Structural verification of linear dynamic models, based on multiple experiment data

This paper examines the problem of structural verification of linear dynamic models, based on multiple independent experiment data. The unmodeled structure is assumed to be an unknown deterministic or stochastic process additive to an assumed baseline model. It is shown that a new state space model can be derived in which the one-step residual sequence is treated as measurements, and the unmodeled process is the input sequence. Incorporating spectral analysis techniques commonly used in signal processing, a methodology, generalized state disturbance approach, is proposed for systematic evaluation of model structures. The applicability of this approach is demonstrated, based on a 19 state inertial guidance model with various types of unmodeled structures.