Advancements in precision optics polishing through AI-driven predictive models

The increasing demand for high quality optics and wafer surfaces requires precise and reliable polishing processes. Surface shape and roughness are critical to performance in various industries. Although traditional polishing relies heavily on expert intuition, there is a growing need for data-driven approaches to improve accuracy. To this end, artificial intelligence (AI) models, including transformers, long short-term memory (LSTM) networks, convolutional neural networks (CNNs), U-Nets, and temporal convolutional networks (TCNs), were investigated. In this study, sensor data from 14 polishing experiments is used to predict material removal, with AI models optimized using the Keras Tuner and evaluated by five-fold time series cross-validation. The results show that the Transformer model provides the best balance between prediction accuracy and inference efficiency, outperforming other models in accurately predicting material removal.