Real-time visual servo module This module constitutes a closed-loop feedback loop in the machining process. It includes a fixedly mounted high-speed industrial camera and an optical cooperative target pre-fixed on the cylinder. During machining, the camera continuously captures images of the target, and a visual pose measurement algorithm (such as the PnP algorithm) is used to calculate the instantaneous spatial pose of the cylinder in real time. This pose information is compared with the theoretical pose of the command, and the resulting error signal is sent to a real-time motion controller. This controller dynamically adjusts the motion commands of each axis in a high-frequency (≥1kHz) closed-loop manner to compensate for trajectory errors caused by clamping, vibration, and other factors, ensuring machining accuracy.

[0027] III. Specific Workflow of the Method The implementation steps of the method of the present invention are described in detail below with reference to the embodiments.

[0028] Step S1: Integrated Data Acquisition and Benchmark Establishment The silicon carbide cylindrical blank is clamped onto the worktable of the multi-axis motion actuator, ensuring its rotation axis is aligned with the machine tool coordinate system reference. The high-precision 3D scanning module is activated, controlling the sensor to scan the cylindrical surface along the planned path. Using hand-eye calibration technology, the scanner coordinate system is unified with the machine tool coordinate system, thus obtaining a complete, high-precision 3D point cloud P_total located in the machine tool coordinate system. This point cloud serves as the geometric reference for all subsequent processing and machining.

[0029] Step S2: Cooperative identification and feature extraction of macro- and micro-defects The data processing and recognition modules perform the following tasks in parallel: Macroscopic Deviation Analysis: The macroscopic alignment unit calls the ICP algorithm to register P_total with the design model. After registration, the normal deviation field D_map of the entire surface is calculated. A threshold is set (e.g., +0.02mm), and all "protruding defect regions" are extracted from D_map. The area, maximum height, centroid coordinates, and boundary profile of each region are calculated.

[0030] Intelligent detection of microscopic defects: The microscopic intelligent recognition unit preprocesses P_total (such as downsampling and denoising) before inputting it into the trained deep learning model. To handle large-scale point clouds, a sliding window segmentation strategy is employed. The model infers for each local window, outputting the defect category probability and instance label for each point. Post-processing clustering yields the category, 3D bounding box, point cloud mask, and depth information for each microscopic defect (pore, crack).

[0031] Feature fusion: The results of the above two methods are fused to generate a structured "global defect feature table". This table records each defect's unique ID, type, three-dimensional coordinates, key geometric parameters (such as height, diameter, and depth), and status markers (such as grindable / requires manual intervention) in a list format.

[0032] Step S3: Defect Feature-Driven Adaptive Path Planning The intelligent planning module parses the defect feature table and generates processing instructions based on the following rules: Rule 1 (Macroscopic Protrusion): For a protrusion of height H, determine the single-layer safe grinding depth ap based on the material removal process library (e.g., for the silicon carbide material, ap can be selected between 0.02 mm and 0.05 mm). Then, the number of grinding layers N = ceil(H / ap). For each layer, generate a parallel scan line tool path on the plane of that height, covering the contour of the protrusion area, with the row spacing of adjacent paths ensuring a certain overlap rate (e.g., not less than 30%).

[0033] Rule 2 (Microscopic Defects): For a pore with radius r and depth h, generate a helical milling path with the defect center as the axis. The helical radius R is slightly larger than r (e.g., R = r + 0.1 mm), the total helical depth is slightly larger than h, and the pitch P is determined based on the single axial depth of cut ae (e.g., ae ≤ 0.005 mm). For cracks, generate a similar series of helical or scanning paths along their direction.

[0034] Finally, the path optimizer sorts, connects, and checks for collisions among all instructions, outputting the final "composite machining CNC program".

[0035] Step S4: Real-time vision servo closed-loop precision machining The machining process is executed. During this process, the real-time vision servo module works synchronously: Image acquisition and pose calculation: A high-speed camera captures images of the cooperative target at a fixed frequency (e.g., ≥100Hz). The image processing unit calculates the target's pose in the camera coordinate system in real time through feature extraction and matching, and then converts it to obtain the actual pose P_actual(t) of the cylinder in the machine tool coordinate system.

