Diagnosis of Fuel Cell Health Status with Deep Sparse Auto-Encoder Neural Network

TL;DR

The paper tackles online diagnosis of fuel cell health using high-frequency impedance, a metric difficult to deploy with conventional electrochemical impedance spectroscopy. It adopts a Deep Sparse Auto-Encoder (DSAE) that enforces sparsity via KL divergence, with the loss given by $J_{total}=J(W,b)+\psi\sum_{k=1}^{q} KL(\xi \| \xi_k)$ and $KL(\xi \| \xi_k)=\xi \log\frac{\xi}{\xi_k}+(1-\xi)\log\frac{1-\xi}{1-\xi_k}$, while configuring a 10-input, 3-output network with two hidden layers (32 and 16 neurons) using ReLU activations and Adam optimization. The approach achieves over 92% accuracy in Python and, when deployed on an FPGA, yields a hardware recognition rate of around 90%, demonstrating feasible real-time FC health monitoring. This work provides a scalable, cost-effective alternative to expensive online impedance testing and supports robust health assessment across varying operating conditions. The FPGA deployment further enables embedded, real-time monitoring in automotive and energy systems, contributing to safer and more reliable fuel cell operation.

Abstract

Effective and accurate diagnosis of fuel cell health status is crucial for ensuring the stable operation of fuel cell stacks. Among various parameters, high-frequency impedance serves as a critical indicator for assessing fuel cell state and health conditions. However, its online testing is prohibitively complex and costly. This paper employs a deep sparse auto-encoding network for the prediction and classification of high-frequency impedance in fuel cells, achieving metric of accuracy rate above 92\%. The network is further deployed on an FPGA, attaining a hardware-based recognition rate almost 90\%.

Figures (14)

• Figure 1: The structure of DSAE network
• Figure 2: The training process of DSAE network
• Figure 3: Training and validation accuracy
• Figure 4: Mean squared error trend
• Figure 5: The confusion matrix