Regression modeling of optical properties for optical design in high-volume white LED manufacturing

In high-volume manufacturing of white light-emitting diodes (LEDs), maintaining consistent luminous flux and chromaticity is essential for stable optical performance and process consistency. Variations in the radiant flux of the blue LED chip and the concentrations of yellow (YAG:Ce) and red (SCASN) phosphors can lead to significant performance deviations, increasing production costs and reducing product uniformity. This study proposes a data-driven regression framework to predict three key optical outputs-total luminous flux (lm) and chromaticity coordinates (x, y)-from structural parameters, which may facilitate early detection of deviations and support timely process adjustments. We evaluate multiple-input single-output (MISO), multiple-input multiple-output (MIMO), and a hybrid MIMO-LR cascade configuration that reduces MIMO output dimensionality. Five regression models were examined, including multilayer perceptron (MLP), radial basis function network (RBFN), support vector regression (SVR), random forest (RF), and linear regression (LR). Despite the presence of a clear discontinuity in the input-output space caused by SCASN inclusion, all MIMO-based models achieved high prediction accuracy [Formula: see text] for total luminous flux and the chromaticity x coordinate, effectively capturing the piecewise optical behavior. The results indicate that such predictive models can support the analysis of complex relationships between structural parameters and optical outputs, and have the potential to contribute to design optimization and process improvement in large-scale LED manufacturing.