The　dimensionality　disaster　problem　exist　in　pattern　recognition　process,　the　fault　diagnosis　method　on　nonlinear　feature　kernel　extraction　is　presented　here.　The　fisher　linear　discriminant　analysis　method　is　extended　to　nonlinear　fields　by　kernel　technology,　the　original　feature　space　is　mapped　into　observation　space　for　features　linearization,　The　fault　pattern　is　classified　by　fisher　linear　discriminant　analysis　in　minimum　criteria,　The　problem　on　nonlinear　feature　separability　and　small　sample　size　are　solved.　an　fault　diagnosis　illustration　verifies　this　method.　©　2011　IEEE.