In　view　of　the　diffieulty　in　seleeting　die　deeomposition　layer　K　and　penalty　faetor　a　of　variational　mode　deeomposition　(VMD),　a　subtraction-average-based　optimizer　(SABO)　is　proposed　to　optimize　the　parameters.　Firstly,　the　SABO　is　used　to　optimize　K　and　a,　Output　the　optimal　parameter　eombination,　and　Substitute　it　into　VMD　to　deeompose　the　original　Vibration　signal　into　K　modal　eomponents.　Then,　the　maximum　envelope　kurtosis　is　used　as　the　index　to　extraet　the　eomponent　with　the　largest　kurtosis　among　the　K　modal　eomponents　as　the　optimal　eomponent,　and　the　eigenveetor　sample　set　is　construeted　by　ealeulating　the　relevant　time-domain　and　entropy　theory　eharaeteristie　parameters　of　the　optimal　eomponent.　Finally,　the　eigenveetor　sample　set　is　input　into　the　support　veetor　maehine　(SVM)　with　mesh　seareh　and　5-fold　eross-Validation　for　lault　diagnosis.　To　verify　the　effeetiveness　of　this　method,　experiments　were　condueted　using　the　bearing　dataset　from　Case　Western　Reserve　University.　The　experimental　results　show　that　the　elassifieation　effeet　of　the　method　is　better,　and　the　aeeuraey　rate　is　99.44%.　Based　on　the　bearing　data　set　experiments　of　three　different　working　eonditions　in　Jiangnan　University,　the　final　fault　diagnosis　aeeuraey　rate　reaehes　more　than　95%.　©　2024　Chinese　Society　for　Measurement.　All　rights　reserved.