Abstract

Wind turbines are critical components in renewable energy systems, where early fault detection is essential to ensure reliable operation and reduce maintenance costs. Vibration-based monitoring using multichannel signals provides rich information about the dynamic behavior of the system, although it also introduces challenges related to high dimensionality and feature redundancy. This paper proposes a machine learning-based methodology for fault classification that combines Discrete Wavelet Transform (DWT) for time–frequency feature extraction with Linear Discriminant Analysis (LDA) for dimensionality reduction within a structured processing pipeline. The approach incorporates a Group K-Fold cross-validation strategy to prevent data leakage and ensure a reliable evaluation when working with segmented signals. Experimental results show that the proposed framework achieves high classification performance, reaching a mean accuracy of 98.84±1.16% and a weighted F1-score of 0.9905±0.0089 using a Support Vector Machine (SVM) classifier over five Group K-Fold splits. The results also indicate that dimensionality reduction plays a critical role in improving class separability, having a greater impact than the specific choice of wavelet transform. Findings demonstrate that the proposed DWT–LDA-based approach provides an effective solution for rotor imbalance detection in the laboratory-scale wind turbine evaluated in this study.

IPC Classification

Keywords