Accurate remaining useful life (RUL) prediction is essential for condition‐based maintenance and safe operation of turbofan engines. Although hybrid convolutional and recurrent architectures have shown promising results, many existing studies emphasize single‐run predictive accuracy while providing limited evidence on robustness, reproducibility, and engineering interpretability. To address this gap, this study presents a hybrid Convolutional Neural Network–Gated Recurrent Unit (CNN–GRU) framework for RUL prediction on the NASA C‐MAPSS benchmark and evaluates it under a reliability‐oriented experimental protocol. The CNN component extracts local degradation patterns from multivariate sensor sequences, whereas the GRU component captures longer‐term temporal dependencies associated with progressive engine deterioration. In contrast to accuracy‐only reporting, the proposed evaluation protocol includes controlled comparisons with standalone CNN and GRU baselines, a corrected component‐level ablation design, repeated training runs for stability analysis, and explainability assessment using SHAP and LIME. Experiments are reported not only on FD001 but also across FD002–FD004 in order to assess behavior under multiple operating conditions and fault modes. The hybrid model achieves the best overall predictive performance among the tested architectures while maintaining low run‐to‐run variability. Explainability results further indicate that the model relies on degradation‐sensitive sensors associated with thermodynamic changes in turbofan operation. Overall, the findings suggest that the proposed framework provides a reproducible, interpretable, and competitively accurate approach for benchmark RUL prediction, while also clarifying its current limitations with respect to real‐world deployment.
Yıldırım Özüpak (2026) studied this question.