[0036] Closed-loop error compensation: In each control cycle (e.g., ≤20ms), the motion controller reads the theoretical pose P_desired(t) of the current command and calculates the pose error E(t) = P_desired(t) - P_actual(t). This error is processed by a feedforward-feedback composite control algorithm to generate a real-time compensation amount ΔC(t) for each axis servo drive. The compensation algorithm comprehensively considers the dynamic characteristics of the system and trajectory look-ahead information to suppress tracking errors.

[0037] Through this closed-loop control, the trajectory tracking accuracy of the grinding head relative to the theoretical surface of the cylinder is significantly improved, which can be better than 0.01mm.

[0038] Step S5: Quality Verification and Iterative Optimization (Optional) After processing is complete, the 3D scanning module can be restarted to inspect the ground surface. The inspection results are compared with the target value. If the target value is not fully met, the system can automatically generate supplementary processing instructions for residual defects and repeat steps S3-S40 until the quality requirements are met or the preset maximum number of iterations is reached.

[0039] IV. Key Technologies and Algorithm Examples To enable those skilled in the art to reproduce the algorithm, the following provides a specific implementation example of the core algorithm: 1. Training Examples of Microscopic Defect Recognition Model Data preparation: Collect a large amount of point cloud data of silicon carbide workpieces containing real defects. Each point cloud is manually annotated with detailed information, including: semantic label of each point (background, pore, crack), ID of each defect instance, and its smallest bounding box.

[0040] Model and Training: The PointNet++ model (MSG version) was built using the PyTorch framework. Multi-scale features were extracted from the input point cloud through farthest point sampling (FPS) and grouping. To address the extremely low proportion of positive samples (defect points) in the silicon carbide defect point cloud, FocalLoss was used in the segmentation task. Its parameters can be set to α=0.75, γ=2.0 to reduce the weight of a large number of simple background samples and focus on difficult-to-separate defect points. The Adam optimizer was used with an initial learning rate of 0.001. The model was trained for hundreds of epochs on a dataset that incorporated rotation, translation, and jitter enhancements until the model's mean intersection-over-union (mIoU) on the validation set stabilized above 90%.

[0041] 2. Real-time visual servo closed-loop control embodiment System modeling and controller design: First, the "machine tool-vision" system is identified to obtain its approximate transfer function. Based on this model, a proportional-integral (PI) feedback controller is designed, and the parameters are determined through simulation and field tuning.

[0042] Feedforward and Predictive Compensation: To improve dynamic tracking performance, feedforward control based on an inverse dynamics model is introduced to calculate the compensation force in advance based on the speed and acceleration of the command. Simultaneously, multi-step trajectory look-ahead is implemented to predict the contour error that may be caused by curvature changes in the future path and perform pre-compensation.

[0043] Integration Implementation: The above algorithm is written as a real-time task and deployed in a motion controller running a real-time operating system (such as Linux with PREMPT_RT patch) to ensure strict periodic execution timing.

[0044] V. Adaptation Instructions for Different Application Scenarios The solution described in this invention can be adapted to various working conditions by adjusting parameters: Large-size cylinders: By optimizing scanning path planning and point cloud stitching algorithms, we ensure the integrity and accuracy of data under large sizes.

[0045] High surface quality requirements: Reduce single-layer grinding depth and increase path overlap rate in the planning, and add fine grinding or polishing process after macro grinding.

[0046] For deep and narrow defects: For deep cracks or pores, the peck milling strategy can be used instead of continuous spiral, that is, segmented feeding and retraction to facilitate chip removal and avoid grinding wheel clogging.

[0047] The above description is merely a preferred embodiment of the present invention, but the scope of protection of the present invention is not limited thereto. Any equivalent substitutions or modifications made by those skilled in the art within the scope of the technology disclosed in the present invention, based on the technical solution and inventive concept of the present invention, should be covered within the scope of protection of the present invention.

