A method for detecting stress distribution in tempered glass

Abstract

This invention relates to the field of tempered glass stress testing technology, and discloses a method and system for detecting stress distribution in tempered glass. The method includes: generating an original interferogram containing fringe contrast distribution and depth-encoded color levels through multi-angle layered scanning using polarized light modulation; extracting local phase jump points to mark candidate stress concentration areas to obtain a stress feature map; using Gaussian difference filtering to remove scattered light noise and retaining effective interference signals based on the light intensity attenuation curve to obtain stress interference data; constructing a three-dimensional phase matrix by eliminating phase truncation through a phase unpacking algorithm; extracting slope abrupt changes in the stress gradient curve to determine internal microcrack defects; and calculating the ratio of the stress concentration coefficient in the defect area to that in the surrounding area to determine the anomaly in stress distribution. This method achieves accurate detection of internal stress distribution and defect depth location in tempered glass, improving the accuracy and reliability of quality control.

Description

Technical Field

[0001] This invention relates to the field of tempered glass stress testing technology, and in particular to a method and system for detecting stress distribution in tempered glass. Background Technology

[0002] Currently, tempered glass develops a complex stress distribution within its interior during the manufacturing process. This stress state directly determines the glass's strength and safety performance. When the stress distribution is uneven or there is localized stress concentration, the glass may spontaneously shatter during use, posing a serious safety hazard.

[0003] Traditional stress detection methods rely primarily on surface observation, making it difficult to accurately obtain the stress distribution at different depths within the glass. Due to the thickness of tempered glass, there is often a significant difference between surface and deep stress, and internal defects such as microcracks are typically hidden deep within the glass, making these potential hazards undetectable by surface inspection alone. When using optical interferometry for deep scanning, reflected light signals from different transmission depths superimpose, forming complex interference fringe patterns containing scattered light noise and background light interference. These noise components severely affect the accuracy of extracting the true stress signal. While some solutions incorporate multi-angle scanning and intelligent sensors, they lack effective mechanisms for separating scattered noise from the reflected light signals, resulting in the deep stress signal being overwhelmed by noise.

[0004] In summary, existing technologies suffer from the technical problem that the internal stress distribution of tempered glass cannot be accurately detected due to noise interference from scattered light and periodic phase truncation. Summary of the Invention

[0005] This invention provides a method and system for detecting stress distribution in tempered glass, thereby solving the technical problem that the internal stress distribution of tempered glass cannot be accurately detected due to scattered light noise interference and phase periodic truncation in the prior art.

[0006] In a first aspect, to solve the above-mentioned technical problems, the present invention provides a method for detecting the stress distribution of tempered glass, comprising:

[0007] The phase difference signal of the reflected light from the tempered glass is acquired, and the phase difference signal of the reflected light is preprocessed to obtain the original interferogram.

[0008] Local phase anomaly features are extracted from the original interferogram. If the local phase anomaly features meet the preset stress concentration judgment conditions, a stress feature map is obtained.

[0009] The stress feature map is subjected to noise suppression processing to obtain stress interferometry data;

[0010] The stress interference data is subjected to phase unpacking processing to obtain continuous phase data, and a three-dimensional phase matrix is ​​constructed based on the continuous phase data;

[0011] Extract the stress gradient variation curve of the three-dimensional phase matrix along the thickness direction, identify the local slope abrupt change feature in the stress gradient variation curve, and if the local slope abrupt change feature meets the preset microcrack judgment condition, then determine the internal microcrack defect.

[0012] The spatial distribution characteristics of the internal microcrack defects are extracted to obtain the spatial range of the defects;

[0013] Based on the spatial range of the defect, stress concentration difference analysis is performed. If the stress concentration difference meets the preset anomaly judgment conditions, the stress distribution anomaly result of the tempered glass is obtained.

[0014] In a second aspect, the present invention provides a system for detecting the stress distribution of tempered glass, comprising: a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the program to implement the steps of the method described above.

[0015] Thirdly, the present invention provides a computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the steps of the method described above.

[0016] Compared with the prior art, the present invention has the following beneficial effects:

[0017] (1) This invention combines polarization light modulation with multi-angle layered penetration scanning to perform optical path difference conversion and depth encoding on the phase difference signal of reflected light generated at different transmission depths, generating an original interference spectrum containing stripe contrast distribution and depth-encoded color levels. This breaks through the limitation of existing technologies that cannot perform layered analysis when incident from a single angle, and realizes complete stress signal acquisition from the surface to the deep layer.

[0018] (2) This invention extracts local phase jump features from the interferogram and marks the candidate region of stress concentration based on the comparison between the gray gradient field and the background light intensity baseline. It uses Gaussian difference filtering to separate high-frequency scattered light noise and retains the effective interference signal according to the light intensity attenuation curve. Then, it uses a quality-oriented three-dimensional region growth phase unpacking algorithm to eliminate the phase periodic truncation and construct a three-dimensional phase distribution matrix that reflects the difference between surface and deep stress. This solves the problems of scattered light noise interference and phase continuity loss, and realizes the accurate restoration of pure layered stress data.

[0019] (3) This invention extracts the stress gradient change curve along the thickness direction, identifies the local slope change characteristics to determine the internal microcrack defects, calculates the ratio of the stress concentration coefficient of the defect area to the surrounding normal area to perform stress concentration difference analysis, and realizes the accurate positioning of the depth coordinates and lateral expansion range of the internal microcrack defects, providing a reliable basis for judging the abnormal internal stress distribution for the quality and safety assessment of tempered glass. Attached Figure Description

[0020] Figure 1 This is a schematic flowchart of a method for detecting stress distribution in tempered glass according to the first embodiment of the present invention; Detailed Implementation

[0021] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0022] Reference Figure 1 The first embodiment of the present invention provides a method for detecting the stress distribution of tempered glass, comprising the following steps:

[0023] S1, Acquire the phase difference signal of the reflected light from the tempered glass, preprocess the phase difference signal of the reflected light to obtain the original interferogram;

[0024] S2, extract the local phase anomaly features from the original interferogram; if the local phase anomaly features meet the preset stress concentration judgment conditions, then a stress feature map is obtained.

[0025] S3, perform noise suppression processing on the stress feature map to obtain stress interference data; perform phase unpacking processing on the stress interference data to obtain continuous phase data, and construct a three-dimensional phase matrix based on the continuous phase data;

[0026] S4, extract the stress gradient change curve of the three-dimensional phase matrix along the thickness direction, identify the local slope change feature in the stress gradient change curve, and if the local slope change feature meets the preset microcrack judgment condition, then determine the internal microcrack defect.

[0027] S5, extract the spatial distribution characteristics of the internal microcrack defects to obtain the defect spatial range; perform stress concentration difference analysis based on the defect spatial range, and if the stress concentration difference meets the preset anomaly judgment conditions, then obtain the stress distribution anomaly result of the tempered glass.

[0028] In step S1, the phase difference signal of the reflected light from the tempered glass is acquired, and the phase difference signal is preprocessed to obtain the original interferogram, including:

[0029] S11 uses polarization modulation to control the vibration direction of the incident beam and performs multi-angle layered penetration scanning on the tempered glass to obtain layered reflected light signals.

[0030] S12, extract the phase difference signal of the reflected light corresponding to different transmission depths in the layered reflected light signal;

[0031] S13, perform optical path difference conversion and depth encoding on the reflected light phase difference signal to obtain layered interference data;

[0032] S14, Generate an original interferogram containing fringe contrast distribution and depth coding information based on the layered interferometric data.

[0033] In step S11, the intelligent sensor control module drives the digitally controlled liquid crystal phase retarder to polarize the incident laser beam with a wavelength of 632nm. A dynamic voltage of 500mV to 2500mV is applied to the digitally controlled liquid crystal phase retarder, causing the polarization direction of the incident beam to change continuously and linearly between 0° and 90° with a modulation period of 10ms. The modulated polarized light is then input into a high-precision servo galvanometer system, which controls the incident angle to change in 0.5° increments between -45° and +45°. The refraction path of the incident beam inside the glass is determined based on the refractive index parameter of the tempered glass, and layered penetration scanning is performed along the thickness direction of the tempered glass at different incident angles. For tempered glass with a thickness of 6 mm, virtual detection layers are divided according to the layer spacing of 50 μm to 200 μm, and 60 virtual detection layers are divided according to the layer spacing of 100 μm, so that the modulated polarized light penetrates the layers of different depths in sequence and generates reflection at the interlayer interface. The reflected light intensity signal returned by each virtual detection layer is received synchronously through the photoelectric detection array to obtain the layered reflected light signal.

