Main bearing cap grinding system based on machine vision and robot cooperation and control method

Abstract

The invention discloses a main bearing cap grinding system based on machine vision and robot cooperation and a control method, and relates to the technical field of machining. The main bearing cap grinding system comprises the steps that an image of a main bearing cap is collected, and image preprocessing is carried out; extracting edge contour information of the main bearing cover from the preprocessed image by adopting an edge detection algorithm; generating a polishing path of the robot according to the edge contour information of the main bearing cover and a path planning algorithm; controlling the robot to drive the grinding head to grind according to a preset path according to the grinding path; in the polishing process, motion state information of the robot, position information of a polishing head and image information of a polishing surface are obtained in real time, and a polishing path and polishing parameters are automatically adjusted through a feedback control algorithm. Through precise control and real-time feedback, the grinding precision and the surface consistency are ensured, manual intervention is reduced, the labor intensity is reduced, and the device is suitable for batch production, can cope with main bearing covers of different specifications and supports the flexible manufacturing requirement.

Description

Technical Field

[0001] The present invention relates to the field of mechanical processing technology, and more particularly to a main bearing cap grinding system and a control method based on collaboration between machine vision and a robot. Background Art

[0002] In the field of mechanical manufacturing, the grinding process of main bearing caps plays a vital role. The quality of the grinding directly affects the overall performance and reliability of the mechanical equipment. Traditional main bearing cap grinding methods often rely on manual operation, which has many shortcomings, such as high labor intensity, relatively low work efficiency, and difficulty in ensuring sufficient grinding accuracy.

[0003] Although some automated grinding equipment has appeared on the market, which has improved the grinding efficiency to a certain extent, they still face the challenge of achieving precise grinding path planning and adaptive adjustment when processing main bearing caps with complex shapes, which leads to unstable grinding quality.

[0004] Therefore, how to propose a main bearing cap grinding system and control method based on the collaboration of machine vision and robots to effectively adapt to the grinding of main bearing caps with complex shapes and improve the grinding efficiency and accuracy is an urgent problem that technical personnel in this field need to solve. Summary of the Invention

[0005] In view of this, the present invention provides a main bearing cap grinding system and control method based on the collaboration of machine vision and robots, which effectively adapts to the grinding of main bearing caps with complex shapes, improves the grinding efficiency and precision, and realizes the automation and intelligence of the grinding process. To achieve the above objectives, the present invention adopts the following technical solutions:

[0006] A main bearing cap grinding system based on machine vision and robot collaboration, comprising:

[0007] Preprocessing module: used to collect images of the main bearing cap and perform image preprocessing;

[0008] Edge detection module: used to extract the edge contour information of the main bearing cap using the edge detection algorithm on the pre-processed image;

[0009] Path planning module: used to generate the robot's grinding path based on the edge profile information of the main bearing cap and the path planning algorithm;

[0010] Execution module: used to control the robot to drive the grinding head to grind according to the predetermined path according to the grinding path;

[0011] Feedback adjustment module: used to obtain the robot's motion state information, the position information of the grinding head and the image information of the grinding surface in real time during the grinding process, and automatically adjust the grinding path and grinding parameters through the feedback control algorithm.

[0012] Optionally, a main bearing cap grinding control method based on collaboration between machine vision and robots includes:

[0013] Collect images of the main bearing cap and perform image preprocessing;

[0014] The edge detection algorithm is used to extract the edge contour information of the main bearing cap from the preprocessed image;

[0015] Generate the robot's grinding path based on the edge profile information of the main bearing cap and the path planning algorithm;

[0016] According to the grinding path, the robot is controlled to drive the grinding head to grind along the predetermined path;

[0017] During the grinding process, the robot's motion state information, the grinding head's position information, and the image information of the grinding surface are obtained in real time, and the grinding path and grinding parameters are automatically adjusted through the feedback control algorithm.

[0018] Optionally, the image preprocessing includes: performing grayscale conversion, noise reduction and binarization operations on the collected image.

[0019] Optionally, extracting edge contour information of the main bearing cap by using an edge detection algorithm includes: extracting edge contour information of the main bearing cap by using a Canny edge detection algorithm, and determining edge contour shape and position information of the main bearing cap.

[0020] Optionally, the generating of the robot's grinding path according to the edge profile information of the main bearing cover and the path planning algorithm includes: using a rapidly expanding random tree RRT algorithm combined with an improved A algorithm for path planning, and planning a feasible path from the starting point to the target point by using the rapidly expanding random tree RRT algorithm; the improved A algorithm smoothes the path based on an improved heuristic function and a search strategy.

