Method and System for Detecting Warpage of Glass Substrates
By using servo-driven reciprocating scanning and least squares fitting, the problem of online detection of warpage and flatness of the entire surface of glass substrates was solved, achieving low-cost and high-efficiency full-surface detection, applicable to a variety of materials.
Patent Information
- Application Number
- CN202611609892.4
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2026-10-19
- Publication Date
- 2026-11-17
AI Technical Summary
Existing technologies make it difficult to achieve online detection of the warpage and flatness of the entire surface of glass substrates on dynamic conveyor lines. In particular, the fixed single-line method is insufficient for local detection, and the area array scanning method is costly and slow, making it difficult to meet the requirements for online deployment.
A servo-driven reciprocating scanning method is adopted to obtain a set of scattered points through the synchronous movement of the upper and lower probes. Combined with regular grid and least squares fitting, the warpage and flatness of the entire surface of the glass substrate can be detected.
It enables online full-surface inspection of glass substrates, with low hardware cost, fast inspection speed, reliable results, applicability to various materials, and meets real-time requirements.
Smart Images

Figure CN123111124A_ABST
Abstract
Description
Technical Field
[0001] This invention relates to the field of glass testing, and in particular to a method and system for detecting the warpage of a glass substrate. Background Technology
[0002] The warpage and surface flatness of glass substrates are core indicators for measuring their processing quality. Excessive warpage can lead to deviations in subsequent processes such as coating, while poor flatness can affect optical performance and bonding effect. Therefore, it is crucial to achieve online full inspection of warpage and flatness during the glass substrate production process. Currently, the commonly used glass warpage detection methods in the industry mainly include the following: (1) Fixed single-line / dual-probe method: The laser displacement sensor is fixedly installed on one side of the conveyor line, and a single line segment is sampled in the edge area of the glass substrate. After eliminating tilt and thickness errors by differential measurement of the upper and lower dual probes, the warpage is calculated by the range method (i.e., the maximum height value minus the minimum height value in the sampling area). This method is simple in structure and low in cost, but it can only reflect the morphological characteristics of local line segments of the glass substrate and cannot evaluate the global warpage and surface flatness of the entire glass. (2) Coordinate measuring machine (CMM) sampling method: The CMM is used to perform discrete point contact measurement on the surface of the glass substrate, with an accuracy of up to the micrometer level. However, this method belongs to the offline sampling inspection mode, and the inspection time per piece is long (about 30 minutes / piece), which cannot meet the needs of online full inspection of the production line. (3) Area array structured light / laser profilometer full-area scanning method: Using an area array camera in conjunction with structured light projection or line laser profilometer to scan the glass substrate in full area can obtain dense three-dimensional point cloud with high accuracy. However, the equipment is expensive, sensitive to environmental vibration and lighting conditions, and the scanning + data processing time is long, making it difficult to achieve real-time online inspection on high-speed conveyor lines. (4) Manual visual / feeler gauge inspection method: Relies on operators to observe visually or use feeler gauges to measure the gap between the glass and the platform. It is highly subjective, has low accuracy, cannot be quantified, and is inefficient, and has been gradually phased out.
[0003] In summary, existing technologies generally suffer from a contradiction between "local detection and global quality assessment": while fixed single-line methods can achieve online detection, they can only cover one line segment, and the amount of information is insufficient to support the overall board quality judgment; area array scanning methods, although providing comprehensive information, are costly, slow, and difficult to deploy online. How to achieve online detection of the warpage and flatness of the entire surface of a glass substrate on a dynamic conveyor line under limited hardware costs is a problem that urgently needs to be solved in this field. Summary of the Invention
[0004] This invention addresses the technical problem of online detection of warpage of glass substrates by providing a method and system for detecting the warpage of glass substrates.
