Geodetic Institute
Logo Geodetic Institute
 
Abschlussarbeiten
Machine learning-based damage identification of oscillating structures using features derived from robust time domain modal parameter identification approach

Machine learning-based damage identification of oscillating structures using features derived from robust time domain modal parameter identification approach

Led by:  Mohammad Omidalizarandi
E-Mail:  zarandi@gih.uni-hannover.de
Year:  2024

The structural health monitoring (SHM) provides valuable information about the structure by assessing the current state of the health of a structure, detecting unsafe conditions and unexpected behaviour or structural damages using regular measurements over time (Dawson, 1976). In this master thesis, a machine learning approach is applied, developed and implemented for detecting anomalies such as abnormal behaviours and damages in oscillating civil engineering structures. For this purpose, first, the dynamic response of the oscillating structure such as eigenfrequencies, eigenforms and modal damping (known as modal parameters) are assumed to be estimated in advance. Robust, accurate, and automatic estimation of the modal parameters are obtained based on robust time domain modal parameter identification (RT-MPI) approach proposed in Omidalizarandi et al. (2020). The eigenfrequencies and damping ratio coefficients are estimated based on an observation model consisting of a damped harmonic oscillation model, an autoregressive (AR) model of coloured measurement noise, and a stochastic model in the form of the heavy-tailed family of scaled t-distributions with unknown degree of freedom and scale factor. The aforementioned three parametric models are jointly adjusted by means of a generalised expectation maximisation (GEM) algorithm. The eigenforms are characterised in a subsequent step, and by using the estimated parameters from the GEM algorithm (Omidalizarandi, 2020).
In order to perform damage detection, in addition to the estimated modal parameters, AR model
coefficients are also estimated, which are then input into an Artificial Neural Network (ANN) (de
Lautour et al. 2010). The optimal selection of the AR model order allows to select the uncorrelated
features within the ANN. The aforementioned modal parameters together with AR model  coefficients are estimated for a healthy structure, which allow to train ANN and to detect damages. Optionally, localisation of damage parts is also of great interest.


References:

Omidalizarandi, M., Herrmann, R., Kargoll, B., Marx, S., Paffenholz, J. A., & Neumann, I. (2020). A
validated robust and automatic procedure for vibration analysis of bridge structures using MEMS
accelerometers. Journal of Applied Geodesy, 14(3), 327-354.

Omidalizarandi, Mohammad: Robust deformation monitoring of bridge structures using MEMS
accelerometers and image-assisted total stations. München 2020. Deutsche Geodätische

Kommission: C (Dissertation): Heft Nr. 859. publikationen.badw.de/de/047037691



de Lautour, O. R., & Omenzetter, P. (2010). Damage classification and estimation in experimental
structures using time series analysis and pattern recognition. Mechanical Systems and Signal
Processing, 24(5), 1556-1569.