TFT (Thin Film Transistor) material display manufacturing process method

Abstract

The invention relates to the technical field of TFT (Thin Film Transistor) material display manufacturing data processing, and provides a TFT material display manufacturing process method which is improved aiming at the problem that point defect detection is not timely in the prior art. And training a TFT material display manufacturing process control model and a point defect detection model by using the neural network model. In the production process, operation data are collected in real time and input into a point defect detection model, and when a point defect is detected, a result is analyzed immediately and process parameters are adjusted. Meanwhile, the control model is optimized through a genetic algorithm, and real-time adjustment and feedback of process parameters are achieved. The production efficiency and the quality of finished products are effectively improved, automatic defect detection and treatment are achieved, and remarkable beneficial effects are achieved.

Description

Technical Field

[0001] The present invention relates to the technical field of data processing for TFT material display production, and particularly to a process method for TFT material display production. Background Art

[0002] The TFT (Thin Film Transistor) material display production process is an advanced liquid crystal display manufacturing technology. It is based on the precise combination of two substrates - the color filter substrate and the TFT array substrate. On the TFT array substrate, millions of thin film transistors are fabricated through a series of complex processes such as cleaning, thin film deposition, photolithography, and etching. These transistors act as circuit switches and are responsible for transmitting image signals to each pixel. The color filter substrate is responsible for providing the three primary colors of red, green, and blue light, and realizes color display by precisely controlling the light transmittance. When a voltage is applied to the TFT array substrate, the thin film transistors conduct or cut off, controlling the orientation and light transmittance of the liquid crystal molecules, thereby displaying an image. TFT displays are widely used in mobile phones, tablet computers, televisions, and various electronic devices due to their advantages such as high resolution, high brightness, wide viewing angle, and fast response speed, bringing users a clear and vivid visual experience.

[0003] The TFT material display production process has the following technical pain points in historical data processing. Point defects are common quality problems on TFT displays, manifested as abnormal pixel writing or holding functions, and abnormal brightness that cannot be controlled by the driving voltage. In the existing TFT material display production process, point defects on the TFT display cannot be detected in a timely manner during the production process. To solve this technical problem, the present invention provides a process method for TFT material display production. Summary of the Invention

[0004] Aiming at the deficiencies of the prior art, the present invention provides a process method for TFT material display production, which solves the problem that in the existing TFT material display production process, point defects on the TFT display cannot be detected in a timely manner.

[0005] To solve the above technical problems, the specific technical solution of the present invention is as follows: The present invention provides a process method for TFT material display production, including: Step S101, obtaining the historical data of the TFT material display production process. The historical data of the TFT material display production process includes substrate historical data, thin film transistor production historical data, color filter production historical data, point defect detection historical data, defect repair historical data, substrate combination historical data, final detection historical data, production environment historical data, and quality control historical data; Step S102: Divide the substrate historical data, thin film transistor manufacturing historical data, color film manufacturing historical data, substrate combination historical data, production environment historical data, and quality control historical data into a control model training set and a control model validation set. Use the control model training set and the control model validation set to train a neural network model to obtain a TFT material display manufacturing process control model; Step S103: Divide the point defect detection historical data, defect repair historical data, and final detection historical data into a detection model training set and a detection model validation set. Use the detection model training set and the detection model validation set to train a neural network model to obtain a point defect detection model; Step S104: Collect the TFT material display manufacturing operation data, substitute the TFT material display manufacturing operation data into the point defect detection model to obtain a point defect detection result. The point defect detection result includes the existence of point defects and the non-existence of point defects. If there are point defects, go to step S105; Step S105: Analyze the point defect detection result to obtain the analyzed point defect detection result. Substitute the analyzed point defect detection result into the TFT material display manufacturing process control model to obtain the TFT material display manufacturing process control parameters. Use the TFT material display manufacturing process control parameters as the real-time TFT material display manufacturing process adjustment parameters and transmit the real-time TFT material display manufacturing process adjustment parameters to the TFT material display manufacturing equipment side.

[0006] Further, in the TFT material display manufacturing process method of the present invention, step S105 further includes: Receive the operation data of the TFT material display manufacturing equipment side executing the real-time TFT material display manufacturing process adjustment parameters to obtain the feedback data of the TFT material display manufacturing equipment side; Substitute the feedback data of the TFT material display manufacturing equipment side into the point defect detection model. If the point defect detection result output by the point defect detection model is the non-existence of point defects, stop detecting the feedback data of the TFT material display manufacturing equipment side; If the point defect detection result output by the point defect detection model is the existence of point defects, optimize the TFT material display manufacturing process control model using a genetic algorithm.

[0007] Further, the TFT material display manufacturing process method of the present invention further includes: Optimize the TFT material display manufacturing process control model using a genetic algorithm to determine the optimization objective; Set the basic parameters of the genetic algorithm. The basic parameters of the genetic algorithm include population size, number of generations, crossover probability, and mutation probability; Initialize the population, randomly generate a set of TFT material display manufacturing process parameters as the initial population, and each parameter combination represents an adjustment parameter, where the adjustment parameters include temperature, pressure, time, and material ratio; Use the point defect detection model to evaluate each solution in the initial population, calculate its fitness value, and select some excellent individuals as parents for crossover and mutation operations according to the fitness value.

[0008] Furthermore, the TFT material display manufacturing process method of the present invention further includes: Crossover: perform a crossover operation on the selected parents to generate new offspring solutions; Mutation: perform a mutation operation on the offspring solutions; Use the point defect detection model to evaluate the newly generated offspring solutions, calculate their fitness values, update the population, and replace part or all of the old population with the newly evaluated offspring solutions to form a new generation population; Check whether the termination condition of the genetic algorithm is reached. If the termination condition is met, select the solution with the highest fitness value from the population as the optimal process parameter combination and output it to the TFT material display manufacturing equipment side for application.

[0009] Furthermore, in the TFT material display manufacturing process method of the present invention, step S101 includes: The substrate historical data includes substrate type, substrate size, substrate batch number, substrate production date, substrate surface condition, and substrate material characteristics; The thin film transistor manufacturing historical data includes thin film deposition parameters, photolithography parameters, etching parameters, transistor size, transistor position information, and transistor electrical performance parameters; The color film manufacturing historical data includes filter color, filter thickness, filter light transmittance, filter position information, and filter uniformity; The point defect detection historical data includes defect type, defect position, defect size, defect quantity, and detection time; The defect repair historical data includes repair method, repair result, repair time, and post-repair performance verification; The substrate combination historical data includes combination method, combination accuracy, overall size after combination, and combination time; The final inspection historical data includes display screen resolution, display screen brightness, display screen contrast, display screen color performance, display screen response time, display screen viewing angle, and display screen overall appearance; The production environment historical data includes production workshop temperature, production workshop humidity, production equipment status, operator information, and production batch number: The quality control historical data includes quality control standards, quality control results, and quality control records.

