Structural steel plate damage detection using DFT spectral energy and artificial neural network

Abstract

In this paper, simple methods for crack identification in steel plates and their classification based on the frame based frequency domain features is presented. Based upon the boundary conditions and experimental modal analysis, two simple experimental methods are designed to measure the vibration at different positions of the steel plate. The plate is excited by an impulse signal and made to vibrate. The propagated vibration signals are then recorded. The signal is transformed into frequency domain by computing the Discrete Fourier Transformation (DFT). The frequency spectral bands are identified and the spectral energy is extracted as features. The condition of the steel plate namely healthy or faulty is associated with the extracted features to form a final feature vector. Two simple neural network models were developed, trained using Backpropagation (BP) and Radial Basis Function (RBF) algorithms. The results and the effectiveness of the system are validated through simulation.