A method and system for laser scanning detection of glass surface defects

Abstract

The application discloses a kind of laser scanning detection method and system of glass surface defect, belong to optical intelligent detection technical field.This method establishes multi-angle laser scanning optical path, makes incident angle in 15° to 75° range continuously adjustable, and drives light beam to carry out two-dimensional scanning;Polarization light path system is configured, incident laser is converted into single polarization state linearly polarized light, and reflected light is divided into two ways by beam splitter at detection end, respectively for monitoring total light intensity and measuring light intensity after polarization analysis, and performing polarization state orthogonal alignment process, so that the transmission axis of polarization analysis element and the polarization direction of mirror reflection component are orthogonal;Double-mode scanning detection is executed, low-angle incidence mode is used to quickly establish overall topography profile, and high-angle incidence mode is used for fine detection of specific area.The application can effectively suppress the mirror reflection interference of high-reflectivity glass surface, avoid signal saturation, and realize high-precision and high-efficiency glass surface flatness detection.

Description

Technical Field

[0001] This application belongs to the field of optical intelligent inspection technology, specifically relating to a laser scanning inspection method for glass surface defects. Background Technology

[0002] Laser scanning inspection technology is an important tool in the field of precision measurement. By emitting a laser beam and receiving the reflected signal after its interaction with the surface of an object, it enables non-contact, high-precision measurement of the surface morphology of the object under test. This technology is widely used in industrial manufacturing, construction, and materials science, providing objective data support for quality control and process optimization. During laser scanning inspection, the reconstruction of the surface morphology depends on the effective calculation of the reflected light signal; its core lies in establishing an accurate mapping relationship between the incident laser and the reflected signal.

[0003] Currently, laser scanning-based surface flatness detection methods are mainly designed for rough surfaces with significant diffuse reflection characteristics, such as building walls and concrete components. These materials can scatter incident laser energy relatively uniformly in all directions, ensuring that the detector receives a sufficiently strong reflected signal to complete the topography reconstruction. However, when the object under test is a highly reflective transparent material such as glass, the interaction mechanism between the laser and the material surface changes fundamentally. Glass surfaces have extremely high specular reflectivity and extremely low diffuse reflection components, and some laser energy will penetrate into the material's interior, forming stray light pollution in the measurement optical path after reflection from the bottom surface.

[0004] Existing laser scanning detection schemes generally employ perpendicular or near-perpendicular incidence to simplify the optical path structure and reduce algorithm complexity. However, this approach introduces serious technical drawbacks when applied to glass surfaces: under perpendicular incidence, most of the laser energy is concentrated and reflected by the mirror surface, causing the signal intensity received by the detector to far exceed the normal operating threshold, resulting in signal saturation and glaring specular glare. Simultaneously, the specularly reflected light carries almost no effective information about the surface microstructure, while the scattered signal, which truly reflects surface undulations, is submerged in strong background noise. Furthermore, the transparency of glass introduces stray light interference, further degrading the quality of the measurement signal. These problems are coupled, making it difficult for existing technologies to obtain effective flatness detection data on highly reflective transparent materials.

[0005] Therefore, there is an urgent need to develop a laser scanning detection method that can effectively suppress specular reflection interference, separate effective scattering signals, and obtain complete surface morphology, taking into account the optical properties of highly reflective materials such as glass. Summary of the Invention

[0006] The purpose of this invention is to provide a laser scanning detection method for glass surface defects, which can effectively solve the technical problems of signal saturation, specular reflection interference, and incomplete acquisition of surface morphology data that occur in the prior art when detecting the surface flatness of high reflectivity transparent materials.

[0007] To achieve the above objectives, the technical solution adopted by the present invention is as follows:

[0008] A laser scanning detection method for glass surface defects includes the following specific steps:

[0009] A multi-angle laser scanning optical path is established, and the laser beam collimated by the collimating lens group is incident on the multi-angle adjustable laser incident scanning device. The incident angle of the laser beam is adjusted by the precision angle adjustment mechanism in the multi-angle adjustable laser incident scanning device, so that the incident angle of the incident laser reaching the glass surface to be tested is continuously adjustable within the preset working range of 15° to 75°. The laser beam is driven to perform two-dimensional scanning motion on the glass surface to be tested by the scanning motion mechanism.

[0010] A polarization beam splitter system is configured, and a polarization beam splitter with an extinction ratio better than 1000:1 is set in the incident light path to convert the incident laser into linearly polarized light with a single polarization state. A beam splitter detection light path is constructed at the detection end, and the reflected light is split into two paths by a beam splitter. One path is used to monitor the total light intensity, and the other path passes through a rotatable polarization analysis element to measure the light intensity. By performing a polarization state orthogonal alignment process, the orthogonal extinction position is locked, so that the transmission axis direction of the polarization analysis element is orthogonal to the polarization direction of the specular reflection component of the glass surface.

[0011] Dual-mode scanning detection is performed by adjusting the laser incident angle to the range of 15° to 30° and executing the low-angle incident mode to quickly establish the overall morphological profile of the glass surface under test; the laser incident angle is adjusted to the range of 50° to 75° and the high-angle incident mode is executed to perform fine detection on specific areas and obtain high-precision micro-morphological data.

[0012] Furthermore, the precision angle adjustment mechanism includes an angle adjustment base, a rotary support shaft, a precision lead screw drive pair driven by a stepper motor or a servo motor, and an angle feedback sensor.

[0013] The precision lead screw transmission pair converts the rotational motion of the stepper motor or servo motor into the linear motion of the lead screw nut, and converts the linear displacement of the lead screw nut into the angular displacement of the rotating support shaft through the cooperation of the lever and the eccentric pin, so as to change the propagation direction of the laser beam.

[0014] The angle feedback sensor uses a circular grating encoder or a magnetic grating encoder to detect the actual angular position of the rotating support shaft in real time and output a position signal. The position signal is fed back to the driver of the stepper motor or servo motor to form a closed-loop control, with an angle resolution better than 0.1°.

[0015] Furthermore, the configured polarization beam splitting optical path system specifically includes:

[0016] The polarizing beam splitter is set in the optical path between the laser source and the glass surface to be tested. The polarizing beam splitter adopts a multi-layer thin film coating structure and is mounted on a mounting bracket equipped with an angle fine adjustment mechanism. The pitch angle and rotation angle of the angle fine adjustment mechanism are both ±5°, and the adjustment resolution is 0.001°, which is used to precisely adjust the polarizing beam splitter to meet the Brewster angle condition.

[0017] At the detection end, a focusing lens group is set to converge the reflected light. Behind the focusing lens group, the beam splitter is set to split the converged reflected light into a first beam and a second beam. The first beam is directly incident on the first photodetector for real-time monitoring of the total light intensity of the reflected light. A polarization analysis element is set on the optical path of the second beam. The polarization analysis element is a polarizer with an extinction ratio parameter better than 1000:1 and is mounted on a precision rotating stage driven by a stepper motor through a worm gear transmission mechanism. The beam passes through the polarization analysis element and is then incident on the second photodetector.

[0018] Furthermore, the polarization state orthogonal alignment process is specifically as follows:

[0019] The data processing unit controls the stepper motor to drive the precision rotary table to rotate the polarization analysis element continuously within the range of 0° to 180° in a preset angle step of 0.01°, and records the output voltage signal of the first photodetector and the output voltage signal of the second photodetector at each rotation angle in real time, and calculates the real-time light intensity ratio between the second photodetector and the first photodetector.

[0020] The data processing unit searches for the orthogonal extinction position where the light intensity ratio reaches its minimum value from the data sequence of all recorded rotation angles and light intensity ratios, and controls the stepper motor to lock the precision rotary table at the searched orthogonal extinction position.

[0021] Furthermore, in the dual-mode scanning detection, the mode switching adopts an adaptive triggering switching method:

[0022] During the low-angle incident mode scanning process, the data processing unit performs surface roughness estimation, defect feature identification, and flatness deviation threshold determination on the real-time acquired morphology data. When the analysis results show that the evaluation result of a certain sub-region exceeds the preset threshold, the sub-region is automatically marked as the region of interest and a scanning trigger command is generated.

[0023] After receiving the scan trigger command, the scanning control system smoothly switches the laser incident angle at the boundary of the sub-region to the angle range corresponding to the high-angle incident mode to perform a secondary fine scan. After the scan is completed, it switches back to the low-angle incident mode to resume full-area scanning.

