Time　synchronous　averaging　(TSA)　can　attenuate　incoherent　and　non-synchronous　components,　thereby　effectively　reducing　noise　interference　and　extracting　periodic　signal　features.　However,　strong　noise　in　the　wind　turbine　gearbox　and　pulse　interference　generated　by　the　equipment　will　seriously　affect　the　attenuation　process　and　the　periodicity　of　the　signal.　To　solve　this　problem,　this　paper　proposes　a　weighted　subspace　local　averaging　(WSLA)　method,　which　can　segment　the　signal　according　to　the　gear　rotation　period　and　decompose　and　reconstruct　it　into　a　new　matrix.　The　pure　signal　subspace　matrix　is　established　by　decomposing　the　matrix　to　achieve　noise　reduction.　In　addition,　a　weight　distribution　method　is　proposed　to　reduce　the　weight　of　incoherent　pulses　in　the　synchronous　averaging　process,　so　as　to　extract　more　accurate　periodic　signal　waveforms.　The　signal　waveform　is　plotted　as　a　gear　circle　diagram,　which　can　intuitively　display　the　health　status　of　the　gear.　By　using　TSA　and　WSLA　methods　to　process　analog　signals　separately,　the　results　show　that　the　SNR　after　WSLA　processing　is　10.5dB　higher　than　that　after　TSA　processing.　In　order　to　be　more　convincing,　the　size　of　the　noise　and　the　amplitude　of　unexpected　impacts　were　increased,　and　the　results　showed　that　the　SNR　after　WSLA　processing　was　11.5dB　higher　than　that　after　TSA　processing.　Verified　the　effectiveness　of　the　WSLA　method.　And　applied　to　two　fault　diagnosis　cases:　the　vibration　signal　of　the　gearbox　of　a　3MW　wind　turbine;　the　vibration　signal　measured　by　the　gearbox　test　bench.　The　results　show　that　the　proposed　method　can　effectively　extract　periodic　features　from　signals　interfered　by　strong　background　noise　and　incoherent　pulses.　The　advantage　of　WSLA　is　that　it　first　performs　noise　reduction,　and　then　by　assigning　data　weights,　it　can　better　reduce　the　influence　of　pulses　unrelated　to　the　gear　rotation　period,　which　is　beneficial　for　identifying　correct　fault　information.　The　gear　circular　diagram　can　more　conveniently　locate　the　fault　location.　©　2001-2012　IEEE.