Real-time quality defect detection method and system for OCA (Optical Clear Adhesive)

Abstract

The invention relates to the technical field of optical material quality detection, and discloses a real-time quality defect detection method and system for OCA optical cement. The method comprises the following steps: acquiring infrared thermal imaging data and colloid thickness data of an optical adhesive sample, and establishing an original thermal imaging data set; solving a temperature value of each point in the colloid according to the original thermal imaging data set to obtain temperature gradient field data, and drawing to obtain a temperature distribution model in the colloid; calculating to obtain a temperature average change rate reflecting heat conduction characteristics of each region; screening candidate defect regions; constructing and normalizing a multi-dimensional feature vector; generating a standard defect feature vector; generating a detailed defect classification result; and calculating geometric parameters and severity rating of each defect area to obtain an optical cement quality detection report. The method improves the reliability of the detection result of the internal defect of the optical cement.

Description

Technical Field

[0001] This invention relates to the field of optical material quality inspection technology, and in particular to a real-time quality defect detection method and system for OCA optical adhesive. Background Technology

[0002] Optical adhesive (OCA), a key material in display technology, is widely used in the manufacturing of touchscreens and LCDs. Its quality directly affects the optical performance and user experience of the product. With the increasing demands for display quality in consumer electronics, the quality of optical adhesive directly impacts the final product's optical performance and user experience, making its quality inspection an indispensable part of the manufacturing process.

[0003] In existing technologies, traditional inspection methods such as optical microscopy and machine vision primarily target surface defects. However, for hidden defects within colloids, such as microbubbles, impurities, and uneven thickness, traditional optical methods based on surface morphology are difficult to detect directly. These internal defects alter the local thermal conductivity of the material, but existing technologies lack the means to accurately invert the internal temperature field from measurable surface or overall thermal signals and establish a quantitative correspondence between it and the defect type and severity. This results in low reliability of inspection results, failing to meet the quality control requirements of high-precision manufacturing. Summary of the Invention

[0004] To address the aforementioned technical problems, this invention provides a real-time quality defect detection method and system for OCA optical adhesive, enabling quantitative analysis of defect types and improving the reliability of detection results.

[0005] In a first aspect, to solve the above-mentioned technical problems, the present invention provides a real-time quality defect detection method for OCA optical adhesive, comprising: Infrared thermal imaging data and colloid thickness data of optical adhesive samples were acquired, and three-dimensional data were constructed to obtain a raw thermal imaging dataset containing complete spatial information. Temperature gradient field data is obtained by solving the temperature values ​​of each point inside the colloid based on the original thermal imaging dataset, and a temperature distribution model inside the colloid is drawn based on the temperature gradient field data. The average temperature change rate, which reflects the heat conduction characteristics of each region, is calculated from the temperature values ​​in the temperature distribution model. Regions with an average temperature change rate exceeding a preset change rate threshold are selected as candidate defect regions. For each location point in the candidate defect region, the temperature gradient features, conversion rate features, and spatial distribution features in its surrounding neighborhood are extracted to construct a multi-dimensional feature vector. The multi-dimensional feature vector is then normalized by combining the colloid thickness data to generate a standard defect feature vector. Based on the standard defect feature vector and the pre-established defect feature library, similarity matching is performed to obtain preliminary defect classification results; Based on the preliminary defect classification results, abnormal bubble regions are extracted, and a decision tree classifier is used to further classify the abnormal bubble regions to generate detailed defect classification results. Spatial clustering analysis is performed on the detailed defect classification results to obtain a defect proximity relationship dataset. Multiple single defects continuously distributed in the defect proximity relationship dataset are merged into a composite defect region. The geometric parameters and severity rating of each defect region are calculated to obtain an optical adhesive quality inspection report.

[0006] In one optional implementation, the acquisition of infrared thermal imaging data and colloid thickness data of the optical adhesive sample, and the creation of a three-dimensional data set to obtain a raw thermal imaging dataset containing complete spatial information, includes: By performing line-by-line infrared scanning on the surface of the optical adhesive sample, the radiation value of each point in the colloid is obtained and the coordinates of the colloid surface are recorded to obtain infrared thermal imaging data. Based on the radiation value, the radiation value is converted using the Stepan-Boltzmann law to obtain the temperature value at each point; Based on the coordinates of the colloid surface, the thickness of the colloid surface is measured using triangulation to obtain the thickness data of the colloid. The infrared thermal imaging data and colloid thickness data are used to construct a three-dimensional dataset, and the original thermal imaging dataset is obtained by using the colloid surface coordinates as an index.

[0007] In one optional implementation, the step of obtaining temperature gradient field data by solving for the temperature values ​​at various points inside the colloid based on the original thermal imaging dataset, and drawing a temperature distribution model inside the colloid based on the temperature gradient field data, includes: The original thermal imaging dataset is processed using cubic spline interpolation. If the distance between adjacent points exceeds the preset grid, the interpolation nodes are densified to obtain the initial temperature field distribution matrix of the colloid. The heat conduction partial differential equation for calculating the initial temperature field distribution matrix is ​​discretized using the finite difference method to obtain the numerical solution of the temperature field. When the temperature gradient change at the boundary nodes exceeds the preset value, the time step is further reduced and iterated until the numerical solution of the temperature field converges to the thermal equilibrium state. Spatial differentiation calculation is performed on the numerical solution of the temperature field, and the high temperature difference part is marked to obtain the temperature gradient field data. Based on the temperature gradient field data, the intersection point of the isothermal surface and the preset grid boundary is determined by the contour tracing algorithm, thereby obtaining the temperature distribution model inside the colloid.

[0008] In one optional implementation, the average temperature change rate reflecting the thermal conductivity characteristics of each region is calculated from the temperature values ​​in the temperature distribution model, and regions with an average temperature change rate exceeding a preset threshold are selected as candidate defect regions, including: The temperature distribution model is sampled to obtain a continuous time series. A sliding window is used to continuously sample the temperature values ​​through an observation window with a fixed time length. The temperature values ​​in the continuous time series are extracted to obtain a standardized time series temperature matrix. By performing differential operations, the temperature values ​​of adjacent time points in the time-series temperature matrix are subtracted point by point, and the location of abnormal temperature difference is identified by the change in values, resulting in a spatial distribution record table containing the location of abnormal temperature difference. Based on the spatial distribution record table, an average filter is used to smooth the temperature difference change sequence within the spatial region, and the average temperature change rate per unit time for each spatial region is calculated. The average temperature change rate is compared with a preset change rate threshold, and the region where the average temperature change rate is greater than the preset change rate threshold is selected as a candidate defect region.

