Steam　turbine　is　an　important　energy　conversion　equipment　in　the　power　industry.　The　steam　turbine　fault　detection　method　combined　with　information　technology　is　an　important　means　to　realize　intelligent　operation　and　maintenance.　Aiming　at　the　shortcomings　of　traditional　fault　detection　methods,　such　as　low　accuracy,　poor　adaptability,　and　heavy　reliance　on　manual　experience,　this　paper　presented　a　method　of　steam　turbine　rotor　fault　detection　based　on　deep　convolutional　neural　network　to　achieve　end-to-end　detection,　which　was　evaluated　by　using　fault　rotor　numerical　model,　including　three　simple　fault　classification　of　rotor　unbalance,　parallel　misalignment　and　angular　misalignment,　as　well　as　the　detection　of　the　location　and　degree　of　the　three　faults.　And　the　method　implements　multi-task　collaborative　detection.　The　effect　of　snr　and　channel　numbers　on　the　detection　performance　of　neural　networks　was　discussed.　Fault　detection　accuracy　is　100%,　the　average　detection　accuracy　of　position　and　degree　is　not　less　than　96.47%.　The　method　proposed　in　this　paper　can　realize　the　direct　mapping　of　vibration　signals　from　multiple　measurement　points　to　fault　features.　It　is　free　of　the　traditional　methods　dependence　on　artificial　experience　and　signal　processing　skills,　and　has　the　characteristics　of　high　accuracy　and　robustness.　©　2021　Chin.　Soc.　for　Elec.　Eng.