Hybrid temporal convolutional network-reservoir computing model for enhanced remaining useful life prediction in...

Area of Science:

• Aerospace Engineering
• Artificial Intelligence
• Machine Learning

Background:

• Predictive maintenance in aerospace is critical for safety and cost reduction.
• Conventional deep learning models (LSTM, CNN) face challenges like high computational costs and difficulty capturing long-term degradation.
• Accurate Remaining Useful Life (RUL) prediction is essential for effective aeroengine health monitoring.

Purpose of the Study:

• To introduce a novel hybrid deep learning model integrating Temporal Convolutional Networks (TCNs) and Reservoir Computing.
• To address limitations of existing models in RUL prediction, such as computational expense and sensitivity to noise.
• To enhance the reliability and reduce maintenance costs in aerospace through improved predictive maintenance.

Main Methods:

• Developed a hybrid deep learning model combining TCNs and Reservoir Computing.
• Utilized the NASA C-MAPSS dataset, a standard benchmark for RUL estimation.
• Evaluated performance using Root Mean Squared Error (RMSE) and a penalty-based Prognostics and Health Management (PHM) score.

Main Results:

• The hybrid model achieved competitive RMSE values across different data subsets (FD001-FD004).
• Significantly improved PHM scores were obtained, indicating enhanced timeliness in failure prediction.
• The proposed hybrid architecture outperformed standalone TCN and Reservoir Computing components and other benchmark methods.

Conclusions:

• The hybrid TCN-Reservoir Computing model offers a practical, real-time solution for aeroengine health monitoring.
• The approach demonstrates superior performance in RUL prediction compared to existing methods.
• This advancement contributes to improved aerospace reliability and reduced maintenance expenses.