Research

We determine how an edge crack alters the waves and vibrations of a beam, providing the theoretical basis for detecting and sizing cracks.

We develop analytical and semi-analytical formulations for the modal and wave behaviors of beam-like waveguides with an edge crack, for which shear deformation and rotary inertia require a Timoshenko-type description. The crack is represented by a local compliance derived from linear elastic fracture mechanics (LEFM). At the wave level, closed-form scattering coefficients are derived under axial–bending–shear coupling for both propagating and evanescent incident waves, and their consistency is verified through the power-flux balance. At the modal level, wave-based descriptions yield explicit frequency equations that clarify when and why the sensitivity of modal frequencies to a crack increases, providing a basis for both forward prediction and inverse crack identification.

We embed the governing mechanics into machine learning so that reliable predictions can be made from sparse measurements.

We develop physics-informed learning frameworks for structural problems in which the governing equations are known but prediction and parameter estimation become ill-posed under sparse or noisy measurements. Mechanical constraints and evolution laws are embedded in the learning process to obtain data-efficient and physically consistent models for response prediction and state/parameter inference. Current applications include the time-dependent behavior of prestressed concrete members and the assessment of multiple incipient cracks from broadband modal frequencies measured with piezoelectric wafers.

We use field measurements to estimate the actual prestressing force and camber of PSC members, narrowing the gap between design assumptions and as-built behavior.

We develop models and estimation procedures for the initial and long-term states of prestressed concrete members, whose effective prestress and camber are governed by construction conditions and by time-dependent creep, shrinkage, and relaxation. Mechanics-based formulations of time-dependent prestress loss are combined with field measurements: tendon elongation is used to estimate the tendon force and friction coefficients that define the initial state of post-tensioned members, and strain measurements are used to reconstruct camber, which is difficult to measure directly.

We transmit and receive ultrasonic guided waves with thin piezoelectric sensors bonded to a structure to detect hidden damage.

We investigate the excitation, propagation, and scattering of guided waves generated and sensed by surface-bonded piezoelectric (PZT) wafers for active structural health monitoring. The work includes time-reversal processes of Lamb waves, reference-free damage diagnosis based on electromechanical signals, efficient spectral element modeling of coupled PZT–structure systems, and wave-based interpretation of electromechanical admittance. The same sensing principle has recently been extended to in-situ monitoring of the evolution of dynamic properties in cement-based materials.

We establish statistical criteria that can reliably declare damage despite measurement noise and changing operating conditions.

We develop statistical frameworks for deciding whether damage has occurred from measured damage features that are inevitably affected by noise and operational variability. Because such decisions are governed by the tail of the feature distribution, extreme value statistics are used to establish baseline distributions and decision thresholds, including parameter estimation of the generalized extreme value (GEV) distribution and the asymptotic convergence of its domain of attraction. These statistical tools are combined with time-reversal processes to enable baseline-free damage diagnosis through consecutive outlier analysis.

We develop numerically stable methods to back-calculate structural properties and states from incomplete, noisy measurements.

We develop numerically stable inverse-analysis and system-identification methods to infer structural parameters and states from limited, noise-contaminated measurements. Emphasis is placed on regularization design (Tikhonov, 1-norm, and spatial regularization) and on the rational selection of regularization parameters. Applications include the identification of inclusions in finite bodies, the detection of abrupt damage induced by an earthquake using a moving time window, and the reconstruction of displacement from measured acceleration in low-frequency dominant structures.