Method for automatically identifying defects of glass substrate of liquid crystal screen module

Abstract

The invention belongs to the technical field of glass substrate detection, and discloses a method for automatically identifying defects of a glass substrate of a liquid crystal screen module. Comprising the following steps: S1, collecting defect data of the glass substrate in real time by detection equipment of a detection assembly line, and carrying out normalization processing on the data; according to the method, the upgrading coefficient prediction model and the degradation coefficient prediction model are set, the models accurately quantify the real-time performance levels of each detection node and each verification node, and the system periodically and automatically executes role conversion based on the output of the models; higher quality responsibility is endowed; and meanwhile, the verification assembly line with lower performance and larger deviation is degraded into a detection assembly line or a triggered maintenance alarm, so that the verification assembly line is ensured to be always composed of the current most excellent equipment, the verification reliability is maintained, the on-demand and dynamic allocation of assembly line resources is realized, and limited high-quality resources are used for the most critical verification link.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of glass substrate detection, and more particularly to a liquid crystal screen module glass substrate defect automatic identification method. BACKGROUND

[0002] With the rapid development of display technology, the flatness, purity and internal structure quality of the glass substrate of the liquid crystal screen module are very high, and micron-level point defects, scratches, bubbles or internal stress unevenness can cause problems such as bright and dark spots, unevenness in the final display panel, resulting in huge economic losses. Therefore, automatic defect detection of 100% of the glass substrate during the manufacturing process has become an indispensable key link.

[0003] The current automatic optical detection system mainly relies on high-resolution visible light cameras to collect substrate images, and uses complex image processing algorithms to identify and classify defects. However, the performance of the detection equipment itself fluctuates and lacks a correction mechanism. Different detection equipment has differences in detection performance due to factors such as factory calibration, use wear, environmental changes, and will drift over time. The existing system usually regards each device as an independent and static detection unit, and lacks online evaluation and mutual verification mechanism for the reliability of the detection results of the equipment itself. The resource allocation of the current detection system is rigid, and the roles of all detection nodes in the production line are fixed, and cannot be dynamically adjusted according to real-time production data, equipment status and quality requirements. The detection equipment with excellent performance cannot undertake higher-level review tasks, and the performance-declined equipment is still performing critical detection, resulting in the inability to continuously optimize the efficiency and accuracy of the overall detection system. SUMMARY

[0004] The present application is a liquid crystal screen module glass substrate defect automatic identification method to solve the problems in the background art.

[0005] To achieve the above purpose, the present application provides the following technical scheme: a liquid crystal screen module glass substrate defect automatic identification method, comprising the following steps: S1: The detection equipment of the detection pipeline collects the defect data of the glass substrate in real time, and performs normalization processing on the data; S2: Generate a defect score for each glass substrate according to the normalized data, grade each glass substrate according to the defect score, and sort them into different areas; S3: Generate an input score for each verification pipeline according to the input data of the verification pipeline, and input the high-level defect glass substrate and the medium-level defect glass substrate determined by the detection pipeline to the corresponding verification pipeline according to the input score, and the verification pipeline re-evaluates the defects of the input glass substrate; S4: Calculate the deviation rate according to the multiple defect scores of the glass substrate, and determine the final grade of the glass substrate according to the deviation rate; S5: Obtain historical upgrade coefficient data and construct an upgrade coefficient prediction model, predict the upgrade coefficient of the detection pipeline using the upgrade coefficient prediction model, set an upgrade threshold, and decide whether the detection pipeline is upgraded according to the upgrade threshold and the upgrade coefficient of the detection pipeline; S6: Obtain historical downgrade coefficient data and construct a downgrade coefficient prediction model, predict the downgrade coefficient of the verification pipeline using the downgrade coefficient prediction model, set a downgrade threshold, and decide whether the verification pipeline is downgraded according to the downgrade threshold and the downgrade coefficient of the verification pipeline.

[0006] Further, a plurality of pipelines are provided in the glass substrate defect detection workshop, the pipelines include a plurality of detection pipelines and a plurality of verification pipelines, and the verification pipelines account for thirty percent of all pipelines. The detection pipeline is used to detect and identify defects of the glass substrate, and the verification pipeline is used to detect the high-grade defect glass substrate and the medium-grade defect glass substrate determined by the detection pipeline again. The defect data includes surface gray variance, thermal distribution gradient, fluorescence intensity peak, edge sharpness, and color consistency of the glass substrate. The surface gray variance is obtained by acquiring the surface image of the glass substrate through the visible light camera and calculating the variance value of the pixel gray of the entire area using an image processing algorithm.

[0007] In the formula, is the normalized surface gray variance, is the measured surface gray variance, is the historical maximum variance; The thermal distribution gradient is obtained by scanning the glass substrate through the infrared thermal imager and calculating the average gradient amplitude of the temperature field.

[0008] In the formula, is the normalized thermal distribution gradient, is the measured thermal distribution gradient, is the maximum gradient; The fluorescence intensity peak is obtained by exciting and collecting the glass substrate image through the ultraviolet fluorescence imager, and the maximum fluorescence intensity value of all pixel points in the image is counted.

[0009] In the formula, is the normalized fluorescence intensity peak, is a measured fluorescence intensity peak value, is a maximum fluorescence intensity peak value; An edge image of the glass substrate is captured by a visible light camera, and the image definition of the edge region is calculated to obtain the edge sharpness, and the edge sharpness is normalized:

[0010] In the formula, is a normalized edge sharpness, is a measured edge sharpness, is a maximum edge sharpness; An RGB image is collected by a visible light camera, and the standard deviations of R, G and B channels in the effective area of the substrate are calculated respectively, and then the average value is obtained to obtain the color consistency, and the color consistency is normalized:

[0011] In the formula, is a normalized color consistency value, is a measured edge color consistency value, is a maximum color consistency value.

[0012] Further, the process of generating a defect score of each glass substrate according to the normalized data includes: The defect score S of the glass substrate is:

[0013] In the formula, , , and are weight coefficients, which are trained according to historical data.

[0014] Further, the process of grading each glass substrate according to the defect score includes: According to historical defect score data, a suitable first defect score threshold and a second defect score threshold are set, the historical defect score data refers to a data set of past glass substrate defect scores, the defect score of the glass substrate is compared with the first defect score threshold and the second defect score threshold, and the glass substrate is divided into glass substrates of different defect levels; When the defect score of the glass substrate is less than or equal to the first defect score threshold, the glass substrate is determined as a low defect glass substrate; When the first defect score threshold is less than the defect score of the glass substrate and less than the second defect score threshold, the glass substrate is determined as a medium defect glass substrate; When the second defect score threshold is less than or equal to the defect score of the glass substrate, the glass substrate is determined as a high defect glass substrate. A defect level label is assigned to the glass substrate, which will be bound with the unique identification code of the glass substrate. When the glass substrate with the unique identification code enters the sorting area of the flow line, the code reader reads its information, and according to the defect level bound therewith, sends a sorting instruction containing the target flow direction to the sorting manipulator. The end of the sorting area is usually provided with a plurality of storage areas corresponding to glass substrates of different defect levels. The sorting area sorts the glass substrate to the corresponding storage area based on the received sorting instruction.