[0009] In one optional implementation, for each location point in the candidate defect region, temperature gradient features, transformation rate features, and spatial distribution features within its surrounding neighborhood are extracted to construct a multi-dimensional feature vector. This multi-dimensional feature vector is then normalized using the colloid thickness data to generate a standard defect feature vector, including: A three-dimensional neighborhood search algorithm is used to establish a cubic neighborhood window at each location point in the candidate defect region to obtain a temperature distribution data set containing the neighborhood spatial structure. Based on the temperature distribution data set, the temperature distribution in the neighborhood is analyzed by using the gradient operator to obtain the temperature gradient changes in each direction, thereby obtaining the temperature gradient component values ​​in the lateral, longitudinal, and depth directions. Based on the temperature gradient component values, the average temperature change rate, and the spatial coordinate information of each location point, an initial multidimensional feature vector containing gradient features, rate features, and location features is constructed. By combining the colloid thickness data and comparing the values ​​that exceed the preset standard thickness range, a thickness ratio coefficient is determined. The feature components in the initial multidimensional feature vector are then weighted according to the thickness ratio coefficient to obtain the adjusted multidimensional feature vector. The adjusted multidimensional feature vector is normalized and mapped to a preset standard range using the maximum and minimum value normalization method, thereby obtaining the normalized feature value. The normalized feature values ​​are subjected to a vector concatenation operation. The vector concatenation operation combines various features into a unified multidimensional vector according to a predetermined order of temperature gradient features, conversion rate features, and spatial distribution features, thereby obtaining the defect feature vector.

[0010] In one optional implementation, similarity matching is performed based on the standard defect feature vector and a pre-established defect feature library to obtain preliminary defect classification results, including: Based on the standard defect feature vector, multi-dimensional extraction is performed to obtain feature components of each dimension; Calculate the Euclidean distance between each feature component and the standard feature vector in the pre-established defect feature library; The Euclidean distance value is compared with a preset distance threshold. If the Euclidean distance value is less than the preset distance threshold, the feature vector corresponding to the Euclidean distance value is marked as a candidate match. Based on the candidate matching items, the cosine value of the angle between the feature vectors is calculated using cosine similarity. The final matching degree value is obtained by comprehensively scoring based on the weight coefficient of the cosine value in the defect feature database. After sorting the final matching scores in descending order, the best matching result with the highest matching score is obtained. Based on the best matching result, the matching value is converted into a confidence score using a confidence conversion function, and the corresponding defect type label is extracted from the defect feature library to determine the preliminary defect classification result.

[0011] In one optional implementation, abnormal bubble regions are extracted based on the preliminary defect classification results, and a decision tree classifier is used to refine the classification of the abnormal bubble regions to generate detailed defect classification results, including: When the preliminary defect classification result includes internal bubble defects, the bubble position is three-dimensionally located and obtained by laser scanning. When a bubble is located outside the preset center area, the location is marked as an abnormal bubble point, and the thickness is sampled based on the bubble location to obtain the colloid thickness distribution. The results are combined to obtain a first dataset containing the bubble location and thickness distribution. Temperature sampling is performed on the abnormal bubble points, and the temperature difference of the abnormal bubble points is detected by infrared thermal imaging scanning to obtain the temperature difference value of the abnormal bubble points. If the temperature difference value exceeds the preset temperature fluctuation threshold, the abnormal temperature area is marked, and the correspondence between the abnormal temperature and the structural defects is determined in the first dataset to obtain a second dataset containing the abnormal temperature area. The second dataset is further classified using a decision tree classifier. After receiving a multi-dimensional feature vector containing positional offset, thickness variation, and temperature anomaly, the decision tree classifier performs subtype identification through branch judgment logic to obtain detailed defect classification results.

[0012] In one optional implementation, spatial clustering analysis is performed on the detailed defect classification results to obtain a defect proximity relationship dataset. Multiple consecutively distributed single defects in the defect proximity relationship dataset are merged into composite defect regions. The geometric parameters and severity rating of each defect region are calculated to obtain an optical adhesive quality inspection report, including: Based on the detailed defect classification results, the spatial proximity distance between each defect point is measured to obtain the spacing value between each adjacent defect. If the distance between two defect points is less than a preset proximity threshold, they are marked as an associated defect pair, and all associated defect pairs are counted to obtain a defect proximity relationship dataset. Based on the distribution continuity judgment of the associated defect pairs in the defect proximity relationship dataset, if three or more associated defect pairs form a continuous path in space, the path tracing pair is used to analyze the path of these points distributed along a straight line, and the multiple continuously distributed associated defect pairs are merged into a unified composite defect region. Extract the geometric parameters of the composite defect region, including the normalized defect region area, shape complexity, and distribution density; A comprehensive score is calculated based on preset weights and the geometric parameters, and a severity rating is obtained by classifying the comprehensive score into levels. The defect location, geometric parameters, and severity rating of the composite defect area are output as an optical adhesive quality inspection report.

[0013] Secondly, the present invention provides a real-time quality defect detection system for OCA optical adhesive, comprising: The data acquisition module is used to acquire infrared thermal imaging data and colloid thickness data of optical adhesive samples, and to build a three-dimensional data to obtain a raw thermal imaging dataset containing complete spatial information. The temperature analysis module is used to solve the temperature values ​​of each point inside the colloid based on the original thermal imaging dataset to obtain temperature gradient field data, and to draw a temperature distribution model inside the colloid based on the temperature gradient field data. The region filtering module is used to calculate the average temperature change rate reflecting the heat conduction characteristics of each region from the temperature values ​​in the temperature distribution model, and to filter regions whose average temperature change rate exceeds a preset change rate threshold as candidate defect regions. The feature extraction module is used to extract temperature gradient features, conversion rate features and spatial distribution features in the surrounding neighborhood of each location point in the candidate defect region to construct a multi-dimensional feature vector, and to normalize the multi-dimensional feature vector by combining the colloidal thickness data to generate a standard defect feature vector. The defect matching module is used to perform similarity matching based on the standard defect feature vector and the pre-established defect feature library to obtain preliminary defect classification results; The detailed classification module is used to extract abnormal bubble regions based on the preliminary defect classification results, and to use a decision tree classifier to further classify the abnormal bubble regions to generate detailed defect classification results. The report generation module is used to perform spatial clustering analysis on the detailed defect classification results to obtain a defect proximity relationship dataset. Multiple single defects that are continuously distributed in the defect proximity relationship dataset are merged into a composite defect region. The geometric parameters and severity rating of each defect region are calculated to obtain an optical adhesive quality inspection report.

[0014] Thirdly, the present invention provides an electronic device, including a processor, a memory, and a computer program stored in the memory, wherein the processor executes the computer program to implement any of the above-described methods for real-time quality defect detection of OCA optical adhesive.

[0015] Fourthly, the present invention provides a computer-readable storage medium storing a computer program that, when executed by a processor, implements any of the above-described methods for real-time quality defect detection of OCA optical adhesive.

[0016] Compared with the prior art, the present invention has the following beneficial effects: (1) This invention realizes direct and visual analysis of the internal thermal conductivity characteristics of OCA optical adhesive by solving the heat conduction equation to invert the internal three-dimensional temperature gradient field, providing a new physical observation dimension for detecting internal hidden defects and breaking through the limitations of traditional surface detection.

[0017] (2) By integrating the spatiotemporal change rate of temperature, multidimensional feature vector extraction and normalization, a feature system that can accurately quantify the thermal and geometric properties of defects was constructed, realizing automatic classification and severity rating of defects.

