Prediction of Crack Propagation Behavior in Silicon Carbide Ceramics Using Artificial Neural Networks

ABSTRACT In this study, a combined approach of artificial neural networks and finite element simulation is employed to investigate crack propagation behavior of silicon carbide ceramics. The research objects include three‐point bending specimens featuring prefabricated cracks and Vickers indentation specimens. Firstly, the extended finite element method and the cohesive zone model are utilized to determine the crack propagation paths of different specimens and complete the model validation through corresponding experiments. Subsequently, the image recognition technology is applied to accurately extract the crack path data, thereby constructing a high‐quality training dataset. On this basis, two feedforward neural network models are established for the two types of specimens, respectively, followed by training, validation, and prediction analysis. The results demonstrate that the established neural network models achieve favorable approximation accuracy for finite‐element–simulated crack paths within the training parameter range. For the crack path prediction of three‐point bending specimens and Vickers indentation specimens, the root mean square errors are no more than 0.0484 mm and 0.2698 μm, respectively. This study provides a reliable data‐driven solution for the crack propagation analysis of silicon carbide ceramics.