Grey-box modeling of a small-scale helicopter using physical knowledge and Bayesian Techniques

Identification experiments for small-scale helicopters are usually difficult to implement, and the data collected are to some extent non-informative and insufficient. This makes it impossible to identify a parametric model accurate enough for controller design. This paper presents a Bayesian method for identification modeling a small-scale helicopter. Priori physical knowledge is fully applied to simplify the dynamics and obtain two parametric state-space models for both longitudinal and lateral motions. The unknown parameters are explicitly expressed in a more reasonable way. A Bayesian maximum a posteriori (MAP) estimation is formed and translated into a constrained nonlinear optimization problem, which is solved by a Lagrange multiplier method using a DFP-based quasi-Newton recursive algorithm. A ldquosinchrdquo algorithm is applied to map the direct continuous-time domain parameterization problem into the discrete-time domain. The continuous-time state-space model acquired shows good prediction performance and is suitable for controller design.