Defect detection method, system and equipment for LCD display screen production

Abstract

The invention provides a flaw detection method, system and equipment for LCD display screen production. The method comprises the following steps: when the LCD is sealed, acquiring a dispensing port image of sealing and dispensing equipment, and predicting the dispensing amount to obtain the predicted dispensing amount; performing a glue amount redundancy defect analysis decision and a glue amount insufficiency defect analysis decision to obtain a redundancy defect analysis coefficient and an insufficiency defect analysis coefficient; predicting the thickness unevenness of the LCD to obtain a first defect parameter; performing cured liquid crystal damage prediction to obtain a second flaw parameter; and according to the first defect parameter and the second defect parameter, generating a defect parameter of the LCD as a defect detection result. The technical problem that in the prior art, the defect detection accuracy and efficiency are not high in the LCD sealing and dispensing process is solved, and the technical effect of improving the defect detection accuracy and efficiency of LCD display screen production through dispensing opening image analysis and multi-parameter defect prediction is achieved.

Description

Technical Field

[0001] The present invention relates to the field of LCD production detection, and particularly to a method, system and device for detecting defects in the production of LCD displays. Background Art

[0002] With the rapid development of display technology, LCD displays have been widely used in various electronic devices. In the production process of LCDs, sealing and dispensing is one of the key technological processes, and its quality directly affects the display effect and service life of LCDs. In the prior art, the defect detection of LCD sealing and dispensing mainly relies on manual sampling inspection or post-production finished product inspection, and it is difficult to detect defects in real time and efficiently during the production process. Some improved methods attempt to judge whether the dispensing amount is qualified by setting a fixed threshold. However, due to the dynamic changes in the operating state of the dispensing equipment, especially the blockage that may be formed due to the solidification of residual glue at the dispensing port, the accuracy of this simple threshold judgment method is limited. Therefore, there are problems of low accuracy and efficiency in defect detection during the LCD sealing and dispensing process in the prior art. Summary of the Invention

[0003] The present invention aims at the technical problems of low accuracy and efficiency in defect detection during the LCD sealing and dispensing process in the prior art, and provides a method, system and device for detecting defects in the production of LCD displays to solve the problems.

[0004] The technical solutions of the present invention for solving the above technical problems are as follows: In a first aspect, the present invention provides a method for detecting defects in the production of LCD displays, including: when sealing the LCD, collecting an image of the dispensing port of the sealing and dispensing equipment, predicting the dispensing amount, and obtaining a predicted dispensing amount; according to the predicted dispensing amount and a preset dispensing amount, performing a glue amount redundancy defect analysis decision and a glue amount shortage defect analysis decision, and obtaining a redundancy defect analysis coefficient and a shortage defect analysis coefficient; according to the redundancy defect analysis coefficient, predicting the thickness non-uniformity of the LCD based on the predicted dispensing amount, and obtaining a first defect parameter; according to the shortage defect analysis coefficient, predicting the damage of the cured liquid crystal based on the predicted dispensing amount, and obtaining a second defect parameter; generating a defect parameter of the LCD according to the first defect parameter and the second defect parameter as a defect detection result.

[0005] Optionally, when sealing the LCD, collecting an image of the dispensing port of the sealing and dispensing equipment, predicting the dispensing amount, and obtaining a predicted dispensing amount, includes: when sealing the LCD, collecting an image of the dispensing port of the sealing and dispensing equipment; inputting the image of the dispensing port into a pre-trained dispensing amount predictor, and outputting to obtain a predicted dispensing amount.

[0006] Optionally, the training step of the dispensing amount predictor includes: collecting a set of sample dispensing port images based on the LCD display production data within a historical time period, and collecting the dispensing amount under each sample dispensing port image, which is labeled as the sample dispensing amount, to obtain a set of sample dispensing amounts; constructing a dispensing amount predictor based on a convolutional neural network; using the set of sample dispensing port images and the set of sample dispensing amounts to perform supervised training and testing on the dispensing amount predictor, and completing pre-training after the accuracy rate is qualified.

[0007] Optionally, based on the predicted dispensing amount and the preset dispensing amount, perform glue amount redundancy defect analysis decision-making and glue amount shortage defect analysis decision-making to obtain a redundancy defect analysis coefficient and a shortage defect analysis coefficient, including: obtaining the preset dispensing amount of the LCD seal, and collecting the maximum dispensing amount error deviating from the preset dispensing amount according to the dispensing data within a historical time period; calculating the difference between the predicted dispensing amount and the preset dispensing amount, and calculating the ratio to the maximum dispensing amount error to obtain an error coefficient; obtaining a preset redundancy defect analysis coefficient and a preset shortage defect analysis coefficient; calculating the sum of the preset redundancy defect analysis coefficient and the error coefficient to obtain a redundancy defect analysis coefficient, and calculating the difference between the preset shortage defect analysis coefficient and the error coefficient to obtain a shortage defect analysis coefficient.

[0008] Optionally, based on the redundancy defect analysis coefficient, perform uneven thickness prediction of the LCD based on the predicted dispensing amount to obtain a first defect parameter, and based on the shortage defect analysis coefficient, perform cured liquid crystal damage prediction based on the predicted dispensing amount to obtain a second defect parameter, including: using the redundancy defect analysis coefficient, combined with the number of N uneven thickness prediction branches in the pre-trained uneven thickness predictor, calculate to obtain U, where N is a positive integer and U is a positive integer less than or equal to N; randomly select U uneven thickness prediction branches, input the predicted dispensing amount, predict and output U uneven thickness prediction parameters, and calculate the mean value to obtain a first defect parameter; using the shortage defect analysis coefficient, combined with the number of N liquid crystal damage prediction branches in the pre-trained liquid crystal damage predictor, calculate to obtain U; randomly select U liquid crystal damage prediction branches, input the predicted dispensing amount, predict and output U liquid crystal damage prediction parameters, and calculate the mean value to obtain a second defect parameter.

