Nonlinear dynamic neural network model for rocket propulsion systems

It is not only very essential for control system design but also for failure detection and diagnosis to set up a realtime, precise and reliable dynamic model of liquid propellant rocket propulsion system. The feed-forward neural network if successfully trained, can map the inputs to the desired outputs, so recent years have seen an extensive amount of research to explore its approximation properties. On the basis of studying RBF(Radial Basis Function) neural networks' theory and system mechanism, a nonlinear dynamic neural networks' model for liquid propellant rocket's propulsion system with multi-inputs and multi-outputs was built. During the modeling, necessary dynamic information was included and parameters of model were also well-chosen. The contrastive results of outputs of the model and measuring data of one real test-firing demonstrates that the model is of many advantages, such as short computational time, better real-time property and good precision. The model is very well fit for the applications of real time condition monitoring, fault diagnosis and control system design of liquid propellant rocket's propulsion system.