Robot in-situ line laser measurement system and calibration method

Abstract

The invention discloses a robot in-situ line laser measurement system and a calibration method, a six-degree-of-freedom robot is adopted to carry a line laser sensor, a human-computer interaction interface is developed, and the measurement efficiency and reliability are improved; besides, in order to improve the measurement precision, a neural network optimized by an intelligent algorithm is adopted to compensate the positioning error of the tail end of the robot, a calibration model is constructed by measuring a standard cylinder, and solving is carried out through a penalty function method, so that the measurement attitude of the laser sensor is obtained, and the three-dimensional coordinates of a measurement point are analyzed. According to the invention, the measurement efficiency can be improved, and the measurement precision and reliability can be ensured, so that the adaptive capacity for the measurement requirements of aerospace large complex components is enhanced.

Description

TECHNICAL FIELD

[0001] The application belongs to the technical field of laser measurement and machining integration in intelligent manufacturing, relates to robot in-situ line laser measurement accuracy analysis and measurement path planning, and particularly relates to a robot in-situ line laser measurement system and a calibration method. BACKGROUND

[0002] In the laser measurement and machining integration technology, the workpiece profile information is detected in real time by a laser sensor, and a model for machining is generated, so the laser measurement accuracy and stability are particularly important. The current related background technologies mainly include robot error analysis, point laser sensor triangulation modeling and path generation technologies, but the efficiency is obviously limited. Therefore, a line laser in-situ measurement system calibration, error modeling and path planning method is needed to ensure the reliability of the measurement accuracy and improve the stability of the measurement path.

[0003] The prior art only targets the point light source form of the laser sensor, and the measurement efficiency is low; the original error formula is not applicable to the line laser sensor, and the coupling analysis of errors in the in-machine measurement process is not fully considered, and the in-machine measurement accuracy and efficiency of the line laser cannot be solved.

[0004] Therefore, when facing the detection of large and complex workpieces, the low measurement efficiency of the point laser sensor limits its application, so the line laser sensor is used as an information sensing means. However, the robot in-situ line laser measurement system has the technical problems of insufficient accuracy and suboptimal path.

[0005] According to the search, the Chinese invention patent with the publication number CN102825602A discloses an industrial robot self-calibration device based on a PSD, two PSDs are respectively installed on two plates of a V-shaped clamp, the V-shaped clamp is placed in the reachable range of the robot, the projection spot and the reflection spot of the laser beam emitted by the laser fixed at the end of the robot are respectively positioned at the center point positions of the two PSDs when the robot is at different positions, and the robot zero deviation self-calibration and the space pose self-calibration are performed through two virtual constraint lines. The self-calibration device has the advantages of simple structure, simple installation and operation, high positioning accuracy, and can simultaneously realize the self-calibration of the robot space pose and zero deviation; the application also provides an industrial robot self-calibration method based on a PSD.

[0006] The technical comparison between the present application and the above-mentioned patent is as follows:

[0007] "An industrial robot self-calibration device based on PSD": two PSDs (position sensitive detector) cooperate with a laser, through the projection / reflection of the laser beam to locate the center point of the PSD, establish two virtual constraint lines, and realize passive calibration of robot zero deviation and spatial pose. Its core relies on the geometric alignment of the physical spot, and does not involve error modeling or dynamic compensation.

[0008] The present application: an integrated system of line laser sensor and robot is constructed, the SSA-BP neural network is used to predict the end positioning error of the robot (the compensation accuracy is improved by 22.76%-54.55%), the standard cylindrical geometric constraint model and the penalty function method are combined to optimize, and high-precision solution of the hand-eye calibration matrix is realized (the distance from the center point of the ellipse to the axis of the cylinder is less than 0.2 mm). The technical scheme contains an active closed loop of "error modeling-compensation-calibration", which breaks through the limitation of PSD relying on geometric alignment only.

[0009] 2. Application scenarios and precision differences

[0010] "An industrial robot self-calibration device based on PSD": suitable for pose rough calibration of conventional industrial robots (such as repeated positioning accuracy repair), and the calibration accuracy is limited by the physical resolution of PSD (sub-millimeter level), which cannot meet the high-precision measurement demand of large curved surfaces in aerospace (such as aircraft skin) (the three-dimensional coordinate measurement accuracy of the present application reaches 0.1 mm level).

[0011] The present application: for complex curved surface in-situ measurement, through line laser scanning (the efficiency is more than 10 times of that of point laser) and error compensation, efficiency and accuracy are considered, and it is especially suitable for curved surface detection requiring dynamic path adjustment (such as a planning algorithm for measuring path adaptive curved surface curvature).

[0012] 3. Core technical innovation

[0013] "An industrial robot self-calibration device based on PSD": without error compensation model, the calibration result only reflects the static pose deviation, and cannot solve the kinematic cumulative error of the robot (such as joint flexibility deformation).

[0014] The present application: for the first time, the BP neural network optimized by squirrel search algorithm is introduced into robot positioning error compensation, a nonlinear mapping of joint angle-end error is established (6 joint angles are input, and 3-dimensional error components are output), dynamic error prediction is realized; at the same time, the hand-eye calibration adopts cylindrical surface constraint+RANSAC ellipse fitting, breaks through the sparseness of PSD single point positioning, and uses the high-density point cloud of line laser scanning to improve the robustness of calibration.

