Experimental verification of detection and prediction of multiple cracks by vibrations, FEM and ANN

Abstract

Purpose – The purpose of this paper is to address the determination of crack location and depth of multiple transverse cracks by monitoring natural frequency and its prediction using Artificial Neural Networks (ANN). An alternative to the existing NDTs is suggested. Design/methodology/approach – Modal analysis is performed to extract the natural frequency. Analysis is performed for two cases of cracks. In first case, both cracks are perpendicular to axis. In second case, both cracks are inclined to vertical plane and also inclined with each other. Finite element method (FEM) is performed using ANSYSTM software which is theoretical basis. Experimentation is performed using Fast Fourier Transform (FFT) analyzer on simply supported stepped rotor shaft and cantilever circular beam with two cracks each. Findings – The results of FEM and experimentation are validated and are in good agreement. The error in crack detection by FEM is in the range of 3-15 percent while 5-20 percent by experimentation. The database obtained by modal analysis is used to train the network of ANN which predicts crack characteristics. Validity of method is investigated by comparing the predictions of ANN with FEM and experimentation. The results are in good agreement with error of 7-16 percent between ANN and FEM while 9-21 percent between ANN and experimental analysis. Originality/value – It envisages that the method is capable. It is an effective as well as an alternate method of fault detection in beam/rotating element to the existing methods.