Operational State Recognition of a Rotating Machine Based on Measured Mechanical Vibration Data

Digital twin is a relatively new concept. Also, it lacks a formal definition and can be applied in virtually any field of technology. Considering digital twins of rotating machines, and especially the in-service phase of their lifecycle, a digital twin should produce valuable information for the owner and operator of the application. The information produced by a digital twin should be accurate, up-to-date, and available anywhere. These requirements act as limiting factors for the complexity of the digital twin and promote the need for efficient data transfer, data acquisition and especially data processing methods at the source of information.
This study investigates how these requirements can be fulfilled in continuous, near real-time operational state recognition of a gas engine genset. Therefore, the objective of this study is to provide a data-based model for operational state recognition and detection of abnormal operation of a gas engine generating set in near real-time.
Two different types of machine learning models for the state recognition of the generating set are presented. The first, a classification model, can identify the current power output level of the generating set using the measured mechanical vibration data. The second, a novelty detection model, can detect abnormal operation of the generating set, in fault situations, at a specific power output level. A two-step state recognition model can be built by combining the classification and novelty detection models.