[0005] In a first aspect, a method for detecting the warpage of a glass substrate is provided, for detecting a glass substrate passing through a detection station at a uniform speed, comprising: Obtain the X-direction position of the glass substrate at time t, x(t) = v·t; where v is the velocity of the glass substrate in the X direction. The upper and lower probes, respectively set on the upper and lower sides of the testing station, will reciprocate in a consistent manner according to the specified amplitude and frequency. Obtain a first set of scattered points collected by the upper and lower probes after the reciprocating motion; each sampling point in the first set of scattered points is: P_i = (x_i, y_i, ΔH_i); where x_i is the coordinate of the X direction, y_i is the coordinate of the direction that forms a 90-degree angle with the X direction, ΔH_i is the effective height deviation value obtained by differential calculation using the synchronous measurement values of the upper and lower probes; i is the sequence number of the sampling point; After the glass substrate is moved for a specified time, a regular grid is constructed on the first plane of the inspection station. By combining the first scatter set and the regular grid, the height values of the grid nodes in the regular grid outside the first scatter set are estimated, and the complete two-dimensional surface height matrix of the first plane of the glass substrate is obtained. The optimal reference plane is obtained by fitting the two-dimensional surface height matrix using the least squares method. The warpage of the first plane is obtained based on the optimal reference plane and the two-dimensional surface height matrix.
[0006] Preferred options also include: The flatness of the first plane is obtained by combining the complete two-dimensional surface height matrix of the first plane, the optimal reference plane, and the total number of grid nodes.
[0007] Preferably, the reciprocating motion includes: the upper probe and the lower probe forming a periodic trajectory on the first plane; the periodic trajectory includes: triangular wave, sawtooth wave, sine wave and cosine wave.
[0008] Preferably, the step of combining the first scatter set and the regular grid to estimate the height values of the grid nodes in the regular grid outside the first scatter set, and obtaining the complete two-dimensional surface height matrix of the first plane of the glass substrate, includes:
[0009]
[0010]
[0011] Where k=1,2,...,N are the sampling points in the N known first scattered point sets closest to the target grid node (N is the preset number of nearest neighbors), d_k is the Euclidean distance between the sampling point in the kth first scattered point set and the target grid node, p is the distance decay exponent (recommended value p=2), Z_k is the ΔH_i value of the sampling point k in the kth first scattered point set; Z_g is the height value of the target grid node; the complete two-dimensional surface height matrix is Z_g(x, y), where x and y are the numbers of all grid nodes in the first plane.
[0012] Preferably, the sampling points in the set of N known first scattered points closest to the target grid node are obtained through KD-Tree query.
[0013] Preferably, obtaining the warpage of the first plane based on the optimal reference plane and the two-dimensional surface height matrix includes: Calculate the offset distance of each grid node from the optimal reference plane as follows: d_j = Z_g(j) z_fit(x_j, y_j); Using the aforementioned deviation distance, the warpage of the first plane is calculated as follows: W = max(d_j) min(d_j); Where z_fit(x_j, y_j) is the optimal reference plane, Z_g(j) is the height value of the target mesh node, W is the warpage of the first plane, and d_j is the offset distance.
[0014] Preferably, the step of combining the complete two-dimensional surface height matrix of the first plane, the optimal reference plane, and the total number of mesh nodes to obtain the flatness of the first plane includes:
[0015]
[0016] Where RMSE is the flatness; M is the total number of grid nodes in the regular grid; and d_j is the deviation distance.
[0017] Secondly, a glass substrate warpage detection system is also provided for detecting glass substrates passing through a detection station at a uniform speed, comprising: First acquisition unit: used to acquire the X-direction position of the glass substrate at time t, x(t) = v·t; where v is the velocity of the glass substrate in the X direction; Setting unit: Used to make the upper and lower probes, which are respectively set on the upper and lower sides of the detection station, reciprocate in a consistent manner according to a specified amplitude and frequency; The second acquisition unit is used to acquire a first set of scattered points collected by the upper and lower probes after the reciprocating motion. Each sampling point in the first set of scattered points is: P_i = (x_i, y_i, ΔH_i); where x_i is the coordinate of the X direction, y_i is the coordinate of the direction that forms a 90-degree angle with the X direction, ΔH_i is the effective height deviation value obtained by differential calculation using the synchronous measurement values of the upper and lower probes, and i is the sequence number of the sampling point. Planning unit: used to construct a regular grid on the first plane of the inspection station after the glass substrate has been moved for a specified time; First calculation unit: used to estimate the height value of the grid node in the regular grid outside the first scatter set by combining the first scatter set and the regular grid, and to obtain the complete two-dimensional surface height matrix of the first plane of the glass substrate; Fitting unit: used to perform least squares fitting on the two-dimensional surface height matrix to obtain the optimal reference plane; The second calculation unit is used to obtain the warpage of the first plane based on the optimal reference plane and the two-dimensional surface height matrix.
