Macro-micro combined robot machining system and machining track planning method thereof

Abstract

The application discloses a kind of macro-micro combination robot machining system and its machining track planning method in the field of robot intelligent manufacturing technology, including six degrees of freedom robot, three degrees of freedom high-precision machining center, line laser sensor measurement module;Six degrees of freedom robot is fixed on shock-absorbing pedestal by bolt, to realize the large stroke movement of machining center, three degrees of freedom high-precision machining center is fixed on six degrees of freedom robot end flange by bolt, realizes three directions high-precision positioning and processing by servo motor and electric spindle, line laser sensor measurement module is installed on three degrees of freedom high-precision machining center end manipulator;The application also discloses a kind of corner point extraction method based on fitting straight line and a kind of machining track planning method based on point cloud registration.The application has the advantages of large workspace and high machining precision, realizes intelligentization in processing, improves in-machine measurement and processing efficiency.

Description

TECHNICAL FIELD

[0001] The present application relates to a macro-micro combined robot machining system and a machining trajectory planning method thereof in the field of robot intelligent manufacturing technology. BACKGROUND

[0002] In the field of aerospace and future customized intelligent manufacturing, there are often single-piece small-batch production or batch production scenarios of large components and complex parts, which are suitable for robot machining. For the machining of large components, compared with the use of large five-axis machining centers with large occupied space and high operating cost, robot machining is flexible and easy to expand, can realize multi-robot collaborative machining, and is convenient for building an integrated intelligent system of measurement, machining and quality monitoring. For complex parts that need to be produced in single-piece or small batch, such parts often have uneven machining allowance distribution and high surface quality requirements, and some parts cannot be machined by machine tools and need the assistance of skilled workers. This problem can also be solved by building an intelligent robot machining system with autonomous learning and advanced perception. However, robots with larger workspaces often have low positioning accuracy, which affects the machining accuracy. SUMMARY

[0003] The purpose of the present application is to provide a macro-micro combined robot machining system and a machining trajectory planning method thereof, which can realize flexible, large workspace and high machining precision robot machining, and rely on a line laser sensor measurement module to extract features from the surface of the workpiece to be machined and generate machining trajectories through point cloud processing.

[0004] To achieve the above purpose, the present application provides a macro-micro combined robot machining system, which comprises a six-degree-of-freedom robot, a three-degree-of-freedom high-precision machining center and a line laser sensor measurement module. The six-degree-of-freedom robot is fixed on a shock-absorbing pedestal, the three-degree-of-freedom high-precision machining center is fixed on the flange at the end of the six-degree-of-freedom robot, a machining platform is arranged below the three-degree-of-freedom high-precision machining center, and the line laser sensor measurement module is installed on the end effector of the three-degree-of-freedom high-precision machining center.

[0005] Compared with the prior art, the present application has the advantages that the six-degree-of-freedom robot is used to realize the movement of the machining center between specified stations, large-stroke motion and pose adjustment, and the macro positioning position error is compensated by the three-degree-of-freedom high-precision machining center, so that the system has both a large workspace and high machining precision. The line laser sensor measurement module is a non-contact on-machine measurement technology, which has the advantages of low cost, high measurement accuracy, high acquisition efficiency and stable and reliable measurement results. It can intelligently identify the features to be machined according to the automatically generated point cloud, and realize autonomous machining of the small three-axis machining center.

[0006] As a further improvement of the present application, the three-degree-of-freedom high-precision machining center is provided with an X-axis moving pair, a Y-axis moving pair, a Z-axis moving pair and an end effector, the X-axis moving pair, the Y-axis moving pair and the Z-axis moving pair are all realized linear motion by servo motor driving screw nut, and the end effector is connected with the Z-axis moving pair.

[0007] In this way, the movement of the end effector in three directions is realized through the X-axis moving pair, the Y-axis moving pair and the Z-axis moving pair, thereby improving the working space of machining.

[0008] As a further improvement of the present application, the X-axis moving pair includes an X-axis motor, an X-direction static platform, the Y-axis moving pair includes a Y-axis motor, a Y-direction static platform, the Z-axis moving pair includes a Z-axis motor, a Z-direction static platform, the X-direction static platform is connected with the six-degree-of-freedom robot end flange, the X-axis motor is installed on the X-direction static platform and connected with the Y-direction static platform through a screw nut, the Y-axis motor is installed on the Y-direction static platform and connected with the Z-direction static platform through a screw nut, the Z-axis motor is installed on the Z-direction static platform and connected with a Z-direction motion platform of the end effector through a screw nut, and an electric spindle is arranged on the Z-direction motion platform and connected with a milling cutter.

