Intermediate Stage Fault Classification for Wind Turbine Gearbox: A Comparative Analysis of Distance Metrics with k-NN Model

Intelligent fault classification is an essential component of failure diagnoses in wind turbine gearboxes. Unfortunately, the intermediate stage of the gear set is still one of the most neglected components in the wind turbine system when it comes to intelligent fault identification. The k-NN algorithm is a widely applied intelligent classification model known for its high accuracy and simplicity. However, this approach's effectiveness heavily depends on the distance metrics used. This paper proposes determining the best distance metric with its optimal 'k' value that produces the most efficient k-NN model for fault classification in this overlooked intermediate stage. 'k' values ranging from 1 to 14 were experimented with eight distance metrics, and their results were compared to determine the optimal model. Based on the optimal 'k' value obtained through cross-validation, the Euclidean distance metric had precision, accuracy, and computation time of 99.90%, 99.85%, and 0.50sec, respectively, followed closely by Cosine, with 99.90%, 99.84%, and 0.04sec, respectively. The authors concluded that the k-NN algorithm using Euclidean and Cosine distance metrics with 12 nearest neighbours is the most reliable for intelligent fault classification with wind turbine gearbox vibration datasets.