ADT will unveil its revolutionary framework for Physics-Enhanced Machine Learning (PEML), directly addressing the industry’s critical data bottleneck in the drive to more automated turbomachinery design.
Overcoming the AI Data Bottleneck via 3D Inverse Design
Traditional data-driven machine learning models require massive datasets, often spanning thousands of costly, high-fidelity Computational Fluid Dynamics (CFD) simulation runs, simply to "learn" basic physical laws and fluid constraints. This computational penalty is severely exacerbated by conventional CAD-based geometric parameters, which alter surfaces blindly and yield non-physical or unmanufacturable results.
ADT’s TURBOdesign Suite resolves this bottleneck by deploying 3D Inverse Design as a foundational, physics-guaranteed filter for AI training. Rather than adjusting raw geometric coordinates, the framework uses aerodynamic loading parameters and circulation distributions to directly compute the 3D blade shapes. This methodology yields fundamental advantages:
- 20x Dimensionality Reduction: The number of parameters required to describe a fully 3D turbomachinery blade is up to 20 times less than when using traditional geometry-based direct design methods.
- Guaranteed Physical Consistency: Because the inverse design formulation inherently matches the specified work input and mass flow rate, the search space is limited entirely to valid, high-performing designs from the very first iteration.
- Lean-Data Machine Learning: By eliminating non-physical designs from the matrix, ADT’s proprietary Reactive Response Surface (RRS) optimizer builds exceptionally high-accuracy surrogate models using fewer than 100 high-fidelity CFD training samples, rather than thousands. Hence reducing the training time by two orders of magnitude.
Universal CAE and Multidisciplinary Integration
To further accelerate industrial deployment, the latest release introduces universal integration with leading commercial CAE suites, enabling automated synthetic data generation factories. TURBOdesign1 features direct, automated coupling for all turbomachinery applications into Ansys Fluent, alongside established workflows for Ansys CFX, Siemens Simcenter STAR-CCM+, and Cadence Fidelity/Fine Turbo. The system seamlessly handles meshing orchestration, execution, and automatic extraction of training data maps back into the design view.
Technical Paper Presentation
ADT will present a cutting-edge peer-reviewed paper demonstrating the real-world application of these algorithms to cryogenic aerospace propulsion systems:
- Session: 36-08
- Paper ID: GT2026-177081
- Title: Machine Learning based Optimization of a LH2 Turbopump with Combined Inducer-Impeller Configuration
- Presentation Schedule: Thursday, June 18, 8:00 AM – 10:00 AM
- Authors: Mohammad Mahdi Ghorani, Melvin Joseph & Prof. Mehrdad Zangeneh
Complete the form to:
✅ Schedule a Meeting: Secure a time to speak with Prof. Zangeneh or Lorenzo Bossi at the event.
✅ Get the Research: Receive a digital copy of the paper after the conference concludes.
✅ See TURBOdesign Suite's PEML Workflow in Action: Learn how to overcome AI data bottleneck with Physics-Enhanced 3D Inverse Design.
