This paper presents the nonlinear dynamic mathematical models of the liquiel rocket engine with turbopump system, which can be used for the development of model-based fault defection. The Extended Kalmcm Filter (KEF) of the engine system has been designed and realized on computer. According to the statistical characteristics of the residuals, innovation-hased detection algorithm has been studied to defect the faults of the engine. The paper also discussed a few methods of reducing the expenses in KEF Calculation and the effects of belief factur and the degrees of freedom on the detection ability of the algorithm. The research provides the important basis for funther study of the model-based real-time detectim algorithm.