Real-Time Fault Diagnosis in Wind Turbines via Latent Covariance PLS and Enhanced Contribution Analysis

Abstract

With the increasing reliance on wind energy systems, the development of intelligent and accurate monitoring strategies has become essential to ensure their reliability, efficiency, and operational safety. This study presents an enhanced fault detection and diagnosis (FDD) framework specifically designed for complex, nonlinear systems. The approach leverages a Latent Covariance Partial Least Squares (LC-PLS) model in combination with a refined variable contribution analysis method. Unlike traditional Principal Component Analysis (PCA) and Partial Least Squares (PLS) techniques, the proposed LC-PLS method effectively captures the underlying latent correlations among process variables while providing improved fault separability using a Mahalanobis distance-based monitoring index. The integrated detection–isolation scheme is validated using publicly available benchmark wind turbine datasets, demonstrating superior diagnostic performance under various fault scenarios. Results indicate high fault detection rates, reduced false alarm probabilities, and significantly improved fault isolation accuracy. These outcomes highlight the robustness and real-time applicability of the proposed LC-PLS and contribution-based framework in modern wind energy monitoring systems.