[0021] Optionally, the method of controlling the robot to drive the grinding head to perform grinding according to a predetermined path based on the grinding path includes: converting the planned path points into joint space trajectories recognized by the robot, smoothing the motion trajectory using quintic polynomial interpolation or B-spline interpolation, setting corresponding contact forces according to different grinding areas, and adjusting the rotation speed and feed rate of the grinding head.

[0022] Optionally, the real-time acquisition of the robot's motion state information includes: installing high-precision encoders at the robot's joints, the encoders obtaining the position, speed and acceleration information of each joint in real time by recording the rotation angle and number of pulses, deploying inertial measurement units at the robot's base and key parts, integrating accelerometers, gyroscopes and magnetometers, and monitoring the robot's posture changes and linear motion parameters in three-dimensional space in real time.

[0023] Optionally, the step of obtaining the position information of the grinding head includes: installing a laser displacement sensor at the connection part of the grinding head, using the principle of laser triangulation to measure the position and information of the grinding head relative to the workpiece or the reference coordinate system in real time, and combining it with the six-dimensional force sensor installed at the end of the robotic arm to sense the magnitude and direction of the contact force during the grinding process, and obtaining the working position and posture of the grinding head through comprehensive judgment through force feedback conversion.

[0024] Optionally, it also includes: preprocessing the robot motion state information, grinding head position information and grinding surface image information collected by the sensor, using multi-sensor information fusion technology to fuse the preprocessed motion state information, grinding head position information and grinding surface image information to construct a grinding work scene model.

[0025] Optionally, the automatic adjustment of the grinding path and grinding parameters through the feedback control algorithm includes: designing a feedback control algorithm based on the working scene model, adopting the model predictive control MPC algorithm, establishing a dynamic model of the grinding process according to the current grinding state and the preset grinding target, predicting the impact of the grinding path and parameter changes on the grinding effect in the future period, combining the adaptive control algorithm, and adjusting the control parameters in real time according to the actual grinding situation. When unevenness or defects are detected on the grinding surface, the grinding path is replanned through the path planning algorithm to avoid the defective area or focus on the defect. According to the feedback of the grinding force and the image analysis results, the rotation speed, feed speed, and pressure grinding parameters of the grinding head are adjusted.

[0026] It can be seen from the above technical solutions that, compared with the prior art, the present invention discloses a main bearing cap grinding system and control method based on the collaboration of machine vision and robots, which has the following beneficial effects:

[0027] This invention proposes a main bearing cap grinding system based on the collaboration of machine vision and robotics. The system comprises a preprocessing module for capturing and preprocessing images of the main bearing cap; an edge detection module for extracting the edge profile of the main bearing cap using an edge detection algorithm from the preprocessed image; a path planning module for generating a robot grinding path based on the main bearing cap edge profile information and the path planning algorithm; an execution module for controlling the robot to drive the grinding head along a predetermined path according to the grinding path; and a feedback adjustment module for automatically adjusting the grinding path and parameters using a feedback control algorithm to obtain real-time information on the robot's motion state, the grinding head's position, and the image of the grinding surface during the grinding process. The system improves image quality through grayscaling, noise reduction, and binarization, providing clear data for subsequent analysis. Canny edge detection accurately extracts the main bearing cap's contour, determining its shape and position, providing a precise basis for path planning. Utilizing a combination of RRT and an improved A algorithm, a feasible path is first generated using RRT, and then the improved A algorithm is used to optimize path smoothness, reducing motion jitter and improving grinding efficiency. Contact force, rotation speed and feed rate are set according to different grinding areas to achieve fine grinding. The robot posture, grinding head position and contact force information are obtained in real time through encoders, IMUs, laser displacement sensors and six-dimensional force sensors. Through model predictive control (MPC) combined with adaptive algorithms, the grinding path and parameters are dynamically adjusted to predict the grinding effect and optimize the control strategy. Through image analysis of the polished surface, uneven or defective areas are automatically identified. The path is automatically adjusted for defective areas to achieve obstacle avoidance or focused grinding, thereby improving surface quality consistency. The integrated preprocessing, edge detection, path planning, execution and feedback adjustment modules work together to form a closed-loop control system. The fusion of multi-sensor data constructs a three-dimensional working model to provide comprehensive information support for intelligent decision-making. BRIEF DESCRIPTION OF THE DRAWINGS

[0028] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the following briefly introduces the drawings required for use in the embodiments or the description of the prior art. Obviously, the drawings described below are merely embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on the provided drawings without paying any creative work.

[0029] Figure 1 This is a flow chart of a main bearing cap grinding control method based on collaboration of machine vision and robots provided by the present invention.

[0030] Figure 2 This is a structural framework diagram of a main bearing cap grinding system based on collaboration between machine vision and robots provided by the present invention. DETAILED DESCRIPTION

[0031] The following will clearly and completely describe the technical solutions in the embodiments of the present invention in conjunction with the accompanying drawings. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without making creative efforts are within the scope of protection of the present invention.

