Research Motivation
Cement manufacturing accounts for approximately 8% of global CO₂ emissions. Designing sustainable concrete requires civil engineers to replace conservative, over-dimensioned formulas with optimized binder blends.
AstraMix AI couples XGBoost prediction models with SciPy solvers to explore how algorithms can support material science decisions, balancing mechanical performance (compressive strength) against cost and environmental impacts.
Scientific Stack & Core Architecture
The platform is constructed on a modern decoupled stack:
- Inference Pipeline: Pre-trained XGBoost Regressors are validated against standard linear and decision-tree architectures to ensure the highest predictive accuracy (R² = 0.917).
- SLSQP Optimization Solver: Bounded optimization searches are formulated using SciPy's Sequential Least Squares Programming. Constraints are locked to physical parameters (e.g. maximum water-cement ratio limits).
- Sustainability Metrics: Carbon and cost estimation models rely on standardized emission factors (kg CO₂ / kg material) and raw materials market pricing indices.
Creator & Founder
AstraMix AI was founded and developed by Nikolay Khachatryan as an academic and research portfolio project to demonstrate the intersection of data science, machine learning, and civil engineering material design.