Abstract: In order to completely enhance the safety of liquid rocket engine turbopump,aiming at its start-up process,a method for vibration fault detection was presented.Firstly,the vibration data sampled by equal time interval were resampled and the data of equiangular interval were obtained to drive out the speed fluctuation influence on vibration features.Secondly,the vibration feature was selected and calculated.Finally,the nonlinear relational model between vibration feature and rotational speed was founded by support vector regression.Once the deviation between the new test data and the model is over a limit,it shows evidence of turbopump fault.The detection time of 2.2s test data is 0.0470s.It shows that the model can be used for the turbopump fault detection in start-up process.