[0032] The embodiment of the present invention discloses a main bearing cap grinding system based on machine vision and robot collaboration, such as Figure 2 Shown, including:

[0033] Preprocessing module: used to collect images of the main bearing cap and perform image preprocessing;

[0034] Edge detection module: used to extract the edge contour information of the main bearing cap using the edge detection algorithm on the pre-processed image;

[0035] Path planning module: used to generate the robot's grinding path based on the edge profile information of the main bearing cap and the path planning algorithm;

[0036] Execution module: used to control the robot to drive the grinding head to grind according to the predetermined path according to the grinding path;

[0037] Feedback adjustment module: used to obtain the robot's motion state information, the position information of the grinding head and the image information of the grinding surface in real time during the grinding process, and automatically adjust the grinding path and grinding parameters through the feedback control algorithm.

[0038] Furthermore, a main bearing cap grinding control method based on machine vision and robot collaboration is proposed. Figure 1 Shown, including:

[0039] Collect images of the main bearing cap and perform image preprocessing;

[0040] The edge detection algorithm is used to extract the edge contour information of the main bearing cap from the preprocessed image.

[0041] Generate the robot's grinding path based on the edge profile information of the main bearing cap and the path planning algorithm;

[0042] According to the grinding path, the robot is controlled to drive the grinding head to grind along the predetermined path;

[0043] During the grinding process, the robot's motion state information, the grinding head's position information, and the image information of the grinding surface are obtained in real time, and the grinding path and grinding parameters are automatically adjusted through the feedback control algorithm.

[0044] Furthermore, the image preprocessing includes: performing grayscale conversion, noise reduction and binarization operations on the collected image.

[0045] Furthermore, the extracting edge contour information of the main bearing cap by using an edge detection algorithm includes: extracting edge contour information of the main bearing cap by using a Canny edge detection algorithm, and determining edge contour shape and position information of the main bearing cap.

[0046] Furthermore, the generation of the robot's grinding path based on the edge profile information of the main bearing cover and the path planning algorithm includes: using the rapidly expanding random tree RRT algorithm combined with the improved A algorithm for path planning, and planning a feasible path from the starting point to the target point through the rapidly expanding random tree RRT algorithm; the improved A algorithm smoothes the path based on the improved heuristic function and search strategy.

[0047] In a specific embodiment, the Rapidly Expanding Random Tree (RRT) algorithm is combined with an improved A algorithm for path planning. The RRT algorithm has high search efficiency in high-dimensional spaces and complex environments, and can quickly find a feasible path from the starting point to the target point. The improved A algorithm excels in path optimization, making the path smoother and more efficient by improving the heuristic function and search strategy. Combining the two ensures that a feasible path is quickly found and the path is optimized to meet the requirements of main bearing cap grinding path planning. The specific steps include:

[0048] (1) Data collection and processing

[0049] 1) Use a high-precision 3D scanner to obtain a 3D model of the main bearing cap, maintaining a scanning accuracy within ±0.05mm to ensure accurate edge profile information. Import the resulting point cloud data into professional reverse engineering software (such as Geomagic DesignX) for pre-processing operations such as point cloud noise reduction and stitching to remove noise points and improve data quality.

[0050] 2) Based on the polishing process requirements, the surface area to be polished is identified. Through manual interaction or image recognition algorithms, the area to be polished is marked in the 3D model, defining the target range for subsequent path planning.

[0051] 3) Set the key parameters of the grinding head: the grinding head diameter is 10 mm, the grinding speed range is 50-100 mm / s, and the grinding force range is 5-15 N. These parameters will serve as constraints for path planning.

[0052] (2) Building an environmental model

[0053] 1) The octree method was used to model the 3D model and workspace of the main bearing cap. A rectangular bounding box was constructed around the main bearing cap, enclosing its workspace. This bounding box was then divided into multiple octree nodes of equal size. Based on the polishing accuracy requirements, the minimum node size was set to 1 mm × 1 mm × 1 mm. Recursive partitioning was performed until a preset stopping condition was met, forming a hierarchical octree structure.

[0054] 2) In the octree model, nodes corresponding to the solid parts of the main bearing caps are marked as obstacle nodes, and areas without solid parts in the workspace are marked as free nodes. The octree model represents the environmental information of the workspace and provides an intuitive spatial structure for path planning.

[0055] (3) Path generation

[0056] 1) RRT algorithm generates initial path

[0057] Determine the starting and target points. The starting point is set at the initial position of the grinding head, and the target points are set appropriately based on the boundaries of the surface to be ground and the grinding sequence. Starting from a corner of the surface to be ground, set the target points in a clockwise or counterclockwise direction to ensure that the entire surface is covered.

