Aiming　at　the　problem　of　features　extraction　of　planetary　gearbox　being　difficult,　a　fault　diagnosis　method　based　on　parameter　optimized　variational　mode　decomposition　(VMD)　and　multi-domain　manifold　learning　was　proposed.　Firstly,　the　salp　swarm　optimization　variational　mode　decomposition　(SSO-VMD)　was　utilized　to　decompose　and　reconstruct　signals　to　reduce　noise　interference.　Then,　fault　features　were　extracted　from　multi-domain,　and　the　improved　supervised　self-organizing　incremental　learning　neural　network　boundary　punctuation　isometric　mapping　(ISSL-Isomap)　algorithm　was　used　to　reduce　dimension,　and　acquire　low-dimensional　fault　features.　Finally,　the　artificial　bee　colony　support　vector　machine　(ABC-SVM)　multi-fault　classifier　was　used　to　do　diagnosis　and　identification.　The　SSO-VMD　was　compared　with　the　empirical　mode　decomposition　(EMD),　and　the　superiority　of　SSO-VMD　was　verified　with　simulation　signal　analysis　results.　The　proposed　fault　diagnosis　method　was　applied　in　planetary　gearbox　fault　diagnosis　test　analysis.　Results　showed　that　the　multi-domain　feature　extraction　is　better　than　the　feature　extraction　in　single-domain　including　time　domain,　frequency　one　and　scale　one;　the　dimension　reduction　effect　of　ISSL-Isomap　is　better　than　those　of　Isomap,　t-distributed　stochastic　neighborhood　embedding,　linear　discriminant　analysis,　weighted　Isomap　and　supervised　Isomap;　the　fault　recognition　rate　of　the　proposed　method　reaches　100%,　and　it　can　effectively　recognize　various　types　working　conditions　of　planetary　gearbox.　©　2021,　Editorial　Office　of　Journal　of　Vibration　and　Shock.　All　right　reserved.