[0009] Optionally, the pre-training step of the thickness non-uniformity predictor includes: according to the LCD sealing data within the historical time, collecting a set of sample predicted dispensing amounts, and collecting the amplitude of thickness non-uniformity of the LCD under different sample predicted dispensing amounts, and labeling to obtain a set of sample thickness non-uniformity parameters; randomly selecting with replacement to obtain the first thickness non-uniformity training data from the set of sample predicted dispensing amounts and the set of sample thickness non-uniformity parameters; continuing to randomly select with replacement to obtain N pieces of thickness non-uniformity training data; using deep learning to construct N thickness non-uniformity prediction branches, and respectively using the N pieces of thickness non-uniformity training data for supervised training and testing until the accuracy rate is qualified to complete the pre-training.

[0010] Optionally, generating the defect parameter of the LCD according to the first defect parameter and the second defect parameter as the defect detection result includes: generating the defect parameter of the LCD according to the first defect parameter and the second defect parameter; using the defect parameter as the defect detection result for display and reminder.

[0011] In a second aspect, the present invention provides a defect detection system for LCD display production, including: a dispensing amount prediction module, configured to collect an image of the dispensing port of the sealing dispensing device during LCD sealing, perform dispensing amount prediction, and obtain a predicted dispensing amount; a glue amount analysis module, configured to perform glue amount redundancy defect analysis decision-making and glue amount shortage defect analysis decision-making according to the predicted dispensing amount and a preset dispensing amount, and obtain a redundancy defect analysis coefficient and a shortage defect analysis coefficient; a defect prediction module, configured to perform thickness non-uniformity prediction of the LCD based on the predicted dispensing amount according to the redundancy defect analysis coefficient to obtain a first defect parameter, and perform cured liquid crystal damage prediction based on the predicted dispensing amount according to the shortage defect analysis coefficient to obtain a second defect parameter; a result generation module, configured to generate a defect parameter of the LCD according to the first defect parameter and the second defect parameter as the defect detection result.

[0012] In a third aspect, the present application provides an electronic device, which includes: a processor; a memory for storing executable instructions of the processor; wherein, the processor is configured to execute a defect detection method for LCD display production provided by the present application.

[0013] The beneficial effects of the present invention are: When sealing the LCD, the dispensing port image of the dispensing and sealing equipment is collected to predict the dispensing volume and obtain the predicted dispensing volume. The working state image of the dispensing equipment can be obtained in real time, and the actual dispensing volume can be predicted based on image analysis, solving the problem that the traditional method cannot timely know the dispensing volume and providing basic data for subsequent defect analysis. According to the predicted dispensing volume and the preset dispensing volume, decisions on redundant defect analysis of dispensing volume and insufficient defect analysis of dispensing volume are made to obtain the redundant defect analysis coefficient and the insufficient defect analysis coefficient. By comparing the difference between the predicted dispensing volume and the preset dispensing volume, the defect risk coefficients that may be caused in the cases of excessive and insufficient dispensing volume are calculated respectively, providing parameter support for subsequent specific defect prediction. According to the redundant defect analysis coefficient, based on the predicted dispensing volume, the uneven thickness prediction of the LCD is carried out to obtain the first defect parameter; at the same time, according to the insufficient defect analysis coefficient, based on the predicted dispensing volume, the cured liquid crystal damage prediction is carried out to obtain the second defect parameter. Through targeted defect prediction for different types of dispensing abnormalities: when the dispensing volume is excessive, it may lead to uneven liquid crystal thickness, and when the dispensing volume is insufficient, it may lead to damage to the liquid crystal during the curing process, thereby obtaining the corresponding defect parameters. According to the first defect parameter and the second defect parameter, the defect parameter of the LCD is generated as the defect detection result, realizing the accurate and efficient detection of defects on the LCD display screen.

[0014] Through the above technical solution, the present application realizes real-time defect prediction and detection during the LCD sealing and dispensing process, improves the accuracy and efficiency of defect detection, and avoids the problems of low accuracy and efficiency caused by relying on manual sampling inspection or later finished product inspection in the traditional method. Brief Description of the Drawings

[0015] Figure 1 It is a schematic flow chart of a method for detecting defects in the production of an LCD display screen provided by the present invention; Figure 2 It is a schematic structural diagram of a system for detecting defects in the production of an LCD display screen provided by the present invention; Figure 3 It is a schematic structural diagram of an electronic device provided by the present invention.

[0016] In the drawings, the components represented by each reference numeral are as follows: Dispensing volume prediction module 11, dispensing volume analysis module 12, defect prediction module 13, result generation module 14, electronic device 200, memory 210, processor 220, computer program 211. Detailed Embodiment

[0017] Next, the technical solutions in the embodiments of the present invention will be clearly and completely described in conjunction with the accompanying drawings in the embodiments of the present invention. Obviously, the described embodiments are only a part of the embodiments of the present invention, rather than all the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative work belong to the scope of protection of the present invention.

[0018] In the description of the present invention, the terms "first" and "second" are only used for descriptive purposes and cannot be construed as indicating or implying relative importance or implicitly specifying the quantity of the indicated technical features. Thus, the features defined with "first" and "second" may explicitly or implicitly include one or more of the described features. In the description of the present invention, "a plurality of" means two or more, unless otherwise specifically defined.

