UC Berkeley

This dissertation develops computational methods for the nonlinear modeling of frame members and their use in rapid infrastructure assessment following extreme events. The work is motivated by the need to convert recorded structural motions into quantitative assessments of condition at the speed and scale required for regional operations. In this setting, physics-based assessment requires beam and column models that can represent geometric nonlinearity and multiaxial inelastic material response, without sacrificing the efficiency of one-dimensional frame analysis. At the same time, data-driven assessment requires a computational framework in which recorded motions, system identification procedures, structural simulation, and model calibration may operate within a common workflow. In Part I, a general framework is developed for the constitutive modeling of beam sections with cross-sectional warping. The framework is formulated in the setting of geometrically exact beam theory with an arbitrary number of global warping fields and remains applicable to linearizations of that model, including classical Timoshenko beam theory. Enhanced section strains are introduced with amplitudes determined locally at each cross section, thereby improving the three-dimensional strain field used for multiaxial constitutive integration while preserving the standard one-dimensional equilibrium equations. The proposed section models may therefore be used with existing beam finite element implementations without modification of the element state determination procedure. Two specializations are developed. For members for which warping boundary effects are insignificant, the framework yields warping-free section models that recover Saint--Venant continuum solutions in the elastic regime, including Poisson-coupled in-plane warping effects, and are compatible with classical six-field beam elements. For restrained torsion, one torsional warping field is retained globally and an additional correction is resolved locally, yielding a seven-field beam theory that restores pointwise equilibrium in the elastic regime and rectifies the equilibrium defect of standard single-field torsion-warping theory. The resulting formulations are organized for finite element implementation through a separation of frame, section, and material models. Numerical studies verify exact agreement with Saint--Venant solutions in the elastic regime and demonstrate improved response in shear-dominated inelastic problems, restrained torsion, evaluation of the Wagner term, and finite-deformation benchmarks. In Part II, a framework is developed for operational structural health monitoring that integrates physics-based simulation and data-driven inference within a common assessment environment. The framework is implemented in BRACE2, developed in collaboration with the California Department of Transportation to deliver post-earthquake assessments for instrumented highway bridges in California. Structural assessment is organized around five abstractions, Assets, Events, Predictors, Metrics, and Evaluations, and realized through a four-layer architecture separating presentation, orchestration, domain, and infrastructure concerns. This organization permits heterogeneous predictors to operate through a common event-evaluation workflow, reduces their outputs to a shared metric representation with explicit provenance, and assembles those metrics into actionable reports for closure and inspection decisions. A second contribution of Part II is a compositional formulation for computing analytic gradients of finite element responses. In this formulation, a host-level parameterization constructs the solver parameters required by the analysis kernel, kernel-level direct differentiation furnishes derivatives with respect to those parameters, and full sensitivities with respect to the original modeling variables are recovered by the chain rule. The resulting construction supports model updating, reliability analysis, and integration with automatic-differentiation toolchains employed in machine learning. These developments furnish a computational environment in which recorded motions, nonlinear structural simulation, differentiable calibration, and operational decision-making are brought into a single workflow for rapid infrastructure assessment. Following a pilot deployment covering 22 bridges, an implementation of the proposed framework is being advanced toward operational use by approximately forty engineers of the California Department of Transportation, with coverage extending across the state highway network.