Experimental Investigations on Crack Detection Using Modal Analysis and Prediction of Properties for Multiple Cracks by Neural Network

Abstract

In the present study, a method is proposed for detection and prediction of properties of multiple transverse cracks on simply supported stepped rotor shaft. Two cases of cracks are considered. Initially, both cracks are perpendicular to axis. Later, both cracks are inclined to vertical plane and also inclined with each other. Modal analysis is performed to extract natural frequency and mode shapes. Finite element method (FEM) is treated as basis for numerical analysis. For validation, experimentation is performed using fast Fourier transform analyzer. Based on natural frequency, cracks are detected. The results of FEM and experimentation are found in agreement. Crack properties are predicted in forward technique using artificial neural networks (ANN). The database of natural frequencies is used to train the network of ANN to predict the crack properties. Applicability of the method is verified by comparing the predictions of ANN with FEM and experimentation. The predictions of ANN and results given by FEM and experimentation are found in agreement. It envisages that the method is competent, suitable and would be alternate to the existing methods. It enhances the performance of structural integrity assessment and online conditioning and monitoring.