Separation trajectory safety evaluation platform based on embedded neural network data prediction

Aiming at the high dynamic multi-body separation problem of hypersonic vehicle, intelligent prediction of aerodynamic characteristics and simulations of separation trajectory were carried out, which provided technical support for separation system design, separation window selection, separation scheme optimization and evaluation. Typical states were selected to carry out numerical simulations and grid force measurement in wind tunnel tests, thus an aerodynamic database was established. Embedded neural network with high and low fidelity data was used for the learning and training processes, and the structure of neural network was optimized by the genetic algorithm, leading to the error between the predicted aerodynamic coefficients and those of the grid force measurement less than 5%. The fourth-order Runge-Kutta method was used to solve the six-degree-of-freedom motion equation of aircraft, and the separation trajectory simulation method based on neural network training results was established. The Monte Carlo analysis and sensitivity analysis were carried out under different initial separation conditions to evaluate the main factors affecting separation safety. Compared with the capture trajectory system (CTS) simulation results in wind tunnel, the platform is reliable in separation trajectory prediction, and the cost is low, which can quickly improve the trajectory simulation ability and better support the separation scheme design.