[0019] In the description of the present invention, the term "for example" is used to mean "serving as an example, illustration, or explanation". Any embodiment described as "for example" in the present invention is not necessarily construed as being more preferred or having more advantages than other embodiments. In order for any person skilled in the art to implement and use the present invention, the following description is given. In the following description, details are set forth for purposes of explanation. It should be understood that those of ordinary skill in the art can recognize that the present invention can be implemented without using these specific details. In other instances, well-known structures and processes are not elaborated in detail to avoid unnecessary details from obscuring the description of the present invention. Therefore, the present invention is not intended to be limited to the embodiments shown, but is consistent with the broadest scope that conforms to the principles and features disclosed in the present invention.

[0020] Embodiment 1, as Figure 1 shown, the embodiment of the present invention provides a method for detecting defects in the production of an LCD display screen, including: S100: When sealing the LCD, collect the image of the dispensing port of the dispensing and sealing device, perform prediction of the dispensing amount, and obtain the predicted dispensing amount.

[0021] Specifically, when the LCD liquid crystal panel is sealed, the image acquisition device is used to perform real-time image acquisition on the dispensing port of the dispensing and sealing device to obtain the dispensing port image, which reflects the state of the dispensing port of the dispensing and sealing device. The state of the dispensing port affects the actual dispensing amount. When the residual glue at the dispensing port of the dispensing and sealing device solidifies and forms a blockage, it will significantly affect the output of the actual dispensing amount. After obtaining the dispensing port image, the actual state of the dispensing port is processed through image analysis technology to predict the dispensing amount to be output under the current dispensing port state, thereby obtaining the predicted dispensing amount, which provides necessary parameters for subsequent glue amount redundancy defect analysis decision-making and glue amount shortage defect analysis decision-making.

[0022] S200: According to the predicted glue dispensing amount and the preset glue dispensing amount, a glue amount redundant defect analysis decision and a glue amount insufficient defect analysis decision are performed to obtain a redundant defect analysis coefficient and an insufficient defect analysis coefficient.

[0023] Specifically, after obtaining the predicted glue dispensing amount, it is compared and analyzed with the preset glue dispensing amount in the LCD display screen sealing process, and the glue amount redundant defect analysis decision and the glue amount insufficient defect analysis decision are respectively carried out. Among them, the preset glue dispensing amount refers to the standard glue dispensing amount predetermined according to the product design requirements and production process standards in the LCD display screen sealing process, which represents the optimal glue dispensing amount that should be used in the sealing process under ideal conditions. The glue amount redundant defect analysis decision is aimed at the situation where the predicted glue dispensing amount exceeds the preset glue dispensing amount. In this case, the excessive amount of glue will cause the liquid crystal to be squeezed, resulting in uneven thickness of the LCD display screen and affecting the display effect. The glue amount insufficient defect analysis decision is aimed at the situation where the predicted glue dispensing amount is lower than the preset glue dispensing amount. Insufficient glue dispensing may cause the glue dispensing to cure prematurely during the curing process, allowing ultraviolet rays to directly irradiate the liquid crystal and cause damage to the liquid crystal.

[0024] Through the redundant glue defect analysis decision and the insufficient glue defect analysis decision, the redundant defect analysis coefficient and the insufficient defect analysis coefficient are calculated respectively. These two coefficients reflect the risk level of corresponding defects that may occur under the current dispensing state, and provide a parameter basis for the subsequent steps of LCD thickness unevenness prediction and solidified liquid crystal damage prediction, thereby improving the pertinence and accuracy of defect detection.

[0025] S300: According to the redundant defect analysis coefficient and based on the predicted glue dispensing amount, the uneven thickness of the LCD is predicted to obtain a first defect parameter; according to the insufficient defect analysis coefficient and based on the predicted glue dispensing amount, the solidified liquid crystal damage is predicted to obtain a second defect parameter.

[0026] Specifically, first, the redundant defect analysis coefficient is used to predict the uneven thickness of the LCD based on the predicted glue dispensing amount. When the glue dispensing amount is too much, the excess glue will squeeze the liquid crystal, resulting in uneven thickness of the liquid crystal layer in various areas of the LCD display, which in turn affects the consistency of the display effect. Through this prediction process, the first defect parameter is obtained, which quantitatively represents the degree of uneven thickness that may occur under the current glue dispensing state. At the same time, the insufficient defect analysis coefficient is used to predict the damage to the solidified liquid crystal based on the predicted glue dispensing amount. When the glue dispensing amount is insufficient, the glue may be cured prematurely during the sealing and curing process, causing ultraviolet rays to directly irradiate the liquid crystal, thereby causing damage to the liquid crystal material. Through this prediction process, the second defect parameter is obtained, which quantitatively represents the degree of liquid crystal damage that may occur under the current glue dispensing state.

[0027] Through the first defect parameter and the second defect parameter, the defect risk of the LCD display screen that may be caused by the current dispensing state is reflected from two directions of excessive dispensing amount and insufficient dispensing amount respectively, providing a comprehensive parameter basis for the subsequent generation of defect detection results.

[0028] S400: Generate a defect parameter of the LCD as the defect detection result according to the first defect parameter and the second defect parameter.

[0029] Specifically, according to the first defect parameter predicted from the uneven thickness and the second defect parameter predicted from the cured liquid crystal damage, through a specific data fusion algorithm, a defect parameter of the LCD that can comprehensively reflect the sealing quality of the LCD display screen is generated as the defect detection result. This defect parameter takes into account both the problem of uneven thickness that may be caused by excessive dispensing amount and the problem of liquid crystal damage that may be caused by insufficient dispensing amount, thus realizing a comprehensive evaluation of the sealing quality of the LCD display screen. For example, the defect risks of uneven thickness and liquid crystal damage are weighted and combined, and appropriate weights are given to different types of defects according to the current production requirements. In some product models, more attention may be paid to display uniformity, so the weight of the defect parameter of uneven thickness will be higher; while in other product models, more attention may be paid to the liquid crystal life, so the weight of the defect parameter of liquid crystal damage will be higher. Through the fusion of defect parameters, it can flexibly adapt to the quality control requirements of different products and provide more accurate defect detection results.