[0058] Initialize the RRT tree and add the starting point to the tree. In each iteration, randomly sample a point x in the workspace. rand , then find the distance x in the RRT tree rand The nearest node x near .

[0059] From x near Along x near with x rand The connection direction of the new node x is generated with a fixed step size (set to 5mm) new . Check x new Is it in the free node area? If so, add it to the RRT tree and establish a connection with x near connection; if x new If it is located in the obstacle node area, it will be resampled to generate a new node.

[0060] Repeat the above steps until the target point is included in the RRT tree, and then an initial path from the starting point to the target point is obtained.

[0061] 2) Improved A algorithm optimization path

[0062] The initial path generated by the RRT algorithm is discretized, and the nodes on the path are used as search nodes for the improved A algorithm. An improved heuristic function h(n) is defined, which combines the Euclidean distance from the node to the target point and the smoothness of the path to guide the search towards a more optimal direction.

[0063] h(n)=α×d euclidean (n, goal) + β × s(n);

[0064] Among them, α and β are weight coefficients, which can be adjusted according to actual conditions, α = 0.8, β = 0.2, d euclidean (n, goal) is the Euclidean distance from node n to the goal point, and s(n) is the smoothness measure of the path at node n.

[0065] A bidirectional search strategy is employed, searching both the starting point and the target point simultaneously, reducing the search space and improving search efficiency. During the search process, paths are evaluated and screened based on constraints such as grinding head size, grinding speed, and grinding force. The distances between adjacent nodes on the path are checked to ensure they meet the grinding speed and grinding head size requirements. If not, the path is adjusted.

[0066] When the paths of the bidirectional search meet in the middle, the optimized polishing path is obtained.

[0067] (4) Path verification and adjustment

[0068] The generated grinding path was imported into the robot simulation software (Gazebo) and verified by simulation using a 3D model of the main bearing cap and the robot model. During the simulation, the distance between the grinding head and the main bearing cap was monitored in real time to check for collision risks. Simultaneously, different grinding speeds and forces were simulated to verify whether the grinding path met the required grinding process requirements, such as uniform coverage of the surface to be polished and whether the surface roughness after polishing met the standard.

[0069] Adjust the path based on the simulation results. If a collision risk is found, return to the path generation step, adjust the parameters of the RRT algorithm or the improved A algorithm, and regenerate the path;

[0070] If the polishing effect does not meet the process requirements, the constraints of the path planning algorithm are optimized, the value range of the polishing speed and polishing force are adjusted, the path is generated again and simulation verification is performed until the polishing path meets all requirements.

[0071] Furthermore, the method of controlling the robot to drive the grinding head to grind according to a predetermined path based on the grinding path includes: converting the planned path points into joint space trajectories recognized by the robot, using quintic polynomial interpolation or B-spline interpolation to smooth the motion trajectory, setting corresponding contact forces according to different grinding areas, and adjusting the rotation speed and feed rate of the grinding head.

[0072] In a specific embodiment, the step of controlling the robot to drive the grinding head to perform grinding along a predetermined path according to the grinding path specifically includes:

[0073] 1. Data Collection and Preprocessing

[0074] (1) 3D data acquisition

[0075] Use an industrial 3D scanner or a high-precision camera to scan the main bearing cap in all directions to obtain its point cloud data.

[0076] (2) Point cloud data processing

[0077] The acquired point cloud data is denoised to remove outliers.

[0078] Perform point cloud simplification operations to reduce data volume while preserving key geometric features.

[0079] Complete the point cloud registration work and unify the data obtained from multiple scans into the same coordinate system.

[0080] (3) Edge contour extraction

[0081] Edge detection algorithms, such as the Canny algorithm and the Sobel operator, are used to identify the edge contours of the main bearing cap. Based on curvature analysis, areas requiring polishing, such as sharp edges and fillets, are accurately extracted.

[0082] 2. Polishing Path Planning

[0083] (1) Coordinate system definition

[0084] Clarify the workpiece coordinate system and ensure that the conversion relationship between it and the robot base coordinate system is clear and accurate. Determine the TCP (Tool Center Point), which is the position and posture of the grinding head.

[0085] (2) Path generation strategy

[0086] Equidistant offset method: Generate a path parallel to the edge contour according to the set offset distance.

[0087] Layer cutting method: For complex surfaces, divide them into multiple plane layers and plan the polishing path layer by layer.

[0088] Curvature Adaptive Method: Dynamically adjusts path spacing and polishing parameters based on surface curvature.

[0089] (3) Path parameter settings

[0090] Determine a reasonable line spacing, generally within the range of 0.5-2mm, and the specific value should be determined according to the surface roughness.

