Gain Scheduling Using Deep Neural Network for Morphing Aircraft

A methodology is presented in this study for the automated gain scheduling of multiple linear systems utilizing deep neural networks. The focus is on a morphing aircraft's short-period dynamic motion and the application of a PID controller. The morphing aircraft dynamics system can be effectively modeled as a linear parameter-varying system in relation to the morphing parameters. As the morphing parameters change, different linear systems are considered. To address this, a deep neural network structure is employed, integrating an LPV system dynamics simulator. The gain scheduler neural network is trained to optimize the unit step response, enabling it to output control gains based on the morphing parameters. A performance comparison is conducted between the scheduler trained using the proposed methodology and a manually-tuned scheduler.