Abstract: A neural network twice classifier is provided which utilizes fuzzy sets as fault patterns of a liquid propellant rocket engine.Each fuzzy set is an aggregate of fuzzy hyperbodies.A fuzzy hyper-body is an n-dimensional hyper-sphere defined by a radius and a center with a corresponding membership function at first learning of fuzzy neural network and a part of the hyper-sphere defined by an included angle,a center and a direction vector at second learning.The training isolation accuracy of the twice learning algorithm is higher than that of the once learning algorithm,and the twice learning of fuzzy neural network also enhances sensitivity to rocket fault.The twice learning algorithm can learn nonlinear fault pattern boundaries in two passes through the input data and provides the ability to incorporate new fault messages in succession and refine existing fault classes without retraining.The emulation of its application indicates that the fuzzy hyper-body neural network can be successfully employed in the fault detection and isolation of the turbo-pump feed liquid rocket engine.