[0091] Set the appropriate grinding speed, usually 50-200 mm / s, which can be adjusted according to the material characteristics.

[0092] Set the overlap rate, recommended to be between 30%-50% to ensure uniformity of grinding.

[0093] (4) Path optimization processing

[0094] Smooth the path to reduce unnecessary starts, stops, and sudden changes in direction. Plan reasonable starting and exit points to avoid leaving marks on the workpiece surface. Perform collision detection to prevent collisions between the robot and fixtures and peripheral equipment.

[0095] 3. Robot motion control trajectory generation

[0096] Convert the planned path points into joint space trajectories that the robot can recognize. Use quintic polynomial interpolation or B-spline interpolation to ensure the smoothness of the motion trajectory.

[0097] Process parameter configuration: Set the appropriate contact force for each grinding area, ranging from approximately 5-50N. Adjust the grinding head speed, typically between 3000-10000rpm. Determine the appropriate feed rate, typically between 0.05-0.3mm. Program generation and debugging: Generate executable program code for the robot, such as ABB's RAPID or KUKA's KRL. Perform simulation verification in a virtual environment to check for interference and optimize motion time. Perform teaching and fine-tuning on the actual equipment to ensure program accuracy.

[0098] 4. Grinding process monitoring and adjustment force feedback control

[0099] (1) A six-dimensional force sensor is used to monitor contact force in real time, and closed-loop control is performed using a PID algorithm. When contact force fluctuates, the robot's posture or feed rate is automatically adjusted. Visual feedback correction is used to dynamically detect edge position during the grinding process and perform real-time path compensation. Surface roughness is detected and grinding parameters are dynamically adjusted based on the detection results.

[0100] (2) Abnormal situation handling: When abrasive wear is detected, the abrasive replacement program is automatically triggered. If a collision or overload occurs, the robot stops immediately and issues an alarm.

[0101] 5. System Integration and Verification Hardware Integration

[0102] Integrate the robot, grinding tool, force sensor, and vision system, and perform communication debugging. Surface roughness of polished workpieces is inspected and evaluated using the Ra parameter. Edge consistency is checked, and chamfer dimensions are measured to ensure compliance. Material removal is quantified and analyzed to ensure machining accuracy. This optimizes cycle time and improves production efficiency.

[0103] Furthermore, the real-time acquisition of the robot's motion state information includes: installing high-precision encoders at the robot's joints, the encoders obtaining real-time position, velocity and acceleration information of each joint by recording the rotation angle and number of pulses, deploying inertial measurement units at the robot's base and key positions, integrating accelerometers, gyroscopes and magnetometers, and monitoring the robot's posture changes and linear motion parameters in three-dimensional space in real time.

[0104] Furthermore, the step of obtaining the position information of the grinding head includes: installing a laser displacement sensor at the connection part of the grinding head, using the principle of laser triangulation to measure the position and information of the grinding head relative to the workpiece or the reference coordinate system in real time, and combining with the six-dimensional force sensor installed at the end of the robotic arm to sense the magnitude and direction of the contact force during the grinding process, and obtaining the working position and posture of the grinding head through comprehensive judgment through force feedback conversion.

[0105] Furthermore, it also includes: preprocessing the robot motion state information, grinding head position information and grinding surface image information collected by the sensor, using multi-sensor information fusion technology to fuse the preprocessed motion state information, grinding head position information and grinding surface image information to construct a grinding work scene model.

[0106] In a specific embodiment, the multi-sensor information fusion technology is used to fuse the pre-processed motion state information, the grinding head position information and the grinding surface image information to construct the grinding work scene model. The specific steps include:

[0107] 1. Sensor Data Fusion and Filtering

[0108] (1) Kalman filter algorithm

[0109] 1) Equation of state: x k =Ax k-1 +Bu k +w k ;

[0110] Among them, x k is the state vector at the kth moment, A is the state transfer matrix, B is the control input matrix, u k is the control input, wk is the process noise, obey

[0111] 2) Observation equation: z k =Hx k +v k ;

[0112] Among them, z k is the observation vector, H is the observation matrix, v k is the observation noise, obey

[0113] 3) Iteration steps:

[0114] Prediction stage:

[0115] Update phase:

[0116]

[0117] Among them, K k is the Kalman gain, P k is the posterior covariance matrix.

[0118] (2) Particle filter algorithm steps

[0119] 1) Particle initialization: randomly generate N particles Weight

[0120] 2) Importance Sampling: Adopting system dynamics model as proposal distribution

[0121] 3) Weight update: in, is the observation likelihood function;

[0122] 4) Resampling: remove low-weight particles, copy high-weight particles, and generate a new particle set

[0123] 5) State Estimation

[0124] 2. Image Information Processing and Feature Extraction

[0125] (1) Edge detection (Canny algorithm)

[0126] 1) Gaussian filtering denoising: Convolve with the image to remove noise.

