<p>Accurate prediction of Remaining Useful Life (RUL) for aeroengines is critical for ensuring flight safety and optimizing maintenance. However, existing deep learning models often struggle to effectively capture multi-scale degradation patterns and long-term dependencies, particularly under varying operating conditions. To address these challenges, we propose Multi-scale Temporal-Gated Attention (MTGA), a novel hybrid deep learning framework designed to enhance RUL prediction accuracy and robustness. MTGA uniquely integrates a Multi-scale Temporal Convolutional Network (MTCN) for extracting degradation features across different temporal resolutions, a Gated Recurrent Unit (GRU) to model long-term dependencies efficiently, and an Improved Self-Attention Mechanism (ISA) to dynamically refine feature importance while suppressing noise. Unlike conventional hybrid models that apply standard attention mechanisms, MTGA’s ISA enhances interpretability by adaptively weighting input features based on their relevance to degradation progression. Extensive evaluations of the Commercial Modular Aero-Propulsion System Simulation (C-MAPSS) dataset demonstrate that MTGA consistently outperforms state-of-the-art methods such as IMDSSN, PI-CNN, and ATCN. For instance, MTGA achieved RMSE reductions of 19.43% and 25.73% over IMDSSN on FD002 and FD004 datasets, respectively, and Score reductions of up to 50.17% compared to ATCN on FD004, particularly under complex operational conditions. The framework’s ability to balance predictive accuracy and computational efficiency makes it well-suited for real-time predictive maintenance in aerospace applications. These findings highlight MTGA’s potential to improve aviation maintenance and safety strategy decision-making.</p>