Structural Dynamics with Uncertainty: Design and Identification Using Intelligent Inference

Structural design and identification are two import aspects in engineering practice. The former aims at enhancing the functional performance based on optimizing the rational design objective. The latter is commonly employed in determining the health status, i.e., the occurrence, location and severity of damage in the structure monitored. In reality, these efforts are inevitably influenced by various uncertainties, i.e., modelling error, measurement noise, and environmental/operational variation etc. In order to achieve effective and robust structural design and identification based on the dynamic responses, in this dissertation a series of inter-related tasks are undertaken, including: A robust design optimization approach in the mean of intentional substructural mistuning to mitigating the negative consequence of random uncertainty in nominally periodic structures, e.g., engine bladed disks, is formulated. Structural dynamic behaviours are often sensitive with respect to surface geometry variation. Based on the NURBS finite element which in theory is conformal with the underlying NURBS geometry, an efficient sensitivity analysis-based inverse optimization is formulated. To further enable NURBS finite element based forward- and inverse- analyses for very large scale structures, a new model order-reduction technique based on the component mode synthesis strategy is formulated. At the presence of uncertainty/variation, model identification/updating, which plays an important role in design and structural health monitoring, should be carried out in the probabilistic sense. In this part of dissertation, Bayesian inference is integrated with Gaussian process to enable the direct updating of structural finite element model within commercial package by using measurement data. Under certain scenarios, system identification has to be performed with minimal baseline information. An improved mass-response method is formulated, which can identify the supporting stiffness of a bridge-type structure without a priori knowledge To fundamentally enhance the probabilistic prediction efficiency, in the final part of the dissertation a two-level Gaussian process based updating of order-reduced model is synthesized.