[0127] 2) Calculate the gradient magnitude and direction:

[0128]

[0129] 3) Non-maximum suppression: retain the local maximum in the gradient direction and suppress non-edge points.

[0130] 4) Double threshold screening edge: define high and low thresholds T h and T l , connecting continuous edges.

[0131] (2) Image segmentation (Otsu threshold method)

[0132] 1) Calculate the grayscale histogram p i .

[0133] 2) Traverse the threshold t and calculate the inter-class variance:

[0134]

[0135] 3) Select The largest t is used as the segmentation threshold.

[0136] (3) Feature extraction (SIFT algorithm key points)

[0137] 1) Scale space construction: L(x, y, σ) = G(x, y, σ) * I(x, y). Gaussian kernels G(·, σ) of different scales are convolved with the image to generate the scale space.

[0138] 2) Extreme point detection: Compare the current pixel with the neighborhood (26 points) in the scale space to find the extreme points.

[0139] 3) Key point positioning: Fit a three-dimensional quadratic function through Taylor expansion to eliminate low-contrast points and edge response points.

[0140] 4) Direction assignment: Calculate the gradient histogram of the key point neighborhood and assign the main direction and auxiliary direction.

[0141] 5) Feature descriptor generation: Divide the key point neighborhood into 4×4 sub-regions, calculate the gradient amplitude and direction of each sub-region, and generate a 128-dimensional feature vector.

[0142] 3. Multi-sensor information fusion model

[0143] (1) Data registration, time and space alignment

[0144] 1) Time synchronization: Linear interpolation is used to unify the timestamps of sensor data with different frequencies.

[0145] 2) Spatial transformation: Transform multi-sensor data into the same coordinate system through the coordinate transformation matrix T: P world =T·P sensor , the IMU data is transformed into the robot base coordinate system through the rotation and translation matrix.

[0146] (2) Weighted fusion algorithm

[0147] Define the confidence weight w of each sensor data i , w1 is the motion state weight, w2 is the position weight, w3 is the image feature weight, and ∑w i =1, the fused state vector is:

[0148] X fused =w1X motion +w2X position +w3X image , where the weights can be obtained by the sensor noise covariance matrix ∑ i Dynamic Adjustment:

[0149] In this embodiment, the encoder + IMU fusion is used to estimate the robot end position (x, y, z) and posture (θ x ,θ y ,θ z ) to eliminate motion noise. Use Canny edge detection to identify the workpiece outline, combined with SIFT feature matching to locate the grinding area. Position data and visual features are weightedly fused to generate a 3D scene model, guiding path planning algorithms (such as the A algorithm) to generate collision-free grinding trajectories.

[0150] Furthermore, the automatic adjustment of the grinding path and grinding parameters through the feedback control algorithm includes: designing a feedback control algorithm based on the working scene model, adopting the model predictive control MPC algorithm, establishing a dynamic model of the grinding process according to the current grinding state and the preset grinding target, predicting the impact of the grinding path and parameter changes on the grinding effect in the future period, combining the adaptive control algorithm, and adjusting the control parameters in real time according to the actual grinding situation. When unevenness or defects are detected on the grinding surface, the grinding path is replanned through the path planning algorithm to avoid the defective area or focus on the defect. According to the feedback of the grinding force and the image analysis results, the rotation speed, feed speed, and pressure grinding parameters of the grinding head are adjusted.

[0151] In a specific embodiment, the automatic adjustment of the grinding path and grinding parameters by the feedback control algorithm specifically includes:

[0152] 1. Construction of dynamic model of polishing process

[0153] (1) State space expression

[0154] The state vector of the polishing system is expressed as:

[0155] Among them, p k represents the position, vk is the feed rate, f k is the contact force, e k is the surface error, e k Indicates the rate of change of error.

[0156] The control input vector is: Where ωk is the grinding head speed, v fk is the feed rate, F setk is the set pressure.

[0157] The discrete state space model is expressed as:

[0158] x k+1 =Ax k +Bu k +w k ;

[0159] y k =Cx k +Du k +v k ;

[0160] Where A, B, C, and D are system matrices, and w k and v k are process noise and measurement noise, respectively.

[0161] (2) Polishing effect prediction model

[0162] The prediction model of material removal is:

[0163]

[0164] Among them, K p , α, β, and γ are material-related parameters.

[0165] The surface roughness prediction model is:

[0166]

[0167] Among them, R a0 is the initial roughness, K r is the roughness coefficient.

