Damage detection and model updating of a steel frame structure by measured strain and acceleration for improving seismic performance assessment

Abstract

For risk-informed decision-making on necessary actions such as repair, retrofit, and disaster prevention planning for existing structures, it is important to predict quantitatively the response to a future disturbance like an earthquake, and evaluate its consequences (structural and/or non-structural damage), considering uncertainties. For this purpose, an analytical model is quite useful, but generally there exists a discrepancy between an analytical model for design and the counterpart real structure due to: (a) modelling error and over-idealization; (b) variation in construction accuracy; (c) retrofit or conversion; (d) degradation over time and/or damage due to disturbances like earthquakes. It is therefore necessary to timely detect such a discrepancy using observation records and continually update the model so that it allows more precise prediction of seismic response.

Most existing studies regarding structural health monitoring and model updating make use of modal properties such as natural frequencies and mode shape displacements, which are usually identified from measured accelerations. For the case of detecting localized stiffness reduction (e.g. the damage of a column-base connection in steel structures), however, it may be more effective to utilize modal information in terms of not only acceleration or displacement but stress or strain in each structural element.

This paper presents a methodology for damage localization and model updating of a steel frame structure using both acceleration and strain, and validates the proposed methodology by conducting a full-scale vibration experiment. The test structure is a one-bay one-story steel frame on which both accelerometers and strain gauges are installed. Manually exciting the structure yielded the free vibration responses for several damage cases, where the anchor bolts of a column base were gradually relaxed to simulate the different damage cases. Stochastic subspace identification was then applied to the acceleration and strain response time histories to estimate mode shape displacements and strains for different vibration modes. These modal properties allow us to compute the bending moment distribution for each column and update the bending stiffness of the column base.

The results have shown that by the obtained modal properties one can detect the stiffness reduction of the damaged column base and quantitatively evaluate the degree of reduction. The model updated by the proposed method is expected to reduce the uncertainty in the predictions of response to a future disturbance and improve seismic performance assessment.