Three-dimensional system calibration method

Abstract

The application discloses a three-coordinate system calibration method, and belongs to the technical field of three-coordinate technology, comprising the following steps: arranging a plurality of reference components in the measuring space of a three-coordinate measuring machine, and setting a plurality of mark points on each reference component; continuously measuring each mark point to obtain reference coordinate data; establishing a space coordinate system according to the parameters of the measuring machine and the arrangement positions of the reference components, and calculating the theoretical coordinate data of each mark point; comparing the reference coordinate data with the theoretical coordinate data under a theoretical scenario to obtain the error values of each mark point in the X-axis, Y-axis and Z-axis directions, and constructing an error distribution function based on the error values; and the application improves the measurement reliability of the three-coordinate measuring machine in the high-precision field of mechanical manufacturing, aerospace and the like through the global coverage of the reference layout, the scene self-adaptive adjustment, the regional error modeling and the closed-loop verification mechanism, and is especially suitable for the measurement scenarios with high precision requirements for edge regions and complex working conditions.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of three-coordinate technology, in particular to a three-coordinate system calibration method. BACKGROUND

[0002] As a high-precision measuring device, the three-coordinate measuring machine directly depends on the accuracy of the coordinate system in the fields of mechanical manufacturing, aerospace, automobile industry, etc., and thus accurate calibration is the key to ensuring reliable measurement results.

[0003] The existing calibration method has many problems, for example, a single reference part (such as a standard ball, a calibration block) is often used, and due to the inherent defects of the measuring machine such as mechanical clearance and guide rail straightness error, it is difficult to fully reflect the error distribution in the measurement space, especially when the error in the edge area is significantly different from that in the center area, it cannot be effectively corrected. SUMMARY

[0004] In order to solve the problems of the prior art, the present application provides a three-coordinate system calibration method, which comprises the following steps:

[0005] A plurality of reference assemblies are arranged in the measurement space of the three-coordinate measuring machine according to a predetermined spatial distribution rule, and a plurality of mark points are arranged on each reference assembly;

[0006] The measurement reference coordinate data and the theoretical coordinate data of the mark points are obtained, and an error distribution function is constructed according to the relationship between the measurement reference coordinate data and the theoretical coordinate data;

[0007] The scene parameters of the measurement scene are adjusted, and the error distribution function is corrected based on the adjusted measurement scene; if there are two or more measurement scenes at the same time, the scene parameters of different measurement scenes are analyzed for multi-scene matching, and the scene parameters are adjusted based on the analysis results;

[0008] The compensation amount is calculated according to the corrected error distribution function and the three-coordinate measuring machine parameters, and the dynamic calibration is realized based on the compensation amount adjustment of the motion trajectory.

[0009] Further, the measurement reference coordinate data is obtained by controlling the measurement head to measure each mark point according to a predetermined trajectory, continuously measuring each mark point multiple times, and taking the average value as the actual measurement reference coordinate data;

[0010] The method for obtaining the theoretical coordinate data is to establish a spatial coordinate system according to the measuring machine parameters and the arrangement position of the reference assembly, and to calculate the theoretical coordinate data of each mark point;

[0011] The measurement scene at least includes a high-temperature scene mode, a high-speed scene mode and a heavy-load scene mode;

[0012] When the temperature of any point in the measurement space is greater than or equal to the set temperature, the high-temperature scene mode is started;

[0013] When the servo motor feedback speed is greater than or equal to the set speed, the high-speed scene mode is started;

[0014] When the workpiece load sensor feedback mass is greater than or equal to the set weight, the heavy load scene mode is started;

[0015] The scene parameters of the different measurement scenes are subjected to multi-scene matching analysis, and scene parameter adjustment is performed based on the analysis results, including:

[0016] The measurement input parameters are obtained, the input parameters are classified according to the input parameters, and are assigned to the corresponding measurement scenes, the scene parameters with high error proportion under multiple measurement scenes are obtained, and the measurement scene corresponding to the scene parameters with high error proportion is optimized to adjust the scene parameters;

[0017] The scene parameters of the measurement scene are adjusted, and further include: if there are two or more measurement scenes at the same time, the load parameters of the multiple measurement scenes are first adjusted;

[0018] The error distribution function is corrected based on the adjusted measurement scene, including:

[0019] The temperature, humidity and air pressure data of the adjusted measurement scene are collected in real time by the sensor, input into the compensation model, and the coordinate correction amount of the marker point is output, and the corrected error distribution function is obtained based on the coordinate correction amount;

[0020] The compensation amount is calculated according to the corrected error distribution function and the measuring machine parameters, and the dynamic calibration is realized based on the compensation amount adjustment of the motion trajectory.

[0021] Further, the reference assembly includes a fixed reference unit and a dynamic reference unit;

[0022] The fixed reference unit is arranged according to a predetermined spatial distribution rule, and the dynamic reference unit is arranged along the key nodes of the measuring machine motion trajectory and can slide along the guide rail,

[0023] The marker points of the fixed reference unit are processed by micro-nano level ruling process, and the marker points of the dynamic reference unit are integrated with micro temperature sensors.

[0024] Further, the measurement space is a three-dimensional rectangular space defined based on X-axis, Y-axis and Z-axis, and the measurement space is a cuboid or a rectangular cuboid structure, in which twelve edges are intersection lines formed by the intersection of six faces, and are divided into three groups in parallel to the coordinate axes, each group has four edges.

