In　this　paper,　the　gearbox　as　the　research　object,　under　the　conditions　of　strong　interference　signal　acquisition　and　analysis,　denoising,　eigenvalue　extraction,　pattern　recognition　and　fault　prediction　has　very　important　practical　significance.　Aiming　at　the　problem　of　end　effect　in　Empirical　Mode　Decomposition　(EMD),　an　improved　method　for　the　continuation　of　the　even-extended　cosine　window　function　(EECW)　of　signal　sequences　is　proposed.　Firstly,　the　signal　sequence　is　continually　extended　to　realize　the　smooth　transition　between　the　extension　data　and　the　original　signal,　avoiding　the　jump　of　the　instantaneous　frequency　of　the　signal.　Secondly,　there　is　the　problem　of　continuation　error　for　this　extension　method.　(Or　slow　speed)　to　the　internal　development　of　data　to　ensure　the　correct　decomposition　of　the　signal　valid　data　to　improve　the　accuracy　of　the　decomposition　of　the　signal　to　achieve　the　improvement　of　EMD　algorithm.　Through　the　simulation　analysis　and　fault　diagnosis,　it　is　shown　that　the　method　can　effectively　suppress　EMD　endpoint　effect　and　realize　the　effective　diagnosis　of　rotating　machinery　fault.　The　experimental　results　show　that　the　proposed　method　has　high　precision　and　good　performance