[0010] Further, for the TFT material display manufacturing process method of the present invention, the step S102 includes: For substrate historical data, thin-film transistor manufacturing historical data, color filter manufacturing historical data, substrate combination historical data, production environment historical data, and quality control historical data, use a random number generator or a data partitioning tool to randomly partition the data into a training set and a validation set, and perform statistics on the partitioned training set and validation set to obtain a control model training set and a control model validation set.

[0011] Further, for the TFT material display manufacturing process method of the present invention, the step S102 includes: Preprocess the control model training set and the control model validation set, and convert the data into a format that can be processed by a neural network; Input the preprocessed control model training set into the neural network for model training. Calculate the predicted value through forward propagation, update the network weights through backward propagation, and iterate for multiple epochs until the loss function converges or reaches the preset number of training epochs; After each training epoch, use the control model validation set to validate the neural network model, and save the trained neural network model.

[0012] Further, for the TFT material display manufacturing process method of the present invention, the step S103 includes: Receive the data volume and the requirements for model training. According to the data volume and the requirements for model training, determine the partitioning ratio of the training set and the validation set, select the random partitioning method, randomly divide the data into a detection model training set and a detection model validation set, and store the partitioned detection model training set and detection model validation set in the database respectively.

[0013] Further, for the TFT material display manufacturing process method of the present invention, the step S103 includes: Preprocess the detection model training set and the detection model validation set, and convert the data into a format that can be processed by a neural network; Input the preprocessed detection model training set into the neural network to start model training, calculate the predicted value through forward propagation, and calculate the gradient according to the loss function through backward propagation to update the weights, and iterate for multiple epochs until the loss function converges or reaches the preset number of training epochs; After each training epoch, use the detection model validation set to validate the model, evaluate the saved model using the detection model validation set. If the model performance meets the expectations, it can be deployed to the point defect detection system to achieve automated detection.

[0014] Further, in the TFT material display manufacturing process method of the present invention, step S104 includes: Collect operation data in real time from the production line of TFT material display manufacturing. The operation data collected in real time includes substrate information, thin film transistor manufacturing parameters, color film manufacturing parameters, combination process parameters, production environment data, and factors affecting product quality. Perform necessary preprocessing on the collected TFT material display manufacturing operation data, load a trained and validated point defect detection model, and substitute the preprocessed TFT material display manufacturing operation data into the point defect detection model. The point defect detection model calculates based on the input data and outputs the result of point defect detection. Check the result of point defect detection output by the point defect detection model. If the result indicates the existence of point defects, there are point defects in the product, and the process should enter step S105 for subsequent processing. If the result indicates the non-existence of point defects, no point defects are detected in the individual product.

[0015] Advantages of the present invention: By collecting real-time operation data during the manufacturing process of TFT material displays and inputting it into a trained point defect detection model, the present invention can predict and identify point defects in the production process in real time, thereby discovering potential quality problems earlier and reducing the generation of defective products. Through an automated defect detection system, the need for manual intervention is reduced, and the possibility of human errors is lowered. When point defects are detected, the system can automatically trigger subsequent processing procedures, such as marking, isolation, or rework.

[0016] Based on the point defect detection results, the present invention inputs the parsed point defect information into the TFT material display manufacturing process control model, obtains real-time adjustment parameters, and transmits them to the production equipment side. The present invention uses a genetic algorithm to optimize the TFT material display manufacturing process control model. Through continuous iteration and evolution, the prediction accuracy of the model is improved. At the same time, the feedback data is input into the point defect detection model again for verification, forming a closed-loop optimization mechanism. Through real-time monitoring, automated detection, and process parameter optimization, the present invention significantly reduces waste and defective product rates in the production process, improves production efficiency and finished product quality, and reduces production costs and manpower requirements.

[0017] The present invention comprehensively collects and analyzes historical data and real-time data of the TFT material display manufacturing process, providing rich data resources for enterprises. The data is not only used for model training and optimization but also provides strong support for the production decisions of enterprises. The present invention significantly improves the quality control level and production efficiency in the manufacturing process of TFT displays, bringing significant economic benefits and competitive advantages to enterprises. Description of the Drawings

[0018] To more clearly illustrate the technical solution of the present invention, the following will briefly introduce the drawings required in the embodiments. Obviously, for those of ordinary skill in the art, without creative efforts, other drawings can also be obtained based on the drawings.

[0019] Figure 1 It is a schematic flowchart of the manufacturing process method for TFT material display provided by the embodiment of the present invention. Specific Embodiments

[0020] To make the objectives, technical solutions, and advantages of the present invention clearer, the following will clearly and completely describe the technical solutions of the present invention in conjunction with specific embodiments and corresponding drawings of the present invention. Obviously, the described embodiments are only a part of the embodiments of the present invention, rather than all of them. Based on the embodiments of the present invention, all other embodiments obtained by those of ordinary skill in the art without creative efforts belong to the scope of protection of the present invention. The following will detail the technical solutions provided by each embodiment of the present invention with reference to the drawings. To better understand the objectives of the present invention, the following will further describe the present invention in detail.

[0021] The present invention provides a manufacturing process method for TFT material display, including: Step S101, obtaining historical data of the manufacturing process for TFT material display. The historical data of the manufacturing process for TFT material display includes substrate historical data, thin film transistor manufacturing historical data, color filter manufacturing historical data, point defect detection historical data, defect repair historical data, substrate combination historical data, final inspection historical data, production environment historical data, and quality control historical data; The substrate historical data includes the type, specification, material, supplier information, production date, batch number of the substrate, and various physical and chemical property data of the substrate during the manufacturing process. It is extracted from the database of the substrate supplier or the production department, or collected in real time through sensors and monitoring systems related to substrate production.

[0022] The thin film transistor manufacturing historical data includes the manufacturing parameters of the thin film transistor (such as deposition temperature, pressure, time, etc.), equipment information, operator information, environmental data during the manufacturing process (such as temperature, humidity), and quality data of the manufacturing results.

[0023] It is extracted from the monitoring system of the thin film transistor production line, or collected in real time through data recording devices.

[0024] The historical data of color film production includes the color, thickness, uniformity, production parameters (such as coating speed, curing temperature, etc.), material information, and quality data during the production process of the color film. It is extracted from the monitoring system of the color film production line or collected in real time through sensors and data recording devices related to color film production.

[0025] Historical data of point defect detection: Record detailed information such as the type, location, size, quantity, detection time, and detection equipment of point defects, as well as defect images and detection reports. It is extracted from the database of the point defect detection equipment or directly output by the detection equipment.

