Optimizing aircraft engine longevity: A comparative framework for dynamically adaptive predictive maintenance using autoencoders, LSTMs, and Gaussian processes

The aviation industry's reliance on fixed-interval maintenance schedules for aircraft engines poses significant inefficiencies, balancing precariously between excessive resource expenditure and catastrophic safety risks. Traditional frameworks, such as the 50-h inspection cycle, fail to account for real-time engine degradation, often leading to unnecessary downtime or undetected failures. Recent machine learning (ML) advances offer transformative solutions through predictive maintenance (PdM), which dynamically adapts schedules based on engine health data. This study evaluates three ML methodologies: Autoencoders (AEs), Long Short-Term Memory networks (LSTMs), and Gaussian Process Regression (GPR) to construct degradation indicators, predict Remaining Useful Life (RUL), and optimize maintenance timing. This work addresses a critical gap in adaptive maintenance strategies by comparing their efficacy in anomaly detection, temporal modeling, and uncertainty quantification. It offers a data-driven pathway to enhance safety, reduce costs, and extend engine operational life. • Addressing Fixed-Interval Maintenance Limitations to meet the critical balance between safety and operational efficiency. • Dynamically Adaptive Predictive Maintenance with application of Autoencoders (AEs), Long Short-Term Memory networks (LSTMs), and Gaussian Process Regression (GPR). • Comparative Evaluation of Machine Learning Methods in PdM dynamic adaptation. • Data-driven pathway to optimize aircraft engine longevity. • The transformative potential of ML-based predictive maintenance in optimizing aircraft engine life cycle.