[0015] According to the search, the Chinese invention patent with publication number CN109794963A discloses a robot rapid positioning method for curved surface components, which is characterized by the following steps: 1) measuring the reference point by a laser tracker to establish a conversion model between the robot flange, the end effector and the vision measurement system coordinate system, and obtain the calibration parameters; 2) pasting a triangular auxiliary reflective target at the corner point of the curved surface component, the vision measurement system collects the image of the curved surface component, extracts the two-dimensional pixel point coordinates of the auxiliary reflective target and calculates the three-dimensional coordinates of the reflective target in the vision measurement system coordinate system; 3) calculating the current pose parameters of the robot according to the vision measurement coordinates of the curved surface component and the calibration parameters, and comparing with the target pose to calculate the robot motion amount. The application improves the reliability of hand-eye calibration parameters, quickly and accurately positions the information of the curved surface component, and can be widely used for rapid positioning of curved surface components in aviation, automobiles and other fields.

[0016] The technical comparison between the present application and the above-mentioned patent is as follows:

[0017] 1. Essential difference of technical solutions

[0018] "A robot rapid positioning method for curved surface components": relying on a laser tracker and a triangular auxiliary reflective target, the target coordinates are collected by a vision measurement system, and a robot-vision coordinate system conversion model is established. The core is "offline calibration + visual positioning" assisted by external equipment, which needs to paste targets on the curved surface and is not suitable for dynamic or non-contact scenes.

[0019] The present application: proposes in-situ laser self-calibration, without external tracker or target, scans a standard cylinder by the end-of-arm laser of the robot, constructs a constraint model by using the geometric invariance of the cylindrical surface (all scanning point clouds should be on the same cylindrical surface), and optimizes and solves the hand-eye matrix by combining the generalized Lagrange multiplier method (the number of iterations is reduced by 30%, and the convergence stability is improved). The technical solution realizes "sensor-robot" closed-loop self-calibration and is suitable for on-site measurement of large curved surfaces (such as in-situ detection of aircraft panel without disassembly).

[0020] 2. Difference in calibration efficiency and adaptability

[0021] "A robot rapid positioning method for curved surface components": multiple targets (such as curved surface corner points) need to be pasted in a single scene, the calibration time is prolonged with the increase of the number of targets (typically more than 30 minutes), and target occlusion may cause failure.

[0022] The present application: the standard cylinder can be randomly arranged in the robot workspace (only one cylinder is needed), and the calibration can be completed by scanning more than 5 groups of poses (within 10 minutes in actual measurement), and the RANSAC algorithm can automatically filter out noise points (such as scratches on the cylindrical surface), and is suitable for complex industrial environments.

[0023] 3. Breakthrough of measurement path planning

[0024] "Robot rapid positioning method for curved surface component": After positioning, the path depends on the preset trajectory, and the influence of the curvature of the curved surface on the measurement accuracy is not considered (such as the error caused by the deviation of the laser incidence angle on the steep slope).

[0025] The present application: On the basis of calibration, the laser incidence angle constraint is introduced (the optimal posture of the sensor is derived through the error model), and the spiral scanning path suitable for the characteristics of the curved surface is generated (such as the encryption sampling at the curvature mutation), the efficiency of the actual measurement path planning is improved by 40%, and the integrity of the measured point cloud is improved by 65%.

[0026] Summary: Unsubstitutability of the invention

[0027] 1. Error compensation dimension: The patents "Industrial robot self-calibration device based on PSD" and "Robot rapid positioning method for curved surface component" do not involve dynamic compensation of the end positioning error of the robot. The present application realizes joint level error prediction through SSA-BP neural network, which is the first of its kind.

[0028] 2. Calibration adaptability: The patent "Robot rapid positioning method for curved surface component" relies on external equipment and targets. The present application realizes true "in-situ self-calibration" based on the geometric self-constraint of line laser-cylinder, and is suitable for scenes such as aerospace where sticking labels is prohibited.

[0029] 3. Efficiency-precision balance: The trinity design of line laser scanning (efficiency) + error compensation (precision) + path planning (stability) breaks through the single technical focus of the comparison documents (such as the patent "Industrial robot self-calibration device based on PSD" focusing on recalibration speed, and the patent "Robot rapid positioning method for curved surface component" focusing on positioning accuracy), forming a systematic solution for large curved surface measurement. SUMMARY

[0030] To solve the above technical problems, the present application provides a robot in-situ line laser measurement system and a calibration method, which can improve the measurement efficiency, ensure the measurement accuracy and reliability, and enhance the adaptability to the measurement demand of large complex components in aerospace.

[0031] To achieve the above purpose, the technical scheme adopted by the present application is:

[0032] A robot in-situ line laser measurement system, characterized in that: it comprises a robot, a robot flange, a wire drawing sensor, a mounting bracket and a data acquisition card, the robot flange is installed at the end of the robot, the wire drawing sensor is installed on the mounting bracket, the wire drawing sensor is connected to the robot flange through a wire, and the data acquisition card is connected to the wire drawing sensor through a data line.

[0033] As a preferred technical scheme of the present application: four mounting holes are formed in the robot flange, which are A hole, B hole, C hole and D hole respectively, a straight rod spherical joint is mounted in each mounting hole, and the wire drawing sensor is connected to the straight rod spherical joint in each mounting hole through a wire.

[0034] A robot in-situ line laser measurement calibration method, characterized by comprising the following steps:

[0035] S1, a robot in-situ line laser measurement scene is arranged;

[0036] S2, a robot end position coordinate is measured;

[0037] S3, a neural network optimized by an intelligent algorithm is used to compensate for the positioning error of the robot end;

[0038] S4, a line laser sensor and a robot pose hand-eye calibration device are designed;

[0039] S5, a hand-eye calibration constraint model is established;

[0040] S6, feature point extraction and data processing are performed;

[0041] S7, a hand-eye calibration model is established and solved.