[0026] The historical data of defect repair includes the method of defect repair, repair time, repair result, parameter records during the repair process, and information of the repair operator. It is extracted from the defect repair record system or collected in real time through the monitoring system of the repair equipment.

[0027] Historical data of substrate combination: Record the process parameters of substrate combination (such as alignment accuracy, lamination pressure, temperature, etc.), quality data of the combination result, and operation data of the combination equipment. It is extracted from the monitoring system of the substrate combination production line or collected in real time through data recording devices.

[0028] The historical data of final inspection includes the quality inspection data of the final product (such as display effect, brightness, contrast, color saturation, etc.), inspection time, inspection equipment, and the determination of the inspection result. It is extracted from the database of the final inspection equipment or directly output by the inspection equipment.

[0029] Historical data of production environment: Record the environmental parameters during the production process (such as temperature, humidity, cleanliness, etc.) and the operation data of environmental monitoring equipment. It is extracted from the environmental monitoring system or collected in real time through environmental sensors.

[0030] The historical data of quality control includes various data during the quality control process, such as quality control standards, inspection frequencies, records of handling non-conforming products, information of quality control personnel, etc. It is extracted from the quality control record system or obtained through the database of the quality control department.

[0031] By comprehensively collecting and organizing these historical data, it can provide strong data support for the improvement and optimization of the TFT material display production process, thereby improving production efficiency and product quality.

[0032] Step S102, divide the substrate historical data, thin film transistor production historical data, color film production historical data, substrate combination historical data, production environment historical data, and quality control historical data into a control model training set and a control model validation set, and use the control model training set and the control model validation set to train a neural network model to obtain a TFT material display production process control model; The neural network model is trained using a three - layer fully connected neural network. The number of neurons in the input layer is the same as the feature dimension, the number of neurons in the hidden layer is 128, the activation function is ReLU, and the output layer is the Sigmoid function.

[0033] The collected substrate historical data, thin - film transistor manufacturing historical data, color - film manufacturing historical data, substrate combination historical data, production environment historical data, and quality control historical data are sorted and cleaned to improve data quality and consistency. Then, these data are randomly divided into two parts: the control model training set and the control model validation set.

[0034] The control model training set is used to train the neural network model so that it can learn the rules and characteristics of the TFT material display manufacturing process. The training set should contain enough data samples to enable the model to fully learn and generalize to new data.

[0035] The control model validation set is used to verify the performance and accuracy of the trained model. The validation set does not participate in the model training process and is only used to evaluate and test the model after training.

[0036] The neural network model is trained using the control model training set. The training process includes the following key steps: According to the complexity of the TFT material display manufacturing process and data characteristics, define a suitable neural network structure, including the number of layers, neurons, and activation functions of the input layer, hidden layer, and output layer. Pre - process the data in the training set, such as data normalization, feature extraction, etc., to make it suitable for the input requirements of the neural network. Input the pre - processed data into the neural network and calculate the output value of the network through forward propagation. Use a suitable loss function (such as mean squared error, cross - entropy, etc.) to calculate the difference between the network output value and the true value, that is, the loss value. Calculate the gradients of the parameters in each layer of the network according to the loss value and update the network weights through the backpropagation algorithm to reduce the loss value. Repeat the above process of forward propagation, loss calculation, and backpropagation until the loss function converges or reaches the preset number of training epochs.

[0037] After training, the control model validation set is used to verify the trained model. By calculating the performance metrics (such as accuracy, recall, F1 - score, etc.) of the model on the validation set, evaluate the performance and accuracy of the model. If the performance of the model meets the expected requirements, it can be used as the TFT material display manufacturing process control model.

[0038] In summary, through key steps such as data division, model training, and model validation in step S102, the neural network model is used to learn and model the historical data of the TFT material display manufacturing process, thereby obtaining the TFT material display manufacturing process control model.

[0039] In step S103, divide the point defect detection historical data, defect repair historical data, and final detection historical data into a detection model training set and a detection model validation set, and use the detection model training set and the detection model validation set to train a neural network model to obtain a point defect detection model. From the collected point defect detection historical data, defect repair historical data, and final detection historical data, carefully perform data cleaning and sorting to ensure the accuracy and integrity of the data. Subsequently, these data are systematically divided into two independent parts: a detection model training set and a detection model validation set.

[0040] Detection model training set: This data set is specifically used to train the neural network model to help the model learn how to effectively identify the point defect features on the TFT display screen. It should contain diverse defect samples so that the model can comprehensively learn and adapt to various defect situations.

[0041] Detection model validation set: This data set is used to verify the performance of the trained model to ensure the accuracy and reliability of the model in actual applications. It does not participate in the model training process and is only used to evaluate the detection effect of the model after the model training is completed.

[0042] Use the detection model training set to conduct special training on the neural network model to build a point defect detection model.

[0043] The training process is as follows: According to the specific requirements and data characteristics of point defect detection, carefully design the architecture of the neural network, including determining the structure of the input layer, hidden layer (possibly including multiple layers), and output layer, as well as selecting appropriate activation functions, etc. Perform necessary preprocessing on the data in the training set, such as data standardization, feature extraction, and enhancement, etc., to make the data meet the input standards of the neural network and improve the detection ability of the model.

[0044] Through the forward propagation algorithm, input the preprocessed data into the neural network and calculate the output of the network. Subsequently, use an appropriate loss function (for example, the cross-entropy loss function can be used for binary classification problems) to quantify the gap between the network output and the true label. Then, through the backpropagation algorithm, calculate the gradients of the network parameters according to the loss value and update the network weights to gradually reduce the loss.

[0045] Continuously repeat the above processes of forward propagation, loss calculation, and backpropagation until the loss function converges to a stable value or reaches the preset number of training epochs. During this process, hyperparameters such as the learning rate and optimizer can be adjusted in a timely manner to further improve the training effect of the model.

[0046] After the training is completed, use the validation set of the detection model to verify and evaluate the model. By calculating metrics such as the accuracy, recall rate, and F1 score of the model on the validation set, comprehensively evaluate the detection performance of the model. If the model performance meets the expectations, it can be deployed to the actual production environment for real-time monitoring and prediction of point defects.

[0047] Step S104, collect the TFT material display production operation data, substitute the TFT material display production operation data into the point defect detection model, and obtain the point defect detection result. The point defect detection result includes the existence of point defects and the non-existence of point defects. If there are point defects, go to step S105; Through various sensors, monitoring devices, and data recording systems, collect the operation data during the production of the TFT material display in real time. These data should cover all key parameters related to point defect detection, such as the substrate status, environmental parameters during the production of thin film transistors, color film production quality, substrate combination accuracy, etc.

