LAUSR

Ambient vibration is an unknown excitation source that may produce stationary or non-stationary signals. Under such circumstances, traditional feature extraction techniques may not yield relevant features to damage and not provide reliable results of damage identification. The main objective of this article is to propose a data-driven method based on the concept of statistical pattern recognition for locating damage under ambient vibration and unpredictable signal nature in terms of simultaneously stationary and non-stationary behavior. This method is generally comprised of a three-level hybrid algorithm for feature extraction and new statistical distance metrics for feature analysis. The proposed feature extraction method aims at providing new damage-sensitive features in three levels including (1) analyzing the nature of measured vibration signals in terms of stationarity or non-stationarity, and normalizing non-stationary signals by detrending and differencing techniques, (2) modeling each vibration signal by an Autoregressive Moving Average (ARMA) model along with extracting the model residuals, and (3) estimating the power spectral density of residual samples as a new spectral-based feature. To identify the location of damage via spectral-based features, this article proposes two new spectral-based measures called Jeffery’s and Smith’s distances. The major contributions of this study include proposing a new feature extraction method for dealing with the problem of unpredictable vibration nature and introducing two new distance metrics for damage identification. Experimental vibration measurements of a well-known laboratory structure are utilized to verify the proposed methods. Results demonstrate that these approaches succeed in accurately extracting relevant features and locating damage under ambient vibration and unpredictable signal nature.