Detection of rotor cracks: comparison of an old model-based approach with a new signal-based approach

Abstract

The detection of shaft cracks within operating rotors is an old subject of scientific research. Several signal- and model-based approaches are developed. Here two approaches used for crack detection in rotating machinery, model-based and signal-based approaches, are compared. Strength and weak points are discussed and compared for the two approaches using two representative applicable methods, in order to achieve a comparative overview of these two available techniques. The PI-observer-based method is considered, as modern model-based technique, to give indication about possibilities and limitations of such kind of methods. A novel signal-based approach is introduced, based on SVM and wavelets as an example for a modern machine learning technique. The concepts of severity estimation and service life prediction are investigated in the proposed approach. Furthermore, a brief comparative discussion is presented in the contribution, including ideas for combination of the introduced approaches, in order to achieve more comprehensive and more robust monitoring system applied to the detection of shaft cracks in rotating machines.