[0009] In this way, each axis motor drives the linear motion of the corresponding platform through the screw nut mechanism, the Y-axis static platform can move in the X direction, the Z-axis static platform can move in the Y direction, and the electric spindle can ascend and descend in the Z direction through the Z-direction motion platform, so that the milling cutter can realize movement in a larger space range.

[0010] As a further improvement of the present application, the linear laser sensor measurement module includes a linear motor, an angle-adjustable rotary cylinder and a linear laser sensor, the linear motor is installed on the Z-direction motion platform, the angle-adjustable rotary cylinder is connected with the linear motor, and the linear laser sensor is installed on the angle-adjustable rotary cylinder, and the linear laser plane of the linear laser sensor is parallel to the Z-axis of the machining platform.

[0011] In this way, the linear motor and the electric spindle are installed on the same platform and can move up and down along the Z-axis, the linear laser sensor is installed on the angle-adjustable rotary cylinder and can move up and down along the Z-axis controlled by the linear motor, thereby realizing the adjustment of the visual depth of the sensor.

[0012] In order to achieve the above-mentioned purpose, the present application further provides a machining trajectory planning method of a macro-micro combined robot machining system, which comprises the following steps,

[0013] Step 1: obtaining original point cloud data by a linear laser sensor, judging the laser sensor deflection angle error by the sawtooth characteristics of the straight edges in the visual point cloud, and correcting the position of the linear laser sensor by adjusting the rotary cylinder adjusting bolt to ensure the position accuracy of the point cloud;

[0014] Step 2, scan the standard part to obtain a template point cloud data set for registration;

[0015] Step 3, divide the feature area of the workpiece to be measured, and form the target point cloud of different feature areas by overlapping scanning through the line laser sensor. Select the obvious vertical corner feature in the workpiece process reference area as the coarse registration reference point, and obtain the coarse registration reference point coordinates and the machining trajectory point coordinates in the template point cloud;

[0016] Step 4, perform point cloud preprocessing on the line laser point cloud by using the straight-through filtering and voxel filtering downsampling method without losing the scanning line features;

[0017] Step 5, segment the point cloud near the reference point, and use the fitting straight line angle point method optimized by inner product to extract the target point cloud reference point coordinates. Perform one-step translation coarse registration according to the difference between the target point cloud and the template point cloud reference point coordinates, and the translation transformation matrix output by the coarse registration is T c ;

[0018] Step 6: Perform iterative closest point fine registration using the template point cloud to obtain the registration transformation matrix T f , if the machining trajectory point coordinate matrix selected in the template point cloud is source P trace , the coordinates of the machining trajectory points in the target point cloud are obtained target P trace =T f T c source P trace , and the machining trajectory in the corresponding feature area is generated.

[0019] Compared with the prior art, the present application has the advantages of strong robustness and low time complexity. The voxel filtering downsampling method used in the point cloud preprocessing part can maintain the scanning line features of the line laser point cloud before and after downsampling. The inner product optimized fitting straight line angle point method used for point cloud feature point extraction relies on the edge angle of the corner feature to filter the fitting straight line, and has strong robustness for feature points with obvious corner features. The point cloud coarse registration uses the inner product optimized fitting straight line reference point method to perform one-step translation registration under the condition that the initial attitude difference of the point cloud is small. The method is simple and reliable, avoids the search algorithm and feature description algorithm with high time complexity, and reduces the maximum corresponding point distance required by the iterative closest point algorithm.

[0020] As a further improvement of the present application, the specific content of step 1 is as follows,

[0021] All points acquired by the line laser sensor at the same time are coplanar in the point cloud, i.e. belong to the same scanning line, when the laser plane is not perpendicular to the movement direction of the line laser sensor measurement module, the position error of the points in the point cloud occurs, when the included angle between the laser beam plane and the processing platform XOZ plane is θ, the distance from the rotating cylinder rotating shaft to the laser beam plane is r, and the measurement value of a point in the X direction of the sensor camera coordinate system is The absolute error of the three coordinates between the generated point in the point cloud and the actual measured point is e(x), e(y), e(z), and the position error of the point in the point cloud due to the laser deflection angle is The error is observed by the sawtooth feature of the straight line edge of the visualized point cloud and removed by adjusting the rotating cylinder adjusting bolt.

[0022] As a further improvement of the application, the specific content of step 4 is as follows,

[0023] Step 4.1, a three-dimensional voxel grid is generated, which is equivalent to a closely packed cube set, and the cube set just completely contains the input point cloud, so that a plurality of points are contained in each cube.