[0025] Further, the position of the fixed reference unit is distributed at eight corner points of the measurement space, a cube reference block is arranged at each corner point, a cylindrical reference column is arranged at the midpoint of each of the twelve edges, a conical reference cone is arranged at the center of each of the six faces, and a standard sphere is arranged at the center position;

[0026] The dynamic reference unit adopts a magnetic attraction type connection structure, and a group of two symmetrically distributed spherical reference elements are arranged every 500mm along the guide rails in the directions of the three coordinate axes; the sliding range of the dynamic reference unit covers more than 95% of the measurement space.

[0027] Further, the error value is obtained by comparing the coordinate data, and the error distribution function in the X-axis direction is:

[0028] ΔX(x,y,z)=a1x+b1y+c1z+d1xy+e1xz+f1yz+g1x 2 +h1y 2 +i1z 2 ;

[0029] where ΔX(x,y,z) represents the X-direction error value of the three-coordinate system at any point (x,y,z) in space;

[0030] x, y, and z respectively represent the coordinate values of the point on the X-axis, Y-axis, and Z-axis of the three-coordinate system;

[0031] a1, b1, c1, d1, e1, f1, g1, h1, and i1 represent fitting coefficients.

[0032] Further, the calibration method further comprises a multi-factor compensation model;

[0033] The multi-factor compensation model adopts a hierarchical structure, and the bottom layer is a physical mechanism model and the upper layer is a data-driven model;

[0034] The physical mechanism model is used to input temperature, humidity, and air pressure parameters to calculate a basic correction amount;

[0035] The data-driven model is used to input vibration frequency, power fluctuation, and bottom layer correction residual as input, output dynamic compensation amount through LSTM network, and obtain final coordinate correction amount through weighted fusion of two layer results.

[0036] Further, in the data acquisition step, strain sensors and laser interferometers are arranged inside the measurement head to collect deformation data of the probe under different measurement angles and touch forces in real time, and a probe deformation error model is established;

[0037] The model is linked with the reference coordinate data acquisition process, and when the measurement head trigger angle deviates from the preset threshold value, the deformation compensation algorithm is automatically called to correct the measurement value.

[0038] Further, in each scene mode, three reference mark points not involved in fitting are selected for verification measurement immediately after parameter adjustment is completed;

[0039] If the deviation of the error calculation value of the verification point and the actual measurement value is ≤0.03 mm, it is determined that the adjustment is effective;

[0040] If the deviation is > 0.03 mm, the control module automatically triggers parameter backtracking, restores the current scene parameters to the optimal value before adjustment, and starts supplementary sampling, recalculates the adjustment parameters based on the expanded data set, and repeats the verification until the deviation requirement is met.

[0041] Further, based on the fact that the error characteristics of each point in the measurement space are different, the measurement space is divided into three subspaces, namely a central core area, an edge transition area and a boundary constraint area;

[0042] The central core area is a spherical area with the geometric center of the measurement space as the spherical center and 1 / 3 of the space diagonal as the radius;

[0043] The edge transition area is an annular area between the core area and the boundary;

[0044] The boundary constraint area is a thin layer area close to the inner wall of the measurement space;

[0045] An error transmission coefficient matrix T between the subspaces is set, wherein T[i][j] represents the influence weight of the error of the ith subspace on the jth subspace, 0≤T[i][j]≤1 and ΣT[i][j]=1;

[0046] The core area error transmission coefficient satisfies T[1][1]≥0.8; the edge transition area satisfies T[2][1]+T[2][3]≥0.6; and the boundary constraint area satisfies T[3][3]≥0.7 and T[3][2]≤0.2;

[0047] When constructing the error distribution function, different polynomial degrees are used for different subspaces: a 3rd order polynomial is used for the core area, a 2nd order polynomial is used for the edge transition area, and a 1st order polynomial is used for the boundary constraint area, and the coupling correction of the subspace error is realized through the transmission coefficient matrix.

[0048] The beneficial effects of the present application are:

[0049] 1. The application adopts a layout combining fixed reference units (covering space angle points, edges, face centers and core areas) and dynamic reference units (sliding along the guide rail, covering more than 95% of the area), breaking through the limitation of single reference part that is difficult to reflect the global error. Through the dynamic reference of micro-nano scale line mark points and integrated sensors, the error difference between the center area and the edge area in the measurement space can be accurately captured, especially solving the problem of insufficient error correction of the edge area in traditional methods.

[0050] 2. The application designs different scene modes for different working conditions such as high temperature, high speed and heavy load, and establishes a priority adjustment mechanism when multiple scenes are triggered (such as the load priority guarantee principle under high temperature and high speed scenes), so that the calibration process can dynamically adapt to the real-time running state of the measuring machine. Combined with the multi-factor compensation model of temperature, humidity, air pressure, vibration and power fluctuation, the measurement precision stability in complex environment is significantly improved.

[0051] 3. The application divides the core area, edge transition area and boundary constraint area based on the topological structure of the measurement space, fits the error characteristics of the corresponding area through differentiated polynomial order, and realizes the coupling correction of the subspace by using the error transmission coefficient matrix, which accurately matches the nonlinear, linear drift and other characteristics of the error of different areas, and improves the error description precision compared with the traditional unified model.

[0052] Overall, the application improves the measurement reliability of three-coordinate measuring machines in high-precision fields such as mechanical manufacturing and aerospace through the global coverage of the reference layout, scene adaptive adjustment, regional error modeling and closed-loop verification mechanism, especially suitable for measurement scenes with high edge area precision requirements and complex working conditions. BRIEF DESCRIPTION OF DRAWINGS

[0053] Fig. 1 The overall principle framework diagram provided by the application;

[0054] Fig. 2 The compensation model principle framework diagram provided by the application;

[0055] Fig. 3 The space partition error modeling principle framework diagram provided by the application. DETAILED DESCRIPTION

[0056] The technical solutions in the embodiments of the application will be described clearly and completely below with reference to the drawings in the embodiments of the application. Obviously, the described embodiments are only part of the embodiments of the application, not all. Based on the embodiments in the application, all other embodiments obtained by those skilled in the art without creative labor are within the scope of protection of the application.