[0015] Further, the process of generating the input score of each verification flow line according to the input data of the verification flow line includes: The input data includes transportation distance, verification flow line load, historical accuracy rate and average processing time; The transportation distance is obtained by measuring the distance between the detection flow line and the verification flow line, and the transportation distance is normalized:

[0016] In the formula, is the normalized transportation distance, is the measured transportation distance, is the maximum transportation distance; The verification flow line load refers to the real-time glass substrate arrangement waiting quantity of the verification flow line, and the verification flow line load is normalized:

[0017] In the formula, is the normalized verification flow line load, is the measured verification flow line load, is the maximum verification flow line load; The historical accuracy rate refers to the historical verification accuracy proportion of the verification flow line, that is, the correctness rate of the determination result of the defect level of the glass substrate. The historical accuracy rate is obtained by dividing the number of glass substrates with incorrect determination by the total number of verified glass substrates, and the historical accuracy rate is normalized:

[0018] In the formula, is the normalized historical accuracy rate, is the measured historical accuracy rate, is the maximum historical accuracy rate; The average processing time refers to the time consumed by the verification flow line to verify a single glass substrate, and the processing time is normalized:

[0019] In the formula, the normalized average processing time, the measured average processing time, the maximum average processing time; the input score P:

[0020] wherein, 、 、 is a weight coefficient, which is trained according to historical data.

[0021] Further, the senior and intermediate defect glass substrates determined by the detection line according to the input score are input to the corresponding verification line, and the process of the verification line for re-scoring the input glass substrate includes: arranging the senior and intermediate defect glass substrates in the detection line storage area in order from high to low according to the input score of each verification line corresponding to the detection line, and inputting the senior and intermediate defect glass substrates in the detection line storage area to the verification line corresponding to the highest input score of the detection line; the verification line re-scoring the input glass substrate to obtain a second defect score of the glass substrate, and re-classifying the glass substrate according to the second defect score, the first defect score threshold and the second defect score threshold, and sorting the glass substrate into the corresponding storage area of the verification line.

[0022] Further, the deviation rate is calculated according to the multiple defect scores of the glass substrate, and the process of determining the final grade of the glass substrate according to the deviation rate includes: setting the defect score generated by the detection line as the first defect score, setting the defect score generated by the verification line as the second defect score, obtaining the score difference by subtracting the second defect score from the first defect score, obtaining the deviation rate of the two scores by dividing the absolute value of the score difference by the maximum value of the two scores, and setting the deviation rate as the first deviation rate; setting a suitable deviation rate threshold according to historical deviation rate data, wherein the historical deviation rate data refers to a data set of the deviation rate of the two defect scores of the glass substrate in the past, and comparing the first deviation rate of the glass substrate with the deviation rate threshold; when the first deviation rate is less than the deviation rate threshold, the detection results of the glass substrate by the detection line and the verification line are both correct, and the glass substrate needs to be repaired; when the first deviation rate is greater than the deviation rate threshold, the glass substrate is input to other verification lines for re-scoring to obtain a third defect score, and the deviation rate of the first defect score and the third defect score is calculated to obtain a second deviation rate, the deviation rate of the second defect score and the third defect score is calculated to obtain a third deviation rate, and the second deviation rate and the third deviation rate are compared with the deviation rate threshold. If the second deviation rate is less than the deviation rate threshold and the third deviation rate is greater than or equal to the deviation rate threshold, the detection process is error-free, the first verification process is error-free, and the defect level of the glass substrate is maintained according to the second deviation rate, which is the judgment result of the detection process. If the second deviation rate is greater than or equal to the deviation rate threshold and the third deviation rate is less than the deviation rate threshold, then the detection pipeline is faulty, the first verification pipeline is error-free, and the defect level of the glass substrate remains the same as the judgment result of the first verification pipeline based on the third deviation rate. If both the second and third deviation rates are greater than or equal to the deviation rate threshold, it cannot be determined whether the testing line and the first verification line have made a detection error. The glass substrate needs to be fed into other verification lines, and the deviation rate should be obtained by following the above steps. Then, it should be verified whether the testing line, the first verification line, and the second verification line have made a detection error.

[0023] Furthermore, the process of acquiring historical upgrade coefficient data and constructing an upgrade coefficient prediction model, and then using this model to predict the upgrade coefficient of the inspection pipeline, includes: The upgrade coefficient refers to the probability coefficient of upgrading the detection pipeline to a verification pipeline. The overall stability of the detection results of the verification pipeline is better than that of the detection pipeline. Factors affecting the upgrade coefficient of the testing pipeline include: average deviation rate, accuracy rate, equipment age, maintenance score, and number of historical upgrades; The average deviation rate refers to the average deviation rate of the inspection process for high-grade defect glass substrates and medium-grade defect glass substrates. Accuracy rate refers to the ratio of the number of accurate defect level determination results for glass substrates by the inspection line to the total number of glass substrates. Equipment age refers to the length of time the equipment has been used; Maintenance rating refers to a rating based on the quality of equipment maintenance, derived from maintenance logs and records; Historical upgrade count refers to the number of times this testing line has been upgraded in the past; Obtain historical upgrade coefficient data for a single testing pipeline. The historical upgrade coefficient data includes the average deviation rate, accuracy rate, equipment age, maintenance score, number of historical upgrades, and historical upgrade coefficient of the single testing pipeline. Based on the average deviation rate, accuracy rate, equipment age, maintenance score, number of historical upgrades and corresponding historical upgrade coefficients of a single detection pipeline in different historical upgrade coefficient data, an upgrade coefficient prediction set is generated and divided into the first training set and the first test set. Construct a first convolutional neural network, using the average deviation rate, accuracy rate, equipment age, maintenance score, and number of historical upgrades from different historical upgrade coefficient data in the first training set as the input data of the first convolutional neural network, and using the corresponding historical upgrade coefficients in the first training set as the output data of the first convolutional neural network; The first convolutional neural network is trained to obtain the first initial convolutional neural network. The first initial convolutional neural network is validated using the first test set. The first initial convolutional neural network that outputs a value less than or equal to the preset first test error threshold is used as the upgrade coefficient prediction model. The average deviation rate, accuracy rate, equipment age, maintenance score, and historical upgrade count of all testing lines are input into the upgrade coefficient prediction model to obtain the predicted upgrade coefficients for all testing lines.

