On sequencing the feature extraction techniques for online damage characterization

Abstract

The current state of the health-monitoring technology lacks a generalized and definitive approach to the identification and localization of mechanical damage in structural materials. In past decades, several signal-processing tools have been used for solving different health-monitoring problems but the commutability of the tools between different problems has been restricted. The fundamental reasons for this shortcoming have never been investigated in detail. A thorough study is presented in this article employing almost all promising feature extraction tools on a representative problem—a plate with rivet holes. The cracks around rivet holes in a joint panel of a steel truss bridge are very difficult to detect. Although well established, Lamb wave–based nondestructive evaluation techniques are revisited and new tools are developed to address this issue. The simulation of scattered ultrasonic wave field is carried out using the finite element method. This ultrasonic wave field is further analyzed to evaluate the integrity of the structure using various feature extraction techniques. The joint time–frequency–energy representation is obtained from ultrasonic signals recorded at various locations on the plate (joint panel) and used to extract damage-sensitive features. Those features were then used to formulate a new damage parameter for better visualization of the crack. The results are shown to demonstrate the comparative effectiveness of these techniques. It is concluded that any particular feature extraction technique cannot detect all possible sizes and orientations of the crack. It is suggested that the statistical occurrence and pattern of the crack must be visualized through few selective feature extraction techniques in a sequence.