Time-domain model identification of structural dynamics from spatially dense 3D vision-based measurements

Abstract

A novel approach is presented for the time-domain system identification of structural dynamic components exploiting the high spatial density of vision-based measurements. By using spatially dense measurements, the number of spatial measurement points is much larger than the dimensionality of the underlying dominant dynamics (i.e., the number of measurement points needed to meet observability of the targeted dynamics). This opens up the potential to develop new experimental identification methods that use this spatial overdetermination which were out of reach with conventional discrete sensors. The new approach presented in this paper directly extracts a relatively noise-free, low-order set of dynamic states by projecting the spatially dense time-domain measurements on a low-order dominant deformation motion basis. The basis is constructed by applying an over-complete singular value decomposition on the measurements. These dynamic states together with their numerically calculated first and second order time-derivatives are then used in a two-step identification of the structural dynamic parameters. Here, the structure of the model to identify is based on a-priori physical knowledge of the underlying set of partial differential equations, unlike a purely data-driven method. This approach has the benefits of providing a mathematical formulation that is straightforward to understand and implement and is time-efficient to solve. The presented approach is experimentally validated on a clamped plate as well as on a flexibly suspended plate which requires a correction of the rigid body motion. The identified models have shown to provide an accuracy up to 1 × 10−5 m with respect to the dominant measured motion components in both validation cases.