[0024] Step 4.2, the size of the voxel cube is accurately adjusted to be less than or equal to the spacing of the line laser point cloud scanning line, so that all points in the same voxel cube are coplanar, i.e. belong to the same scanning line.

[0025] Step 4.3, the centroid of all points in each cube is calculated, and the centroid point is used to approximate the points in the cube, and the centroid point belongs to the same scanning line, so that the down-sampled point cloud without losing the scanning line features of the original point cloud is obtained.

[0026] As a further improvement of the application, the specific content of step 5 is as follows,

[0027] Step 5.1, the point cloud near the corner point is segmented;

[0028] Step 5.2, a plane with the most inner points is fitted by random consistency sampling segmentation;

[0029] Step 5.3, the boundary points are judged by using the R-neighborhood maximum angle method based on normal estimation for the plane obtained in step 5.2;

[0030] Step 5.4, the segmentation boundary caused by step 5.1 is removed;

[0031] Step 5.5, the first straight line with the most inner points is extracted from the remaining boundary point cloud by using random consistency sampling fitting, and the inner points of the straight line are removed from the remaining boundary point cloud;

[0032] Step 5.6, a second straight line is fitted from the remaining point cloud, and the inner points of the second straight line are removed from the remaining point cloud;

[0033] Step 5.7, find the inner product of two straight line direction vectors;

[0034] Step 5.8: compare the inner product result with a threshold value, if within the threshold value, jump to 5.9, otherwise jump to 5.5;

[0035] Step 5.9, calculate the intersection point of the two spatial straight lines or a foot point of the common perpendicular line as the reference point coordinate output;

[0036] Step 5.10, according to the difference between the target point cloud and the template point cloud reference point coordinates, generate a translation transformation matrix, and translate the template point cloud to align with the target point cloud reference point, complete the coarse registration.

[0037] As a further improvement of the application, the specific content of step 5.2 is as follows,

[0038] The random consistency sampling algorithm records the indexes of the inner points and the outer points, as well as the number of in-plane points, which is used as a standard to evaluate the plane model. Then the process is repeated, and the minimum number of iterations required can be calculated by the expected error rate, the number of inner points, the number of original point clouds, and the number of sample subsets. The number is a probability value. Assuming that the total set of original point clouds is N, and the number of inner points is n inliers , the probability of selecting an inner point each time is Assuming that n points are needed for each sampling to estimate the model, the probability that these selected points are inner points is w n . Let the probability that the random consistency sampling algorithm only selects inner points be p, then 1-p=(1-w n ) k , the number of iterations is For plane model segmentation, n=3 is taken here. When the proportion of inner points in the original point cloud decreases, i.e. w decreases, the number of iterations increases. A reasonable distance threshold is set to reduce the number of iterations and increase the number of inner points. The number of inner points is guaranteed by the extraction of the region of interest in the previous step, and the number of inner points in the main feature plane can be made to account for more than 50% through straight-through filtering.

[0039] As a further improvement of the application, the specific content of step 5.7 is as follows,

[0040] In order to handle the noise points caused by the rough surface of the measured object, remove unnecessary straight lines, and on the basis of fitting two spatial straight lines to find the corner points, a method is proposed to optimize and remove redundant straight lines by taking the inner product of the straight line as the condition; the inner product of the two straight line direction vectors is calculated, and the threshold is set to-0.5≤ε i ≤0.5, i.e. the intersection angle between the two straight lines is between 60°≤α≤120°. BRIEF DESCRIPTION OF DRAWINGS

[0041] Figure 1This is a three-dimensional structural diagram of the present invention.

[0042] Figure 2 This is a partial schematic diagram of the present invention, which includes a three-degree-of-freedom high-precision machining center and a line laser sensor measurement module.

[0043] Figure 3 This is a schematic diagram of the sensor scanning path planning of the present invention.

[0044] Figure 4 This invention relates to the principle of voxel filtering downsampling without losing scan line features during point cloud processing.

[0045] Figure 5 The flowchart for extracting the region of interest in this invention is shown below.

[0046] Figure 6 Flowchart for extracting plane consistency for random sampling.

[0047] Figure 7 Flowchart for boundary extraction procedure.

[0048] Figure 8 Flowchart of a program for finding the corner points of two fitted lines for inner product optimization.

[0049] Figure 9 Flowchart of the coarse registration procedure for point clouds.