[0042] As a preferred technical scheme of the present application: step S1 is specifically as follows:

[0043] The wire drawing sensor is connected to the robot end through the robot flange, and a data acquisition card capable of simultaneously collecting two-dimensional measurement data of the robot end position and the wire drawing sensor is developed, the data acquisition card is connected to a man-machine interface, real-time display of data curves at different times is realized, and three-dimensional coordinate values of the measurement points are calculated for evaluating the measurement accuracy.

[0044] As a preferred technical scheme of the present application: step S2 is specifically as follows:

[0045] Four wire drawing sensors are used to convert the spatial position point information into length data, the displacement data collected are converted into three-dimensional positions through the man-machine interface combined with the corresponding settlement algorithm, and the calculation formula is as follows:

[0046] Suppose the position coordinates of the robot end measured point in the position measurement device coordinate system are , , , , the value of is determined by calculating the geometric relationship:

[0047] (1);

[0048] (2);

[0049] (3).

[0050] As a preferred technical scheme of the application: step S3 is specifically as follows:

[0051] S31, a BP neural network model based on squirrel algorithm optimization is proposed, a mapping relationship between joint angles of the robot and end error is constructed and learned and fitted, and the specific steps are as follows:

[0052] S311, squirrel algorithm

[0053] S3111, population initialization;

[0054] S3112, evaluate the fitness of each squirrel position;

[0055] S3113, sort the comfort of squirrel position in ascending order;

[0056] S3114, find squirrels on hickory trees, oak trees and ordinary trees;

[0057] S3115, update all squirrel positions;

[0058] S3116, determine whether all individuals have been updated, if yes, go to the next step, if not, return to step S3115;

[0059] S3117, calculate seasonal constant S c ;

[0060] S3118, determine whether seasonal constant S c ≤S min , if yes, go to the next step, if not, return to step S3112;

[0061] S3119, seasonal conversion, randomly reset the squirrel position according to the formula;

[0062] S31110, update S min value;

[0063] S31111, determine whether the end condition is met, if yes, enter the BP neural network model, if not, return to step S3112;

[0064] S312, BP neural network

[0065] S3121, determine the input and output vectors of the neural network, and normalize the sample data;

[0066] S3122, determine the neural network structure, select the activation function and the number of hidden layer nodes;

[0067] S3123, training the neural network;

[0068] S3124, judging whether the test network meets the generalization condition, if yes, entering the next step, if not, returning to step S3122;

[0069] S3125, saving the neural network structure, and the training is completed;

[0070] S32, the input layer of the BP neural network is set to 6 nodes, respectively corresponding to the angle values of the six joints of the robot, and the output layer is set to 3 nodes, for expressing the error components of the end position in the three-dimensional space;

[0071] S33, comparative analysis of the positioning error before compensation, BP network compensation and SSA-BP algorithm compensation.

[0072] As a preferred technical scheme of the application, step S4 is specifically as follows:

[0073] A standard cylinder-based calibration method is used for pose hand-eye calibration, specifically as follows:

[0074] S41, the standard cylinder is randomly arranged in the effective working space of the ABB robot;

[0075] S42, the spatial pose of the end effector of the ABB robot is adjusted, so that the plane emitted by the line laser sensor can intersect with the side surface of the standard cylinder, thereby forming an elliptical cross-sectional curve.

[0076] As a preferred technical scheme of the application, step S5 is specifically as follows:

[0077] S51, changing the end pose of the ABB robot;

[0078] S52, judging whether the laser intersects with the side surface of the standard cylinder, if yes, entering the next step, if not, returning to step S51;

[0079] S53, processing the profile data to extract the coordinates of the center of the ellipse;

[0080] S54, establishing a constraint optimization equation with the constraint that the center of the ellipse profile is on the axis of the cylinder;

[0081] S55, solving the constraint optimization equation to obtain an accurate calibration transformation matrix and the position of the standard cylinder.

[0082] As a preferred technical scheme of the application, step S6 is specifically as follows:

[0083] After obtaining the point cloud data of the standard cylinder side surface measurement, the original data is filtered and down-sampled, and since the intersection line of the laser plane and the standard cylinder side wall is in the shape of an ellipse, a robust ellipse curve fitting is introduced on the basis of the filtered point cloud to accurately extract the contour equation:

[0084] (4).

[0085] As a preferred technical scheme of the application, step S7 is specifically as follows:

[0086] S71, the spatial distance of the center point of the fitted ellipse contour to the cylinder axis is taken as an approximate evaluation standard of the distance between the overall contour and the surface of the standard cylinder, and the coordinates of the center point in the three-dimensional space are further calculated by means of the ellipse fitting result, and the specific solving process is shown in formula (5):

[0087] (5);

[0088] In the formula, each parameter is the coefficient of the ellipse equation;

[0089] S72, the center point of the ellipse is converted from the sensor coordinate system to the robot base coordinate system through homogeneous transformation of coordinates, and is obtained, and the specific expression is shown in formula (6):

[0090] (6);

[0091] In the formula, matrix represents the transformation matrix of the robot end coordinate system to the base coordinate system; represents the pose transformation matrix of the line laser sensor coordinate system relative to the robot end coordinate system, and the to-be-calibrated parameters are ;

[0092] S73, assuming that a point on the cylinder axis is known , the coordinates of the point in the robot base coordinate system are , and the direction vector of the cylinder axis is , then the parametric equation of the axis is represented as:

[0093] (7);

[0094] S74, the center point of the ellipse is connected with a point on the cylinder axis to form a line segment , and the projection length of the line segment on the cylinder axis is :

[0095] (8);

[0096] then the point is the square of the distance to the cylinder axis is:

[0097] (9);