[0030] The generated defect parameter of the LCD as the final result of defect detection can be used to guide the quality control decision-making on the production line. Through this parameter, the potential defect risks in the LCD production process can be detected in time, and corresponding adjustment measures can be taken to improve the production yield and product quality of the LCD display screen.

[0031] Furthermore, during the LCD sealing, collect the dispensing port image of the sealing dispensing equipment, perform dispensing amount prediction, and obtain the predicted dispensing amount, including: S110: During the LCD sealing, collect the dispensing port image of the sealing dispensing equipment; S120: Input the dispensing port image into a pre-trained dispensing amount predictor to output and obtain the predicted dispensing amount.

[0032] In a feasible implementation manner, during the LCD sealing process, an image acquisition device is used to perform real-time image acquisition on the dispensing port of the dispensing and gluing equipment to obtain a dispensing port image. This dispensing port image contains key information such as the current opening state of the dispensing port and the residual glue situation, and this information is closely related to the actual dispensing amount. Subsequently, the obtained dispensing port image is input into a pre-trained dispensing amount predictor for processing. The dispensing amount predictor is an algorithm model specifically used to predict the dispensing amount. By analyzing the features in the dispensing port image, such as the opening size, shape, and residual glue distribution, it outputs the predicted dispensing amount. This dispensing amount predictor has been fully trained and can accurately predict the amount of glue that will be output during the actual dispensing process based on the features of the dispensing port image, obtaining the predicted dispensing amount.

[0033] Through the above steps, it is possible to obtain an accurate prediction of the dispensing amount to be used before actual dispensing, providing a reliable data basis for subsequent defect analysis and decision-making. By means of pre-prediction, the problem in traditional methods that the suitability of the dispensing amount can only be detected after dispensing is effectively solved, realizing the preposition of defect detection.

[0034] Furthermore, the training steps of the dispensing amount predictor include: S131: According to the LCD display production data within the historical time, collect a set of sample dispensing port images, and collect the dispensing amount under each sample dispensing port image, which is labeled as the sample dispensing amount, to obtain a set of sample dispensing amounts; S132: Based on the convolutional neural network, construct a dispensing amount predictor; S133: Use the set of sample dispensing port images and the set of sample dispensing amounts to perform supervised training and testing on the dispensing amount predictor. After the accuracy rate is qualified, the pre-training is completed.

[0035] In a preferred implementation manner, first, based on the LCD display production data within the historical time, a training data set is constructed. Specifically, a large number of sample dispensing port images are collected from the LCD display production data within the historical time to form a set of sample dispensing port images. At the same time, for each sample dispensing port image, its corresponding actual dispensing amount is recorded as the sample dispensing amount to form a set of sample dispensing amounts. In this way, a complete training data set containing input features (dispensing port images) and output labels (sample dispensing amounts) is obtained.

[0036] Next, a dispensing amount predictor is constructed based on convolutional neural network technology. Convolutional neural network is a deep learning algorithm used to process image data. Through multi-layer convolution and pooling operations, it can effectively extract feature information in the image and is suitable for extracting features related to the dispensing amount from the dispensing port image. After that, the constructed dispensing amount predictor is supervised and trained using the obtained sample dispensing port image set and sample dispensing amount set. During the training process, the sample dispensing port image is input, and the dispensing amount predictor is asked to output the predicted dispensing amount, which is then compared with the known sample dispensing amount, and the error is calculated and the parameters of the dispensing amount predictor are adjusted accordingly. At the same time, test verification is carried out to ensure that the accuracy of the dispensing amount predictor meets the predetermined standard and complete the entire pre-training process.

[0037] Through a training method based on a large amount of historical data, a high-precision dispensing amount predictor can be built, providing reliable support for the detection of sealing defects in LCD displays.

[0038] Furthermore, it is characterized in that, according to the predicted glue dispensing amount and the preset glue dispensing amount, a glue amount redundant defect analysis decision and a glue amount insufficient defect analysis decision are performed to obtain a redundant defect analysis coefficient and an insufficient defect analysis coefficient, including: S210: Obtaining a preset glue dispensing amount for LCD sealing, and collecting a maximum glue dispensing amount error that deviates from the preset glue dispensing amount based on glue dispensing data in a historical period; S220: Calculate the difference between the predicted dispensing amount and the preset dispensing amount, and calculate the ratio of the predicted dispensing amount to the maximum dispensing amount error to obtain an error coefficient; S230: Obtaining a preset redundant defect analysis coefficient and a preset insufficient defect analysis coefficient; S240: Calculate the sum of the preset redundant defect analysis coefficient and the error coefficient to obtain a redundant defect analysis coefficient, and calculate the difference between the preset insufficient defect analysis coefficient and the error coefficient to obtain an insufficient defect analysis coefficient.

[0039] In a preferred embodiment, first, the preset glue dispensing amount of the LCD sealing process is obtained, and the preset glue dispensing amount is a standard glue dispensing amount predetermined based on product design and process requirements. At the same time, based on the statistical analysis of the glue dispensing data in the historical time, the maximum glue dispensing amount error that deviates from the preset glue dispensing amount is collected. The maximum glue dispensing amount error serves as a reference for subsequent calculations and reflects the maximum fluctuation range of the glue dispensing amount during the production process. Then, the difference between the current predicted glue dispensing amount and the preset glue dispensing amount is calculated, and the difference is calculated by ratio with the maximum glue dispensing amount error to obtain the error coefficient. The error coefficient obtained is a value with positive and negative signs. When the predicted glue dispensing amount is greater than the preset glue dispensing amount, the error coefficient is positive; when the predicted glue dispensing amount is less than the preset glue dispensing amount, the error coefficient is negative. For example, if the predicted glue dispensing amount is 10% less than the preset glue dispensing amount by the maximum error, the error coefficient is -10%.

