Self-Driving Cars Safer by Analysis

Vehicles kill 1.25 million a year worldwide, says the World Health Organization. Do the math to see that’s someone being killed every two minutes. But for some us—engineers who are trying to design vehicles that drive themselves—the time and distance between vehicle-related deaths is too much.

There's far too many miles between accidents to suit the testing of self-driving vehicles. The distance that must be logged before the public and lawmakers are assured that the cars are safe enough is estimated to be 8.8 billion miles, according to Rand in the recent study “Driving to Safety.” That's 400 years with a fleet of a hundred cars driving 25 mph around the clock every single day.

Google, the leader in road and track testing for its self-driving vehicles, has managed to log a mere 2 million miles in a purely autonomous mode, a period of six years in which its vehicles were involved in 14 collisions. That's not our fault, said Google.

Ok, Google, that's great. But at this rate, it will take over 25,000 years to log 8.8 billion miles.

Clearly, we don't have that kind of time. Sandeep Sovani, director of the automotive industry for ANSYS, offers simulation as way to speed things up. Most of those billions of miles can be simulated, said Sovani.

Sovani, who has a PhD in mechanical engineering from Purdue, lists himself as working in the intersection of the automotive industry and simulation. He has studied electric vehicles as well as explored our relationships with cars from a philosopher's and futurist's perspective. For the last few years, Sovani has turned his attention to self-driving vehicles.

“It's a closed-loop control system, but a complicated one,” said Sovani. “In fact, it may be the most complicated control system in the world.”

If you thought airplanes and spacecraft would be more complicated, you are wrong. There's relatively few threats in deep space. Even nightmare situations for fighter pilots are more of a test of reaction time than a complicated threat analysis. Bogeys at 6 o'clock will pale in comparison to a parking lot in a shopping mall on Black Friday.

Though a self driving car has only about a dozen sensors (radar, cameras, LiDAR, etc.), the combinations of stimuli and possible situations an automobile encounters is mind boggling.

Systems that have to simulate the various vehicle environments have to have virtual trees, curbs, even people. They'd have to simulate conditions, some merciless, that would test the limits of professional race car drivers. When do race car drivers encounter black ice, for example?

ANSYS Tools

Well known for mechanical simulation in the automotive industry, ANSYS first solved for stress, strain and fatigue in car bodies and parts. ANSYS started solving for how cars move through air after acquiring Fluent in 2006 for $398 million. However, simulating for how cars moved in traffic involves mastery of another discipline: electromagnetics.

Sensing your proximity to other objects in a moving vehicle requires bouncing electromagnetic waves off them. It's what your eye-brain combination does with electromagnetic waves in the visible spectrum. It's what radar, LiDAR and computers do for the rest of the spectrum.

Electromagnetics by Acquisition

ANSYS acquired Ansoft for $832 million in 2008. Almost a neighbor (both companies are in or near Pittsburgh), Ansoft's main products, Maxwell and Simplorer, instantly made ANSYS an electronics and electromagnetics simulation vendor. Little did both companies know at that time how important this acquisition would be in positioning ANSYS for its role in self-driving vehicles.

Additional acquisitions have added to the ANSYS toolset for self-driving vehicle design and analysis. Esterel Technologies was acquired for its SCADE software in 2012. ANSYS SCADE generates control software following a model-based system engineering(MBSE) solution. SCADE Suite, SCADE Display, SCADE Test and SCADE LifeCycle will help further a design and fine-tune the embedded systems that have turned the modern automobile into a rolling computer system.

Delcross Technology was acquired in 2015 for technology that was to become ANSYS HFSS SBR+ (shooting and bouncing ray), which solves for 3D antenna and radar radiations

For the radar sensors that detect positions of other nearby vehicles, ANSYS offers ANSYS HFSS, which stands for high-frequency electronic field simulation. HFSS, along with HBR+, will predict in 3D the strength and shape of electromagnetic waves that result from direct transmission and bounces of nearby objects to make sure car-mounted sensors have adequate coverage and no blind spots.

ANSYS Simplorer software allows model-based design in various industries such as industrial machinery, aerospace, electronics—and automotive. It can serve as a platform for vehicle simulation, said Sovani. To simulate self-driving vehicles, an engineer would use Simplorer to do the MBSE and then use SCADE to write the code for the control software that is found on a vehicle.

Why ANSYS?

A number of robust programs exist to simulate self-driving cars, including TASS International's PreScan and Mechanical Simulation's CarSim. It’s a rapidly growing space, concedes Sovani. What does ANSYS bring to the party?

Sovani suggests that ANSYS, with its strength in simulation—finite element analysis (FEA) and computational fluid dynamics, in particular—is well positioned for analysis of other types. Acquisitions have led to ANSYS now having a full toolbox for electromagnetic analysis. As automobiles become increasingly electronic, electronics simulation also becomes increasingly important.

While other companies are providing self-driving vehicle simulation software that relies on simple analytical or empirical models, ANSYS offers its strength in rigorous computational analysis of the building blocks: the sensors in the self-driving vehicles.

We can accurately simulate and predict the performance of sensors, says Sovani. For this, we use FEA.

Self-Driving Vehicles Get Started

The race for the self-driving car is not a new one. The U.S. government agency, Defense Advanced Research Projects Agency, set up a million-dollar challenge for autonomous vehicles to cross a stretch of the Mojave Desert in 2004. Vaunted teams from Caltech, Carnegie Mellon University and other institutions literally crashed and burned, not one even completing the first eight miles of the 110-mile course—what Popular Science labelled a debacle. But in 2006, Stanford’s tricked-out VW Touareg (named Stanley) crossed the finish line of a tortuous 132-mile desert course to win a $2M grand prize.

Since then, it's been a frenzy of activity. Members of the Stanford team were hired by Google. The “search giant turned car maker” was an example to computer manufacturers (Apple) and even chip makers (NVIDIA) to get into the race. Suddenly, it became apparent that the vehicles we had grown up with were less about traditional automotive technology of pistons and compression ratios than they were about computers and software. Why not let the computer and software companies have a go at it? What had Detroit even done lately? All the action was taking place in Silicon Valley and Boston's tech corridors and campuses, and more and more, the focus of an industry was turning to a patch of land in western Pennsylvania at the north end of the Allegheny Mountains.

ANSYS' relationship with research institutions is further indication of ANSYS keeping an eye on the road ahead. A geographic stroke of luck put ANSYS in Canonsburg, Penn., close to Carnegie Mellon University in Pittsburgh. The university is a hotbed of self-driving vehicle research, and the relationship with ANSYS has been a close one. ANSYS is to fund the building and construction of a three-story, 30,000-square-foot research center with a maker space on the ground floor. Its stated goal is to have ANSYS engineers educate undergraduate students in the use of simulation, a field of study that universities typically reserve for graduate students and faculty. The building is scheduled to open its doors in 2018.