AWS for Industries

University of Michigan student team develops an energy efficient solar car with High Performance Computing (HPC) on AWS

Driving a car powered only by light rays from the sun, across a 3000 km wide continent, in the shortest possible time, is no easy task. The University of Michigan’s solar car team is heading off to do exactly that, with help from High Performance Computing (HPC) on Amazon Web Services (AWS). The student team has developed the Astrum solar powered vehicle from scratch all the way from first concepts to parts manufacturing, assembling and vehicle testing. The team is racing Astrum across the Australian outback to compete in the 2023 Bridgestone World Solar Challenge on October 22, 2023.

Pictured University of Michigan Solar Car Team’s Astrum race carPictured: University of Michigan Solar Car Team’s Astrum race car.

Understanding the team’s unique challenges

The Bridgestone World Solar Challenge is a 3,000 kilometer (1,800 miles) solar car race across the Australian outback. On October 22, 2023, over 50 teams travel from Darwin at the northern coast of the continent, to Adelaide in the south, across the desolate outback. The event attracts hundreds of students with the goal of designing and building a car that is solely powered by the sun, and putting it to the test in one of the most challenging environments on Earth. What makes the event truly stand out is that it brings students together to develop new and innovative solutions to tackle the challenges of climate change.

The University of Michigan Solar Car Team – a completely student run organization with over a hundred students and a 33-year legacy – has been competing in solar car races since 1990. Each season, the team designs and builds a brand-new solar car right from concept and design, to manufacturing, assembly, and testing. Over the 33 years that the team has competed in solar car races, they have built 17 entirely solar powered vehicles, the latest one being 2023’s Astrum. In addition to building the vehicle, the team also has to develop a winning race strategy by optimizing various parameters within the limits of the race rules.

Solar car racing is a complex sport, where 35-hour-long challenges are won and lost by performance differences of less than five minutes. Every minor detail that makes a difference in aerodynamics and energy consumption must be optimized. Astrum consumes about half the power of a typical toaster oven and has the drag coefficient similar to that of a typical passenger car’s side-view mirror. It is quite literally a game of milliwatts.

To position itself for success, the team had to tackle two crucial tasks. First, was reducing the aerodynamic drag on the vehicle by optimizing its shape in order to achieve the fastest possible driving speed. Secondly, the team had to make strategic decisions about race tactics in order to gain the most advantage by optimizing a large set of parameters.

Perfecting vehicle technology with High Performance Computing (HPC) on AWS

The University of Michigan Solar Car Team needed to perform thousands of simulations to optimize the vehicle’s aerodynamics and to evaluate thousands of race scenarios. They used HPC on AWS for these extensive tasks.

Vehicle aerodynamics optimization

The team’s aerodynamics group used Siemens Simcenter StarCCM+ software to run Computational Fluid Dynamics simulations of airflow around the car with computational meshes containing 50 to 200 million cells. Because the team needed to run hundreds of simulations to try out various car shape configurations in order to find ones with the lowest drag, they setup an automated workflow on AWS starting with a web app built with AWS Amplify. Whenever a simulation has to be run, the teams’ aerodynamicists log into the app and upload the input geometry which gets stored on Amazon Simple Storage Service (Amazon S3). In the backend, the team has connected the app to AWS ParallelCluster that creates and operates a HPC cluster comprised of Amazon Elastic Compute Cloud (Amazon EC2) hpc6a.48xlarge instances and Amazon FSx for Lustre scratch file system. The Slurm job scheduler is used to manage job submission queues. Whenever simulation jobs are submitted, AWS ParallelCluster spins up compute instances with AWS Auto Scaling groups, to run the jobs. The results of the simulations are stored in Amazon S3, with metadata extracted using AWS Lambda and stored in MongoDB. The team’s aerodynamicists are able to review the results in the web app and make decisions for changing the vehicle’s shape to reduce its aerodynamics drag.