[0024] Furthermore, after acquiring the topographic data of the low-angle and high-angle incident modes, polarization state analysis was performed:

[0025] The polarization state of the reflected light is described by the Stokes vector method. The polarization analysis element is rotated sequentially to 0°, 90°, 45° and 135° while the voltage signal output by the second photodetector is recorded simultaneously. The components of the Stokes vector are calculated and the degree of polarization is obtained.

[0026] The Mueller matrix method is used to obtain four sets of linearly independent input Stokes vectors by changing the polarization state of the incident light at least four times and measuring the corresponding output Stokes vectors, and then solving for all elements of the Mueller matrix.

[0027] A mapping relationship between polarization degree and surface roughness is established. Using a polarization degree threshold of 0.3 to 0.5 as a criterion, the light intensity data of sampling points with a polarization degree less than the preset polarization degree threshold are marked as valid scattered signals. The light intensity data of sampling points with a polarization degree greater than or equal to the preset polarization degree threshold are calculated and replaced with equivalent scattered light intensity through the mapping relationship.

[0028] Furthermore, after polarization state analysis, multi-angle topography data are fused to generate a unified three-dimensional topography model:

[0029] A weighted fusion method is adopted to evaluate the data quality of the topography data acquired by the low-angle incident mode and the high-angle incident mode based on the signal-to-noise ratio, sampling density and effective data coverage, and calculate the comprehensive data quality evaluation score. The fusion weight is calculated based on the comprehensive data quality evaluation score, and the two types of topography data are weighted and summed to obtain the weighted fused topography data.

[0030] A Bayesian fusion method is adopted, using low-angle incident pattern topography data as prior information and a Gaussian Markov random field model as the prior distribution, and high-angle incident pattern topography data as observation information and a Gaussian noise observation model is established. The Markov chain Monte Carlo sampling algorithm is used to calculate the posterior distribution, and the mean of all topography samples is taken as the topography data after Bayesian fusion.

[0031] Furthermore, the flatness parameters are calculated based on a unified three-dimensional topography model:

[0032] The deviation of each sampling point relative to the ideal reference plane is calculated by fitting an ideal reference plane using the least squares method and singular value decomposition.

[0033] The deviation data were preprocessed by removing outliers using the 3σ criterion and filling in missing data using the nearest neighbor interpolation method.

[0034] Global smoothness parameters are calculated based on the preprocessed deviation, including arithmetic mean deviation, root mean square deviation, maximum peak-to-valley difference, kurtosis coefficient, and uniformity coefficient.

[0035] The surface of the glass to be tested is uniformly divided into multiple rectangular sub-regions in the horizontal plane. The local flatness parameters of each rectangular sub-region are calculated, and the Gaussian filtering algorithm is used to smooth the local flatness parameters.

[0036] Furthermore, the scanning motion mechanism adopts a gantry structure or a cantilever structure, and is composed of an X-axis linear module and a Y-axis linear module with a movement accuracy better than 1μm and a repeatability positioning accuracy better than 0.5μm. The scanning path of the scanning motion mechanism can be selected from grating scanning mode, spiral scanning mode or adaptive path planning mode.

[0037] A laser scanning detection system for glass surface defects includes:

[0038] Laser source, used to emit a laser beam;

[0039] A collimating lens group is used to collimate the light beam emitted from the laser source into a parallel light beam;

[0040] A multi-angle adjustable laser incident scanning device includes a precision angle adjustment mechanism and a scanning motion mechanism. The precision angle adjustment mechanism is used to adjust the incident angle of the laser beam so that the incident angle reaching the glass surface to be tested is continuously adjustable within the range of 15° to 75°. The angle resolution of the precision angle adjustment mechanism is better than 0.1°. The scanning motion mechanism is used to drive the laser beam to perform two-dimensional scanning motion on the glass surface to be tested.

[0041] A polarization beam splitting optical path system includes a polarization beam splitter with an extinction ratio better than 1000:1 disposed in the incident light path, and a beam splitting detection optical path disposed in the detection end; the beam splitting detection optical path includes a beam splitter, a first photodetector, a polarization analysis element driven to rotate by a stepper motor, and a second photodetector, wherein the beam splitter is used to split the reflected light into a first beam and a second beam.

[0042] The data processing unit is used to perform polarization state orthogonal alignment, polarization state analysis and processing, multi-angle topography data fusion, and flatness parameter calculation.

[0043] The scanning control system has a core controller that internally runs a scanning trajectory planning module, a motor motion control module, a data acquisition synchronization module, and a mode switching decision module, which are used to control the scanning motion mechanism, the laser incident angle, and the execution and switching of the dual-mode scanning strategy.

[0044] In summary, this application includes at least one of the following beneficial technical effects:

[0045] 1. This invention utilizes the synergistic effect of a multi-angle adjustable laser incident scanning device and a polarization beam splitter system to dynamically adjust the incident angle within a preset angle range and separate S-polarized light from P-polarized light using a high-performance polarization beam splitter that meets preset extinction ratio performance indicators. This effectively blocks the specular reflection component from entering the detector, avoiding detector signal saturation and specular glare problems caused by high reflectivity, and effectively extracts the true scattering signal from the glass surface, laying the foundation for subsequent morphology data processing.

[0046] 2. This invention utilizes a dual-mode scanning strategy to comprehensively leverage the wide-range, rapid scanning capability of the low-angle incident mode and the high-precision local scanning capability of the high-angle incident mode. Combined with a multi-angle data fusion algorithm of weighted fusion or Bayesian fusion, it generates a unified three-dimensional morphological model of the glass surface, compensating for the deficiencies of data blind spots and insufficient accuracy under a single incident angle, thus ensuring both detection efficiency and accuracy requirements.

[0047] 3. This invention establishes a mapping relationship between the polarization state change of reflected light and surface roughness through a polarization state analysis algorithm, distinguishes between the components of directly reflected light and scattered light, and extracts the true surface scattering signal. Combined with subsequent processing steps such as smoothing, error correction, and zoning identification and grading evaluation, it significantly improves the calculation accuracy of glass surface flatness parameters and the reliability of test results, providing objective and accurate technical support for the quality evaluation of glass products. Attached Figure Description

[0048] Figure 1 A schematic diagram of the overall technical solution for laser scanning detection of defects on glass surfaces;

[0049] Figure 2 This is a schematic diagram illustrating the core principle of a multi-angle laser scanning and polarization beam splitting optical path system.

[0050] Figure 3 This is a schematic diagram illustrating the execution flow of the dual-mode scanning strategy;

[0051] Figure 4 This is a schematic diagram of the processing logic framework for polarization state analysis and multi-angle data fusion.

[0052] Figure 5 This is a flowchart illustrating the process of calculating flatness parameters and evaluating zones. Detailed Implementation

[0053] To make the objectives, technical solutions, and advantages of this invention clearer, the following description is provided in conjunction with the appendix. Figure 1-5 The present invention will be further described in detail below with reference to specific embodiments.

[0054] Firstly, this invention provides a laser scanning detection method for glass surface defects. This method integrates core components such as a multi-angle adjustable laser incident scanning device, a polarization beam splitter optical path system, a dual-mode scanning strategy, a polarization state analysis algorithm, and a multi-angle data fusion algorithm, forming a complete technical solution for glass surface flatness detection. The steps are described in detail below.

[0055] First, for step S1, a multi-angle laser scanning optical path is established, which includes the following sub-steps.

[0056] Step S101: The multi-angle adjustable laser incident scanning device is arranged on one side of the glass surface to be tested. The laser beam emitted from the laser source is first collimated by a collimating lens group to form a parallel beam before entering the multi-angle adjustable laser incident scanning device.

[0057] Step S102: A precision angle adjustment mechanism is installed on the beam transmission path of the multi-angle adjustable laser incident scanning device. The precision angle adjustment mechanism includes an angle adjustment base, a rotating support shaft, a precision lead screw drive pair, and an angle feedback sensor. The angle adjustment base is fixedly mounted on the optical platform or scanning frame, and the rotating support shaft is supported on the angle adjustment base by a high-precision rolling bearing, thereby ensuring the smoothness of the rotation process and the repeatability of the positioning accuracy.

