Neural networks for the burn back performance of solid propellant grains

In this study, the design of a solid propellant grain is optimized using neural networks. Variables must be balanced while designing a solid propellant grain to achieve the required performance. An optimized design is proposed for solid propellant grains with improved efficiency based on a neural network. Burning surface training datasets for grains created using various design variable values are obtained. Deep neural network, recurrent neural network, long short-term memory (LSTM), bidirectional LSTM, gated recurrent unit (GRU), and GRU-FC models are trained on the aforementioned training datasets, and their prediction accuracies are compared. The post-training model accuracy is evaluated by varying the amount of training data for the neural network that achieved the highest accuracy. By training a neural network using burning surface data for the target grain, the design variable values are predicted, and the model accuracy is verified.