A method and apparatus for in-process measurement of a workpiece on a grinding machine

Abstract

The application discloses a workpiece in-situ measurement method and device for grinding machine tool, and relates to the technical field of precision machining and measurement. The method comprises collecting an image sequence of a transverse section of a rotary workpiece and an image sequence of multiple stations along an axial direction; a one-dimensional gray profile is established along a normal direction for each pixel-level edge point in each frame of image; a parameterized fitting is performed on the error function model by introducing a linear background term, and a sub-pixel edge position is solved; a circle fitting is performed on a point set of the same height of the transverse section to reconstruct a transverse section line; a distance field is constructed for an axial profile point set of adjacent stations to establish a bidirectional Chamfer registration target function, and rigid transformation parameters are solved to realize accurate splicing of the axial profile; according to the reconstructed transverse section line and the axial profile line, a workpiece topography error is evaluated, and a detection result is output; the method realizes high-precision in-situ measurement of the transverse section and the axial profile of a large-size rotary part under a grinding environment.

Description

Technical Field

[0001] This invention relates to the field of precision machining and measurement technology, specifically to a method and apparatus for in-situ measurement of workpieces for grinding machine tools. Background Technology

[0002] Grinding rotating parts (including cylinders, frustums, and their combinations of rotating surfaces) is widely used in energy equipment, aerospace, and precision manufacturing. Their geometric quality is typically evaluated primarily by the morphological errors of the transverse cross-sections (such as roundness and coaxiality) and the morphological errors of the axial cross-sections (such as straightness and profile). To reduce the error chain introduced by repeated clamping, off-machine transfer, and secondary positioning, and to improve the efficiency of the machining-inspection closed loop, online / on-machine profile measurement in machine tool environments has gradually become an important technological direction in related fields. Under actual grinding conditions, rotating surfaces often exhibit smooth profile characteristics with low texture and low feature density. Simultaneously, interference factors such as strong reflections, coolant atomization, dust, and micro-vibrations exist in the machine tool environment, making it difficult to guarantee the signal stability, anti-interference ability, and measurement repeatability of non-contact online measurements.

[0003] Existing measurement methods include contact and non-contact measurements. Contact measurements (such as dial indicators, trigger probes, and coordinate measuring machines) typically achieve high accuracy, but their measurement efficiency is low. On-machine measurement involves complex path planning and interference avoidance, and there is a risk of contact deformation on thin-walled parts or flexible clamping components, making it difficult to meet the requirements of online high-frequency feedback and closed-loop correction in grinding. Non-contact measurement methods such as laser triangulation and structured light, while offering speed advantages, are prone to speckle noise, unstable echoes, or multipath reflections on highly reflective surfaces after grinding and in dusty / foggy media environments. Interference from radiation causes contour point drift and reduced repeatability, making it difficult to stably support micron-level online measurements. Furthermore, in situations where machine space and field of view are limited, axial contours typically require multi-field / multi-position acquisition and stitching reconstruction. The stitching process is highly sensitive to target features and matching stability. Vision-based contour detection schemes often rely on edge extraction, which is susceptible to systematic biases caused by factors such as pixel quantization, low contrast, edge widening, and noise disturbances. These biases are amplified into deviations in the evaluation of roundness, contour accuracy, and other morphological errors. Especially for low-feature targets such as smooth generatrices, if axial multi-field stitching relies on local features such as texture, corner points, or manual markings, unstable matching, misalignment, and accumulated errors can easily occur, making continuous and reliable reconstruction of axial cross-sections difficult. Simultaneously, some on-machine transverse cross-section measurement schemes still require external turntables or additional angle reference devices to achieve rotational measurement or establish angular correspondences, complicating system integration and adjustment, and introducing new clamping or positioning error chains, which is detrimental to the integrated implementation and consistency assurance of on-machine measurements.

