Madison, Wisconsin—Dec 15, 2021—Convergent Science, Argonne National Laboratory, and Parallel Works received the 2021 HPCwire Readers’ Choice Award for the Best Use of High Performance Data Analytics & Artificial Intelligence. The team was nominated for developing a scalable, automated, and adaptive machine learning-genetic algorithm (ML-GA) workflow and demonstrating its capability to significantly accelerate virtual prototyping for the optimization of an advanced heavy-duty internal combustion engine design.

Virtual prototyping allows engineers to perform a more comprehensive design optimization and save on the costs associated with building physical prototypes. However, for complex machines such as internal combustion engines, virtual design optimization can take up to several months due to the multitude of design parameters to consider and the computational expense of running many sequential simulations.

The ML-GA software technology, which also won a 2021 R&D 100 Award, was developed by scientists at Argonne National Laboratory to address these challenges. The ML-GA algorithm couples active learning, ensemble ML-driven surrogate models, and genetic algorithms with computational fluid dynamics (CFD) simulations within an end-to-end framework. By leveraging an ensemble ML technique, known as Super Learner, along with on-the-fly optimization of the ML hyperparameters, ML-GA markedly reduces the amount of simulation training data required for developing accurate ML surrogate models. In addition, ML-GA’s active learning feature intelligently selects the best possible design points to simulate during each successive design iteration. The result is that the ML-GA approach converges to the global design optimum much faster, while also lowering the number of CFD simulations required in the process.

Argonne scientists coupled the ML-GA approach with Convergent Science’s CONVERGE CFD software to perform a design optimization of a heavy-duty gasoline compression ignition (GCI) engine. The study included multiple control variables, such as fuel injection timing, fuel spray targeting, fuel injection pressure, injector geometry, and initial in-cylinder thermodynamics and flow conditions. The goal of the study was to maximize engine efficiency while adhering to emissions standards and the mechanical limits of the engine. The ML-GA approach sped up the optimization process by ten times compared to the current industry standard.

“ML-GA offers the capability to drastically shrink product design cycles and costs for industry,” said Research Scientist Pinaki Pal, who is leading the ML-GA development effort at Argonne. “If you are able to design a product in a much shorter time frame, you are also accelerating its market delivery and deployment.”

While the Argonne team demonstrated the ML-GA workflow on a GCI engine optimization problem, the approach can be used in a wide range of industries, from aerospace to manufacturing to oil and gas. To make the technology available for commercial use, ML-GA was integrated into Parallel Works’ HPC cloud platform. This integration and commercialization effort also recently won Argonne and Parallel Works the Federal Laboratory Consortium’s (FLC) Midwest Regional Award for Excellence in Technology Transfer.

“The Parallel Works team is excited about integrating the ML-GA technology from our collaboration with Argonne into our existing suite of workflow orchestration tools and hybrid HPC easy-access software. Our new Learner Works product family allows easy, scalable access for customers in government, industry, and academia to realize the cost-saving benefits of this novel machine learning technology,” said Parallel Works CEO, Michael Wilde.

The commercialized ML-GA technology within Parallel Works’ HPC platform offers exciting opportunities to CONVERGE users.

“There’s a lot of demand from our clients for optimization techniques, especially with the increasing availability and capability of computing resources,” said Dan Probst, Senior Principal Engineer at Convergent Science. “ML-GA is a really important technology because it offers such a significant speedup in the design optimization process, and it can benefit clients across a wide range of industries.”

The development and commercialization of ML-GA was funded by the U.S. Department of Energy (DOE)’s Vehicle Technologies Office (VTO) via the Technology Commercialization Fund (TCF). VTO is part of DOE’s Office of Energy Efficiency and Renewable Energy (EERE).

If you want to learn more about the automated ML-GA approach, see this International Journal of Engine Research article.

About Convergent Science

Headquartered in Madison, Wisconsin, Convergent Science is a global leader in computational fluid dynamics (CFD) software. Our mission is to enable our customers to perform revolutionary CFD simulations by creating accurate, versatile, user-friendly software and providing unparalleled support.

Our flagship product, CONVERGE, is an innovative CFD software that eliminates the grid generation bottleneck through autonomous meshing and features a suite of advanced physical models, fully coupled detailed chemistry, and the ability to easily accommodate moving geometries. CONVERGE is revolutionizing the CFD industry and shifting the paradigm toward predictive CFD.

For more information about Convergent Science please visit


Published December 15, 2021