[0050] The components include: 1. Six-DOF robot; 2. X-axis prismatic joint; 3. Y-axis prismatic joint; 4. Z-axis prismatic joint; 5. End effector; 6. Linear laser sensor measurement module; 7. Machining platform; 2-1 X-axis motor; 2-2 X-axis stationary platform; 3-1 Y-axis motor; 3-2 Y-axis stationary platform; 4-1 Z-axis motor; 4-2 Z-axis stationary platform; 5-1 electric spindle; 5-2 Z-axis motion platform; 6-1 linear motor; 6-2 rotary cylinder; 6-3 linear laser sensor. Detailed Implementation

[0051] The present invention will be further described below with reference to the accompanying drawings:

[0052] like Figures 1-2 The macro-micro combined robotic machining system shown includes a six-degree-of-freedom robot 1, a three-degree-of-freedom high-precision machining center, and a line laser sensor 6-3 measurement module 6. The six-degree-of-freedom robot 1 is fixed on a vibration-damping platform, and the three-degree-of-freedom high-precision machining center is fixed on the end flange of the six-degree-of-freedom robot 1. A machining platform 7 is set below the three-degree-of-freedom high-precision machining center, and the line laser sensor 6-3 measurement module 6 is installed on the end effector 5 of the three-degree-of-freedom high-precision machining center.

[0053] The three-degree-of-freedom high-precision machining center is equipped with an X-axis prismatic joint 2, a Y-axis prismatic joint 3, a Z-axis prismatic joint 4, and an end effector 5. The X-axis prismatic joint 2, the Y-axis prismatic joint 3, and the Z-axis prismatic joint 4 are all driven by servo motors to achieve linear motion of the lead screw and nut. The end effector 5 is connected to the Z-axis prismatic joint 4.

[0054] X-axis gliding pair 2 includes X-axis motor 2-1 and X-axis stationary platform 2-2; Y-axis gliding pair 3 includes Y-axis motor 3-1 and Y-axis stationary platform 3-2; Z-axis gliding pair 4 includes Z-axis motor 4-1 and Z-axis stationary platform 4-2. X-axis stationary platform 2-2 is connected to the end flange of the six-degree-of-freedom robot 1. X-axis motor 2-1 is mounted on X-axis stationary platform 2-2 and connected to Y-axis stationary platform 3-2 via a lead screw and nut. Y-axis motor 3-1 is mounted on Y-axis stationary platform 3-2 and connected to Z-axis stationary platform 4-2 via a lead screw and nut. Z-axis motor 4-1 is mounted on Z-axis stationary platform 4-2 and connected to Z-axis motion platform 5-2 of the end effector 5 via a lead screw and nut. Z-axis motion platform 5-2 is equipped with an electric spindle 5-1, which is connected to a milling cutter.

[0055] The line laser sensor 6-3 measurement module 6 includes a linear motor 6-1, an angle-adjustable rotary cylinder 6-2, and a line laser sensor 6-3. The linear motor 6-1 is mounted on the Z-axis motion platform 5-2. The angle-adjustable rotary cylinder 6-2 is connected to the linear motor 6-1. The line laser sensor 6-3 is mounted on the angle-adjustable rotary cylinder 6-2. The line laser plane of the line laser sensor 6-3 is parallel to the Z-axis of the processing platform 7.

[0056] like Figures 3-9 The method for machining trajectory planning in a macro-micro combined robot machining system, as shown, includes the following steps:

[0057] Step 1: Acquire raw point cloud data using a line laser sensor. Determine the laser sensor's angular error by visualizing the jagged features of the straight line edges in the point cloud. Correct the line laser sensor's pose by adjusting the rotating cylinder's adjusting bolts to ensure the point cloud's positional accuracy.

[0058] All points acquired by the line laser sensor at the same time are coplanar in the point cloud, meaning they belong to the same scan line. When the laser plane is not perpendicular to the movement direction of the line laser sensor's measurement module, positional errors occur in the points in the point cloud. When the angle between the laser beam plane and the XOZ plane of the processing platform is θ, and the distance from the rotation axis of the rotary cylinder to the laser beam plane is r, the measured value of a certain point in the X direction of the sensor camera coordinate system is... The absolute errors of the three coordinates between the generated point in the point cloud and the actual measured point are e(x), e(y), and e(z). Then, the positional error of that point in the point cloud due to the laser deflection angle is... The error is observed by visualizing the sawtooth feature of the straight edge of the point cloud and removed by adjusting the rotation cylinder adjustment bolt.