[0004] In summary, existing technologies cannot simultaneously achieve high-precision transverse section line measurement, stable splicing measurement of axial section lines, and highly reliable sub-pixel contour point extraction of smooth contours of large-size grinding rotating bodies under machine tool in-situ conditions, resulting in the inability to meet the requirements for morphology error assessment. Summary of the Invention

[0005] To address the shortcomings of existing technologies in simultaneously achieving high-precision transverse section line measurement, stable splicing measurement of axial section lines, and highly reliable sub-pixel contour point extraction of smooth contours of large-sized grinding rotating bodies under machine tool in-situ conditions, this invention proposes an in-situ workpiece measurement method and device for grinding machine tools. This method enables online measurement of the transverse section line of the rotating body; stable splicing measurement of the axial section line of the rotating body; and high-precision sub-pixel contour point extraction to meet the requirements of morphological error assessment, thereby solving the problems existing in the prior art.

[0006] A method for in-situ measurement of workpieces for use in a grinding machine tool includes the following steps: Under synchronous triggering conditions of the machine tool, image sequences of one revolution of the transverse cross section of the rotating workpiece and image sequences of multiple stations along the axial direction are acquired to establish an image dataset; For each frame of the image dataset, a one-dimensional grayscale profile is established along its normal direction for the pixel-level edge points. The one-dimensional grayscale profile is parametrically fitted using an error function ERF model that incorporates a linear background term to establish the profile model. The profile model is then solved to generate the coordinates of each sub-pixel contour point. Based on a preset coordinate reference, the coordinates of all sub-pixel contour points are transformed to a unified coordinate system, and the circumferential point sets of each transverse section and the generatrix point sets of each axial workstation are constructed respectively. Least square circle fitting is performed on the circumferential point sets of each transverse section, and the minimum regional roundness error is calculated based on the radial deviation of each contour point relative to the fitted circle to reconstruct the transverse section line. A distance field is constructed for the axial generatrix point sets of adjacent workstations, and a two-way Chamfer registration objective function is established. A multi-step hierarchical search strategy is used to solve the two-way Chamfer registration objective function to achieve the splicing of axial contours and the reconstruction of continuous generatrixes. Based on the reconstructed transverse section lines and axial contour lines, the shape error of the rotating workpiece is evaluated to output the inspection results.

[0007] Furthermore, based on a preset coordinate reference, the coordinates of all sub-pixel contour points are transformed to a unified coordinate system, and the circumferential point set of each transverse section and the generatrix point set of each axial workstation are constructed respectively, specifically including: For each frame of an image derived from a transverse cross-section image sequence, sub-pixel contour points are extracted at the selected cross-section location and transformed to a unified coordinate reference to obtain the sub-pixel edge points of the same height of the point set of that cross-section, thereby generating the circumferential point set of each transverse cross-section. For each frame of an image derived from an axial image sequence, all sub-pixel contour points contained therein constitute the initial contour point set for that workstation. Based on the established coordinate reference, the initial contour point set is transformed from the image coordinate system to the machine tool coordinate system, resulting in the contour point set of that workstation in the machine tool coordinate system. According to the axial coordinates of the machine tool when the image of that workstation was acquired, each contour point in the contour point set is assigned axial position information, resulting in a point set with axial coordinates. The point sets with axial coordinates are stored in the order of the workstations, resulting in segmented axial contour point sets. The adjacent point sets contain overlapping areas in physical space.

[0008] Furthermore, the method of employing a multi-step hierarchical search strategy to solve the bidirectional Chamfer registration objective function to achieve accurate splicing of the axial profile and reconstruction of continuous generatrices specifically includes the following steps: For the axial section line image sequence, a reference segment contour point set is obtained for each workstation. The set of segments to be spliced ; From the reference segment outline point set Perform a distance transformation to obtain the distance field Similarly, a distance transformation is performed on the point set Q of the spliced ​​segment to obtain the distance field. ; In rigid body position Construct a bidirectional Chamfer objective function ;in, For the segments to be spliced, As a reference contour point, Forward transformation, This is an inverse transformation; Using a multi-step hierarchical search, at the first step... Layer, translation step size Angle step size In the candidate set The above enumeration, take use smallest And using it as the center, narrow the search range and step size of the next layer, and in the final layer... Using the initial value, further minimize it using local least squares. Finally, the optimal splicing result is obtained; among them, These are the candidate value sets for the translation in the x and y directions, respectively. , For the minimum translation amount, , For the maximum translation, For the set of candidate values ​​for the rotation angle, , These are the minimum angle and the maximum angle, respectively.