[0098] After bringing equation (8) into equation (9), the value of can be obtained, and after introducing the cylinder axis, the calibration parameters are expanded to , wherein the unit vector constraint condition needs to be met as shown in equation (10):

[0099] (10);

[0100] By adjusting different poses of the robot end, a plurality of groups of laser scanning profile data can be collected, for each group of scanning data, the RANSAC algorithm is used to extract the ellipse center point , and the hand-eye transformation matrix and the parameters of the cylinder axis are preliminarily estimated, which are used as the initial values of subsequent optimization calculation;

[0101] S75, after obtaining the initial parameters, the square of the distance between each ellipse center point and the cylinder axis is further calculated , and all the ellipse center points should strictly fall on the cylinder axis, which is as follows:

[0102] Through iterative optimization, the estimated values of the hand-eye calibration matrix and the cylinder axis parameters are constantly updated, so that the square of the distance between each ellipse center point and the cylinder axis is gradually reduced and tends to zero, and thus the hand-eye calibration problem is converted into an optimization problem with a unit vector constraint, and the objective function is:

[0103] (11);

[0104] S76, by setting a penalty factor, the original constraint condition is introduced into the objective function as a penalty term, so that the original constraint optimization problem is equivalent to a solvable unconstrained optimization model, which is as follows:

[0105] The penalty factor is set by the generalized Lagrange multiplier method, which introduces and dynamically adjusts the value of the Lagrange multiplier when the penalty factor is moderate, so that the optimization process can gradually approach the optimal solution,

[0106] and the standard form of the constrained optimization problem is:

[0107] (12)

[0108] Where, for dimensional parameter vector; represents the objective function to be optimized; represents an equality constraint;

[0109] The generalized Lagrange multiplier method is used to solve constrained optimization problems. The specific steps are as follows:

[0110] S761, Initialization

[0111] Select the starting point , the multiplier vector , initial penalty factor , allowable error threshold , penalty factor adjustment coefficient , scale factor , and the number of iterations ;

[0112] S762, Unconstrained Optimization

[0113] Construct an augmented Lagrangian function and use the Powell algorithm to solve the unconstrained problem to obtain the optimal parameter value of the current iteration :

[0114] (13);

[0115] S763, termination judgment

[0116] If the current parameters meet the accuracy requirements, that is, , then terminate the iteration and output As the final solution, otherwise, go to the next step;

[0117] S764, Penalty Factor Update

[0118] Adjust the penalty factor according to the size of the constraint residual. If the constraint error is large, , then increase the penalty factor ,otherwise, ;

[0119] S765, Multiplier Update

[0120] make , , return and continue iterating until the termination condition is met.

[0121] Compared with the prior art, the present invention has the following beneficial effects:

[0122] The application adopts a six-degree-of-freedom robot to carry a line laser sensor, develops a human-computer interaction interface, and improves the measurement efficiency and reliability; in addition, in order to improve the measurement accuracy, a neural network optimized by an intelligent algorithm is used to compensate for the positioning error of the robot end, a calibration model is constructed by measuring a standard cylinder, a penalty function method is used to solve, and the measurement posture of the laser sensor is obtained, and then the three-dimensional coordinates of the measurement points are analyzed. The application can improve the measurement efficiency, ensure the measurement accuracy and reliability, and enhance the adaptability to the measurement demand of large complex components for aerospace. BRIEF DESCRIPTION OF DRAWINGS

[0123] Figure 1 It is a schematic diagram of the robot in-situ line laser measurement system structure in the application;

[0124] Figure 2 It is a schematic diagram of the straight rod spherical joint connection in the application;

[0125] Figure 3 It is a coordinate detection device construction flowchart in the application;

[0126] Figure 4 It is a data acquisition card platform schematic diagram in the application;

[0127] Figure 5 It is a schematic diagram of the line sensor calibration measurement in the application;

[0128] Figure 6 It is a BP neural network training flowchart in the application;

[0129] Figure 7 It is a BP neural network error compensation flowchart in the application;

[0130] Figure 8 It is a table of statistical data of position error in each direction in the application;

[0131] Figure 9 It is a hand-eye calibration scheme flowchart in the application;

[0132] Figure 10 It is a line laser scanning part profile in the application;

[0133] Figure 11 It is a schematic diagram of the robot end position coordinate measurement in the application;

[0134] Figure 12 It is a schematic diagram of the robot end position coordinate measurement in the application;

[0135] Figure 13 It is a comparison and analysis diagram of compensation before BP network compensation and SSA-BP algorithm compensation positioning error in the application;

[0136] Figure 14 Fig. 1 is a schematic diagram of a line laser sensor and a robot pose hand-eye calibration device according to the present application.

[0137] List of reference signs:

[0138] 1. robot; 2. robot flange; 3. wire sensor; 4. mounting bracket; 5. data acquisition card; 6. straight bar spherical joint; 7. sensor wire end; 8. guide rail; 9. mirror; 10. interference mirror; 11. laser interferometer; 12. line laser sensor; 13. standard cylinder; 14. cylinder center axis. DETAILED DESCRIPTION

[0139] The present application will be further described in conjunction with the accompanying drawings and specific embodiments:

[0140] As shown in Figures 1-5 , the present application proposes a robot in-situ line laser measurement system, which comprises a robot 1, a robot flange 2, a wire sensor 3, a mounting bracket 4, and a data acquisition card 5, the robot flange 2 is installed at the end of the robot 1, the wire sensor 3 is installed on the mounting bracket 4, the wire sensor 3 is connected to the robot flange 2 through a wire, and the data acquisition card 5 is connected to the wire sensor 3 through a data line.

