Deep learning-based optimization using a limited number of experiments for roll-to-roll UV embossing

Optoelectronic components are becoming more complex and functional, but they must be miniaturized to achieve portability. The roll-to-roll UV embossing process is used to produce a light guide film to replace the light guide plate in a backlight module. To improve replicability, this study uses a deep learning-based optimization (DLBO) system. Deep learning requires a large amount of experimental data, but it is impractical for manufacturing. The DLBO system comprises a data generator, a model generator, and an optimal factor generator. The data generator generates massive labeled data by using a limited number of Taguchi experiments to predict the quality characteristic for each combination. The model generator produces an optimal neural network model that conforms to the input data. The optimal factor generator uses the model that is produced by the model generator as the objective function to determine the optimal factors. Verification experiments show that a DLBO system can successfully predict the height of a microstructure. This system also determines the optimal process conditions. The measured height of the microstructure is 1.27 mu m higher than the result with Taguchi optimization, and the replicability is 90%. For the optimal conditions using the DLBO system, the roll-to-roll UV embossed film has the best light guiding properties. The results show that the optimal conditions that are determined using the DLBO system produce light guide films with better optical qualities than the Taguchi method.

Automated summary

Optoelectronic components are becoming more complex and functional, and deep learning-based optimization (DLBO) system is used to improve replicability. The DLBO system, including data generator, model generator, and optimal factor generator, successfully predicts and determines the optimal process conditions to produce roll-to-roll UV embossed film with the best light guiding properties. The results show that the DLBO system produces light guide films with better optical qualities than the traditional Taguchi method.