[0018] Preferred options also include: The third calculation unit is used to combine the complete two-dimensional surface height matrix of the first plane, the optimal reference plane, and the total number of grid nodes to obtain the flatness of the first plane.
[0019] Preferably, the reciprocating motion includes: the upper probe and the lower probe forming a periodic trajectory on the first plane; the periodic trajectory includes: triangular wave, sawtooth wave, sine wave and cosine wave.
[0020] Beneficial effects: (1) Online full-surface inspection capability. This invention expands the inspection area of the traditional fixed dual-probe method from a single line segment to the entire surface of the glass substrate by introducing servo-driven reciprocating scanning. The mathematical controllability of the scanning trajectory allows the sampling points to be distributed regularly in space, avoiding blind spots from random sampling and ensuring effective coverage of the entire surface morphology. At the same time, the inspection process is completed entirely during the normal transport of the glass substrate, without the need for line stoppage or offline operation, truly achieving online full inspection, with a single-piece inspection cycle controlled within 10 seconds.
[0021] (2) Lower system hardware cost. This invention only requires adding a servo linear scanning module (servo motor + linear guide) to the existing fixed dual-probe method. The incremental hardware cost is far lower than that of area array structured light or high-density line laser profilometer solutions. The upper and lower probes continue to use industrial-grade laser displacement sensors such as Keyence, without the need to upgrade to more expensive area array sensors. The overall system modification cycle is short and it is easy to install and deploy on existing production lines.
[0022] (3) High computational efficiency, meeting real-time requirements.
[0023] (4) The measurement results are reliable, and tilt and thickness errors are systematically eliminated. This invention inherits the core advantage of the dual-probe differential method—by simultaneously measuring and differentiating with the upper and lower probes, the systematic errors caused by the overall tilt and thickness tolerance of the glass substrate are naturally eliminated, without the need for additional tilt calibration and thickness compensation steps. The least squares plane fitting after interpolation reconstruction further removes the residual low-order trend terms, making the baselines of warpage (range method) and flatness (RMSE method) consistent and the results reproducible.
[0024] (5) Wide applicability and strong scalability. The core of the method of this invention is a universal "line scan dimensionality enhancement" framework that does not depend on specific glass materials or sizes. By simply adjusting parameters such as scanning amplitude A, frequency f, and grid spacing h, it can be adapted to online detection of surface morphology of sheet materials such as glass substrates, metal plates, ceramic substrates, and PCB circuit boards of different specifications. Attached Figure Description
[0025] Figure 1 Figure 1 shows a first embodiment of a method for detecting the warpage of a glass substrate. Figure 2 Figure 1 shows a second embodiment of a method for detecting the warpage of a glass substrate. Figure 3 This is a front view and a top view schematic diagram of a glass substrate warpage detection system; Figure 4 A top view of the glass substrate (including sampling points and grid nodes). Figure 5 This is the interpolated surface mesh diagram; Figure 6 Figure 1 shows a first embodiment of a glass substrate warpage detection system; Figure 7 This is a second embodiment of a glass substrate warpage detection system.
[0026] Explanation of reference numerals in the attached figures: 1. A glass substrate warpage detection system; 11. First acquisition unit; 12. Setting unit; 13. Second acquisition unit; 14. Planning unit; 15. First calculation unit; 16. Fitting unit; 17. Second calculation unit; 18. Third calculation unit. Detailed Implementation
[0027] Terminology Explanation: Warpage: refers to the degree of overall bending or twisting deformation of the entire glass substrate in three-dimensional space. It reflects the global shape deviation of the substrate and is usually quantified by the range between the maximum and minimum deviation.
