A reliability-aware and explainable CNN–BiGRU–Attention framework for remaining useful life prediction of turbofan engines

The accurate and trustworthy estimation of Remaining Useful Life (RUL) is vital for maintenance guided by risk and the safe operation of turbofan engines. In safety-critical prognostics, prediction models are required to accomplish not just great precision but also to show durability, comprehensibility, and relevance to engineering. This investigation presents a Reliability-Aware CNN–BiGRU–Attention model for predicting Remaining Useful Life based on the NASA C-MAPSS turbofan engine deterioration dataset. The recommended architecture merges one-dimensional convolutional layers for the extraction of local degradation-sensitive patterns, bidirectional gated recurrent units for understanding temporal dependencies in both directions within sensor sequences, and an attention mechanism to prioritize the most critical time steps during RUL estimation. Unlike standard CNN–GRU models, the proposed system explicitly combines bidirectional temporal learning, attention-based sequence prioritization, multi-run robustness assessment, and explainable AI. SHAP and LIME are applied to offer comprehensive and specific interpretations of sensor contributions, while attention weights are further scrutinized to augment temporal understanding. All C-MAPSS subsets, including FD001, FD002, FD003, and FD004, undergo experimentation using an engine-wise data partitioning method to eliminate temporal leakage. The outcomes reveal that the recommended CNN–BiGRU–Attention model attains superior accuracy and stability in RUL predictions relative to CNN, GRU, BiGRU, CNN–GRU, and CNN–BiGRU reference models. The evaluation of statistical robustness through numerous independent executions substantiates that the proposed model retains low variability amongst chaotic training scenarios. The research reveals that the recommended framework delivers an exact, trustworthy, and elucidative decision-support mechanism for turbofan engine predictive maintenance.