Advanced Design Technology Blog

Physics-Enhanced Machine Learning for High-Temperature Heat Pump Compressor Design

Rich Evans — Thu, 03 Sep 2026 12:43:10 GMT

Physics-Enhanced Machine Learning (PEML) accelerates High-Temperature Heat Pump compressor optimization by coupling 3D Inverse Design with surrogate modelling. Using just 15 physics-based input parameters and 40 design candidates in the training database, PEML eliminated AI hallucinations, delivering multi-point efficiency and range gains with under 1.1% prediction error.

Goal: The team used Physics-Enhanced Machine Learning (PEML) for an automated multi-point, multi-objective optimization of an HTHP compressor.

Meanline Optimization: Using TURBOdesign Pre, they integrated cycle codes with meanline compressor design. With TURBOdesign Pre run in scripted batch mode, they set up and explored a design space specifically looking for the optimum range for the coefficient of performance (COP).

The Power of PEML: By using 3D Inverse Design as the generative engine, the system's inputs control the fluid physics (like blade loading), and the software solves for the corresponding 3D blade shape. This means 100% of the generated designs land near the target duty point, eliminating physically impossible outputs.

Efficiency: The entire 3D blade shape could be described using just 15 parameters. Because the data space is clean and bounded by physics, the system built a highly accurate surrogate model using fewer than 100 data points to deliver multi-point, multi-objective optimal designs - which were then verified by CFD.

High Temperature Heat Pumps as a Transformational Technology
Why Standard AI Fails in Turbomachinery
The Breakthrough: Physics-Enhanced Machine Learning (PEML)
Case Study: Optimizing an HTHP Compressor Stage

- Conclusion: High-Fidelity Results at a Fraction of the Cost

Physics-Enhanced Machine Learning for Design of a Rocket Engine Turbopump

Rich Evans — Thu, 03 Sep 2026 10:49:49 GMT

This blog describes how a Machine Learning algorithm can successfully design a high performance, single component rocket engine turbopump.

The Problem: Rocket turbopumps typically use a separate inducer blade row upstream of the main impeller to minimize cavitation. However, this separated configuration causes problematic flow interactions at off-design conditions, leading to flow separation, backflow, and drooping H-Q curves.

The Goal: The objective is to design a combined, single-component turbopump that eliminates these row interactions and reduces weight, while maintaining the same performance and cavitation margins as a separated design.

The Solution: Unlike standard Artificial Neural Networks that require massive datasets and can hallucinate physically impossible geometries, PEML uses 3D Inverse Design where the inputs explicitly control the fluid physics, such as blade loading.

The Process: The PEML system optimized the combined turbopump using only 13 parameters. By utilizing a Reactive Response Surface algorithm to build an algebraic surrogate model, the system required just 57 simulated designs (amounting to 171 CFD runs over 12.5 hours) to find the global optima.

The Results: The optimized single-component design successfully matched or exceeded the stage efficiency of the baseline separated design across multiple operating points. Furthermore, it achieved the same or better cavitation margins. This demonstrates that PEML offers highly accurate, paradigm-shifting performance gains for turbomachinery at a fraction of the computational cost.

What is Physics-Enhanced Machine Learning (PEML)?
How to use PEML to design a rocket turbopump?
Physics-Enhanced Machine Learning is the way to paradigm shifting performance gains in turbomachinery

Designing a Micro Gas Turbine for Standby Emergency Power Generation

Rich Evans — Thu, 30 Jul 2026 14:07:31 GMT

ADT and Siemens have partnered to demonstrate a rapid, blank-page-to-concept design workflow for an emergency Micro Gas Turbine (MGT).

By coupling Siemens Simcenter Amesim with TURBOdesign Suite, the joint workflow dramatically accelerates development:

System Sizing: Amesim defines the mission profile and passes requirements to TURBOdesign Pre to instantly generate matched meanline designs and performance maps.
3D Blade Design: TURBOdesign1 turns these specs into optimized, 3D compressor and turbine blade geometries in just a few hours.
Multi-Domain Co-Simulation: The matching fluid and solid models are seamlessly exported to Simcenter for automated aerodynamic, structural, and rotor-dynamic analysis.

This highly integrated, end-to-end loop slashes complex turbomachinery development timelines from months to days.

