With the newest release of NETL’s carbonaceous chemistry for computational modeling, or C3M, software, researchers have leveraged machine learning approaches to overcome one of the biggest drains to computational resources when modeling advanced energy systems. Version 19.1 of C3M introduces the Machine Learning Accelerated Stabilized Explicit Variable Load (MLA-STEV) software that solves complex chemical reaction equations much faster than previous iterations, drastically shortening design time and significantly reducing research and development costs.

“The MLA-STEV solver could be used to help accelerate the design of cleaner and more efficient energy systems like gasifiers,” said Dirk VanEssendelft, Ph.D., referring to an energy technology that converts organic material such a coal into useful fuels and chemicals.

VanEssendelft works on NETL’s multiphase flow science (MFS) team, which investigates multiphase flow of different phases of matter in energy systems (e.g., coal ash and hot gases in a gasifier) using powerful computational fluid dynamics (CFD) software like the Lab’s Multiphase Flow with Interphase eXchanges (MFiX).

Reactors based on multiphase flow are difficult to design, scale up and operate. Physical experiments to guide this process are increasingly expensive and sometimes impossible to perform. So, a high-fidelity computational tool like MFiX is crucial to accurately simulate these systems. However, important data concerning chemical reactions must also be integrated into the model.

“That’s where C3M and the MLA-STEV solver come in,” said Terry Jordan, who also works on the MFS team. “C3M acts as a platform to generate the chemical reaction models. Solving the chemistry in CFD models accounts for a large portion of the total computational power used for CFD modeling. With the MLA-STEV solver, that computational need is reduced to nearly zero, effectively cutting CFD modeling time in half in many cases.”

The MLA-STEV solver achieves its accelerated capabilities through a merger of MFiX with Google’s TensorFlow, a machine learning framework. TensorFlow allows solvers like MLA-STEV to run on multiple operating devices, including graphical processing units, which allow for many more equations to be solved in parallel.

“Prior to developing the MLA-STEV solver, we knew which chemical equations we needed to consider,” said VanEssendelft. “But, solving them was incredibly time intensive and computationally expensive. Now, with the latest release of C3M, researchers can run the new solver on a modified MFiX build to model reactive multiphase flow systems in record time. Faster modeling means that we can help bring these more efficient and environmentally benign energy technologies online much sooner.”

While this successful research has greatly increased the speed of solving chemical reactions, the team hopes to accelerate CFD models even further by apply artificial intelligence and machine learning techniques to also solve the fluid portion of the modeling effort.