Remaining Useful Life Estimation Using Deep Convolutional Generative Adversarial Networks Based on an Autoencoder...

Area of Science:

• Engineering
• Computer Science
• Artificial Intelligence

Background:

• Accurate Remaining Useful Life (RUL) prediction is vital for system reliability and Prognostics and Health Management (PHM).
• Traditional methods often struggle with complex degradation patterns in systems like turbofan engines.

Purpose of the Study:

• To propose an integrated deep learning approach for enhanced RUL prediction in turbofan engines.
• To improve feature extraction capabilities for better prognostic accuracy.

Main Methods:

• An integrated deep learning model combining an Autoencoder (AE) and a Deep Convolutional Generative Adversarial Network (DCGAN) for pretraining.
• Utilizing a Long Short-Term Memory (LSTM) network in the fine-tuning stage to capture sequential dependencies for RUL prediction.
• Validation on the NASA Commercial Modular Aero-Propulsion System Simulation (C-MAPSS) dataset.

Main Results:

• The integrated AE-DCGAN-LSTM approach demonstrated superior feature extraction capabilities.
• Achieved excellent prediction performance for RUL, outperforming existing state-of-the-art methods.
• Verified effectiveness on the C-MAPSS dataset.

Conclusions:

• The proposed data-driven prognostic method offers a promising new approach for RUL prediction.
• The integrated deep learning framework provides an efficient scheme for feature extraction in PHM.
• This research advances the field of predictive maintenance for critical components.