Application of Particle Filter Technique for Lifetime Determination of a LED Driver

Assessing product reliability through its degradation study is gaining wider acceptance as it provides insight to its degradation physics and shorten the required test time. In this paper, a degradation study for a linear-mode LED driver is conducted, and it is found that the measurement errors incorporated in the degradation study can reduce the accuracy of its lifetime estimation and that the errors cannot be reduced by using conventional regression method such as nonlinear least squares (NLS) and nonlinear mixed-effect estimation (NLME). To improve the estimation accuracy, a particle filter (PF) is implemented and combined with NLS for a single test unit, and a PF is combined with NLME for grouped test units. With this combination, the minimum test time is only one-fifth of the conventional method and the coefficient of variation for t_50% is reduced by 71% as compared with using NLME alone. With the proposed methods, we can also determine the remaining useful life of a product in situ as illustrated in this paper.