[0024] Furthermore, the process of acquiring historical degradation coefficient data and constructing a degradation coefficient prediction model, and then using this model to predict and verify the degradation coefficient of the pipeline, includes: The degradation coefficient refers to the probability coefficient of a verification pipeline being downgraded to a testing pipeline. Factors affecting the degradation factor of the validation pipeline include: accuracy, cross-bias rate, equipment age, maintenance score, and number of historical degradations; Accuracy rate refers to the ratio of the number of accurate defect level determination results of the verification line for glass substrates to the total number of glass substrates. Cross-bias ratio refers to the average deviation rate between the verification pipeline and other verification pipelines when scoring the same glass substrate; Equipment age refers to the length of time the equipment has been used; Maintenance rating refers to a rating based on the quality of equipment maintenance, derived from maintenance logs and records; Historical downgrade count refers to the number of times this verification pipeline has been downgraded in the past; Obtain historical degradation coefficient data for a single validation pipeline. The historical degradation coefficient data includes the accuracy, cross-bias rate, equipment age, maintenance score, number of historical degradations, and historical degradation coefficient of the single validation pipeline. Based on the accuracy, cross-bias rate, equipment age, maintenance score, number of historical degradations and corresponding historical degradation coefficients of a single validation pipeline in different historical degradation coefficient data, a degradation coefficient prediction set is generated and divided into a second training set and a second test set. Construct a second convolutional neural network, using the accuracy, cross-bias ratio, equipment age, maintenance score, and number of historical degradations from different historical degradation coefficients in the second training set as the input data of the second convolutional neural network, and using the corresponding historical degradation coefficients in the second training set as the output data of the second convolutional neural network; The second convolutional neural network is trained to obtain the second initial convolutional neural network. The second initial convolutional neural network is validated using the second test set. The second initial convolutional neural network that outputs a second test error threshold less than or equal to the preset second test error threshold is used as the degradation coefficient prediction model. The accuracy, cross-bias ratio, equipment age, maintenance score, and historical degradation count of all validation pipelines are input into the degradation coefficient prediction model to obtain the predicted degradation coefficients for all validation pipelines.

[0025] Furthermore, an appropriate upgrade coefficient threshold is set based on historical upgrade coefficient data, where the historical upgrade coefficient data refers to the data set of upgrade coefficients of previous detection pipelines, and the upgrade coefficients of all detection pipelines are compared with the upgrade coefficient threshold. The detection pipelines with upgrade coefficients greater than the upgrade coefficient threshold are arranged in descending order of upgrade coefficient, and the top 20% of X detection pipelines are selected to be upgraded into validation pipelines. Set an appropriate degradation coefficient threshold based on historical degradation coefficient data. The historical degradation coefficient data refers to the data set of degradation coefficients of previous verification pipelines. Compare the degradation coefficients of all verification pipelines with the degradation coefficient threshold. The validation pipelines are arranged in descending order of their degradation coefficients. The first X validation pipelines are selected, and it is determined whether the degradation coefficient of the Xth validation pipeline is greater than the degradation coefficient threshold. If the degradation coefficient is greater than the threshold, the first X validation pipelines are downgraded to detection pipelines and maintained in a timely manner. If the degradation coefficient of the Xth validation pipeline is less than the threshold, only the first Y validation pipelines with degradation coefficients greater than the threshold need to be downgraded to detection pipelines and maintained in a timely manner. Only the first Y detection pipelines can be upgraded to validation pipelines to ensure that the ratio of detection pipelines to validation pipelines remains unchanged.

[0026] The technical effects and advantages of the automatic defect identification method for glass substrate of LCD screen module of the present invention are as follows: (1) By setting up an upgrade coefficient prediction model and a downgrade coefficient prediction model, the model accurately quantifies the real-time performance level of each detection node and verification node. Based on the output of the model, the system automatically performs role switching on a regular basis: upgrading the high-performance, stable and reliable detection pipeline to the verification pipeline and giving it greater quality responsibility; at the same time, downgrading the verification pipeline with declining performance and large deviation to the detection pipeline or triggering maintenance alarms, ensuring that the verification pipeline is always composed of the best equipment at present, maintaining the reliability of verification, realizing the on-demand and dynamic allocation of pipeline resources, so that limited high-quality resources are used for the most critical verification links, which not only significantly improves the quality control level of the entire production line, but also reduces the quality risks and maintenance costs caused by the uneven performance of equipment in the long run, and achieves the synergistic growth of quality and efficiency.

[0027] (2) By introducing a verification pipeline and performing mandatory secondary scoring on substrates with defects initially judged to be of high or medium grade, cross-verification of test results is achieved. By setting a deviation rate, the consistency of the two test results can be quantified fairly and robustly. When the deviation rate is below the threshold, the system is certain that the test is correct and can proceed with subsequent processing. When the deviation rate exceeds the limit, the system does not simply report an error, but automatically initiates an arbitration process, sending the glass substrate to a third-party verification pipeline for a third test. By calculating multiple sets of deviation rates, it determines whether the responsible party made a mistake in the initial test or a mistake in the retest, effectively identifying and eliminating accidental errors caused by equipment fluctuations, environmental interference, or algorithm misjudgment, and intercepting and correcting the unavoidable false detections and missed detections in traditional systems from the source. Attached Figure Description

[0028] Figure 1 This is a schematic diagram of the system of the present invention. Detailed Implementation

[0029] 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 of ordinary skill in the art without creative effort are within the scope of protection of the present invention.

[0030] Reference Figure 1 An automatic defect identification method for glass substrate of LCD screen module includes the following steps: S1: The inspection equipment in the inspection line collects defect data of the glass substrate in real time and performs normalization processing on the data; S2: Generate a defect score for each glass substrate based on the normalized data, classify each glass substrate according to the defect score, and sort them into different areas. S3: Generate an input score for each verification pipeline based on the input data of the verification pipeline. Based on the input score, input the high-defect glass substrates and medium-defect glass substrates determined by the detection pipeline into the corresponding verification pipeline. The verification pipeline performs a second defect score on the input glass substrates. S4: The deviation rate is calculated based on multiple defect scores of the glass substrate, and the final grade of the glass substrate is determined based on the deviation rate. S5: Obtain historical upgrade coefficient data and build an upgrade coefficient prediction model. Use the upgrade coefficient prediction model to predict the upgrade coefficient of the detection pipeline, set the upgrade threshold, and decide whether to upgrade the detection pipeline based on the upgrade threshold and the upgrade coefficient of the detection pipeline. S6: Obtain historical degradation coefficient data and build a degradation coefficient prediction model. Use the degradation coefficient prediction model to predict the degradation coefficient of the verification pipeline, set a degradation threshold, and decide whether to degrade the verification pipeline based on the degradation threshold and the degradation coefficient of the verification pipeline.

