As　the　carrier　would　undergo　angular　and　linear　swaying　in　various　dynamic　disturbance　cases,　rotary　strapdown　inertial　navigation　system　(Rotary　SINS)　is　difficult　to　realize　self-alignment.　In　order　to　solve　this　problem,　a　new　self-alignment　method　based　on　inertial　reference　frame　was　proposed.　Firstly,　based　on　the　coarse　alignment　results　in　the　inertial　frame,　the　initial　fine　alignment　error　models　of　a　dualaxis　rotation　modulation　of　rotary　SINS　in　the　inertial　frame　is　derived.　Then,　aiming　at　the　noise　uncertainty　of　the　observation　model,　by　changing　the　embedded　manner　of　fading　factors,　this　paper　proposes　an　improved　adaptive　Kalman　filter　based　on　multiple　fading　factors.　Finally,　simulation　results　show　that　the　proposed　method　can　effectively　solve　the　self-alignment　problem　of　rotary　SINS　in　dynamic　disturbance　cases,　realizing　rapid　and　accurate　self-alignment.　Copyright　©　2014　Acta　Automatica　Sinica.　All　rights　reserved.