[0141] Four mounting holes are provided on the robot flange 2, which are A hole, B hole, C hole and D hole respectively, a straight bar spherical joint 6 is installed in each mounting hole, and the wire sensor 3 is connected to the straight bar spherical joint 6 in each mounting hole through a wire.

[0142] A robot in-situ line laser measurement calibration method, comprising the following steps:

[0143] S1, the robot in-situ line laser measurement scene arrangement, specifically as follows:

[0144] The wire sensor 3 is connected to the end of the robot 1 through the robot flange 2, and a data acquisition card 5 capable of simultaneously collecting two-dimensional measurement data of the end position of the robot 1 and the wire sensor 3 is developed, the data acquisition card 5 is connected to a man-machine interface, real-time display of data curves at different times is realized, and three-dimensional coordinate values of measurement points are calculated for evaluating measurement accuracy.

[0145] S2, measurement of the end position coordinates of the robot 1, specifically as follows:

[0146] As shown in Figure 11 , four wire sensors 3 are used to convert spatial position point information into length data, and the displacement data collected are converted into three-dimensional positions through a man-machine interface combined with corresponding settlement algorithms, and the calculation formula is as follows:

[0147] Assuming that the position coordinates of the robot 1 end point measured point in the position measurement device coordinate system are , the value of , , is calculated through geometric relationship, and the position coordinates of the measured point are determined:

[0148] (1);

[0149] (2);

[0150] (3)

[0151] S3, the neural network optimized by the intelligent algorithm compensates the positioning error of the robot 1 end point, as shown in Figure 6 , 7 , 8, 12, and the specific steps are as follows:

[0152] S31, a BP neural network model based on squirrel algorithm optimization is proposed, the mapping relationship between the joint angles of the robot 1 and the end error is constructed, and learning and fitting are performed, and the specific steps are as follows:

[0153] S311, squirrel algorithm

[0154] S3111, population initialization;

[0155] S3112, evaluate the fitness of each squirrel position;

[0156] S3113, sort the comfort of squirrel position in ascending order;

[0157] S3114, find squirrels on hickory trees, oak trees and ordinary trees;

[0158] S3115, update all squirrel positions;

[0159] S3116, judge whether all individuals have been updated, if yes, go to the next step, if not, return to step S3115;

[0160] S3117, calculate the seasonal constant S c ;

[0161] S3118, judge whether the seasonal constant S c ≤S min , if yes, go to the next step, if not, return to step S3112;

[0162] S3119, season conversion, randomly reset the squirrel position according to the formula;

[0163] S31110, update the value of S min ;

[0164] S31111, determine whether the end condition is met, if so, enter the BP neural network model, if not, return to step S3112;

[0165] S312, BP neural network

[0166] S3121. Determine the neural network input and output vectors and normalize the sample data;

[0167] S3122. Determine the neural network structure, select the activation function and the number of hidden layer nodes;

[0168] S3123, training neural network;

[0169] S3124: Determine whether the test network meets the generalization condition. If so, proceed to the next step; if not, return to step S3122.

[0170] S3125, save the neural network structure, and the training is completed;

[0171] S32, setting the input layer of the BP neural network to 6 nodes, corresponding to the angle values ​​of the six joints of the robot 1, and setting the output layer to 3 nodes, used to express the error components of the end position in the three-dimensional space;

[0172] S33. Comparative analysis of positioning errors before compensation, BP network compensation and SSA-BP algorithm compensation.

[0173] like Figure 13 As shown in Figure 2, experimental results demonstrate that both the BP neural network alone and the SSA-BP neural network combined with SSA optimization can effectively improve positioning accuracy. Specifically, SSA-BP achieves 22.76%, 54.55%, and 39.85% improvements in accuracy in the three directions, respectively, compared to the original BP model. This demonstrates that the proposed SSA-BP-based robot positioning error compensation strategy significantly improves the absolute positioning accuracy of Robot 1, validating its feasibility and effectiveness in practical applications.

[0174] The design of S4, line laser sensor 12 and robot 1 posture hand-eye calibration device is as follows:

[0175] like Figure 14 As shown in the figure, a standard cylinder-based calibration method is used for hand-eye calibration, as follows:

[0176] S41, randomly arranging the standard cylinders 13 in the effective working space of the robot 1;

[0177] S42. Adjust the spatial posture of the end effector of the robot 1 so that the emission plane of the line laser sensor 12 intersects with the side surface of the standard cylinder 13, thereby forming an elliptical cross-sectional curve.

[0178] S5. Establish a hand-eye calibration constraint model. Since the standard cylinder's position in space remains unchanged, all collected contour data is distributed on the side surface of the same standard cylinder. Based on this geometric constraint relationship, a corresponding cylindrical surface constraint model can be constructed. Subsequently, the Lagrange multiplier method is combined to solve this constraint optimization problem, thereby obtaining an accurate hand-eye transformation matrix and the specific position of the cylinder in space.

[0179] like Figure 9 As shown, the details are as follows:

[0180] S51, changing the end position of robot 1;

[0181] S52, determine whether the laser intersects with the side surface of the standard cylinder 13, if so, proceed to the next step, if not, return to step S51;

[0182] S53, processing the contour data to extract the ellipse center coordinates;

[0183] S54, establishing a constrained optimization equation based on the constraint that the center of the ellipse contour is on the cylinder axis;

[0184] S55. Solve the constraint optimization equation to obtain the accurate calibration transformation matrix and the position of the standard cylinder 13.