[0057] Please refer to Figs. 1-3The application provides a three-coordinate system calibration method, which comprises the following steps:

[0058] A plurality of reference assemblies are arranged in a measurement space of a three-coordinate measuring machine according to a preset spatial distribution rule, and a plurality of mark points are arranged on each reference assembly;

[0059] Reference coordinate data acquisition: a measurement head is controlled to measure each mark point according to a preset track, each mark point is measured multiple times in succession, and an average value is taken as reference coordinate data obtained by actual measurement;

[0060] Theoretical coordinate data acquisition: a space coordinate system is established according to parameters of the measuring machine and arrangement positions of the reference assemblies, and theoretical coordinate data of each mark point is calculated;

[0061] In a theoretical scenario, the reference coordinate data and the theoretical coordinate data are compared to obtain error values of each mark point in X-axis, Y-axis and Z-axis directions, and an error distribution function is constructed based on the error values;

[0062] Based on a current scenario, different scenario modes are started, and the error distribution function is adjusted correspondingly based on the corresponding scenario modes;

[0063] The measurement scenarios at least include a high-temperature scenario mode, a high-speed scenario mode and a heavy-load scenario mode;

[0064] When the temperature of any point in the measurement space is greater than or equal to a set temperature, the high-temperature scenario mode is started;

[0065] When the feedback speed of a servo motor is greater than or equal to a set speed, the high-speed scenario mode is started;

[0066] When the feedback mass of a workpiece bearing sensor is greater than or equal to a set weight, the heavy-load scenario mode is started;

[0067] Multi-scenario matching analysis is performed on scenario parameters of the different measurement scenarios, and scenario parameter adjustment is performed based on the analysis results, including:

[0068] Input parameters are obtained, the parameters are classified according to the input parameters, and the parameters are allocated to corresponding measurement scenarios, scenario parameters with high error proportions in multiple measurement scenarios are obtained, and the measurement scenarios corresponding to the scenario parameters with high error proportions are optimized to adjust the scenario parameters;

[0069] Adjusting the scenario parameters of the measurement scenarios also includes: if two or more measurement scenarios exist at the same time, the load parameters in the multiple measurement scenarios are first adjusted;

[0070] The error distribution function is corrected based on the adjusted measurement scenarios, including:

[0071] The temperature, humidity and air pressure data of the adjusted measurement scene are collected in real time by the sensor, input into the compensation model, and the coordinate correction amount is output, and the corrected error distribution function is obtained based on the coordinate correction amount;

[0072] The compensation amount is calculated based on the corrected error distribution function and the measuring machine parameters, and the dynamic calibration is realized by adjusting the motion trajectory based on the compensation amount;

[0073] The scene parameters of the measurement scene are adjusted, and the error distribution function is corrected based on the adjusted measurement scene. If there are two or more measurement scenes at the same time, multi-scene matching analysis is performed on the scene parameters of different measurement scenes, and scene parameter adjustment is performed based on the analysis result;

[0074] The compensation amount is calculated based on the corrected error distribution function and the three-coordinate measuring machine parameters, and the dynamic calibration is realized by adjusting the motion trajectory based on the compensation amount.

[0075] When two or more scene modes are triggered at the same time, the controller analyzes the scene parameters corresponding to the triggered scene, and adjusts the scene parameters with high error proportion based on the analysis result;

[0076] When the high-temperature and high-speed scenes are started at the same time, the load adjustment is prioritized, and then the corresponding parameters are optimized and adjusted according to the priority of high temperature>high speed;

[0077] Since the measurement accuracy of the three-coordinate measuring machine depends on the stability of the mechanical structure (guide rail, screw, support frame), the load directly acts on the mechanical core component, and its influence is fundamental and irreversible:

[0078] When the load exceeds the limit, it will cause the guide rail to bend, the screw pitch to deform, and the support frame to deform slightly. These are structural errors that, once occurred, will directly change the geometric reference of the measurement space, and this deformation cannot be reversed by subsequent temperature and speed parameter adjustment. Therefore, ensuring load adjustment is to maintain the mechanical accuracy reference of the measuring machine, and to provide a reliable premise for subsequent temperature and speed parameter optimization.

[0079] The error sources of high-temperature and high-speed scenes are different, and their influence range and duration on the measurement space are significantly different, which determines that high temperature needs to be corrected first:

[0080] High temperature error: global and persistent effect on the accuracy reference itself. High temperature scenarios affect all key components of the measuring machine through the "thermal expansion effect: the guide length changes due to thermal expansion, the reference assembly marker point coordinates drift due to thermal expansion, and the measuring head probe deforms due to thermal expansion and contraction. This error is globally covered (the coordinate reference of the entire measurement space is offset), and continuously accumulates (if the temperature does not decrease, the error will continuously expand), and if not corrected in advance, it will cause the error correction of the subsequent high-speed scene to lose the accurate coordinate reference.