[0031] It should be further explained that, in the specific implementation process, the glass substrate defect inspection workshop is equipped with multiple production lines, including multiple inspection production lines and multiple verification production lines. The verification production lines account for 30% of all production lines. The inspection production lines are used to detect and identify defects in the glass substrates, and the verification production lines are used to re-inspect the high-level defect glass substrates and medium-level defect glass substrates identified by the inspection production lines. The defect data includes the surface grayscale variance of the glass substrate, thermal distribution gradient, fluorescence intensity peak, edge sharpness, and color consistency. Images of the glass substrate surface are captured using a visible light camera. Image processing algorithms are used to calculate the variance of pixel gray levels across the entire area to obtain the surface gray level variance. This surface gray level variance is then normalized.

[0032] In the formula, To normalize the surface grayscale variance, This represents the measured surface grayscale variance. This represents the historical maximum variance, specifically 1000. The temperature distribution map of the glass substrate is obtained by scanning it with an infrared thermal imager. The average gradient magnitude of the temperature field is calculated to obtain the thermal distribution gradient, which is then normalized.

[0033] In the formula, For the normalized heat distribution gradient, This is the measured heat distribution gradient. The maximum gradient is 10℃ / mm; Images of the glass substrate are excited and acquired using an ultraviolet fluorescence imager. The peak fluorescence intensity is obtained by counting the maximum fluorescence intensity value of all pixels in the image, and then the peak fluorescence intensity is normalized.

[0034] In the formula, The normalized fluorescence intensity peak value is... This represents the measured peak fluorescence intensity. The maximum fluorescence intensity peak value is 255; The edge image of the glass substrate is captured by a visible light camera, and the edge sharpness is calculated by measuring the image clarity of the edge region. The edge sharpness is then normalized.

[0035] In the formula, To normalize edge sharpness, This represents the measured edge sharpness. This represents the maximum edge sharpness, specifically 100. RGB images are acquired using a visible light camera. The standard deviations of the R, G, and B channels within the effective area of ​​the substrate are calculated, and their average values ​​are then used to obtain color consistency. Color consistency is then normalized.

[0036] In the formula, This is a normalized color consistency value. This represents the measured value for consistent edge color. This is the maximum color consistency value, specifically 50.

[0037] It should be further explained that, in the specific implementation process, the process of generating a defect score for each glass substrate based on the normalized data includes: Defect score S of glass substrate:

[0038] In the formula, , , and These are weighting coefficients, obtained from training on historical data, and set to 0.25, 0.2, 0.2, 0.15, and 0.2 respectively. If the defect data of a certain glass substrate is: It is 500. It is 5. It is 100. It is 50. If the value is 25, then the defect score S of the glass substrate is 0.478.

[0039] It should be further explained that, in the specific implementation process, the process of grading each glass substrate according to the defect score includes: Based on historical defect scoring data, set appropriate first and second defect scoring thresholds. The historical defect scoring data refers to the data set of previous glass substrate defect scores. Compare the defect scores of the glass substrate with the first and second defect scoring thresholds to classify the glass substrates into glass substrates with different defect levels. If the defect score of the glass substrate is less than or equal to the first defect score threshold, the glass substrate is judged to be a low-level defect glass substrate. If the first defect score threshold is less than the glass substrate defect score and the second defect score threshold, the glass substrate is judged to be a medium-level defect glass substrate. If the second defect score threshold is less than or equal to the defect score of the glass substrate, the glass substrate is judged to be a high-defect glass substrate. A defect level label is assigned to each glass substrate, and this defect level label is bound to the unique identification code of the glass substrate. When a glass substrate with a unique identification code enters the sorting area of ​​the production line, the barcode scanner reads its information and sends a sorting instruction containing the target flow direction to the sorting robot according to the defect level it is bound to. At the end of the sorting area, there are usually multiple storage areas, each corresponding to a glass substrate with a different defect level. Based on the received sorting instructions, the sorting area sorts the glass substrates to the corresponding storage areas.

[0040] It should be further explained that, in the specific implementation process, the process of generating an input score for each validation pipeline based on the input data of the validation pipeline includes: The input data includes transportation distance, verification pipeline load, historical accuracy, and average processing time. The transport distance is obtained by measuring the distance between the testing line and the verification line, and then normalized.

[0041] In the formula, To normalize transport distance, This is the measured transportation distance. The maximum transport distance is 100m. Verification pipeline load refers to the number of glass substrates waiting to be arranged in the verification pipeline in real time. The verification pipeline load is then normalized.

[0042] In the formula, To normalize the verification pipeline load, To verify the pipeline load through actual testing, To maximize the verification pipeline load, specifically 50 units; Historical accuracy refers to the percentage of accurate verifications in the history of the verification pipeline, that is, the correctness of the judgment results on the defect level of the glass substrate. The historical accuracy is obtained by dividing the number of glass substrates with incorrect judgments by the total number of glass substrates verified. The historical accuracy is then normalized.

[0043] In the formula, To normalize historical accuracy, This represents the historical accuracy rate as measured. This represents the maximum historical accuracy, specifically 100%. Average processing time refers to the time spent processing a single glass substrate in the verification pipeline. The processing time is normalized as follows:

[0044] In the formula, To normalize the average processing time, This represents the measured average processing time. This represents the maximum average processing time, specifically 60 seconds. Input rating P:

[0045] In the formula, , , These are weighting coefficients, derived from historical data training, and set to 0.25, 0.25, 0.35, and 0.15 respectively. If the input data for a certain verification pipeline relative to a certain testing pipeline is: It is 20m. There are 10. 95%, If the time is 30 seconds, then the input score P of the verification pipeline is equivalent to that of the detection pipeline, which is 0.8075.

[0046] It should be further explained that, in the specific implementation process, based on the input score, the high-grade defect glass substrates and medium-grade defect glass substrates determined by the inspection line are input into the corresponding verification line. The verification line performs a second defect score on the input glass substrates, which includes: Based on the input scores of each verification pipeline equivalent to the inspection pipeline from high to low, the high-defect glass substrates and medium-defect glass substrates in the storage area of ​​the inspection pipeline are put into the verification pipeline with the highest input score of that inspection pipeline. The verification line performs a second defect score on the incoming glass substrates to obtain a second defect score. Based on the second defect score, the first defect score threshold, and the second defect score threshold, the glass substrates are graded again and sorted into the corresponding storage area of ​​the verification line.