[0018] (3) The present invention is based on a unified heat conduction physical model, and each step is closely connected, from physical inversion, anomaly screening, feature quantification to intelligent identification. This significantly improves the detection accuracy, quantitative analysis capability and overall reliability of the results for defects such as internal bubbles and uneven thickness. Attached Figure Description

[0019] Figure 1This is a flowchart illustrating the real-time quality defect detection method for OCA optical adhesive provided in the first embodiment of the present invention; Figure 2 This is a schematic diagram of the structure of the real-time quality defect detection system for OCA optical adhesive provided in the second embodiment of the present invention. Detailed Implementation

[0020] 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.

[0021] To address the challenge of accurately detecting and quantifying hidden defects (such as microbubbles and uneven thickness) within OCA optical adhesives, this invention proposes a defect detection method based on internal thermal conductivity analysis. It's worth noting that internal defects alter the local thermal conductivity of the adhesive, creating a unique temperature gradient field in infrared thermal imaging. This invention reconstructs the three-dimensional temperature field and analyzes its dynamic changes to determine the spatial location and type of internal defects. This invention organically integrates multiple steps, including data interpolation, solving physical equations, feature extraction, and pattern recognition. First, spatial interpolation and high-dispersion solving of the thermal conductivity equations allow for the reconstruction of the true three-dimensional internal temperature gradient field from surface infrared data. Second, by analyzing the spatiotemporal rate of change of the temperature field, regions with abnormal thermal conductivity are identified. Then, multi-dimensional feature vectors are extracted from these abnormal regions to quantify the thermal and geometric properties of the defects. Finally, by matching and classifying with a known defect database, accurate identification and quantitative assessment of defects are achieved. This series of steps constitutes a complete and highly reliable detection scheme capable of quantifying internal defects.

[0022] Reference Figure 1 The first embodiment of the present invention provides a real-time quality defect detection method for OCA optical adhesive, the specific steps of which are as follows: S101: Acquire infrared thermal imaging data and colloid thickness data of the optical adhesive sample, and perform three-dimensional data construction to obtain the original thermal imaging dataset containing complete spatial information. S102, based on the original thermal imaging dataset, the temperature gradient field data is obtained by solving the temperature values ​​of each point inside the colloid, and a temperature distribution model inside the colloid is drawn based on the temperature gradient field data. S103, calculate the average temperature change rate reflecting the heat conduction characteristics of each region by calculating the temperature values ​​in the temperature distribution model, and screen regions whose average temperature change rate exceeds the preset change rate threshold as candidate defect regions. S104. For each location point in the candidate defect region, extract the temperature gradient features, conversion rate features, and spatial distribution features in its surrounding neighborhood to construct a multi-dimensional feature vector, and normalize the multi-dimensional feature vector by combining it with the colloid thickness data to generate a standard defect feature vector. S105, perform similarity matching based on the standard defect feature vector and the pre-established defect feature library to obtain preliminary defect classification results; S106, Extract abnormal bubble regions based on the preliminary defect classification results, and use a decision tree classifier to refine the classification of the abnormal bubble regions to generate detailed defect classification results; S107, Spatial clustering analysis is performed on the detailed defect classification results to obtain a defect proximity relationship dataset. Multiple single defects continuously distributed in the defect proximity relationship dataset are merged into a composite defect region. The geometric parameters and severity rating of each defect region are calculated to obtain an optical adhesive quality inspection report.

[0023] In step S101, infrared thermal imaging data and colloid thickness data of the optical adhesive sample are acquired, and three-dimensional data are established to obtain the original thermal imaging dataset containing complete spatial information.

[0024] In one implementation, a thermal imaging sensor array scans the optical adhesive sample line by line according to a preset rasterized scanning path. The sensor array contains 256×256 pixels, each corresponding to a spatial resolution of 0.1 micrometers. When a temperature change on the adhesive surface is detected to exceed a preset threshold of 2 degrees Celsius, the system automatically records the X and Y coordinates of that location.

[0025] In this embodiment, the laser ranging device uses the triangulation principle to detect the thickness of the area where temperature data has been collected, achieving a measurement accuracy of 0.01 micrometers. When the thickness difference between adjacent measurement points exceeds the tolerance range of 0.05 micrometers, the system initiates a bilinear interpolation algorithm to complete the missing data.

[0026] In one possible implementation, a three-dimensional data structure is established based on the temperature and thickness data. Spatial coordinates are stored as index keys in a data matrix. The three-dimensional data structure uses a hash table for storage, with spatial coordinates (x, y) as index keys, corresponding to the stored temperature and thickness information, resulting in an original thermal imaging dataset containing complete spatial information. This storage method enables rapid retrieval of complete spatial information at any location, providing efficient data access capabilities for subsequent temperature field reconstruction.

[0027] The depth calibration process considers the influence of colloid thickness on infrared thermometry. Since the attenuation of the infrared signal in the colloid is thickness-dependent, a thickness correction coefficient is introduced to standardize the original temperature data. For example, for typical thickness ranges of OCA optical adhesives (e.g., 50 μm to 300 μm), a thickness-correction coefficient relationship can be established through experimental calibration or theoretical models. In a preferred embodiment, the correction coefficient is set to 1.0 when the baseline thickness is 100 μm; when the thickness increases to 200 μm, the correction coefficient can be adjusted to 0.92. This standardization effectively reduces measurement errors caused by thickness inhomogeneity and improves the comparability of temperature data from different thickness regions.

[0028] In step S102, temperature gradient field data is obtained by solving for the temperature values ​​of each point inside the colloid based on the original thermal imaging dataset, and a temperature distribution model inside the colloid is drawn based on the temperature gradient field data, including: Based on the original thermal imaging dataset, cubic spline interpolation is used to perform spatial interpolation on the temperature distribution of the colloidal surface and interior to obtain an initial temperature field distribution matrix covering the entire colloidal region. Based on the initial temperature field distribution matrix, the three-dimensional heat conduction partial differential equation is numerically discretized using the finite difference method to obtain the numerical solution of the temperature field that satisfies the thermal equilibrium condition. By performing spatial differentiation on the numerical solution of the temperature field using the gradient operator, temperature gradient field data reflecting the heat flow distribution characteristics inside the colloid are obtained. Based on the temperature gradient field data, a temperature distribution model inside the colloid is plotted using linear interpolation. The temperature distribution model includes an isothermal surface distribution map.

[0029] In this embodiment, a cubic spline interpolation algorithm is first used to spatially interpolate the temperature of the colloid surface and interior. For example, when the coordinates of adjacent measurement point A are detected to be 10.5 micrometers and the temperature is 43.2 degrees Celsius, and the coordinates of measurement point B are detected to be 15.8 micrometers and the temperature is 41.7 degrees Celsius, the distance between the two points is 5.3 micrometers, which exceeds the requirement of a preset grid size of 2.0 micrometers. The system automatically inserts densification nodes at positions of 12.0 micrometers and 14.0 micrometers, and calculates the corresponding temperature values ​​of 42.6 degrees Celsius and 42.1 degrees Celsius through spline functions, forming a uniformly distributed initial temperature field distribution matrix.