[0081] High-speed error: instantaneous and local effect on the accuracy of the motion process. The high-speed error of the high-speed scene is mainly caused by dynamic response lag. When the servo motor moves at high speed, inertia will cause the measuring head to be positioned with overshoot, and the guide will vibrate to produce instantaneous deviation. This error is instantaneous (only occurs at the moment of high-speed movement) and local (only affects the measurement points along the motion trajectory), and can be quickly alleviated by reducing the speed and optimizing the acceleration and deceleration curve. Compared with the global destruction of the reference by high temperature, the influence range of high-speed error is smaller and more reversible. Under the premise of correcting the high temperature error, optimizing the high-speed parameters can more accurately eliminate dynamic interference and will not cause correction failure due to reference offset.

[0082] When arranging the reference assembly in the measurement space, the marker point position should be planned according to the principles of uniform distribution and key node strengthening in combination with the travel range of the measuring machine, to ensure that the spatial distance between adjacent marker points does not exceed 100 times the minimum resolution of the measuring machine, so as to avoid error interpolation distortion.

[0083] In the reference coordinate acquisition stage, the number of continuous measurements for each marker point is set to 5-10 times, and the random error can be controlled within 0.001 mm by taking the mean value after removing the abnormal values outside 3 times the standard deviation. The theoretical coordinate calculation needs to call the factory parameters of the measuring machine (such as guide spacing, screw pitch, reduction ratio, etc.), and establish the coordinate system in combination with the installation and positioning error of the reference assembly (≤0.002 mm). When constructing the error distribution function, the cross errors between X, Y, and Z axes (such as the perpendicularity error of the XY plane) need to be covered. In the scene mode adjustment, the temperature threshold is usually set to 30°C (the material thermal expansion coefficient changes significantly when the environmental temperature exceeds this value), the servo motor speed threshold is 80% of the rated speed, and the workpiece weight threshold is 70% of the rated load of the measuring machine (to prevent structural deformation from exceeding the limit). In the multi-factor compensation link, the temperature and humidity, and air pressure data are collected at a frequency of 10 Hz, the response delay of the compensation model is controlled within 50 ms to ensure the real-time performance of dynamic calibration, and finally the calibration accuracy of any point in space is improved to within ±0.003 mm by correcting the pulse equivalent of the motion controller to adjust the trajectory.

[0084] The compensation model can be implemented by an existing model, for example, the model is composed of a physical mechanism layer and a data-driven layer, and global error correction is realized by dynamic weighted fusion.

[0085] For example, taking thermal error compensation of a high-precision three-coordinate measuring machine as an example, the physical mechanism is based on a thermal expansion formula to calculate a basic correction amount: ΔT phy =α·ΔT·L; wherein, α is a material thermal expansion coefficient, ΔT is a temperature change, and L is a guide rail length; the data-driven layer adopts a 3-layer LSTM network, input parameters include a temperature gradient, a vibration frequency spectrum (1-1000 Hz), and a physical layer residual error, and an output dynamic compensation amount ΔT final =0.7·ΔT phy +0.3·ΔT data .

[0086] In some embodiments, the reference assembly includes a fixed reference unit and a dynamic reference unit;

[0087] The fixed reference unit is arranged according to a preset spatial distribution rule, and the dynamic reference unit is arranged at key nodes along a motion trajectory of the measuring machine and can slide along a guide rail,

[0088] The mark points of the fixed reference unit are processed by a micro-nano scale ruling process, and the mark points of the dynamic reference unit are integrated with micro temperature sensors.

[0089] The fixed and dynamic combined architecture realizes full-dimensional coverage of the measurement space. The arrangement of the fixed reference unit needs to be based on the geometric characteristics of the measurement space, for example, in a cuboid space of 1000mm×800mm×600mm, reference units need to be arranged at 8 corner points, 12 edge midpoint (500mm, 400mm, 300mm away from the edge end), and 6 face centers (500mm, 400mm away from each side), and the material of the reference units is selected from low thermal expansion coefficient invar (linear expansion coefficient ≤1.5×10 -6The dynamic reference unit adopts a high-precision linear guide rail slider structure, and the movement accuracy along the X, Y and Z axis rails is ≤0.002 mm / 100 mm, the adsorption force of the magnetic attraction type connection is ≥50 N (to prevent loosening during measurement), the measurement range of the NTC temperature sensor integrated with the mark point is -10℃~80℃, the accuracy is ±0.1℃, and the temperature data can be output to the controller in real time, so as to provide a data source for local error correction in a high-temperature scene. The micro-nano level scale line of the fixed reference adopts a femtosecond laser processing technology, the scale line width is ≤500 nm, the depth is ≤1 μm, the edge roughness Ra is ≤10 nm, and the identification accuracy of the optical measuring head is ensured; the spherical mark point of the dynamic reference adopts a ceramic material (hardness HRC60 or more), and the surface roundness is ≤0.0005 mm, so that the deformation amount when the measuring head is contacted is reduced.

[0090] In some embodiments, the measurement space is a three-dimensional rectangular space defined based on an X axis, a Y axis and a Z axis, and the measurement space is a cubic or cuboid structure, wherein twelve edges are intersection lines formed by the intersection of six surfaces, and are divided into three groups in the direction parallel to the coordinate axes, and each group has four edges.

[0091] In some embodiments, the position of the fixed reference unit is distributed at eight corner points of the measurement space, a square reference block is arranged at each corner point, a cylindrical reference column is arranged at the midpoint of each of the twelve edges, a conical reference cone is arranged at the center of each of the six surfaces, and a standard ball is arranged at the center position.

[0092] The dynamic reference unit adopts a magnetic attraction type connection structure, and a group of two symmetrically distributed spherical reference elements are arranged every 500 mm along the guide rails in the direction of the three coordinate axes; the sliding range of the dynamic reference unit covers more than 95% of the measurement space.

