Combination of Bayesian networks and FEM models to predict reliability of LED systems

Intelligent LED lighting systems can save up to 80% of energy compared to incandescent lighting systems. In order to provide these products at reasonable costs, integration and miniaturization are important steps [1]. Another attractive feature of LED systems is the claimed long life expectancy. The design lifetime of LED luminaries is typically 25 to 50 thousand hours [2, 3]. The long design lifetime, in combination with short-cycled technological innovations in LED packaging, pose challenges to traditional test-in reliability engineering approaches. Innovations in LED packaging are required to meet future cost target requirements at proven and designed-in lifetimes. Integral design of complex systems — such as LED systems merging optical, thermal, electrical, and mechanical disciplines — becomes more difficult [4, 9] as reliability engineering builds on the higher specialization per discipline. In this work we explore novel approaches for reliability engineering.