[0059] Step 2, scan the standard part to obtain a template point cloud data set for registration;

[0060] Step 3, divide the feature area of the workpiece to be measured, and form the target point cloud of different feature areas by overlapping scanning through the line laser sensor. Select the obvious vertical corner feature in the workpiece process reference area as the coarse registration reference point, and obtain the coarse registration reference point coordinates and the machining trajectory point coordinates in the template point cloud;

[0061] Step 4, point cloud preprocessing of the line laser point cloud is performed by through filtering and voxel filtering downsampling method without losing the scanning line features;

[0062] Step 4.1, a three-dimensional voxel grid is generated, which is equivalent to a closely packed cube set, and the cube set completely contains the input point cloud, so that each cube contains a number of points.

[0063] Step 4.2, the voxel cube edge length is accurately adjusted to be less than or equal to the spacing of the line laser point cloud scanning line, so that all points in the same voxel cube are coplanar, i.e. belong to the same scanning line.

[0064] Step 4.3, the centroid of all points in each cube is calculated, and the centroid point is used to approximate the points in the cube, and the centroid point belongs to the same scanning line, so that the downsampled point cloud without losing the original point cloud scanning line features is obtained.

[0065] Step 5, segment the point cloud near the reference point, and use the fitting straight line angle point method optimized by inner product to extract the target point cloud reference point coordinates. One-step translation coarse registration is performed according to the reference point coordinate difference between the target point cloud and the template point cloud, and the translation transformation matrix output by the coarse registration is T c ;

[0066] Step 5.1, segment the point cloud near the corner point;

[0067] Step 5.2, randomly and uniformly sample to fit the plane with the most inliers;

[0068] The random and uniform sampling algorithm records the indexes of inliers and outliers, as well as the number of inliers in the plane, which is used as the standard for evaluating the plane model. Then the process is repeated, and the minimum number of iterations required can be calculated by the expected error rate, the number of inliers, the number of original point clouds, and the number of sample subsets. The number is a probability value. Assuming that the total set of original point clouds is N, and the number of inliers is n inliers , the probability of randomly selecting an inlier each time is Assuming that n points are needed to estimate the model each time sampling, the probability that these selected points are all inliers is w n , assuming that the probability that the random uniform sampling algorithm selects only inliers is p, then 1-p = (1-w n ) k , then the number of iterations is For planar model segmentation, here n = 3, when the proportion of inliers in the original point cloud decreases, that is, w decreases, the number of iterations increases; the distance threshold is set reasonably to reduce the number of iterations and increase the number of inliers, and the number of inliers is ensured by extracting the region of interest in the previous step, and the number of inliers in the main feature plane can be made to account for more than 50% through the straight-through filter.

[0069] Step 5.3, using the R-neighbor maximum angle method based on normal estimation to judge the boundary points of the plane obtained in step 5.2;

[0070] Step 5.4, remove the segmentation boundary caused by step 5.1;

[0071] Step 5.5, use random uniform sampling to fit and extract the straight line with the most inliers from the remaining boundary point cloud, and remove the inliers of the straight line from the remaining boundary point cloud;

[0072] Step 5.6, fit the second straight line from the remaining point cloud, and remove the inliers of the second straight line from the remaining point cloud;

[0073] Step 5.7, find the inner product of the direction vectors of the two straight lines;

[0074] In order to handle the noise points caused by the rough surface of the measured object, remove the unnecessary straight lines, and on the basis of the intersection point of the two spatial straight lines, a method is proposed to optimize and remove the redundant straight lines by setting the inner product of the straight lines as a condition; the inner product of the direction vectors of the two straight lines is set to be -0.5≤ε i ≤0.5, that is, the intersection angle of the two straight lines is between 60°≤α≤120°.

[0075] Step 5.8: compare the inner product result with the threshold value, if it is within the threshold value, go to step 5.9, otherwise go to step 5.5;

[0076] Step 5.9, calculate the intersection point or a foot point of the common perpendicular of the two spatial straight lines as the reference point coordinates output;

[0077] Step 5.10, generate a translation transformation matrix according to the difference between the reference point coordinates of the target point cloud and the template point cloud, translate the template point cloud, and align it with the reference point of the target point cloud to complete the coarse registration.

[0078] Step 6, use the template point cloud to perform iterative closest point fine registration to obtain the registration transformation matrix T fIf the coordinate matrix of the processing trajectory points selected in the template point cloud is source P trace This allows us to obtain the coordinates of the processing trajectory points in the target point cloud. target P trace =T f T c source P trace This generates the processing trajectory within the corresponding feature region.