[0030] Secondly, the finite difference method transforms the three-dimensional heat conduction partial differential equation of the initial temperature field distribution matrix into a system of algebraic equations for numerical solution. Taking the coordinates of nodes inside the colloid at 8.0 μm, 6.5 μm, and 1.2 μm as an example, the temperature at this point is affected by the heat conduction of the surrounding six adjacent nodes. When the rate of change of the temperature gradient at the boundary nodes reaches 0.85 degrees Celsius per μm, exceeding the convergence threshold of 0.5 degrees Celsius per μm, the system adjusts the time step from 0.01 seconds to 0.005 seconds and re-iterates the calculation. After 15 iterations, the numerical solution of the temperature field converges to the thermal equilibrium state, and the node temperature stabilizes at 42.8 degrees Celsius, yielding a numerical solution that satisfies thermal equilibrium.

[0031] Then, the gradient operator obtains the temperature gradient field data by calculating the partial derivatives of the temperature field in the X, Y, and Z directions. For example, when the temperature in the detection area drops from 44.5 degrees Celsius to 39.2 degrees Celsius within a 1.5-micrometer distance, the calculated temperature gradient amplitude is 3.53 degrees Celsius per micrometer. Since this value exceeds the critical thermal stress value of 2.8 degrees Celsius per micrometer for optical adhesive materials, the system automatically marks this area as a region of high temperature difference variation, indicating that there may be internal defects or stress concentration at this location.

[0032] It should be noted that the contour tracing algorithm uses linear interpolation to determine the intersection points of the isotherm surface and the grid boundary. Taking the 42°C isotherm as an example, when the temperature at grid node A is 41.3°C and the temperature at node B is 43.1°C, the 42°C isotherm point is located on the line connecting the two nodes, at a distance of 0.39 times the side length of point A. By connecting all the 42°C intersection points, a complete isotherm surface distribution map is formed.

[0033] In step S103, the average temperature change rate reflecting the heat conduction characteristics of each region is calculated from the temperature values ​​in the temperature distribution model. Regions with an average temperature change rate exceeding a preset threshold are selected as candidate defect regions, including: A continuous time series is obtained by sampling according to the temperature distribution model. The temperature values ​​in the continuous time series are segmented and extracted using a sliding window to obtain a standardized time series temperature matrix. By performing differential operations, the temperature values ​​of adjacent time points in the time-series temperature matrix are subtracted point by point to obtain a spatial distribution record table containing the locations of abnormal temperature differences. A moving average filter is used to smooth the temperature difference change sequence in each spatial region of the spatial distribution record table, and the average temperature change rate per unit time of each spatial region is calculated. The average temperature change rate is compared with a preset change rate threshold, and the region where the average temperature change rate is greater than the preset change rate threshold is determined as a candidate defect region.

[0034] In this embodiment, the sliding window method first performs continuous sampling of historical temperature data by setting a fixed observation window to obtain a standardized time-series temperature matrix. For example, when the system detects that a monitoring point inside the colloid is missing temperature records for 0.15 seconds and 0.25 seconds within the time period from 0.1 seconds to 0.3 seconds, a linear interpolation algorithm is used to calculate the temperature at 0.15 seconds as 41.4 degrees Celsius and the temperature at 0.25 seconds as 40.6 degrees Celsius, based on the values ​​of 41.8 degrees Celsius at 0.1 seconds and 40.2 degrees Celsius at 0.3 seconds. This data completion method ensures the continuity and accuracy of time-series analysis, providing a complete data foundation for subsequent difference operations.

[0035] Secondly, differential calculations identify abnormal heat conduction phenomena by calculating the temperature changes at adjacent time points, resulting in a spatial distribution record table containing the locations of temperature differences. Taking temperature monitoring at coordinates 7.2 μm, 4.8 μm, and 2.1 μm inside the colloid as an example, when the temperature at this point drops sharply from 43.6 degrees Celsius to 41.1 degrees Celsius within 0.05 seconds, the absolute value of the temperature difference reaches 2.5 degrees Celsius, exceeding the critical thermal stability parameter of 1.8 degrees Celsius for the optical adhesive material. The system automatically marks this location as an abnormal change point and records its three-dimensional coordinates and the magnitude of the change. This real-time monitoring mechanism can promptly capture abnormal heat conduction phenomena inside the colloid, providing crucial data support for defect identification.

[0036] Then, a moving average filter is used to smooth the temperature difference change sequences in each spatial region within the spatial distribution record table, and the average temperature change rate per unit time for each spatial region is calculated. When a 2 μm × 2 μm detection region contains 5 abnormal temperature difference points with original change rates of 3.2, 4.1, 2.8, 3.7, and 3.5 degrees Celsius per second, a smoothed sequence of 3.4, 3.5, and 3.3 degrees Celsius per second is obtained after a 3-point moving average filter. The system further calculates the average change rate for this region to be 3.4 degrees Celsius per second, and the maximum change rate to be 3.5 degrees Celsius per second. This data smoothing process effectively reduces the interference of measurement noise on the analysis results and improves the reliability of the thermal conductivity parameters.

[0037] Finally, the average temperature change rate is compared with a preset change rate threshold. Regions where the average temperature change rate exceeds the preset threshold are identified as candidate defect regions. In this method, the threshold comparison method filters and classifies conversion rate parameters by setting defect judgment criteria. When the average change rate of a region within the colloid reaches 4.2 degrees Celsius per second, exceeding the preset defect judgment criterion of 3.0 degrees Celsius per second, the system marks this region as a suspected defect location. Simultaneously, the system examines the adjacent regions within a 1.5-micron radius and finds that the adjacent locations also exhibit a high-rate change characteristic of 3.8 degrees Celsius per second. Based on the connectivity analysis principle, the system identifies the entire 4-micron × 3-micron connected region as a potential defect region, generating three-dimensional location information including starting coordinates of 5.2 microns, 3.1 microns, and 1.8 microns, and ending coordinates of 9.2 microns, 6.1 microns, and 1.8 microns. This spatial correlation analysis method can accurately locate the distribution range of defects within the colloid, providing precise location guidance for quality assessment and process optimization.

[0038] In step S104, for each location point in the candidate defect region, temperature gradient features, conversion rate features, and spatial distribution features within its surrounding neighborhood are extracted to construct a multi-dimensional feature vector. This multi-dimensional feature vector is then normalized using the colloid thickness data to generate a standard defect feature vector, including: A three-dimensional neighborhood search algorithm is used to establish a cubic neighborhood window at each location point in the candidate defect region to obtain a temperature distribution data set containing the neighborhood spatial structure. Based on the temperature distribution data set, the temperature change amplitude in each direction is calculated using a gradient operator to obtain the temperature gradient component values ​​in the lateral, longitudinal, and depth directions. Based on the temperature gradient component values, the average temperature change rate, and the spatial coordinate information of each location point, an initial multidimensional feature vector containing gradient features, rate features, and location features is constructed. Based on the colloid thickness data, a thickness ratio coefficient is determined, and the feature components in the initial multidimensional feature vector are weighted and adjusted according to the thickness ratio coefficient to obtain the adjusted multidimensional feature vector. The adjusted multidimensional feature vector is normalized and mapped to obtain the normalized feature values. The normalized feature values ​​are concatenated into vectors to obtain a standard defect feature vector that is combined in a predetermined order.

