High performance concrete (HPC) has strict requirements for mix ratio accuracy and performance stability.
Traditional iterative trial mixing methods based on ACI specifications are difficult to efficiently explore high-dimensional nonlinear interaction relationships between multi-component materials.
Existing machine learning prediction models mostly remain in the offline computing stage, and generally suffer from insufficient interpretability of the “black box” and poor adaptation to the heterogeneity of mix proportion samples, making it difficult to directly support real-time mix proportion adjustment on site.
.