MTCAT: A Modern Temporal Convolution and Enhanced Attention Transformer Model for Remaining Useful Life Prediction of Aerospace Self-Lubricating Bearings

As mission-critical components in aircraft flap actuation systems, aerospace self-lubricating bearings play a pivotal role in ensuring operational safety during critical flight phases. Conventional machine learning approaches for remaining useful life (RUL) prediction suffer from limited time-series and nonlinear modeling capabilities. Especially individual models, built on a single algorithmic paradigm, struggle to address the complex, multidimensional challenges of RUL prediction. To solve the above challenge, this study proposes an innovative hybrid deep learning method integrating modernized temporal convolutional networks (ModernTCNs) with attention-enhanced Transformer (MTCAT) for RUL prediction. First, ModernTCN employs optimized dilated convolutions to capture multiscale local temporal patterns efficiently. Second, an attention-enhanced Transformer leverages multihead self-attention to model long-range dependencies and global degradation trends. Finally, the MTCAT is validated on the custom test rig, replicating real-world conditions, where it achieves a 30%–50% reduction in root mean square error (RMSE) compared to baseline models. By autonomously learning hierarchical features from raw data, the MTCAT model minimizes manual feature engineering, demonstrating robustness and scalability for RUL prediction.