[0185] S6. Feature point extraction and data processing, as follows:

[0186] After obtaining the point cloud data measured on the side of the standard cylinder 13, in order to reduce the influence of noise on the fitting results, the original data needs to be filtered and downsampled. Since the intersection line between the laser plane and the side wall of the standard cylinder 13 is elliptical, the random sampling consensus algorithm (RANSAC) is introduced on the basis of the filtered point cloud to perform robust elliptical curve fitting, so as to accurately extract its contour equation:

[0187] (4).

[0188] The pseudo code of the RANSAC algorithm is as follows:

[0189]

[0190] S7. Establish and solve the hand-eye calibration model as follows:

[0191] S71、Since the calibration method relies on the geometric characteristics of the side surface of the standard cylinder 13, the core purpose is to minimize the total distance from each point on the laser scanning profile to the side wall of the standard cylinder 13. In order to simplify the calculation process, the spatial distance between the center point of the fitted ellipse profile and the axis of the cylinder is usually taken as the approximate evaluation standard of the distance between the overall profile and the surface of the standard cylinder 13. With the help of the ellipse fitting result, the coordinates of the center point in the three-dimensional space can be further calculated, and the specific solving process is shown in formula (5):

[0192] (5);

[0193] In the formula, each parameter is the coefficient of the ellipse equation;

[0194] S72, the center point of the ellipse is converted from the sensor coordinate system to the robot base coordinate system through homogeneous transformation of coordinates, and is obtained, and the specific expression is shown in formula (6):

[0195] (6);

[0196] In the formula, the matrix represents the transformation matrix of the robot end coordinate system to the base coordinate system; represents the pose transformation matrix of the line laser sensor coordinate system relative to the robot end coordinate system, and the calibration parameters are . Due to the structural characteristics of the line laser sensor 12, the component of the measured center point of the ellipse along the Y axis in the sensor coordinate system is always zero. Therefore, the elements and in the matrix have no effect on the final solution, so the calibration parameters are ;

[0197] S73, assuming that a point on the axis of the cylinder is known , the coordinates of the point in the robot base coordinate system are , and the direction vector of the axis of the cylinder is , then the parametric equation of the axis is:

[0198] (7);

[0199] S74, the center point of the ellipse is connected with a point on the axis of the cylinder to form a line segment , and the projection length of the line segment on the axis of the cylinder is :

[0200] (8);

[0201] then the point Distance to the cylinder axis is:

[0202] (9);

[0203] After substituting equation (8) into equation (9), the value of can be obtained. After introducing the cylinder axis, the calibration parameters are extended to , where the unit vector constraint condition is shown in equation (10):

[0204] (10);

[0205] By adjusting different poses of the robot 1 end, multiple sets of laser scanning profile data can be collected. For each set of scanning data, the RANSAC algorithm is used to extract the ellipse center point , and the hand-eye transformation matrix and the cylinder axis parameters are preliminarily estimated, which are used as the initial values for subsequent optimization calculation;

[0206] S75, after obtaining the initial parameters, the distance square of each ellipse center point to the cylinder axis is further calculated , and all ellipse center points should strictly fall on the cylinder axis. Ideally, all ellipse center points should strictly fall on the cylinder axis, i.e. the distance should be close to zero. To achieve this goal, the following steps are taken:

[0207] Through iterative optimization, the estimated values of the hand-eye calibration matrix and the cylinder axis parameters are constantly updated, so that the distance square of each ellipse center point to the cylinder axis is gradually reduced and tends to zero. Thus, the hand-eye calibration problem is converted into an optimization problem with a unit vector constraint, and the objective function is:

[0208] (11);

[0209] The penalty function method is a strategy widely used to solve constrained nonlinear optimization problems. This method introduces the original constraint conditions into the objective function as a penalty term by setting a penalty factor, thereby converting the original constrained optimization problem into a solvable unconstrained optimization model. However, the penalty function method has obvious defects: in order to ensure that the constraint condition is strictly met, the penalty factor usually needs to be constantly increased in the iteration process, and eventually tends to infinity. This will cause the augmented objective function to have a "sick" phenomenon, reduce the numerical stability, and further affect the convergence speed and solving effect of the unconstrained optimization algorithm.

[0210] The generalized Lagrange multiplier method combines the advantages of the traditional Lagrange multiplier method and the penalty function method. When the penalty factor is moderate, the method introduces and dynamically adjusts the numerical value of the Lagrange multiplier, so that the optimization process can gradually approach the optimal solution without relying on a large penalty factor, thereby effectively avoiding the problem of ill-conditioned objective function in the penalty function method.

[0211] S76, by setting the penalty factor, the original constraint condition is introduced into the objective function as a penalty term, so that the original constrained optimization problem is equivalent to a solvable unconstrained optimization model, and the specific steps are as follows:

[0212] The generalized Lagrange multiplier method sets the penalty factor. When the penalty factor is moderate, the method introduces and dynamically adjusts the numerical value of the Lagrange multiplier, so that the optimization process can gradually approach the optimal solution without relying on a large penalty factor, thereby effectively avoiding the problem of ill-conditioned objective function in the penalty function method,

[0213] In addition, the introduction of the multiplier reduces the sensitivity of the algorithm to the initial parameters to some extent, improves the global convergence performance. This feature is particularly critical in the hand-eye calibration problem, which can significantly improve the stability and success rate of optimization solution, and provides strong support for obtaining accurate calibration results. The standard form of the constrained optimization problem is:

[0214] (12)

[0215] where, is a dimensional parameter vector; denotes the objective function to be optimized; denotes the equality constraint condition;

[0216] The generalized Lagrange multiplier method is used to solve the constrained optimization problem, and the specific steps are as follows:

[0217] S761, initialization

[0218] Select the initial point , the multiplier vector , the initial penalty factor , the allowed error threshold , the penalty factor adjustment coefficient , the proportion factor , and the iteration number ;

[0219] S762, unconstrained optimization

[0220] Construct augmented Lagrangian function, and solve the unconstrained problem by using Powell algorithm to obtain the optimal parameter value of the current iteration :

[0221] (13);

[0222] S763, termination determination

[0223] If the current parameter meets the accuracy requirement, i.e. , terminate the iteration and output as the final solution, otherwise, go to the next step;

[0224] S764, penalty factor update

[0225] Adjust the penalty factor according to the size of the constraint residual. If the constraint error is large, , increase the penalty factor , otherwise, ;

[0226] S765, multiplier update

[0227] Let , , return to continue iteration until the termination condition is met.

