Fault Diagnostics for Rotating Machinery using Deep Learning Ensemble Model of Frequency Spectrum, Spectrogram, and Orbit

Abstract

The Prognosis and Health Management (PHM) research field is rising in modern industry to minimize the down-time of a mechanical system. In the power plant, the needs in PHM for the rotating machinery in the power plant are increasing since rotating machinery such as a generator, turbine, and pump are carried out an important role. However, a complicated sensory system for diagnosing rotating equipment in power plants is scarcely established. Although the plant could stop running due to a malfunction in rotating equipment, rotating equipment in power plants usually lacks a real-time machine condition monitoring system. For this reason, it is hard to detect the fault of equipment in advance. Therefore, establishing a diagnosis system for rotating equipment is essential to minimize the downtime in the operation process in power plants. The proposed fault diagnosis system aims to enhance the diagnosis performance by combining multi-classifier. Spectrum analysis, spectrogram analysis, and orbit analysis are used to conduct a single network. These are the most commonly used analysis techniques for vibration data. Deep ensemble learning networks and combination networks are used to improve the diagnosis performance.