Comparing Various Types of Artificial Neural Network Metamodels in Finocyl Grain Design

Grain design is the most important part of solid rocket motor design. In this paper, the goal is implementing and comparing various Artificial Neural Network metamodels in Finocyl grain design based on predetermined objective function with respect to thrust history or pressure history in order to satisfy various thrust performance requirements through an innovative design approach using genetic algorithm optimization method. The classical sampling method is used for design space filling. The level set method is used for simulating the evolution of burning surface in the propellant grain. An algorithm is developed beside the level set code that prepares the initial grain configuration using Pro/Engineer to export generated models to level set code. The zero dimensional method with considering the effect of erosive burning is used to perform internal ballistic analysis. The meta-model is used to surrogate the level set method in optimization design loop. The surrogate method is based on artificial neural network based on Multi Layer Perceptron with comparing various training algorithm and Radial Basis Function. Finally, a case study is done to verify the proposed algorithm. Observed results show that grain design method reduced the design time significantly and this algorithm can be used to design of any grain configuration.