The　performance　degradation　assessment　(PDA)　of　rolling　element　bearings　is　a　necessary　link　to　ensure　the　reliability　of　high-end　equipment.　However,　traditional　health　indicators　(HIs)　are　not　sensitive　to　early　defects,　and　there　are　often　large　local　fluctuations　in　the　later　stage　of　degradation.　Hence,　this　paper　propose　a　novel　PDA　method　to　obtain　HIs　with　early　warning　capability　and　monotone　trend.　Firstly,　an　improved　graph　spectrum　reconstruction　method　is　proposed　to　enhance　the　characteristics　of　signals.　The　random　phase　space　reconstruction　strategy　is　introduced　to　solve　the　problem　of　large-scale　graph　Laplacian　matrix　decomposition.　Then,　the　spectrums　of　the　enhanced　signals　are　characterized,　namely　the　high-dimensional　degradation　features　in　frequency　domain　are　extracted　and　smoothed　by　Kalman　filter.　Finally,　the　Laplacian　Eigenmaps　is　used　to　extract　the　intrinsic　degeneration　manifolds　from　the　high-dimensional　degradation　features　as　the　established　HIs.　Life　cycle　degradation　data　and　quantitative　failure　data　are　analyzed　to　verify　the　effectiveness　of　the　proposed　method.　Compared　with　other　state-of-art　methods,　the　results　show　that　the　HIs　established　by　the　proposed　method　reflect　the　degradation　earlier　and　have　obvious　degradation　trend.　It　effectively　realizes　the　mapping　between　degradation　and　HI.　©　2021