Data-driven Transport Modeling for Tracer Injection and Mixing Time Screening in Stirred Tank Reactors

This repository contains a **complete research data package** for the first experimental validation of transport-based recurrence CFD (rCFD) methodology applied to liquid mixing in stirred tank reactors. The dataset includes M-Star lattice Boltzmann simulations providing velocity field databases, rCFD implementation with calibrated diffusion parameters, experimental conductivity probe measurements, systematic parametric injection location optimization (90 cases), post-processing workflows, and comprehensive validation results demonstrating 3,400× computational speedup versus ANSYS Fluent while achieving closer agreement with experimental data than conventional high-fidelity CFD methods.

**Research Context**: This dataset addresses a critical computational bottleneck in stirred tank reactor (STR) design and optimization. Conventional CFD methods (ANSYS Fluent, M-Star) require days to weeks to simulate mere seconds of mixing time, rendering parametric studies and real-time process optimization impractical for industrial applications. Recurrence CFD (rCFD) exploits pseudo-periodic flow patterns in turbulent stirred tank flows to achieve orders-of-magnitude computational acceleration while preserving predictive accuracy.

**Key Innovation**: First rigorous experimental validation of rCFD for liquid mixing, demonstrating that calibrated rCFD (operating on coarse 4.7M lattice point M-Star flow fields) predicts mixing time closer to experimental measurements (τ₉₅=24.5±1.0 s vs 24.9±2.7 s experimental) than high-resolution M-Star (29.1±1.4 s, 90M nodes) or ANSYS Fluent (29.2±2.9 s, 3.4M cells). The resulting computational efficiency enables a 90-case parametric design space exploration in 2.25 hours—a task requiring 320 days with Fluent.

This dataset contains M-Star LBM simulations (v3.12.21, D3Q19 lattice with Smagorinsky LES, Cs=0.1) executed on NVIDIA H200 GPU with 768 GB RAM, producing velocity databases at 4.7×10⁶ lattice points (coarse resolution) and 9.0×10⁷ points (high-resolution reference). The M-Star database comprises 900 frames sampled at 200 Hz (Δt=0.005 s) over t=25-29.5 s, establishing a recurrence window of τ_rec=4.5 s. The rCFD implementation requires ANSYS Fluent 2019 R3 or later with Visual Studio 2017+ (Windows) or GCC 4.8+ (Linux) for UDF compilation, operating on a 1.8×10⁶ cell FVM mesh with time step Δt_rCFD=0.025 s and calibrated face-swap diffusion parameter f_swap=0.0825, generating a ~45 GB cell-to-cell velocity database. The reactor configuration consists of a 30 L stirred tank (96 mm ID × 600 mm height) with dual impellers (PBT + RT) operating at 200 rpm in deionized water (998.2 kg/m³) under isothermal single-phase conditions. Post-processing workflows utilize Python 3.8+ (numpy, pandas, matplotlib, scipy, pyvista, opencv-python) and ParaView 5.11+ for VTK visualization. Minimum system requirements for data analysis include 16 GB RAM, 4 CPU cores, and 50 GB storage; rCFD execution requires 64 GB RAM, 32 cores, and 200 GB SSD; reproducing M-Star simulations requires NVIDIA H200 GPU equivalent, 768 GB RAM, and 500 GB storage.