The objective of this study was to investigate the attitude algorithm and guidance and control algorithm which are core technologies necessary for the development of smart munitions. Also, each algorithm was designed based on neural networks in accordance with the latest academic trends. In the case of smart munitions, the direct application of widely used algorithms related to guidance, control, and navigation is unreasonable. In terms of the operation mechanism, it is unreasonable because not only are various sensors activated during flight after fire, but the rapidly rotating airframe and the nose that rotates at a specific angular velocity for control move individually. Moreover, the operation perspective of utilizing conventional munitions as intelligent munitions by attaching a kit is a weak solution because such kits have to be applicable to conventional munitions of varying platforms, and there is the aerodynamic characteristic of conventional munitions having significant shape deviation because they are cheaply manufactured.
The scope of this research covered the core algorithms necessary for the development of smart munitions with the characteristics described above, and it was specifically focused on the in-flight attitude algorithm and adaptive control design which can be applied with regard to aerodynamic error.
Additionally, since the final performance of the subject system can be determined through the circular error probability (CEP), an algorithm to create guidance commands and CEP analysis through Monte Carlo simulation were investigated.
The major contribution of this study include the design of a Kalman filter for in-flight attitude estimation and the composition of a neural-network-based adaptive attitude prediction filter to enhance the designed Kalman filter. As a result, the neural-network-based adaptive attitude prediction filter reduced the time delay effect of the Kalman filter.
For the control algorithm design, the longitudinal/directional axis coupling effect, which is an issue in classical control techniques, was reduced using a neural-network-based model inversion controller, and L1 adaptive control was applied to improve the performance of the conventional neural-network-based adaptive controller. In addition, analysis of the time delay margin was performed, and its results were presented.
In the case of the subject system, it was determined that designing guidance commands using the conventional PNG method is unreasonable, so in this study, guidance commands were created using a sensitivity method that utilizes neural-network-based impact point prediction.
All the designed algorithms were composed using a MATLAB-based integrated simulation. Monte Carlo simulation for the error condition of the aerodynamic and initial launch conditions was conducted, and the results are presented.