[0079] In this invention, such as Figure 1 As shown, the system includes a six-degree-of-freedom robot 1, a three-degree-of-freedom high-precision machining center (including three sliding pairs and an electric spindle 5-2), and a line laser sensor measurement module 6. The six-degree-of-freedom robot 1 is bolted to a vibration-damping platform to enable the machining center to perform large-stroke movements; the three-degree-of-freedom high-precision machining center is bolted to the end flange of the six-degree-of-freedom robot 1 and has three sliding pairs.

[0080] like Figure 2 As shown, all linear motion is achieved by a servo motor driving a lead screw nut. The electric spindle 5-1, which serves as the end effector, is parallel to the Z-axis of the machining platform 7 and is fixed on the Z-axis motion platform 5-2. Different tools can be replaced according to different machining objects. The line laser sensor measurement module 6 is installed on the end effector 5 of the three-degree-of-freedom high-precision machining center. The line laser plane is parallel to the Z-axis of the machining platform 7 and the visual depth is adjusted by a linear motor 6-1.

[0081] The three-degree-of-freedom high-precision machining center consists of an X-axis motor 2-1, an X-axis stationary platform 2-2, a Y-axis motor 3-1, a Y-axis stationary platform 3-2, a Z-axis motor 4-1, and a Z-axis stationary platform 4-2. The end effector 5 consists of an electric spindle 5-1 and a Z-axis motion platform 5-2. Each axis motor drives the linear motion of the platform through a lead screw and nut mechanism. The Y-axis stationary platform 3-2 can move along the X-axis, the Z-axis stationary platform 4-2 can move along the Y-axis, and the electric spindle 5-1 can move up and down along the Z-axis.

[0082] The line laser sensor measurement module 6 consists of a linear motor 6-1, an angle-adjustable rotary cylinder 6-2, and a line laser sensor 6-3. The linear motor 6-1 is mounted on the same platform as the electric spindle 5-1 and can move up and down along the Z-axis. The line laser sensor 6-3 is mounted on the angle-adjustable rotary cylinder 6-2 and is controlled by the linear motor 6-1, allowing it to move up and down along the Z-axis to adjust the sensor's visual depth.

[0083] The six-degree-of-freedom robot 1 is used to enable the machining center to move between designated workstations, perform large-stroke movements, and adjust its posture. The macroscopic positioning error is compensated by the three-degree-of-freedom high-precision machining center.

[0084] The linear laser sensor measurement module 6 is synchronously measured by a three-degree-of-freedom high-precision machining center, and a high-precision surface point cloud of a workpiece to be machined is formed by relying on pulse counting of a motor encoder of the three-degree-of-freedom high-precision machining center.

[0085] The linear laser sensor measurement module 6 is installed on the three-degree-of-freedom high-precision machining center end effector 5, the linear laser plane is parallel to the Z-axis of the machining platform 7, the visual depth is adjusted through a linear motor 6-1, and the linear laser sensor 6-3 body is installed on the slider of the linear motor 6-1, and the visual angle is adjusted through a rotary cylinder 6-2.

[0086] The nine-axis macro-micro combined robot milling machining system based on point cloud processing has the six-degree-of-freedom robot 1 only used for realizing movement of the machining center between specified stations, large-stroke movement and pose adjustment, and the three-degree-of-freedom high-precision machining center is used for compensating macro positioning position error, so that the machining center has a larger working space and higher machining precision.

[0087] The nine-axis macro-micro combined robot milling machining system based on point cloud processing has the six-degree-of-freedom robot 1 only used for realizing movement of the machining center between specified stations, large-stroke movement and pose adjustment, and the three-degree-of-freedom high-precision machining center is used for compensating macro positioning position error, so that the machining center has a larger working space and higher machining precision.

[0088] When the machining system is used, the robot needs to be fixed on the shock-absorbing pedestal through bolts. Since the pose of the small three-axis CNC machining platform, the workpiece and the tooling platform is not determined, the tool coordinate system is calibrated by using the 4-point method to determine the TCP point of the tool, the tool coordinate system attitude is determined by using the ABC world coordinate system method, and then the pose of the workpiece coordinate system is determined by using the known TCP point and the 3-point method.

[0089] The six-degree-of-freedom robot is controlled by the control program stored in the host computer to move to the designated work station, and after the posture of the on-line laser sensor is adjusted, the linear motor controls the sensor to extend downward so that the measured feature is within the laser measurement range (75-125 mm). The host computer sends a command to turn on the laser and start triggering through 485 wired communication. At this time, the on-line laser sensor starts to continuously shoot and store the two-dimensional profile data of the measured part surface. The three-degree-of-freedom high-precision machining center moves along the Y-axis positive direction according to the programmed scanning path, and the Y-axis motor encoder pulse count is used as the trigger signal for taking the profile graph. Every time a certain number of pulses is reached, the host computer sends a profile graph taking instruction to the sensor, reads the latest two-dimensional profile data, and records the total displacement of the sensor along the Y-axis direction. Finally, the displacement of the sensor along the Y-axis direction and the two-dimensional profile data containing (X, Z) coordinates are combined to obtain the Z-direction perspective surface profile data of the measured part.