[0034] It should be noted that the incident beam wavelength is 632nm to match the optical resolution requirements for detecting the birefringence phase difference inside the tempered glass; the dynamic voltage is set from 500mV to 2500mV, which is determined based on the voltage-phase delay calibration curve of the digitally controlled liquid crystal phase delay unit used in this embodiment. Before detection, nine calibration voltage points of 500mV, 750mV, 1000mV, 1250mV, 1500mV, 1750mV, 2000mV, 2250mV, and 2500mV are applied to the liquid crystal phase delay unit, and the phase delay and polarization direction change corresponding to each calibration voltage point are recorded. The calibration voltage points and phase delay are linearly fitted. When the linear fitting determination coefficient R² is not less than 0.98 and the polarization direction can cover the effective modulation range from 0° to 90°, the dynamic voltage modulation range of 500mV to 2500mV is determined. If the R² of the calibration curve is less than 0.98, the liquid crystal phase delay unit preheating and calibration process should be repeated, or the voltage points with abnormal responses should be removed and refitted. Voltage-phase delay calibration should be performed when the equipment is first started, when the liquid crystal phase delay unit is replaced, when the light source wavelength changes, or when the ambient temperature changes by more than 5°C. During continuous batch testing, a rapid verification calibration should be performed every 50 pieces of tempered glass tested or after 4 hours of continuous operation.

[0035] If a different model of liquid crystal phase delay unit is used, the corresponding effective modulation voltage range is re-determined through the same voltage-phase delay calibration process.

[0036] In step S12, the layered reflected light signal obtained in step S11 is input to the photoelectric signal acquisition unit. The photoelectric signal acquisition unit synchronously acquires the layered reflected light signal at a sampling rate of 10000Hz and groups the acquired reflected light intensity sequence according to the incident angle, polarization direction, and virtual detection layer number. A fast Fourier transform is performed on each group of reflected light intensity sequence to convert the time-domain reflected light intensity sequence into a frequency-domain interference light intensity spectrum. High-frequency noise components and background light DC components are removed from the frequency-domain interference light intensity spectrum, and effective frequency components related to the coherent superposition of ordinary and extraordinary light are retained. The phase delay value between ordinary and extraordinary light is extracted from the effective frequency components and arranged according to the corresponding virtual detection layer number to obtain the reflected light phase difference signal corresponding to different transmission depths.

[0037] In step S13, the reflected light phase difference signal corresponding to different virtual detection layers is extracted, and the reflected light phase difference signal is converted into optical path difference according to the incident beam wavelength. The calculation method is to multiply the reflected light phase difference by the incident beam wavelength, divide by two and multiply by pi to obtain the cumulative optical path difference. The cumulative optical path difference of each detection layer is corrected for phase continuity, and the cumulative optical path difference is limited to the effective mapping range of 0nm to 500nm. The cumulative optical path difference is mapped to the stripe contrast gray value, and the depth coordinates of the corresponding detection layer are mapped to the depth-coded color level. The stripe contrast gray value, depth-coded color level, detection layer number and spatial pixel coordinates are structurally integrated to obtain layered interference data.

[0038] It should be noted that the fringe contrast grayscale value mapping range is 0 to 255, and the depth encoding range is 0 mm to 6 mm. By encoding the cumulative optical path difference and the depth coordinates simultaneously, the layered interferometric data can simultaneously characterize the phase change caused by stress and the depth position of the phase change.

[0039] In step S14, the layered interference data obtained in step S13 is read, and the fringe contrast grayscale values ​​of each virtual detection layer are written into a two-dimensional image matrix according to the spatial pixel coordinates; the depth-coded color levels corresponding to different detection layers are superimposed on the two-dimensional image matrix so that the same pixel position carries both fringe contrast information and depth-coded information; the layered interference data generated under each incident angle are spatially registered, and the fringe responses under the same spatial position but different incident paths are fused; the calculation method is to average the fringe contrast grayscale values ​​corresponding to multiple incident angles under the same spatial position to obtain the fused fringe contrast at that spatial position; the original interference pattern is generated by rendering based on the fused fringe contrast, depth-coded color levels and spatial pixel coordinates. The original interference pattern contains fringe contrast distribution, depth-coded information, detection layer number and spatial position index.

[0040] It should be noted that the resolution of the original interferogram is determined based on the accuracy of identifying microcrack defects inside the tempered glass. The resolution of the spectrum is 4096 by 4096 pixels to ensure that small phase transition regions can be identified at the pixel level. The fringe contrast distribution in the spectrum is used to characterize the changes in interference intensity caused by the phase difference of the reflected light, and the depth encoding information is used to characterize the specific transmission depth layer of the tempered glass from which the corresponding interference fringes originate.

[0041] In step S2, local phase anomaly features are extracted from the original interferogram. If the local phase anomaly features meet the preset stress concentration judgment conditions, a stress feature map is obtained, including:

[0042] S21, extract the fringe spacing variation data and phase jump data from the original interferogram to obtain local phase anomaly data;

[0043] S22, Perform spatial coordinate mapping and grayscale gradient transformation on the local phase anomaly data to obtain local phase anomaly features;

[0044] S23, compare the local phase anomaly features with the preset stress concentration judgment conditions. If the preset stress concentration judgment conditions are met, mark the stress concentration candidate region.

[0045] S24, perform pixel-level mapping processing based on the stress concentration candidate region to obtain a stress feature map.

[0046] In step S21, the original interferogram obtained in step S1 is read. The original interferogram includes fringe contrast distribution, depth coding information, detection layer number, and spatial pixel coordinates. The fringe center lines are extracted from the original interferogram in both row and column directions to obtain the pixel position sequence of adjacent interference fringes. The fringe spacing between adjacent fringes is extracted point by point based on the pixel position sequence, and the fringe spacing is compared with the standard fringe spacing under normal stress-free conditions. The calculation method is to subtract the current fringe spacing from the standard fringe spacing and then divide by the standard fringe spacing to obtain the fringe spacing change rate. At the same time, the Laplace second derivative detection is performed on the original interferogram to extract the local phase jump positions corresponding to the zero-crossing points of the second derivative, thus obtaining phase jump data. The fringe spacing change rate, phase jump positions, depth coding information, and spatial pixel coordinates are structurally integrated to obtain local phase anomaly data.

[0047] It should be noted that the standard fringe spacing is determined based on the interference fringe distribution of untempered annealed glass samples of the same batch and material, and is set to 50 pixels. The untempered annealed glass sample is a sheet of glass with the same thickness, material formula, and surface finish as the tempered glass to be tested before tempering. Since no permanent residual compressive stress from the tempering process has formed inside, it can be used as a reference sample for the baseline interference fringes. The fringe spacing change rate threshold is determined based on the backtesting results of the interference spectrum of defect-free tempered glass samples and known stress concentration samples, and is set to 15%. Specifically, defect-free samples and known stress concentration samples of tempered glass of the same thickness and process are collected, and the fringe spacing change rate distribution in normal areas and stress concentration areas is statistically analyzed. The cutoff point with the lowest combined false positive and false negative rates is used as the fringe spacing change rate threshold. When the fringe spacing change rate is below 12%, some real stress concentration areas are easily missed; when the fringe spacing change rate is above 18%, slight stress concentration areas are easily excluded. Therefore, in this embodiment, 15% is taken as the preferred threshold.

[0048] When the local fringe spacing shrinks to below 42 pixels, the fringe spacing change rate exceeds 15%, indicating that there is a phase anomaly in this area caused by internal stress disturbance. The zero-crossing point of the second derivative of Laplace is used to locate the phase jump position in the fringe grayscale curve caused by abrupt changes in brightness, so that the local phase anomaly data simultaneously contains fringe compression information and phase jump information.

[0049] In step S22, the phase jump position, fringe spacing change rate, depth coding information, and spatial pixel coordinates are extracted from the local phase anomaly data, and the phase jump position is mapped to the two-dimensional pixel coordinate system of the original interferogram. An 11x11 pixel analysis window is constructed with each phase jump position as the center, and the gray value distribution and fringe contrast distribution are extracted within the analysis window. The Sobel operator is used to calculate the horizontal and vertical gray value changes within the analysis window, and the horizontal and vertical gray value changes are converted into gray value gradient fields. The calculation method is to first sum the squares of the horizontal and vertical gray value changes, and then take the arithmetic square root of the summation result to obtain the gray value gradient amplitude. The fringe spacing change rate, phase jump coordinates, gray value gradient amplitude, depth coding information, and analysis window range are combined to obtain the local phase anomaly features.

[0050] It should be noted that the grayscale gradient field is used to characterize the degree of abrupt change in light intensity in local phase transition regions. The larger the grayscale gradient amplitude, the more drastic the change in brightness of the interference fringes at that location, and the higher the probability of internal stress concentration. Through spatial coordinate mapping and grayscale gradient transformation, local phase anomaly data can be converted into quantitative features that can be used to determine stress concentration.

[0051] In step S23, the fringe spacing change rate, gray-level gradient magnitude, phase jump coordinates, and depth coding information are extracted from the local phase anomaly features, and the background light intensity baseline of the original interferogram is read. The fringe spacing change rate is compared with the fringe spacing change rate threshold, and the gray-level gradient magnitude is compared with the background light intensity baseline multiple threshold. It is determined whether the phase jump coordinate is located within a continuous abnormal fringe region. The calculation method is to divide the gray-level gradient magnitude by the background light intensity baseline to obtain the gray-level gradient multiple. If the fringe spacing change rate is greater than 15%, the gray-level gradient multiple is greater than 2, and the phase jump coordinate is located within a continuous abnormal fringe region, then the local phase anomaly features are determined to meet the preset stress concentration judgment conditions. The pixel area covered by the corresponding analysis window is marked as a stress concentration candidate area, centered on the phase jump coordinate that meets the judgment conditions.