[0028] Smoothness: This usually refers to the degree of microscopic unevenness of the measured surface within a specified area, focusing on evaluating the local undulations of the surface. Unlike warpage, which focuses on global curvature, smoothness pays more attention to the short-wavelength ripples and roughness characteristics of the surface.
[0029] This invention addresses the technical problem of online detection of warpage of glass substrates by providing a method and system for detecting the warpage of glass substrates.
[0030] Firstly, such as Figure 1 As shown, a method for detecting the warpage of a glass substrate is provided, used to detect a glass substrate passing through a detection station at a uniform speed, comprising: S1: Obtain the X-direction position of the glass substrate at time t: x(t) = v·t; where v is the velocity of the glass substrate in the X-direction. Roller conveyor line: Used to carry and transport the glass substrate, allowing it to pass through the detection station at a uniform speed along the X-direction. A position encoder is installed on the roller conveyor line to acquire the X-direction displacement of the glass substrate in real time and record this displacement. This displacement x(t) is used as the basis for subsequent calculations. The position encoder is installed on the drive roller of the roller conveyor line (or at an encoder mounting position coaxially connected to the roller), and the forward distance of the glass substrate along the X-direction is calculated by recording the angular displacement of the roller rotation.
[0031] S2: The upper and lower probes, respectively positioned on the upper and lower sides of the inspection station, reciprocate in unison according to a specified amplitude and frequency. Servo-driven linear scanning module: Installed above the roller conveyor line across the width direction (Y direction) of the glass substrate, including a servo motor, linear guide rail, and slide table, it drives the probe assembly to reciprocate along the Y direction. Dual-probe laser displacement sensor assembly: Installed on the servo slide table, including an upper probe above the glass substrate and a lower probe below the glass substrate. The upper and lower probes are coaxially opposed, synchronously acquiring the distance values from the upper and lower surfaces to the probes. Both probes are fixedly mounted on the same servo slide table, reciprocating synchronously along the Y direction (perpendicular to the glass width direction of the conveying direction) with the slide table. Because the upper and lower probes are coaxially opposed, their positions in the Y direction remain consistent, preventing relative displacement. The advantages of this design are: firstly, it ensures strict spatial alignment during differential dual-probe operation; secondly, only one servo drive system (one motor + one guide rail) is needed to drive both probes simultaneously.
[0032] S3: Obtain the first scatter set of data collected by the upper and lower probes after the reciprocating motion; each sampling point in the first scatter set is: P_i = (x_i, y_i, ΔH_i); where x_i is the coordinate in the X direction, y_i is the coordinate in the Y direction at a 90-degree angle to the X direction, ΔH_i is the effective height deviation value obtained by differential calculation using the synchronous measurement values of the upper and lower probes; i is the sequence number of the sampling point; wherein, for each sampling point, differential calculation is performed using the synchronous measurement values of the upper and lower probes to eliminate the systematic errors introduced by the overall tilt of the glass substrate and the thickness tolerance, and the effective height deviation value is obtained: ΔH_i = H_top(t_i). H_bottom(t_i) d_ref; where d_ref is the nominal thickness of the glass substrate. After differential processing, each sampling point constitutes a complete three-dimensional data point: P_i = (x_i, y_i, ΔH_i). During the movement of the glass substrate along the X-direction by the roller conveyor line, the upper and lower probes reciprocate along a position mounted above the roller conveyor line spanning the width direction (Y-direction) of the glass substrate. Simultaneously, the upper and lower probes acquire the distance between the probe and the plate (other than the plate whose warp is being tested) in real time. Combined with the X-direction coordinates and the Y-direction coordinates of the upper and lower probes, sampling points are obtained. All sampling points acquired during the sampling period constitute the first scatter set.