[0009] Furthermore, the error function ERF model that introduces a linear background term is expressed as: ; in, The position of a one-dimensional sub-pixel edge along the normal direction. Characterizes the edge transition width. This represents the grayscale step amplitude. For linear background items, For sampling window, The background grayscale constant, It is the background grayscale linear shift coefficient.

[0010] Furthermore, by defining residuals for the established profile model, the objective function is constructed using the Huber robust cost function, and the weights are calculated using iterative reweighted least squares (IRLS) to construct a weighted least squares problem. The weighted Levenberg-Marquardt LM algorithm is used to iteratively calculate the parameter vector of the profile model, output the sub-pixel edge positions, and thus obtain the coordinates of the sub-pixel edge points.

[0011] Furthermore, the process of calculating the parameter vector of the anatomical surface model also includes outputting a fitting quality index after each iteration converges, and generating the confidence level of the corresponding sub-pixel contour points based on the fitting quality index.

[0012] Furthermore, the residual is the difference between the observation point and the sub-pixel ideal edge point.

[0013] The present invention also includes a workpiece in-situ measurement device for a grinding machine tool, comprising: The machine tool spindle is used to clamp and drive the rotating workpiece to rotate. Grinding wheel holder, on which a grinding wheel is mounted; A follow-up measuring bracket is fixed on the grinding wheel frame; The camera and parallel light source are coaxially arranged vertically on the servo measurement bracket to form a coaxial transmission imaging structure. A dust extraction hood is installed below the working area of ​​the grinding wheel and the rotating workpiece, and above the parallel light source, to absorb grinding dust. A control unit is connected to the camera and the machine tool CNC system; the control unit is used to execute the workpiece in-situ measurement method for a grinding machine tool.

[0014] Furthermore, the parallel light beam emitted from the parallel light source passes through the outer contour of the rotating workpiece from bottom to top, forming a high-contrast light-dark boundary edge, which is then received and imaged by the camera.

[0015] Furthermore, the dust collection hood is a shell structure with an opening at the top, which is used to ensure that the parallel light beam can pass through while collecting dust.

[0016] This invention provides a method for in-situ measurement of workpieces for grinding machine tools, which has the following beneficial effects: This invention, based on the traditional error function (ERF) edge transition model, explicitly introduces a linear background term, which effectively describes and compensates for grayscale drift caused by slow changes in illumination during grinding. This avoids systematic edge positioning deviations due to uneven illumination, significantly improving the accuracy and stability of edge extraction. Addressing grayscale sampling anomalies caused by interference factors such as dust, fogging, strong reflections, and micro-vibrations in the grinding environment, this invention employs the Huber robust cost function to construct the objective function. It then adaptively reduces the weight of outlier sampling points through iterative reweighted least squares (IRLS), effectively suppressing the interference of outliers on edge positioning and ensuring the reliability and repeatability of sub-pixel edge extraction. Simultaneously, it constructs a distance field from the contour point set, using point-to-point distances... Minimum distance describes geometric relationships, completely avoiding the dependence of traditional stitching methods on local features such as texture, corners, and manual markings. This fundamentally solves the problem of unstable matching of low-feature surfaces. A bidirectional Chamfer distance is used to construct the registration objective function, and a multi-step, layered coarse-to-fine search is used to solve for translation and small angle compensation. Finally, local least-squares fine-tuning can be performed to obtain a stable stitching result. Without relying on texture, corners, or manual markings, it helps to reduce misalignment and cumulative errors. This method enables high-precision transverse section line measurement, stable stitching measurement of axial section lines, and highly reliable sub-pixel contour point extraction of smooth contours of large-size grinding rotating bodies under machine tool in-situ conditions, thus meeting the requirements for shape error assessment. Attached Figure Description

[0017] Figure 1 This is a schematic diagram of an in-situ measurement system for large-size grinding rotating bodies in an embodiment of the present invention; Figure 2 This is a schematic diagram of the in-situ measurement method for large-size grinding rotating bodies in an embodiment of the present invention; Figure 3 This is a schematic diagram illustrating the principle of sub-pixel edge extraction in an embodiment of the present invention; Figure 4 This is a schematic diagram illustrating the image registration and stitching principle in an embodiment of the present invention. Figure 5 This is a schematic diagram illustrating the principle of reconstructing the transverse cross-section line in an embodiment of the present invention; Figure 1 In the middle: 1-Machine tool spindle / rotary axis; 2-Rotating workpiece; 3-Grinding wheel; 4-Tool module / grinding wheel holder; 4a-Follow-up measuring bracket; 5-Telecentric lens + camera; 6-Parallel light source; 7-Parallel beam; 8-Dust collection hood; 9-Industrial control computer. Detailed Implementation

[0018] The technical solutions of the present invention will be clearly and completely described below with reference to the accompanying drawings of the embodiments of the present invention. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments.

