MACHINE LEARNING METHOD FOR CRACK CLASSIFICATION USING NATURAL FREQUENCIES

Abstract

By using an analytical approach, the research paper aims to present a method for locating and classifying cracks in beam-like structures, made of steel. By applying known equations, the training data consisting of Relative Frequency Shifts (RFS’s) values are calculated for multiple damage scenarios considering transverse and branched cracks. After the RFS database is created, the MATLAB software is used to train a feedforward artificial neural network (ANN) that will be able to predict the crack’s location, type and evaluate its severity. The results show that the described model has a high accuracy in determining if the crack is in incipient state, or it has further penetrated the material and it also can predict the crack location in any of the two states.