[0033] S4: After the glass substrate has moved for a specified time, a regular grid is constructed on the first plane of the inspection station. After the glass substrate has moved for a period of time, a certain area of the glass substrate will obviously pass through the inspection station; this certain area of the glass substrate is defined as the first plane. Simultaneously, a regular grid is constructed based on this first plane. The coordinate set of the grid nodes on the first plane in the X direction is G_x = {x_min, x_min+h, x_min+2h, ..., x_max}, and the coordinate set in the Y direction is G_y = {y_min, y_min+h, y_min+2h, ..., y_max}, where h is the grid spacing, and a recommended value is 0.5~2.0 mm. The total number of grid nodes M = |G_x| × |G_y| S5: By combining the first scatter set and the regular grid, the height values of the grid nodes in the regular grid outside the first scatter set are estimated, and a complete two-dimensional surface height matrix of the first plane of the glass substrate is obtained. Some sampling points in the first scatter set may coincide with the grid nodes of the regular grid, but the vast majority of sampling points and grid nodes do not coincide. In this technical solution, it is necessary to estimate the height values of the grid nodes that do not coincide with the first scatter set based on the sampling points in the first scatter set. With the height values of the sampling points in the first scatter set and the estimated height values of the grid nodes, a complete two-dimensional surface height matrix of the first plane of the glass substrate can be obtained. This matrix can represent the height curve of the first plane.
[0034] S6: Perform least squares fitting on the two-dimensional surface height matrix to obtain the optimal reference plane; S7: The warpage of the first plane is obtained based on the optimal reference plane and the two-dimensional surface height matrix. The warpage is obtained by combining the optimal reference plane of the entire first plane and the two-dimensional surface height matrix.
[0035] Preferred, such as Figure 2 As shown, it also includes: S8: Combine the complete two-dimensional surface height matrix of the first plane, the optimal reference plane, and the total number of grid nodes to obtain the flatness of the first plane.
[0036] Preferably, the reciprocating motion includes: the upper and lower probes forming periodic trajectories on the first plane; the periodic trajectories include: triangular waves, sawtooth waves, sine waves, and cosine waves. By combining the movement speeds of the upper and lower probes with the movement speed of the glass substrate in the X direction, a periodic trajectory can be formed from the first set of scattered points, including periodic waveforms such as triangular waves, sawtooth waves, sine waves, and cosine waves. In this technical solution, as... Figure 3As shown, this periodic waveform is a sine wave. The analysis of other waveforms is the same as that of the sine wave, and will not be repeated here. The servo motor drives the probe assembly to perform sinusoidal reciprocating motion along the Y direction with amplitude A and frequency f. At any time t, the probe's position in the Y direction is: y(t) = A·sin(2π·f·t) Combining the previously obtained X-direction position x(t) = v·t, the triplet (t, H_top, H_bottom) at each sampling time is mapped to the sampling point P_i(x_i, y_i) in the spatial coordinate system: where x_i = v·t_i, y_i = A·sin(2π·f·t_i).
[0037] in, Preferably, when the periodic trajectory is a sine wave, the ratio of the specified scanning frequency f of the upper and lower probes to the moving speed v of the glass substrate should satisfy: f / v ≥ 1 / (4A). The sinusoidal scan sweeps from the peak +A to the trough in the Y direction within half a cycle. A, Total travel = 2A; Half-cycle time = T_scan / 2 = 1 / (2f); Therefore, the average scanning speed in the Y direction is: V_scan = 2A / (1 / (2f)) = 4Af; The forward speed of the glass in the X direction is v. Coverage condition: Scanning speed must be ≥ glass forward speed, i.e., v ≤ 4Af; Physical meaning: The average scanning speed of the upper and lower probes in the Y direction is 4Af. Only when the glass transmission speed v does not exceed this scanning speed can it be guaranteed that the probe "scans fast and the glass moves slowly", and the entire strip width is scanned in each half-cycle without leaving any gaps in detection. If the glass moves too fast (v>4Af), the glass will have already rushed ahead before the probe has finished scanning one amplitude, resulting in gaps between the tracks.
