Neural Network-based High-energy High-intensity Too Root Propellant Performance Prediction

Traditional propellant formula is designed on the basis of a basic formulation, then series of propellants are prepared by varying the contents of different components, its performances are tested and characterized at last. But the effect of different components on the mechanical performance, combustion and energy property are complicated, a lot of human, material and financial resources are wasted in this kind of study. So it will supply valuable guidance and application to achieve the prediction of propellant performances using the existing experimental data. BP (Back Propagation) neural network has unique ability of approximating nonlinear function and generalization, it can find out the complicated relationships between input and output. So it provides a reasonable and effective method for the prediction of propellant performances.On the basis of disposing and analyzing experimental data, the database of the mechanical performance, combustion and energy property of TEGN gun propellant is established, which is used as study sample of neural network. Prediction software based on artificial neural networks is developed through Visual Basic 2008 programming. And the software runs well. It can be used for forecasting pressure stress strain at normal and low temperature respectively, impact intension at normal and low temperature, propellant force, residual volume, burning rate, pressure exponent of TEGN gun propellant. It possesses of better man-machine interactive interface. The user can manage the database, input parameters to train the neural network, examine the train results, forecast sample data, save the forecasting results and so on. In practical train, appropriate structure and train parameters of network model are determined by adjusting different parameters and comparing with the train results, then the reasonable, fast and accurate prediction network model of TEGN gun propellant is established. The experimental result shows that the relative error between predicted value and the measured value is within 10%. Therefore the prediction of TEGN gun propellant performances by this network model is relatively accurate and feasible.