Automated Fault Diagnosis in Rotating Machinery

Abstract

Rotating machinery are an important part of industrial equipment. Their components are subjected to harsh operating environments, and hence experience significant wear and tear. It is necessary that they function efficiently all the time in order to avoid significant monetary losses and down-time. Monitoring the health of such machinery components has become an essential part in many industries to ensure their continuous operation and avoiding loss in productivity. Traditionally, signal processing methods have been employed to analyze the vibration signals emitted from rotating machines. With time, the complexity of machinery components has increased, which makes the process of condition monitoring complex and time consuming, and consequently costly. Hence, a paradigm shift in condition monitoring methods towards data-driven approaches has recently taken place towards reducing complexity in estimation, where the monitoring of machinery is focused on purely data-driven methods.

In this thesis, a novel data-driven framework to condition monitoring of gearbox is studied and illustrated using simulated and experimental vibration signals. This involves analyzing the signal, deriving feature sets and using machine learning algorithms to discern the condition of machinery. The algorithm is implemented on data from a drivetrain dynamics simulator (DDS), equipment designed by Spectraquest Inc. for academic and industrial research purposes. Datasets from pristine state and faulty gearboxes are collected and the algorithms are tested against this data. This framework has been developed to facilitate automated monitoring of machinery in industries, thus reducing the need for manual supervision and interpretation.