Open Access Open Access  Restricted Access Subscription Access

MACHINE LEARNING BASED GAIN SCHEDULING FOR CONTROL OF GUIDED ARTILLERY PROJECTILES

A. Anandaraj, P. P. Parkhi, Pooja Agrawal, Ganapati N. Joshi

Abstract


Guided projectiles operate over a wide range of flight conditions in which significant variations in aerodynamic characteristics, dynamic pressure, mass properties, and control effectiveness require continuous adjustment of autopilot parameters to ensure stable and accurate flight. Conventional gain scheduling has been widely adopted to address these nonlinearities; however, it relies on extensive system linearization, multidimensional lookup tables, and interpolation techniques, resulting in design complexity, computational overhead, memory requirements, and verification effort. This paper proposes a machine learning-based gain scheduling framework for the pitch and roll autopilots of guided projectiles. A validated classical gain-scheduled autopilot is employed as a supervisory controller to generate a comprehensive training dataset through Monte Carlo simulations covering the complete operational flight envelope. The generated dataset is used to train a feed-forward neural network that learns the nonlinear mapping between projectile flight conditions and the corresponding control actions. The trained model eliminates the need for multidimensional gain tables and online interpolation while preserving the control characteristics of the classical controller. The proposed methodology offers reduced computational complexity, improved scalability, lower memory requirements, and simplified integration of additional scheduling variables. The implementation methodology, realization, practical deployment aspects, limitations, and future extensions based on physics-informed and hybrid machine learning approaches are also discussed, demonstrating the potential of the proposed framework for next-generation intelligent guided projectile autopilot systems.


DOI
10.12783/ballistics26/37614

Full Text:

PDF

Refbacks

  • There are currently no refbacks.