[0039] In this embodiment, the three-dimensional neighborhood search algorithm first acquires a set of temperature distribution data by constructing a cubic window around the coordinates of candidate defect points. When the system receives defect points with location coordinates of 8.5 μm, 5.2 μm, and 3.1 μm, the algorithm automatically establishes a cubic neighborhood window with a side length of 2.0 μm for temperature data acquisition. If the window boundary extends 0.3 μm beyond the edge of the colloidal material, the system detects boundary overstepping and automatically adjusts the window size in that direction to 1.7 μm to ensure that the sampling area is completely within the material. This adaptive adjustment mechanism effectively avoids the interference of boundary effects on temperature distribution analysis, ensuring the integrity and accuracy of the neighborhood data.

[0040] Subsequently, spatial variation features are extracted by performing directional analysis on the temperature distribution dataset within the neighborhood using gradient operators. Taking a 2μm × 2μm × 2μm neighborhood window as an example, the system detects a temperature change from 42.3 degrees Celsius to 39.8 degrees Celsius in the lateral direction, calculating a lateral gradient component of 1.25 degrees Celsius per micrometer. In the vertical direction, the temperature decreases from 41.5 degrees Celsius to 40.1 degrees Celsius, with a vertical gradient component of 0.7 degrees Celsius per micrometer. In the depth direction, the temperature increases from 42.1 degrees Celsius to 43.6 degrees Celsius, with a depth gradient component of 0.75 degrees Celsius per micrometer. Combining the previously obtained conversion rate parameter of 3.4 degrees Celsius per second and spatial coordinate information, the system constructs a 12-dimensional initial feature vector matrix containing 6 gradient features, 3 rate features, and 3 location features.

[0041] Then, the detection result of the actual thickness of the colloidal material directly affects the weight allocation of the feature vector. When the detection found that the thickness of the colloidal material at a defect location was 2.8 micrometers, exceeding the standard thickness range of 2.0 to 2.5 micrometers, the system calculated a thickness ratio coefficient of 1.12. Based on this coefficient, the temperature gradient component in the initial feature vector was adjusted, changing the original lateral gradient value of 1.25 to 1.4 and the longitudinal gradient value of 0.7 to 0.78. Subsequently, the maximum-minimum normalization method was used to map the adjusted feature values ​​to the standard range of 0 to 1, ensuring the comparability between features of different dimensions.

[0042] Finally, the vector concatenation operation combines various features into a unified defect feature vector according to a predetermined order of temperature gradient characteristics, conversion rate characteristics, and spatial distribution characteristics. The system arranges the three gradient components (0.85, 0.62, 0.71), the conversion rate feature (0.68), and the spatial coordinate features (0.42, 0.31, 0.55) sequentially to form a 7-dimensional standardized defect feature descriptor. This feature vector can comprehensively reflect the temperature distribution characteristics, dynamic change patterns, and spatial location information of defects inside the colloidal body, providing a standardized data foundation for subsequent defect classification and quality assessment, and significantly improving the accuracy and consistency of defect identification.

[0043] In step S105, similarity matching is performed based on the standard defect feature vector and the pre-established defect feature library to obtain preliminary defect classification results, including: Based on the standard defect feature vector, feature components of each dimension are extracted; Calculate the Euclidean distance between each feature component and the standard feature vector in the pre-established defect feature library; The Euclidean distance value is compared with a preset distance threshold. If the Euclidean distance value is less than the preset distance threshold, the feature vector corresponding to the Euclidean distance value is marked as a candidate match. Based on the candidate matching items, the cosine value of the angle between the feature vectors is calculated using cosine similarity. The final matching degree value is obtained by comprehensively scoring based on the weight coefficient of the cosine value in the defect feature database. After sorting the final matching scores in descending order, the best matching result with the highest matching score is obtained. Based on the best matching result, the matching value is converted into a confidence score using a confidence conversion function, and the corresponding defect type label is extracted from the defect feature library to determine the preliminary defect classification result.

[0044] In this embodiment, after the vector distance calculator receives the 7-dimensional standardized defect feature vector, it first performs numerical extraction operations on the feature components of each dimension. The system sequentially reads the temperature gradient components (0.85, 0.62, 0.71), the conversion rate feature (0.68), and the spatial coordinate features (0.42, 0.31, 0.55) from the input feature vector.

[0045] Subsequently, the Euclidean distance formula is used to calculate the spatial distance between each dimension of feature components and the standard feature templates in the pre-established defect feature library. The Euclidean distance value is compared with a preset distance threshold. If the Euclidean distance value is less than the preset distance threshold, the feature vector corresponding to the Euclidean distance value is marked as a candidate match. When the system calculates and finds that the distance value of a bubble defect template is 0.23, which is less than the preset similarity threshold of 0.35, the template is marked as a candidate match.

[0046] Subsequently, when the weighted similarity calculation method precisely measures each candidate match, the system uses a cosine similarity algorithm to calculate the cosine value of the angle between feature vectors. Taking the bubble defect candidate as an example, the calculated cosine similarity value is 0.87. Combined with the weight coefficient of 1.2 for this feature template in the defect feature database, a comprehensive score is calculated, resulting in a final matching degree of 0.91 for this candidate. Simultaneously, the system calculates a cosine similarity of 0.76 and a weight coefficient of 0.95 for the crack defect candidate, resulting in a comprehensive matching degree of 0.82.

[0047] Then, the matching scores of all candidate matches are sorted in descending order to obtain the best matching result with the highest matching score. Finally, based on the best matching result, the corresponding defect type label is extracted from the defect feature library to determine the preliminary defect classification result. Since the matching score exceeds the preset classification threshold of 0.80, the system extracts the corresponding defect type label information from the feature library and determines the specific classification category of the current defect as "internal bubble defect".

[0048] It should be noted that the confidence transformation function maps the matching score of 0.91 to the confidence scoring system. Through linear mapping, the confidence score corresponding to this matching score is 92.3. Since this score is higher than the preset confidence threshold of 85, the system marks the classification result as high confidence. The final complete classification report includes key information such as defect type "internal bubble defect", confidence score of 92.3, matching score of 0.91, and classification confidence level of "high".