[0058] In step S103, the precision lead screw drive pair is driven by a stepper motor or a servo motor, and the output shaft of the motor is connected to the lead screw through a coupling. When the lead screw rotates, it drives the lead screw nut to move linearly along the lead screw axis. A lever is fixed on the lead screw nut, and the other end of the lever slides with an eccentric pin on the rotating support shaft, thereby converting the linear displacement of the lead screw nut into the angular displacement of the rotating support shaft.

[0059] The rotating support shaft further drives the reflector or prism assembly mounted on it to rotate, changing the propagation direction of the laser beam so that the incident laser reaching the glass surface to be tested has a continuously adjustable incident angle.

[0060] Step S104: The angle feedback sensor is selected from circular grating encoders or magnetic grating encoders to detect the actual angular position of the rotating support shaft in real time and output a position signal.

[0061] The position signal is fed back to the driver of the stepper motor or servo motor, and compared with the commanded position to form a closed-loop control, ensuring the accuracy and stability of angle adjustment.

[0062] The selection of stepper motors or servo motors must meet the requirement of angular resolution better than 0.1°. For example, when using a stepper motor with a lead screw of 2mm and a step angle of 1.8°, and with a 16-microstep driver, the linear displacement of the lead screw nut for each step is 0.00125mm, which translates to an angular displacement of approximately 0.01125° on the rotary support shaft, meeting the 0.1° resolution requirement. If a servo motor is used, its built-in 17-bit encoder can achieve position feedback with an angular resolution on the order of 0.001°.

[0063] Step S105: The preset working range of the laser incident angle is set to 15° to 75°, with a preferred working angle range of 30° to 45°. Within the preset working range, the incident laser and the glass surface are in an oblique relationship. Part of the incident energy returns in the form of specular reflection, while the other part of the energy penetrates into the glass or propagates in various directions in the form of scattering.

[0064] The oblique incidence geometry alters the spatial distribution characteristics of the reflected light, thus avoiding the problem of detector saturation caused by most of the energy being concentrated in the mirror reflection when incident perpendicularly.

[0065] Step S106: The multi-angle adjustable laser incident scanning device also includes a scanning motion mechanism for realizing two-dimensional scanning motion of the laser beam on the surface of the glass to be tested.

[0066] The scanning motion mechanism adopts a gantry or cantilever structure, consisting of an X-axis linear module and a Y-axis linear module. The movement accuracy of both the X-axis and Y-axis linear modules is better than 1μm, and the repeatability is better than 0.5μm.

[0067] In step S107, the laser beam, after angle adjustment, is projected onto the surface of the glass to be tested and moves along a preset scanning path under the drive of the scanning motion mechanism, completing a full-coverage scan of the entire area to be tested. The scanning path can be planned using a raster scanning mode, a spiral scanning mode, or an adaptive path planning mode to adapt to glass samples of different shapes and sizes.

[0068] After establishing the multi-angle laser scanning optical path, the laser beam can obliquely irradiate the glass surface with high repeatability and continuously adjustable incident angle, and cover the entire area under test in combination with two-dimensional scanning motion. This provides reliable optical front-end conditions for subsequent configuration of the polarization-splitting optical path system, execution of the dual-mode scanning strategy, and extraction of real surface scattering signals through polarization state analysis.

[0069] For step S2, configuring the polarization beam splitting optical path system specifically includes the following sub-steps.

[0070] Step S201: A polarizing beam splitter assembly is installed on the incident light path side in the optical path between the laser source and the glass surface to be tested. The polarizing beam splitter in the polarizing beam splitter assembly adopts a multi-layer thin-film coating structure. By controlling the thickness and refractive index of each thin film, the polarizing beam splitter exhibits different transmittance and reflectance characteristics for S-polarized light and P-polarized light.

[0071] The extinction ratio parameter of a polarizing beam splitter is better than 1000:1, meaning that the ratio of the target polarization component to the orthogonal polarization component transmitted through the polarizing beam splitter is not less than 1000:1. When natural or unpolarized light emitted from a laser source is incident on the surface of the polarizing beam splitter, the S-polarization component dominates in the reflected light, and the P-polarization component dominates in the transmitted light, thereby converting the incident laser into linearly polarized light with a single polarization state.

[0072] The polarizing beam splitter is mounted on a mounting bracket equipped with an angle fine-tuning mechanism. The pitch and rotation angles of the angle fine-tuning mechanism are adjustable within a range of ±5°, with an adjustment resolution of 0.001°. The placement angle of the polarizing beam splitter is precisely adjusted using the angle fine-tuning mechanism to meet the Brewster angle condition or the preset polarization state output requirements.

[0073] The laser beam emitted from the laser source is collimated by a collimating lens group and then incident on a polarizing beam splitter. After transmission or reflection by the polarizing beam splitter, it becomes linearly polarized light with a single polarization state. The linearly polarized light then enters a multi-angle adjustable laser incident scanning device for angle adjustment, illuminating the glass surface to be tested at a preset incident angle.

[0074] Through step S201, linearly polarized light with a pure and stable polarization state is obtained on the incident light path, providing a clear polarization reference for subsequently distinguishing the specular reflection component and the scattering component from the reflected light.

[0075] Step S202: Construct a beam-splitting detection optical path in front of the detector. First, a focusing lens group is set up to converge the light beam reflected back from the surface of the glass to be tested. A beam splitter is set up in the optical path behind the focusing lens group. The beam splitter splits the converged reflected light into a first beam and a second beam according to a preset ratio. The first beam is directly incident on the first photodetector to monitor the total light intensity of the reflected light in real time. The total light intensity voltage signal output by the first photodetector is denoted as... .

[0076] A polarization analysis element assembly is placed in the optical path of the second beam. The polarization analysis element is a polarizer with an extinction ratio better than 1000:1. The polarization analysis element is mounted on a precision rotary stage driven by a stepper motor. The precision rotary stage has a hollow aperture for the beam to pass through. The stepper motor and the precision rotary stage are connected by a worm gear transmission mechanism, which transmits the rotational motion of the stepper motor to the polarization analysis element.

[0077] The second beam passes through the polarization analysis element and is incident on the second photodetector. The light intensity voltage signal output by the second photodetector is denoted as... The data processing unit collects data synchronously. and .

[0078] In step S202, a dual-channel differential detection optical path is constructed at the detection end. One path is used to obtain the light intensity reference, and the other path is used to measure the light intensity after polarization filtering, providing a hardware basis for subsequent search for orthogonal extinction positions.

[0079] Step S203: Before starting the formal scanning and inspection of the glass surface to be tested, a polarization state orthogonal alignment process is performed. The data processing unit controls a stepper motor to drive a precision rotary stage in preset angular steps of 0.01°, causing the polarization analysis element to rotate continuously within the range of 0° to 180°. During the rotation of the polarization analysis element, the data processing unit records the polarization state at each rotation angle in real time. and The numerical values ​​are calculated, and the ratio of the output voltage of the second photodetector to that of the first photodetector is determined, i.e., the real-time light intensity ratio. The calculation formula is:

[0080]

[0081] in, Represents the real-time light intensity ratio. This represents the output voltage signal of the second photodetector. This represents the output voltage signal of the first photodetector.

[0082] Step S204: The data processing unit records the ratios of all rotation angles to light intensity. Search for the light intensity ratio in the data sequence. The angular position of the precision rotary table when the minimum value is reached is defined as the orthogonal extinction position.

[0083] The data processing unit then controls the stepper motor to precisely lock the precision rotary stage at the found orthogonal extinction position, ensuring that the transmission axis of the polarization analysis element is orthogonal to the polarization direction of the specular reflection component of the glass surface. After completing the polarization state orthogonal alignment process, the signal output by the second photodetector during subsequent scanning and detection mainly originates from the scattered light component of the glass surface, while the specular reflection component is blocked.

[0084] Through steps S203 and S204, the automatic orthogonal alignment of the polarization analysis element is completed by data acquisition and closed-loop control, which filters out strong specular reflection signals that would cause detector saturation from a physical level and separates the scattering signal carrying the micro-morphological information of the glass surface.