[0090] Specifically, the linear laser scanning path is as shown in Figure 3 Due to the characteristics of the linear laser, there will be missing points in the areas at both ends of the scanning line, so overlapping scanning is required for the part that may produce missing points to supplement the surface information. Then, in the point cloud processing process, the voxel filtering method is used for downsampling, removing repeated points, and reducing the amount of point cloud data. Therefore, the specific scanning path is: the linear laser sensor is driven by the CNC machining platform to move along the Y-axis positive direction in point-by-point motion, as path ① in Figure 3 , and the Y-axis motor encoder pulse count is used as the trigger for step scanning, with a step value of Δy, that is, the CNC machining platform moves Δy, and the linear laser sensor takes a profile picture. Obviously, the step value determines the density of the linear laser point cloud in the Y-axis direction, which has a decisive influence on the accuracy of the final surface profile point cloud. After the linear laser sensor moves a certain distance along the Y-axis direction, the scanning line forms a rectangular area, as shown by the dashed box in Figure 3 . Then the machine tool runs empty to return to the starting point of the scanning, as path ② in Figure 3 . The CNC machining platform drives the linear laser sensor to move along the X-axis positive direction in point-by-point motion, with a moving distance of Δx, as path ③ in Figure 3 , to prepare for the scanning of the next rectangular area. Assuming that the length of the scanning line in the field depth is L, when Δx < L, an overlapping area appears between the two areas, and the width of the overlapping area is L-Δx. Repeat the scanning of the rectangular area until the scanning of the entire measured feature area is completed. The scanning result presents as several rectangular areas with some overlapping areas, Figure 3 In order to make the areas clear, different styles of dashed lines are used to draw adjacent rectangular areas.

[0091] Specifically, the line laser sensor measurement module adopts a method of point cloud hard splicing based on CNC platform XY motor encoder pulse counting, that is, point cloud is organized based on the line laser sensor coordinate provided by the motor encoder, Figure 3 The coordinates of any point in the point cloud are shown in the figure, wherein (x C ,z C ) is the coordinate of the point in the line laser sensor camera coordinate system, (x0, y0, z0) is the coordinate of the origin of the camera coordinate system in the machine tool machining coordinate system when the line laser sensor is located at the starting point of scanning, and the point is located at the n-th scanning line in the scanning area, so the offset of the point relative to the starting point of the scanning line in the Y direction is nΔy, and the x0 and y0 coordinates are obtained by pulse counting of the CNC machining platform motor encoder.

[0092] The present application is not limited to the above-mentioned embodiments, and on the basis of the technical solutions of the present disclosure, those skilled in the art can make some substitutions and deformations to some technical features according to the disclosed technical content without creative labor, and these substitutions and deformations are all within the protection scope of the present application.

Claims

1. A machining trajectory planning method for a macro-micro combined robotic machining system, characterized in that: Includes the following steps, Step 1: Acquire raw point cloud data using a line laser sensor. Determine the laser sensor's angular error by visualizing the jagged features of the straight line edges in the point cloud. Correct the line laser sensor's pose by adjusting the rotating cylinder's adjusting bolts to ensure the point cloud's positional accuracy. Step 2: Scan the standard part to obtain the template point cloud dataset for registration; Step 3: Divide the workpiece to be tested into feature regions, and use a line laser sensor to perform overlapping scanning to form target point clouds of different feature regions; select obvious vertical corner features in the process reference region of the workpiece to be tested as coarse registration reference points, and obtain the coordinates of the coarse registration reference points and the coordinates of the processing trajectory points in the template point cloud. Step 4: Perform point cloud preprocessing on the line laser point cloud using a pass-through filtering and voxel filtering downsampling method that does not lose scan line features; Step 5: Segment the point cloud near the reference point, and extract the coordinates of the reference point in the target point cloud using the inner product optimization method for fitting straight lines to find corner points; perform a one-step coarse registration based on the difference between the reference point coordinates of the target point cloud and the template point cloud, and the translation transformation matrix output by the coarse registration is... ; Step 6: Perform iterative close-point fine registration using the template point cloud to obtain the registration transformation matrix. If the coordinate matrix of the processing trajectory points selected in the template point cloud is This allows us to obtain the coordinates of the processing trajectory points in the target point cloud. This generates the processing trajectory within the corresponding feature region; The specific details of step 1 are as follows: All points acquired by the line laser sensor at the same time are coplanar in the point cloud, meaning they belong to the same scan line. When the laser plane is not perpendicular to the movement direction of the line laser sensor measurement module, positional errors occur in the points in the point cloud. When the angle between the laser beam plane and the XOZ plane of the processing platform is θ, the distance from the rotation axis of the rotary cylinder to the laser beam plane is... The measured value of a certain point in the X direction of the sensor-camera coordinate system is The absolute errors of the three coordinates between the generated point in the point cloud and the actual measured point are e(x), e(y), and e(z). Then, the positional error of this point in the point cloud due to the laser deflection angle is... This error was observed by visualizing the jagged features of the straight edges of the point cloud and removed by adjusting the adjusting bolt of the rotating cylinder.