[0093] In some embodiments, the error value is obtained by comparing the coordinate data, and the error distribution function in the X axis direction is:

[0094] ΔX(x,y,z)=a1x+b1y+c1z+d1xy+e1xz+f1yz+g1x 2 +h1y 2 +i1z 2 ;

[0095] Wherein, ΔX(x,y,z) represents the X direction error value of the three coordinate system at any point (x,y,z) in space;

[0096] x, y and z respectively represent the coordinate values of the point on the X axis, Y axis and Z axis of the three coordinate system;

[0097] a1, b1, c1, d1, e1, f1, g1, h1 and i1 represent fitting coefficients.

[0098] In the X-axis direction error distribution function, the first order term (a1x, b1y, c1z) mainly reflects the linear error of the measuring machine (such as the proportional error of the X-axis direction caused by uniform wear of the guide rail); the cross term (d1xy, e1xz, f1yz) reflects the coupling error between axes (such as the xy plane cross error caused by the perpendicularity error of X-Y axis); the quadratic term (g1x 2 , h1y 2 , i1z 2 ) reflects the nonlinear error (such as the x 2 term error caused by the parabolic bending of the guide rail). The fitting coefficients (a1 to i1) are solved by the least square method, for example, the X-axis error values of 100 reference marks are fitted to minimize the sum of squares of fitting residuals, and the typical coefficient range is a1∈[-1e -6 ,1e -6 ] (unit 1 / mm), d1∈[-1e -9 ,1e -9 ] (unit 1 / (mm 2 )), and the specific value changes with the model and the degree of wear. The error distribution function of Y-axis and Z-axis is consistent with that of X-axis (such as ΔY(x,y,z)=a2x+b2y+c2z+d2xy+e2xz+f2yz+g2x 2 +h2y 2 +i2z 2 ), and three functions can completely describe the error values of any point (x, y, z) in space in three coordinate axes, for example, the X-axis error ΔX=0.002mm of a point (500, 500, 500) can be calculated by the function to calculate the error influence of the surrounding area, which provides a quantitative basis for subsequent compensation.

[0099] In some embodiments, the calibration method further comprises a multi-factor compensation model.

[0100] The multi-factor compensation model adopts a hierarchical structure, and the bottom layer is a physical mechanism model and the upper layer is a data-driven model.

[0101] The physical mechanism model is used to input temperature, humidity, and air pressure parameters to calculate a basic correction amount.

[0102] The data-driven model is used to take vibration frequency, power fluctuation, and bottom layer correction residual as input, and outputs a dynamic compensation amount through an LSTM network, and the final coordinate correction amount is obtained by weighted fusion of the results of the two layers.

[0103] The underlying physical mechanism model is based on known environmental influence rules, such as the thermal expansion of steel materials resulting in an X-axis error change of about 1.2e-5 mm / mm (i.e. 0.012 mm for a 1 m length) for every 1℃ change in temperature, the air refractive index change resulting in a laser measurement error of about 0.0005 mm / m for every 10% change in humidity, and the error of about 0.0003 mm / m caused by the air density change corresponding to every 1 kPa change in air pressure. The basic correction amount can be directly calculated through these physical formulas.

[0104] The upper-layer data-driven model adopts an LSTM network (containing 3 layers of hidden layers, each layer having 64 neurons), the input parameters include the vibration frequency (1-1000 Hz, sampling rate 1 kHz), power fluctuation (voltage change within ±5%), and the correction residual error of the underlying model (i.e. the difference between the actual error and the basic correction amount), and the model is trained through historical data (1000 or more calibration samples) to make the prediction error of the dynamic compensation amount ≤0.001 mm. In the weighted fusion of the results of the two layers, the weight of the physical mechanism model is 0.7 (high certainty), and the weight of the data-driven model is 0.3 (compensation of unknown interference), and the output frequency of the final coordinate correction amount is consistent with the sensor acquisition frequency (10 Hz), ensuring real-time response in the case of environmental changes, such as an additional correction of 0.005-0.01 mm of thermal expansion error in a high-temperature environment.

[0105] In some embodiments, in the data acquisition step, by setting strain sensors and laser interferometers inside the measurement head, deformation data of the probe under different measurement angles and touch forces are collected in real time to establish a probe deformation error model;

[0106] The model is linked with the reference coordinate data acquisition process, and when the measurement head trigger angle deviates from the preset threshold, the deformation compensation algorithm is automatically called to correct the measurement value.

[0107] The strain sensor (accuracy 1με) inside the measurement head is pasted at the root of the probe rod (diameter 3 mm, length 50 mm) to monitor the bending strain in real time; the laser interferometer (resolution 0.01 μm) measures the displacement change of the probe tip through a reflecting mirror, and the deformation value under different angles (0°-90°, interval 5°) and touch forces (0.1-1 N, interval 0.1 N) can be calculated by fusing the data of the two. The probe deformation error model adopts a quadratic function form: ΔL(θ, F) = k1θ 2 +k2F 2+k3θF (where θ is the measurement angle, F is the touch force, and k1, k2, and k3 are fitting coefficients), for example, when θ = 60° and F = 0.5 N, ΔL = 0.002 mm. When the model is linked with the reference coordinate acquisition, a preset angle threshold of ±5° (that is, when the deviation between the actual angle of the measurement head and the theoretical angle exceeds 5°), the compensation algorithm is automatically called, and the deformation influence is eliminated by correcting the measurement value (such as correcting the measured coordinate x to x-ΔL・cosθ).