[0085] Step S2 completes the full configuration of the polarization beam splitter system. At the light source end, a high extinction ratio polarization beam splitter and an angle fine-tuning mechanism are used to purify the incident laser into linearly polarized light with a single polarization state. At the detector end, a beam splitter detection optical path is constructed, including a beam splitter, dual photodetectors, a polarization analysis element, and a precision rotating stage. Based on this, through a polarization state orthogonal alignment process, the orthogonal extinction position is automatically searched and locked, making the transmission axis of the polarization analysis element orthogonal to the polarization direction of the mirror-reflected light.

[0086] The entire optical path system blocks the mirror-reflected light from entering the detector at the hardware level, effectively avoiding detector signal saturation. At the same time, it separates the scattered light signal that can reflect the morphology of the glass surface, providing a high-quality raw optical signal for implementing the polarization state analysis algorithm in the subsequent step S4.

[0087] Then, for step S3, dual-mode scanning detection is performed, which includes the following sub-steps.

[0088] Step S301: Adjust the laser incident angle to the range of 15° to 30° and execute the low-angle incident scanning mode. The laser beam is obliquely incident on the glass surface to be tested at a small incident angle, and the projected area of ​​the light spot on the glass surface is large, covering an area approximately 2 to 3 times the area of ​​the incident light spot.

[0089] At the same scanning speed, the low-angle incident mode can cover a larger detection area, enabling rapid establishment of the overall morphological profile of the glass surface under test. The morphological data acquired by the low-angle incident mode is denoted as... .

[0090] Step S302: Adjust the laser incident angle to the range of 50° to 75° and execute the high-angle incident scanning mode. The laser beam is obliquely incident on the glass surface to be tested at a large incident angle, the projected area of ​​the light spot on the glass surface is small, and the angle between the incident light and the surface normal is large.

[0091] The spatial distribution of reflected light is more sensitive to changes in surface microstructure, and the scattered signal contains more high-frequency information. High-angle incidence modes offer higher surface topography resolution, enabling fine-grained detection of specific regions and acquisition of high-precision microstructure data. The topography data acquired by high-angle incidence modes is denoted as... .

[0092] Step S303: Based on the detection requirements, intelligently switch or coordinate between the two scanning modes. The switching strategy is divided into preset trigger switching and adaptive trigger switching.

[0093] The preset trigger switching is executed according to the pre-set scanning sequence. First, the low-angle incident mode is executed to perform a full-area scan. After completion, it switches to the high-angle incident mode to perform a fine scan of the pre-specified area of ​​interest.

[0094] The adaptive triggering switching process is as follows: During the low-angle incident mode scanning process, the data processing unit analyzes the real-time acquired morphology data. The analysis includes surface roughness estimation, defect feature identification, and flatness deviation threshold determination.

[0095] When the analysis results show that the surface roughness estimate of a certain sub-region exceeds the preset roughness threshold, or the defect feature recognition algorithm detects a suspected defect, or the average height deviation of this sub-region relative to the neighborhood exceeds the preset flatness deviation threshold, the data processing unit automatically marks this sub-region as a region of interest and generates a scanning trigger command for a high-angle incident mode.

[0096] After receiving the scan trigger command, the scanning control system pauses the low-angle incident mode at the boundary of this sub-region and switches the laser incident angle to the angle range corresponding to the high-angle incident mode, performing a second fine scan of this sub-region. After the scan of this sub-region is completed, the scanning control system switches the laser incident angle back to the low-angle incident mode and resumes full-area scanning.

[0097] The collaborative working mode allows low-angle and high-angle incident modes to operate simultaneously or alternately over time. The two modes share the scanning motion mechanism through time-segmentation or spatial segmentation. Time-segmentation means executing the two scanning modes separately within different time periods; spatial segmentation means dividing the glass surface under test into several strip or block regions, with the scanning motion mechanism using the low-angle incident mode for region 1 and the high-angle incident mode for region 2 during a single unidirectional scan stroke.

[0098] Step S304: The core controller of the scanning control system adopts an industrial-grade programmable logic controller or a high-performance embedded processor. The core controller internally runs a scanning trajectory planning module, a motor motion control module, a data acquisition and synchronization module, and a mode switching decision module.

[0099] The scanning trajectory planning module reads the geometric parameters of the glass under test and the preset scanning resolution to generate a scanning path. The scanning path meets the dual requirements of coverage integrity and scanning efficiency.

[0100] The motor motion control module sends motion commands to the X-axis linear module, Y-axis linear module and precision angle adjustment mechanism, with a control cycle better than 1ms and a position control accuracy better than 1μm.

[0101] The data acquisition synchronization module coordinates the timing relationship between the laser, the first photodetector, the second photodetector, and the scanning motion mechanism. The module sends trigger pulses to the laser and sampling pulses to the first and second photodetectors, while simultaneously latching the position encoder readings of the X-axis and Y-axis linear modules. This ensures that the position coordinates, incident angle, and total light intensity voltage signal output by the first photodetector are recorded at each sampling point. and the polarization intensity voltage signal output by the second photodetector Accurate correlation is achieved to form light intensity data pairs labeled with spatial coordinates and incident angle.

[0102] The mode switching decision module receives an adaptive trigger signal or a preset trigger signal from the data processing unit and generates a mode switching command. The mode switching command includes the target incident angle, the switching start time, and the switching transition duration. The switching process is controlled by the acceleration and deceleration curve of a stepper motor or servo motor, so that the laser incident angle smoothly transitions from the current angle to the target angle, avoiding data distortion caused by motion disturbances or changes in the optical path at the moment of switching.

[0103] Step S3 involves executing two complementary scanning modes, low-angle and high-angle, and intelligently connecting the modes using three methods: preset triggering, adaptive triggering, or collaborative working, to acquire basic topographic data with complementary coverage and distinct levels of accuracy. and This provides input for data fusion in step S5.

[0104] Step S4 involves polarization state analysis, which includes the following sub-steps.

[0105] Step S401: The polarization state of the reflected light is described using the Stokes vector method. The Stokes vector contains four components, namely... , , , . It represents the total intensity of the reflected light. The intensity difference between horizontally polarized light and vertically polarized light is defined as... ,in This represents the light intensity measured after passing through a polarization analysis element with a transmission axis of 0°. This represents the light intensity measured after passing through a polarization analysis element with a transmission axis of 90°.

[0106] The intensity difference characterizing the difference between light polarized at 45° and light polarized at 135° is defined as... ,in This represents the light intensity measured after passing through a polarization analysis element with a transmission axis of 45°. This represents the light intensity measured after passing through a polarization analysis element with a transmission axis of 135°.

[0107] The intensity difference between right-handed and left-handed circularly polarized light is defined as... ,in Represents the intensity of right-handed circularly polarized light. This represents the intensity of left-handed circularly polarized light.

[0108] The actual measurement of the Stokes vector is accomplished using a rotating polarization analysis element. The data processing unit controls a stepper motor to drive a precision rotary stage, sequentially rotating the transmission axis of the polarization analysis element to 0°, 90°, 45°, and 135°, while simultaneously recording the voltage signal output from the second photodetector. This signal is then converted using the photoelectric conversion coefficient to obtain... , , and .

[0109] Subsequently, the data processing unit calculates each component of the Stokes vector according to the following formula:

[0110]

[0111]

[0112]

[0113] for Component (defined as) , The intensity of right-handed circularly polarized light. (For left-hand circularly polarized light intensity), a quarter-wave plate needs to be inserted in the optical path in front of the polarization analysis element during measurement. The intensity of right-hand and left-hand circularly polarized light is measured separately. and Then calculate according to the above definition. .

[0114] Step S402: The polarization state transformation of the reflected light is characterized using the Mueller matrix method. Mueller matrix. It is a 4×4 real matrix, its elements middle, Represents row index, Represents column index, and The values ​​of are all from 1 to 4. Mueller matrix. Describe the input Stokes vector With output Stokes vector The linear transformation relationship between them is as follows:

[0115] in, The Stokes vector represents the linearly polarized light before it strikes the glass surface under test. The Stokes vector represents the reflected light after reflection from the surface of the glass under test. The Mueller matrix represents the surface of the glass to be tested.

[0116] Muller matrix The method for obtaining the polarization state of the incident light is as follows: by changing the polarization state of the incident light at least four times using a multi-angle adjustable laser incident scanning device, four sets of linearly independent polarization states are obtained. and measure the corresponding The Mueller matrix is ​​obtained by constructing a system of linear equations. All 16 elements. Mueller matrix. The physical meaning can be broken down into forward transmission coefficient, reflection coefficient, polarization degree change, and phase delay.

