On the detection of change-points in structural deformation analysis

Abstract

In structural deformation analysis the behaviour of the monitored structure is typically described in a dynamic model, by deducing a weighting or transfer function. Its parameters can be estimated from the recorded influencing and deformation signals, using system theory and time series or regression analysis. The analysing functions or the adjustment models can be determined using the entire data set, if the assumption of stationarity up to the 2 order is fulfilled. This is the case if the monitoring activity extends over a long time and the influences on the structure maintain their statistical properties from a long-term point of view. However due to the higher recording rates made possible by some modern sensors also shortterm deformations and influencing factors like wind or traffic, which expose a more irregular pattern, can be registered and included in the investigation of the structures’ behaviour. In these cases the stationarity assumption needs a more careful analysis. The data segments with homogeneous statistical properties have to be identified and different model parameters have to be estimated for each of them. This paper deals with an approach for the detection of the change-points in the statistical properties of the data. The method is based on the likelihood function. In this approach the change-points are estimated by minimising a penalised contrast function. To get a better insight in the behaviour of the structure when several effects overlap, the signals are first decomposed using the Discrete Wavelet Transform. The change-point method is applied to the obtained signal components. The performance of the approach is assessed by analysing simulated and real data, recorded during the monitoring of a wind energy turbine and a vertical lift bridge.