Forecasting of abnormalities in optical glass lens centering using acoustic emissions

Abstract This study aims to develop a predictive monitoring system to reduce defects and improve production efficiency in the optical lens-centering process. Centering is a critical manufacturing process of optical glass lens to align the geometric and optical axes of a lens through edge grinding. A lens’s quality after centering depends on the process parameters, size differences of the lens blanks, and condition of the grinding wheel and machine. Edge cracks and circularity errors may unexpectedly generate during processing. This study developed an acoustic emission (AE) signal trend prediction monitoring system for centering process to reflect on process performance and avoid product loss. Down-sampling and exponential weighted moving average (EWMA) were applied to signal preprocess and analysis. The prediction algorithms based on long short-term memory (LSTM) were trained and compared with actual process data to evaluate their predictive performance. The CuDNNLSTM model had the highest performance with R 2 = 0.9833 and prediction time = 36 s. Both showed excellent predictive capabilities which aid in preventing errors from happening. It is the most suitable for centering process monitoring with AE sensor among the compared algorithms. According to the final verification in real manufacturing, the prediction model built with CuDNNLSTM could improve yield rate from 64 to 92% and total production time reduced by 32.9%. Moreover, the prediction model built by CuDNNLSTM with EWMA could perfectly predict circularity errors and reduce total production time by 43.0%.