Conditional GAN for Generation of 3D SOFC Electrode Microstructure Dataset

The conditional generative adversarial network (Conditional GAN) model is developed to generate synthetic porous microstructures of solid oxide fuel cell anode. To control the volume fractions of the synthetic anode microstructures, volume fraction loss is defined and used for training the generator in addition to the training using the adversarial loss. The synthetic microstructures are generated with a specified volume fraction of the phases and compared with those from the real anode microstructures in terms of microstructural parameters. It is found that the volume fractions of the synthetic microstructures are successfully controlled by the proposed additional training procedure for the generator, and the microstructural parameters, such as surface area density and TPB density, are close to those from the real microstructures.