[0052] It should be noted that the preset stress concentration judgment conditions are determined based on the fringe compression and intensity abrupt change characteristics of the stress concentration region inside the tempered glass in the interferogram. Specifically, these include a fringe spacing change rate greater than 15%, a gray-level gradient amplitude greater than twice the background light intensity baseline, and a phase jump point located in a continuous abnormal fringe region. The background light intensity baseline is obtained by statistically analyzing the average gray-level value of unmarked abnormal regions in the original interferogram. The gray-level gradient multiple threshold of 2 is determined by backtesting the gray-level gradient distribution of defect-free samples and known stress concentration samples. The gray-level gradient multiples of normal regions, scattered light interference regions, and known stress concentration regions are statistically analyzed respectively. When the threshold is lower than 1.8 times, some ordinary background light fluctuations are easily misjudged as stress concentration candidate regions. When the threshold is higher than 2.2 times, some weak stress concentration regions are easily missed. Therefore, in this embodiment, 2 times is taken as the preferred threshold. The fringe spacing change rate and gray-level gradient multiple are used together in the judgment to avoid misjudgment caused by relying solely on a single fringe compression or a single intensity abrupt change.

[0053] In step S24, a blank pixel matrix with the same size as the original interferogram is established, and the resolution of the blank pixel matrix is ​​consistent with that of the original interferogram. The stress concentration candidate regions marked in step S23 are read, and the corresponding pixels are written into the blank pixel matrix according to the spatial coordinate range of the stress concentration candidate regions. Pixels belonging to the stress concentration candidate regions are assigned a highlight mark value, and pixels not belonging to the stress concentration candidate regions are assigned a background mark value. The calculation method is to set the pixel value of the candidate region to 255 and the pixel value of the non-candidate region to 0 to obtain a binary stress mapping matrix. The binary stress mapping matrix is ​​bound to the depth coding information, and the fringe spacing change rate, gray-level gradient magnitude and phase jump coordinates corresponding to each stress concentration candidate region are retained to obtain a stress feature map containing pixel-level stress mapping.

[0054] It should be noted that the stress feature map is used to convert local phase anomaly regions in the original interferogram into pixel-level stress mapping results that can be used for subsequent noise suppression processing. At the same time, the depth encoding information is preserved to ensure that subsequent steps can distinguish whether the stress concentration candidate region originates from the surface, transition layer, or deep layer of the tempered glass.

[0055] In step S3, noise suppression processing is performed on the stress feature map to obtain stress interferometric data; phase unpacking processing is performed on the stress interferometric data to obtain continuous phase data; and a three-dimensional phase matrix is ​​constructed based on the continuous phase data, including:

[0056] S31, extract the blurred edge region in the stress feature map to obtain the stress region to be denoised;

[0057] S32, perform scattered light noise separation on the interference fringe data in the stress region to be denoised to obtain candidate interference signals;

[0058] S33, match the candidate interference signal with the preset light intensity attenuation rule, and retain the valid interference signal that satisfies the preset light intensity attenuation rule;

[0059] S34, perform layered mapping processing based on the effective interference signal to obtain stress interference data;

[0060] S35, extract the truncated phase value from the stress interference data, identify the phase jump region between adjacent spatial positions, and obtain the phase truncation boundary;

[0061] S36, perform phase compensation processing on the phase truncation boundary to obtain continuous phase data;

[0062] S37, extract the surface phase information and deep phase information from the continuous phase data to obtain the interlayer phase difference characteristics;

[0063] S38. Based on the continuous phase data and the interlayer phase difference characteristics, spatial mapping is performed to construct a three-dimensional phase matrix.

[0064] In step S31, the stress feature map obtained in step S2 is read. The stress feature map includes candidate area pixels, non-candidate area pixels, depth encoding information, stripe spacing change rate, gray-level gradient magnitude and phase jump coordinates. Mathematical morphological boundary extraction processing is performed on the candidate area pixels in the stress feature map. First, dilation operation is performed on the candidate area, then erosion operation is performed on the candidate area, and the difference area between the dilation result and the erosion result is extracted as the candidate boundary area.

[0065] Within the candidate boundary region, the grayscale transition width and stripe contrast variation range are read point by point. The region with a grayscale transition width between 3 and 8 pixels is defined as the edge blur region. The grayscale transition width is obtained by subtracting the erosion candidate region boundary position from the dilated candidate region boundary position. The spatial pixel coordinates, depth coding information, stripe contrast data and phase jump coordinates corresponding to the edge blur region are integrated to obtain the stress region to be denoised.

[0066] When the grayscale transition width is less than 3 pixels, this area is usually a clear fringe boundary and is not the primary target for noise reduction. When the grayscale transition width is greater than 8 pixels, this area usually contains a large range of background light variations and needs to be further filtered in subsequent steps using light intensity attenuation rules. By first locking the blurred edge areas, over-filtering of complete and effective interference fringes can be avoided.

[0067] In step S32, the original interference fringe density data, fringe contrast data, grayscale transition width, and depth encoding information are extracted from the stress region to be denoised. A Gaussian difference filtering algorithm is used to perform dual-scale filtering on the original interference fringe density data. First, a small-scale Gaussian kernel with a standard deviation of 1.0 to 1.5 is used to convolve the original interference fringe density data, with the standard deviation of the small-scale Gaussian kernel set to 1.2, to obtain the detail response map. Then, a large-scale Gaussian kernel with a standard deviation of 3.0 to 4.0 is used to convolve the original interference fringe density data, with the standard deviation of the large-scale Gaussian kernel set to 3.5, to obtain the background response map. The calculation method is to subtract the background response map from the detail response map to obtain the Gaussian difference response map. High-frequency scattered light noise components are extracted from the Gaussian difference response map, and these components are removed from the total signal matrix of the stress region to be denoised, resulting in candidate interference signals.

[0068] It should be noted that the standard deviation of the small-scale Gaussian kernel (1.2) and the standard deviation of the large-scale Gaussian kernel (3.5) were determined based on the filtering comparison results of the scattered light noise sample and the clear interference fringe sample. Combined filtering tests were conducted using 1.0, 1.2, and 1.5 as candidate standard deviations for the small scale, and 3.0, 3.5, and 4.0 as candidate standard deviations for the large scale. The evaluation indicators were fringe edge preservation, background light suppression rate, and the improvement in interference fringe signal-to-noise ratio. When the small-scale standard deviation is 1.2 and the large-scale standard deviation is 3.5, it can suppress low-frequency scattered background while preserving fringe edge details, and the improvement in interference fringe signal-to-noise ratio is superior. Therefore, this is the preferred value in this embodiment. The noise spatial frequency threshold is determined based on the high-frequency distribution characteristics of scattered light in the interference fringe density, and is set to 50 line pairs / mm. When the corresponding spatial frequency in the Gaussian difference response is higher than 50 line pairs / mm, this part of the signal is judged as a scattered light noise component and removed from the total signal.

[0069] In step S33, the candidate interference signal obtained in step S32 is read, and the light intensity distribution sequence of the candidate interference signal at different depth encoding positions is extracted. The light intensity distribution sequence is matched with a preset light intensity attenuation rule. The preset light intensity attenuation rule is a light intensity change relationship that decreases exponentially with increasing transmission depth, expressed as I(z)=I0·e^(-αz), where I(z) represents the reflected light intensity at transmission depth z, I0 represents the initial reflected light intensity, α represents the light intensity attenuation coefficient, and z represents the transmission depth. The light intensity distribution sequence of the candidate interference signal is fitted to the ideal exponential light intensity attenuation curve using a nonlinear least squares fitting method, and the Pearson correlation coefficient between the two is calculated. The calculation method is to divide the covariance of the corresponding points of the candidate interference signal light intensity sequence and the ideal exponential light intensity attenuation curve by the product of their standard deviations to obtain the Pearson correlation coefficient. If the Pearson correlation coefficient is greater than 0.85, and the fitted light intensity attenuation coefficient is within the range of the reference attenuation coefficient, then the candidate interference signal is determined to satisfy the preset light intensity attenuation rule and is retained as a valid interference signal.

