Abstract
With their advantages of high efficiency, long lifetime, compact size and being free of mercury, ultraviolet light-emitting diodes (UV LEDs) are widely applied in disinfection and purification, photolithography, curing and biomedical devices. However, it is challenging to assess the reliability of UV LEDs based on the traditional life test or even the accelerated life test. In this paper, radiation power degradation modeling is proposed to estimate the lifetime of UV LEDs under both constant stress and step stress degradation tests. Stochastic data-driven predictions with both Gamma process and Wiener process methods are implemented, and the degradation mechanisms occurring under different aging conditions are also analyzed. The results show that, compared to least squares regression in the IESNA TM-21 industry standard recommended by the Illuminating Engineering Society of North America (IESNA), the proposed stochastic data-driven methods can predict the lifetime with high accuracy and narrow confidence intervals, which confirms that they provide more reliable information than the IESNA TM-21 standard with greater robustness.