Construction of a Diffuser-Plate and Thickness-Distribution Dataset and Performance Evaluation of an Image-to-Image Translation Surrogate Model for Organic Film Deposition in OLED Lighting

In this paper, we construct a dataset consisting of paired diffuser-plate images and thickness-distribution images, and evaluate the computational cost and prediction accuracy of an image-to-image translation surrogate model that directly pre-dicts thickness-distribution images from diffuser-plate drawings. In conventional fluid simulation, computing thickness-distribution images from diffuser-plate drawings entails a substantial computational cost. The proposed image-to-image translation surrogate model significantly reduces the computational cost of predicting thickness-distribution images by employing a deep learning-based image-to-image translation method, compared with fluid simulation. Experimental results showed that predicting a single thickness-distribution image required an average of 3.3 hours with fluid simulation, whereas the proposed image-to-image translation surrogate model required an average of 122.0 milliseconds. In the image-to-image translation surrogate model, experimental results confirmed that Pix2Pix, which is one of the image-to-image translation techniques, achieved smaller errors relative to fluid simulation than the other evaluated methods.