[0038] Preferably, the step of combining the first scatter set and the regular grid to estimate the height values of the grid nodes in the regular grid outside the first scatter set, and obtaining the complete two-dimensional surface height matrix of the first plane of the glass substrate, includes:
[0039]
[0040]
[0041] Where k=1,2,...,N are the sampling points in the N known first scattered point sets closest to the target grid node (N is the preset number of nearest neighbors), d_k is the Euclidean distance between the sampling point in the kth first scattered point set and the target grid node, p is the distance decay exponent (recommended value p=2), Z_k is the ΔH_i value of the sampling point k in the kth first scattered point set; Z_g is the height value of the target grid node; the complete two-dimensional surface height matrix is Z_g(x, y), where x and y are the numbers of all grid nodes in the first plane. For the regular surface Z_g(x, y) obtained by the above interpolation, the least squares method is used to fit the best reference plane. The plane equation is expressed as: z_fit = a0+ a1· x + a2· y. Using all M grid nodes as data points, an overdetermined system of equations is constructed and the coefficient vector [a0, a1, a2] is solved: [a0, a1, a2] = argmin Σ j [Z_g(j) a0 a1·x j a2·y j ]²; where the summation ranges from j=1 to M. This least squares problem can be solved directly using the normal equation method or the singular value decomposition (SVD) method. Figure 4 As shown, any target grid node that does not coincide with a point on the sine wave trajectory, i.e., an empty circle point or a grid point to be interpolated, needs to find the sampling point in the nearest first scatter point set, i.e., a solid point, and the Euclidean distance can be used to calculate d_k.
[0042] Preferably, the sampling points in the N known first scatter point sets closest to the target grid node are obtained through KD-Tree queries. A KD-Tree (K-Dimensional Tree) data structure is constructed for the planar coordinates (x_i, y_i) of all sampling points in the first scatter point set P to accelerate the nearest neighbor search in the subsequent interpolation process. The KD-Tree recursively divides the two-dimensional space into a binary tree structure, making the time complexity of the k-nearest neighbor search O(k·log N).
[0043] Preferred, such as Figure 5 As shown, obtaining the warpage of the first plane based on the optimal reference plane and the two-dimensional surface height matrix includes: Calculate the offset distance of each grid node from the optimal reference plane as follows: d_j = Z_g(j) z_fit(x_j, y_j); Using the aforementioned deviation distance, the warpage of the first plane is calculated as follows: W = max(d_j) min(d_j); Where z_fit(x_j, y_j) is the optimal reference plane, Z_g(j) is the height value of the target mesh node, W is the warpage of the first plane, and d_j is the offset distance. Figure 5 It is the interpolated surface mesh.
[0044] Preferably, the step of combining the complete two-dimensional surface height matrix of the first plane, the optimal reference plane, and the total number of mesh nodes to obtain the flatness of the first plane includes:
[0045]
[0046] Where RMSE is the flatness; M is the total number of grid nodes in the regular grid; and d_j is the deviation distance.
[0047] Secondly, such as Figure 6 As shown, a glass substrate warpage detection system 1 is also provided for detecting glass substrates passing through the detection station at a uniform speed, comprising: First acquisition unit 11: used to acquire the X-direction position x(t) = v·t of the glass substrate motion corresponding to time t; Setting unit 12: used to make the upper probe and lower probe, which are respectively set on the upper and lower sides of the detection station, reciprocate in a consistent manner according to a specified amplitude and a specified frequency; The second acquisition unit 13 is used to acquire a first set of scattered points collected by the upper and lower probes after the reciprocating motion; each sampling point in the first set of scattered points is: P_i = (x_i, y_i, ΔH_i); where x_i is the coordinate of the X direction, y_i is the coordinate of the direction that forms a 90-degree angle with the X direction, ΔH_i is the effective height deviation value obtained by differential calculation using the synchronous measurement values of the upper and lower probes; i is the sequence number of the sampling point; Planning unit 14: used to construct a regular grid on the first plane of the detection station after the glass substrate has been moved for a specified time; First calculation unit 15: used to estimate the height value of the grid node in the regular grid outside the first scatter set by combining the first scatter set and the regular grid, and obtain the complete two-dimensional surface height matrix of the first plane of the glass substrate. Fitting unit 16: used to perform least squares fitting on the two-dimensional surface height matrix to obtain the optimal reference plane; Second calculation unit 17: used to obtain the warpage of the first plane based on the optimal reference plane and the two-dimensional surface height matrix.
