Identification Of Health Status Stages Of Wind Turbine High Speed Shaft Bearing With Deep Learning

Abstract

Mechanical components in wind turbines with an unstable operating environment under variable weather conditions are at a very high risk of wear. This situation brings about sudden unexpected stops of components and high maintenance costs. It is of great importance to plan appropriate maintenance times in order to ensure continuity in energy production, prevent unexpected unplanned downtime and minimize maintenance costs. Therefore, before a component failure occurs, the health status must be carefully monitored and maintenance periods must be planned according to the wear and tear process. In this paper, in order to evaluate the health status of a real wind turbine high-speed shaft bearing, the healthy, degradation and fault stages of the bearing are identified by a deep learning based classification model. In the proposed study, vibration data obtained from a real wind turbine high-speed shaft have been used. The study basically consists of the steps of extracting the features of the vibration data, selecting the features that will effectively reveal the health process of the bearing, obtaining the health index by integrating the selected features, and classifying the health index into stages with the LSTM deep learning model. In the study where four different health stages are defined, an accuracy of 99% has been obtained on the test data.