[0019] When performing non-contact contour detection on the transverse cross-section and axial generatrix of a large-sized grinding rotating body under machine tool in-situ conditions, the following problems exist: ① Smooth, low-texture contours make edge positioning susceptible to noise and gradual changes in illumination, and amplify pixel quantization errors; ② Axial multi-view / multi-station contour reconstruction is prone to unstable matching due to lack of texture / corner points, resulting in misalignment and cumulative errors; ③ Shape error evaluation needs to be consistent with the accuracy of the obtained contour point set, outlier handling, and fusion strategies to ensure micron-level repeatability.

[0020] Based on this, the present invention proposes a workpiece in-situ measurement method for grinding machine tools, such as... Figure 1 As shown, the method specifically includes the following steps: S1. Data Acquisition and Unification: Under synchronous triggering conditions of the machine tool, acquire image sequences of one circumference of the transverse section, as well as image sequences of multiple stations along the axial direction (with overlapping areas for adjacent stations); determine the rotation center and coordinate reference through calibration to achieve coordinate unification of contour points of different frames / different stations.

[0021] S2, Sub-pixel contour point extraction (ERF error function model + linear background + Huber-IRLS + weighted LM): The coordinates of each pixel-level edge point obtained from the coarse detection of the workpiece contour edge are defined as follows: Calculate the image gradient and define the unit normal. (Along the grayscale transition direction). Establish one-dimensional sampling coordinates and observations along the normal direction. Given a sampling half-width L>0 (unit: pixels) and a sampling point number N≥3, define equal spacing:

[0022] ; Normal sampling point coordinates: ; For non-integer pixel positions The observation points were obtained using bilinear interpolation. This forms a one-dimensional grayscale profile sample set. In some embodiments, the following can be used: Interpolation encryption and pre-smoothing are performed to reduce the impact of pixel quantization and noise perturbation.

[0023] To adapt to gradual changes in on-board illumination and anomalous disturbances, this invention explicitly introduces a linear background term based on the existing ERF transition model. Establish a cross-sectional model: ; in The position of a one-dimensional sub-pixel edge along the normal direction. Characterizes the edge transition width. This represents the grayscale step amplitude. The background grayscale constant, It is the background grayscale linear shift coefficient.

[0024] Let the residual This is the difference between the observed point and the ideal sub-pixel edge point. To suppress the influence of outlier sampling points caused by factors such as dust, fogging, and reflection, this invention uses Huber robust cost to construct the objective function:

[0025] Minimize Huber robust cost (threshold δ>0): ; Overall optimization problem: .

[0026] The weights are obtained through IRLS (Iterative Reweighting). Constructing a weighted least squares form , Corresponding IRLS weights (when : ; In the In the next iteration, let Jacobian defines it as: ; Weighted LM increment Solve using the following formula: ; And update ; in The LM damping factor can be adaptively adjusted according to the descent / backtracking rule to ensure convergence stability. After iterative convergence, the following is obtained: Output sub-pixel edge point coordinates:

[0027] ; RMS is available. These parameters are used as fitting quality indicators for subsequent fusion and outlier removal.

[0028] S3. Transverse Cross-Section Line Reconstruction and Error Assessment: For each frame of the transverse cross-section line image sequence, extract pixel contour points at the selected cross-section location and transform them to a unified coordinate reference to form a set of sub-pixel edge points at the same height around the cross-section. .

[0029] Define residual: , where 'a' refers to the radius.

[0030] Can be Least squares fitted circle : .