[0048] Before substituting the collected operation data into the point defect detection model, it may be necessary to perform certain preprocessing on the data. This includes data cleaning (removing outliers, filling missing values, etc.), data standardization (converting the data into a format and range acceptable to the model), and feature extraction (extracting features useful for point defect detection from the original data).

[0049] Substitute the preprocessed TFT material display production operation data into the already trained point defect detection model. The model will, based on the input data, utilize the knowledge and rules it learned during the training process to predict the point defect situation in the current production process.

[0050] Based on the output of the model, determine whether there are point defects in the current production process. There are usually two possibilities for the point defect detection result: No point defects: If the model prediction result indicates that there are no point defects in the current production process, then the production can continue, and at the same time, continue to monitor and collect data for subsequent further analysis.

[0051] There are point defects: If the model prediction result indicates that there are point defects in the current production process, then it is necessary to immediately enter the next step (step S105) for processing.

[0052] If there are point defects, go to step S105. This step may include detailed analysis, location, recording of the defects, and subsequent repair work. At the same time, the relevant information about this point defect (such as defect type, location, size, etc.) can also be added to the point defect detection historical data for subsequent further training and optimization of the model.

[0053] Through the real-time monitoring and prediction in step S104, point defect problems in the TFT material display manufacturing process can be detected in a timely manner.

[0054] Step S105: Analyze the point defect detection results to obtain the analyzed point defect detection results. Substitute the analyzed point defect detection results into the TFT material display manufacturing process control model to obtain the TFT material display manufacturing process control parameters. Use the TFT material display manufacturing process control parameters as the real-time TFT material display manufacturing process adjustment parameters and transmit the real-time TFT material display manufacturing process adjustment parameters to the TFT material display manufacturing equipment side.

[0055] Receive the operation data of the TFT material display manufacturing equipment side executing the real-time TFT material display manufacturing process adjustment parameters to obtain the feedback data of the TFT material display manufacturing equipment side; Substitute the feedback data of the TFT material display manufacturing equipment side into the point defect detection model. If the point defect detection result output by the point defect detection model is that there are no point defects, stop detecting the feedback data of the TFT material display manufacturing equipment side; If the point defect detection result output by the point defect detection model is that there are point defects, optimize the TFT material display manufacturing process control model using the genetic algorithm.

[0056] Optimize the TFT material display manufacturing process control model using the genetic algorithm and determine the optimization goal; Set the basic parameters of the genetic algorithm. The basic parameters of the genetic algorithm include population size, number of generations, crossover probability, and mutation probability; Initialize the population, randomly generate a set of TFT material display manufacturing process parameters as the initial population. Each parameter combination represents adjustment parameters, and the adjustment parameters include temperature, pressure, time, and material ratio; Use the point defect detection model to evaluate each solution in the initial population, calculate its fitness value, and select some excellent individuals as parents for crossover and mutation operations according to the fitness value.

[0057] Crossover: Perform crossover operations on the selected parents to generate new offspring solutions; Mutation: Perform mutation operations on the offspring solutions; Use the point defect detection model to evaluate the newly generated offspring solutions, calculate their fitness values, update the population, and replace part or all of the old population with the newly evaluated offspring solutions to form a new generation population; Check whether the termination condition of the genetic algorithm is reached. When the standard deviation of the population fitness is less than 0.01 or the number of iterations reaches 500 generations, terminate the algorithm. If the termination condition is met, select the solution with the highest fitness value from the population as the optimal process parameter combination and output it to the TFT material display manufacturing equipment for application.

[0058] Analysis of point defect detection results: Analyze the output of the point defect detection model in detail to clarify the type of defects (such as bright spots, dark spots, color spots, etc.), location, size, and possible causes (such as material inhomogeneity, improper process parameters, etc.). Save the analyzed point defect detection results in a structured format for subsequent analysis and processing.

[0059] Obtaining process control parameters: Take the analyzed point defect detection results as input and substitute them into the TFT material display manufacturing process control model. The model calculates the adjusted process control parameters, such as temperature, pressure, time, material ratio, etc., based on the defect situation, combined with historical data and process knowledge.

[0060] Take the calculated process control parameters as real-time adjustment parameters and transmit them to the TFT material display manufacturing equipment through the communication interface. After receiving the parameters, the equipment automatically or manually adjusts to the new process parameters to ensure the stability of the production process and product quality.

[0061] Collect the operation data of the TFT material display manufacturing equipment in real time after executing the new process parameters, including production status, product quality, equipment performance, etc. Substitute the feedback data into the point defect detection model to perform defect detection again to verify the effectiveness of the new process parameters.

[0062] If the result output by the point defect detection model shows that there are no point defects, it means that the new process parameters are effective, and the detection of the current feedback data can be stopped, and the production process can continue to be monitored. If there are still point defects, start the genetic algorithm to optimize the TFT material display manufacturing process control model.

[0063] Genetic algorithm optimization process: Determine the optimization objective: Clearly define that the optimization objective is to reduce or eliminate point defects and improve product quality.

[0064] Set basic parameters: Reasonably set the basic parameters of the genetic algorithm according to the problem complexity and computing resources, such as population size (i.e., the number of solutions considered simultaneously), number of generations (i.e., the number of iterations), crossover probability (i.e., the probability of generating offspring by crossover of parents), and mutation probability (i.e., the probability of offspring mutation).

[0065] Initialize the population: Randomly generate a set of TFT material display manufacturing process parameters as the initial population, and each parameter combination represents a possible solution.

[0066] Evaluation and Selection: Use the point defect detection model to evaluate each solution in the initial population and calculate its fitness value (i.e., the degree of defect reduction). Select some excellent individuals as parents according to the fitness value for subsequent crossover and mutation operations.

[0067] Crossover Operation: Perform crossover operations on the selected parents to generate new offspring solutions by combining gene segments of different parents, so as to increase the diversity of solutions.

[0068] Mutation Operation: Perform mutation operations on the offspring solutions to explore new solution spaces by randomly changing some gene values.

[0069] Update Population: Use the point defect detection model to evaluate the newly generated offspring solutions, calculate their fitness values, and update the population according to the fitness values. Replace some or all of the individuals with lower fitness values in the old population with the newly evaluated offspring solutions to form a new generation population.

[0070] Termination Condition Check: Check whether the termination conditions of the genetic algorithm are met, such as reaching the maximum number of generations, convergence of fitness values, or finding a solution that meets the requirements. If the termination conditions are met, select the solution with the highest fitness value from the population as the optimal process parameter combination.

[0071] Application of Optimal Process Parameters: Output the optimized optimal process parameter combination to the TFT material display production equipment for application. The equipment adjusts according to the new process parameters and continuously monitors the production process to steadily improve the product quality.

