The research is to obtain the vibration signals and BDC signal of cylinder head surface without disassembling the engine, then section on vibration signal in the time domain energy reasonablely according to crankshaft rotation cycle, and extract the signal in the time domain energy as the engine state characteristic value. We establish the training set of engine under different engine fault states, and use support vector machine method for engine fault diagnosis and recognition. The results show that the method has an important significance on diagnosis of engine fault type and fault location.