[0228] In the above steps, the Powell algorithm is a classic derivative-free optimization method, which mainly relies on the iterative update of the conjugate direction to accelerate the convergence process of searching for the optimal solution. The algorithm performs one-dimensional line search along several preset directions, and dynamically adjusts these directions in each iteration to approach the local minimum point. Initially, these directions can be coordinate axis directions or other linearly independent initial direction sets. As the iteration progresses, the algorithm will replace some directions that perform poorly or no longer have the conjugate property with new directions constructed based on the displacement between the current optimal point and the original starting point, in order to maintain the effectiveness and linear independence of the direction set. However, during the search process, some conjugate directions may lose their original conjugate properties due to increased linear correlation, which may lead to search failure. To solve this problem, Powell proposed an improved scheme to dynamically adjust the directions in the conjugate direction set, thereby ensuring the effectiveness of the search process and ultimately converging to the correct optimal solution. A significant advantage of the Powell algorithm is that it does not require the derivative of the objective function, so even if the objective function is not differentiable or the derivative is not continuous, the method still works normally. In addition, the algorithm has strong ability in local optimization problems and can quickly and accurately find the local optimal solution. Due to these characteristics, the Powell algorithm is considered as an efficient and practical direct search method, and is widely used in various practical scenarios, with high application value.

[0229] The above merely describes the preferred embodiments of the present application, but does not constitute any other form of limitation to the present application, and any modification or equivalent change made according to the technical essence of the present application still falls within the scope of the present application.

Claims

1. A robot in-line laser measurement system, characterized by: The invention comprises a robot (1), a robot flange (2), a wire drawing sensor (3), a mounting frame (4), and a data acquisition card (5), wherein the robot flange (2) is mounted on the end of the robot (1), the wire drawing sensor (3) is mounted on the mounting frame (4), the wire drawing sensor (3) is connected to the robot flange (2) via a wire drawing, and the data acquisition card (5) is connected to the wire drawing sensor (3) via a data line.

2. The robot in-line laser measurement system according to claim 1, characterized in that: The robot flange (2) is provided with four mounting holes, namely hole A, hole B, hole C and hole D, each of which is provided with a straight rod spherical joint (6), and the wire sensor (3) is connected to the straight rod spherical joint (6) in the four mounting holes via a wire.

3. A robot in-line laser measurement calibration method according to any one of claims 1-2, characterized in that: The steps include: S1, robot (1) in-position line laser measurement scene arrangement; S2, robot (1) end position coordinate measurement; S3, using a neural network optimized by an intelligent algorithm to compensate for the positioning error of the robot (1); S4, line laser sensor (12) and robot (1) posture hand-eye calibration device design; S5. Establish a hand-eye calibration constraint model; S6, feature point extraction and data processing; S7. Establish and solve the hand-eye calibration model.

4. A robot in-line laser measurement calibration method according to claim 3, characterized in that: Step S1 is specifically as follows: The wire sensor (3) is connected to the end of the robot (1) through the robot flange (2), and a data acquisition card (5) is developed that can simultaneously acquire the position of the end of the robot (1) and the two-dimensional measurement data of the wire sensor (3). The data acquisition card (5) is connected to a human-computer interaction interface, displays data curves at different times in real time, and calculates the three-dimensional coordinate values ​​of the measurement points for evaluating the measurement accuracy.

5. The robot in-line laser measurement calibration method according to claim 3, characterized in that: Step S2 is specifically as follows: Four wire sensors (3) are used to convert spatial position point information into length data. The collected displacement data is converted into three-dimensional position through the human-computer interaction interface combined with the corresponding settlement algorithm. The calculation formula is as follows: Assume that the position coordinates of the measured point at the end of the robot (1) in the coordinate system of the position measuring device are , calculated through geometric relationships , , The value of is used to determine the position coordinates of the measured point: （1）； （2）； （3）。 6. The robot in-line laser measurement calibration method according to claim 3, characterized in that: Step S3 is as follows: S31. A BP neural network model based on the squirrel algorithm optimization is proposed to construct the mapping relationship between the robot (1) joint angles and the terminal error and to learn and fit it. The details are as follows: S311, Squirrel Algorithm S3111, population initialization; S3112, evaluate the fitness of each squirrel's position; S3113. Sort the comfort levels of the squirrels' positions in ascending order; S3114. Find the squirrels in hickory trees, acorn trees, and regular trees; S3115, update all squirrel positions; S3116: Determine whether all individuals have been updated. If so, proceed to the next step. If not, return to step S3115. S3117, calculate the seasonal constant S c ; S3118, determine the seasonal constant S c ≤S min If yes, go to the next step, if no, return to step S3112; S3119, seasonal changes, randomly reset the squirrels' positions according to the formula; S31110, Update S min value; S31111, determine whether the end condition is met, if so, enter the BP neural network model, if not, return to step S3112; S312, BP neural network S3121. Determine the neural network input and output vectors and normalize the sample data; S3122. Determine the neural network structure, select the activation function and the number of hidden layer nodes; S3123, training neural network; S3124: Determine whether the test network meets the generalization condition. If so, proceed to the next step; if not, return to step S3122. S3125, save the neural network structure, and the training is completed; S32, the input layer of the BP neural network is set to 6 nodes, corresponding to the angle values ​​of the six joints of the robot (1), and the output layer is set to 3 nodes, used to express the error component of the end position in the three-dimensional space; S33. Comparative analysis of positioning errors before compensation, BP network compensation and SSA-BP algorithm compensation.

