A health index-based approach for fuel cell lifetime estimation

Area of Science:

• Energy Systems Engineering
• Materials Science
• Computational Science

Background:

• Accurate remaining useful life (RUL) prediction for fuel cells is vital for operational efficiency and safety.
• Existing health indicator (HI) extraction methods struggle under dynamic operating conditions with variable loads.
• Challenges exist in optimizing prediction models and achieving high accuracy for fuel cell RUL estimation.

Purpose of the Study:

• To develop a robust prediction framework for fuel cell RUL under dynamic conditions.
• To propose novel methods for extracting effective health indicators (HI) from operational data.
• To enhance the accuracy and adaptability of RUL prediction models.

Main Methods:

• Health indicators (HI) were extracted using a combined approach of complete ensemble empirical mode decomposition with adaptive noise, power spectral density, and energy analysis (CPE).
• A time convolution network with adaptive Bayesian optimization (AB-TCN) was employed for RUL prediction.
• Random forest was utilized to identify effective feature parameters for training the AB-TCN model.

Main Results:

• The extracted HI effectively determined the fuel cell's end-of-life.
• The AB-TCN model achieved accurate RUL estimation with a prediction error of 6.825%.
• The proposed framework demonstrated strong adaptability across various prediction tasks.

Conclusions:

• The developed CPE method successfully extracts health indicators under dynamic fuel cell operation.
• The AB-TCN model provides a highly accurate and adaptable solution for fuel cell RUL prediction.
• This framework offers a significant advancement in monitoring fuel cell health and predicting operational lifespan.