NanoLCA-DT

Multi-Objective Optimization Digital Twin for
Sustainable Nano-Modified Construction Materials
Developed by
Dr. Ali Keyvanfar & Dr. Arezou Shafaghat
A² Learning Studio · Kennesaw State University

NanoLCA-DT is a conceptual prototype demonstrating a proposed research framework for AI-driven multi-objective optimization of nano-modified construction materials. The platform integrates ML surrogate models, cradle-to-gate lifecycle assessment (ISO 14040), and NSGA-II Pareto optimization across three objectives: lifecycle cost, embodied carbon, and durability. It further envisions deployment as open, callable infrastructure through Skills, Model Context Protocol (MCP) servers, and REST APIs — enabling any AI agent to perform sustainable materials decision-making.

This mock demonstration was created to communicate the research vision for candidacy in the Cornell University AI4S Visiting Professorship program (CAISI, 2026–2028). Not every function or computation in this prototype is fully operational — the platform is designed to illustrate the scope, architecture, and user experience of the proposed 2-year collaboration, not to serve as a production research tool. Surrogate model predictions shown here are based on simplified physics-informed approximations calibrated to published literature, not on trained ML models.

Voice narration is provided via ElevenLabs TTS. The platform requires no installation, login, or external dependencies — it runs entirely in the browser as a single HTML file.

🔊
OverviewReady

NanoLCA-DT

Multi-Objective Optimization Digital Twin

AI-Driven Multi-Objective Optimization for Sustainable Nano-Modified Construction Materials

A computational decision-support framework integrating ML surrogate models, cradle-to-gate lifecycle assessment, and Pareto-optimal mix design — deployed as interoperable Skills, APIs, and MCP servers for agentic AI platforms.

3
Objectives
LCA
Integrated
MCP
Ready
N=5000
Monte Carlo

Pareto Optimizer

Tri-objective optimization across Cost, Embodied Carbon, and Durability. Explore the Pareto frontier — every point is a viable mix design. Your current design is plotted as a red diamond.

🧬

Grey-Box Digital Twin

Hybrid mechanistic + data-driven architecture. ML surrogates trained on published experimental data, constrained by physics-based cement hydration models.

🔌

Skills / MCP / API

Infrastructure for agentic AI. Any LLM can call optimize_mix(constraints) and receive Pareto-optimal, LCA-validated mix designs.

Research Framing Context

The Problem: Cement production accounts for ~8% of global CO₂ emissions. Nano-additives can improve durability and reduce cement content — but optimal mix design requires navigating a high-dimensional tradeoff space across cost, carbon, and performance.

The Gap: Materials informatics has transformed battery and photovoltaic discovery — but construction-scale cementitious materials remain outside the AI-for-materials pipeline.

The Contribution: A grey-box digital twin that trains ML surrogates on nano-cementitious data, integrates LCA (ISO 14040), solves tri-objective optimization via NSGA-II, and deploys everything as callable infrastructure (Skills, MCPs, APIs).

Target Venues: Nature Sustainability, Nature Computational Science, Science Advances.

Proposed by Dr. Ali Keyvanfar (KSU) for Cornell AI4S Visiting Professorship · A² Learning Studio · 2026–2028
Mix Design Space Input ?
Lifecycle Cost
$/m³
Embodied CO₂
kgCO₂eq/m³
Durability Index
0–100
28d f'c
MPa
Cl⁻ Diff.
×10⁻¹² m²/s
Carbonation
mm/√yr
Self-Heal
% recovery
Pareto Frontier — Cost vs CO₂ (bubble size = Durability) Run optimization to visualize
Cradle-to-Gate LCA Breakdown ISO 14040 ?
ComponentMass (kg/m³)EF (kgCO₂/kg)Contribution

🤖 Decision Intelligence

Adjust the mix parameters on the left panel, then click "Run NSGA-II Optimization" to generate the Pareto frontier. Your current design will appear as a red diamond — the AI will analyze whether you're on the efficient frontier or if better tradeoffs exist.