[0108] In some embodiments, under each scene mode, three reference markers not involved in fitting are selected for verification measurement immediately after parameter adjustment is completed: if the deviation between the error calculation value of the verification point and the actual measurement value is ≤0.03 mm, it is determined that the adjustment is effective; if the deviation is >0.03 mm, the control module automatically triggers parameter backtracking, restores the current scene parameters to the optimal value before adjustment, and starts supplementary sampling, recalculates the adjustment parameters based on the expanded data set, and repeats the verification until the deviation requirement is met.

[0109] The verification and backtracking mechanism of the calibration parameters is established to ensure the effectiveness of the scene mode adjustment. The selection of the verification point needs to meet the principles of “not involved in fitting” and “uniformly distributed in space”, for example, three markers (located in the central core area, the edge transition area, and the boundary constraint area) are randomly selected from the dynamic reference unit, the deviation threshold between the error calculation value (obtained by the error distribution function) and the actual measurement value (the average of 5 repeated measurements) is set to 0.03 mm (1 / 3 of the maximum permissible error MPE based on the measuring machine). When the deviation is out of limit, the control module backtracks to the optimal value before adjustment (i.e., the parameters of the last verification passed) through the parameter snapshot stored in the EEPROM (saved every 10s), and starts supplementary sampling - increases the reference markers by 20% (e.g., from 100 to 120), expands the spatial coverage of the data set, and recalculates the adjustment parameters by the least squares method. In the repeated verification process, if the deviation requirement is not met for 3 times in a row, an alarm is triggered and the reference assembly state is prompted for inspection (such as whether it is loose or worn out). This closed-loop mechanism improves the long-term stability (within 24 hours) of the calibration result to within ±0.005 mm, and the reliability is improved by 40% compared with the traditional method without verification.

[0110] In some embodiments, based on the fact that the error characteristics of each point in the measurement space are different, the measurement space is divided into three subspaces: a central core area, an edge transition area, and a boundary constraint area.

[0111] The central core area is a spherical region with the geometric center of the measurement space as the center and 1 / 3 of the space diagonal as the radius.

[0112] The edge transition area is an annular region between the core area and the boundary.

[0113] The boundary constraint region is a thin layer region close to the inner wall of the measurement space;

[0114] An error transfer coefficient matrix T between the subspaces is set, wherein T[i][j] represents an influence weight of an error of an i-th subspace on a j-th subspace, 0≤T[i][j]≤1 and ΣT[i][j]=1;

[0115] The core region error transfer coefficient satisfies T[1][1]≥0.8; the edge transition region satisfies T[2][1]+T[2][3]≥0.6; and the boundary constraint region satisfies T[3][3]≥0.7 and T[3][2]≤0.2;

[0116] In constructing the error distribution function, different polynomial degrees are used for different subspaces: a 3-order polynomial is used for the core region, a 2-order polynomial is used for the edge transition region, and a 1-order polynomial is used for the boundary constraint region, and the coupling correction of the error of the subspaces is realized through the transfer coefficient matrix.

[0117] wherein the error modeling is optimized based on the space partitioning theory, and the error description accuracy of different regions is improved. The calculation method of the spherical region of the central core region is: the geometric center coordinates (x0, y0, z0) = (750 mm, 500 mm, 400 mm), the space diagonal length is ≈1972mm, and the 1 / 3 diagonal length is ≈657mm, so the core region range is (x-x0) 2 +(y-y0) 2 +(z-z0) 2 ≤657 2; the edge transition zone is a region from the core zone to a region 100 mm away from the boundary; the boundary constraint zone is a thin layer close to the inner wall (such as x≤100 mm or x≥1400 mm, etc.). The typical values of the error transfer coefficient matrix T are: T[1][1]=0.85, T[1][2]=0.1, T[1][3]=0.05 (the core zone mainly affects itself); T[2][1]=0.3, T[2][2]=0.1, T[2][3]=0.6 (the edge zone is affected by the core and the boundary); T[3][1]=0.05, T[3][2]=0.15, T[3][3]=0.8 (the boundary zone is mainly self-constrained). The basis for selecting the polynomial order is: the nonlinear error ratio of the core zone is >60% (3-order polynomial is needed), the linear and nonlinear error ratio of the edge zone is about 4:6 (2-order polynomial is needed), and the linear error ratio of the boundary zone is >80% (1-order polynomial is needed). The error of each subspace is coupled and corrected by the matrix T (such as edge zone error=0.3×core zone error+0.6×boundary zone error), so that the error fitting accuracy of the edge region is improved from ±0.01 mm to ±0.006 mm, and the boundary region is improved from ±0.015 mm to ±0.008 mm, solving the problem of "good center fitting and large edge deviation" of the traditional unified model.

[0118] It is worth noting that the present application solves the problems of "single reference, insufficient regional error description, and poor environmental adaptability" in traditional calibration through the collaborative mechanism of "global reference sampling → regional error modeling → scene dynamic adaptation → multi-factor compensation → closed-loop verification". The core innovation lies in: integrating static mechanical error and dynamic environmental disturbance into a unified correction framework, achieving accurate quantization and real-time compensation of error through spatial partitioning and scene priority, and finally stabilizing the calibration accuracy of any point in the measurement space within ±0.003 mm, and maintaining long-term reliability under extreme conditions such as high temperature and high speed.

