The　fault　diagnosis　of　power　converter　plays　a　decisive　role　in　the　intelligent　and　stable　operation　of　DC　microgrid.　Aiming　at　the　nonlinearity　of　fault　output　of　converter　power　transistor　and　the　difficulty　of　feature　extraction,　a　combination　of　variational　mode　decomposition　and　kernel　density　estimation　was　proposed.　Firstly,　the　power　converter　output　signal　was　collected.　Secondly,　the　signal　was　subjected　to　variational　mode　decomposition　to　decompose　the　complex　signal　into　a　series　of　sub-signals,　and　each　modal　component　was　extracted　as　a　feature　vector.　Finally,　the　fault　diagnosis　was　realized　by　means　of　the　kernel　density　estimation　classifier.　The　experimental　results　showed　that　the　method　reduced　the　diagnostic　cost　and　improved　the　diagnostic　accuracy,　and　the　method　was　feasible　and　effective.　©　2019　IEEE.