[0047] It should be further explained that, in the specific implementation process, the deviation rate is calculated based on multiple defect scores of the glass substrate, and the process of determining the final grade of the glass substrate based on the deviation rate includes: The defect score generated by the inspection line is set as the first defect score, and the defect score generated by the verification line is set as the second defect score. The first defect score is subtracted from the second defect score to obtain the score difference. The absolute value of the score difference is divided by the maximum value of the two scores to obtain the deviation rate between the two scores. This deviation rate is set as the first deviation rate. Set an appropriate deviation rate threshold based on historical deviation rate data. The historical deviation rate data refers to the data set of deviation rates of two previous defect scores of the glass substrate. Compare the first deviation rate of the glass substrate with the deviation rate threshold. If the first deviation rate is less than the deviation rate threshold, then the test results of both the detection line and the verification line for the glass substrate are correct, and the glass substrate needs to be reworked. If the first deviation rate is greater than the deviation rate threshold, the glass substrate is put into another verification pipeline for defect scoring again to obtain a third defect score. The deviation rate between the first and third defect scores is calculated to obtain the second deviation rate. The deviation rate between the second and third defect scores is calculated to obtain the third deviation rate. The second and third deviation rates are compared with the deviation rate threshold. If the second deviation rate is less than the deviation rate threshold and the third deviation rate is greater than or equal to the deviation rate threshold, the detection process is error-free, the first verification process is error-free, and the defect level of the glass substrate is maintained according to the second deviation rate, which is the judgment result of the detection process. If the second deviation rate is greater than or equal to the deviation rate threshold and the third deviation rate is less than the deviation rate threshold, then the detection pipeline is faulty, the first verification pipeline is error-free, and the defect level of the glass substrate remains the same as the judgment result of the first verification pipeline based on the third deviation rate. If both the second and third deviation rates are greater than or equal to the deviation rate threshold, it cannot be determined whether the testing line and the first verification line have made a detection error. The glass substrate needs to be fed into other verification lines, and the deviation rate should be obtained by following the above steps. Then, it should be verified whether the testing line, the first verification line, and the second verification line have made a detection error.

[0048] It should be further explained that, in the specific implementation process, the process of acquiring historical upgrade coefficient data and constructing an upgrade coefficient prediction model, and then using the upgrade coefficient prediction model to predict the upgrade coefficient of the inspection pipeline, includes: The upgrade coefficient refers to the probability coefficient of upgrading the detection pipeline to a verification pipeline. The overall stability of the detection results of the verification pipeline is better than that of the detection pipeline. Factors affecting the upgrade coefficient of the testing pipeline include: average deviation rate, accuracy rate, equipment age, maintenance score, and number of historical upgrades; The average deviation rate refers to the average deviation rate of the inspection process for high-grade defect glass substrates and medium-grade defect glass substrates. Accuracy rate refers to the ratio of the number of accurate defect level determination results for glass substrates by the inspection line to the total number of glass substrates. Equipment age refers to the length of time the equipment has been used; Maintenance rating refers to a rating based on the quality of equipment maintenance, derived from maintenance logs and records; Historical upgrade count refers to the number of times this testing line has been upgraded in the past; Obtain historical upgrade coefficient data for a single testing pipeline. The historical upgrade coefficient data includes the average deviation rate, accuracy rate, equipment age, maintenance score, number of historical upgrades, and historical upgrade coefficient of the single testing pipeline. Based on the average deviation rate, accuracy rate, equipment age, maintenance score, number of historical upgrades and corresponding historical upgrade coefficients of a single detection pipeline in different historical upgrade coefficient data, an upgrade coefficient prediction set is generated and divided into the first training set and the first test set. Construct a first convolutional neural network, using the average deviation rate, accuracy rate, equipment age, maintenance score, and number of historical upgrades from different historical upgrade coefficient data in the first training set as the input data of the first convolutional neural network, and using the corresponding historical upgrade coefficients in the first training set as the output data of the first convolutional neural network; The first convolutional neural network is trained to obtain the first initial convolutional neural network. The first initial convolutional neural network is validated using the first test set. The first initial convolutional neural network that outputs a value less than or equal to the preset first test error threshold is used as the upgrade coefficient prediction model. The average deviation rate, accuracy rate, equipment age, maintenance score, and historical upgrade count of all testing lines are input into the upgrade coefficient prediction model to obtain the predicted upgrade coefficients for all testing lines. In an embodiment of the present invention, the predicted upgrade coefficient of a single inspection pipeline is obtained through an upgrade coefficient prediction model. The predicted upgrade coefficient is related to the average deviation rate, accuracy rate, equipment age, maintenance score, and historical upgrade count. The level of the average deviation rate directly affects the level of the prediction upgrade coefficient. The higher the average deviation rate, the worse the detection consistency of the detection line and the lower the prediction upgrade coefficient. Therefore, the average deviation rate and the prediction upgrade coefficient are negatively correlated. The accuracy rate directly affects the prediction upgrade coefficient. The higher the accuracy rate, the higher the prediction upgrade coefficient. Therefore, the accuracy rate and the prediction upgrade coefficient are positively correlated. The age of the equipment directly affects the predicted upgrade coefficient. The older the equipment, the more outdated the equipment in the testing line, the lower its performance, and the lower the predicted upgrade coefficient. Therefore, the equipment age and the predicted upgrade coefficient are negatively correlated. The maintenance score directly affects the predicted upgrade coefficient. A higher maintenance score indicates higher reliability of the testing pipeline and a higher predicted upgrade coefficient. Therefore, the maintenance score and the predicted upgrade coefficient are positively correlated. The number of historical upgrades directly affects the predicted upgrade coefficient. The more historical upgrades, the more upgrades the testing pipeline has undergone, the higher its reliability, and the higher the predicted upgrade coefficient. Therefore, the number of historical upgrades and the predicted upgrade coefficient are positively correlated.

