Time-Frequency Feature Fusion Combining Convolutional Neural Networks and Deep Attention Fuzzy Cognitive Maps for Mechanical Component Fault Recognition

Accurate fault diagnosis of mechanical components is critical for ensuring industrial productivity and equipment reliability. However, traditional methods often suffer from noise interference and limited capability in capturing and integrating essential features from complex signals. To address these limitations, this study proposes a novel time-frequency feature fusion framework that combines convolutional neural networks (CNN), Transformer architectures, and deep attention fuzzy cognitive maps (DAFCM). Specifically, one-dimensional vibration signals are first transformed into two-dimensional time-frequency representations via Fast Fourier Transform (FFT). Both the original and transformed features are simultaneously processed through CNN and Transformer modules to extract highly discriminative representations. The extracted features are then fused using DAFCM, which leverages fuzzy relationships and attention mechanisms to enhance feature interactions and improve fault pattern separability. Extensive experiments conducted on multiple mechanical fault datasets demonstrate that the proposed method consistently outperforms conventional approaches and single-modal feature extraction methods. The results highlight the potential of the proposed framework for advancing intelligent fault diagnosis and its applicability in practical engineering scenarios.