EMPLOYING MACHINE LEARNING TO DETECT POST RESONANCE BACKWARD WHIRL IN A FAULTY ROTOR SYSTEM

Machine Learning (ML) algorithms have several applications in various fields of science. One of the major applications of ML is to recognize patterns in input signals and classify the input accordingly to various classes. Employing ML with structural health monitoring tools to identify faults has become an emerging area in research. Faults in rotor systems generally cause asymmetries in the rotor shaft stiffness which further results in generating higher harmonic components. One of the recent tools that has been prevalently used to identify faults is the post-resonance backward whirl phenomena (Po-BW). It can be said that any asymmetry in a rotor system should be affecting the dynamic of the whirling response of the system which can be further identified by detecting the Po-BW in the response. A two degree of freedom (2DOF) Jeffcott rotor (JC) model with a healthy and faulty shaft is used here to establish feature datasets. The obtained datasets are later used as input to train a deep learning network in the form of Alexnet to classify the input images based on the faults such as an open crack or/and unbalance related faults. The obtained results has indicated to classification accuracy of above 99%.