[0072] Specifically, for the TFT material display production process method described in the present invention, the step S101 includes: The substrate historical data includes substrate type, substrate size, substrate batch number, substrate production date, substrate surface condition, and substrate material characteristics; The thin film transistor production historical data includes thin film deposition parameters, lithography parameters, etching parameters, transistor size, transistor position information, and transistor electrical performance parameters; The color filter production historical data includes filter color, filter thickness, filter transmittance, filter position information, and filter uniformity; The point defect detection historical data includes defect type, defect position, defect size, defect quantity, and detection time; The defect repair historical data includes repair method, repair result, repair time, and performance verification after repair; The substrate combination historical data includes combination method, combination accuracy, overall size after combination, and combination time; The final inspection historical data includes the display screen resolution, display screen brightness, display screen contrast, display screen color performance, display screen response time, display screen viewing angle, and the overall appearance of the display screen; The production environment historical data includes the temperature of the production workshop, humidity of the production workshop, status of production equipment, operator information, and production batch number: The quality control historical data includes quality control standards, quality control results, and quality control records.

[0073] Substrate historical data: Substrate type: Record the type of substrate used, such as glass substrate, plastic substrate, etc., as well as the model and specifications of the substrate.

[0074] Substrate size: Accurately measure and record the length, width, and thickness of the substrate to make the substrate size meet the production requirements.

[0075] Substrate batch number: Assign a unique batch number to each batch of substrates for easy tracking and traceability.

[0076] Substrate production date: Record the production date of the substrate to understand the storage time and aging condition of the substrate.

[0077] Substrate surface condition: Describe the cleanliness, flatness, scratch condition, etc. of the substrate surface to make the substrate surface quality meet the production requirements.

[0078] Substrate material properties: Record the physical and chemical properties of the substrate material, such as heat resistance, corrosion resistance, light transmittance, etc., to provide a basis for setting subsequent process parameters.

[0079] Historical data of thin - film transistor fabrication: Thin - film deposition parameters: Detail and record parameters such as temperature, pressure, deposition rate, deposition time, etc. during the thin - film deposition process to ensure the quality and thickness uniformity of the thin film.

[0080] Lithography parameters: Include the type of photoresist, coating thickness, exposure time, development time, etc. to ensure the accuracy and clarity of the lithography pattern.

[0081] Etching parameters: Record the type, concentration, etching time, etc. of the etching solution to control the depth and shape of the etching.

[0082] Transistor size: Accurately measure the length, width, and height of the transistor to make the transistor size meet the design requirements.

[0083] Transistor position information: Record the specific position of the transistor on the substrate for easy subsequent circuit connection and detection.

[0084] Electrical performance parameters of transistors: Test and record electrical performance parameters such as the current-voltage characteristics, switching speed, breakdown voltage of transistors to ensure that the performance of transistors meets the usage requirements.

[0085] Historical data of color film production: Color of the filter: Record the color of the filter, such as red, green, blue, etc., to ensure the color accuracy of the color film.

[0086] Thickness of the filter: Measure and record the thickness of the filter to ensure the stable optical performance of the filter.

[0087] Transmittance of the filter: Test the transmittance of the filter to meet the brightness requirements of the display screen.

[0088] Position information of the filter: Record the position of the filter on the substrate to ensure the correct correspondence between the color film and the transistors.

[0089] Uniformity of the filter: Evaluate the distribution uniformity of the filter on the substrate to avoid color deviation.

[0090] Historical data of point defect detection: Type of defect: Classify and record the types of point defects, such as bright spots, dark spots, color spots, etc.

[0091] Position of defect: Accurately record the position coordinates of the defect on the display screen for subsequent repair and verification.

[0092] Size of defect: Measure and record the size of the defect, such as diameter, area, etc.

[0093] Number of defects: Count the number of point defects in each batch of products to evaluate the stability of the production process.

[0094] Detection time: Record the time of defect detection to trace the link and cause of defect generation.

[0095] Historical data of defect repair: Repair method: Record the repair methods adopted, such as laser repair, chemical repair, etc.

[0096] Repair result: Evaluate and record the effect after repair, such as whether the defect is completely eliminated, whether new defects are generated, etc.

[0097] Repair time: Record the time of the repair operation to evaluate the repair efficiency.

[0098] Performance verification after repair: Conduct performance tests on the repaired products to ensure that the repair operation has no negative impact on the product performance.

[0099] Historical data of substrate combination: Combination method: Record the combination method of the substrates, such as single-piece combination, multi-piece combination, etc.

[0100] Combined accuracy: Measure and record the accuracy after the substrates are combined to ensure that the combined product meets the design requirements.

[0101] Overall size after combination: Measure the overall size of the product after combination to meet the requirements of packaging and transportation.

[0102] Combination time: Record the time of substrate combination to track the production progress.

[0103] Final inspection historical data: Display screen resolution: Test and record the resolution of the display screen to make the image clear and delicate.

[0104] Display screen brightness: Measure the brightness of the display screen to meet the visibility requirements in the usage environment.

[0105] Display screen contrast ratio: Test the contrast ratio of the display screen to make the color levels distinct.

[0106] Display screen color performance: Evaluate the color reproduction and saturation of the display screen.

[0107] Display screen response time: Measure the response time of the display screen to meet the requirements of dynamic image display.

[0108] Display screen viewing angle: Test the viewing angle range of the display screen to obtain good visual effects at different angles.

[0109] Overall appearance of the display screen: Inspect the appearance quality of the display screen, such as the frame, surface flatness, scratches, etc.

[0110] Production environment historical data: Temperature in the production workshop: Record the temperature in the production workshop to make the production environment meet the equipment requirements.

[0111] Humidity in the production workshop: Measure the humidity in the production workshop to avoid the influence of too high or too low humidity on product quality.

[0112] Status of production equipment: Record the operating status, maintenance records, and fault conditions of the production equipment.

[0113] Operator information: Record the name, employee number, training status, etc. of the operator to trace the responsibilities and problems in the production process.

[0114] Production batch number: Assign a unique batch number to each batch of products for easy tracking and recall.

[0115] Quality control historical data: Quality control standards: Establish and record the quality control standards and procedures to ensure the controllability and consistency of the production process.

[0116] Quality control results: Record the results of quality control inspections, such as the passing rate of spot checks, the handling of non-conforming products, etc.

[0117] Quality control records: Detail the inspection data, analysis reports, and improvement measures during the quality control process for continuous improvement of production quality.

[0118] Specifically, for the TFT material display manufacturing process method described in the present invention, the step S102 includes: For substrate historical data, thin-film transistor manufacturing historical data, color filter manufacturing historical data, substrate combination historical data, production environment historical data, and quality control historical data, use a random number generator or data partitioning tool to randomly partition the data into a training set and a validation set, and perform statistics on the partitioned training set and validation set to obtain a control model training set and a control model validation set.