[0031] The fitted center c and radius a are used to characterize the geometric features of the cross section. c allows for slight drift relative to the calibration center of rotation to absorb the effects of rotation center errors and noise disturbances, thus improving the stability of the reconstructed transverse cross section line. Roundness can be defined by the minimum zone circle (MZC) as the minimum bandwidth of two concentric envelope circles: .

[0032] S4, Axial Profile Stitching (Distance Field + Two-Way Chamfer + Layered Search): Point set and rigid body transformation: Let the reference contour point be m=1, M. Reference segment outline point set Similarly, let the segments to be spliced ​​be... n=1, M. Set of segments to be spliced .

[0033] pose parameters Translation Rotation .

[0034] Forward transformation: .

[0035] Inverse transform: .

[0036] The distance field is defined as a set of points Construct the distance function: .

[0037] In engineering implementation, distance transformation is used to obtain the result on the grid. For non-grid points Bilinear interpolation is used to obtain the values.

[0038] Similarly, from structure Therefore, a bidirectional Chamfer objective function is constructed: ;

[0039] Target: .

[0040] Finally, a multi-step hierarchical search is employed at the [number]th step. Layer, translation step size Angle step size In the candidate set The above enumeration, take use smallest And using this as the center, narrow down the search range and step size for the next layer. In the final layer... Using the initial value, further fine-tuning is performed using local least squares to continue minimizing. Ultimately, the best stitching result was obtained; among them, These are the candidate value sets for the translation in the x and y directions, respectively. , For the minimum translation amount, , For the maximum translation, For the set of candidate values ​​for the rotation angle, , These are the minimum angle and the maximum angle, respectively.

[0041] Based on the above methods, this invention proposes an in-situ workpiece measurement system for grinding machine tools, which completes the detection of the contour shape error of large-sized rotating bodies under machine conditions. Specifically, it includes: machine tool spindle / rotation axis 1, rotating workpiece 2, grinding wheel 3, tool module / grinding wheel holder 4, follow-up measurement bracket 4a, telecentric lens + camera 5, parallel light source / collimating backlight 6, parallel beam 7, dust suction hood 8, and control and calculation unit 9.

[0042] The rotating workpiece 2 is clamped on the machine tool spindle / rotation axis 1 and rotates coaxially with the spindle. The tool module / grinding wheel holder 4 mounts the grinding wheel 3 and provides the motion support required for grinding wheel machining. The follow-up measuring bracket 4a is fixed on the column or horizontal extension arm of the tool module / grinding wheel holder 4, so that the telecentric lens + camera 5 and the parallel light source / collimating backlight 6 are arranged coaxially and vertically to form a coaxial transmission imaging structure. The telecentric lens + camera 5 is located above with the lens facing down, and the parallel light source / collimating backlight 6 is located below with the light emitting upward. The parallel beam 7 passes through the outer contour of the workpiece 2 from bottom to top to form a high-contrast light and dark boundary edge and is emitted by the telecentric lens + camera 5. The image is received; the dust hood 8 is set below the working area formed by the grinding wheel 3 and the workpiece 2, and above the parallel light source / collimating backlight 6. The dust hood 8 is a shell structure with an opening at the top to ensure that the parallel light beam 7 can pass through the opening without blocking the light path. At the same time, the dust hood 8 is connected to an external dust collection device through the dust suction pipe to absorb grinding dust. The dust hood 8 can be represented by a dashed line in the structural diagram; the control and calculation unit 9 is connected to the telecentric lens + camera 5 through a data link, and is connected to the machine tool CNC system through synchronous trigger / interface to realize synchronous image acquisition, contour reconstruction, stitching registration, shape evaluation and result output.

[0043] The working process of this invention is as follows Figure 1As shown: When the machine tool performs grinding operations, the machine tool spindle / rotary axis 1 drives the rotating workpiece 2 to rotate around the spindle axis, and the grinding wheel 3 works with the tool module / grinding wheel head 4 and rotates at high speed; the tool module / grinding wheel head 4 performs horizontal feed according to process requirements (feed direction is perpendicular to the workpiece rotation axis), and the follow-up measuring bracket 4a drives the telecentric lens + camera 5 and the parallel light source / collimating backlight 6 to move synchronously, so that the coaxial transmission light path is always aligned with the measured contour area of ​​the workpiece 2, and the parallel beam 7 passes through the outer contour of the workpiece 2 to form a stable light and dark boundary edge; the dust generated by grinding is absorbed by the dust collection hood 8 below the working area, and the opening structure on the dust collection hood 8 ensures that the parallel beam 7 passes through and collects dust at the same time; the control and calculation unit 9 triggers image acquisition in conjunction with the CNC system through synchronous triggering / interface, ensuring the consistency of angle sampling or displacement sampling, and performs sub-pixel contour point extraction, transverse cross-section line reconstruction, axial cross-section line splicing and morphological error evaluation on the acquired image, outputting error curves, key indicators and qualified judgment results, and, if necessary, further outputting machining compensation amount for closed-loop correction.

