Research on accurate prediction method of display quality features based on augmented hybrid modeling

Quality prediction in the Thin-Film Transistor Liquid Crystal Display (TFT-LCD) production process is crucial for reducing production costs and optimizing the manufacturing process. For the process quality, a deep learning-based in process product prediction method is constructed. First, feature selection based on machine learning methods is performed to reduce the feature dimension of sequence data and filter sensitive features. Then, a hybrid model based on multiscale convolutional neural network (CNN) and long and short-term memory network (LSTM), is constructed for extracting the deep features of the feature set more comprehensively, efficiently, and stably, while capturing the local details and longrange dependencies of the data. Finally, an enhanced multiscale temporal attention mechanism is introduced and enables the hybrid model to efficiently control the flow of feature information during the forward propagation process and further feature screening to realize accurate product quality prediction. The proposed deep learning hybrid model integrates CNN and LSTM, and enhanced multiscale temporal attention mechanism, which improves the ability of capturing key features and modeling temporal dependencies. Validated by actual production data, the hybrid model proposed in this paper achieves a coefficient of determination of 0.814 and a mean square error of 0.00787, realizing the accurate prediction of product quality feature values.