Fault detection in cracked structures under moving load through a recurrent-neural-networks-based approach

Abstract

The current work is based on the development of an indirect approach in the domain of Recurrent Neural Networks (RNNs) to identify and quantify cracks on a multi-cracked cantilever beam structure subjected to transit mass. At first, the responses of the multi-cracked structure subjected to transit load are determined using fourth order Runge-Kutta numerical method and finite element analysis (FEA) has been executed using ANSYS software to authenticate the employed numerical method. The existences and positions of cracks are identified from the measured dynamic excitation of the structure. The crack severities are found out by FEA as forward problem. The modified Elman's Recurrent Neural Networks (ERNNs) approach has been implemented as inverse problem to predict the locations and severities of cracks in the structure by applying Levenberg-Marquardt (LM) back propagation algorithm. The present analogy has been carried out in a supervised manner to check the convergence of the proposed algorithm. The proposed ERNNs method converge good results with those of theory and FEA.