Neural network based modal identification of structural systems through output-only measurement

This paper proposes the application of neural networks for output-only modal identification of structural systems. Four frequency-dependent indicators, based on specific properties of the spectral tensor of vibration measurements, are defined and employed to build a likelihood function for the presence of structural resonances. Subsequently an artificial neural network (ANN), fed with the four indexes, was built and adopted to assess structural eigenvalues and eigenmodes. The proposed technique was tested on a three-storey steel-frame. After training the trained ANN was able to assess eigenvalues and eigenmodes, as demonstrated by comparison of the obtained results with those provided by literature methods.