[0040] Then, the preset redundant defect analysis coefficient and the preset insufficient defect analysis coefficient are obtained. Among them, the preset redundant defect analysis coefficient refers to the benchmark analysis strength set for the uneven thickness of LCD caused by excessive glue dispensing under standard working conditions. The coefficient is pre-set according to product characteristics, process requirements and historical data analysis results. For example, it is set to 50%, which means that under standard conditions, 50% of the computing resources will be allocated for thickness unevenness prediction; the preset insufficient defect analysis coefficient refers to the benchmark analysis strength set for the liquid crystal damage problem caused by insufficient glue dispensing under standard working conditions. The coefficient is also pre-set according to product characteristics, process requirements and historical data analysis results. For example, it is set to 50%, which means that under standard conditions, 50% of the computing resources will be allocated for liquid crystal damage prediction. After that, the final redundant defect analysis coefficient and insufficient defect analysis coefficient are obtained by different calculation methods. For the redundant defect analysis coefficient, the preset redundant defect analysis coefficient is added to the error coefficient, for example, 50%+10%=60%. For the insufficient defect analysis coefficient, the preset insufficient defect analysis coefficient is subtracted from the error coefficient, for example, 50%-10%=40%. Since a large amount of glue dispensed will cause the liquid crystal to be squeezed and uneven thickness will occur, the above calculation method will increase the redundant defect analysis coefficient when the predicted glue dispensed amount is large, and more resources will be allocated for uneven thickness prediction; at the same time, since the risk of ultraviolet damage to the liquid crystal is small due to a large amount of glue dispensed, the insufficient defect analysis coefficient will be reduced accordingly, which will reduce the resources used for liquid crystal damage prediction. Through this dynamic adjustment mechanism, defect detection can be carried out in a targeted manner according to the current glue dispensed state, improving detection efficiency and accuracy.

[0041] Further, according to the redundant defect analysis coefficient, based on the predicted glue dispensing amount, the uneven thickness of the LCD is predicted to obtain a first defect parameter, and according to the insufficient defect analysis coefficient, based on the predicted glue dispensing amount, the solidified liquid crystal damage is predicted to obtain a second defect parameter, including: S310: using the redundant defect analysis coefficient and combining the number of N thickness unevenness prediction branches in the pre-trained thickness unevenness predictor, calculate and obtain U, where N is a positive integer and U is a positive integer less than or equal to N; S320: Randomly select U thickness unevenness prediction branches, input the predicted glue dispensing amount, predict and output U thickness unevenness prediction parameters, calculate the mean, and obtain the first defect parameter; S330: using the insufficient defect analysis coefficient and combining the number of N liquid crystal damage prediction branches in the pre-trained liquid crystal damage predictor to calculate and obtain U; S340: Randomly select U liquid crystal damage prediction branches, input the predicted dispensing volume, predict and output U liquid crystal damage prediction parameters, calculate the mean value, and obtain the second defect parameter.

[0042] In a preferred embodiment, the idea of ensemble learning is adopted to dynamically adjust the number of prediction branches, achieving reasonable allocation of computing resources and optimization of prediction accuracy.

[0043] First, using the redundant defect analysis coefficient, combined with the total number of N thickness non-uniformity prediction branches in the pre-trained thickness non-uniformity predictor, calculate the actual number of prediction branches U to be used. Here, N is a positive integer representing the total number of all thickness non-uniformity prediction branches in the thickness non-uniformity predictor; U is also a positive integer and satisfies U ≤ N, representing the number of branches actually to be called for this prediction. The larger the redundant defect analysis coefficient, the larger the calculated U value, which means more computing resources will be allocated for thickness non-uniformity prediction. Then, randomly select U branches from the N thickness non-uniformity prediction branches, input the predicted dispensing volume as the input parameter into these U prediction branches, and obtain U thickness non-uniformity prediction parameters. Subsequently, calculate the mean value of these U thickness non-uniformity prediction parameters as the first defect parameter. By randomly selecting multiple thickness non-uniformity prediction branches and taking the mean value, the deviation of a single thickness non-uniformity prediction branch can be reduced, improving the stability and accuracy of the overall prediction.

[0044] Similarly, using the insufficient defect analysis coefficient, combined with the total number of N liquid crystal damage prediction branches in the pre-trained liquid crystal damage predictor, calculate the actual number of prediction branches U to be used. The larger the insufficient defect analysis coefficient, the larger the calculated U value, which means more computing resources will be allocated for liquid crystal damage prediction. Then, randomly select U liquid crystal damage prediction branches from the N liquid crystal damage prediction branches, input the predicted dispensing volume into these U liquid crystal damage prediction branches, obtain U liquid crystal damage prediction parameters, and calculate the mean value of these U liquid crystal damage prediction parameters as the second defect parameter.

[0045] Through dynamic prediction based on ensemble learning, the computing resource allocation can be flexibly adjusted according to the current dispensing state, improving the computing efficiency while ensuring prediction accuracy, providing an efficient and reliable technical means for detecting production defects of LCD displays.

[0046] Furthermore, the pre-training steps of the thickness non-uniformity predictor include: S351: According to the LCD sealing data within the historical time, collect a set of sample predicted dispensing volumes, and collect the amplitude of thickness non-uniformity of the LCD under different sample predicted dispensing volumes, and label to obtain a set of sample thickness non-uniformity parameters; S352: Randomly select the first thickness non-uniformity training data from the sample predicted dispensing volume set and the sample thickness non-uniformity parameter set with replacement; S353: Continue to randomly select N pieces of thickness non-uniformity training data with replacement; S354: Use deep learning to construct N thickness non-uniformity prediction branches, and use the N pieces of thickness non-uniformity training data for supervised training and testing respectively until the accuracy rate is qualified to complete the pre-training.