[0168] 2. Model Predictive Control (MPC) Design

[0169] (1) The prediction time domain of the rolling optimization problem is N p , the control time domain is N c (N c ≤N p ), the optimization objective function is:

[0170]

[0171] Among them, e k+i|k =y ref -y k+i|k is the tracking error, Q, R, P are weight matrices, Δu k+i|k =i k+i|k -u k+i-1|k It is the control increment.

[0172] (2) Constraints Control input constraints are: u min ≤u k+i|k ≤u max ;

[0173] The control input change rate constraint is: Δu min ≤Δu k+i |k≤Δu max ;

[0174] The state constraint is: x min ≤x k+i|k ≤x max ;

[0175] (3) Quadratic Programming (QP) Problem: The MPC problem is transformed into a standard quadratic programming form:

[0176]

[0177] in,

[0178] 3. Adaptive Control Parameter Adjustment

[0179] (1) The online identification of model parameters is performed using the recursive least squares method (RLS):

[0180]

[0181] in, is the parameter estimate, φ k is the regression vector, λ is the forgetting factor, P k is the covariance matrix.

[0182] (2) Adaptive weight adjustment based on system uncertainty σ k Dynamically adjust the weight matrix:

[0183]

[0184] in, Q0, R0 are the initial weight matrices, and σ0 is the reference uncertainty.

[0185] 4. Defect Detection and Path Replanning

[0186] (1) Surface defect detection

[0187] Extract surface features through image processing:

[0188] Where I(x, y) is the surface image, is the feature extraction function, Θ is the feature parameter, and D(x, y) is the defect distribution map.

[0189] (2) The RRT path planning algorithm defines the state space χ and the obstacle area χ obs :

[0190] 1) Initialization:

[0191] 2) Iterative process:

[0192] sampling:

[0193] Find the closest point:

[0194] Generate a new node: x new =Steer(x near , x rand ,δ);

[0195] Collision detection: If Collision Free (xnear, xnew), then find the nearest neighbor point:

[0196] Select the parent node:

[0197] Update the tree:

[0198] Rewire: To X near For each node x in C(x new )+c(x new , x)<C(x), then update the parent node.

[0199] (3) Path smoothing is done using Bezier curves:

[0200]

[0201] Among them, P i are the control points and n is the curve order.

[0202] 5. Parameter Adaptive Adjustment Strategy

[0203] (1) Grinding parameter adjustment rules

[0204] Parameter adjustment rules based on fuzzy logic:

[0205]

[0206] in, is a fuzzy inference system, and ∏ is a fuzzy rule set.

[0207] (2) Multi-objective optimization

[0208] Construct a multi-objective function:

[0209] min u [w1·||e k ||+w2·Time(u)+w3·Energy(u)+w4·Wear(u)];

[0210] Among them, w1, w2, w3, and w4 are weight coefficients, and Time, Energy, and Wear represent processing time, energy consumption, and tool wear, respectively.

[0211] 6. Algorithm Implementation Process

[0212] (1) State initialization: Get x0, set u0

[0213] (2) Iterative control (k = 0, 1, 2, ...);

[0214] State measurement: Get x k ;

[0215] Defect detection: calculate D(x, y);

[0216] Parameter Identification: Update

[0217] Model prediction: Calculate x k+i|k and y k+i|k (i=0,1,...,N p -1);

[0218] Optimization solution: Solve the QP problem to get

[0219] Path planning: Generate a new path if necessary

[0220] Control Execution: Application

[0221] Time update: k=k+1.

[0222] By combining the predictive capability of MPC, the environmental adaptability of adaptive control, and the flexibility of intelligent path planning through algorithms, parameters can be adjusted in real time according to the workpiece characteristics and grinding status to ensure grinding quality and efficiency.

[0223] The various embodiments in this specification are described in a progressive manner, with each embodiment focusing on the differences from other embodiments. Reference can be made to the common and similar parts between the various embodiments. For the devices disclosed in the embodiments, since they correspond to the methods disclosed in the embodiments, the description is relatively simple, and the relevant parts can be referred to the method description.

[0224] The above description of the disclosed embodiments is intended to enable one skilled in the art to implement or use the present invention. Various modifications to these embodiments will be readily apparent to one skilled in the art, and the general principles defined herein may be implemented in other embodiments without departing from the spirit or scope of the present invention. Therefore, the present invention is not limited to the embodiments shown herein but is intended to conform to the widest scope consistent with the principles and novel features disclosed herein.