[0070] It should be noted that the preset light intensity attenuation rule is determined based on the optical propagation characteristics of tempered glass, where reflected light attenuates with increasing transmission depth. Under incident light conditions at a wavelength of 632 nm, the reference light intensity attenuation coefficient of the tempered glass sample is obtained through calibration using defect-free tempered glass samples from the same batch, with a value range of 0.08 mm⁻¹ to 0.35 mm⁻¹, preferably 0.18 mm⁻¹ to 0.25 mm⁻¹ as the effective fitting range. Specifically, the calibration method involves selecting defect-free tempered glass samples of the same thickness, material, and surface treatment process, collecting reflected light intensity sequences along a transmission depth from 0 mm to 6 mm, and performing exponential attenuation fitting on the light intensity sequences to obtain the reference light intensity attenuation coefficient range for this batch of tempered glass at a wavelength of 632 nm. The Pearson correlation coefficient threshold of 0.85 is determined based on the fitting backtesting results of the effective interference signal samples and the scattered noise signal samples. When the threshold is 0.80, some random scattered noise is mistakenly retained as effective interference signals; when the threshold is 0.90, some deep weak reflection effective signals are mistakenly rejected; when the threshold is 0.85, a balance can be achieved between the noise rejection rate and the retention rate of deep effective signals. Therefore, 0.85 is chosen as the preferred threshold in this embodiment. By simultaneously limiting the range of the Pearson correlation coefficient and the light intensity attenuation coefficient, the situation where noise contamination causes the exponential fitting results to have a high correlation but deviate from the physical attenuation characteristics of the material can be avoided. If the glass thickness, surface treatment process, or light source wavelength is changed, the same sample calibration and threshold backtesting method can be used to redetermine the Pearson correlation coefficient threshold and the reference light intensity attenuation coefficient range.

[0071] In step S34, the effective interference signals retained in step S33 are extracted, and the spatial pixel coordinates, depth encoding information, cumulative optical path difference, fringe contrast grayscale value, and phase jump coordinates corresponding to the effective interference signals are read. According to the layered structure along the thickness direction of the tempered glass, the effective interference signals are mapped to different virtual detection layers, and the layer position of each effective interference signal is determined based on the depth encoding information. The effective interference signals within the same layer are written into a two-dimensional matrix according to their spatial pixel coordinates. The two-dimensional matrices of different depth levels are stacked along the thickness direction to obtain stress interference data in a three-dimensional matrix format. The calculation method is to divide the depth coordinates of the effective interference signal by the layer interval to obtain the corresponding virtual detection layer number. The stress interference data includes the intensity of the effective interference signal, phase jump coordinates, fringe contrast grayscale value, cumulative optical path difference, and depth layer number.

[0072] It should be noted that the layer mapping process is used to restore the effective interference signal after noise suppression to the detection layer structure in the thickness direction of the tempered glass. The layer spacing is 100 μm, consistent with the layer penetration scan in step S1.

[0073] In step S35, the stress interference data obtained in step S3 is read. The stress interference data includes the effective interference signal intensity, phase jump coordinates, fringe contrast grayscale value, cumulative optical path difference, and depth level number. The truncated phase value in the stress interference data is extracted layer by layer according to the depth level number, and the truncated phase value is limited to the phase wrap-around range of -3.14 to +3.14. The truncated phase values ​​of adjacent pixels are read in the horizontal and vertical directions within the same depth layer, and the truncated phase values ​​of adjacent voxels are read between adjacent depth layers. The calculation method is to subtract the truncated phase value of the previous spatial position from the truncated phase value of the later spatial position to obtain the phase difference between adjacent spatial positions. If the absolute value of the phase difference is greater than 3.14, it is determined that there is a periodic phase truncation between the corresponding adjacent spatial positions, and the position is marked as a phase jump region. The continuously distributed phase jump regions are connected by boundary to obtain the phase truncation boundary.

[0074] It should be noted that the absolute phase difference threshold of 3.14 is used to identify abrupt changes in spatial locations exceeding half a phase period. When the absolute phase difference between adjacent voxels is greater than 3.14, it indicates that there is a periodic phase truncation caused by the wavelength selection range at that location, which requires integer period compensation in subsequent steps.

[0075] In step S36, phase compensation processing is performed on the phase truncation boundary to obtain continuous phase data. Specifically, for the phase truncation boundary obtained in step S41, a quality-oriented three-dimensional region growth phase unpacking algorithm is used to restore the continuous phase. First, the phase derivative variance of each voxel point in the stress interference data is calculated, and a quality map is generated based on the phase derivative variance. In the quality map, voxel points with a phase derivative variance lower than 0.2 are selected as high-reliability seed points, and the unpacking path is expanded outward in order of quality value from high to low. During the expansion process, periodic compensation is performed on adjacent voxels that pass through the phase truncation boundary. The calculation method is as follows: when the difference between the phase value of the next voxel and the phase value of the previous voxel is greater than 3.14, 6.28 is subtracted from the phase value of the next voxel; when the difference is less than -3.14, 6.28 is added to the phase value of the next voxel to obtain the compensated continuous phase value. Phase compensation of all voxel points is completed sequentially according to the unpacking path to obtain continuous phase data.

[0076] It should be noted that the quality map is used to limit the expansion order of the phase unpacking path. Voxel points with a variance value below 0.2 are used as high-reliability seed points to prioritize unpacking from positions with stable phase changes and low noise influence. The phase derivative variance threshold of 0.2 is determined based on the phase unpacking backtesting results of defect-free tempered glass samples, samples with known stress concentrations, and samples containing pre-existing microcracks. Unpacking tests are conducted using candidate thresholds of 0.1, 0.2, and 0.3, respectively, and the unpacking residual, phase continuity recovery rate, and defect region boundary preservation are used as evaluation indicators.

[0077] When the threshold is 0.1, the number of selectable seed points is insufficient, and the unpacking path is easily blocked by local noise regions. When the threshold is 0.3, some voxel points with large phase fluctuations will be mistakenly selected as seed points, causing the unpacking error to spread to the surrounding areas. When the threshold is 0.2, it can balance the number of seed points and the reliability of the initial phase. Therefore, 0.2 is taken as the preferred threshold in this embodiment. The phase compensation amount is 6.28 because the phase wrapping period is two pi. By adding or subtracting an integer multiple of 6.28 to the phase value at the truncation boundary, the phase jump caused by periodic truncation can be eliminated, and the true continuous phase surface can be restored.

[0078] In step S37, the continuous phase data obtained in step S36 is read, and the surface phase region, intermediate phase region, and deep phase region are divided according to the depth encoding information in the thickness direction of the tempered glass. The continuous phase data in the range of 0 to 50 μm near the incident surface is determined as the surface phase information, the continuous phase data in the range of 50 μm to half the glass thickness is determined as the intermediate phase information, and the continuous phase data in the range of half the glass thickness to the bottom surface of the glass is determined as the deep phase information. The phase gradient in the surface phase information, intermediate phase information, and deep phase information is extracted along the thickness direction. The calculation method is to subtract the continuous phase value of the previous depth position from the continuous phase value of the later depth position, and then divide by the distance between the two depth positions to obtain the phase gradient. The surface phase gradient, intermediate phase gradient, and deep phase gradient are compared point by point, and the positions where the phase delay difference per micrometer exceeds 0.05 radians are extracted to obtain the interlayer phase difference characteristics.

[0079] It should be noted that the phase range for the intermediate layer, from 50 μm to half the glass thickness, is used to characterize the stress change in the transition region from the surface compressive stress zone to the interior of the tempered glass. The phase range for the deep layer, from half the glass thickness to the bottom surface, is used to characterize the stress change in the deeper regions of the tempered glass. For a 6 mm thick tempered glass, half the thickness corresponds to 3000 μm; therefore, the phase range for the intermediate layer is 50 μm to 3000 μm, and the phase range for the deep layer is 3000 μm to 6000 μm. The threshold for phase retardation difference per micrometer is set at 0.05 radians, determined based on the allowable range of normal stress gradient differences between the surface, intermediate, and deep layers. When the phase retardation difference between any two adjacent depth regions exceeds 0.05 radians / μm, it indicates a significant interlayer stress difference at that location, which can serve as a characteristic basis for subsequent analysis of the thickness-direction stress gradient variation curve.

[0080] In step S38, the spatial pixel coordinates, depth encoding information, continuous phase values ​​and inter-layer phase difference features in the continuous phase data are read, and a three-dimensional data structure is established with horizontal pixel coordinates, vertical pixel coordinates and depth level numbers as indexes. The depth level numbers are consistent with the virtual detection layer numbers in step S11. For tempered glass with a thickness of 6 mm and a layer spacing of 100 μm, 60 depth layers are retained along the thickness direction.

[0081] The continuous phase values ​​corresponding to each spatial pixel coordinate are written into the corresponding depth level according to their depth encoding information, and the inter-layer phase difference features are written into the feature field of the corresponding pixel position. The phase values ​​of missing depth levels are filled in by linear interpolation of adjacent depth levels. The calculation method is to use the average of the continuous phase values ​​of the previous depth level and the continuous phase values ​​of the next depth level to obtain the interpolated phase value of the missing depth level. Spatial resampling is performed according to the matrix size of 1024 x 1024 x 60 to map the continuous phase data and inter-layer phase difference features to a unified spatial coordinate system and construct a three-dimensional phase matrix.

