Chou, Hai-TangHai-TangChouCheng, Ting-YuTing-YuChengYAO-YANG TSAIKe, Kun-ChengKun-ChengKeSEN-YEU YANG2025-07-312025-07-312025-01-0100323888https://www.scopus.com/record/display.uri?eid=2-s2.0-105008208431&origin=resultslisthttps://scholars.lib.ntu.edu.tw/handle/123456789/730868The rapid advancement of virtual reality technology has heightened the demand for high-quality microlens arrays, necessitating precise optical quality control. This study investigates injection compression molding to produce microlens arrays, using in-mold pressure and temperature sensors to collect critical process data. Five quality indicators, based on signal features highly correlated with optical quality (correlation up to 0.8), were developed to train a multilayer perceptron (MLP) neural network. The MLP classifies optical quality by predicting grayscale values, derived from optical fringes via birefringence measurements, as quality metrics. Moldex3D simulations, validated by the Taguchi method and variance analysis, optimized melt temperature and compression gap in full-factorial experiments, generating 250 datasets. After outlier removal and hyperparameter tuning, the MLP achieved a test accuracy of 95.5% (up from 86.4%), outperforming traditional methods. This work demonstrates the effectiveness of artificial intelligence in classifying microlens array optical quality, enabling sustainable manufacturing for virtual reality applications.falseinjection compression moldingneural network modelopticssimulation[SDGs]SDG9[SDGs]SDG12journal article10.1002/pen.700112-s2.0-105008208431