[0119] Global reference sampling solves the problem of incomplete sampling by constructing a global reference network, and uses fixed and dynamic reference components to construct a three-dimensional sampling network:

[0120] The fixed reference unit covers 8 corner points (cubic reference block), 12 edge midpoints (cylindrical reference column), 6 face centers (conical reference cone), and the center (standard sphere) of the measurement space. Carbon steel material (low thermal expansion coefficient) and micro-nano line process are adopted to ensure the high accuracy of the static reference point (positioning error ≤0.001 mm).

[0121] The dynamic reference unit is equipped with symmetrical spherical reference pieces (magnetic attraction type connection, sliding range covering 95% of the area) along the X / Y / Z axis guide rail every 500 mm, integrated with micro temperature sensors (accuracy ±0.1℃), and can capture dynamic errors in real time along the motion trajectory.

[0122] Spatial partition error modeling is used to improve the accuracy of region-specific error description. The measurement space is divided into three subspaces according to error characteristics, and different modeling is used:

[0123] Central core area (geometric center as the center of the sphere, 1 / 3 space diagonal as the radius): nonlinear error ratio > 60%, 3-order polynomial fitting (focus on capturing mechanical deformation error).

[0124] Edge transition zone (annular area between core area and boundary): linear and nonlinear error ratio 4:6, 2-order polynomial (balance two kinds of error).

[0125] Boundary constraint area (close to the inner wall 100mm thin layer): linear error ratio > 80%, 1-order polynomial (highlight the linear drift of guide rail).

[0126] Through error transmission coefficient matrix T (such as core area self-influence ≥80%, boundary area affected by edge area ≤20%), realize the coupling correction of subspace error, make the fitting accuracy of edge area from ±0.01mm to ±0.006mm.

[0127] Dynamic adaptation to different scenarios, optimize error model for different working conditions.

[0128] Single scene trigger: when temperature ≥30℃ (high temperature mode), servo speed ≥80% rated value (high speed mode) or workpiece weight ≥70% rated load (heavy load mode), automatically adjust the weight of the corresponding influence item in the error distribution function (such as increasing the thermal expansion related coefficient when high temperature).

[0129] Multi-scene superposition: through controller analysis error ratio, priority correction of dominant error (such as high temperature and high speed scene, according to load priority, high temperature > high speed priority parameter adjustment).

[0130] Through strain sensor and laser interferometer, real-time acquisition of probe deformation data at different angles (0°-90°) and touch force (0.1-1N), when the angle deviates ±5, automatically call model correction measurement value, reduce 70%-90% probe error.

[0131] Multi-factor dynamic compensation: eliminate the influence of environmental and dynamic interference, integrate physical law and data-driven correction.

[0132] Physical mechanism layer: through the physical formula calculation of temperature (1.2e-5mm / mm thermal expansion per ℃), humidity (0.0005mm / m refractive index error per 10% change), air pressure (0.0003mm / m error per kPa change), basic correction amount.

[0133] Data-driven layer: LSTM network (3 layers of hidden layer) is adopted, vibration frequency (1-1000Hz), power fluctuation (±5%) and physical layer residual error are input, and dynamic compensation (prediction error≤0.001mm) is output.

[0134] Weighted fusion: physical layer weight 0.7 (high certainty) + data layer weight 0.3 (compensate unknown interference), output coordinate correction amount at 10Hz frequency, real-time update error distribution function (response delay≤50ms).

[0135] Closed-loop verification, after adjusting each scene, select 3 reference points (cover core area, edge area, boundary area) not participating in fitting to verify, if error calculation value and measured value deviation≤0.03mm, it is determined to be effective; if the deviation is out of limit, automatically backtrack to the optimal parameters before adjustment (stored through EEPROM snapshot), supplement 20% reference point sampling and recalculate parameters until the requirements are met.

[0136] The above only describes the preferred embodiments of the present application and is not intended to limit the present application, any modification, equivalent replacement, improvement, etc. made within the spirit and principle of the present application shall be included in the protection scope of the present application.

Claims

1. A calibration method for a three-coordinate system, characterized in that, The method includes the following steps: Within the measurement space of the coordinate measuring machine, multiple reference components are arranged according to a preset spatial distribution rule, and multiple marker points are set on each reference component; Obtain the measurement reference coordinate data and theoretical coordinate data of the marked points, and construct an error distribution function based on the relationship between the measurement reference coordinate data and the theoretical coordinate data; The method for obtaining the measurement reference coordinate data is as follows: control the measuring head to measure each marker point according to a preset trajectory, measure each marker point multiple times continuously, and take the average value as the reference coordinate data obtained from the actual measurement; The method for obtaining the theoretical coordinate data is as follows: a spatial coordinate system is established based on the parameters of the measuring machine and the arrangement of the reference components, and the theoretical coordinate data of each marker point is calculated. The measurement scenarios include at least: high temperature scenario mode, high speed scenario mode, and heavy load scenario mode; When the temperature at any point in the measurement space is greater than or equal to the set temperature, the high temperature scene mode is activated. When the servo motor feedback speed is greater than or equal to the set speed, the high-speed scene mode is activated. When the workpiece load sensor reports a mass ≥ the set weight, the heavy load scenario mode is activated. Multi-scenario matching analysis is performed on scenario parameters for different measurement scenarios, and scenario parameters are adjusted based on the analysis results, including: Obtain measurement input parameters, classify scenarios based on the input parameters, and assign them to corresponding measurement scenarios. Obtain scenario parameters with high error proportions under multiple measurement scenarios, and optimize the adjustment of scenario parameters corresponding to scenario parameters with high error proportions. Adjusting the scene parameters for the measurement scenario also includes: if there are two or more measurement scenarios at the same time, first adjust the load parameters under multiple measurement scenarios; The error distribution function is corrected based on the adjusted measurement scenario, including: The temperature, humidity, and air pressure data of the measured scene are collected in real time by sensors and input into the compensation model. The coordinate correction of the marked points is output, and the corrected error distribution function is obtained based on the coordinate correction. The compensation amount is calculated based on the corrected error distribution function and the measuring machine parameters, and the motion trajectory is adjusted based on the compensation amount to achieve dynamic calibration. The error value is obtained by comparing coordinate data, and the error distribution function in the X-axis direction is: ΔX(x,y,z)=a1x+b1y+c1z+d1xy+e1xz+f1yz+g1x 2 +h1y 2 +i1z 2 ; Where ΔX(x,y,z) represents the X-direction error value of the three-coordinate system at any point (x,y,z) in space; x, y, z represent the coordinates of the point on the X-axis, Y-axis, and Z-axis of the three-coordinate system, respectively; a1, b1, c1, d1, e1, f1, g1, h1, i1 represent the fitting coefficients; The scene parameters of the measurement scenario are adjusted, and the error distribution function is corrected based on the adjusted measurement scenario. If there are two or more measurement scenarios at the same time, multi-scenario matching analysis is performed on the scene parameters of different measurement scenarios, and the scene parameters are adjusted based on the analysis results. The compensation amount is calculated based on the corrected error distribution function and the parameters of the coordinate measuring machine, and the motion trajectory is adjusted based on the compensation amount to achieve dynamic calibration.