[0082] It should be noted that the 3D phase matrix size is 1024 x 1024 x 60 to preserve the phase distribution details of the 60 layers in the thickness direction while reducing the computational cost in planar space. 1024 x 1024 corresponds to the planar spatial resolution after resampling, and 60 corresponds to the 60 virtual detection layers with a thickness of 6 mm and a layer spacing of 100 μm. The 3D phase matrix does not compress the 60 depth layers into 3 representative layers, but rather retains the continuous phase values ​​of each virtual detection layer in the thickness direction to ensure that the subsequent stress gradient variation curve can reflect the local phase fluctuations of different depth layers. The surface layer, intermediate layer, and deep layer are only used as statistical partitions for interlayer phase difference characteristics. The surface layer corresponds to 0 to 50 μm, the intermediate layer to 50 μm to half the glass thickness, and the deep layer to half the glass thickness to the bottom surface. The mean, variance, and gradient mean of the continuous phase values ​​within these partitions can be calculated as auxiliary statistical features, but these statistical features do not replace the 60-layer 3D phase matrix.

[0083] In step S4, the stress gradient variation curve along the thickness direction of the three-dimensional phase matrix is ​​extracted, and local slope abrupt changes in the stress gradient variation curve are identified. If the local slope abrupt changes satisfy the preset microcrack determination conditions, then an internal microcrack defect is determined, including:

[0084] S41, extract stress phase data along the thickness direction of the three-dimensional phase matrix, and perform curve fitting processing on the stress phase data to obtain the stress gradient change curve;

[0085] S42, perform local slope identification on the stress gradient change curve to obtain local slope abrupt change characteristics;

[0086] S43, compare the local slope abrupt change feature with the preset microcrack determination condition. If the local slope abrupt change feature meets the preset microcrack determination condition, then mark the microcrack candidate position.

[0087] S44, based on the candidate locations of the microcracks, perform connected component screening and location consistency verification to determine the internal microcrack defects.

[0088] In step S41, the three-dimensional phase matrix obtained in step S3 is read. The three-dimensional phase matrix includes horizontal pixel coordinates, vertical pixel coordinates, depth level numbers, continuous phase values, and inter-layer phase difference features. For each planar pixel coordinate in the three-dimensional phase matrix, the continuous phase values ​​corresponding to layers 1 to 60 are extracted sequentially along the thickness direction of the tempered glass to obtain stress phase data containing 60 depth sampling points. The 60 depth sampling points correspond one-to-one with the 60 virtual detection layers in step S11, and no surface, transition layer, or deep layer compression is performed on the thickness direction data. The stress phase data is extended by thickness direction interpolation to convert the phase points in the 60 discrete depth levels into continuous depth sampling points. A cubic spline interpolation algorithm is used to perform curve fitting on the continuous depth sampling points to obtain the stress gradient change curve corresponding to the planar pixel coordinate. The calculation method is to subtract the continuous phase value of the previous depth sampling point from the continuous phase value of the subsequent depth sampling point, and then divide by the distance between the two depth sampling points to obtain the thickness direction phase gradient. The sequence of phase gradient changes with depth is used as the stress gradient change curve.

[0089] It should be noted that the number of planar pixel coordinates in the three-dimensional phase matrix is ​​determined by the matrix size of 1024 by 1024, corresponding to 1,048,576 planar pixel coordinates; 60 virtual detection layers are retained in the thickness direction to ensure that the stress gradient change curve comes directly from the complete layered scan data, and to avoid smoothing out the local phase abrupt change near the microcrack after compressing the 60 layers of depth information into a small number of representative layers.

[0090] Cubic spline interpolation is only used for continuous point supplementation between adjacent virtual detection layers, and is not used to deduce the complete thickness direction curve from the three control points of the surface layer, transition layer, and deep layer. The thickness direction sampling interval is determined jointly based on the accuracy of internal microcrack depth positioning, the computational cost of phase unpacking, and the minimum identifiable abrupt change duration depth, with a value ranging from 1μm to 5μm, preferably 1μm; among which, the 1μm sampling interval is used to identify short-distance slope changes corresponding to shallow or fine microcracks, and the 5μm sampling interval is used to reduce the computational cost when inspecting large areas of glass in batches.

[0091] In this embodiment, the microcrack depth positioning accuracy is required to be controlled within the range of ±1μm to ±2μm, and the preset microcrack judgment condition requires that the slope change must be maintained within a continuous depth interval of not less than 2μm. To ensure that the 2μm change interval contains at least 3 continuous sampling points, namely the start point, the middle point, and the end point of the change, the thickness direction sampling interval is preferably 1μm. If the sampling interval is greater than 2μm, only 1 sampling point may be obtained within the continuous 2μm change interval, making it impossible to stably determine the depth of the change. If the sampling interval is less than 1μm, the computational load will increase significantly, while the improvement in the microcrack depth positioning accuracy of this embodiment will be limited. The slope peak threshold of 0.15, the slope change amplitude threshold of 0.12, and the change duration depth threshold of 2μm in step S53 are all obtained based on the calibration of the three-dimensional phase matrix with 60 depth layers retained. The calibration sample and the actual detection process use the same 60-layer depth data, the same cubic spline interpolation method, and the same 1μm continuous sampling interval, thereby ensuring that the calibration threshold is consistent with the slope feature distribution during actual detection.

[0092] In step S42, the stress gradient change curve obtained in step S41 is read, and a first-order difference processing is performed on the stress gradient change curve along the thickness direction to obtain a slope sequence. A sliding window with a length of 3μm to 8μm is set on the slope sequence, with a sliding window length of 5μm, and the sliding window is moved along the thickness direction in steps of 1μm. In each sliding window, the initial slope value, peak slope value, slope change duration depth, and slope change direction are extracted. The calculation method is to subtract the initial slope from the peak slope in the sliding window to obtain the local slope change amplitude. If the slope change amplitude continues to increase in multiple adjacent sliding windows, the corresponding depth coordinates, slope peak value, change amplitude, and duration depth are recorded to obtain the local slope change characteristics.

[0093] It should be noted that the sliding window length, ranging from 3μm to 8μm, is determined based on the local stress release range of the internal microcrack in the thickness direction. When the window length is less than 3μm, single-point phase noise is easily misidentified as a slope abrupt change; when the window length is greater than 8μm, short-distance stress abrupt changes caused by local microcracks are easily smoothed and weakened. In this embodiment, the sliding window length is 5μm to cover the local stress release area in the vicinity of the microcrack, while also considering the sensitivity of abrupt change identification and the noise resistance. The sliding step size is 1μm to maintain consistency with the thickness direction sampling interval in step S51, ensuring the accuracy of microcrack depth coordinate positioning. Since adjacent windows overlap by 4μm when the 5μm sliding window moves with a step size of 1μm, corresponding to an overlap rate of 80%, the slope abrupt change detection results generated by adjacent windows are redundant-free and merged.

[0094] If the difference in peak depth between adjacent windows does not exceed 2 μm, the slope change direction is consistent, and the corresponding planar pixel coordinates belong to the same eight-neighbor connected range, it is determined to be a repeated detection result of the same microcrack at different window positions, and the window with the largest slope peak is retained as the representative window. If the difference in peak depth between windows is greater than 2 μm, or the slope change direction is inconsistent, or the corresponding planar pixel coordinates do not belong to the same eight-neighbor connected range, it is retained as different local slope abrupt change features for subsequent multi-microcrack candidate location judgment. The local slope abrupt change features include at least the abrupt change initiation depth, the abrupt change peak depth, the slope change amplitude, the slope peak value, and the abrupt change duration depth, used to characterize whether the stress gradient curve shows a sharp change caused by internal discontinuous structures.

[0095] In step S43, the local slope mutation features obtained in step S42 are extracted. The local slope mutation features include mutation initiation depth, mutation peak depth, slope mutation amplitude, slope peak value, and mutation duration depth. The slope peak value is compared with a preset slope peak value threshold, the slope mutation amplitude is compared with a preset mutation amplitude threshold, and the mutation duration depth is compared with a preset duration depth threshold. The calculation method is to subtract the preset slope peak value threshold from the slope peak value to obtain the slope peak value exceeding the limit. If the slope peak value is greater than 0.15, the slope mutation amplitude is not less than 0.12, and the mutation duration depth is not less than 2μm, then the local slope mutation features are determined to meet the preset microcrack determination conditions. The mutation peak depth that meets the determination conditions is used as the microcrack candidate depth coordinate, and the corresponding planar pixel coordinates are bound to the microcrack candidate depth coordinates to mark the microcrack candidate positions.

