Interpretable Fault Prognostics for Switch Mode Power Supplies

Switch mode power supplies present light weight power conversion solutions, but degradation affecting sub-components have been known to transition the converter to move to unstable dynamic operations. This paper presents a method to build interpretive prognostics for switch mode power supplies with electromagnetic input filters by modeling sub-component degradation trajectories and using discrete event simulation to generate lifecycle data associated with the system impedances to use as inputs into machine learning based prognostics to make interpretable remaining useful life predictions. As a usage case, a buck-boost switch mode power supply with parasitic elements for all components and time-dependent degradation concerning the input and output filtering capacitor is analyzed.