Grey-Box Digital Twin Architecture Framework

Hybrid mechanistic + data-driven architecture. Physics-based models (Abrams' law, pozzolanic reaction kinetics, Fick's diffusion) provide constraints; ML surrogates (trained on 2,400+ published data points) provide fast, accurate predictions. LCA is integrated at every evaluation step.

📥

Input Layer

Mix composition + environmental conditions + service life target

🧠

ML Surrogate

Predicts f'c, permeability, porosity, carbonation, crack healing from composition

🌍

LCA Engine

Cradle-to-gate per ISO 14040. Embodied carbon + energy + water

📊

NSGA-II

Tri-objective: min Cost, min CO₂, max Durability

↓ ↓ ↓ ↓
🔌

MCP Server

Any AI agent calls optimize_mix(constraints) via tool use

📋

Claude Skill

Structured workflow with branching logic for mix optimization

🌐

REST API

POST /api/v1/optimize → Pareto front + LCA report

↓ ↓ ↓
👩‍🔬

Researcher Interface

Interactive exploration of Pareto frontier, drill into mix compositions, compare LCA scenarios

🏭

Industry Decision-Maker

"Given my budget, local materials, and carbon target — what's my optimal mix?"

How the Grey-Box Model Works Technical

Mechanistic Component (White Box)

Abrams' Law: f'c ∝ (1/w:b)^k — strength–water relationship
Pozzolanic Kinetics: SiO₂ + Ca(OH)₂ → C-S-H gel — nano-silica reaction
Fick's 2nd Law: ∂C/∂t = D·∂²C/∂x² — chloride diffusion
Carbonation: x = K·√t — carbonation depth progression
These physics models constrain the search space and provide interpretability.

Data-Driven Component (Black Box)

ML Surrogates: XGBoost + Gaussian Process regression trained on published experimental data from nano-SiO₂ cement studies
Feature Space: Particle size, concentration, SSA, w/b, SCM type/%, cure age, exposure class
Targets: f'c, Cl⁻ diffusivity, porosity, carbonation rate, crack healing %
Fast inference (~1ms) enables population-based optimization over thousands of candidates.

Agentic Infrastructure Concept Future

Just as Claude uses Skills and MCPs to access calendars, databases, and code execution — materials researchers need callable, composable computational tools. This project creates the first MCP infrastructure for sustainable construction material optimization.

📋 Claude Skill YAML
# SKILL.md — NanoLCA Optimizer name: nanolca-mix-optimizer triggers: "optimize concrete mix" "sustainable mix design" "low-carbon concrete" # Workflow: # 1. Collect constraints # 2. Run NSGA-II optimization # 3. Present Pareto frontier # 4. User selects design # 5. Generate LCA report
🔌 MCP Server Protocol
// MCP Tool Definition { "name": "optimize_mix", "parameters": { "max_cost": number, "max_co2": number, "min_strength": number, "service_life": number, "nano_additive": enum } }
🌐 REST API Endpoint
POST /api/v1/optimize // Request { "constraints": { "strength": ">=40 MPa", "co2": "<=350 kg", "budget": "<=180 $" } } // Returns: Pareto front + LCA
Why This Matters Impact

Today: A researcher wanting to optimize a nano-modified concrete mix must manually search literature, build custom models, run separate LCA calculations, and iterate by hand. This takes weeks and is non-reproducible.

With NanoLCA-DT: A researcher's AI agent calls optimize_mix(strength≥40, co2≤350) and receives Pareto-optimal, LCA-validated mix designs in seconds.

Session Data Export Research

Export complete session data including all mix configurations, optimization results, and comparison data for research validation and reproducibility.

Comparison Table Results
📊

Add mix designs from the Optimizer tab using the "+ Add Current to Comparison" button.