Linear model identification for rotorcraft using adaptive learning

Convergence to a unique identification result with an optimum model structure is a goal in rotorcraft system identification. Whether using time domain or frequency domain methods, achievement of this target requires additional tools, startup procedures/algorithms or a-priori information about the plant. In this paper, an adaptive learning based methodology is proposed to improve parameter convergence. The bounded convergence is guaranteed and robust to initial conditions even when there exist redundant derivatives in the initial state-space structure. A converged solution is obtained as a starting point and a typical bias-variance trade-off is performed. The effectiveness of the method is demonstrated through the identification of a Level-D class high-fidelity nonlinear helicopter model. The converged solution and the reduced order model can also be used in other system identification methods/algorithms as a starting identification state.