Ehrenhofer, AdrianLehmann, ThomasWallmersperger, ThomasUfert, MartinJayaraman, AdhithyaAdhithyaJayaraman2024-05-212024-05-212024-02-19https://publica.fraunhofer.de/handle/publica/468556A Proton-Exchange-Membrane/ Polymer-Electrolyte-Membrane Fuel Cell (PEMFC) is an electrochemical device that converts the chemical energy of hydrogen and oxygen into electrical energy through a chemical reaction and, therefore, could play a vital role in the transition to a more sustainable and environment-friendly energy system. Fuel cell models generally consist of physics-based mathematical/analytical equations with semiempirical coefficients determined by adapting to experimental data based on various optimization algorithms. However, because the PEMFC exhibits highly non-linear dynamics, many models operate within a restricted parameter space or omit effects that can only be adequately characterized by other models. The current thesis is focused on implementing an Artificial Neural Network (ANN)/ black box model that fits synthetic data generated from two of the physics-based fuel cell models, namely the Generic MATLAB Model (GMM) and State Space Model (SSM), to predict the output response of a fuel cell under the varying operating conditions provided as input. Principal component analysis (PCA) is used to reduce the dimensions of the output voltage space along the directions of maximum covariance between the data points. The Principal Component Output Network (PCoNet) based surrogate model used in this thesis is a custom architecture based on ANN used to train on the synthetic data. In addition, techniques for meta-modeling are being explored to enhance the efficiency of black box models under different operating conditions. These techniques are based on data augmentation and transfer learning architectures. The findings indicate that Neural Networks (NNs) have the potential to be used as surrogate models. The models have shown promise in overcoming some limitations encountered during NN training with limited data.enfuel cellsfuel cell modelingneural networkmaster thesis