7. The robot in-line laser measurement calibration method according to claim 3, characterized in that: Step S4 is specifically as follows: A standard cylinder-based calibration method is used for hand-eye calibration, as follows: S41, randomly arranging the standard cylinders (13) in the effective working space of the robot (1); S42. By adjusting the spatial posture of the end effector of the robot (1), the emission plane of the line laser sensor (12) can intersect with the side surface of the standard cylinder (13), thereby forming an elliptical cross-sectional curve.

8. The robot in-line laser measurement calibration method according to claim 3, characterized in that: Step S5 is specifically as follows: S51, changing the end position of the robot (1); S52, determining whether the laser intersects with the side surface of the standard cylinder (13), if so, proceeding to the next step, if not, returning to step S51; S53, processing the contour data to extract the ellipse center coordinates; S54, establishing a constrained optimization equation based on the constraint that the center of the ellipse contour is on the cylinder axis; S55. Solve the constrained optimization equation to obtain the accurate calibration transformation matrix and the position of the standard cylinder (13).

9. The robot in-line laser measurement calibration method according to claim 3, characterized in that: Step S6 is specifically as follows: After obtaining the point cloud data of the side measurement of the standard cylinder (13), the original data is filtered and downsampled. Since the intersection line between the laser plane and the side wall of the standard cylinder (13) is elliptical, a random sampling consistency algorithm is introduced on the basis of the filtered point cloud to perform robust elliptical curve fitting, thereby accurately extracting its contour equation: （4）。 10. The robot in-line laser measurement calibration method according to claim 3, characterized in that: Step S7 is specifically as follows: S71. The spatial distance from the center point of the fitted elliptical contour to the cylinder axis is used as an approximate evaluation standard for the distance between the overall contour and the surface of the standard cylinder (13). With the help of the elliptical fitting result, the coordinates of its center point in three-dimensional space are further calculated. The specific solution process is shown in formula (5): （5）； The parameters in the formula are the coefficients of the ellipse equation; S72, the center point of the ellipse Through the homogeneous transformation of coordinates from the sensor coordinate system to the robot base coordinate system, we get , the specific expression is shown in formula (6): （6）； Among them, the matrix Represents the transformation matrix from the robot's end coordinate system to the base coordinate system; It represents the pose transformation matrix of the line laser sensor coordinate system relative to the robot end coordinate system. The parameters to be calibrated are ; S73. Assume that a point on the axis of the cylinder is known , its coordinates in the robot base coordinate system are , the direction vector of the cylinder axis is , then the parametric equation of the axis is expressed as: （7）； S74, the center point of the ellipse A point on the cylinder axis Connect to form line segments , its projected length on the cylinder axis for: （8）； Points The square of the distance to the cylinder axis for: （9）； After substituting formula (8) into formula (9), we can get After the cylindrical axis is introduced, the calibration parameters are expanded to , where the unit vector constraint condition must be satisfied as shown in formula (10): （10）； By adjusting the different postures of the end of the robot (1), multiple sets of laser scanning contour data can be collected. For each set of scanning data, the RANSAC algorithm is used to extract the center point of the ellipse. , preliminarily estimate the parameters of the hand-eye transformation matrix and the cylinder axis, and use them as the initial values ​​for subsequent optimization calculations; S75. After obtaining the initial parameters, further calculate the center point of each ellipse Square of the distance from the cylinder axis , and make the center points of all ellipses All should fall strictly on the axis of the cylinder, as follows: By iterative optimization, the hand-eye calibration matrix and the estimated values ​​of the cylinder axis parameters are continuously updated so that the center points of each ellipse are Square of the distance to the cylinder axis Gradually decreases and approaches zero, thus the hand-eye calibration problem is transformed into an optimization problem with unit vector constraints, and its objective function is: （11）； S76. By setting a penalty factor, the original constraint conditions are introduced into the objective function as a penalty term, thereby converting the original constrained optimization problem into a tractable unconstrained optimization model for solution, as follows: The penalty factor is set by the generalized Lagrange multiplier method. When the penalty factor remains moderate, the value of the Lagrange multiplier is introduced and dynamically adjusted, so that the optimization process can gradually approach the optimal solution. Its standard form for constrained optimization problems is: （12） Where, for dimensional parameter vector; represents the objective function to be optimized; represents an equality constraint; The generalized Lagrange multiplier method is used to solve optimization problems with constraints. The specific steps are as follows: S761, Initialization Select the starting point , the multiplier vector , initial penalty factor , allowable error threshold , penalty factor adjustment coefficient , scale factor , and the number of iterations ; S762, Unconstrained Optimization Construct an augmented Lagrangian function and use the Powell algorithm to solve the unconstrained problem to obtain the optimal parameter value of the current iteration : （13）； S763, termination judgment If the current parameters meet the accuracy requirements, that is, , then terminate the iteration and output As the final solution, otherwise, go to the next step; S764, Penalty Factor Update Adjust the penalty factor according to the size of the constraint residual. If the constraint error is large, , then increase the penalty factor ,otherwise, ; S765, Multiplier Update make , , return and continue iterating until the termination condition is met.

Images