Component level signal segmentation method for multi-component fault detection in a wind turbine gearbox

Abstract

Condition monitoring of a modern wind turbine gearbox is quite challenging as it comes with multiple stages which operate at different frequencies. A gearbox is made up of multiple components and fault diagnosis (single or multi-component) could be challenging owing to the interaction between the mating parts and the damaged component. In this investigation, a simplified signal segmentation technique that segments the non-stationary vibration signals to match a specific speed stage and component within a multi-stage gearbox is proposed. This technique improves the features within the dataset and allow even simpler algorithms to be more effective while performing fault diagnosis. The segmentation approach is also evaluated for its robustness with three different machine-learning algorithms, namely Decision tree, Support Vector Machine and Deep Neural Network. The overall classification accuracy of the datasets prepared with the proposed approach is found to be 97%, which is higher when compared to the conventional approach.