[0117] Step S403: Establish the mapping relationship between polarization state parameters and surface roughness. The mapping relationship is established by experimental calibration.

[0118] The experimental calibration process is as follows: Select a set of standard glass samples with known surface roughness parameters. The roughness parameters of the standard glass samples include the arithmetic mean roughness. Maximum peak and valley height and root mean square roughness For each standard glass sample, the scanning detection process from steps S1 to S3 is executed to obtain the Stokes vector corresponding to each sampling point and calculate the degree of polarization. polarization degree The calculation formula is:

[0119]

[0120] in, Represents the degree of polarization. Represents the total intensity of the reflected light. This represents the intensity difference between horizontally polarized light and vertically polarized light. This represents the intensity difference between light polarized at 45° and light polarized at 135°. This represents the intensity difference between right-handed and left-handed circularly polarized light.

[0121] The polarization degree of each standard glass sample With known arithmetic mean roughness To link them together, forming a system of... Value and Correspondence The data pairs consist of values. Using data pairs from all standard glass samples, an empirical formula is established through cubic polynomial fitting. ,in , , , Let be the coefficients of the polynomial to be fitted. The least squares method is used to solve for these coefficients, specifically: Let the total number of... The first standard sample, the first The polarization degree of each sample is Arithmetic mean roughness is Construct the objective function To make him , , , The partial derivatives are zero, and the coefficients are obtained by solving the system of linear equations. As an example, a typical set of calibration data is as follows: , , , The cubic polynomial obtained by least-squares fitting is: In practical applications, the polynomial coefficients need to be recalculated based on the measured data of the standard samples used. This empirical formula... This refers to the mapping relationship between polarization state parameters and surface roughness.

[0122] Step S404: Extract the actual surface scattering signal. For each sampling point in the scanning detection, the data processing unit calculates the degree of polarization according to step S401. The calculated degree of polarization With preset polarization threshold Comparison, The value ranges from 0.3 to 0.5.

[0123] When polarization degree Less than the preset polarization degree threshold When the reflected light at this sampling point is determined to be dominated by the scattering component, the light intensity data of this sampling point is directly marked as an effective scattering signal for subsequent topography reconstruction.

[0124] When polarization degree Greater than or equal to the preset polarization degree threshold At that time, it was determined that the reflected light from this sampling point was severely interfered with by the specular reflection component.

[0125] At this point, the data processing unit invokes the mapping relationship established in step S403. The degree of polarization of the sampling point The corresponding surface roughness estimate is calculated, and the light intensity data of this sampling point is replaced with the equivalent scattered light intensity generated by the scattering model based on the surface roughness estimate, in order to compensate for specular reflection interference.

[0126] Step S4 establishes a complete mathematical description of the polarization state of the reflected light using the Stokes vector method and the Mueller matrix method. The mapping relationship between the degree of polarization and the surface roughness is determined by standard sample experiments. The scattering signal and specular reflection interference are separated by the polarization degree threshold. The effective scattering signal or the corrected equivalent scattering signal of each sampling point is used as the output to provide high-quality input data after polarization state analysis for the multi-angle topography data fusion in step S5.

[0127] For step S5, the multi-angle topographic data is fused, which specifically includes the following sub-steps.

[0128] Step S501: A weighted fusion method is used to fuse the topography data obtained from the low-angle incident mode and the topography data obtained from the high-angle incident mode.

[0129] First, the topographic data acquired from the low-angle incident mode. Topographic data obtained from high-angle incident mode Data quality assessments were conducted separately. The assessment metrics included signal-to-noise ratio (SNR), sampling density, and effective data coverage. SNR was obtained by analyzing the power ratio of the light intensity signal to the noise signal at each sampling point; a higher SNR indicates more reliable data. Sampling density was obtained by counting the number of sampling points per unit area; a higher sampling density indicates higher spatial resolution. Effective data coverage was obtained by counting the proportion of effective sampling points to the total number of sampling points; effective sampling points are those that meet the preset light intensity range and polarization state conditions.

[0130] The above three data quality assessment indicators are denoted as: Signal-to-noise ratio (SNR) index Sampling density index Effective data coverage index The three data quality assessment indicators were weighted and summed to obtain the comprehensive data quality assessment score. Topographic data acquired using the low-angle incident mode. The overall data quality assessment score is recorded as Topographic data acquired by high-angle incident mode The overall data quality assessment score is recorded as . and The calculations all use the same weighting formula, with For example, the calculation formula is:

[0131]

[0132] in, The signal-to-noise ratio (SNR) metric representing topographic data under low-angle incidence mode. The sampling density index represents the topography data under low-angle incidence mode. This represents the effective data coverage index for topographic data under low-angle incidence mode. Weighting coefficients. , , As a preset value, it satisfies For example, take , , .

[0133] Then, based on the comprehensive data quality assessment score and Calculate the fusion weights. The fusion weights for the low-angle incident mode. Fusion weights corresponding to high-angle incident modes The calculation formula is:

[0134]

[0135]

[0136] in, The fusion weights represent the topographic data of low-angle incident modes. The fusion weights represent the topographic data of high-angle incident modes. The overall data quality assessment score represents the topographic data of the low-angle incident mode. A comprehensive data quality assessment score representing the topographic data of high-angle incident modes. Fusion weights. and satisfy .

[0137] Finally, the morphology data are weighted and summed using fusion weights to obtain the weighted fused morphology data. The calculation formula is:

[0138]

[0139] in, This represents the morphological data after weighted fusion. This represents the topographic data acquired using the low-angle incident mode. This represents the topographic data acquired using the high-angle incident mode.

[0140] When the signal-to-noise ratio and effective data coverage of the high-angle incident mode are both better than those of the low-angle incident mode, Greater than , Greater than High-angle incident mode topography data dominates the fusion results.

[0141] Step S502: The topography data obtained from the low-angle incident mode and the topography data obtained from the high-angle incident mode are fused using the Bayesian fusion method.

[0142] Bayesian fusion methods model the multi-angle data fusion problem as a Bayesian inference problem. Topographic data acquired using low-angle incident patterns is used as prior information, denoted as... , This describes the initial estimation of the surface topography of the glass under test in the absence of high-angle incident mode topography data. Topography data acquired using the high-angle incident mode is used as the observation information and is denoted as... , Provides the likelihood function under the condition of high-angle incident mode topography data.

[0143] The specific calculation process of the Bayesian fusion method is as follows.

[0144] Step 1: Obtaining topographic data from the low-angle incident mode Preprocessing is performed to generate a gridded depth map, which is then used as the prior distribution. Sampling implementation. Prior distribution. The model is chosen to be a Gaussian Markov random field model, in which the height value of each grid point follows a joint Gaussian distribution, and the height difference between adjacent grid points has a smoothing constraint. The smoothing parameter in the model (i.e., the precision of the height difference between adjacent grid points) is set to 0.01.

[0145] Step 2: Establish an observation model Observation model Description based on actual appearance Generate high-angle incident mode topographic observation data Likelihood relation. Observation model. A Gaussian noise model is adopted, that is, it is assumed that the topographic observation data of the high-angle incident mode are... In true appearance Given that it follows a Gaussian distribution with a mean of 1,0 ... The interpolation at the corresponding high-angle sampling point location has a standard deviation equal to the measurement noise standard deviation of the high-angle incident mode topography data. . The value is estimated through the following steps: Under high-angle incidence mode, a uniform area of ​​10mm × 10mm in size and known to be defect-free is selected on the surface of the glass to be tested, and the output voltage signal of the second photodetector at all sampling points in this area is collected. Calculate the region The sample standard deviation, which is used as the standard deviation. The estimated value. If a uniform area cannot be predetermined on the surface of the glass to be tested, a standard glass sample (roughness) should be used before the formal scanning. Perform the same high-angle incidence mode, acquire and calculate its... Standard deviation as .

[0146] Step 3: Calculate the posterior distribution using the Markov chain Monte Carlo sampling algorithm. Posterior distribution The calculation is based on Bayes' theorem:

[0147]

[0148] in, This represents topographic observation data under a known high-angle incident mode. True morphology under the conditions The posterior probability distribution, Represents the likelihood function. Represents the prior distribution, This represents the normalization constant.