[0049] In step S106, abnormal bubble regions are extracted based on the preliminary defect classification results. A decision tree classifier is then used to refine the classification of these abnormal bubble regions, generating detailed defect classification results, including: When the preliminary defect classification result includes internal bubble defects, the bubble position is three-dimensionally located to obtain the bubble position. When a bubble is located outside the preset center area, the location is marked as an abnormal bubble point, and the thickness is sampled based on the bubble location to obtain the colloid thickness distribution. The results are combined to obtain a first dataset containing the bubble location and thickness distribution. Temperature difference values ​​of abnormal bubble points are obtained by sampling the temperature of the abnormal bubble points. If the temperature difference value exceeds the preset temperature fluctuation threshold, the abnormal temperature area is marked, and the correspondence between abnormal temperature and structural defects is determined in the first dataset to obtain a second dataset containing the abnormal temperature area. The second dataset is further refined using a decision tree classifier to obtain detailed defect classification results. First, a spatial coordinate extractor uses laser scanning technology to precisely locate air bubbles inside the colloid. After establishing a three-dimensional coordinate system, the system spatially maps each detected bubble defect. Second, when the system detects an internal bubble defect, the spatial coordinate extractor first determines the center point coordinates of the bubble to be 15.2 μm on the X-axis, 8.7 μm on the Y-axis, and 3.4 μm on the Z-axis. A thickness gauge simultaneously scans this area point by point, measuring the thickness distribution around the bubble at 2.8 μm, 3.1 μm, and 2.9 μm. The system's preset center region range is 10 to 20 μm on the X-axis and 5 to 15 μm on the Y-axis. Since the bubble's coordinates are near the boundary, deviating from the ideal center position by more than 4.5 μm, it is marked as an abnormal bubble point. Then, a temperature sensor array uses infrared thermal imaging technology to monitor the temperature difference in the defect area in real time. After receiving the temperature data, the thermal imaging data processor calculates the temperature gradient distribution of the abnormal bubble area. Specifically, the system detected a temperature distribution around the bubble of 42.3 degrees Celsius in the core area and 38.7 degrees Celsius in the edge area, with a temperature difference of 3.6 degrees Celsius. When the temperature difference exceeded the preset fluctuation threshold of 2.5 degrees Celsius, the abnormal temperature area marking program was automatically activated, marking the area as a temperature difference abnormality point and establishing a direct correspondence between temperature abnormalities and bubble defects. Subsequently, a multi-dimensional feature fusion processor comprehensively quantified and evaluated the bubble position offset of 4.5 micrometers, the thickness non-uniformity coefficient of 0.23, and the temperature difference fluctuation amplitude of 3.6 degrees Celsius. After receiving the multi-dimensional feature vector containing position offset, thickness variation, and temperature anomaly, the decision tree classifier performed subtype identification through branch judgment logic. When the positional offset is greater than 4 micrometers and the temperature difference fluctuation exceeds 3 degrees Celsius, the system classifies the defect as an "edge-aggregated bubble defect". In contrast, if the positional offset is less than 2 micrometers but the thickness non-uniformity coefficient is greater than 0.3, it is classified as a "center-thickness bubble defect". This multi-dimensional comprehensive analysis method can achieve refined identification of defect characteristics, providing more accurate data support for subsequent quality control and process optimization.

[0050] In step S107, spatial clustering analysis is performed on the detailed defect classification results to obtain a defect proximity relationship dataset. Multiple consecutively distributed single defects in the defect proximity relationship dataset are merged into composite defect regions. The geometric parameters and severity rating of each defect region are calculated to obtain an optical adhesive quality inspection report, including: Based on the detailed defect classification results, the spatial proximity distance between each defect point is measured to obtain the spacing value between each adjacent defect. If the distance between two defect points is less than a preset proximity threshold, they are marked as an associated defect pair, and all associated defect pairs are counted to obtain a defect proximity relationship dataset. Based on the distribution continuity judgment of the associated defect pairs in the defect proximity relationship dataset, if three or more associated defect pairs form a continuous path in space, then the multiple continuously distributed associated defect pairs are merged into a unified composite defect region. Extract the geometric parameters of the composite defect region, including the normalized defect region area, shape complexity, and distribution density; A comprehensive score is calculated based on preset weights and the geometric parameters, and a severity rating is obtained by classifying the comprehensive score into levels. The defect location, geometric parameters, and severity rating of the composite defect area are output as an optical adhesive quality inspection report.

[0051] In this embodiment, the distance between defect points is first analyzed using spatial coordinate differences. Assuming the system detects two bubble defects with coordinates (12.5, 7.8, 4.2) and (13.0, 8.0, 4.5), the calculated Euclidean distance is approximately 0.6 micrometers. If the preset proximity threshold is 1.0 micrometer, the two defect points are marked as an associated defect pair because the distance is less than the threshold. The system generates a dataset containing the 0.6-micrometer distance and the association marker. This method quantifies spatial proximity, clearly identifying potential associations between defects and providing a data foundation for subsequent analysis. Then, the continuity detector uses a path tracing algorithm to determine the continuity of the defect distribution. Five defect points are detected, with coordinates (10.0, 5.0, 3.0), (10.5, 5.2, 3.1), (11.0, 5.5, 3.2), (11.5, 5.7, 3.3), and (12.0, 6.0, 3.4). Path tracing algorithm analysis revealed that these points are distributed along a straight path, with spacing less than 1.2 micrometers, forming a continuous distribution chain. The composite defect region marking program was initiated, merging these five points into a single composite defect region. The boundary data was recorded as a cuboid region, ranging from 10.0 to 12.0 micrometers along the X-axis, 5.0 to 6.0 micrometers along the Y-axis, and 3.0 to 3.4 micrometers along the Z-axis. This continuity analysis effectively identifies defect clusters, facilitating subsequent centralized processing. Subsequently, a geometric measuring instrument calculated geometric parameters based on the composite defect region boundary data. Assuming a composite defect region boundary of 10.0 to 12.5 micrometers along the X-axis, 6.0 to 8.0 micrometers along the Y-axis, and 3.0 to 3.5 micrometers along the Z-axis, the calculated length was 2.5 micrometers, the width was 2.0 micrometers, and the area was 5.0 square micrometers. The shape descriptor further extracted features, finding the region's shape to be approximately elliptical, with a shape complexity coefficient of 0.85. If the preset area limit is 4.0 square micrometers, this region is marked as a large-area defect type due to exceeding the area limit. This geometric feature extraction provides a quantitative basis for defect classification, enhancing analytical accuracy. Finally, the severity assessor comprehensively scores the geometric parameters. Assuming a defect area of ​​5.2 square micrometers, a shape complexity coefficient of 0.9, and a distribution density of 1.2 defect points per square centimeter, the system uses multi-parameter weighting for calculation, with weights of 0.4 for area, 0.3 for shape complexity, and 0.3 for distribution density, resulting in a comprehensive score of 0.92. If the severity defect threshold is 0.8, the area is rated as high severity. The final inspection report records the area's coordinates, geometric parameters, and severity rating. This multi-dimensional assessment method clearly distinguishes defect severity through quantitative indicators, providing a reliable basis for quality control.

[0052] Reference Figure 2 The second embodiment of the present invention provides a real-time quality defect detection system for OCA optical adhesive, comprising: The data acquisition module is used to acquire infrared thermal imaging data and colloid thickness data of optical adhesive samples, and to build a three-dimensional data to obtain a raw thermal imaging dataset containing complete spatial information. The temperature analysis module is used to solve the temperature values ​​of each point inside the colloid based on the original thermal imaging dataset to obtain temperature gradient field data, and to draw a temperature distribution model inside the colloid based on the temperature gradient field data. The region filtering module is used to calculate the average temperature change rate reflecting the heat conduction characteristics of each region from the temperature values ​​in the temperature distribution model, and to filter regions whose average temperature change rate exceeds a preset change rate threshold as candidate defect regions. The feature extraction module is used to extract temperature gradient features, conversion rate features and spatial distribution features in the surrounding neighborhood of each location point in the candidate defect region to construct a multi-dimensional feature vector, and to normalize the multi-dimensional feature vector by combining the colloidal thickness data to generate a standard defect feature vector. The defect matching module is used to perform similarity matching based on the standard defect feature vector and the pre-established defect feature library to obtain preliminary defect classification results; The detailed classification module is used to extract abnormal bubble regions based on the preliminary defect classification results, and to use a decision tree classifier to further classify the abnormal bubble regions to generate detailed defect classification results. The report generation module is used to perform spatial clustering analysis on the detailed defect classification results to obtain a defect proximity relationship dataset. Multiple single defects that are continuously distributed in the defect proximity relationship dataset are merged into a composite defect region. The geometric parameters and severity rating of each defect region are calculated to obtain an optical adhesive quality inspection report.