[0049] It should be further explained that, in the specific implementation process, the process of obtaining historical degradation coefficient data and building a degradation coefficient prediction model, and then using the degradation coefficient prediction model to predict and verify the degradation coefficient of the pipeline, includes: The degradation coefficient refers to the probability coefficient of a verification pipeline being downgraded to a testing pipeline. Factors affecting the degradation factor of the validation pipeline include: accuracy, cross-bias rate, equipment age, maintenance score, and number of historical degradations; Accuracy rate refers to the ratio of the number of accurate defect level determination results of the verification line for glass substrates to the total number of glass substrates. Cross-bias ratio refers to the average deviation rate between the verification pipeline and other verification pipelines when scoring the same glass substrate; Equipment age refers to the length of time the equipment has been used; Maintenance rating refers to a rating based on the quality of equipment maintenance, derived from maintenance logs and records; Historical downgrade count refers to the number of times this verification pipeline has been downgraded in the past; Obtain historical degradation coefficient data for a single validation pipeline. The historical degradation coefficient data includes the accuracy, cross-bias rate, equipment age, maintenance score, number of historical degradations, and historical degradation coefficient of the single validation pipeline. Based on the accuracy, cross-bias rate, equipment age, maintenance score, number of historical degradations and corresponding historical degradation coefficients of a single validation pipeline in different historical degradation coefficient data, a degradation coefficient prediction set is generated and divided into a second training set and a second test set. Construct a second convolutional neural network, using the accuracy, cross-bias ratio, equipment age, maintenance score, and number of historical degradations from different historical degradation coefficients in the second training set as the input data of the second convolutional neural network, and using the corresponding historical degradation coefficients in the second training set as the output data of the second convolutional neural network; The second convolutional neural network is trained to obtain the second initial convolutional neural network. The second initial convolutional neural network is validated using the second test set. The second initial convolutional neural network that outputs a second test error threshold less than or equal to the preset second test error threshold is used as the degradation coefficient prediction model. The accuracy, cross-bias rate, equipment age, maintenance score, and historical degradation count of all validation pipelines are input into the degradation coefficient prediction model to obtain the predicted degradation coefficients for all validation pipelines. In an embodiment of the present invention, the predicted degradation coefficient of a single validation pipeline is obtained through a degradation coefficient prediction model. The predicted degradation coefficient is related to accuracy, cross-bias rate, equipment age, maintenance score, and historical degradation count. The accuracy rate directly affects the prediction degradation coefficient. The higher the accuracy rate, the lower the prediction degradation coefficient. Therefore, the accuracy rate and the prediction degradation coefficient are negatively correlated. The cross-bias rate directly affects the prediction degradation coefficient. The higher the cross-bias rate, the more isolated and unreliable the detection results of the validation pipeline are, and the higher the prediction degradation coefficient is. Therefore, the cross-bias rate and the prediction degradation coefficient are positively correlated. The age of the equipment directly affects the predicted degradation coefficient. The older the equipment, the older the equipment in the validation pipeline, the lower the performance of the equipment, and the higher the predicted degradation coefficient. Therefore, the equipment age is positively correlated with the predicted degradation coefficient. The maintenance score directly affects the predicted degradation coefficient. A higher maintenance score indicates higher reliability of the verification pipeline and a lower predicted degradation coefficient. Therefore, the maintenance score and the predicted degradation coefficient are negatively correlated. The number of historical degradations directly affects the predicted degradation coefficient. The more historical degradations, the more degradations the validation pipeline undergoes, the lower its reliability, and the higher the predicted degradation coefficient. Therefore, the number of historical degradations and the predicted degradation coefficient are positively correlated.

[0050] It should be further explained that, in the specific implementation process, an appropriate upgrade coefficient threshold is set according to the historical upgrade coefficient data. The historical upgrade coefficient data refers to the data set of upgrade coefficients of previous detection pipelines. The upgrade coefficients of all detection pipelines are compared with the upgrade coefficient threshold. The detection pipelines with upgrade coefficients greater than the upgrade coefficient threshold are arranged in descending order of upgrade coefficient, and the top 20% of X detection pipelines are selected to be upgraded into validation pipelines. Set an appropriate degradation coefficient threshold based on historical degradation coefficient data. The historical degradation coefficient data refers to the data set of degradation coefficients of previous verification pipelines. Compare the degradation coefficients of all verification pipelines with the degradation coefficient threshold. The validation pipelines are arranged in descending order of their degradation coefficients. The first X validation pipelines are selected, and it is determined whether the degradation coefficient of the Xth validation pipeline is greater than the degradation coefficient threshold. If the degradation coefficient is greater than the threshold, the first X validation pipelines are downgraded to detection pipelines and maintained in a timely manner. If the degradation coefficient of the Xth validation pipeline is less than the threshold, only the first Y validation pipelines with degradation coefficients greater than the threshold need to be downgraded to detection pipelines and maintained in a timely manner. Only the first Y detection pipelines can be upgraded to validation pipelines to ensure that the ratio of detection pipelines to validation pipelines remains unchanged.

[0051] The above description is merely a specific embodiment of this application, but the scope of protection of this application is not limited thereto. Any variations or substitutions that can be easily conceived by those skilled in the art within the scope of the technology disclosed in this application should be included within the scope of protection of this application. Therefore, the scope of protection of this application should be determined by the scope of the claims.

[0052] In conclusion, the above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the protection scope of the present invention.

Claims

1. A method for automatic identification of defects in the glass substrate of a liquid crystal display module, characterized in that, Includes the following steps: S1: The inspection equipment in the inspection line collects defect data of the glass substrate in real time and performs normalization processing on the data; S2: Generate a defect score for each glass substrate based on the normalized data, classify each glass substrate according to the defect score, and sort them into different areas. S3: Generate an input score for each verification pipeline based on the input data of the verification pipeline. Based on the input score, input the high-defect glass substrates and medium-defect glass substrates determined by the detection pipeline into the corresponding verification pipeline. The verification pipeline performs a second defect score on the input glass substrates. S4: The deviation rate is calculated based on multiple defect scores of the glass substrate, and the final grade of the glass substrate is determined based on the deviation rate. S5: Obtain historical upgrade coefficient data and build an upgrade coefficient prediction model. Use the upgrade coefficient prediction model to predict the upgrade coefficient of the detection pipeline, set the upgrade threshold, and decide whether to upgrade the detection pipeline based on the upgrade threshold and the upgrade coefficient of the detection pipeline. S6: Obtain historical degradation coefficient data and build a degradation coefficient prediction model. Use the degradation coefficient prediction model to predict the degradation coefficient of the verification pipeline, set a degradation threshold, and decide whether to degrade the verification pipeline based on the degradation threshold and the degradation coefficient of the verification pipeline.