[0149] The Markov chain Monte Carlo sampling algorithm uses the Metropolis-Hastings sampler, with the following parameter settings: the total number of iterations for the Markov chain is set to 10,000, with the first 2,000 iterations discarded as a burn-in period, and only samples from the last 8,000 iterations used for statistical inference. Convergence is determined using the Gelman-Rubin latent scale reduction factor, calculated every 500 iterations. When this factor is less than 1.1, the chain is considered convergent, and the iteration can be terminated early. The proposed distribution of the morphology height parameter adopts a Gaussian distribution, with its standard deviation initially set to 0.05 μm and dynamically adjusted according to the acceptance rate to maintain the acceptance rate between 0.2 and 0.4. The Euclidean distance between the morphology means of two adjacent iterations is less than [value missing]. When convergence is reached, sampling is stopped. From the posterior distribution... A large number of morphology samples are generated through sampling, and the mean of all morphology samples is taken as the morphology data after Bayesian fusion. .

[0150] Step S503: The weighted and fused topography data Morphological data after fusion with Bayesian methods By combining the data, a unified three-dimensional topographic model of the glass surface under test is generated. The synthesis method is as follows: for the sub-region covered by the high-angle incident mode scan, a 3D topographic model is used. Directly using the topographic data after Bayesian fusion For sub-regions covered by the low-angle incident mode but not covered by the high-angle incident mode, the 3D topography model... Using weighted fusion of topographic data .

[0151] Unified three-dimensional topography model It is stored in the form of point cloud data. The point cloud data is stored using a three-dimensional coordinate array, with the coordinates of each sampling point... Corresponding to the horizontal position of the glass surface to be tested Height information ,in For sampling point index, The value ranges from 1 to , This refers to the total number of sampling points contained in the 3D topography model. Alternatively, it refers to a unified 3D topography model. It is stored in the form of grid data. The grid data adopts a regular grid format, dividing the glass surface under test into... OK The grid consists of rectangular sections, with each vertex storing its corresponding height value. The grid data is then interpolated to connect discrete sampling points into a continuous surface, creating a unified 3D topographic model. The data format is compatible with the smoothing process, error correction, and flatness parameter calculation in step S6.

[0152] By employing multi-index data quality assessment and weight allocation in the weighted fusion method, and prior modeling, likelihood modeling, and posterior inference in the Bayesian fusion method, large-scale topographic data from low-angle incident modes and high-precision topographic data from high-angle incident modes are fused to generate a unified three-dimensional topographic model of the glass surface under test. This provides complete basic data for calculating the flatness parameters in step S6.

[0153] Finally, step S6 is to calculate the flatness parameters, which includes the following sub-steps.

[0154] Step S601: Determine the ideal reference plane. The unified 3D topography model generated in step S5... The spatial coordinates of all sampling points are read from the database. The total number of sampling points is denoted as [missing information]. , No. The coordinates of each sampling point are: ,in For sampling point index, The value ranges from 1 to , and Representing the first The horizontal and vertical coordinates of each sampling point in the horizontal plane Representing the The height value of each sampling point.

[0155] The ideal reference plane is defined as the ideal plane that most closely approximates the actual morphology of the glass surface to be measured, and is expressed using plane equations. It means that among them , , The normal vector components of the ideal reference plane, Let be the distance from the ideal reference plane to the origin. The ideal reference plane is fitted using the least squares method, with the objective function being to minimize the sum of the squares of the perpendicular distances from all sampled points to the ideal reference plane. Plane parameters. , , , The following optimization problem was solved to obtain the following:

[0156]

[0157] in, , , The normal vector components of the ideal reference plane, The distance of the ideal reference plane from the origin of the coordinate system. , , For the first Spatial coordinates of each sampling point Let be the total number of sampling points. The above optimization problem is solved using singular value decomposition. The calculation process is as follows: First, construct... A coordinate matrix with 3 rows and 3 columns, the first... Behavior ,in , , All sampling points coordinate, coordinate, The arithmetic mean of the coordinates is calculated; then, singular value decomposition is performed on this coordinate matrix, and the right singular vector corresponding to the smallest singular value is taken as the normal vector. Finally, based on the normal vector and the mean of the sampling point coordinates Calculate the intercept .

[0158] Step S602: Calculate the deviation of each sampling point relative to the ideal reference plane. Deviation at each sampling point Representing the The perpendicular distance from each sampling point to the ideal reference plane is calculated using the following formula:

[0159]

[0160] in, Representing the The deviation of each sampling point , , The normal vector components of the ideal reference plane, The distance of the ideal reference plane from the origin of the coordinate system. , , For the first Spatial coordinates of each sampling point. Deviation. absolute value Indicates the first The degree to which each sampling point deviates from the ideal reference plane.

[0161] Step S603: Preprocess the deviation data. Preprocessing includes removing outliers and filling in missing data.

[0162] Outlier removal uses the 3σ criterion, and the specific steps are as follows: Calculate all... Deviation at each sampling point arithmetic mean and standard deviation ,in The mean of the deviation. The standard deviation represents the amount of deviation. This will satisfy... The sampled points are marked as outliers, and these outliers are removed from the sampled point set. The total number of sampled points after removing outliers is denoted as . .

[0163] Missing data is filled using nearest neighbor interpolation, specifically as follows: For sampling points that were replaced with equivalent scattered light intensity due to severe specular reflection interference identified in step S4 polarization state analysis, or sampling points that did not acquire valid data during the scanning process due to optical path obstruction, a search is conducted for these sampling points within the horizontal plane. Take the nearest neighbor valid sampling points. The arithmetic mean of the height values ​​of the nearest neighbor valid sampling points is used as the height fill value for this sampling point. The value ranges from 4 to 8.

[0164] Step S604: Calculate the global flatness parameters. Based on the preprocessed... Deviation of each valid sampling point Calculate the following global flatness parameters.

[0165] Arithmetic mean deviation , defined as the arithmetic mean of the absolute values ​​of the deviations of all valid sampling points, is calculated using the following formula:

[0166]

[0167] in, This represents the arithmetic mean deviation, in μm. This represents the total number of valid sampling points after removing outliers. Representing the The deviation of each sampling point.

[0168] Root mean square deviation , defined as the square root of the arithmetic mean of the squares of the deviations of all valid sampling points, is calculated using the following formula:

[0169]

[0170] in, This represents the root mean square deviation, in μm. This represents the total number of valid sampling points after removing outliers. Representing the The deviation of each sampling point.

[0171] Maximum peak-valley difference , defined as the difference between the maximum and minimum deviations among all valid sampling points, is calculated using the following formula:

[0172]

[0173] in, Represents the maximum peak-to-valley difference, in μm. This represents the maximum value of the deviation across all valid sampling points. This represents the minimum deviation of all valid sampling points.

[0174] Kurtosis coefficient , defined as the third central moment of the deviation distribution, is used to describe the symmetry and skewness of the surface morphology, and is calculated using the following formula:

[0175]

[0176] in, Represents the kurtosis coefficient. This represents the total number of valid sampling points after removing outliers. Representing the The deviation of each sampling point The mean of the deviation. The standard deviation represents the amount of deviation.

[0177] Uniformity coefficient , defined as the fourth central moment of the deviation distribution, is used to describe the sharpness of the surface morphology, and is calculated using the following formula:

[0178]

[0179] in, Represents the uniformity coefficient. This represents the total number of valid sampling points after removing outliers. Representing the The deviation of each sampling point The mean of the deviation. The standard deviation represents the amount of deviation.

[0180] Step S605: Calculate local flatness parameters. Divide the glass surface to be tested evenly into horizontal planes. There are three rectangular sub-regions, each with a preset size. For each rectangular sub-region, the deviation of all valid sampling points within that sub-region is extracted, and the arithmetic mean deviation of that sub-region is calculated according to the formula in step S604. Root mean square deviation and maximum peak-to-valley difference ,get Set local flatness parameters. The local flatness parameters are arranged according to the horizontal position of the rectangular sub-regions to generate a flatness distribution map.

