Bayesian-Driven Optimization of MDCNN-LSTM-RSA: A New Model for Predicting Aeroengine RUL

Accurate estimation of remaining useful life (RUL) in aeroengines is essential for improving safety, reducing operational costs, and optimizing maintenance strategies within the aviation sector. This study introduces a novel Bayesian optimization (BO) multiscale dilated convolutional neural networks long short-term memory (LSTM) residual self-attention framework, which synergistically combines multiscale dilated convolutional neural networks to extract intricate spatial features, LSTM networks to model temporal dependencies, residual self-attention to enhance feature stability and relevance, and BO to automate and refine hyperparameter selection. Traditional methods often fall short due to limited feature extraction capabilities and reliance on manual parameter tuning, challenges that this integrated approach effectively addresses. The framework’s performance is rigorously evaluated using the C-MAPSS dataset, a widely recognized benchmark, across its diverse subdatasets, revealing substantial enhancements in predictive accuracy and reduced error metrics. An ablation study systematically assesses the individual contributions of each component, confirming their collective impact on overall performance. Furthermore, a detailed full life cycle time series analysis for a representative engine demonstrates the model’s ability to precisely track degradation patterns over its entire operational duration, offering clear evidence of its predictive reliability. Compared to earlier techniques that depend solely on convolutional or recurrent neural networks, this framework exhibits superior adaptability and precision, with promising potential for extension to other machine learning architectures and a broader range of datasets. The approach mitigates common pitfalls such as overfitting and instability, providing a stable foundation for real-world deployment. This research significantly advances the field of RUL prediction by delivering an automated, scalable, and robust tool tailored for the complex dynamics of aeroengine systems, paving the way for future explorations into adaptive maintenance technologies and enhanced prognostic methodologies.