Global LPV model identification of flapping-wing dynamics using flight data

Biologically-inspired flapping-wing micro aerial vehicles are characterised by nonlinear, unsteady aerodynamics and complex dynamics, both highly challenging to model. To take full advantage of the flight capabilities of such vehicles, it is necessary to obtain insight into their dynamics in the different flyable conditions, and to provide adequate control in all of these conditions. Nonetheless, the dynamics are typically only considered in a single flight regime, and controllers are frequently tuned for a particular flight condition. Due the high complexity of flapping flight and limited availability of accurate free-flight data, global models are not yet readily available, particularly models based on free-flight data and suitable for practical applications. This paper demonstrates an approach to obtain a global dynamic model for a flapping-wing micro aerial vehicle. To allow for standard linear control and systems theory to be applied, the nonlinear dynamics are approximated using a linear parameter-varying (LPV) approach based on a set of local linear models. The scheduling parameters, and the parameters in the underlying local models, are determined using system identification methods applied to free-flight data collected on a real test platform, and covering a significant part of the flight envelope. The proposed approach allows for modelling of the vehicle and prediction of the dominant dynamic properties across the considered part of the flight envelope, using a total of 16 parameters, as opposed to the starting point of 46 local models with 12 parameters each. The use of a single model adapting to the flight condition provides flexibility and continuous coverage, and is therefore highly useful for simulation and control applications. While in the explored part of the flight envelope the nonlinearity was found to be limited, such that a weighted average model may be sufficient for some applications, the LPV model provides a higher accuracy and more consistent performance across the conditions considered. Additionally, the approach is shown to be promising and is expected to be adaptable to cover more significant variation. Improvements could be obtained through more extensive flight envelope coverage, more accurate measurement and more informative identification data.