[0047] In a preferred embodiment, to train the thickness non-uniformity predictor, first, according to the LCD sealing data collected within the historical time, collect the sample predicted dispensing volume set as the input feature of the training data. At the same time, collect the actual measurement data of the LCD thickness non-uniformity phenomenon corresponding to different sample predicted dispensing volumes, label these data as sample thickness non-uniformity parameters, and form a sample thickness non-uniformity parameter set as the output label of the training data. In this way, the correspondence between the input feature (predicted dispensing volume) and the output label (thickness non-uniformity parameter) is established. Then, use the sampling method with replacement to randomly select data from the sample predicted dispensing volume set and the sample thickness non-uniformity parameter set to construct the first piece of thickness non-uniformity training data. Here, sampling with replacement means that the same sample may be selected multiple times, and this sampling method helps to increase the diversity of the training data.

[0048] Subsequently, continue to use the same sampling method with replacement to randomly select data from the sample predicted dispensing volume set and the sample thickness non-uniformity parameter set to construct a total of N pieces of thickness non-uniformity training data. Although each piece of training data is derived from the same original data set, due to the characteristics of random sampling, there are differences between each piece of training data, and this difference is the basis for the effect of ensemble learning. Then, use deep learning technology to construct N thickness non-uniformity prediction branches with the same structure but independent parameters. Each thickness non-uniformity prediction branch uses a corresponding piece of thickness non-uniformity training data for supervised training and testing until the prediction accuracy rate of each branch reaches the preset standard to complete the entire pre-training process.

[0049] It should be noted that a similar method can also be used for the pre-training process of the liquid crystal damage predictor. That is, based on historical data, sample predicted dispensing volume and corresponding liquid crystal damage parameters are also collected, multiple pieces of training data are constructed through sampling with replacement, and multiple independent liquid crystal damage prediction branches are trained to form a complete liquid crystal damage predictor.

[0050] Through this pre-training method based on ensemble learning, a predictor model with high robustness and prediction accuracy can be constructed, providing reliable technical support for defect detection in the LCD display production process.

[0051] Further, according to the first defect parameter and the second defect parameter, generate the defect parameter of the LCD as the defect detection result, including: S410: Generate the defect parameter of the LCD according to the first defect parameter and the second defect parameter; S420: Use the defect parameter as the defect detection result for display and reminder.

[0052] In a feasible implementation, summarize the obtained first defect parameter and second defect parameter to generate a comprehensive LCD defect parameter. This defect parameter comprehensively considers the problem of uneven thickness that may be caused by excessive dispensing volume (characterized by the first defect parameter) and the problem of liquid crystal damage that may be caused by insufficient dispensing volume (characterized by the second defect parameter), and can comprehensively reflect the current sealing quality status of the LCD display screen.

[0053] Subsequently, use the generated defect parameter as the final defect detection result, display it through the human-machine interface of the production line control system, and trigger the corresponding reminder mechanism when necessary. This display and reminder mechanism enables production personnel to timely understand the possible sealing defect risks of the currently produced LCD display screens, and then take necessary adjustment measures, such as adjusting the dispensing equipment parameters, cleaning the blockage of the dispensing nozzle, replacing the dispensing equipment, etc., so as to effectively prevent the generation of defective products.

[0054] Through the above steps, the conversion from the defect parameter to the defect detection result is realized, and the detection result is organically combined with the production control system to form a complete closed-loop for the detection and prevention of the sealing defects of the LCD display screen, effectively improving the production yield and product quality of the LCD display screen.

[0055] Embodiment 2, as Figure 2 shown, based on the same inventive concept as the defect detection method for the production of an LCD display screen provided in Embodiment 1, the embodiment of the present invention further provides a defect detection system for the production of an LCD display screen, including: A dispensing volume prediction module 11, configured to collect an image of the dispensing nozzle of the sealing dispensing equipment during LCD sealing, perform dispensing volume prediction, and obtain the predicted dispensing volume; A glue volume analysis module 12, configured to perform glue volume redundancy defect analysis decision-making and glue volume insufficiency defect analysis decision-making according to the predicted dispensing volume and the preset dispensing volume, and obtain a redundancy defect analysis coefficient and an insufficiency defect analysis coefficient; A defect prediction module 13, configured to perform uneven thickness prediction of the LCD based on the predicted dispensing volume according to the redundancy defect analysis coefficient to obtain a first defect parameter, and perform cured liquid crystal damage prediction based on the predicted dispensing volume according to the insufficiency defect analysis coefficient to obtain a second defect parameter; A result generation module 14, configured to generate a defect parameter of the LCD as a defect detection result according to the first defect parameter and the second defect parameter.

[0056] Further, the dispensing amount prediction module 11 includes the following execution steps: When sealing the LCD, collect an image of the dispensing nozzle of the sealing dispensing device; Input the image of the dispensing nozzle into a pre-trained dispensing amount predictor, and output and obtain the predicted dispensing amount.

[0057] Further, the dispensing amount prediction module 11 further includes the following execution steps: According to the production data of the LCD display screen within the historical time, collect a set of sample dispensing nozzle images, and collect the dispensing amount under each sample dispensing nozzle image, label it as the sample dispensing amount, and obtain a set of sample dispensing amounts; Based on a convolutional neural network, construct a dispensing amount predictor; Use the set of sample dispensing nozzle images and the set of sample dispensing amounts to perform supervised training and testing on the dispensing amount predictor, and after the accuracy rate is qualified, complete the pre-training.

