A transformer-based method for aircraft engine RUL prediction integrating dual-layer attention with BiLSTM

• Proposes SBi-Transformer, a novel RUL prediction model integrating dual-layer attention and BiLSTM to enhance both global and local temporal feature extraction. • Achieves state-of-the-art performance on NASA C-MAPSS datasets, demonstrating superior accuracy and robustness under complex operating conditions and multiple fault modes. • Ablation studies validate strong synergies among Transformer, sparse attention, and BiLSTM modules, significantly improving prediction performance and uncertainty estimation. Addressing the computational bottlenecks and limited local sensitivity of standard Transformers in processing long sequences, this study introduces the SBi-Transformer—an enhanced architecture tailored for aircraft engine remaining useful life (RUL) prediction. Structurally, the model incorporates a dual-layer attention mechanism designed to simultaneously reduce complexity and sharpen local feature extraction. A distinct architectural departure is the elimination of the conventional decoder; instead, a Bidirectional LSTM (Bi-LSTM) is employed to model encoder outputs directly. This integration explicitly captures bidirectional degradation trends, thereby enriching the local temporal context. Empirical evaluations on NASA’s C-MAPSS dataset confirm that the SBi-Transformer outperforms contemporary baselines in both prediction accuracy and uncertainty estimation. Systematic ablation studies further validate the architectural synergy, revealing how the sparse attention and Bi-LSTM modules complementarily drive performance gains. notably, the model demonstrates robust adaptability under complex fault modes, with confidence intervals that dynamically converge toward the true degradation trajectory. Consequently, this research offers a computationally efficient, high-precision solution with significant potential for practical engineering prognostics.