[0119] Preprocess the control model training set and the control model validation set, and convert the data into a format that can be processed by a neural network; Input the preprocessed control model training set into the neural network for model training. Calculate the predicted values through forward propagation, update the network weights through backward propagation, and iterate for multiple epochs until the loss function converges or reaches the preset number of training epochs; After each training epoch, use the control model validation set to validate the neural network model and save the trained neural network model.

[0120] In the TFT material display manufacturing process, to establish an accurate quality control model, it is necessary to effectively partition, preprocess, and train the collected historical data. Specifically, the step S102 in the TFT material display manufacturing process method described in the present invention includes the following key steps: Random partitioning: Use a random number generator or data partitioning tool (such as the train_test_split function in scikit-learn) to randomly partition the substrate historical data, thin-film transistor manufacturing historical data, color filter manufacturing historical data, substrate combination historical data, production environment historical data, and quality control historical data into a training set and a validation set. The partitioning ratio can be determined according to the actual situation. Usually, the training set accounts for a larger proportion (such as 70%-80%), and the validation set accounts for a smaller proportion (such as 20%-30%).

[0121] Perform statistics on the partitioned training set and validation set to confirm the consistency of the data distribution, so that the training set and validation set can represent the characteristics of the overall data set. At the same time, generate a control model training set and a control model validation set for subsequent model training and validation.

[0122] Data preprocessing: Check for missing values, outliers, or duplicate values in the dataset and perform corresponding processing. For missing values, methods such as filling (e.g., mean filling, median filling) or deletion can be used; for outliers, they can be corrected or deleted according to business logic; for duplicate values, they are directly deleted to avoid affecting model training.

[0123] According to business requirements and model requirements, perform feature extraction and transformation on the original data. For example, continuous variables can be discretized, categorical variables can be encoded (such as one-hot encoding), and feature scaling (such as normalization, standardization) can be performed, etc., to improve the performance and stability of the model.

[0124] Convert the preprocessed data into a format that can be processed by the neural network, such as the tensor form, and make the dimensions and types of the data match the input requirements of the neural network model.

[0125] According to the task requirements and data characteristics, select a suitable neural network architecture (such as convolutional neural network CNN, recurrent neural network RNN, fully connected neural network, etc.), and set parameters such as the number of network layers, the number of neurons, and activation functions.

[0126] Forward propagation: Input the preprocessed control model training set into the neural network and obtain the predicted values through layer-by-layer calculation.

[0127] Loss calculation: Use a suitable loss function (such as mean squared error MSE, cross-entropy loss, etc.) to calculate the difference between the predicted values and the true values.

[0128] Backward propagation: According to the gradient of the loss function, update the weights and biases of the neural network through the backward propagation algorithm to reduce the prediction error.

[0129] Iterative training: Repeat the processes of forward propagation, loss calculation, and backward propagation, and iteratively train for multiple epochs until the loss function converges or reaches the preset number of training rounds.

[0130] Hyperparameter tuning: During the training process, methods such as grid search, random search, or Bayesian optimization can be used to tune the hyperparameters of the neural network (such as learning rate, batch size, regularization parameter, etc.) to improve the performance of the model.

[0131] Model validation: After each training epoch, use the control model validation set to validate the neural network model and evaluate the generalization ability of the model. The performance of the model can be evaluated by calculating metrics such as loss, accuracy, and F1 score on the validation set.

[0132] Model selection: According to the validation results, select the model with the best performance as the final quality control model.

[0133] Model saving: Save the trained neural network model in a specific format (such as.h5,.pth, etc.) for subsequent loading and inference. At the same time, save the training parameters and configuration information of the model for model reproduction and update.

[0134] Specifically, for the TFT material display manufacturing process method described in the present invention, step S103 includes: Receive the data volume and the requirements for model training. According to the data volume and the requirements for model training, determine the division ratio of the training set and the validation set. Select the random division method to randomly divide the data into a detection model training set and a detection model validation set, and store the divided detection model training set and detection model validation set in the database respectively.

[0135] Preprocess the detection model training set and the detection model validation set, and convert the data into a format that can be processed by the neural network; input the preprocessed detection model training set into the neural network to start model training. Calculate the predicted value through forward propagation, and calculate the gradient according to the loss function through backward propagation to update the weights. Iteratively train for multiple epochs until the loss function converges or reaches the preset number of training rounds. After each training epoch, use the detection model validation set to verify the model. Evaluate the saved model on the detection model validation set. If the model performance meets the expectations, it can be deployed to the point defect detection system to achieve automated detection.

[0136] Data division: Receive the specific requirements regarding the data volume and model training, including information such as the expected model accuracy, training time, available data volume, etc.

[0137] Determination of division ratio: According to the data volume and the requirements for model training, determine the division ratio of the training set and the validation set. Usually, the training set accounts for a relatively large proportion (such as 60%-80%) for model training; the validation set accounts for a relatively small proportion (such as 20%-40%) for model verification and performance evaluation.

[0138] Random division: Select the random division method to make the distribution of data in the training set and the validation set uniform. A random number generator or data division tools (such as functions in Python's NumPy library or scikit-learn library) can be used to achieve the random division of data.

[0139] Data storage: Store the divided detection model training set and detection model validation set in the database respectively for subsequent data processing and model training.

[0140] Clean the detection model training set and the validation set to remove missing values, outliers, or duplicate values to improve the data quality.

[0141] According to the task requirements of point defect detection, extract useful features and perform necessary feature transformations, such as normalization, standardization, one-hot encoding, etc., to improve the performance of the model.

[0142] Convert the data into a format that can be processed by the neural network, such as the form of tensors, and make the dimensions, types, and ranges of the data match the input requirements of the neural network model.

[0143] According to the characteristics of point defect detection and data features, select a suitable neural network architecture (such as Convolutional Neural Network CNN, Recurrent Neural Network RNN, or hybrid neural network, etc.), and set parameters such as the number of network layers, the number of neurons, activation functions, loss functions, etc.

[0144] Input the preprocessed detection model training set into the neural network and obtain the predicted values through layer-by-layer calculation.

[0145] Use a suitable loss function (such as cross-entropy loss, mean squared error, etc.) to calculate the difference between the predicted value and the true value.

[0146] Backpropagation and weight update: According to the gradient of the loss function, calculate the gradients of each layer of neurons through the backpropagation algorithm, and update the weights and biases of the network to reduce the prediction error.

[0147] Repeat the processes of forward propagation, loss calculation, and backpropagation, and iterate for multiple epochs until the loss function converges or reaches the preset number of training rounds. At the same time, monitor indicators such as the loss value and accuracy during the training process to evaluate the training progress and performance of the model.

