To　solve　the　small-sample　pattern　recognition　problem　of　mechanical　equipment　fault　and　improve　classification　ability,　a　new　hybrid　diagnosis　model　of　support　vector　machine　(SVM)　based　on　fuzzy　feature　extraction　with　empirical　mode　decomposition　(EMD)　is　proposed,　where　these　intrinsic　mode　components　are　extracted　with　EMD　from　original　signals　and　converted　into　fuzzy　feature　vectors,　and　then　the　mechanical　fault　can　be　diagnosed.　The　extracted　fuzzy　feature　vectors　are　input　into　the　multi-classification　SVM　to　detect　the　different　abnormal　cases.　This　model　is　applied　to　the　classification　of　turbo-generator　set　under　3　operating　conditions.　Testing　results　show　that　the　classification　accuracy　of　the　proposed　model　(100%　classification　success　rate)　is　greatly　improved　compared　with　the　SVM　without　feature　extraction　(53.33%　classification　success　rate)　and　with　the　SVM　extracting　the　fuzzy　feature　from　wavelet　packets　(86.67%　classification　success　rate),　and　the　faults　of　turbo-generator　set　can　be　correctly　and　rapidly　diagnosed.