Damage identification technique based on time series models for LANL and ASCE benchmark structures

摘要

A damage identification technique for structural health monitoring (SHM) using auto-regressive time series models is presented and its effectiveness is demonstrated using two benchmark problems. The proposed damage detection scheme consists of two phases. In the first phase, the presence of damage (ie the exact time instant of damage) is identified using the prediction errors of auto-regressive (AR) and auto-regressive with exogenous input (ARX) models, constructed from the current data and the corresponding matched reference data. Once the presence of damage is identified, the second phase is activated in which a vector auto-regressive model (VAR) is employed to extract the spatial damage feature in order to precisely locate the spatial single/multiple damage. Two benchmark problems for damage detection, an eight degrees of freedom (8-DOF) system from Los Alamos National Laboratory (LANL) and the Phase I ASCE benchmark problem from the IASC-ASCE Structural Health Monitoring Task Group, are used to demonstrate the proposed damage identification technique. The results of the studies carried out in this paper clearly indicate that the proposed two-phase model is robust in identifying the time instant of damage, as well as the spatial location of single/multiple damages in the structure, with operational variability and practical levels of measurement noise.