AI enhanced fault indicators vs. classical bearing monitoring – example results of bearing tests and transferability to wind turbines

Abstract. Condition monitoring of drive trains is indispensable in the operation of wind turbines. Early knowledge of faults allows for maintenance planning and in-situ counter-measures and thus reduces operational costs. Current commercial methods include significant human supervision and interpretation of measurement data. Larger fleets of assets raise the need for enhanced methods that require reduced supervision and less manual interaction. The present work verifies two ways of using artificial intelligence that fulfill this requirement. These are normal-behaviour models and high-level indicators. The verification includes test data analysis of small-scale bearing tests of Ø100 mm thrust bearings and considerations of transfer to wind turbines. In bearing tests, enhanced monitoring gives comparable or significantly earlier warnings than classical monitoring. In three of five performed tests, the warning thresholds were passed at comparable times, in two, the warnings were significantly earlier and clearer with the enhanced monitoring. As classical monitoring benefits most from the simplified test environment, it is reasonable to assume an even more pronounced advantage for enhanced monitoring in more complex machines like wind turbines.