Damage Detection Using Identified Uncertain Modal Parameters from Operational Modal Analysis

Modal parameters identified from vibration responses under operational conditions inherently involve uncertainties due to measurement noise, inputs, and varying operational conditions. Hence, damage detection using such identified modal parameters should adopt a statistical framework that incorporates such uncertainties. This study develops a statistical approach to detect structural damage using the Damage Locating Vector (DLV) method. The uncertainty in modal parameters may be obtained either through multiple test datasets or from a single dataset using appropriate modal identification techniques. Using a single dataset, the uncertainty of the identified modal parameters is propagated using first order Taylor series expansion to compute the mean and variance of the cumulative stress (CS) obtained using DLV. The computed CS is then compared with that of baseline (undamaged) condition using Mahalanobis squared distance to identify the location of damage. If multiple baseline datasets are available, the probability distributions of Normalized Cumulative Stresses (NCS) in different structural elements can be obtained using DLV. The NCS distributions obtained for the current state of the structure are then compared with those obtained from a baseline state using the Jensen–Shannon divergence, based on which the structural element may be detected to be in same condition as baseline or damage. The method is validated using two examples. The first example is a continuous bridge under moving heavy train load, with the variability in the identified modal parameters induced by time-varying nature of the train-bridge system. In the second example, a laboratory scale steel frame is subjected to a suite of base excitations, where the uncertainty in the identified modal parameters is induced by the uncertainty in the applied base excitations applied through shaking table. It is seen that, in both the examples, the methods could correctly detect the damaged structural elements.