APPLICATION OF MACHINE LEARNING TO PREDICTION OF TURBINE ROTOR VIBRATION IN STEAM POWER PLANT

Abstract

Abstract

Presented in this study is a predictive approach to the maintenance of turbine rotors in thermal power plants. Using a supervised machine learning technique, a model that could predict future vibrations was developed on the simulation platform MATLAB. Historical data on the vibration symptoms of the turbine-generator couple in a generating unit of a steam power plant was employed in the model to predict the future technical condition of the plant component after the model had already been trained with a portion of the turbine-generator section's operational data. Distribution of the test values of the data about the lines of regression was obtained by quantitative analysis; likewise, the model's ability to correctly predict items that were not used in the training process was also measured. Performance evaluation of the model shows mean square error and mean absolute error of 0.000013691 and 0.0025, respectively, at training, with 0.0011 and 0.0037, respectively, at prediction. Future maintenance needs of the turbine rotor can thus be determined by comparing predictions with the vibration safety threshold of the rotor. Operators of modern power plants can leverage the approach of this study to model and plan maintenance schemes that best suit individual unitofpowerplants, rather than premising maintenance of plant components on the rule-of-thumb.

Keywords:    turbine rotor, machine learning, power plant, predictive maintenance, electricity generation