[0048] Preferred, such as Figure 7 As shown, it also includes: The third calculation unit 18 is used to combine the complete two-dimensional surface height matrix of the first plane, the optimal reference plane, and the total number of grid nodes to obtain the flatness of the first plane.
[0049] Preferably, the reciprocating motion includes: the upper probe and the lower probe forming a periodic trajectory on the first plane; the periodic trajectory includes: triangular wave, sawtooth wave, sine wave and cosine wave.
[0050] Beneficial effects: (1) Online full-surface inspection capability. This invention expands the inspection area of the traditional fixed dual-probe method from a single line segment to the entire surface of the glass substrate by introducing servo-driven reciprocating scanning. The mathematical controllability of the scanning trajectory allows the sampling points to be distributed regularly in space, avoiding blind spots from random sampling and ensuring effective coverage of the entire surface morphology. At the same time, the inspection process is completed entirely during the normal transport of the glass substrate, without the need for line stoppage or offline operation, truly achieving online full inspection, with a single-piece inspection cycle controlled within 10 seconds.
[0051] (2) Lower system hardware cost. This invention only requires adding a servo linear scanning module (servo motor + linear guide) to the existing fixed dual-probe method. The incremental hardware cost is far lower than that of area array structured light or high-density line laser profilometer solutions. The upper and lower probes continue to use industrial-grade laser displacement sensors such as Keyence, without the need to upgrade to more expensive area array sensors. The overall system modification cycle is short and it is easy to install and deploy on existing production lines.
[0052] (3) High computational efficiency, meeting real-time requirements.
[0053] (4) The measurement results are reliable, and tilt and thickness errors are systematically eliminated. This invention inherits the core advantage of the dual-probe differential method—by simultaneously measuring and differentiating with the upper and lower probes, the systematic errors caused by the overall tilt and thickness tolerance of the glass substrate are naturally eliminated, without the need for additional tilt calibration and thickness compensation steps. The least squares plane fitting after interpolation reconstruction further removes the residual low-order trend terms, making the baselines of warpage (range method) and flatness (RMSE method) consistent and the results reproducible.
[0054] (5) Wide applicability and strong scalability. The core of the method of this invention is a universal "line scan dimensionality enhancement" framework that does not depend on specific glass materials or sizes. By simply adjusting parameters such as scanning amplitude A, frequency f, and grid spacing h, it can be adapted to online detection of surface morphology of sheet materials such as glass substrates, metal plates, ceramic substrates, and PCB circuit boards of different specifications.
[0055] Finally, it should be noted that any modification or equivalent substitution of some or all of the technical features based on the device structure and the technical solutions of the embodiments of the present invention, without departing from the corresponding technical solutions of the present invention, shall fall within the patent scope of the device structure and the embodiments of the present invention.
Claims
1. A method for detecting the warpage of a glass substrate, used to detect a glass substrate passing through a detection station at a uniform speed, characterized in that, include: Obtain the X-direction position of the glass substrate at time t, x(t) = v·t; where v is the velocity of the glass substrate in the X direction. The upper and lower probes, respectively set on the upper and lower sides of the testing station, will reciprocate in a consistent manner according to the specified amplitude and frequency. Obtain a first set of scattered points collected by the upper and lower probes after the reciprocating motion; each sampling point in the first set of scattered points is: P_i = (x_i, y_i, ΔH_i); where x_i is the coordinate of the X direction, y_i is the coordinate of the direction that forms a 90-degree angle with the X direction, ΔH_i is the effective height deviation value obtained by differential calculation using the synchronous measurement values of the upper and lower probes; i is the sequence number of the sampling point; After the glass substrate is moved for a specified time, a regular grid is constructed on the first plane of the inspection station. By combining the first scatter set and the regular grid, the height values of the grid nodes in the regular grid outside the first scatter set are estimated, and the complete two-dimensional surface height matrix of the first plane of the glass substrate is obtained. The optimal reference plane is obtained by fitting the two-dimensional surface height matrix using the least squares method. The warpage of the first plane is obtained based on the optimal reference plane and the two-dimensional surface height matrix.