[0053] This invention also provides an electronic device. The electronic device includes a processor, a memory, and a computer program stored in the memory and executable on the processor, such as a real-time quality defect detection program for OCA optical adhesive. When the processor executes the computer program, it implements the steps in the various embodiments of the real-time quality defect detection method for OCA optical adhesive described above, for example... Figure 1 The step S101 shown. Alternatively, when the processor executes the computer program, it implements the functions of each module / unit in the above-described device embodiments, such as the defect matching module.

[0054] For example, the computer program may be divided into one or more modules / units, which are stored in the memory and executed by the processor to complete the present invention. The one or more modules / units may be a series of computer program instruction segments capable of performing a specific function, which describe the execution process of the computer program in the electronic device.

[0055] The electronic device may be a desktop computer, laptop, handheld computer, or smart tablet, etc. The electronic device may include, but is not limited to, a processor and memory. Those skilled in the art will understand that the above components are merely examples of electronic devices and do not constitute a limitation on the electronic device. It may include more or fewer components than described above, or combine certain components, or different components. For example, the electronic device may also include input / output devices, network access devices, buses, etc.

[0056] The processor can be a Central Processing Unit (CPU), or other general-purpose processors, digital signal processors (DSPs), application-specific integrated circuits (ASICs), field-programmable gate arrays (FPGAs), or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc. A general-purpose processor can be a microprocessor or any conventional processor. The processor is the control center of the electronic device, connecting all parts of the electronic device via various interfaces and lines.

[0057] The memory can be used to store the computer programs and / or modules. The processor implements various functions of the electronic device by running or executing the computer programs and / or modules stored in the memory and by calling data stored in the memory. The memory may mainly include a program storage area and a data storage area. The program storage area may store the operating system, at least one application program required for a function (such as sound playback function, image playback function, etc.), etc.; the data storage area may store data created according to the use of the mobile phone (such as audio data, phonebook, etc.). In addition, the memory may include high-speed random access memory, and may also include non-volatile memory, such as hard disk, memory, plug-in hard disk, smart media card (SMC), secure digital (SD) card, flash card, at least one disk storage device, flash memory device, or other volatile solid-state storage device.

[0058] Wherein, if the modules / units integrated in the electronic device are implemented as software functional units and sold or used as independent products, they can be stored in a computer-readable storage medium. Based on this understanding, all or part of the processes in the methods of the above embodiments of the present invention can also be implemented by a computer program instructing related hardware. The computer program can be stored in a computer-readable storage medium, and when executed by a processor, it can implement the steps of the various method embodiments described above. The computer program includes computer program code, which can be in the form of source code, object code, executable files, or certain intermediate forms. The computer-readable medium can include: any entity or device capable of carrying the computer program code, recording media, USB flash drives, portable hard drives, magnetic disks, optical disks, computer memory, read-only memory (ROM), random access memory (RAM), electrical carrier signals, telecommunication signals, and software distribution media, etc. It should be noted that the content included in the computer-readable medium can be appropriately added or removed according to the requirements of legislation and patent practice in the jurisdiction. For example, in some jurisdictions, according to legislation and patent practice, computer-readable media do not include electrical carrier signals and telecommunication signals.

[0059] 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 relationship between modules indicates that they have a communication connection, which can be implemented as one or more communication buses or signal lines. Those skilled in the art can understand and implement this without creative effort. The specific embodiments described above further illustrate the purpose, technical solution, and beneficial effects of this invention. It should be understood that the above descriptions are merely specific embodiments of this invention and are not intended to limit the scope of protection of this invention. In particular, for those skilled in the art, any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of this invention should be included within the scope of protection of this invention.

Claims

1. A real-time quality defect detection method for OCA optical adhesive, characterized in that, include: Infrared thermal imaging data and colloid thickness data of optical adhesive samples were acquired, and three-dimensional data were constructed to obtain a raw thermal imaging dataset containing complete spatial information. Temperature gradient field data is obtained by solving the temperature values ​​of each point inside the colloid based on the original thermal imaging dataset, and a temperature distribution model inside the colloid is drawn based on the temperature gradient field data. The average temperature change rate, which reflects the heat conduction characteristics of each region, is calculated from the temperature values ​​in the temperature distribution model. Regions with an average temperature change rate exceeding a preset change rate threshold are selected as candidate defect regions. For each location point in the candidate defect region, the temperature gradient features, conversion rate features, and spatial distribution features in its surrounding neighborhood are extracted to construct a multi-dimensional feature vector. The multi-dimensional feature vector is then normalized by combining the colloid thickness data to generate a standard defect feature vector. Based on the standard defect feature vector and the pre-established defect feature library, similarity matching is performed to obtain preliminary defect classification results; Based on the preliminary defect classification results, abnormal bubble regions are extracted, and a decision tree classifier is used to further classify the abnormal bubble regions to generate detailed defect classification results. Spatial clustering analysis is performed on the detailed defect classification results to obtain a defect proximity relationship dataset. Multiple single defects continuously distributed in the defect proximity relationship dataset are merged into a composite defect region. The geometric parameters and severity rating of each defect region are calculated to obtain an optical adhesive quality inspection report.

2. The real-time quality defect detection method for OCA optical adhesive according to claim 1, characterized in that, Temperature gradient field data is obtained by solving for the temperature values ​​at various points inside the colloid based on the original thermal imaging dataset, and a temperature distribution model inside the colloid is plotted based on the temperature gradient field data, including: Based on the original thermal imaging dataset, cubic spline interpolation is used to perform spatial interpolation on the temperature distribution of the colloidal surface and interior to obtain an initial temperature field distribution matrix covering the entire colloidal region. Based on the initial temperature field distribution matrix, the three-dimensional heat conduction partial differential equation is numerically discretized using the finite difference method to obtain the numerical solution of the temperature field that satisfies the thermal equilibrium condition. By performing spatial differentiation on the numerical solution of the temperature field using the gradient operator, temperature gradient field data reflecting the heat flow distribution characteristics inside the colloid are obtained. Based on the temperature gradient field data, a temperature distribution model inside the colloid is plotted using linear interpolation. The temperature distribution model includes an isothermal surface distribution map.

