Interpretable fault diagnosis framework for offshore wind turbine gearbox based on AFS and signal analysis theory

Fault diagnosis of offshore wind turbine gearboxes is crucial for extending equipment lifespan and reducing maintenance costs. However, traditional data-driven methods often emphasize model accuracy and computational efficiency while neglecting model interpretability. To address this issue, an interpretable fault diagnosis framework based on Axiomatic Fuzzy Set (AFS) theory and signal analysis theory is proposed, referred to as AFSBWFA. This framework aims to maintain high fault recognition accuracy while enhancing the transparency and comprehensibility of the computational process. The proposed framework consists of four main components: data acquisition, signal preprocessing, feature selection, and pattern recognition. First, based on the obtained fault data set, we introduce an efficient method for raw signal denoising and reconstruction, referred to as BWF, combining Black Kite Algorithm (BKA), Wavelet Packet Decomposition (WPD), and Feature Mode Decomposition (FMD). Next, leveraging entropy theory in signal analysis, we design a two-dimensional time-frequency domain feature extraction method based on Multiscale Fuzzy Entropy (MFE), denoted as MFETF. Finally, by delving into AFS theory, we develop a novel conceptual classifier based on EI algebra, namely AFSCC, to achieve accurate identification of fault patterns in offshore wind turbine gearboxes. The effectiveness of the proposed framework was validated using a private dataset provided by Dalian Maritime University (DMU) and a public dataset from Beijing Jiaotong University (BJU). Experimental results demonstrate that the framework exhibits excellent interpretability and achieves 100 % diagnostic accuracy across different datasets. Comparative analysis with existing advanced diagnostic methods indicates that the proposed framework outperforms shallow machine learning algorithms in terms of evaluation metrics and achieves comparable performance to deep learning approaches. Furthermore, it incorporates rich semantic information, offering a novel technical reference for future research on fault diagnosis of gearboxes in offshore wind turbines.