Algorithm Research And Software Development Of LCD Glass Substrate Thickness Prediction Based On Data Driven

With the gradual improvement of China's manufacturing Internet of Things and informatization layout,and the steady implementation of the national strategy of intelligent manufacturing,product quality prediction and traceability analysis based on data-driven have become a core part of intelligent manufacturing in the field of industrial product quality management.The liquid crystal display panel is the mainstream device of the new display technology in the digital and network era,and is widely used in smart phones,LCD TVs,tablet computers and various types of smart display appliances.As the key upstream material for liquid crystal display panels,the thickness of the glass substrate is a key indicator in the quality evaluation of the glass substrate,which has a great influence on the performance of the liquid crystal display panel.The existing glass substrate thickness measurement method has high cost and poor real-time performance.It is difficult to achieve efficient and comprehensive thickness measurement,and it is impossible to trace the abnormal processing parameters of the glass substrate,which may easily cause the defective glass substrate to flow into the liquid crystal display panel subsequent processing steps,increasing manufacturing costs.Regarding the issue above,the manufacturing process data of the glass substrate is used in this paper to in-depth study of key algorithms in glass substrate thickness prediction,and the glass substrate thickness prediction software system is designed and developed.The main work is as follows:(1)Analyzed the current status and pain points of the thickness measurement technology of the liquid crystal glass substrate,and an overall framework with data driving as the core technology is proposed.At the same time,the architectural pattern and functional modules of the software system are determined.(2)The characteristics of data in the production process of liquid crystal glass substrates are studied,a differential feature selection method based on Pearson correlation coefficient is proposed,and the multi-collinearity of the data features is studied.(3)For small sample and high-dimensional data features,ridge regression,support vector regression,XGBoost regression and Light GBM regression algorithms are used to build a single prediction model,and the grid search method,cross-validation method and Bayesian optimization algorithm are used for hyperparameter optimization.Finally,multiple metrics are used to analyze the prediction results of a single prediction model.(4)Three model fusion algorithms of equal weight fusion,weighted fusion based on squared error minimization and Stacking ensemble learning fusion are compared,algorithm testing and analysis are conducted.Among them,the model fusion algorithm based on Stacking ensemble learning has obvious advantages in prediction performance.(5)The glass substrate thickness prediction software system developed using Pyqt5 and Python programming language,which can realize many functions such as data import,feature engineering,model training,model prediction,traceability analysis,model self-update and data management;After the function and performance test of the software system,the software system runs stably and reliably.