Modal parameter identification based on Singular Value Decomposition and backward prediction

Modal parameter identification is crucial for mechanical system control and surveillance. The application of time series model has been investigated in this issue for long time. However, nonstationary property of vibration makes it difficulty to find an accurate time series model to fit the measured data. We explore in the paper a backward prediction based method to identify modal parameters. Technique of Singular Value Decomposition for a Hankel matrix is adopted to deal with ill-conditioning occurred in linear system equation solving. The neural network was used to find the Singular Value Decomposition. Numerical examples are presented to show the effectiveness of our method in modal parameter identification.