An Optimized Neural Network Approach for Rapid Unpressurized Compartment Venting Predictions

Abstract

An examination has been performed of using artificial Neural Networks to predict the unpressurized compartment pressure time history for spacecraft, launch vehicle, and air vehicle applications. In this paper, the development of and results from using artificial Neural Networks to predict the unpressurized compartment pressure time history for a space capsule reentering the Earth’s atmosphere is presented. A Neural Network-based approach was selected for their ability to mimic the compartment pressure time histories using only a small number of high fidelity venting solutions. This approach employs simple multi-layer feed-forward Neural Network architectures with inputs based on the freestream static pressure and Mach time histories, pressure time derivatives and pressure moving averages. The time periods over which the pressure averages are calculated were optimized using a Genetic Algorithm. For repressurization of the Apollo aft bay during reentry, the Neural Networks correctly identified the worst three trajectories, and 18 of the worst 20 cases, out of a 100,000 set of dispersed trajectory data. For the worst case venting trajectory the Neural Network predictions were within 1.3% and 0.25 seconds the peak of the deltapressure value and time-at-peak-pressure, respectively. This new tool runs in the Matlab environment, and after training the Neural Network can predict the compartment pressure time history for a new trajectory within a few seconds.