Planetary　gear　trains　equipped　in　wind　turbine　often　run　under　slow　speed　and　non-stationary　load　condition.　The　incipient　gear　faults　in　a　wind　turbine　gearbox　can　hardly　be　detected　yet　might　cause　tremendous　loss.　In　order　to　detect　the　incipient　faults,　a　resultant　vibration　signal　model　is　proposed　to　characterize　the　faulty　features　of　a　single　stage　planetary　gear　train　working　under　non-stationary　load　conditions.　For　this　purpose,　an　analytical　dynamic　model　is　developed.　By　introducing　the　crack-induced　mesh　stiffness　and　varying　load　into　the　dynamic　model,　the　vibration　responses　of　the　system　are　predicted.　Based　on　this,　a　resultant　vibration　signal　model　is　developed　in　the　form　of　weighted　summation　of　mesh　vibration　signals.　With　the　resultant　model,　the　vibration　signals　of　an　example　system　are　simulated　and　analyzed.　The　simulation　results　indicate　that　varying　load　and　tooth　crack　make　the　system＇s　vibration　signals　become　extremely　complicated　in　both　time　and　frequency　domains.　The　incipient　tooth　crack　induced　impulse　vibration　signals　are　too　weak　to　be　identified　in　the　time　domain　but　can　be　detected　from　the　order　spectrum.　The　simulation　results　from　the　resultant　signal　model　are　verified　by　the　test　rig　experimental　measurements.　©　2018　Elsevier　Ltd