Correlating between atomic structure and mechanical properties in cobalt-doped silicate glasses via molecular dynamics and machine learning

In this study, the effect of cobalt (Co) doping on the atomic structure and mechanical properties of silicate-based glasses with the composition of 60SiO₂-(36–x)CaO–4P₂O₅–xCoO (mol%) in the range of x = 0 to 20 (labeled as Co0 to Co20) was investigated using molecular dynamics (MD) simulation and machine learning (ML) models. For a detailed analysis of the short-range structure, radial distribution functions (RDF), bond angle distribution (BAD), and coordination number (CN) were used. The results showed that with increasing CoO content, the Co–O bond length became shorter and the network strength increased. Meanwhile, the network connectivity (NC) value increased, indicating a strengthening of the structure's integrity. The investigation of the ion clustering parameter (R factor) showed that for Co5 and Co8, the distribution of Ca²⁺ and Co²⁺ ions was more uniform, and the structure had higher stability. The mechanical properties also showed that with increasing CoO, the Young's modulus increased continuously, and the ultimate tensile strength (UTS) varied in the range of 7.36 to 7.97 GPa. This behavior was due to the increase in network integrity and the formation of stronger Co–O bonds. In the data-driven analysis section, multiple linear regression (MLR) models were developed based on the mechanical data to predict structural parameters, including NC and Qⁿ distribution (Q⁰–Q⁴). The cross-validation (LOOCV) investigations showed that the models demonstrated promising correlations and reasonable predictive capability within the studied composition range, allowing accurate structural prediction from mechanical data. Overall, the combination of physical and data-driven methods in this study provides a new perspective for the targeted design of high-strength functional Co-doped silicate-base glasses to be utilized as biomaterials in tissue engineering applications.