Ibrahim Syed, the University of Michigan Solar Car Team’s member who setup the aerodynamics simulation workflow on AWS, says, “Prior to using AWS it took our engineers 8 to 10 hours to run each aerodynamics simulation on the computers that we had access to on-premises. On AWS, we can scale up the simulations to run in parallel on 10 to 16 Amazon EC2 hpc6a.48xlarge compute nodes, comprising of a total of 960 to 1536 physical compute cores, and complete the simulations in just 1 to 2 hours. This has enabled our team to run over 1000 aerodynamics simulations, whereas on-premises we would have been able to run only a few dozens. With AWS we were able to explore the design space much more extensively to find the most optimum shape for the vehicle’s nose cone. As a result, we were able to reduce the aerodynamic drag on our vehicle by over 30%. Our team’s previous generation race vehicle, named Electrum, had a 2.5 m2 solar panel array, but the latest generation vehicle, Astrum, has a much bigger 4 m2 array. Even though the array size has roughly doubled, we have been able to reduce the frontal area of the vehicle, which is a key contributing factor to overall drag force, thanks to the aerodynamic improvements that we have made.”

Picture More aerodynamic nose-cone for the year-2023 Astrum solar car (left) compared to the year-2019 Electrum car (right and side)

Picture: More aerodynamic nose-cone for the year-2023 Astrum solar car (left) compared to the year-2019 Electrum car (right).

Regarding the process of getting started with AWS, Syed adds, “It was super easy to setup the entire workflow on AWS. The documentation was clear, straight-forward, and intuitive. To setup the full aerodynamics simulation automation pipeline on AWS including the web app, AWS ParallelCluster, storage, database, and AWS Lambda functions, it took us the same amount of time as would have been needed for running 2 or 3 simulations. The return-on-investment that we got from setting up the aerodynamics simulation process on AWS was huge – 2 to 3 orders of magnitude greater than the effort we invested.”

Vehicle dynamics and strategy simulations

After setting up and successfully using the workflow for running aerodynamics simulations on AWS, the University of Michigan Solar Car Team also started running vehicle dynamics simulations on AWS. These simulations are done with the software CarSim from Applied Intuition. Each vehicle dynamics simulation is run on a single Amazon EC2 instance with GUI operated via remote desktop. The vehicle dynamics simulations are helping the team improve the static and dynamic stability of the vehicle, accounting for variables such as wind and weather. Syed says, “The team is now in Australia making final preparations for this week’s race. Because of the flexibility of using AWS, the team is able to run vehicle dynamics simulations live from the Australian outback to finetune vehicle parameters based on real-life ground conditions and last-minute design changes. We were not able to do this before we started using AWS.”

The team is also using AWS to run race strategy simulations with a C++ based strategy simulator that the team has developed over the past few years. The simulator considers factors such as weather, physical forces on the car, and shading of solar panels, in order to decide the best tactics for running the race in the shortest possible time. Before using AWS, the team manually ran these simulations on their personal computers, a process that took several days to complete. On AWS, the team is now using the same workflow setup for aerodynamics simulations, to run strategy simulations. Each simulation now takes only about 60 seconds to run and the team has run over 100,000 simulations to date.

Watch this video to learn more about the University of Michigan Solar Car Team.

Read related stories from AWS:

Dr. Sandeep Sovani

Dr. Sandeep Sovani

Dr. Sandeep Sovani a principal go-to-market specialist with Amazon Web Services (AWS). He works with worldwide customers helping them to run their engineering simulation workloads on the cloud with scalable and secure high performance computing solutions from AWS. In his spare time, he loves to travel, hike, photograph, and spend time with family.

Min Kwon

Min Kwon

Min Kwon is a senior undergraduate student majoring in Economics and minoring in Translation Studies and Creative Writing at the University of Michigan. She is a Marketing Specialist for the University of Michigan Solar Car Team, specializing in videography. In her free time, she likes to read, translate, cook, and run.