Use of data-driven phase space models in assessing the strength of a bolted connection in a composite beam

This work explores the role of empirical dynamical models in deducing the level of preload loss in a bolted connection. Specifically, we examine the functional relationship between data gleaned from locations on either side of the connection using nonlinear predictive models. This relationship, as quantified by a measure of prediction error, changes as a function of bolt loosening, thus allowing both the presence and magnitude of the axial load to be identified. The models are based on a phase space portrayal of the system dynamics and require only that the structure's response be low dimensional. The technique is demonstrated experimentally on a composite beam fastened to steel plates with four instrumented bolts. Results are compared to a similar approach using an auto-regressive (AR) modeling technique.