[0148] After each training epoch, use the detection model validation set to verify the neural network model and evaluate the generalization ability of the model. The performance of the model can be evaluated by calculating indicators such as the loss value, accuracy, and F1 score on the validation set. According to the verification results, select the model with the best performance as the final point defect detection model. Conduct a comprehensive performance evaluation on the selected model, including its performance on different datasets, the detection ability for different types of point defects, and the robustness and stability of the model.

[0149] Package and encapsulate the trained point defect detection model to generate a deployable model file (such as in.h5,.pth, etc. formats). Integrate the model into the point defect detection system to achieve automated detection. This includes configuring the input and output interfaces of the model, setting the running parameters of the model, optimizing the inference speed of the model, etc. Conduct comprehensive testing on the integrated point defect detection system to make the functions and performance of the system meet the expectations. Actual production data or simulated data can be used for testing to verify the accuracy and stability of the system. After testing, deploy the point defect detection system to the production line to achieve real-time and efficient point defect detection. At the same time, establish a monitoring and alarm mechanism to promptly detect and handle abnormal situations in the system.

[0150] Specifically, for the TFT material display manufacturing process method described in the present invention, step S104 includes: Real-time collect operation data from the production line of TFT material display manufacturing. The real-time collected operation data includes substrate information, thin film transistor manufacturing parameters, color filter manufacturing parameters, combination process parameters, production environment data, and factors affecting product quality. Perform necessary preprocessing on the collected TFT material display manufacturing operation data, load the already trained and verified effective point defect detection model, and substitute the preprocessed TFT material display manufacturing operation data into the point defect detection model. The point defect detection model calculates based on the input data and outputs the result of point defect detection. Check the result of point defect detection output by the point defect detection model. If the result indicates the existence of point defects, then there are point defects in the product, and the process should enter step S105 for subsequent processing. If the result indicates the non-existence of point defects, then no point defects are detected in the individual product.

[0151] In the TFT material display manufacturing process, in order to achieve real-time point defect detection of products on the production line, step S104 covers the entire process from data collection to model prediction. The specific details are as follows: Real-time collect operation data from the production line of TFT material display manufacturing through sensors, automation devices, or data interfaces, etc. These data should cover substrate information (such as substrate type, size, batch number, etc.), thin film transistor manufacturing parameters (such as deposition speed, temperature, pressure, etc.), color filter manufacturing parameters (such as color ratio, coating thickness, etc.), combination process parameters (such as alignment accuracy, lamination pressure, etc.), production environment data (such as temperature, humidity, cleanliness, etc.), and any other factors that may affect product quality.

[0152] Enable the collected data to be synchronized to the central database or data processing system in real time for subsequent data processing and model prediction. At the same time, establish a data backup and recovery mechanism to prevent data loss or damage.

[0153] Clean the collected TFT material display production operation data to remove invalid data, abnormal data or redundant data, so as to ensure the accuracy and reliability of the data. For example, for missing values, they can be filled or deleted according to business logic; for outliers, they can be identified and processed through statistical methods or machine learning algorithms.

[0154] Extract useful features from the original data according to the requirements of the point defect detection model, and perform necessary feature transformations. This includes data normalization, standardization, one-hot encoding, etc., so that the data format and range match the model input requirements.

[0155] To improve the processing efficiency, the preprocessed data can be cached in memory for quick loading into the point defect detection model. At the same time, establish a data loading mechanism so that the model can obtain the latest data in real time.

[0156] Load the trained and validated point defect detection model from the model library. Make the model version match the process and data characteristics of the current production line. Substitute the preprocessed TFT material display production operation data into the point defect detection model for real-time prediction. The model will calculate based on the input data and output the results of point defect detection. Check the results output by the point defect detection model. If the result is that there are point defects, it indicates that there are point defects in the product, and it is necessary to immediately enter step S105 for subsequent processing (such as marking, isolation, rework, etc.). If the result is that there are no point defects, it indicates that no point defects are detected in a single product, and the subsequent production process can continue or enter the next process.

[0157] During the model prediction process, if data anomalies, model prediction failures or other abnormal situations occur, corresponding exception handling mechanisms should be established, such as retrying, alarming, manual intervention, etc., to ensure the normal operation of the production line and product quality control. Real-time feedback the results of point defect detection to the production line operators or quality management personnel so that they can timely understand the product quality situation and make corresponding decisions or adjustments. At the same time, the results can be stored in the database for subsequent data analysis and quality traceability.

[0158] The present invention first collects various historical data of the TFT material display production process, and these data cover the entire production process from the substrate to the final inspection. Then, use these data to train two neural network models: the TFT material display production process control model, which is used to guide the adjustment of the production process; and the other is the point defect detection model, which is used to monitor and predict point defects in the production process in real time.

[0159] Collect historical data in various aspects including substrates, thin-film transistor fabrication, color filter fabrication, point defect detection, defect repair, substrate combination, final inspection, production environment, and quality control. Use part of the historical data to train a TFT material display fabrication process control model, which can predict the optimal process adjustment plan based on the input production parameters. Use the historical data related to point defect detection to train a point defect detection model, which can accurately identify and predict point defects in production.

[0160] During the production process, collect operation data in real-time and input it into the point defect detection model for detection. If the model predicts the existence of point defects, immediately trigger the subsequent processing flow. Analyze the detected point defects and input the results into the fabrication process control model to obtain process adjustment parameters. Transmit the adjustment parameters to the production equipment in real-time for process adjustment.

[0161] Collect the feedback data after adjustment, input it into the point defect detection model again for verification, and optimize the control model using genetic algorithms if necessary. Clean and convert the format of the collected data to make the data quality meet the input requirements of the neural network model. Through iterative training, enable the model to accurately predict and identify point defects. The model analyzes the production data in real-time and automatically outputs the point defect detection results to achieve an automated detection process. Through real-time monitoring and prediction, the present invention can quickly respond to point defect problems in production, adjust process parameters in real-time, reduce the generation of defective products, and improve production efficiency. Timely discovery and handling of point defects significantly improve the finished product quality of the TFT display screen and reduce the outflow of defective products. The automated defect detection system of the present invention reduces the need for manual intervention, reduces human errors, and improves the automation level and stability of the production process.

[0162] In summary, the present invention realizes automated defect detection and processing in the TFT display screen fabrication process by using a neural network model for real-time monitoring and prediction, thereby significantly improving production efficiency and finished product quality.

