Monitoring seismic damage via accelerometer data alone using Volterra series and genetic algorithm

An application of Volterra series in nonlinear system identification is presented in this paper. This novel approach makes use of accelerometer data alone. The aim is to develop an algorithm that can rapidly detect damage, caused by earthquakes, in infrastructural systems without the need for a parallel transducer system that also monitors relative displacements around predicted damage locations. In addition, this algorithm (here termed Method A ‘Acceleration only’) is shown to be effective for low to high levels of damage. It first extracts the estimated multimodal linear kernel using a genetic algorithm applied to the input/output structural acceleration time-series. This enables a very precise estimate of linear system parameters. It then subsequently extracts quadratic and cubic nonlinear (kernel) terms by making use of multinomial combinations of the wavelet basis of the input signal. Extracted nonlinear kernel acceleration time-series and their standardised cumulative norms are compared with normalised hysteretic dissipated energy which requires both response accelerations and displacements, (here termed method A/D “Accelerations and displacements’). As a heuristic case we investigate the performance of both method A, and method A/D in predicting probable damage in a Bouc-Wen nonlinear system. Results suggest that method A/D and method A are comparable at estimating the likely maximum system ductility. We develop a fragility curve for estimating the probability of damage based on our nonlinear Volterra series intensity measure. Finally, we verify the application of this Volterra series approach against experimental test data from physical laboratory shake-table experiments of reinforced concrete columns and demonstrate that this approach is useable in practice.