A deep learning framework for remaining useful life prediction of turbofan engines with partial sensor failure

During long-term operation, turbofan engine sensors often suffer from partial damage or signal loss caused by complex flight environments, harsh mechanical vibrations, and thermal stresses. Such degradation in sensor reliability leads to incomplete or inaccurate monitoring data, which significantly reduces the precision of remaining useful life (RUL) prediction and poses potential risks to the safety and maintenance scheduling of aero-engines. To address this challenge, this paper proposes a novel generative regression model based on Long Short Term Memory Generative Adversarial Network to achieve robust life prediction under sensor damage conditions. The proposed model first employs a missing-parameter generator to fill in the lost sensor data, which helps restore the integrity of feature inputs. Then, we use an RUL predictor to extract the temporal degradation information from the reconstructed features for more accurate RUL estimation. Experiments conducted on the NASA C-MAPSS dataset, a widely used benchmark for turbofan engine degradation analysis, demonstrate that the proposed model maintains high prediction accuracy even under partial sensor failure scenarios and outperforms multiple baseline methods. The results verify the robustness, generalization, and reliability of the model under complex operating conditions. This study provides a task-oriented and unified modeling approach to improving turbofan engine health management, offering valuable guidance for enhancing system safety, reliability, and predictive maintenance efficiency.