ARMA modelled time-series classification for structural health monitoring of civil infrastructure

Abstract

Structural health monitoring (SHM) is the subject of a great deal of ongoing research leading to the capability that reliable remote monitoring of civil infrastructure would allow a shift from schedule-based to condition-based maintenance strategies. The first stage in such a system would be the indication of an extraordinary change in the structure's behaviour.

A statistical classification algorithm is presented here which is based on analysis of a structure's response in the time domain. The time-series responses are fitted with Autoregressive Moving Average (ARMA) models and the ARMA coefficients are fed to the classifier. The classifier is capable of learning in an unsupervised manner and of forming new classes when the structural response exhibits change. The approach is demonstrated with experimental data from the IASC–ASCE benchmark four-storey frame structure, the Z24 bridge and the Malaysia–Singapore Second Link bridge. The classifier is found to be capable of identifying structural change in all cases and of forming distinct classes corresponding to different structural states in most cases.