Machine learning methods for predictive maintenance using real-time data and time-frequency analysis

Abstract

In this diploma thesis, different techniques of Predictive Maintenance based on Machine Learning are compared. In particular, the Remaining Useful Life of a ball bearing of the shaft of a Wind Turbine was predicted with different methods: Classification algorithms, degradation models and real time updates using a Kalman Filter. In the first half, the theory of ball bearing failure mechanisms, predictive maintenance and machine learning is analyzed. At the second half, different methods are implemented for the prediction of the remaining useful life. Last, the writer comes to a conclusion about the efficiency of each method.