Data-centric input signal design for highly interactive dynamical systems

Data-centric system identification approaches generate a local function approximation from a database of regressors at a given operating point. This paper studies the design of input signals for data-centric identification of highly interactive multivariable systems which show strong gain directionality. The input signal design formulation aims to develop uniform coverage in the output space by addressing the optimal distribution of time-indexed output points under general operating constraints on the manipulated input and measured output signals. The solution of resulting nonconvex quadratic program is proposed using semidefinite and nonlinear programming methods. A numerical example is shown to highlight the benefit of proposed design in comparison to the input design based on Weyl's criterion for data of finite length.