Claims

1. A method of surface finishing a silicon carbide pressureless cylinder, characterized by: Includes the following steps: S1: Collaborative data acquisition steps: Acquire high-precision three-dimensional point cloud data of the outer surface of the cylinder blank; S2: Collaborative identification and parameterized characterization steps: Based on the three-dimensional point cloud data, macroscopic geometric deviation identification and microscopic material defect intelligent identification are performed in parallel, and a defect feature set containing location, type and geometric parameters is output; S3: Adaptive planning step: Based on the defect feature set, generate composite machining instruction sets with different grinding strategies and process parameters for macroscopic protrusion defects and microscopic pores or cracks, respectively. S4: Closed-loop execution steps: Control the grinding head to execute the composite machining instruction set, and at the same time monitor the cylinder position and dynamically adjust the grinding head movement trajectory in real time through the vision sensing system to compensate for system errors in a closed-loop manner.

2. The method of claim 1 wherein: In the collaborative identification and parameterized characterization steps, the macroscopic geometric deviation identification is achieved by performing three-dimensional registration between the three-dimensional point cloud and the design CAD model, and outputting the normal deviation distribution; the microscopic material defect intelligent identification is achieved by directly inputting the three-dimensional point cloud into the trained PointNet++ deep learning model, and outputting the defect category, three-dimensional coordinates and point cloud segmentation results.

3. The method of claim 2 wherein: In the adaptive planning step, the specific method for generating the composite processing instruction set is as follows: For macroscopic protrusion defects, a layered contour grinding command is generated, and the grinding depth of a single layer is adaptively set based on the normal deviation distribution. For microscopic pores or cracks, a fixed-point spiral grinding command is generated, and the radius and depth of the spiral trajectory are adaptively determined based on the three-dimensional coordinates of the defect and the point cloud segmentation results.

4. The method of claim 3, wherein: The single-layer grinding depth of the layered contour grinding command is set between 0.02 mm and 0.05 mm, and the overlap rate of adjacent grinding paths is not less than 30%.

5. The silicon carbide pressureless cylinder surface finishing grinding method according to claim 3, characterized in that: The single grinding depth increment of the fixed-point spiral grinding command is no more than 0.005 mm.

6. The silicon carbide pressureless cylinder surface finishing grinding method according to claim 1, characterized in that: In the closed-loop execution step, the visual sensing system samples at a frequency of not less than 100Hz, the dynamically adjusted closed-loop control period is not greater than 20ms, and the trajectory tracking accuracy is better than 0.01mm.

7. The silicon carbide pressureless cylinder surface finishing grinding method according to claim 2, characterized in that: The single-point measurement accuracy of the three-dimensional point cloud data is better than 0.005 mm.

8. The method for surface finishing grinding of silicon carbide pressureless cylinders according to any one of claims 1-7, characterized in that: The defect feature set is organized in the form of structured data and is used to drive the adaptive planning step.

9. A silicon carbide pressureless cylinder surface finishing grinding equipment, characterized in that: A high-precision 3D scanning module is used to acquire high-precision 3D point cloud data of the outer surface of the cylinder blank; The data processing and recognition module is communicatively connected to the 3D scanning module and is configured to perform the collaborative recognition and parameterized characterization steps. The intelligent planning module, which is communicatively connected to the data processing and recognition module, is configured to execute the adaptive planning steps. Multi-axis motion actuators are used to carry and drive the movement of the grinding head; The real-time vision servo module is communicatively connected to the control terminal of the multi-axis motion actuator and is configured to perform the vision monitoring and trajectory dynamic adjustment functions in the closed-loop execution steps.

10. The silicon carbide pressureless cylinder surface finishing grinding equipment according to claim 9, characterized in that: The data processing and recognition module includes a macroscopic comparison unit and a microscopic intelligent recognition unit arranged in parallel; the macroscopic comparison unit is configured to perform three-dimensional registration and deviation analysis; the microscopic intelligent recognition unit integrates a trained PointNet++ deep learning model, which is used to directly process three-dimensional point cloud data and identify microscopic defects.

Images