Research on fault warning for wind turbine performance based on multi-residual autoencoder fusion

Read full text Summarise with VIRAT
Authors: Ronghua Meng, Zongyi He, Zhengjia Wu, Yingying Zhu, Xuan Du
Cited by: 0
References: 11

Summary

No summary yet. VIRAT can read this paper and write a plain-language summary of its aim, methods, key findings and limitations.

Summarise with VIRAT Free to try with your Science Catalogs account.

Abstract

Conventional fault warning approaches relying on single-state-variable thresholds often lead to false or missed alarms. To overcome this limitation, this study introduces a novel system-level fault warning methodology that integrates residual information from multiple source state variables. First, we develop a quartile method (QM)-based adaptive parameter tuning strategy for Density-Based Spatial Clustering of Applications with Noise (DBSCAN-PA), which effectively cleans abnormal wind speed-power data and overcomes the incomplete filtering issues typical of conventional techniques. Subsequently, the Maximal Information Coefficient (MIC) is employed to select relevant features from multi-source state variables. A hybrid prediction model combining a 1D convolutional neural network with an echo state network (1D-CNN-ESN) is then constructed to mitigate gradient vanishing or explosion problems commonly encountered in deep learning (DL) models for time-series forecasting. Finally, an autoencoder (AE) is utilised to fuse the residuals of multiple state variables, producing reconstructed residual information. A comprehensive residual index is derived through a time-windowed weighted average, enabling holistic fault diagnosis at the wind turbine system level. Case studies conducted on two operational wind turbines demonstrate that the proposed framework achieves effective early fault warning with improved reliability and timeliness.

Fields of Study & Subjects

Machine Fault Diagnosis Techniques Fault Detection and Control Systems Wind Turbine Control Systems

Publication details

Journal / source
Digital Twin
Pages
2602304
Publication date
30 March 2026
Language
EN