Trim of rotorcraft multibody models using a neural-augmented model-predictive auto-pilot

Abstract

The aeromechanical analysis of rotorcraft using comprehensive multibody vehicle models is crucially dependent on the ability to accurately compute the model trim settings. Among the various techniques proposed in the literature, the auto-pilot approach, being independent of the complexity of the model, has the potential to solve trim problems efficiently even for the highly detailed aero-servo-elastic vehicle models that are developed using modern finite-element-based multibody analysis codes. Published proportional auto-pilots show to work well in many practical instances. However, their robustness with respect to the flight condition is often poor, so that they must be accurately tuned. In this paper, an auto-pilot based on adaptive non-linear model-predictive control is proposed. The formulation uses a non-linear reference model of the rotorcraft, which is augmented with an adaptive neural element. The adaptive element identifies and corrects the mismatch between reduced and comprehensive models, thereby improving the predictive capabilities of the controller. The methodology is tested on the wind-tunnel trim of a rotor multibody model and compared to an existing implementation of a classical auto-pilot for comprehensive rotorcraft analysis applications. The proposed controller shows improved stability over the conventional approach without the need for calibration.