Parallel computing enables Argonne to speed simulations

  • 20-Apr-2009 04:41 EDT
Aymeric Rousseau (shown), the Research Engineer responsible for Argonne National Laboratory’s modeling and simulations activities, said The Mathworks’ distributed computing tools have enabled the lab to run simulations and perform analysis at a much faster pace.

Because of ever-increasing time and cost constraints, automotive engineers are relying more and more on computer models to bring new technologies to production. This is particularly the case in powertrain development, where, with the advancement of conventional, hybrid electric, and plug-in hybrid electric powertrains, there are hundreds of thousands of configuration options. Adding further complexity are the many options regarding component technologies and control.

In 1995, Argonne National Laboratory, in collaboration with Chrysler, Ford, and General Motors, developed the Powertrain System Analysis Toolkit (PSAT) as an industry-standard tool to help automotive engineers analyze advanced powertrain configurations. Since making the product widely available in 2003, the use of PSAT for analysis and simulation has expanded worldwide to more than 120 companies and 600 users.

As the complexity of powertrain physics models increased, so too did the amount of time required to run simulations. In recognition of this, The Mathworks announced at the SAE World Congress that Argonne National Lab has begun using its Parallel Computing Toolbox and Matlab Distributed Server to speed the overall analysis.

“To be able to run the studies in a time-effective manner, we realized we needed to distribute the complex simulations to evaluate the vehicle performance,” said Aymeric Rousseau, the Research Engineer responsible for Argonne’s modeling and simulations activities.  “When we looked at the different options of how to distribute the computations, the selection of The Mathworks’ distributed computing toolbox was a natural choice for us because everything was in the same environment.” PSAT was developed using Simulink and Stateflow tools from The Mathworks.

Argonne used The Mathworks parallel computing tools to execute simulations on a 16-node cluster, reducing the overall run time in some cases from two weeks to one day. The Matlab Distributed Computing Server also enabled Argonne to move its simulations from the desktop to a cluster in 1.0 h.

“Using distributed computing, we went from running one set of simulations every two weeks to four sets of simulations every week,” Rousseau said. “So not only are we able to run the simulation faster, we’re able to perform analysis faster and then repeat at different steps and within a week or two do two or three runs of simulations to have better results or make changes in the code to ensure that everything is consistent for the entire hardware.”

Some of the studies being performed by the U.S. Department of Energy include more than 1500 vehicles, each of them sized to meet the vehicle technical specifications and its control parameters tuned to maximize fuel efficiency. Others, focused on development and optimization of control strategies or assumption uncertainties using MonteCarol, require hundreds of runs.

“We can look at processes and studies that before we did not consider because of time constraints or computer constraints,” Rousseau said. “So that opened a whole new horizon for us as far as what we can do from a simulation point of view.”

The Mathworks also announced at the SAE Congress the availability of its Vehicle Network Toolbox, which connects Matlab to a vehicle’s CAN, eliminating the need for additional connectivity tools. This live vehicle communication data bring the power of Matlab to automotive engineers working on simulations, test cells, and prototype vehicles, saving time and providing a more efficient workflow.

HTML for Linking to Page
Page URL
Rate It
4.56 Avg. Rating

Read More Articles On

Related Items

Technical Paper / Journal Article