Diffusion models enable high-fidelity prediction of fuel cell impedance spectrum from short time-domain profiles

Online acquisition of polarization kinetics data from impedance spectrum is critical for fault diagnosis and health assessment of fuel cells to overcome lifetime bottleneck, but it is technically challenging due to stringent measurement requirements and complex evolution under different operating conditions. To address this, we propose an impedance spectrum prediction method based on the denoising diffusion implicit model, which requires only collectable short time-domain profiles and can generate high-fidelity impedance spectrum through a stepwise denoising process. To validate our method, we performed extensive comparative impedance tests under various operating conditions for fuel cells with different power levels, constructing four datasets comprising over 5700 samples in total. Through performance evaluation experiments, robustness tests, and transfer learning experiments, our proposed method has demonstrated satisfactory performance and outperforms existing methods by down to 37.3% on median mean absolute percentage error. As a useful resource, we have released the trained model and the largest datasets of impedance spectrum for fuel cells to date. Reliable fuel cell monitoring needs rapid access to impedance. Here, the authors use a denoising diffusion model to predict high-fidelity impedance spectrum from short signals, beating existing methods and releasing large open datasets.