The multiphase corrosion structure on the surface of underwater ceramic cultural relics was successfully reproduced by the sol-gel process combined with hydrothermal mineralization. XRD analysis showed that the corrosion layer was mainly composed of pyrite (FeS2, 65±3%) and hematite (α-FeOOH). and the Nanhai No.Ⅰ shipwreck sample (FeS2, 55±5%) is attributed to enhanced metabolic conditions in the laboratory. Scanning electron microscopy verified that its layered structure (thickness of 200±30 μm) and porosity (35±5%) were highly consistent with that of the native sample (32±7%) (p=0.15). In order to further realize the accurate quantitative analysis of the corrosion layer structure, the CNN-Transformer hybrid deep learning model was introduced, which realized the automatic segmentation and identification of multiphase structures in scanning electron microscopy (SEM) images (accuracy>94%). Intelligent phase analysis was performed on X-ray diffraction (XRD) sequence data. The analysis results significantly improve the objectivity and accuracy of structural characterization, and provide data support for confirming the key role of the electron transfer mechanism at the interface. By optimizing the sol-gel process, the deviation of the base chemical composition can be controlled to ±0.4wt% (validated by ICP-OES). The two-step sintering process (400℃+800℃) retains 6.8vol% amorphous phase and reproduces the characteristic interface of ancient glazes. The study of microbial mineralization kinetics shows that SRB/FeOB synergistically drives the Fe2+/Fe3+ cycle, forming a structure of coexistence of pyrite and goethite. Accelerated aging experiments have proved that the replicate layer has excellent stability and its chemical behavior is comparable to that of real cultural relics.
Tao et al. (2026) studied this question.