Resource-Efficient Microstructure Optimization of Carbon-Reinforced Ceramic Composites Enabled by an RGB-Inspired Convolutional Neural Network

Enhancing the energy tolerance and thermal stability of carbon-reinforced ceramic composites (CRC) is essential for the reliable operation of high-voltage power equipment, while conventional microstructure optimization strategies rely heavily on empirical trial and error, resulting in excessive material consumption, high processing cost, and limited controllability. In this study, an RGB-inspired convolutional neural network (CNN) framework is proposed for goal-driven microstructure optimization of CRCs and related functional ceramics, with the aim of improving performance while reducing experimental redundancy. Experimentally accessible energy-dispersive X-ray spectroscopy (EDS) elemental mappings are employed as low-dimensional structural surrogates and encoded into multi-channel pseudo-RGB representations that capture the spatial distributions of carbon, ceramic framework, and pore regions. The CNN was trained on a dataset of 10,000 EDS-derived images to improve the robustness of microstructure learning. Within this representation, a convolutional self-learning mechanism iteratively refines the topology of conductive pathways under a hybrid loss function that integrates performance objectives with physical and manufacturability constraints. Rather than directly predicting processing parameters or porosity, the CNN operates within experimentally observed structural manifolds to optimize the spatial redistribution of conductive networks, thereby suppressing electro-thermal localization and enhancing structural continuity. Electro-thermal coupled simulations and pulse-energy tolerance experiments demonstrate that the optimized CRC exhibits significantly improved energy absorption capacity, thermal shock resistance, and resistance stability, with a 78% increase in permissible injected energy and a 57.1% increase in maximum tolerable temperature. By reducing unnecessary experimental iterations and material waste during microstructure optimization, this AI-guided strategy provides a resource-efficient and sustainable pathway for the design of heterogeneous functional ceramics, bridging image-based characterization with performance-oriented materials engineering while preserving physical realism.