[0044] The measurement process of this invention includes: First, camera and system calibration are performed. Using calibration components, the rotation center position is obtained by fitting the trajectory changes of the contour edge in the image. Simultaneously, a correspondence is established between the camera measurement coordinates and the machine tool motion coordinates, providing a common benchmark for the subsequent unified evaluation of transverse and axial cross-sectional lines. Image acquisition is triggered by the synchronous signal of the spindle CNC system, ensuring consistent acquisition angle intervals or axial step distances, improving sampling consistency and repeatability. In the subsequent image acquisition stage, under the condition that the telecentric lens and parallel light source remain stable relative to the workpiece measurement position, the spindle is driven to rotate at fixed angle intervals and synchronously triggers acquisition, obtaining a sequence of transverse cross-sectional line images covering one revolution. Multi-station scanning acquisition along the axial direction is achieved through machine tool axis feed or relative motion of the measurement unit. Overlapping measurement areas are set between adjacent stations to provide registration constraints, obtaining a sequence of axial cross-sectional line images. For pixel-level edge points, the unit normal is first determined by the gradient direction. and take the window in the legal direction. Grayscale profile obtained by sampling For non-integer pixel positions, bilinear interpolation is used to obtain... If necessary, the cross-section is interpolated, refined, and pre-smoothed to reduce the effects of distortion and noise. Subsequently, a one-dimensional ERF imaging transition model is used for parameter fitting; to adapt to gradual changes in illumination and anomalous disturbances under in-flight conditions, a linear background term is explicitly added to the model. The objective function is constructed using Huber robust loss, weights are formed using IRLS, and parameters are iteratively updated within a weighted LM framework, with the damping factor decreasing. The backtracking rule is adaptively adjusted. Sub-pixel edge position parameters are obtained after convergence. Output sub-pixel edge points Simultaneously output residual RMS, Confidence scores are generated using fitting quality metrics for subsequent fusion and outlier suppression. Shape error detection and axial stitching are then performed. For each frame of the transverse cross-sectional image sequence, sub-pixel contour points are extracted at the selected cross-sectional location and transformed to a unified coordinate system, forming a point set of the same height for that cross-section. Subpixel edge points around the week ,right The center *c* and radius *r* of the circle obtained by least-squares fitting are used to characterize the geometric features of the cross-section. *c* allows for slight drift relative to the calibration center of rotation to absorb the effects of rotation center errors and noise disturbances. Simultaneously, the RMS threshold is used to eliminate outliers, improving the stability of the reconstructed transverse cross-section. Roundness can be defined by the minimum zone circle (MZC) as the minimum bandwidth of two concentric envelope circles. Calculation. For the axial section line image sequence, a set of reference segment outline points is obtained for each workstation. The set of segments to be spliced The range field is obtained by performing a range transformation on a reference segment P. Similarly, a distance transformation is performed on the spliced ​​segment Q to obtain the distance field. In rigid body position Construct a bidirectional Chamfer objective function Then, a multi-step hierarchical search is used to find the result in the first step. Layer, translation step size Angle step size In the candidate set The above enumeration, take use smallest And using this as the center, narrow down the search range and step size for the next layer. In the final layer... Using the initial value, further fine-tuning is performed using local least squares to continue minimizing. The final result is a precise stitching, and the final test result report is output, which can be used for error compensation later.