The Challenge: Emergency Standby Power for Critical Infrastructure
The Solution: A Seamless, Blank-Page Concept Workflow
System-Level Simulation & Cycle Matching in Simcenter Amesim
Matching the Radial Inflow Turbine Stage
From Meanline to 3D Geometry: Inverse Design & Co-Simulation
Key Takeaway

What is Physics-Enhanced Machine Learning (PEML)?

Rich Evans — Thu, 30 Jul 2026 13:08:43 GMT

Physics-Enhanced Machine Learning is the only way to create an expert system for the task of detailed, workable and realizable engineering design, based on CAE and simulation. Whilst at the same time restricting the upfront training effort to something manageable that does not require a super-computer and tens thousands of individual simulation datapoints.

TURBOdesign Suite 2026.1 Release

Rich Evans — Thu, 11 Jun 2026 10:31:27 GMT

The 2026.1 release advances Physics-Enhanced Machine Learning (PEML) and universal CAE integration, allowing engineers to build highly accurate optimization models with smaller datasets.

TURBOdesign Suite 2026.1 is the Large Physics Model (LPM) Accelerator.
Expanded Ansys Fluent Integration: Direct automatic integration from inside TURBOdesign1, supporting all machine types, mesh formats, boundary conditions, and fluid states.
Open Machine Learning Framework: The Reactive Response Surface (RRS) machine learning engine now connects with any external platform, including custom in-house codes and open-source packages.
HPC Scalability: Native remote execution settings allow parallelized optimization jobs to be offloaded directly to remote Linux HPC clusters.
Advanced Modeling: Features new Inlet Guide Vane (IGV) compressor models, automated axial thrust calculations, and two-phase flow functionality.
Meanline Design Exploration: Introduces N-dimensional design space sampling (Latin Hypercube) to map variables before setting detailed 3D geometry

Key New Features and Advancements
How to overcome data bottlenecks in AI and machine learning for optimization?
What is Artificial Neural Network (ANN)?
Physics Informed Neural Network (PINN)
Physics Enhanced Machine Learning (PEML)
What are the New Features in TURBOdesign Suite 2026.1?

Blade Loading, Inverse Design and Shape Parameterization

Rich Evans — Fri, 15 May 2026 11:04:37 GMT

Blade Loading Defined: Fundamentally - the pressure difference between the pressure and suction sides of a blade along a meridional pathline. It directly controls the amount of work done or extracted and reveals problematic flow regions like cavitation or shockwaves.

Spanwise Work Control: The non-dimensionalized angular momentum (rVtheta*), and its distribution from hub to shroud (‘spanwise’). It is an input parameter to the 3D Inverse design method and rooted directly in turbomachinery fluid dynamics.

Streamwise Loading Control: The other key input control in 3d Inverse Design. This defines the distribution of work from leading to trailing edge at the hub and shroud . A blade can be "fore-loaded," "aft-loaded," or "mid-loaded," (or some combination across the span) which manages surface pressure/velocity distribution and mitigates cross-passage gradients and other loss mechanisms..

The Inverse Solver: An iterative process that starts with a flow field and evolves a blade shape that will enforce it.

Shape Parameterization: Because designs are defined by aerodynamic controls rather than multiple geometry controls, the design space is low-dimensional and physics-led, making it highly compatible with Machine Learning through Reactive Response Surface optimization.

What is Blade Loading?
What is the difference between Direct Design and 3D Inverse Design?
How do we control the spanwise distribution of work?
How do we control the streamwise blade loading?
How does the 3D Inverse Design work?
How does Inverse Design enable shape parameterization and efficient optimization?
What examples of Machine Learning for turbomachinery have used 3D Inverse Design?
What is recommended further reading on 3D Inverse Design?

Machine Learning for Wastewater Pump Design

Rich Evans — Mon, 23 Mar 2026 15:30:31 GMT

This blog describes how tools from Advanced Design Technology (ADT) use a combination of 3D Inverse Design and Machine Learning (specifically "Reactive Response Surface + CAE") to rapidly design high-performance wastewater pumps. The primary challenge in wastewater pump design is balancing conflicting requirements: high hydraulic efficiency, the ability to pass large solids, and preventing cavitation. By using Machine Learning, engineers can explore vast design spaces to find optimal solutions that meet strict non-clogging constraints in a matter of hours.