[0058] Further, the glue amount analysis module 12 includes the following execution steps: Obtain the preset dispensing amount for sealing the LCD, and according to the dispensing data within the historical time, collect the maximum dispensing amount error deviating from the preset dispensing amount; Calculate the difference between the predicted dispensing amount and the preset dispensing amount, and calculate the ratio with the maximum dispensing amount error to obtain an error coefficient; Obtain a preset redundant defect analysis coefficient and a preset insufficient defect analysis coefficient; Calculate the sum of the preset redundant defect analysis coefficient and the error coefficient to obtain a redundant defect analysis coefficient, and calculate the difference between the preset insufficient defect analysis coefficient and the error coefficient to obtain an insufficient defect analysis coefficient.

[0059] Further, the defect prediction module 13 includes the following execution steps: Use the redundant defect analysis coefficient, combined with the number of N thickness non-uniformity prediction branches in the pre-trained thickness non-uniformity predictor, to calculate and obtain U, where N is a positive integer and U is a positive integer less than or equal to N; Randomly select U thickness non-uniformity prediction branches, input the predicted dispensing amount, predict and output U thickness non-uniformity prediction parameters, and calculate the mean value to obtain the first defect parameter; Use the insufficient defect analysis coefficient, combined with the number of N liquid crystal damage prediction branches in the pre-trained liquid crystal damage predictor, to calculate and obtain U; Randomly select U liquid crystal damage prediction branches, input the predicted dispensing volume, predict and output U liquid crystal damage prediction parameters, calculate the mean value, and obtain the second defect parameter.

[0060] Further, the defect prediction module 13 further includes the following execution steps: According to the LCD sealing data within the historical time, collect a set of sample predicted dispensing volumes, and collect the amplitude of uneven thickness of the LCD under different sample predicted dispensing volumes, and label to obtain a set of sample uneven thickness parameters; Randomly select with replacement to obtain the first uneven thickness training data from the set of sample predicted dispensing volumes and the set of sample uneven thickness parameters; Continue to randomly select with replacement to obtain N pieces of uneven thickness training data; Adopt deep learning to construct N uneven thickness prediction branches, and respectively use the N pieces of uneven thickness training data for supervised training and testing until the accuracy rate is qualified to complete the pre-training.

[0061] Further, the result generation module 14 includes the following execution steps: Generate the defect parameter of the LCD according to the first defect parameter and the second defect parameter; Use the defect parameter as the defect detection result and display a reminder.

[0062] Embodiment 3, please refer to Figure 3 , Figure 3 is the schematic diagram of the embodiment of the electronic device provided by the embodiment of the present invention. As Figure 3 shown, an electronic device 200 provided by an embodiment of the present invention includes a memory 210, a processor 220, and a computer program 211 stored on the memory 210 and executable on the processor 220. When the processor 220 executes the computer program 211, it implements the defect detection method for LCD display production in Embodiment 1.

[0063] It should be noted that in the above embodiments, the descriptions of each embodiment have their own emphases. For the parts not detailedly described in a certain embodiment, reference can be made to the relevant descriptions of other embodiments.

[0064] Those skilled in the art should understand that the embodiments of the present invention can be provided as a method, a system, or a computer program product. Therefore, the present invention can adopt the form of a complete hardware embodiment, a complete software embodiment, or an embodiment combining software and hardware aspects. Moreover, the present invention can adopt the form of a computer program product implemented on one or more computer-usable storage media (including but not limited to disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code.

[0065] The present invention is described with reference to the flowcharts and / or block diagrams of methods, apparatus (systems), and computer program products according to embodiments of the invention. It should be understood that each flow and / or block in the flowcharts and / or block diagrams, and combinations of flows and / or blocks in the flowcharts and / or block diagrams can be implemented by computer program instructions. These computer program instructions can be provided to the processors of general purpose computers, special purpose computers, embedded computers or other programmable data processing devices to produce a machine, such that the instructions executed by the processors of the computer or other programmable data processing devices create means for implementing the functions specified in one flow Figure 1 one or more flows and / or blocks Figure 1 or means for implementing the functions specified in one block or more blocks.

[0066] These computer program instructions can also be stored in a computer-readable memory that can direct a computer or other programmable data processing device to operate in a particular manner, such that the instructions stored in the computer-readable memory produce a manufacture including instruction means that implement the functions specified in one flow Figure 1 one or more flows and / or blocks Figure 1 or means for implementing the functions specified in one block or more blocks.

[0067] These computer program instructions can also be loaded onto a computer or other programmable data processing device, such that a series of operational steps are performed on the computer or other programmable device to produce a computer-implemented process, and thus the instructions executed on the computer or other programmable device provide steps for implementing the functions specified in one flow Figure 1 one or more flows and / or blocks Figure 1 or means for implementing the functions specified in one block or more blocks.

[0068] Although the preferred embodiments of the present invention have been described, additional changes and modifications can be made to these embodiments by those skilled in the art once they learn of the basic inventive concept.

[0069] Obviously, those skilled in the art can make various changes and modifications to the present invention without departing from the spirit and scope of the present invention. Thus, if these modifications and variations of the present invention fall within the scope of the present invention and its equivalent technologies, the present invention is also intended to include these modifications and variations.

Claims

1. A defect detection method for LCD display production, characterized in that: The method comprises: When sealing the LCD, collect the dispensing port image of the sealing dispensing equipment, predict the dispensing amount, and obtain the predicted dispensing amount; According to the predicted glue dispensing amount and the preset glue dispensing amount, a glue amount redundant defect analysis decision and a glue amount insufficient defect analysis decision are performed to obtain a redundant defect analysis coefficient and an insufficient defect analysis coefficient; According to the redundant defect analysis coefficient, based on the predicted glue dispensing amount, the uneven thickness of the LCD is predicted to obtain a first defect parameter; according to the insufficient defect analysis coefficient, based on the predicted glue dispensing amount, the damage of the solidified liquid crystal is predicted to obtain a second defect parameter; According to the first defect parameter and the second defect parameter, a defect parameter of the LCD is generated as a defect detection result.

