UC San Diego

Manmade structures can be described as a collection of smaller substructures joined together at contacting interfaces. Contact nonlinearities present at these interfaces can cause computational bottlenecks in large, high-fidelity finite element models. Consequently, different techniques of model order reduction have emerged, wherein the number of degrees of freedom (DOF), particularly at nonlinear interfaces, is decreased to a manageable level. In the analysis of civil structures, so-called “macro” elements are often employed, which represent entire contact surfaces using uniaxial spring and dashpot elements. Mechanical and aerospace systems, on the other hand, more often use projection-based model order reduction, which transform interface deformations onto a low-dimensional subspace. This research develops high-fidelity and reduced-order models of contacting structures, from large-scale building structures down to small-scale mechanical components. Moat wall pounding, a phenomenon where base-isolated buildings collide with their retaining walls during seismic shaking, is examined as a case study for reduced-order models of civil structures. In the first half of the dissertation, new and existing macro impact elements are verified against experimental data, and then statistically examined in a large parametric study. Following that, two different high-fidelity models of moat wall pounding are developed and compared with corresponding experimentation. Results from these studies indicate that most macro models provide sufficient accuracy for moat wall pounding simulations. Their high-fidelity counterparts, while more difficult to calibrate to experimental data, reveal dynamic behavior that cannot be ascertained with macro elements alone. The second half of the dissertation develops a novel family of methods to reduce the nonlinear interface DOF for preloaded bolted structures. Five such methods are applied to a bolted beam assembly and compared in terms of accuracy and computational savings. Results show that the interface reduction methodology is capable of capturing often-neglected interface kinematics, including transient contact area and joint slip. Furthermore, the methods are tunable to the desired combination of accuracy and computational effort.