Scholarlink

<p>Precisely determining the remaining useful life (RUL) of Aeroengines is crucial to the implementation of predictive maintenance, ensuring the safety of operations, and reducing the maintenance-related expenses in the aviation sector. This paper presents an extensive deep learning architecture that combines state-of-the-art models, including Generative Adversarial Networks (GANs), Variational Autoencoders (VAEs), Convolutional Neural Networks (CNNs), and Bidirectional Long Short-Term Memory (BiLSTM), all trained together using the Lion Optimisation Algorithm (LOA). The GAN component is responsible for generating highly realistic synthetic degradation paths to address data scarcity. Meanwhile, the VAE is the component that discovers the compact latent health representations, which are crucial for describing the degradation process. CNN-BiLSTM, the module for prediction, can recognize both local spatial correlations and long-range temporal dependencies in multivariate sensor signals; it recognizes them effectively. LOA is used for cooperative hyperparameter tuning of the entire structure, performed in parallel, thereby improving not only convergence stability but also overall predictive performance. The system proposed in this article is tested on the NASA C-MAPSS turbofan engine dataset, a standard dataset, across different operational modes and fault scenarios. Experimental results demonstrate that the proposed method delivers highly accurate, competitive, and stable RUL prediction performances even when the degradation environment changes, particularly in late-life and data-scarce operating regimes. In addition, the setup exhibits better capabilities in both degradation representation and latent health-index and sensor-importance analyzes, providing improved interpretability. These findings suggest that the proposed approach is a resilient and efficient solution for Aeroengine prognostics and predictive maintenance applications.</p>

错误：搜索内容不能为空，请输入英文关键词
错误：关键词超出字数限制，请精简
高级检索

An integrated GAN–VAE–CNN–BiLSTM framework with lion optimization for remaining useful life prediction of aeroengines

  • 期刊论文

  • 出版模式：

    开放获取

  • 发表日期：

    2026年6月22日

查看全文
  • Aarti,
  • Gireesh Kumar Devineni,
  • Kranthi Kumar Singamaneni

摘要

Precisely determining the remaining useful life (RUL) of Aeroengines is crucial to the implementation of predictive maintenance, ensuring the safety of operations, and reducing the maintenance-related expenses in the aviation sector. This paper presents an extensive deep learning architecture that combines state-of-the-art models, including Generative Adversarial Networks (GANs), Variational Autoencoders (VAEs), Convolutional Neural Networks (CNNs), and Bidirectional Long Short-Term Memory (BiLSTM), all trained together using the Lion Optimisation Algorithm (LOA). The GAN component is responsible for generating highly realistic synthetic degradation paths to address data scarcity. Meanwhile, the VAE is the component that discovers the compact latent health representations, which are crucial for describing the degradation process. CNN-BiLSTM, the module for prediction, can recognize both local spatial correlations and long-range temporal dependencies in multivariate sensor signals; it recognizes them effectively. LOA is used for cooperative hyperparameter tuning of the entire structure, performed in parallel, thereby improving not only convergence stability but also overall predictive performance. The system proposed in this article is tested on the NASA C-MAPSS turbofan engine dataset, a standard dataset, across different operational modes and fault scenarios. Experimental results demonstrate that the proposed method delivers highly accurate, competitive, and stable RUL prediction performances even when the degradation environment changes, particularly in late-life and data-scarce operating regimes. In addition, the setup exhibits better capabilities in both degradation representation and latent health-index and sensor-importance analyzes, providing improved interpretability. These findings suggest that the proposed approach is a resilient and efficient solution for Aeroengine prognostics and predictive maintenance applications.