[0096] It should be noted that the preset microcrack judgment conditions are determined based on the calibration results of samples containing artificially pre-fabricated microcracks, crack-free control samples, and samples with known stress concentration but no cracks; the slope peak threshold is used to characterize the absolute peak intensity of the local stress gradient, with a value range of 0.12 to 0.18, preferably 0.15; the slope abrupt change amplitude threshold is used to characterize the change amplitude of the slope from the initial value to the peak value within the same sliding window, with a value range of 0.10 to 0.14, preferably 0.12; the abrupt change duration depth threshold is used to exclude instantaneous spikes caused by single-point noise, with a value range of 2μm to 5μm, preferably 2μm. The thickness-direction stress gradient curves of the above samples were backtested, and the slope peak, abrupt change amplitude, and duration depth distribution of the real microcrack region, crack-free region, and ordinary stress concentration region were statistically analyzed. With the goal of minimizing the combined false negative and false positive rates of microcrack identification, the slope peak threshold, slope abrupt change amplitude threshold, and abrupt change duration depth threshold were jointly traversed and calibrated. When the abrupt change duration depth threshold was adjusted from 2μm to 5μm, the system synchronously retested the slope peak threshold and slope abrupt change amplitude threshold to avoid microcrack identification boundary drift caused by changing the duration depth threshold alone.

[0097] In this embodiment, the combined false negative and false positive rates for microcrack identification are lowest when the slope peak threshold is 0.15, the slope abrupt change amplitude threshold is 0.12, and the abrupt change duration depth threshold is 2 μm. Therefore, this set of thresholds is used as the preferred judgment condition. Artificially pre-fabricated microcrack samples are prepared using a combination of laser micro-etching and thermal shock propagation. First, femtosecond laser micro-etching is used to form initial micro-notches in the target depth region inside tempered glass samples of the same batch, thickness, and surface treatment process. Then, localized thermal shock induces the micro-notches to propagate microcracks along the thickness and transverse directions. The length of the artificially pre-fabricated microcracks ranges from 20 μm to 80 μm, the width from 5 μm to 20 μm, and the depth from 50 μm to 500 μm. The crack depth and location are determined by microscopic cross-sectional observation or optical coherence tomography (OCT). In each batch of calibration samples, the number of artificially pre-fabricated microcrack samples shall not be less than 30, the number of crack-free control samples shall not be less than 30, and the number of samples with known stress concentration but no cracks shall not be less than 20. Under the same preparation parameters, the deviation of microcrack length shall be controlled within ±10μm, and the deviation of depth position shall be controlled within ±5μm to ensure the reproducibility of the calibration sample preparation. If the thickness, material composition, tempering process, or surface treatment method of the tempered glass to be tested changes, the corresponding batch of artificially pre-fabricated microcrack samples shall be re-prepared, and the above-mentioned joint threshold calibration procedure shall be repeated.

[0098] In step S44, the candidate microcrack locations obtained in step S43 are read. The candidate microcrack locations include planar pixel coordinates, candidate microcrack depth coordinates, slope peak value, slope change amplitude, and change duration depth. Using the candidate microcrack depth coordinates as the reference plane, an eight-neighbor connected component labeling algorithm is activated in the horizontal space to aggregate candidate points with slope change amplitudes exceeding 0.12 among adjacent pixels. The calculation method is to combine the minimum x-coordinate, maximum x-coordinate, minimum y-coordinate, and maximum y-coordinate of the connected candidate points into the minimum bounding boundary to obtain the horizontal expansion range. The position consistency of candidate points within the same connected component is checked. If the depth coordinate difference of candidate points within the connected component does not exceed 3 μm, and the area of ​​the connected component is not less than the preset minimum defect area, then the connected component is determined as an internal microcrack defect. The depth coordinates, horizontal expansion range, slope peak value, and change amplitude of the internal microcrack defect are written into the defect feature database.

[0099] It should be noted that the eight-neighbor connected component labeling algorithm is used to aggregate spatially adjacent microcrack candidate points into the same defect region; the depth coordinate difference threshold of 3μm is determined based on the thickness direction sampling interval, cubic spline interpolation error, and phase unpacking error, and is used to eliminate pseudo-connected regions caused by noise from different depth layers; when the depth coordinate difference threshold is less than 2μm, some real microcrack regions will be split due to depth estimation deviation; when the depth coordinate difference threshold is greater than 5μm, noise points from different depth layers are easily mistakenly connected. Therefore, 3μm is preferred in this embodiment.

[0100] The preset minimum defect area is determined based on the resolution of the original interferogram, the resampling ratio, and the minimum detectable scale of the internal microcracks. The value ranges from 20 to 40 pixels, with 25 pixels being preferred. When the connected region area is less than 25 pixels, this area is typically an isolated phase noise or scattered light residue point and is not considered a valid internal microcrack defect. The slope abrupt change amplitude threshold of 0.12 is consistent with the preset abrupt change amplitude threshold in step S53, ensuring that all candidate points participating in aggregation within the connected region have obvious stress gradient abrupt change characteristics. Only candidate regions that simultaneously satisfy lateral connectivity, consistent depth and position, and valid area are identified as internal microcrack defects.

[0101] In step S5, the spatial distribution characteristics of the internal microcrack defects are extracted to obtain the defect spatial range; stress concentration difference analysis is performed based on the defect spatial range; if the stress concentration difference meets the preset anomaly judgment conditions, the stress distribution anomaly result of the tempered glass is obtained, including:

[0102] S51, extract the spatial location data of the internal microcrack defects in the three-dimensional phase matrix to obtain spatial distribution characteristics;

[0103] S52, perform thickness direction coordinate mapping on the spatial distribution features to obtain the defect depth layer;

[0104] S53, identify the lateral connected region of the spatial distribution features within the defect depth layer to obtain the defect lateral region;

[0105] S54, merge the defect depth layer and the defect lateral region to obtain the defect spatial range;

[0106] S55, determine the defect analysis area based on the defect spatial range, and extract the stress distribution data of the defect analysis area;

[0107] S56, Perform stress concentration quantization processing on the stress distribution data to obtain defect stress characteristics;

[0108] S57, Extract the normal stress characteristics around the defect analysis area, and compare the defect stress characteristics with the normal stress characteristics to obtain the stress concentration difference;

[0109] S58, if the stress concentration difference meets the preset anomaly judgment condition, then the stress distribution anomaly result of the tempered glass is obtained.

[0110] In step S51, the internal microcrack defects determined in step S4 are read. The internal microcrack defects include candidate depth coordinates, lateral extension range, slope peak value, slope abrupt change amplitude, and connected component pixel set. Based on the planar pixel coordinates in the connected component pixel set, the corresponding lateral coordinates, vertical coordinates, depth level number, continuous phase value, and inter-layer phase difference features are extracted from the three-dimensional phase matrix. All abnormal pixels corresponding to the same internal microcrack defect are sorted according to spatial coordinates to obtain a defect point cloud set. The calculation method is to use the lateral coordinates, vertical coordinates, and depth coordinates of each abnormal pixel to form a three-dimensional coordinate point to obtain the defect point cloud set. The defect point cloud set, slope peak value, slope abrupt change amplitude, connected component area, and depth distribution range are structurally integrated to obtain spatial distribution features.

[0111] It should be noted that spatial distribution features are used to characterize the three-dimensional location of internal microcrack defects within tempered glass. Spatial distribution features include at least the defect point cloud set, the lateral coordinate range, the longitudinal coordinate range, the depth coordinate range, the area of ​​connected regions, the peak slope, and the magnitude of slope abrupt changes. By transforming internal microcrack defects from single candidate points into a defect point cloud set, a spatial basis can be provided for subsequent defect depth layer localization and lateral connected region identification.

[0112] In step S52, the defect point cloud set and depth coordinate range are extracted from the spatial distribution features obtained in step S61, and each abnormal pixel is classified in the thickness direction according to the depth level number in the three-dimensional phase matrix; the depth coordinates of all abnormal pixels in the same internal microcrack defect are statistically analyzed to extract the defect center depth, minimum depth and maximum depth; the calculation method is to average the depth coordinates of all abnormal pixels to obtain the defect center depth; the corresponding thickness direction level is matched according to the defect center depth, and abnormal pixels with a depth coordinate difference of no more than 3μm are classified into the same defect depth layer to obtain the defect depth layer.

[0113] It should be noted that the depth coordinate difference threshold is set to 3 μm, consistent with the position consistency verification threshold in step S54. This threshold is determined based on the thickness direction sampling interval, phase unpacking compensation error, and backtesting error of microcrack candidate depth coordinates. This threshold is used to ensure that abnormal pixels belonging to the same defect depth layer are distributed along the thickness direction of the same microcrack defect, rather than being formed by the misconnection of noise points from different depth layers. The defect center depth is used to characterize the main position of the internal microcrack defect in the thickness direction of the tempered glass, while the minimum and maximum depths are used to characterize the extension range of the defect in the thickness direction.

[0114] In step S53, the spatial distribution features are identified by lateral connectivity within the defect depth layer to obtain the defect lateral region. Specifically, abnormal pixels belonging to the same internal microcrack defect are extracted within the defect depth layer obtained in step S52, and a two-dimensional planar distribution map is established according to the lateral and longitudinal coordinates. The eight-neighbor connected component labeling algorithm is used to identify the connectivity of abnormal pixels in the two-dimensional planar distribution map, and abnormal pixels adjacent in the lateral, longitudinal, or diagonal directions are merged into the same connected region. For each connected region, the minimum lateral coordinate, maximum lateral coordinate, minimum longitudinal coordinate, and maximum longitudinal coordinate are extracted. The calculation method is to subtract the minimum lateral coordinate from the maximum lateral coordinate to obtain the defect length, and subtract the minimum longitudinal coordinate from the maximum longitudinal coordinate to obtain the defect width. Connected regions with an area of ​​not less than 25 pixels and a slope change amplitude of not less than 0.12 are determined as the defect lateral region.