2. The automatic defect identification method for LCD screen module glass substrate according to claim 1, characterized in that, The glass substrate defect inspection workshop is equipped with multiple production lines, including multiple inspection production lines and multiple verification production lines. The verification production lines account for 30% of all production lines. The inspection production lines are used to detect and identify defects in the glass substrates, and the verification production lines are used to re-inspect the high-level defect glass substrates and medium-level defect glass substrates identified by the inspection production lines. The defect data includes the surface grayscale variance of the glass substrate, thermal distribution gradient, fluorescence intensity peak, edge sharpness, and color consistency. Images of the glass substrate surface are captured using a visible light camera. Image processing algorithms are used to calculate the variance of pixel gray levels across the entire area to obtain the surface gray level variance. This surface gray level variance is then normalized. In the formula, To normalize the surface grayscale variance, This represents the measured surface grayscale variance. This represents the largest historical variance. The temperature distribution map of the glass substrate is obtained by scanning it with an infrared thermal imager. The average gradient magnitude of the temperature field is calculated to obtain the thermal distribution gradient, which is then normalized. In the formula, For the normalized heat distribution gradient, This is the measured heat distribution gradient. The maximum gradient; Images of the glass substrate are excited and acquired using an ultraviolet fluorescence imager. The peak fluorescence intensity is obtained by counting the maximum fluorescence intensity value of all pixels in the image, and then the peak fluorescence intensity is normalized. In the formula, The normalized fluorescence intensity peak value is... This represents the measured peak fluorescence intensity. The peak value represents the maximum fluorescence intensity. The edge image of the glass substrate is captured by a visible light camera, and the edge sharpness is calculated by measuring the image clarity of the edge region. The edge sharpness is then normalized. In the formula, To normalize edge sharpness, This represents the measured edge sharpness. Maximum edge sharpness; RGB images are acquired using a visible light camera. The standard deviations of the R, G, and B channels within the effective area of ​​the substrate are calculated, and their average values ​​are then used to obtain color consistency. Color consistency is then normalized. In the formula, This is a normalized color consistency value. This represents the measured value for consistent edge color. This represents the maximum color consistency value.

3. The automatic defect identification method for LCD screen module glass substrate according to claim 2, characterized in that, The process of generating a defect score for each glass substrate based on the normalized data includes: Defect score S of glass substrate: In the formula, , , and These are weighting coefficients, obtained through training based on historical data.

4. The automatic defect identification method for liquid crystal display module glass substrate according to claim 3, characterized in that, The process of grading each glass substrate based on defect scores includes: Based on historical defect scoring data, set appropriate first and second defect scoring thresholds. The historical defect scoring data refers to the data set of previous glass substrate defect scores. Compare the defect scores of the glass substrate with the first and second defect scoring thresholds to classify the glass substrates into glass substrates with different defect levels. If the defect score of the glass substrate is less than or equal to the first defect score threshold, the glass substrate is judged to be a low-level defect glass substrate. If the first defect score threshold is less than the glass substrate defect score and the second defect score threshold, the glass substrate is judged to be a medium-level defect glass substrate. If the second defect score threshold is less than or equal to the defect score of the glass substrate, the glass substrate is judged to be a high-defect glass substrate. A defect level label is assigned to each glass substrate, and this defect level label is bound to the unique identification code of the glass substrate. When a glass substrate with a unique identification code enters the sorting area of ​​the production line, the barcode scanner reads its information and sends a sorting instruction containing the target flow direction to the sorting robot according to the defect level it is bound to. At the end of the sorting area, there are usually multiple storage areas, each corresponding to a glass substrate with a different defect level. Based on the received sorting instructions, the sorting area sorts the glass substrates to the corresponding storage areas.

5. The automatic defect identification method for liquid crystal display module glass substrate according to claim 4, characterized in that, The process of generating an input score for each validation pipeline based on its input data includes: The input data includes transportation distance, verification pipeline load, historical accuracy, and average processing time. The transport distance is obtained by measuring the distance between the testing line and the verification line, and then normalized. In the formula, To normalize transport distance, This is the measured transportation distance. Maximum transport distance; Verification pipeline load refers to the number of glass substrates waiting to be arranged in the verification pipeline in real time. The verification pipeline load is then normalized. In the formula, To normalize the verification pipeline load, To verify the pipeline load through actual testing, To maximize the verification pipeline load; Historical accuracy refers to the percentage of accurate verifications in the history of the verification pipeline, that is, the correctness of the judgment results on the defect level of the glass substrate. The historical accuracy is obtained by dividing the number of glass substrates with incorrect judgments by the total number of glass substrates verified. The historical accuracy is then normalized. In the formula, To normalize historical accuracy, This represents the historical accuracy rate as measured. To achieve the highest historical accuracy; Average processing time refers to the time spent processing a single glass substrate in the verification pipeline. The processing time is normalized as follows: In the formula, To normalize the average processing time, This represents the measured average processing time. This represents the maximum average processing time. Input rating P: In the formula, , , These are weighting coefficients, obtained through training based on historical data.

6. The automatic defect identification method for liquid crystal display module glass substrate according to claim 5, characterized in that, Based on the input scores, high-grade defect glass substrates and medium-grade defect glass substrates determined by the inspection pipeline are fed into the corresponding verification pipeline. The verification pipeline performs a second defect score on the fed glass substrates, which includes: Based on the input scores of each verification pipeline equivalent to the inspection pipeline from high to low, the high-defect glass substrates and medium-defect glass substrates in the storage area of ​​the inspection pipeline are put into the verification pipeline with the highest input score of that inspection pipeline. The verification line performs a second defect score on the incoming glass substrates to obtain a second defect score. Based on the second defect score, the first defect score threshold, and the second defect score threshold, the glass substrates are graded again and sorted into the corresponding storage area of ​​the verification line.

7. The automatic defect identification method for liquid crystal display module glass substrate according to claim 6, characterized in that, The deviation rate is calculated based on multiple defect scores of the glass substrate. The process of determining the final grade of the glass substrate based on the deviation rate includes: The defect score generated by the inspection line is set as the first defect score, and the defect score generated by the verification line is set as the second defect score. The first defect score is subtracted from the second defect score to obtain the score difference. The absolute value of the score difference is divided by the maximum value of the two scores to obtain the deviation rate between the two scores. This deviation rate is set as the first deviation rate. Set an appropriate deviation rate threshold based on historical deviation rate data. The historical deviation rate data refers to the data set of deviation rates of two previous defect scores of the glass substrate. Compare the first deviation rate of the glass substrate with the deviation rate threshold. If the first deviation rate is less than the deviation rate threshold, then the test results of both the detection line and the verification line for the glass substrate are correct, and the glass substrate needs to be reworked. If the first deviation rate is greater than the deviation rate threshold, the glass substrate is put into another verification pipeline for defect scoring again to obtain a third defect score. The deviation rate between the first and third defect scores is calculated to obtain the second deviation rate. The deviation rate between the second and third defect scores is calculated to obtain the third deviation rate. The second and third deviation rates are compared with the deviation rate threshold. If the second deviation rate is less than the deviation rate threshold and the third deviation rate is greater than or equal to the deviation rate threshold, the detection process is error-free, the first verification process is error-free, and the defect level of the glass substrate is maintained according to the second deviation rate, which is the judgment result of the detection process. If the second deviation rate is greater than or equal to the deviation rate threshold and the third deviation rate is less than the deviation rate threshold, then the detection pipeline is faulty, the first verification pipeline is error-free, and the defect level of the glass substrate remains the same as the judgment result of the first verification pipeline based on the third deviation rate. If both the second and third deviation rates are greater than or equal to the deviation rate threshold, it cannot be determined whether the testing line and the first verification line have made a detection error. The glass substrate needs to be fed into other verification lines, and the deviation rate should be obtained by following the above steps. Then, it should be verified whether the testing line, the first verification line, and the second verification line have made a detection error.