Claims

1. A main bearing cap grinding system based on machine vision and robot collaboration, characterized in that: include: Preprocessing module: used to collect images of the main bearing cap and perform image preprocessing; Edge detection module: used to extract the edge contour information of the main bearing cap using the edge detection algorithm on the pre-processed image; Path planning module: used to generate the robot's grinding path based on the edge profile information of the main bearing cap and the path planning algorithm; Execution module: used to control the robot to drive the grinding head to grind according to the predetermined path according to the grinding path; Feedback adjustment module: used to obtain the robot's motion state information, the position information of the grinding head and the image information of the grinding surface in real time during the grinding process, and automatically adjust the grinding path and grinding parameters through the feedback control algorithm.

2. A main bearing cap grinding control method based on machine vision and robot collaboration, characterized in that: include: Collect images of the main bearing cap and perform image preprocessing; The edge detection algorithm is used to extract the edge contour information of the main bearing cap from the preprocessed image. Generate the robot's grinding path based on the edge profile information of the main bearing cap and the path planning algorithm; According to the grinding path, the robot is controlled to drive the grinding head to grind along the predetermined path; During the grinding process, the robot's motion state information, the grinding head's position information, and the image information of the grinding surface are obtained in real time, and the grinding path and grinding parameters are automatically adjusted through the feedback control algorithm.

3. The main bearing cap grinding control method based on machine vision and robot collaboration according to claim 2 is characterized in that: The image preprocessing includes: performing grayscale conversion, noise reduction and binarization operations on the collected image.

4. The main bearing cap grinding control method based on machine vision and robot collaboration according to claim 2 is characterized in that: The extracting edge contour information of the main bearing cap by using an edge detection algorithm includes: extracting edge contour information of the main bearing cap by using a Canny edge detection algorithm, and determining edge contour shape and position information of the main bearing cap.

5. The main bearing cap grinding control method based on machine vision and robot collaboration according to claim 2 is characterized in that: The method of generating a robot grinding path based on the edge profile information of the main bearing cap and the path planning algorithm includes: using a rapidly expanding random tree (RRT) algorithm combined with an improved A algorithm for path planning, planning a feasible path from a starting point to a target point by using the rapidly expanding random tree (RRT) algorithm; and smoothing the path based on an improved heuristic function and a search strategy by the improved A algorithm.

6. The main bearing cap grinding control method based on machine vision and robot collaboration according to claim 2 is characterized in that: The method of controlling the robot to drive the grinding head to grind according to a predetermined path according to the grinding path includes: converting the planned path points into joint space trajectories recognized by the robot, using quintic polynomial interpolation or B-spline interpolation to smooth the motion trajectory, setting corresponding contact forces according to different grinding areas, and adjusting the rotation speed and feed rate of the grinding head.

7. The main bearing cap grinding control method based on machine vision and robot collaboration according to claim 2 is characterized in that: The real-time acquisition of the robot's motion state information includes: installing high-precision encoders at the robot's joints, the encoders obtaining real-time position, velocity, and acceleration information of each joint by recording the rotation angle and pulse number, deploying inertial measurement units at the robot's base and key positions, integrating accelerometers, gyroscopes, and magnetometers, and monitoring the robot's posture changes and linear motion parameters in three-dimensional space in real time.

8. The main bearing cap grinding control method based on machine vision and robot collaboration according to claim 2 is characterized in that: The step of obtaining the position information of the grinding head includes: installing a laser displacement sensor at the connection part of the grinding head, using the principle of laser triangulation to measure the position information of the grinding head relative to the workpiece or the reference coordinate system in real time, combining with the six-dimensional force sensor installed at the end of the robotic arm to sense the magnitude and direction of the contact force during the grinding process, and obtaining the working position and posture of the grinding head through comprehensive judgment through force feedback conversion.

9. The main bearing cap grinding control method based on machine vision and robot collaboration according to claim 8 is characterized in that: Also includes: The robot motion state information, grinding head position information and grinding surface image information collected by the sensor are preprocessed, and the multi-sensor information fusion technology is used to fuse the preprocessed motion state information, grinding head position information and grinding surface image information to construct a grinding work scene model.

10. The main bearing cap grinding control method based on machine vision and robot collaboration according to claim 2 is characterized in that: The automatic adjustment of the grinding path and grinding parameters by the feedback control algorithm includes: designing a feedback control algorithm based on the working scene model, adopting the model predictive control MPC algorithm, establishing a dynamic model of the grinding process according to the current grinding state and the preset grinding target, predicting the impact of the grinding path and parameter changes on the grinding effect in the future period, combining the adaptive control algorithm, and adjusting the control parameters in real time according to the actual grinding situation. When unevenness or defects are detected on the grinding surface, the grinding path is replanned by the path planning algorithm to avoid the defective area or focus on grinding the defects. According to the feedback of the grinding force and the image analysis results, the rotation speed, feed speed, and pressure grinding parameters of the grinding head are adjusted.

Images