3. The real-time quality defect detection method for OCA optical adhesive according to claim 1, characterized in that, The average temperature change rate, reflecting the heat conduction characteristics of each region, is calculated from the temperature values ​​in the temperature distribution model. Regions with an average temperature change rate exceeding a preset threshold are selected as candidate defect regions, including: A continuous time series is obtained by sampling according to the temperature distribution model. The temperature values ​​in the continuous time series are segmented and extracted using a sliding window to obtain a standardized time series temperature matrix. By performing differential operations, the temperature values ​​of adjacent time points in the time-series temperature matrix are subtracted point by point to obtain a spatial distribution record table containing the locations of abnormal temperature differences. A moving average filter is used to smooth the temperature difference change sequence in each spatial region of the spatial distribution record table, and the average temperature change rate per unit time of each spatial region is calculated. The average temperature change rate is compared with a preset change rate threshold, and the region where the average temperature change rate is greater than the preset change rate threshold is determined as a candidate defect region.

4. The real-time quality defect detection method for OCA optical adhesive according to claim 1, characterized in that, For each location point in the candidate defect region, temperature gradient features, transformation rate features, and spatial distribution features within its surrounding neighborhood are extracted to construct a multi-dimensional feature vector. This multi-dimensional feature vector is then normalized using the colloid thickness data to generate a standard defect feature vector, including: A three-dimensional neighborhood search algorithm is used to establish a cubic neighborhood window at each location point in the candidate defect region to obtain a temperature distribution data set containing the neighborhood spatial structure. Based on the temperature distribution data set, the temperature change amplitude in each direction is calculated using a gradient operator to obtain the temperature gradient component values ​​in the lateral, longitudinal, and depth directions. Based on the temperature gradient component values, the average temperature change rate, and the spatial coordinate information of each location point, an initial multidimensional feature vector containing gradient features, rate features, and location features is constructed. Based on the colloid thickness data, a thickness ratio coefficient is determined, and the feature components in the initial multidimensional feature vector are weighted and adjusted according to the thickness ratio coefficient to obtain the adjusted multidimensional feature vector. The adjusted multidimensional feature vector is normalized and mapped to obtain the normalized feature values. The normalized feature values ​​are concatenated into vectors to obtain a standard defect feature vector that is combined in a predetermined order.

5. The real-time quality defect detection method for OCA optical adhesive according to claim 1, characterized in that, Based on the standard defect feature vector and the pre-established defect feature library, similarity matching is performed to obtain preliminary defect classification results, including: Based on the standard defect feature vector, feature components of each dimension are extracted; Calculate the Euclidean distance between each feature component and the standard feature vector in the pre-established defect feature library; The Euclidean distance value is compared with a preset distance threshold. If the Euclidean distance value is less than the preset distance threshold, the feature vector corresponding to the Euclidean distance value is marked as a candidate match. Based on the candidate matching items, the cosine value of the angle between the feature vectors is calculated using cosine similarity. The final matching degree value is obtained by comprehensively scoring based on the weight coefficient of the cosine value in the defect feature database. After sorting the final matching scores in descending order, the best matching result with the highest matching score is obtained. Based on the best matching result, the matching value is converted into a confidence score using a confidence conversion function, and the corresponding defect type label is extracted from the defect feature library to determine the preliminary defect classification result.

6. The real-time quality defect detection method for OCA optical adhesive according to claim 1, characterized in that, Based on the preliminary defect classification results, abnormal bubble regions are extracted. A decision tree classifier is then used to further classify these abnormal bubble regions, generating detailed defect classification results, including: When the preliminary defect classification result includes internal bubble defects, the bubble position is three-dimensionally located to obtain the bubble position. When a bubble is located outside the preset center area, the location is marked as an abnormal bubble point, and the thickness is sampled based on the bubble location to obtain the colloid thickness distribution. The results are combined to obtain a first dataset containing the bubble location and thickness distribution. Temperature difference values ​​of abnormal bubble points are obtained by sampling the temperature of the abnormal bubble points. If the temperature difference value exceeds the preset temperature fluctuation threshold, the abnormal temperature area is marked, and the correspondence between abnormal temperature and structural defects is determined in the first dataset to obtain a second dataset containing the abnormal temperature area. The second dataset is further refined using a decision tree classifier to obtain detailed defect classification results.

7. The real-time quality defect detection method for OCA optical adhesive according to claim 1, characterized in that, Spatial clustering analysis is performed on the detailed defect classification results to obtain a defect proximity relationship dataset. Multiple single defects continuously distributed in the defect proximity relationship dataset are merged into composite defect regions. The geometric parameters and severity rating of each defect region are calculated to obtain an optical adhesive quality inspection report, including: Based on the detailed defect classification results, the spatial proximity distance between each defect point is measured to obtain the spacing value between each adjacent defect. If the distance between two defect points is less than a preset proximity threshold, they are marked as an associated defect pair, and all associated defect pairs are counted to obtain a defect proximity relationship dataset. Based on the distribution continuity judgment of the associated defect pairs in the defect proximity relationship dataset, if three or more associated defect pairs form a continuous path in space, then the multiple continuously distributed associated defect pairs are merged into a unified composite defect region. Extract the geometric parameters of the composite defect region, including the normalized defect region area, shape complexity, and distribution density; A comprehensive score is calculated based on preset weights and the geometric parameters, and a severity rating is obtained by classifying the comprehensive score into levels. The defect location, geometric parameters, and severity rating of the composite defect area are output as an optical adhesive quality inspection report.

8. A real-time quality defect detection system for OCA optical adhesive, characterized in that, include: The data acquisition module is used to acquire infrared thermal imaging data and colloid thickness data of optical adhesive samples, and to build a three-dimensional data to obtain a raw thermal imaging dataset containing complete spatial information. The temperature analysis module is used to solve the temperature values ​​of each point inside the colloid based on the original thermal imaging dataset to obtain temperature gradient field data, and to draw a temperature distribution model inside the colloid based on the temperature gradient field data. The region filtering module is used to calculate the average temperature change rate reflecting the heat conduction characteristics of each region from the temperature values ​​in the temperature distribution model, and to filter regions whose average temperature change rate exceeds a preset change rate threshold as candidate defect regions. The feature extraction module is used to extract temperature gradient features, conversion rate features and spatial distribution features in the surrounding neighborhood of each location point in the candidate defect region to construct a multi-dimensional feature vector, and to normalize the multi-dimensional feature vector by combining the colloidal thickness data to generate a standard defect feature vector. The defect matching module is used to perform similarity matching based on the standard defect feature vector and the pre-established defect feature library to obtain preliminary defect classification results; The detailed classification module is used to extract abnormal bubble regions based on the preliminary defect classification results, and to use a decision tree classifier to further classify the abnormal bubble regions to generate detailed defect classification results. The report generation module is used to perform spatial clustering analysis on the detailed defect classification results to obtain a defect proximity relationship dataset. Multiple single defects that are continuously distributed in the defect proximity relationship dataset are merged into a composite defect region. The geometric parameters and severity rating of each defect region are calculated to obtain an optical adhesive quality inspection report.

9. An electronic device, characterized in that, The method includes a processor, a memory, and a computer program stored in the memory and configured to be executed by the processor, wherein the processor, when executing the computer program, implements the real-time quality defect detection method for OCA optical adhesive as described in any one of claims 1 to 7.

10. A computer-readable storage medium, characterized in that, The computer-readable storage medium includes a stored computer program, wherein, when the computer program is executed, it controls the device on which the computer-readable storage medium is located to perform the real-time quality defect detection method for OCA optical adhesive as described in any one of claims 1 to 7.

Images