Condition Monitoring in a Wind Turbine Planetary Gearbox Using Sensor Fusion and Convolutional Neural Network

Abstract

To meet the requirements of reducing operations and maintenance costs of wind turbines, we present a machine learning framework for early damage detection in gearboxes based on the cyclostationary analysis of sensor data. The application focus is the condition monitoring of wind turbine gearboxes under varying load scenarios, in particular turbulent wind conditions. Faults in the rotating components in the gearbox can leave their signature in the vibrations that can be measured by accelerometers. We analyze data stemming from a simulated vibration response of a 5MW multibody wind turbine model in a healthy and damaged scenarios and under different wind conditions. With cyclostationary analysis applied on acquired sensor data, we generate cyclic spectral coherence maps that highlight signatures related to the fault damage and enable its identification. These maps are used to train convolutional neural networks that then identify faults, including those of small magnitude, in test data with a high accuracy. Benchmark test cases inspired by an NREL study are tested and faults successfully detected.