2. The detection method according to claim 1, characterized in that, Also includes: The flatness of the first plane is obtained by combining the complete two-dimensional surface height matrix of the first plane, the optimal reference plane, and the total number of grid nodes.
3. The detection method according to claim 1, characterized in that, The reciprocating motion includes: the upper probe and the lower probe forming a periodic trajectory on the first plane; the periodic trajectory includes: triangular wave, sawtooth wave, sine wave and cosine wave.
4. The detection method according to claim 3, characterized in that, When the periodic trajectory is a sine wave, the ratio of the scanning frequency f of the upper and lower probes to the movement speed v of the glass substrate should satisfy: f / v ≥ 1 / (4A).
5. The detection method according to claim 1, characterized in that, The step of combining the first scatter set and the regular grid to estimate the height values of grid nodes in the regular grid outside the first scatter set, and obtaining the complete two-dimensional surface height matrix of the first plane of the glass substrate, includes: Where k=1,2,...,N are the sampling points in the N known first scattered point sets closest to the target grid node (N is the preset number of nearest neighbors), d_k is the Euclidean distance between the sampling point in the kth first scattered point set and the target grid node, p is the distance decay exponent (recommended value p=2), Z_k is the ΔH_i value of the sampling point k in the kth first scattered point set; Z_g is the height value of the target grid node; the complete two-dimensional surface height matrix is Z_g(x, y), where x and y are the numbers of all grid nodes in the first plane.
6. The detection method according to claim 5, characterized in that, The sampling points in the set of N known first scattered points closest to the target grid node are obtained through KD-Tree query.
7. The detection method according to claim 1, characterized in that, The process of obtaining the warpage of the first plane based on the optimal reference plane and the two-dimensional surface height matrix includes: Calculate the offset distance of each grid node from the optimal reference plane as follows: d_j = Z_g(j) z_fit(x_j, y_j); Using the aforementioned deviation distance, the warpage of the first plane is calculated as follows: W = max(d_j) min(d_j); Where z_fit(x_j, y_j) is the optimal reference plane, Z_g(j) is the height value of the target mesh node, W is the warpage of the first plane, and d_j is the offset distance.
8. The detection method according to claim 7, characterized in that, The step of combining the complete two-dimensional surface height matrix of the first plane, the optimal reference plane, and the total number of mesh nodes to obtain the flatness of the first plane includes: Where RMSE is the flatness; M is the total number of grid nodes in the regular grid; and d_j is the deviation distance.
9. A glass substrate warpage detection system, used to detect glass substrates passing through a detection station at a uniform speed, characterized in that, include: First acquisition unit: used to acquire the X-direction position of the glass substrate at time t, x(t) = v·t; where v is the velocity of the glass substrate in the X direction; Setting unit: Used to make the upper and lower probes, which are respectively set on the upper and lower sides of the detection station, reciprocate in a consistent manner according to a specified amplitude and frequency; The second acquisition unit is used to acquire a first set of scattered points collected by the upper and lower probes after the reciprocating motion. Each sampling point in the first set of scattered points is: P_i = (x_i, y_i, ΔH_i); where x_i is the coordinate of the X direction, y_i is the coordinate of the direction that forms a 90-degree angle with the X direction, ΔH_i is the effective height deviation value obtained by differential calculation using the synchronous measurement values of the upper and lower probes, and i is the sequence number of the sampling point. Planning unit: used to construct a regular grid on the first plane of the inspection station after the glass substrate has been moved for a specified time; First calculation unit: used to estimate the height value of the grid node in the regular grid outside the first scatter set by combining the first scatter set and the regular grid, and to obtain the complete two-dimensional surface height matrix of the first plane of the glass substrate; Fitting unit: used to perform least squares fitting on the two-dimensional surface height matrix to obtain the optimal reference plane; The second calculation unit is used to obtain the warpage of the first plane based on the optimal reference plane and the two-dimensional surface height matrix.
10. The detection system according to claim 9, characterized in that, Also includes: The third calculation unit is used to combine the complete two-dimensional surface height matrix of the first plane, the optimal reference plane, and the total number of grid nodes to obtain the flatness of the first plane.