[0115] It should be noted that the eight-neighbor connected region labeling algorithm is used to identify the lateral propagation morphology of internal microcracks within the same depth layer. The minimum defect area is set to 25 pixels, determined based on the spatial mapping ratio between the 4096 x 4096 pixel original interferogram and the 1024 x 1024 resampling matrix, used to eliminate tiny pseudo-defect regions formed by isolated phase noise; the slope abrupt change amplitude threshold is set to 0.12, consistent with the preset abrupt change amplitude threshold in step S53, to ensure that abnormal pixels within the lateral connected region all have obvious stress gradient abrupt change characteristics. The lateral defect region is used to characterize the propagation range of internal microcracks in the tempered glass plane direction.

[0116] In step S54, the defect depth layer and the defect transverse region are fused to obtain the defect spatial range. Specifically, the defect depth layer obtained in step S52 and the defect transverse region obtained in step S53 are read, and the center depth, minimum depth, and maximum depth in the defect depth layer are fused with the defect length, defect width, and minimum circumscribed boundary in the defect transverse region. The minimum circumscribed boundary of the defect transverse region is used as the planar range, and the minimum and maximum depths of the defect depth layer are used as the thickness direction range to construct the three-dimensional spatial envelope of the internal microcrack defect. The calculation method is to use the defect length, defect width, and defect thickness range to jointly define the three-dimensional spatial envelope to obtain the defect spatial range. The defect spatial range is bound to the defect number, center depth, slope peak, slope abrupt change amplitude, and connected region area to obtain the defect spatial range that can be used for subsequent stress concentration difference analysis.

[0117] It should be noted that the defect spatial extent is used to characterize the complete spatial area occupied by internal microcracks within tempered glass, including their position in the thickness direction and their lateral planar extension range. By fusing the defect depth layer and the defect lateral region, the problem of incomplete defect localization caused by outputting only a single depth coordinate or a single planar region can be avoided, and a spatial boundary can be provided for subsequent calculations of the stress concentration differences between the defect region and the surrounding normal region.

[0118] In step S55, the defect analysis region is determined based on the defect spatial range, and stress distribution data of the defect analysis region is extracted. Specifically, the defect spatial range obtained in step S5 is read. The defect spatial range includes the defect center depth, minimum depth, maximum depth, defect length, defect width, three-dimensional spatial envelope, slope peak value, slope abrupt change amplitude, and connected region area. Using the three-dimensional spatial envelope as the boundary of the defect analysis region, continuous phase values, interlayer phase difference characteristics, phase gradient values, and corresponding pixel coordinates located within the defect analysis region are extracted from the three-dimensional phase matrix. Based on the photoelastic conversion coefficient and phase retardation of tempered glass, the continuous phase values ​​are converted into stress tensor data. The calculation method is to multiply the phase retardation by the material photoelastic conversion coefficient and then divide by the detection layer thickness to obtain the stress value at the corresponding position. The stress values, stress tensors, depth coordinates, and spatial coordinates of all pixels within the defect analysis region are structurally integrated to obtain the stress distribution data of the defect analysis region.

[0119] It should be noted that the defect analysis region is bounded by the three-dimensional spatial envelope corresponding to the defect's spatial extent. This is to ensure that subsequent stress concentration difference analysis only targets the area actually occupied by the internal microcrack defect, and does not mistakenly include the surrounding normal area into the defect region. The photoelastic conversion factor is determined based on the birefringence stress response calibration results of the tempered glass material to be tested.

[0120] The photoelastic conversion factor for the material is taken as 18.5 MPa·μm / radian. Known bending loads are applied to defect-free glass samples of the same batch, thickness, and surface treatment process. Standard stress values ​​and corresponding phase retardations under different loads are recorded, establishing a linear calibration relationship between phase retardation and stress values. The slope of the linear fit is used as the photoelastic conversion factor. If the glass manufacturer, glass composition, thickness specification, or tempering process is changed, the same four-point bending photoelastic calibration procedure is used again to determine the corresponding photoelastic conversion factor. The thickness of the detection layer remains consistent with the layer spacing of the aforementioned layered scanning. By converting the phase retardation into stress tensor data, optical phase anomalies can be further transformed into quantifiable mechanical stress distribution data.

[0121] In step S56, stress concentration quantization processing is performed on the stress distribution data to obtain the defect stress characteristics. Specifically, the stress distribution data obtained in step S55 is read, and the stress tensor data, stress value, spatial coordinates, and depth coordinates of each pixel within the defect analysis area are extracted. The finite element integration algorithm is used to perform area integration on the stress tensor of all pixels within the defect analysis area, and the local average stress value is obtained based on the nominal cross-sectional area of ​​the defect analysis area. At the same time, the maximum principal stress value, average principal stress value, stress gradient peak value, and stress concentration area are extracted within the defect analysis area. The calculation method is to divide the area integration result of the stress tensor within the defect analysis area by the nominal cross-sectional area of ​​the defect analysis area to obtain the local average stress value. The calculation method is to divide the local maximum principal stress value of the defect area by the theoretical nominal stress value at that depth level to obtain the stress concentration coefficient of the defect area. The local average stress value, local maximum principal stress value, stress concentration coefficient, stress gradient peak value, and defect spatial range are bound together to obtain the defect stress characteristics.

[0122] It should be noted that the stress concentration factor in the defect region is used to characterize the degree of local stress amplification in the region corresponding to the internal microcrack defect. The theoretical nominal stress value is determined based on the standard stress distribution in the defect-free region at the same depth level and is used as a reference benchmark for the degree of stress concentration in the defect region. When the local maximum principal stress value in the defect region is significantly higher than the theoretical nominal stress value, it indicates that the internal microcrack defect has caused local stress concentration, and further comparison with the surrounding normal region is required.

[0123] In step S57, starting from the minimum outer boundary of the defect analysis area determined in step S55, the area is expanded outward at equal intervals of 10μm to 30μm in both the horizontal and vertical directions to generate an annular buffer zone. In this embodiment, the annular buffer zone is expanded outward at equal intervals of 20μm to generate an annular buffer zone, and the area in the annular buffer zone that is not marked as an internal microcrack defect is determined as the surrounding normal area. Continuous phase values, stress tensors, local principal stress values, and depth coordinates are extracted from the surrounding normal area, and stress concentration quantization is performed using the same finite element integration algorithm as in step S56 to obtain normal stress characteristics. The calculation method is to average the stress concentration coefficients of each sampling point in the surrounding normal area to obtain the average stress concentration coefficient of the surrounding normal area. The calculation method is to divide the stress concentration coefficient of the defect area by the average stress concentration coefficient of the surrounding normal area to obtain the stress concentration difference. The stress concentration difference is then bound to the defect number, defect center depth, defect horizontal area, and maximum principal stress value.

[0124] It should be noted that the expansion size of the annular buffer zone is determined based on the lateral dimension of the defect, the range of stress disturbance attenuation, and the stability of the reference area at the same depth, with a value ranging from 10μm to 30μm. When the expansion size is less than 10μm, the surrounding normal area is easily affected by stress disturbance at the defect edge, resulting in a higher reference stress. When the expansion size is greater than 30μm, the reference area may span different optical paths or different local states of materials, causing the comparison benchmark to deviate from the defect neighborhood.

[0125] In this embodiment, the maximum lateral size of the internal microcrack defect is typically less than 50 μm, and the annular buffer zone preferably expands outward by 20 μm to establish a normal stress reference region with the same depth, similar optical path, and similar material state within the defect neighborhood. If the maximum lateral size of the defect is greater than 50 μm, the expansion size of the annular buffer zone can be 0.4 to 0.6 times the maximum lateral size of the defect. The surrounding normal region must exclude marked internal microcrack defect pixels and abnormal points with a slope abrupt change of not less than 0.12 to ensure that the normal stress characteristics can reflect the reference stress state around the defect.

[0126] In step S58, if the stress concentration difference meets the preset anomaly judgment conditions, the result of an abnormal stress distribution of the tempered glass is obtained. Specifically, the stress concentration difference obtained in step S56 is read and input into the defect assessment rule engine; the defect assessment rule engine compares the stress concentration difference with the preset anomaly judgment conditions, which include the stress concentration difference being greater than the anomaly judgment threshold, the local maximum principal stress value in the defect area being higher than the theoretical nominal stress value at the same depth, and the existence of a continuously connected microcrack area within the defect space; the anomaly judgment threshold ranges from 2.30 to 2.80, preferably 2.50 in this embodiment; the calculation method is to subtract the anomaly judgment threshold from the stress concentration difference to obtain the anomaly exceedance; if the anomaly exceedance is positive and the microcrack area within the defect space meets the connectivity validity requirement, the tempered glass is judged to have an abnormal stress distribution; the defect number, defect center depth, defect lateral area, local maximum principal stress value, stress concentration difference, and anomaly judgment conclusion are written into the quality inspection log to obtain the result of an abnormal stress distribution of the tempered glass.