Why invest in wastewater pump design optimization?
TURBOdesign Pre provides the starting point for stage design
3D Inverse Design - the enabling technology for Machine Learning
Including the volute in the performance prediction
Creating a seed design and specifications for running CFD within the ML training
Pump optimization via Machine Learning
TURBOdesign1 creates design choices from the Machine Learning solution
Re-using the training dataset to optimize the pump design for a new set of objectives
Conclusions - Machine Learning for turbomachinery design is now a reality

Maximizing Refrigeration Performance with ML & 3D Inverse Design

Rich Evans — Thu, 26 Feb 2026 15:32:25 GMT

Explore how 3D Inverse Design and Machine Learning are replacing conventional workflows to deliver a new benchmark in refrigeration cycle efficiency and aerodynamic precision..

The Challenge: Conventional methods are not sufficient to meet modern energy efficiency demands in refrigeration applications.
The Solution: Conventional methods are not sufficient to meet modern energy efficiency demands in refrigeration applications.
Rapid Cycle Setup: Engineers can quickly define single or multi-stage cycles, utilize real gas properties (like CO2 or R134a), and instantly generate performance maps and automatic 1D sizing.
3D Inverse Design: Instead of leading with blade shapes, you specify the optimum flow field, and the software computes the exact 3D geometry needed for impellers, diffusers, and volutes.
Machine Learning: A Reactive Response Surface (RRS) algorithm rapidly evaluates the design space to find the optimal multi-point, multi-objective performance in a fraction of the time.
The Result: Faster design times, breakthrough aerodynamic performance, and versatility across all types of refrigerants.

The demand for energy efficiency and sustainability for refrigeration applications has never been higher. Engineers are no longer just looking for "good enough" designs; they need a competitive edge that streamlines development while pushing the boundaries of what is possible.

This is where TURBOdesign Suite changes the game. By moving beyond traditional trial-and-error methods, it offers an integrated, physics-based approach to the entire refrigeration cycle.

Here is how TURBOdesign Suite empowers engineers to design superior components and why its 3D Inverse Design technology is now the benchmark for design capability.

Complete Control of Specifications
Complete Refrigeration Cycle Calculations
3D Inverse Design: Designing by Intent
Full-Cycle Component Control
Machine Learning for Design Optimization
Conclusion - World Leading Refrigeration Cycle Design

TURBOdesign Suite Release 2025.2

Rich Evans — Mon, 01 Dec 2025 14:50:03 GMT

The 2025.2 release is focused on speed, automation, and manufacturing readiness for turbomachinery design.

Faster Optimization: Instantly re-run the Reactive Response Surface (RRS) surrogate model for new optimization objectives/constraints without repeating costly CFD simulations.
Maximum Automation: Introduced a new Python API for scripted control over all TURBOdesign1 settings, enabling greater flexibility and integration into custom workflows.
Expanded CAE: Added support for Cadence FineTurbo and ANSYS Fluent integration, alongside new full-stage analysis workflows (Rotor + Volute).
Production-Ready: Includes a new feature to generate flank millable geometry using user-specified tool paths, ensuring manufacturability upfront.
New Applications: Enhanced concept prediction tools (TD-Pre) with new design modules for waste-water pumps and mixed-flow fans.

The new release of TURBOdesign Suite brings a set of wide-ranging and application specific improvements to enable designers to swiftly create, assess, and improve turbomachinery stages, and to run the design workflow with more automatization, efficiency and clarity.

Faster and more flexible design optimization via Machine Learning
A new Python API and scripting improvements
Full stage analysis and more CAE integration options
Enhanced Features in TD-Pre and TURBOdesign Volute

TURBOdesign Suite integrates end-to-end concept, meanline design, 3D design, volutes and other additional stage components and preparation for manufacture

What is Reactive Response Surface (RRS )?

Rich Evans — Fri, 28 Nov 2025 14:22:04 GMT

The Reactive Response Surface (RRS) optimizer is a Machine Learning method that efficiently finds optimal solutions by intelligently selecting data points for evaluation. Instead of relying on a large, pre-determined dataset, RRS iteratively learns from existing data to decide where to focus the search next, minimizing the cost and time of complex optimization problems requiring high-fidelity simulation. The core components are a probabilistic Response Surface Model (RSM), a Multi-Objective Genetic Algorithm (MOGA), and a Reactive Function that balances exploring the design space with exploiting known good areas. The final verified results show the RRS surrogate model retains the accuracy of high-fidelity simulations with close to identical results (within rounding) between both methods on key performance metrics.

● An Introduction to Reactive Response Surface (RRS)
● What is Response Surface Model (RSM)?
● What is a Multi-Objective Genetic Algorithm (MOGA)?
● What is the Reactive Function?
● How is RRS Verified?
● Key Takeaway