Mohammed A. Alam, Michael H. Azarian, Michael Osterman and Michael Pecht
Center for Advanced Life Cycle Engineering (CALCE)
1103 Engineering Lab Building University of Maryland,
College Park, MD 20742
This paper presents the application of model-based and data-driven approaches for prognostics and health management (PHM) of embedded planar capacitors under elevated temperature and voltage conditions. An embedded planar capacitor is a thin laminate that serves both as a power/ground plane and as a parallel plate capacitor in a multilayered printed wiring board (PWB). These capacitors are typically used for decoupling applications and are found to reduce the required number of surface mount capacitors.
The capacitor laminate used in this study consisted of an epoxy-barium titanate (BaTiO3) composite dielectric sandwiched between Cu layers. Three electrical parameters, capacitance, dissipation factor, and insulation resistance, were monitored in-situ once every hour during testing under elevated temperature and voltage aging conditions. The failure modes observed were a sharp drop in insulation resistance and a gradual decrease in capacitance. An approach to model the time-to-failure associated with these failure modes as a function of the stress level is presented in this paper. Model-based PHM can be used to predict the time-to-failure associated with a single failure mode, consisting of a drop in either insulation resistance or capacitance. However, failure of an embedded capacitor could occur due to either of these two failure modes and was not captured using a single model. A combined model for both these failure modes can be developed but there was a large variance in the time-to-failure data of failures as a result of a sharp drop in insulation resistance. Therefore a data-driven approach, which utilizes the trend and correlation between the parameters to predict remaining life, was investigated to perform PHM.
The data-driven approach used in this paper is the Mahalanobis distance (MD) method that reduces a multivariate data set to a single parameter by considering correlations among the parameters. The Mahalanobis distance method was successful in predicting the failures as a result of a gradual decrease in capacitance. However, prediction of failures as a result of a drop in insulation resistance was generally challenging due to their sudden onset. An experimental approach to address such sudden failures is discussed to facilitate identifying any trends in the parameters prior to failure.
Complete article is available to CALCE Consortium Members.