[0181] Step S606: Smooth the local flatness parameters. A Gaussian filtering algorithm is used to eliminate high-frequency fluctuations in the spatial domain. The Gaussian filter convolution kernel size is 3×3 sub-regions, and the standard deviation is 1.0 times the width of a rectangular sub-region. For the local flatness parameters of each rectangular sub-region in the flatness distribution map, the weighted average of the local flatness parameters of this rectangular sub-region itself and its eight adjacent rectangular sub-regions is taken as the smoothed local flatness parameter. The weights are determined by the Gaussian convolution kernel. The smoothed local flatness parameters are used to identify local defects or abnormal areas in the surface morphology.

[0182] Step S6 fits the ideal reference plane using the least squares method and singular value decomposition, calculates the deviation of each sampling point relative to the ideal reference plane, and after preprocessing by removing outliers and filling missing data, calculates global flatness parameters such as arithmetic mean deviation, root mean square deviation, maximum peak-to-valley difference, kurtosis coefficient, and uniformity coefficient, and divides rectangular sub-regions to calculate and smooth local flatness parameters, outputting complete flatness evaluation results.

[0183] To facilitate understanding, the technical solution of the present invention will be described in detail below with reference to specific application examples. The glass sample to be tested is float glass with dimensions of 1000mm × 800mm, used for building curtain walls, and it is necessary to test whether the surface flatness meets the requirements for superior grade products. The testing equipment in this embodiment adopts the aforementioned laser scanning detection method for glass surface defects, and includes a laser light source, a multi-angle adjustable laser incident scanning device, a polarization beam splitting optical path system, a detector, and a data processing unit.

[0184] The detection process is divided into a calibration stage, a pre-scanning stage, a dual-mode scanning stage, a data fusion stage, and a flatness evaluation stage. In the calibration stage, a standard optical reference sample is used to calibrate the extinction ratio performance of the polarization beam splitter and the response characteristics of the polarization analysis element, and a calibration curve and a compensation parameter database are established. The calibration samples are standard glass plates with known surface roughness, and the roughness parameters Ra are 0.01μm, 0.05μm, 0.1μm, and 0.5μm respectively, covering the expected roughness range of the glass to be measured.

[0185] In the pre-scanning stage, the laser power is set to 5mW and the scanning speed is set to 800mm / s. A quick preview of the glass to be measured is performed to obtain approximate information on the surface reflectivity and uniformity distribution, which is used to adjust parameters such as the laser power and integration time for subsequent scans, avoiding signal saturation or insufficient signal-to-noise ratio.

[0186] In the dual-mode scanning stage, the low-angle incidence mode and the high-angle incidence mode are executed in sequence. In the low-angle incidence mode, the incident angle is set to 25°, the scanning resolution is 1mm×1mm, the scanning speed is 500mm / s, and the total scanning time is 1600 seconds. The low-angle scan covers the entire surface of the glass to be measured, obtaining overall topography profile data and initially identifying 3 local areas with obvious flatness deviations. In the high-angle incidence mode, the incident angle is set to 60°, the scanning resolution is increased to 0.2mm×0.2mm, the scanning speed is reduced to 100mm / s, and only the 3 initially identified abnormal areas are finely scanned, with a total scanning time of 1800 seconds.

[0187] In the data fusion stage, the Bayesian fusion method is used to fuse the low-angle and high-angle scanning data. During the fusion process, the low-angle scanning data is used as prior information, and the high-angle scanning data is used as observation information. The posterior distribution is calculated through the Markov chain Monte Carlo sampling algorithm to generate a unified three-dimensional topography model of the glass surface. The data resolution of the fusion model is 0.2mm in the high-angle scanning area and 1mm in the low-angle scanning area, completely covering the entire surface of the glass to be measured.

[0188] In the flatness evaluation stage, flatness parameters are calculated based on the fused three-dimensional topography model. The least squares method is used to fit the ideal reference plane, and the deviation amount of each sampling point is calculated, obtaining the arithmetic mean deviation Sa = 2.3μm, the root mean square deviation Sq = 3.1μm, and the maximum peak-to-valley difference Sz = 18.7μm. Referring to the flatness grading standard for architectural glass, when Sa≤3μm, it is judged as a first-class product; when 3μm < Sa≤5μm, it is judged as a first-class product; when Sa>5μm, it is judged as a qualified product. The Sa of this sample is 2.3μm, so it is judged as a first-class product. At the same time, local flatness analysis is performed on the 3 abnormal areas, and it is identified that 2 of the areas can reach the first-class product standard after local grinding treatment, and 1 area needs to be downgraded for use or reworked due to excessive local depression depth.

[0189] The technical solution of this invention achieves high-precision flatness detection of high-reflectivity glass surfaces through the coordinated operation of the above six steps. A multi-angle adjustable laser incident scanning device and a polarization beam splitting optical path system eliminate specular reflection interference; a dual-mode scanning strategy balances detection efficiency and accuracy; a polarization state analysis algorithm extracts scattering signals; a multi-angle data fusion algorithm generates a complete and high-precision three-dimensional topography model; and flatness parameter calculation provides an objective and accurate evaluation basis. This technical solution can be widely applied to surface quality inspection in fields such as architectural glass, optical glass, and automotive glass.

[0190] On the other hand, the present invention provides a laser scanning detection system for glass surface defects, comprising:

[0191] Laser source, used to emit a laser beam;

[0192] Collimating lens group, used to collimate the beam emitted from the laser source into a parallel beam;

[0193] The multi-angle adjustable laser incident scanning device includes a precision angle adjustment mechanism and a scanning motion mechanism. The precision angle adjustment mechanism is used to adjust the incident angle of the laser beam so that the incident angle reaching the glass surface to be tested is continuously adjustable within the range of 15° to 75°. The angle resolution of the precision angle adjustment mechanism is better than 0.1°. The scanning motion mechanism is used to drive the laser beam to perform two-dimensional scanning motion on the glass surface to be tested.

[0194] A polarization beam splitting optical path system includes a polarization beam splitter with an extinction ratio better than 1000:1 disposed in the incident light path, and a beam splitting detection optical path disposed in the detection end; the beam splitting detection optical path includes a beam splitter, a first photodetector, a polarization analysis element driven to rotate by a stepper motor, and a second photodetector, and the beam splitter is used to split the reflected light into a first beam and a second beam.

[0195] The data processing unit is used to perform polarization state orthogonal alignment, polarization state analysis and processing, multi-angle topography data fusion, and flatness parameter calculation.

[0196] The scanning control system's core controller internally operates a scanning trajectory planning module, a motor motion control module, a data acquisition and synchronization module, and a mode switching decision module, which are used to control the scanning motion mechanism, the laser incident angle, and the execution and switching of the dual-mode scanning strategy.

[0197] It will be apparent to those skilled in the art that the present invention is not limited to the details of the exemplary embodiments described above, and that the present invention can be implemented in other specific forms without departing from the spirit or essential characteristics of the present invention. Therefore, the embodiments should be regarded as exemplary and non-limiting in all respects.

[0198] Furthermore, it should be understood that although this specification describes embodiments, not every embodiment contains only one independent technical solution. This narrative style is merely for clarity. Those skilled in the art should consider the specification as a whole, and the technical solutions in each embodiment can also be appropriately combined to form other embodiments that can be understood by those skilled in the art.

Claims

1. A laser scanning detection method for glass surface defects, characterized in that, include: A multi-angle laser scanning optical path is established, and the laser beam collimated by the collimating lens group is incident on the multi-angle adjustable laser incident scanning device. The incident angle of the laser beam is adjusted by the precision angle adjustment mechanism in the multi-angle adjustable laser incident scanning device, so that the incident angle of the incident laser reaching the glass surface to be tested is continuously adjustable within the preset working range of 15° to 75°. The laser beam is driven to perform two-dimensional scanning motion on the glass surface to be tested by the scanning motion mechanism. A polarization beam splitter system is configured, and a polarization beam splitter with an extinction ratio better than 1000:1 is set in the incident light path to convert the incident laser into linearly polarized light with a single polarization state. A beam splitter detection light path is constructed at the detection end, and the reflected light is split into two paths by a beam splitter. One path is used to monitor the total light intensity, and the other path passes through a rotatable polarization analysis element to measure the light intensity. By performing a polarization state orthogonal alignment process, the orthogonal extinction position is locked, so that the transmission axis direction of the polarization analysis element is orthogonal to the polarization direction of the specular reflection component of the glass surface. Dual-mode scanning detection is performed by adjusting the laser incident angle to the range of 15° to 30° and executing the low-angle incident mode to quickly establish the overall morphological profile of the glass surface under test; the laser incident angle is adjusted to the range of 50° to 75° and the high-angle incident mode is executed to perform fine detection on specific areas and obtain high-precision micro-morphological data.

