Breathing Crack Detection Using Dynamic Equations and Measurement Data Regression and Filtering Techniques

Abstract

Structural health monitoring (SHM) traditionally starts from a parametric dynamic model representing damages or defects. The analysis of the dynamic model subjected to vibrational excitations provides the relationship between the response and the damage parameters; this relationship is the basis for determining structural damage from the response measurements. In recent years, data-driven and machine learning methods have been successfully applied to various SHM problems. In this paper, we propose a new method that uses both dynamic equations and measurement data regression and filtering techniques to detect breathing cracks. This method circumvents the dynamic analysis of the system model. We conducted a series of empirical studies based on noisy measurement data and compared the results of several regression and filtering algorithms in our simulations. We found that the proposed method was effective in detecting breathing crack with up to 5% noises in measurements. Among the various data processing algorithms, least squares linear regression following a support vector regression filter on noisy measurement data performed the best.