Sensor hybridization using neural networks for rocket terminal guidance

Abstract

The Guidance, Navigation and Control (GNC) of air and space vehicles, including artillery rockets, has been one of the spearheads of research in the aerospace field in recent times. Increasing the accuracy of these ballistic projectiles is the major goal. Using Global Navigation Satellite Systems (GNSS) and inertial navigation systems, accuracy may be detached from range. However, during the terminal stage of flight, when movement is governed by highly changing nonlinear forces and moments, GNC strategies based on these systems cause enormous errors in determining attitude and position. These effects can be diminished using additional sensors, independent of jamming and cumulative errors, such as the quadrant photo-detector semi-active laser. This paper proposes a novel non-linear hybridization algorithm, which is based on neural networks, to feed GNC systems while complexity and costs are reduced. It fuses the information from multi-sensor signals, such as GNSS, inertial navigation systems, and semi-active lasers, to predict the line of sight vector, which joins the target and the projectile and drives the flight. As compared to traditional approaches, the use of a neural network presents the advantage that once the network is trained, it is no longer necessary to know the physical-mathematical foundations that govern the dynamics of flight. Instead, it is the network that learns the dynamics. Six-degree-of-freedom non-linear simulations, which are based on real flight dynamics, are used to train the neural networks in the hybridization process. Once training is finished, the approach is tested and simulated together with modified proportional navigation techniques and control methods. Monte Carlo analysis is conducted to determine the suitability of the closed-loop performance across a full spectrum of uncertainty conditions regarding launch, sensor performance, atmospheric, and thrust.