A multiple regression model framework for designing a UVC LED reactor for point-of-use water treatment

Abstract

UVC LED technology has been introduced for drinking water treatment because of advantages including its flexible wavelength range, smaller form factor, low susceptibility to power cycling and long lifespan. However, designing a reactor utilizing UVC LEDs is still a challenging task. In this study, we proposed a systematic framework for designing a UVC LED reactor for microbial inactivation for point-of-use (POU) water treatment. By combining computational fluid dynamics, optical ray tracing and Lagrangian particle tracing models, we established a baseline multiphysics UV fluence model, and then converted UV fluence to logarithmic reduction values (LRVs) via microbial inactivation kinetic equations. The simulation was validated (one-way ANOVA, p < 0.05) against biodosimetry using bacteriophage Qβ. A two-level full factorial design-of-experiments (DOE) was conducted on eight operating and design variables for the reactor with a total of 256 combinations based on the multiphysics model. Four main effects including the flow rate, water UV transmittance, LED radiant power and viewing angle pattern were identified to have significant impacts on reactor microbial inactivation efficiency. Multiple regression predictive models were derived after DOE analysis to include two-way and three-way interactions among six design variables with predictive accuracies being 82.2% and 90.6%, respectively. Finally, the multiphysics model and multiple regression models were validated again via biodosimetry on a newly constructed reactor and the results showed great agreement (p < 0.05).