8. The automatic defect identification method for glass substrate of LCD screen module according to claim 7, characterized in that, The process of acquiring historical upgrade coefficient data and building an upgrade coefficient prediction model, and then using the upgrade coefficient prediction model to predict the upgrade coefficient of the inspection pipeline includes: The upgrade coefficient refers to the probability coefficient of upgrading the detection pipeline to a verification pipeline. The overall stability of the detection results of the verification pipeline is better than that of the detection pipeline. Factors affecting the upgrade coefficient of the testing pipeline include: average deviation rate, accuracy rate, equipment age, maintenance score, and number of historical upgrades; The average deviation rate refers to the average deviation rate of the inspection process for high-grade defect glass substrates and medium-grade defect glass substrates. Accuracy rate refers to the ratio of the number of accurate defect level determination results for glass substrates by the inspection line to the total number of glass substrates. Equipment age refers to the length of time the equipment has been used; Maintenance rating refers to a rating based on the quality of equipment maintenance, derived from maintenance logs and records; Historical upgrade count refers to the number of times this testing line has been upgraded in the past; Obtain historical upgrade coefficient data for a single testing pipeline. The historical upgrade coefficient data includes the average deviation rate, accuracy rate, equipment age, maintenance score, number of historical upgrades, and historical upgrade coefficient of the single testing pipeline. Based on the average deviation rate, accuracy rate, equipment age, maintenance score, number of historical upgrades and corresponding historical upgrade coefficients of a single detection pipeline in different historical upgrade coefficient data, an upgrade coefficient prediction set is generated and divided into the first training set and the first test set. Construct a first convolutional neural network, using the average deviation rate, accuracy rate, equipment age, maintenance score, and number of historical upgrades from different historical upgrade coefficient data in the first training set as the input data of the first convolutional neural network, and using the corresponding historical upgrade coefficients in the first training set as the output data of the first convolutional neural network; The first convolutional neural network is trained to obtain the first initial convolutional neural network. The first initial convolutional neural network is validated using the first test set. The first initial convolutional neural network that outputs a value less than or equal to the preset first test error threshold is used as the upgrade coefficient prediction model. The average deviation rate, accuracy rate, equipment age, maintenance score, and historical upgrade count of all testing lines are input into the upgrade coefficient prediction model to obtain the predicted upgrade coefficients for all testing lines.

9. The automatic defect identification method for glass substrate of LCD screen module according to claim 8, characterized in that, The process of acquiring historical degradation coefficient data and building a degradation coefficient prediction model, and then using the degradation coefficient prediction model to predict and verify the degradation coefficient of the pipeline, includes: The degradation coefficient refers to the probability coefficient of a verification pipeline being downgraded to a testing pipeline. Factors affecting the degradation factor of the validation pipeline include: accuracy, cross-bias rate, equipment age, maintenance score, and number of historical degradations; Accuracy rate refers to the ratio of the number of accurate defect level determination results of the verification line for glass substrates to the total number of glass substrates. Cross-bias ratio refers to the average deviation rate between the verification pipeline and other verification pipelines when scoring the same glass substrate; Equipment age refers to the length of time the equipment has been used; Maintenance rating refers to a rating based on the quality of equipment maintenance, derived from maintenance logs and records; Historical downgrade count refers to the number of times this verification pipeline has been downgraded in the past; Obtain historical degradation coefficient data for a single validation pipeline. The historical degradation coefficient data includes the accuracy, cross-bias rate, equipment age, maintenance score, number of historical degradations, and historical degradation coefficient of the single validation pipeline. Based on the accuracy, cross-bias rate, equipment age, maintenance score, number of historical degradations and corresponding historical degradation coefficients of a single validation pipeline in different historical degradation coefficient data, a degradation coefficient prediction set is generated and divided into a second training set and a second test set. Construct a second convolutional neural network, using the accuracy, cross-bias ratio, equipment age, maintenance score, and number of historical degradations from different historical degradation coefficients in the second training set as the input data of the second convolutional neural network, and using the corresponding historical degradation coefficients in the second training set as the output data of the second convolutional neural network; The second convolutional neural network is trained to obtain the second initial convolutional neural network. The second initial convolutional neural network is validated using the second test set. The second initial convolutional neural network that outputs a second test error threshold less than or equal to the preset second test error threshold is used as the degradation coefficient prediction model. The accuracy, cross-bias ratio, equipment age, maintenance score, and historical degradation count of all validation pipelines are input into the degradation coefficient prediction model to obtain the predicted degradation coefficients for all validation pipelines.

10. The automatic defect identification method for glass substrate of LCD screen module according to claim 9, characterized in that, Set an appropriate upgrade coefficient threshold based on historical upgrade coefficient data. The historical upgrade coefficient data refers to the data set of upgrade coefficients of previous detection pipelines. Compare the upgrade coefficients of all detection pipelines with the upgrade coefficient threshold. The detection pipelines with upgrade coefficients greater than the upgrade coefficient threshold are arranged in descending order of upgrade coefficient, and the top 20% of X detection pipelines are selected to be upgraded into validation pipelines. Set an appropriate degradation coefficient threshold based on historical degradation coefficient data. The historical degradation coefficient data refers to the data set of degradation coefficients of previous verification pipelines. Compare the degradation coefficients of all verification pipelines with the degradation coefficient threshold. The validation pipelines are arranged in descending order of their degradation coefficients. The first X validation pipelines are selected, and it is determined whether the degradation coefficient of the Xth validation pipeline is greater than the degradation coefficient threshold. If the degradation coefficient is greater than the threshold, the first X validation pipelines are downgraded to detection pipelines and maintained in a timely manner. If the degradation coefficient of the Xth validation pipeline is less than the threshold, only the first Y validation pipelines with degradation coefficients greater than the threshold need to be downgraded to detection pipelines and maintained in a timely manner. Only the first Y detection pipelines can be upgraded to validation pipelines to ensure that the ratio of detection pipelines to validation pipelines remains unchanged.

Images