[0045] Compared with the prior art, the present invention has the following technical advantages: (1) Robust subpixel contour point extraction for in-machine interference: Based on the ERF edge imaging transition model, this invention introduces a linear background term (b0,b1) to reduce the systematic bias caused by gradual changes in illumination. The subpixel edge position t0 is solved by Huber robust cost + IRLS reweighting + weighted LM iteration. At the same time, the residual RMS, |A| / σ and other fitting quality indicators are output to generate point-level confidence, which are used for subsequent fusion and outlier suppression. This makes it more suitable for low texture smooth contours, strong reflection / fogging / dust / micro-vibration noise and outlier interference in the grinding in-machine environment.

[0046] (2) Axial stitching registration that does not rely on texture or corner points: In view of the problem that axial multi-position / multi-viewpoint matching is prone to instability under low feature contours, this invention constructs the contour point set into a distance field (distance transformation), constructs the registration objective function with bidirectional Chamfer statistics, and uses multi-step hierarchical coarse-to-fine search to solve translation and small angle compensation. Finally, local least squares fine-tuning can be performed to obtain a stable stitching result. Under the condition that it does not rely on texture, corner points or manual marking, it helps to reduce misalignment and cumulative error.

[0047] (3) Online detection and index output of transverse cross-section lines for on-machine operation: The image sequence of the cross-section is acquired by synchronous triggering of the CNC axis and stepping by angle. A unified coordinate reference is established with the calibration rotation center. The sub-pixel contour points extracted from each frame are stitched into a set of points of the same height, realizing online detection of transverse cross-section lines without additional angle marking. The set of points of the whole circle is fitted with the center-radius joint least squares. The center is allowed to drift slightly to absorb the rotation center error and noise disturbance. The roundness is defined by the minimum bandwidth of MZC. In the overlapping area, the roundness is evaluated after fusion according to confidence / distance weight and outlier suppression, thereby improving the robustness and repeatability of roundness and related morphological indexes.

[0048] The above description is only a preferred embodiment of the present invention, but the scope of protection of the present invention is not limited thereto. Any equivalent substitutions or modifications made by those skilled in the art within the scope of the technology disclosed in the present invention, based on the technical solution and inventive concept of the present invention, should be covered within the scope of protection of the present invention.

Claims

1. A method for in-situ measurement of workpieces for grinding machine tools, characterized in that, Includes the following steps: Under synchronous triggering conditions of the machine tool, image sequences of one revolution of the transverse cross section of the rotating workpiece and image sequences of multiple stations along the axial direction are acquired to establish an image dataset; For each frame of the image dataset, a one-dimensional grayscale profile is constructed along its normal direction for the pixel-level edge points. An error function ERF model with a linear background term is used to parametrically fit the one-dimensional grayscale profile to establish the profile model; the profile model is then solved to generate the coordinates of each sub-pixel contour point. Based on a preset coordinate reference, the coordinates of all sub-pixel contour points are transformed to a unified coordinate system, and the circumferential point sets of each transverse section and the generatrix point sets of each axial workstation are constructed respectively. Least square circle fitting is performed on the circumferential point sets of each transverse section, and the minimum regional roundness error is calculated based on the radial deviation of each contour point relative to the fitted circle to reconstruct the transverse section line. A distance field is constructed for the axial generatrix point sets of adjacent workstations, and a two-way Chamfer registration objective function is established. A multi-step hierarchical search strategy is used to solve the two-way Chamfer registration objective function to achieve the splicing of axial contours and the reconstruction of continuous generatrixes. Based on the reconstructed transverse section lines and axial contour lines, the shape error of the rotating workpiece is evaluated to output the inspection results.

2. The workpiece in-situ measurement method for a grinding machine tool according to claim 1, characterized in that, Based on a preset coordinate reference, the coordinates of all sub-pixel contour points are transformed to a unified coordinate system, and the circumferential point set of each transverse section and the generatrix point set of each axial workstation are constructed respectively, specifically including: For each frame of an image derived from a transverse cross-section image sequence, sub-pixel contour points are extracted at the selected cross-section location and transformed to a unified coordinate reference to obtain the sub-pixel edge points of the same height of the point set of that cross-section, thereby generating the circumferential point set of each transverse cross-section. For each frame of an image derived from an axial image sequence, all sub-pixel contour points contained therein constitute the initial contour point set for that workstation. Based on the established coordinate reference, the initial contour point set is transformed from the image coordinate system to the machine tool coordinate system, resulting in the contour point set of that workstation in the machine tool coordinate system. According to the axial coordinates of the machine tool when the image of that workstation was acquired, each contour point in the contour point set is assigned axial position information, resulting in a point set with axial coordinates. The point sets with axial coordinates are stored in the order of the workstations, resulting in segmented axial contour point sets. The adjacent point sets contain overlapping areas in physical space.