2. The defect detection method for LCD display production according to claim 1, characterized in that: When sealing the LCD, collect the dispensing port image of the sealing dispensing equipment, predict the dispensing amount, and obtain the predicted dispensing amount, including: When sealing the LCD, collect the dispensing port image of the sealing dispensing equipment; The dispensing port image is input into a pre-trained dispensing amount predictor, and the predicted dispensing amount is obtained as output.

3. The defect detection method for LCD display production according to claim 2, characterized in that: The training steps of the dispensing amount predictor include: According to the LCD display production data in the historical period, a sample dispensing port image set is collected, and the dispensing amount under each sample dispensing port image is collected and marked as the sample dispensing amount to obtain a sample dispensing amount set; Based on convolutional neural network, a dispensing amount predictor is constructed; The sample dispensing port image set and the sample dispensing amount set are used to perform supervised training and testing on the dispensing amount predictor, and the pre-training is completed after the accuracy rate is qualified.

4. The defect detection method for LCD display production according to claim 1, characterized in that: According to the predicted dispensing amount and the preset dispensing amount, a glue amount redundant defect analysis decision and a glue amount insufficient defect analysis decision are made to obtain a redundant defect analysis coefficient and an insufficient defect analysis coefficient, including: Obtaining a preset dispensing amount for LCD sealing, and collecting a maximum dispensing amount error that deviates from the preset dispensing amount based on dispensing data in a historical period; Calculating the difference between the predicted dispensing amount and the preset dispensing amount, and calculating the ratio of the predicted dispensing amount to the maximum dispensing amount error to obtain an error coefficient; Obtaining preset redundant defect analysis coefficients and preset insufficient defect analysis coefficients; The sum of the preset redundant defect analysis coefficient and the error coefficient is calculated to obtain the redundant defect analysis coefficient, and the difference between the preset insufficient defect analysis coefficient and the error coefficient is calculated to obtain the insufficient defect analysis coefficient.

5. The defect detection method for LCD display production according to claim 1, characterized in that: According to the redundant defect analysis coefficient, based on the predicted glue dispensing amount, the uneven thickness of the LCD is predicted to obtain a first defect parameter, and according to the insufficient defect analysis coefficient, based on the predicted glue dispensing amount, the damage of the solidified liquid crystal is predicted to obtain a second defect parameter, including: The redundant defect analysis coefficient is used in combination with the number of N thickness unevenness prediction branches in the pre-trained thickness unevenness predictor to calculate and obtain U, where N is a positive integer and U is a positive integer less than or equal to N; Randomly select U thickness unevenness prediction branches, input the predicted glue dispensing amount, predict and output U thickness unevenness prediction parameters, calculate the mean, and obtain the first defect parameter; The insufficient defect analysis coefficient is used in combination with the number of N liquid crystal damage prediction branches in the pre-trained liquid crystal damage predictor to calculate and obtain U; Randomly select U liquid crystal damage prediction branches, input the predicted glue dispensing amount, predict and output U liquid crystal damage prediction parameters, calculate the mean, and obtain the second defect parameter.

6. The defect detection method for LCD display production according to claim 5, characterized in that: The pre-training step of the thickness unevenness predictor comprises: According to the LCD sealing data in the historical time, the sample predicted dispensing amount set is collected, and the amplitude of uneven thickness of LCD under different sample predicted dispensing amounts is collected, and the sample thickness uneven parameter set is obtained by marking; Randomly selecting with replacement from the sample prediction dispensing amount set and the sample thickness unevenness parameter set to obtain first thickness unevenness training data; Continue to randomly select with replacement to obtain N sets of training data with uneven thickness; Deep learning is used to construct N uneven thickness prediction branches, and the N uneven thickness training data are used to perform supervised training and testing respectively until the accuracy rate is qualified and the pre-training is completed.

7. The defect detection method for LCD display production according to claim 1, characterized in that: Generating a defect parameter of the LCD as a defect detection result according to the first defect parameter and the second defect parameter, including: Generate a defect parameter of the LCD according to the first defect parameter and the second defect parameter; The defect parameters are used as defect detection results for display and reminder.

8. A defect detection system for LCD display production, characterized in that: The method for detecting defects in LCD display screen production according to any one of claims 1 to 7 comprises: The glue dispensing amount prediction module is used to collect the glue dispensing port image of the sealing glue dispensing equipment when the LCD is sealed, predict the glue dispensing amount, and obtain the predicted glue dispensing amount; A glue quantity analysis module, used to make glue quantity redundant defect analysis decisions and glue quantity insufficient defect analysis decisions according to the predicted glue dispensing quantity and the preset glue dispensing quantity, and obtain redundant defect analysis coefficients and insufficient defect analysis coefficients; A defect prediction module, configured to predict uneven thickness of LCD based on the redundant defect analysis coefficient and the predicted glue dispensing amount to obtain a first defect parameter, and to predict damage of solidified liquid crystal based on the insufficient defect analysis coefficient and the predicted glue dispensing amount to obtain a second defect parameter; The result generating module is used to generate the defect parameter of the LCD as the defect detection result according to the first defect parameter and the second defect parameter.

9. An electronic device, characterized in that: include: Memory for storing computer software programs; A processor is used to read and execute the computer software program, thereby implementing the defect detection method for LCD display production described in any one of claims 1-7.

Images