Deep Convolutional Identifier for Dynamic Modeling and Adaptive Control of Unmanned Helicopter

Area of Science:

• Aerospace Engineering
• Control Systems
• Machine Learning

Background:

• Helicopters present significant challenges for system identification due to their complex, nonlinear, and time-varying dynamics.
• Accurate dynamic modeling is crucial for developing effective adaptive control strategies for helicopters.
• Existing methods often struggle to capture the intricate coupling of aerodynamic forces and engine dynamics.

Purpose of the Study:

• To develop a robust system identification method for helicopter dynamic modeling.
• To improve adaptive flight control by accurately representing system uncertainties and hidden states.
• To validate the proposed method using real-world helicopter flight data.

Main Methods:

• System identification formulated as dynamic regression.
• Utilizing a deep convolutional neural network (CNN) to model system uncertainties and hidden states.
• Learning CNN parameters directly from real aerobatic helicopter flight data.
• Integrating the CNN identifier with a backstepping-based adaptive controller.

Main Results:

• The deep convolutional identifier demonstrated strong robustness and high accuracy, even for untrained maneuvers.
• The proposed method effectively captured the dynamic behavior of hidden states and uncertainties.
• Experimental validation using real-world flight data confirmed the identifier's effectiveness.
• The adaptive flight control scheme achieved asymptotic convergence of position and attitude tracking errors.

Conclusions:

• Deep convolutional neural networks offer a powerful approach for helicopter system identification.
• The proposed method enhances the robustness and accuracy of adaptive helicopter control.
• The integrated control scheme guarantees stability and precise tracking performance for unmanned helicopters.