3. The workpiece in-situ measurement method for a grinding machine tool according to claim 1, characterized in that, The method employs a multi-step hierarchical search strategy to solve the bidirectional Chamfer registration objective function, thereby achieving accurate stitching of the axial profile and reconstruction of continuous generatrices. Specifically, this includes the following steps: For the axial section line image sequence, a reference segment contour point set is obtained for each workstation. The set of segments to be spliced ; From the reference segment outline point set Perform a distance transformation to obtain the distance field Similarly, a distance transformation is performed on the point set Q of the spliced ​​segment to obtain the distance field. ; In rigid body position Construct a bidirectional Chamfer objective function ;in, For the segments to be spliced, As a reference contour point, Forward transformation, This is an inverse transformation; Using a multi-step hierarchical search, at the first step... Layer, translation step size Angle step size In the candidate set The above enumeration, take use smallest And using it as the center, narrow the search range and step size of the next layer, and in the final layer... Using the initial value, further minimize it using local least squares. Finally, the optimal splicing result is obtained; among them, These are the candidate value sets for the translation in the x and y directions, respectively. , For the minimum translation amount, , For the maximum translation, For the set of candidate values ​​for the rotation angle, , These are the minimum angle and the maximum angle, respectively.

4. A method for in-situ measurement of workpieces for a grinding machine tool according to claim 1; characterized in that, The error function ERF model that introduces a linear background term is expressed as: ； in, The position of a one-dimensional sub-pixel edge along the normal direction. Characterizes the edge transition width. This represents the grayscale step amplitude. For linear background items, For sampling window, The background grayscale constant, It is the background grayscale linear shift coefficient.

5. A method for in-situ measurement of workpieces for a grinding machine tool according to claim 4, characterized in that, By defining residuals for the established profile model, the objective function is constructed using the Huber robust cost function, and the weights are calculated using iterative reweighted least squares (IRLS) to construct a weighted least squares problem. The weighted Levenberg-Marquardt LM algorithm is used to iteratively calculate the parameter vector of the profile model, output the sub-pixel edge positions, and then obtain the coordinates of the sub-pixel edge points.

6. A method for in-situ measurement of workpieces for a grinding machine tool according to claim 5, characterized in that, The process of obtaining the parameter vector of the anatomical surface model also includes outputting a fitting quality index after each iteration converges, and generating the confidence level of the corresponding sub-pixel contour points based on the fitting quality index.

7. A method for in-situ measurement of workpieces for a grinding machine tool according to claim 5, characterized in that, The residual is the difference between the observation point and the sub-pixel ideal edge point.

8. A workpiece in-situ measuring device for a grinding machine tool, used to implement the method of claim 1, characterized in that, include: The machine tool spindle (1) is used to clamp and drive the rotating workpiece (2) to rotate; Grinding wheel holder (4), on which grinding wheel (3) is installed; The follow-up measuring bracket (4a) is fixed on the grinding wheel frame (4); The camera (5) and the parallel light source (6) are arranged coaxially on the follow-up measurement bracket (4a) to form a coaxial transmission imaging structure; A dust hood (8) is set below the working area of ​​the grinding wheel (3) and the rotating workpiece (2) and above the parallel light source (6) to absorb grinding dust. The control unit (9) is connected to the camera (5) and the machine tool CNC system; the control unit is used to perform the method described in claim 1.

9. A workpiece in-situ measuring device for a grinding machine tool according to claim 8, characterized in that, The parallel light beam (7) emitted from the parallel light source (6) passes through the outer contour of the rotating workpiece (2) from bottom to top to form a high-contrast light and dark boundary edge, and is received and imaged by the camera (5).

10. A workpiece in-situ measuring device for a grinding machine tool according to claim 8, characterized in that, The dust collection hood (8) is a shell structure with an opening at the top, which is used to ensure that the parallel light beam (7) passes through while collecting dust.

Images