DARTS-NetGAB
Published:

DARTS‑NetGAB (Design Automation and Real-Time Simulation using Neural Network Ensembles for Gas-Bearing Supported Turbocompressors) is an integrated computational framework that couples Panel/Bokeh interfaces, ParaturboCAD, and ensemble artificial neural networks for fast iterative modification and computation of performance for gas bearing supported turbocompressors.
Key Features
- Ensemble surrogate models of high‑fidelity data with < 5 % error on isentropic efficiency and pressure ratio (up to 11 % near choke limits)
- Real‑time simulation: ~1 s for coarse (6 195 pts) and ~8.5 s for fine (311 250 pts) performance maps
- Interactive Panel/Bokeh GUI for on‑the‑fly parameter tweaks and instant visualization
- Automated CAD generation via ParaturboCAD/CadQuery: ~7 min per STEP/STL model
Tech Stack
- Python 3.9.17
- Panel 1.3.4 & Bokeh 3.3.2
- TensorFlow 2.10 (EANNs)
- ParaturboCAD (CadQuery 2.3.1)
- pandas 2.1.4, NumPy 1.26.2, cqMore 0.1, VTK/PyVista, trimesh 4.0.5, gmsh 4.11.1
Related Publication
This framework enables engineers to rapidly explore design spaces and optimize turbocompressor performance through an integrated computational approach.