On Modeling of Vibration and Crack Growth in a Rotor for Prognostics

Abstract

Prognostics and health management (PHM) include comprehensive engineering approaches that evaluate the real-time health condition of an asset and predict its future states under the actual operating conditions. This predictive ability would result in efficient maintenance approaches such as Condition Based Maintenance (CBM) that can set maintenance strategies optimally and reduce the life cycle costs. Diagnostics and Prognostics are two major concepts in PHM. Detection, Isolation and Identification of faults are done by diagnostics while prognostics deals with estimation of future states. Mechanical fatigue phenomenon that causes crack initiation and propagation is considered as a common reason for failure in mechanical parts. Hence, diagnostics and prognostics of the crack initiation and propagation have been the subject of many research papers recently.

The current paper presents a diagnostics and prognostics method capable of detecting the crack initiation and propagation in a rotor under cyclic loading. At the first step, the coupled equations of rotor motion and crack growth are obtained. An extended model of Paris–Erdogan equation is used for crack growth modeling. The coupled equations are solved numerically. A set of features are extracted from the dynamic response of the rotor for a range of crack lengths. A dataset is compiled including features of response, operating frequency, crack length and number of cycles remained until reaching the critical crack length. With the objective of generalization of the results, the dataset is used for creating a model using an Artifical Neural Network (ANN). In the trained ANN the inputs are the operating speed and the outputs are the crack length and the remaining useful life (RUL) that address the diagnostics and prognostics objectives, respectively.