Disturbance source identification for multivariable control

A unique method for identifying the directional information of how stochastic disturbances propagate through the dynamics of a linear, multivariable system is presented. No new theoretical results are used to arrive at the method. Rather, it is an integration of existing time and frequency domain identification techniques into an iterative curve fitting procedure. Beginning with an input to output model of the system and time domain output measurements, the method first produces a nonparametric estimate of the output power spectral density and then fits a parameterized model to it using a non-linear least squares tuning algorithm. The main advantage of this procedure is the ability to directly judge and enhance the quality of the estimates by graphically evaluating the differences in the individual nonparametric and parametric spectral density estimates. A simulation, in which a wind gust disturbance that excites an aircraft, is identified illustrates the quality of the estimates produced by the method. The estimate is then used to design controllers which reject the wind gust disturbance. Doing so demonstrates that the method does provide the necessary directional information needed to obtain good levels of disturbance rejection performance