[0127] It should be noted that the anomaly determination threshold is determined based on the backtesting results of stress concentration differences between samples containing pre-fabricated microcracks, crack-free control samples, and historical failure samples. Specifically, the stress concentration differences in crack-free control areas, ordinary stress concentration areas, and areas with confirmed internal microcracks are statistically analyzed, and candidate threshold values ​​of 2.30, 2.50, and 2.80 are used for anomaly determination backtesting. When the threshold value is 2.30, some ordinary stress concentration areas are easily misjudged as abnormal defects; when the threshold value is 2.80, some real microcrack areas are easily missed; when the threshold value is 2.50, a balance can be achieved between the accuracy of anomaly identification and the missed detection rate. Therefore, 2.50 is selected as the preferred anomaly determination threshold value in this embodiment.

[0128] When the ratio of the stress concentration factor in the defect area to the average stress concentration factor in the surrounding normal area is greater than 2.50, it indicates that the stress concentration in the defect area is significantly higher than that in the surrounding reference area, posing a risk of abnormal stress distribution. Using the local maximum principal stress value, stress concentration difference, and defect spatial range as joint anomaly judgment conditions can avoid misjudgment due to single-point stress peaks or local noise. If the thickness of the tempered glass, the tempering process, or the surface treatment process changes, the anomaly judgment threshold can be recalibrated using samples from the same batch without cracks and samples containing pre-fabricated microcracks.

[0129] A second embodiment of the present invention provides a system for detecting the stress distribution of tempered glass, comprising: a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the program to implement the steps of the method described above.

[0130] It should be noted that the tempered glass stress distribution detection system provided in this embodiment of the invention is used to execute all the process steps of the tempered glass stress distribution detection method in the above embodiment. The working principles and beneficial effects of the two are one-to-one, so they will not be described again.

[0131] It should be noted that the device embodiments described above are merely illustrative. The units described as separate components may or may not be physically separate, and the components shown as units may or may not be physical units; that is, they may be located in one place or distributed across multiple network units. Some or all of the modules can be selected to achieve the purpose of this embodiment according to actual needs. Furthermore, in the accompanying drawings of the device embodiments provided by this invention, the connection relationships between modules indicate that they have communication connections, which can be specifically implemented as one or more communication buses or signal lines. Those skilled in the art can understand and implement this without any creative effort.

[0132] The specific embodiments described above further illustrate the purpose, technical solution, and beneficial effects of the present invention. It should be understood that the above descriptions are merely specific embodiments of the present invention and are not intended to limit the scope of protection of the present invention. In particular, it should be noted that any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the scope of protection of the present invention for those skilled in the art.

Claims

1. A method for detecting stress distribution in tempered glass, characterized in that, include: The phase difference signal of the reflected light from the tempered glass is acquired, and the phase difference signal of the reflected light is preprocessed to obtain the original interferogram. Local phase anomaly features are extracted from the original interferogram. If the local phase anomaly features meet the preset stress concentration judgment conditions, a stress feature map is obtained. The stress feature map is subjected to noise suppression processing to obtain stress interference data; the stress interference data is subjected to phase unpacking processing to obtain continuous phase data, and a three-dimensional phase matrix is ​​constructed based on the continuous phase data; Extract the stress gradient variation curve of the three-dimensional phase matrix along the thickness direction, identify the local slope abrupt change feature in the stress gradient variation curve, and if the local slope abrupt change feature meets the preset microcrack judgment condition, then determine the internal microcrack defect. The spatial distribution characteristics of the internal microcrack defects are extracted to obtain the defect spatial range; stress concentration difference analysis is performed based on the defect spatial range, and if the stress concentration difference meets the preset anomaly judgment conditions, the stress distribution anomaly result of the tempered glass is obtained.

2. The method for detecting stress distribution in tempered glass according to claim 1, characterized in that, The process involves acquiring the phase difference signal of the reflected light from the tempered glass, preprocessing the reflected light phase difference signal to obtain the original interferogram, including: The vibration direction of the incident beam is controlled by polarization modulation, and the tempered glass is subjected to multi-angle layered penetration scanning to obtain the layered reflected light signal; Extract the phase difference signals of the reflected light corresponding to different transmission depths from the layered reflected light signals; The reflected light phase difference signal is subjected to optical path difference conversion and depth encoding to obtain layered interference data; Based on the layered interferometric data, an original interferogram containing fringe contrast distribution and depth encoding information is generated.

3. The method for detecting stress distribution in tempered glass according to claim 1, characterized in that, The step of extracting local phase anomaly features from the original interferogram, and obtaining a stress feature map if the local phase anomaly features satisfy a preset stress concentration determination condition, includes: Extract the fringe spacing variation data and phase jump data from the original interferogram to obtain local phase anomaly data; Spatial coordinate mapping and grayscale gradient transformation are performed on the local phase anomaly data to obtain local phase anomaly features; The local phase anomaly features are compared with the preset stress concentration judgment conditions. If the preset stress concentration judgment conditions are met, the stress concentration candidate region is marked. Pixel-level mapping is performed on the candidate stress concentration regions to obtain a stress feature map.

4. The method for detecting stress distribution in tempered glass according to claim 1, characterized in that, The process of performing noise suppression processing on the stress feature map to obtain stress interferometry data includes: Extract the blurred edge regions from the stress feature map to obtain the stress region to be denoised; Scattered light noise separation is performed on the interference fringe data in the stress region to be denoised to obtain candidate interference signals; The candidate interference signals are matched with a preset light intensity attenuation rule, and the valid interference signals that satisfy the preset light intensity attenuation rule are retained. Based on the effective interference signal, layered mapping processing is performed to obtain stress interference data.

5. The method for detecting stress distribution in tempered glass according to claim 1, characterized in that, The step of performing phase unpacking processing on the stress interferometric data to obtain continuous phase data, and constructing a three-dimensional phase matrix based on the continuous phase data, includes: Extract the truncated phase value from the stress interference data, identify the phase jump region between adjacent spatial positions, and obtain the phase truncation boundary; Phase compensation processing is performed on the phase truncation boundary to obtain continuous phase data; Extract surface phase information and deep phase information from the continuous phase data to obtain interlayer phase difference characteristics; A three-dimensional phase matrix is ​​constructed by spatial mapping based on the continuous phase data and the interlayer phase difference characteristics.

6. The method for detecting stress distribution in tempered glass according to claim 1, characterized in that, The process involves extracting the stress gradient variation curve along the thickness direction of the three-dimensional phase matrix, identifying local slope abrupt changes in the stress gradient variation curve, and determining internal microcrack defects if the local slope abrupt changes meet preset microcrack judgment conditions. This includes: Stress phase data is extracted along the thickness direction of the three-dimensional phase matrix, and curve fitting is performed on the stress phase data to obtain the stress gradient change curve. Local slope identification is performed on the stress gradient change curve to obtain local slope abrupt change characteristics; The local slope abrupt change feature is compared with the preset microcrack determination condition. If the local slope abrupt change feature meets the preset microcrack determination condition, the microcrack candidate position is marked. Based on the candidate locations of the microcracks, connected component screening and location consistency verification are performed to determine the internal microcrack defects.

7. The method for detecting stress distribution in tempered glass according to claim 1, characterized in that, The extraction of the spatial distribution characteristics of the internal microcrack defects to obtain the defect spatial range includes: The spatial location data of the internal microcrack defects in the three-dimensional phase matrix are extracted to obtain spatial distribution characteristics; The spatial distribution features are mapped along the thickness direction to obtain the defect depth layer; Within the defect depth layer, the spatial distribution features are identified by lateral connectivity regions to obtain the defect lateral regions. By fusing the defect depth layer and the defect lateral region, the defect spatial range is obtained.

8. The method for detecting stress distribution in tempered glass according to claim 1, characterized in that, The stress concentration difference analysis is performed based on the defect spatial range. If the stress concentration difference meets the preset anomaly judgment conditions, the stress distribution anomaly result of the tempered glass is obtained, including: The defect analysis area is determined based on the defect spatial range, and the stress distribution data of the defect analysis area is extracted. The stress distribution data is subjected to stress concentration quantization processing to obtain the defect stress characteristics; Extract the normal stress characteristics around the defect analysis area, and compare the defect stress characteristics with the normal stress characteristics to obtain the stress concentration difference; If the stress concentration difference meets the preset anomaly judgment condition, then the stress distribution anomaly result of the tempered glass is obtained.

9. A system for detecting stress distribution in tempered glass, comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that, When the processor executes the program, it implements the steps of the method for detecting the stress distribution of tempered glass according to any one of claims 1 to 8.

10. A computer-readable storage medium having a computer program stored thereon, characterized in that, When executed by a processor, the program implements the steps of the method for detecting the stress distribution of tempered glass as described in any one of claims 1 to 8.

Images