Residual-Enhanced Convolutional Transformer for Robust Rolling Bearing RUL Prediction

Rolling bearings serve as critical components in rotating machinery, where accurate remaining useful life (RUL) prediction under variable operating conditions presents a fundamental challenge for industrial predictive maintenance. In order to construct a prediction model with high robustness within a known operating range, this study proposes a Residualenhanced Convolutional Transformer that combines strided convolutions with residual connections for local fault feature extraction and a Transformer encoder for long-term degradation modeling. Through systematic ablation studies, optimal hyperparameters were identified to establish an effective training strategy. Evaluation on the XJTU-SY accelerated lifetime test dataset demonstrates consistent performance with R² scores reaching 77.23 % on the validation dataset and 81.94 % on the testing dataset, which outperforms conventional CNN and ResNet model. The results indicate that the proposed model effectively captures overarching degradation patterns from multi-condition data, confirming its robustness in modeling bearing deterioration and its potential for practical health monitoring systems.