University of Auckland Centre for Earthquake Engineering Research


Time series based structural health monitoring

A simple time series model that relates the current value of the time series to the previous values using a series of coefficients.

In this research, Auto-Regressive (AR) models were used to fit the dynamic response of a structure under earthquake excitation. The AR model is a simple time series model that relates the current value of the time series to the previous values using a series of coefficients. The coefficients of the AR models are used as damage sensitive features ie, the values of coefficients are expected to change with damage.

To successfully quantify and locate damage requires changes in the coefficients to be identified and interpreted using a statistical technique. Artificial Neural Networks (ANNs) are data processing structures that mimic the function of the biological brain. ANNs have been used widely throughout the civil engineering field. The ANN is used to relate patterns in the AR coefficients to damage patterns in the structure.

In 2007 research focused on applying the developed SHM technique to more realistic real-world structures. The structures had more complex damage patterns and possibly exhibit non-linear behaviour:

  • ASCE SHM benchmark structure is a 3.6m high 4-storey 2-bay by 2-bay steel frame structure. Damage is simulated by the removal of bracing elements and/or the loosing of beam connections. Data from the experiments is freely available.
  • RC bridge pier damaged progressively using an electro-magnetic shaker at The University of Auckland. The response of the bridge pier is likely to be non-linear.

Research also focused on developing an online method for tracking damage as it occurs using recursive estimation techniques to identify time series coefficients online. Recursive techniques include forgetting factors, Kalman filter and prediction-error methods. Initially, simple analytical lumped mass models will be used with a prescribed reduction in stiffness. Later more realistic non-linear structural models were planned to be investigated.