Predictive analysis toward the identification of cracks in functional gradient beam structures using an optimization algorithm based on the transit search technique

Abstract

Machine learning techniques can be used for the prediction of cracks in beam type structures. Indeed, these techniques present an important management aspect which consists in developing a maintenance decision model, which can anticipate future failure trends in order to improve the maintenance decision process. The prediction of open cracks on the edges of beams is a problem often encountered in industry and can be detected by considering that the crack is simulated by means of a rotating spring, whose stiffness can be identified by the size of the crack. This article proposes two different techniques for crack detection in Euler-Bernoulli model functional gradient beams. The first technique generates frequency contours from a three-dimensional plot of the crack position and size, and the intersection of distinct mode contours predicts the crack location and size. The second technique uses a meta-heuristic optimization, which is inspired by astrophysics and based on well-known exoplanet discovery methods, to determine crack size and position concurrently. The weighted sum of the squared errors between the measured and computed natural frequencies is utilized to design the objective function in this second strategy. The results reveal that the two crack size and location prediction techniques agree quite well.