Investigation of Solid Oxide Fuel Cell Lifetime Improvement Through an Integrated Simulation, Experimental, and Data-Driven Modelling Approach

Abstract

The global transition toward net zero emissions is accelerating the development of reliable, efficient, and durable energy technologies. Solid oxide fuel cells (SOFCs), known for their high electrical efficiency and fuel flexibility, are strong candidates for stationary and distributed power generation. However, their commercial deployment is hindered by limited operational lifetimes, complex degradation phenomena, and fragmented predictive approaches. Existing tools are often either physics-based but difficult to account for degradation, or data-driven yet limited by data and poor physical behaviour reflection.This thesis presents an integrated methodology that combines high fidelity simulation, experimental data acquisition, and deep learning to predict SOFC degradation and extend system lifetime. A comprehensive modelling framework was developed in Aspen Plus, incorporating reforming route analysis and degradation mitigating system optimisation. Multi stack configurations with anode off gas recycling were assessed, with the three-stacks in series, recycling to the pre-reformer inlet (MS3RPR) configuration delivering the highest system performance, 65.3% efficiency and >78,000 hours of projected lifetime.The experimental work involved the full commissioning and operation of a laboratory SOFC test rig over four days, during which IV and EIS datasets were acquired under controlled conditions. A custom MATLAB-based graphical user interface (GUI) was developed to automate data processing, area specific resistance (ASR) estimation, and degradation trend visualisation. Regression analyses of ASR(t) revealed the linear and non-linear ASR growth, providing experimental insight into voltage degradation behaviour.In parallel, neural network models were trained to predict system performance with a novel pre-processing workflow introduced to improve robustness. The bidirectional long short-term memory (BiLSTM) model achieved and a coefficient of determination, R² = 0.987, mean absolute percentage error, MAPE = 2.4%, and strong prediction stability across the forecast period.The proposed framework demonstrates that the convergence of simulation, experimental data, and machine learning enables accurate, low-cost prediction of SOFC degradation. It lays the foundation for future development of SOFC digital twins, real time data driven models that enable predictive maintenance, control optimisation, and lifecycle management.

Date of Award	Jul 2025
Original language	English
Awarding Institution	Aston University
Supervisor	Amirpiran Amiri (Supervisor) & Vesna Najdanovic (Supervisor)