2. The laser scanning detection method for glass surface defects according to claim 1, characterized in that: The precision angle adjustment mechanism includes an angle adjustment base, a rotary support shaft, a precision lead screw drive pair driven by a stepper motor or a servo motor, and an angle feedback sensor. The precision lead screw transmission pair converts the rotational motion of the stepper motor or servo motor into the linear motion of the lead screw nut, and converts the linear displacement of the lead screw nut into the angular displacement of the rotating support shaft through the cooperation of the lever and the eccentric pin, so as to change the propagation direction of the laser beam. The angle feedback sensor uses a circular grating encoder or a magnetic grating encoder to detect the actual angular position of the rotating support shaft in real time and output a position signal. The position signal is fed back to the driver of the stepper motor or servo motor to form a closed-loop control, with an angle resolution better than 0.1°.

3. The laser scanning detection method for glass surface defects according to claim 1, characterized in that, The configured polarization beam splitting optical path system specifically includes: The polarizing beam splitter is set in the optical path between the laser source and the glass surface to be tested. The polarizing beam splitter adopts a multi-layer thin film coating structure and is mounted on a mounting bracket equipped with an angle fine adjustment mechanism. The pitch angle and rotation angle of the angle fine adjustment mechanism are both ±5°, and the adjustment resolution is 0.001°, which is used to precisely adjust the polarizing beam splitter to meet the Brewster angle condition. At the detection end, a focusing lens group is set to converge the reflected light. Behind the focusing lens group, the beam splitter is set to split the converged reflected light into a first beam and a second beam. The first beam is directly incident on the first photodetector to monitor the total light intensity of the reflected light in real time. A polarization analysis element is set on the optical path of the second beam. The polarization analysis element is a polarizer with an extinction ratio parameter better than 1000:1 and is mounted on a precision rotating stage driven by a stepper motor through a worm gear transmission mechanism. The beam passes through the polarization analysis element and is then incident on the second photodetector.

4. The laser scanning detection method for glass surface defects according to claim 3, characterized in that, The polarization state orthogonal alignment process is specifically as follows: The data processing unit controls the stepper motor to drive the precision rotary table to rotate the polarization analysis element continuously within the range of 0° to 180° in a preset angle step of 0.01°, and records the output voltage signal of the first photodetector and the output voltage signal of the second photodetector at each rotation angle in real time, and calculates the real-time light intensity ratio between the second photodetector and the first photodetector. The data processing unit searches for the orthogonal extinction position where the light intensity ratio reaches its minimum value from the data sequence of all recorded rotation angles and light intensity ratios, and controls the stepper motor to lock the precision rotary table at the searched orthogonal extinction position.

5. The laser scanning detection method for glass surface defects according to claim 1, characterized in that, In the dual-mode scanning detection, the mode switching adopts an adaptive triggering switching method: During the low-angle incident mode scanning process, the data processing unit performs surface roughness estimation, defect feature identification, and flatness deviation threshold determination on the real-time acquired morphology data. When the analysis results show that the evaluation result of a certain sub-region exceeds the preset threshold, the sub-region is automatically marked as the region of interest and a scanning trigger command is generated. After receiving the scan trigger command, the scanning control system smoothly switches the laser incident angle at the boundary of the sub-region to the angle range corresponding to the high-angle incident mode to perform a secondary fine scan. After the scan is completed, it switches back to the low-angle incident mode to resume full-area scanning.

6. The laser scanning detection method for glass surface defects according to claim 1, characterized in that, After acquiring the topographic data for both low-angle and high-angle incident modes, polarization state analysis was performed. The polarization state of the reflected light is described by the Stokes vector method. The polarization analysis element is rotated sequentially to 0°, 90°, 45° and 135° while the voltage signal output by the second photodetector is recorded simultaneously. The components of the Stokes vector are calculated and the degree of polarization is obtained. The Mueller matrix method is used to obtain four sets of linearly independent input Stokes vectors by changing the polarization state of the incident light at least four times and measuring the corresponding output Stokes vectors, and then solving for all elements of the Mueller matrix. A mapping relationship between polarization degree and surface roughness is established. Using a polarization degree threshold of 0.3 to 0.5 as a criterion, the light intensity data of sampling points with a polarization degree less than the preset polarization degree threshold are marked as valid scattered signals. The light intensity data of sampling points with a polarization degree greater than or equal to the preset polarization degree threshold are calculated and replaced with equivalent scattered light intensity through the mapping relationship.

7. The laser scanning detection method for glass surface defects according to claim 6, characterized in that, After polarization state analysis, multi-angle topography data are fused to generate a unified three-dimensional topography model: A weighted fusion method is adopted to evaluate the data quality of the topography data acquired by the low-angle incident mode and the high-angle incident mode based on the signal-to-noise ratio, sampling density and effective data coverage, and calculate the comprehensive data quality evaluation score. The fusion weight is calculated based on the comprehensive data quality evaluation score, and the two types of topography data are weighted and summed to obtain the weighted fused topography data. A Bayesian fusion method is adopted, using low-angle incident pattern topography data as prior information and a Gaussian Markov random field model as the prior distribution, and high-angle incident pattern topography data as observation information and a Gaussian noise observation model is established. The Markov chain Monte Carlo sampling algorithm is used to calculate the posterior distribution, and the mean of all topography samples is taken as the topography data after Bayesian fusion.

8. The laser scanning detection method for glass surface defects according to claim 7, characterized in that, Calculate flatness parameters based on a unified 3D topography model: The deviation of each sampling point relative to the ideal reference plane is calculated by fitting an ideal reference plane using the least squares method and singular value decomposition. The deviation data were preprocessed by removing outliers using the 3σ criterion and filling in missing data using the nearest neighbor interpolation method. Global smoothness parameters are calculated based on the preprocessed deviation, including arithmetic mean deviation, root mean square deviation, maximum peak-to-valley difference, kurtosis coefficient, and uniformity coefficient. The surface of the glass to be tested is uniformly divided into multiple rectangular sub-regions in the horizontal plane. The local flatness parameters of each rectangular sub-region are calculated, and the Gaussian filtering algorithm is used to smooth the local flatness parameters.

9. The laser scanning detection method for glass surface defects according to claim 1, characterized in that, The scanning motion mechanism adopts a gantry structure or a cantilever structure, and is composed of an X-axis linear module and a Y-axis linear module with a movement accuracy better than 1μm and a repeatability positioning accuracy better than 0.5μm. The scanning path of the scanning motion mechanism can be selected from raster scanning mode, spiral scanning mode or adaptive path planning mode.

10. A laser scanning detection system for glass surface defects, characterized in that, For implementing the method as described in any one of claims 1 to 9, comprising: Laser source, used to emit a laser beam; A collimating lens group is used to collimate the light beam emitted from the laser source into a parallel light beam; A multi-angle adjustable laser incident scanning device includes a precision angle adjustment mechanism and a scanning motion mechanism. The precision angle adjustment mechanism is used to adjust the incident angle of the laser beam so that the incident angle reaching the glass surface to be tested is continuously adjustable within the range of 15° to 75°. The angle resolution of the precision angle adjustment mechanism is better than 0.1°. The scanning motion mechanism is used to drive the laser beam to perform two-dimensional scanning motion on the glass surface to be tested. A polarization beam splitting optical path system includes a polarization beam splitter with an extinction ratio better than 1000:1 disposed in the incident light path, and a beam splitting detection optical path disposed in the detection end; the beam splitting detection optical path includes a beam splitter, a first photodetector, a polarization analysis element driven to rotate by a stepper motor, and a second photodetector, wherein the beam splitter is used to split the reflected light into a first beam and a second beam. The data processing unit is used to perform polarization state orthogonal alignment, polarization state analysis and processing, multi-angle topography data fusion, and flatness parameter calculation. The scanning control system has a core controller that internally runs a scanning trajectory planning module, a motor motion control module, a data acquisition synchronization module, and a mode switching decision module, which are used to control the scanning motion mechanism, the laser incident angle, and the execution and switching of the dual-mode scanning strategy.

Images