Claims

1. A TFT material display manufacturing process, characterized in that: include: Step S101, obtaining TFT material display manufacturing process history data, the TFT material display manufacturing process history data includes substrate history data, thin film transistor manufacturing history data, color film manufacturing history data, point defect detection history data, defect repair history data, substrate assembly history data, final detection history data, production environment history data and quality control history data; Step S102, dividing substrate historical data, thin film transistor manufacturing historical data, color film manufacturing historical data, substrate combination historical data, production environment historical data, and quality control historical data into a control model training set and a control model verification set, and using the control model training set and the control model verification set to train a neural network model to obtain a TFT material display manufacturing process control model; Step S103, dividing the point defect detection history data, the defect repair history data and the final detection history data into a detection model training set and a detection model verification set, and using the detection model training set and the detection model verification set to train a neural network model to obtain a point defect detection model; Step S104, collecting TFT material display manufacturing operation data, substituting the TFT material display manufacturing operation data into a point defect detection model to obtain a point defect detection result, the point defect detection result includes the presence of a point defect and the absence of a point defect. If a point defect exists, proceeding to step S105; Step S105, analyzing the point defect detection result to obtain the analyzed point defect detection result, substituting the analyzed point defect detection result into the TFT material display manufacturing process control model to obtain the TFT material display manufacturing process control parameters, using the TFT material display manufacturing process control parameters as real-time TFT material display manufacturing process adjustment parameters, and transmitting the real-time TFT material display manufacturing process adjustment parameters to the TFT material display manufacturing equipment end.

2. The TFT material display manufacturing process according to claim 1, characterized in that: The step S105 further includes: Receive the running data of the TFT material display manufacturing equipment end executing the real-time TFT material display manufacturing process adjustment parameters, and obtain the feedback data of the TFT material display manufacturing equipment end; Substituting the feedback data from the TFT material display manufacturing equipment into the point defect detection model, if the point defect detection result output by the point defect detection model is that there is no point defect, then stopping the detection of the feedback data from the TFT material display manufacturing equipment; If the point defect detection result output by the point defect detection model is that a point defect exists, the TFT material display manufacturing process control model is optimized using a genetic algorithm.

3. The TFT material display manufacturing process according to claim 1, characterized in that: Also includes: Use genetic algorithm to optimize the TFT material display manufacturing process control model and determine the optimization target; Set the basic parameters of the genetic algorithm, including population size, genetic generations, crossover probability, and mutation probability; Initialize the population, randomly generate a set of TFT material display manufacturing process parameters as the initial population, each parameter combination represents an adjustment parameter, and the adjustment parameters include temperature, pressure, time and material ratio; The point defect detection model is used to evaluate each solution in the initial population, calculate its fitness value, and select some excellent individuals as parents according to the fitness value for crossover and mutation operations.

4. The TFT material display manufacturing process according to claim 3, characterized in that: Also includes: Crossover, perform a crossover operation on the selected parent generation to generate a new offspring solution; Mutation, perform mutation operations on offspring solutions; Use the point defect detection model to evaluate the newly generated offspring solution, calculate its fitness value, update the population, and replace part or all of the old population with the newly evaluated offspring solution to form a new generation of population; Check whether the termination condition of the genetic algorithm is met. If the termination condition is met, select the solution with the highest fitness value from the population as the optimal process parameter combination and output it to the TFT material display manufacturing equipment for application.

5. The TFT material display manufacturing process according to claim 1, characterized in that: The step S101 includes: The substrate history data includes substrate type, substrate size, substrate batch number, substrate production date, substrate surface condition, and substrate material properties; The historical data of thin film transistor manufacturing includes thin film deposition parameters, photolithography parameters, etching parameters, transistor size, transistor location information, and transistor electrical performance parameters; Color film production history data includes filter color, filter thickness, filter transmittance, filter position information, and filter uniformity; Point defect detection history data includes defect type, defect location, defect size, defect quantity and detection time; Defect repair history data includes repair methods, repair results, repair time, and post-repair performance verification; The historical data of substrate assembly include assembly method, assembly accuracy, overall size after assembly and assembly time; The final test history data includes display resolution, display brightness, display contrast, display color performance, display response time, display viewing angle, and overall appearance of the display; The historical data of the production environment includes the temperature and humidity of the production workshop, the status of the production equipment, the operator information and the production batch number: Quality control historical data includes quality control standards, quality control results and quality control records.

6. The TFT material display manufacturing process according to claim 1, characterized in that: The step S102 includes: For historical data of substrates, historical data of thin-film transistor production, historical data of color film production, historical data of substrate combinations, historical data of production environment, and historical data of quality control, a random number generator or a data partitioning tool is used to randomly divide the data into training sets and validation sets, and the divided training sets and validation sets are statistically analyzed to obtain control model training sets and control model verification sets.

7. The TFT material display manufacturing process according to claim 4, characterized in that: The step S102 includes: Preprocess the control model training set and control model validation set, and convert the data into a format that can be processed by the neural network; The preprocessed control model training set is input into the neural network for model training. The prediction value is calculated through forward propagation, and the network weight is updated through back propagation. The training is iterated for multiple cycles until the loss function converges or the preset number of training rounds is reached. After each training cycle, the neural network model is verified using the control model verification set, and the trained neural network model is saved.

8. The TFT material display manufacturing process according to claim 1, characterized in that: The step S103 includes: Receive the amount of data and the demand for model training, determine the division ratio of the training set and the verification set according to the amount of data and the demand for model training, select the random division method, randomly divide the data into the detection model training set and the detection model verification set, and store the divided detection model training set and detection model verification set in the database respectively.

9. The TFT material display manufacturing process according to claim 8, characterized in that: The step S103 includes: Preprocess the detection model training set and the detection model verification set, and convert the data into a format for neural network processing; input the preprocessed detection model training set into the neural network, start model training, calculate the prediction value through forward propagation, and calculate the gradient according to the loss function through back propagation, update the weight, and iterate the training for multiple cycles until the loss function converges or the preset number of training rounds is reached; After each training cycle, the model is verified using the detection model validation set, and the saved model is evaluated on the detection model validation set. If the model performance meets expectations, it can be deployed in the point defect detection system to achieve automated detection.

10. The TFT material display manufacturing process according to claim 1, characterized in that: The step S104 includes: Collect operation data in real time from the production line of TFT material display production, including substrate information, thin film transistor production parameters, color film production parameters, combined process parameters, production environment data and factors affecting product quality; Perform necessary preprocessing on the collected TFT material display manufacturing operation data, load the trained and verified effective point defect detection model, and substitute the preprocessed TFT material display manufacturing operation data into the point defect detection model; The point defect detection model calculates and outputs the point defect detection result according to the input data, and checks the point defect detection result output by the point defect detection model. If the result is that a point defect exists, then the point defect exists in the product, and the process should enter step S105 for subsequent processing; If the result is that there is no point defect, no point defect is detected for the individual product.

Images