A System Identification Based Approach For Estimating Modal Frequencies of Bridges Under Environmental Influence

Abstract

For health monitoring of bridge structures, vibration response-based methods offer several advantages due to the global nature of the approach. Vibration based structural health monitoring identify damage by detecting the abnormal changes in the dynamic characteristics such as the frequencies and mode-shapes extracted from the vibration response. However, the environmental fluctuations in temperature, radiation, convection, and humidity may change the dynamic characteristics of a bridge structure sometime more than those caused by structural damage and thus mask the damage effects. A method is being demonstrated to estimate the change in the dynamic characteristics directly from the body temperature measurements to separate them from those caused by the damage. In this paper, we use the stochastic subspace system identification technique to estimate the frequency, given the temperature records at specific locations within the bridge. The data used for development and validation of the presented approach has been generated using a finite element analysis capable of translating thermal environmental condition. In this paper, the finite element analysis setup and the identification approach used are presented. It is observed that the temperature and frequency records have prominent yearly and diurnal trends and the relation between temperature values and modal frequencies is nonlinear. A straightforward use of the stochastic identification approach does not provide acceptable frequency estimations, especially when tested on data for different environmental conditions. However, these observed seasonal and diurnal trends motivate us to use filters to improve the results of the identification model over an extended duration spanning the seasonal variations. The identified model is tested for both reproduction and generalization performances over seasonally varying thermal conditions and, in spite of nonlinear effects, the model is shown to provide very good estimates of the system frequencies in the simulation study even under the influence of large measurement noise.