Due　to　the　complex　working　conditions,　it　is　difficult　to　accurately　extract　the　periodic　pulse　component　of　the　rolling　bearing　vibration　signal.　The　Ramanujan　subspace　(RS)　theory　has　a　good　ability　to　extract　periodic　components　(PCs).　However,　the　theory　divides　signals　by　fixed　period　length,　which　means　that　the　effect　of　extracting　quasi-periodic　signals　is　poor.　Therefore,　this　article　proposes　an　optimized　periodic　mode　decomposition　(OPMD)　method　to　improve　the　defect　and　apply　it　to　bearing　fault　diagnosis.　First,　the　window　length　of　the　intercepted　data　segment　is　determined　by　periodicity　measurement.　Then,　the　pulse　phase　corresponding　to　each　data　segment　is　detected　by　correlation　to　divide　the　data　segment,　and　cubic　spline　interpolation　is　used　for　phase　compensation.　Finally,　the　periodic　pulse　is　extracted　from　the　reconstructed　signal.　Simulation　and　experimental　results　show　that　compared　with　the　existing　RS　methods,　OPMD　can　extract　better　periodic　pulses,　reduce　the　impact　of　quasi-periodic　signals　on　the　extraction　effect,　and　improve　the　adaptability　of　the　algorithm.