Rolling　bearings　are　important　components　in　mechanical　machinery,　and　their　failure　will　directly　affect　the　normal　operation　of　the　machine.　Therefore,　the　analysis　of　mechanical　machinery　vibration　signals　is　crucial　for　ensuring　the　normal　operation　of　the　machinery.　Successive　variational　mode　decomposition　(SVMD)　is　an　important　technique　for　decomposing　a　bearing　stationary　signal　into　its　characteristic　modes　with　a　priori　penalty　factor.　Therefore,　it　cannot　handle　nonstationary　bearing　signals.　To　tackle　the　above　problems,　a　novel　method,　named　successive　variational　nonstationary　mode　decomposition　(SVNMD),　is　developed　in　this　article.　First,　a　new　decomposition　framework　is　proposed　by　adopting　the　constructed　resampling　operator　to　modify　the　optimization　function　of　the　SVMD,　which　eliminates　the　influence　of　frequency　mixing.　Second,　in　order　to　automatically　determine　the　optimal　penalty　factor,　an　iterative　selection　scheme　is　developed,　which　is　free　from　prior　knowledge.　Third,　an　instantaneous　frequency　estimation　theory　is　proposed　to　obtain　the　common　trend　function　of　the　signal.　Finally,　a　time-frequency　representation　with　high-energy　concentration　is　obtained　to　accurately　identify　the　fault　characteristics　of　rolling　bearings.　Both　the　simulation　and　experimental　verification　have　confirmed　the　productiveness　of　the　SVNMD　in　diagnosing　multiple　faults　of　bearings　under　time-varying　speed　conditions.