2. The three-coordinate system calibration method according to claim 1, characterized in that, The reference assembly includes a fixed reference unit and a dynamic reference unit; Fixed reference units are arranged according to a preset spatial distribution rule, while dynamic reference units are set along key nodes of the measuring machine's motion trajectory and can slide along the guide rail. The marking points of the fixed reference unit are processed using micro-nano-level scribing technology, while the marking points of the dynamic reference unit integrate a miniature temperature sensor.

3. The three-coordinate system calibration method according to claim 1, characterized in that, The measurement space is a three-dimensional rectangular space defined by the X-axis, Y-axis, and Z-axis. The measurement space is a cube or cuboid structure, in which the twelve edges are the intersection lines formed by the intersection of the six faces, and are divided into three groups of four edges in the direction parallel to the coordinate axes.

4. The three-coordinate system calibration method according to claim 2, characterized in that, The fixed reference units are distributed at eight corner points of the measurement space. A cube reference block is set at each corner point, a cylindrical reference column is arranged at the midpoint of each of the twelve edges, a conical reference cone is arranged at the center of each of the six faces, and a standard sphere is arranged at the center. The dynamic reference unit adopts a magnetic connection structure, with a set of guide rails arranged every 500mm along the three coordinate axes. Each set contains two symmetrically distributed spherical reference components. The sliding range of the dynamic reference unit covers more than 95% of the measurement space.

5. The three-coordinate system calibration method according to claim 1, characterized in that, The calibration method further includes: a multi-factor compensation model; The multi-factor compensation model adopts a hierarchical structure, with the bottom layer being a physical mechanism model and the upper layer being a data-driven model; The physical mechanism model is used to calculate the basic correction amount based on the input temperature, humidity, and air pressure parameters; The data-driven model takes vibration frequency, power fluctuation and bottom-level correction residual as input, outputs dynamic compensation amount through LSTM network, and the results of the two layers are weighted and fused to obtain the final coordinate correction amount.

6. The three-coordinate system calibration method according to claim 1, characterized in that, In the data acquisition step, by setting a strain sensor and a laser interferometer inside the measuring head, the deformation data of the probe under different measuring angles and tactile forces is collected in real time, and a probe deformation error model is established. The model is linked to the baseline coordinate data acquisition process. When the trigger angle of the measuring head deviates from the preset threshold, the deformation compensation algorithm is automatically invoked to correct the measured value.

7. The three-coordinate system calibration method according to claim 1, characterized in that, In each scenario mode, after the parameters are adjusted, three reference markers that were not involved in the fitting are immediately selected for verification measurement; If the deviation between the calculated error value and the actual measured value at the verification point is ≤0.03mm, the adjustment is deemed effective. If the deviation is greater than 0.03 mm, the control module will automatically trigger parameter backtracking, restore the current scene parameters to the optimal value before adjustment, and start supplementary sampling. Based on the extended dataset, the adjustment parameters will be recalculated and verified repeatedly until the deviation requirement is met.

8. The three-coordinate system calibration method according to claim 1, characterized in that, Based on the different error characteristics of each point in the measurement space, the measurement space is decomposed into three subspaces: the central core area, the edge transition area, and the boundary constraint area. The central core area is a spherical region with the geometric center of the measurement space as the center and 1 / 3 of the space diagonal as the radius; The edge transition zone is a ring-shaped area between the core area and the boundary. The boundary constraint region is a thin layer area close to the inner wall of the measurement space; Define the error propagation coefficient matrix T between subspaces, where T[i][j] represents the influence weight of the error of the i-th subspace on the j-th subspace, 0≤T[i][j]≤1 and ΣT[i][j]=1; The central core area satisfies T[1][1]≥0.8; the edge transition area satisfies T[2][1]+T[2][3]≥0.6; the boundary constraint area satisfies T[3][3]≥0.7 and T[3][2]≤0.2; When constructing the error distribution function, different polynomial orders are used for different subspaces: the core region uses a third-order polynomial, the edge transition region uses a second-order polynomial, and the boundary constraint region uses a first-order polynomial. The coupling correction of subspace errors is achieved through the transfer coefficient matrix.

Images