2. The machining trajectory planning method for a macro-micro combined robot machining system according to claim 1, characterized in that: Step 4 details are as follows: Step 4.1: Generate a 3D voxel grid, which is equivalent to a tightly packed set of cubes that completely contains the input point cloud, so that each cube contains several points. Step 4.2: Precisely adjust the side length of the voxel cube to be less than or equal to the spacing of the line laser point cloud scanning lines, so that all points in the same voxel cube are coplanar, i.e. belong to the same scanning line. Step 4.3: Calculate the centroid of all points within each cube, and approximate the points within the cube with the centroid. The centroid belongs to the same scan line, thus obtaining a downsampled point cloud without losing the scan line features of the original point cloud.

3. A machining trajectory planning method for a macro-micro combined robot machining system according to claim 2, characterized in that: The specific details of step 5 are as follows: Step 5.1, point cloud segmentation near corner points; Step 5.2: Random consistency sampling fits and segments the plane with the most interior points; Step 5.3: For the plane obtained in Step 5.2, use the R-nearest neighbor maximum included angle method based on normal estimation to determine the boundary points; Step 5.4: Remove the segmentation boundaries created in Step 5.1; Step 5.5: Use random consistency sampling to fit and extract the line with the first most interior points from the remaining boundary point cloud, and remove the interior points of the line from the remaining boundary point cloud. Step 5.6: Fit a second straight line from the remaining point cloud and remove the points inside the second straight line from the remaining point cloud; Step 5.7: Calculate the dot product of the direction vectors of the two lines; Step 5.8: Compare the inner product result with the threshold. If it is within the threshold, jump to 5.9; otherwise, jump to 5.

5. Step 5.9: Calculate the intersection point of the two spatial lines or a foot of their common perpendicular, and output the coordinates of the reference point. Step 5.10: Based on the coordinate difference between the reference points of the target point cloud and the template point cloud, generate a translation transformation matrix, translate the template point cloud to align it with the reference point of the target point cloud, and complete the coarse registration.

4. A machining trajectory planning method for a macro-micro combined robot machining system according to claim 3, characterized in that: The specific content of step 5.2 is as follows: The random consistency sampling algorithm records the indices of interior and exterior points, as well as the number of points in the plane, using these as criteria to evaluate the planar model. This process is repeated. The minimum number of iterations required can be calculated using the expected error rate, the number of interior points, the original point cloud points, and the number of points in the sample subset. This number is a probability value. Assuming the total number of original point clouds is N and the number of interior points is... The probability of randomly selecting an interior point each time is... ; Assuming each sampling requires n points to estimate the model, the probability that all selected points are interior points is: Let p be the probability that the random consistency sampling algorithm selects only interior points, then we have Then the number of iterations ; For planar model segmentation, here we take When the proportion of inliers in the original point cloud decreases, i.e., w decreases, the number of iterations increases. Setting a reasonable distance threshold can reduce the number of iterations and increase the number of inliers. The number of inliers is guaranteed by the region of interest extraction in the previous step. Through pass-through filtering, the proportion of inliers in the main feature plane can exceed [a certain threshold]. .

5. A machining trajectory planning method for a macro-micro combined robot machining system according to claim 4, characterized in that: The specific details of step 5.7 are as follows: To address noise points caused by uneven surfaces of the test piece and remove unwanted straight lines, a method for optimizing and eliminating redundant straight lines based on the inner product of straight lines is proposed, building upon the method of fitting straight lines in two spaces to find corner points. Take the inner product of the direction vectors of the two lines and set the threshold as follows. That is, the angle between the two lines is between 60° ≤ 𝛼 ≤ 120°.

Images