The automotive world is becoming much more dependent on modeling than ever before. Automakers are now using more aluminum parts; vehicles have complex computers on board, often working with each other; and there's an increasing need to make cars safer in the age of the distracted driver.
Without extensive modeling and prototyping on a computer, there's no way to design cars that can keep pace with the advancements in the automotive industry and keep us safe.
Take the 2016 Cadillac CT6, a newly designed luxury sedan that starts at $53,495 and recently hit dealerships in the U.S. To keep costs low and hit production schedules, Cadillac engineers say they ran more than 200,000 simulations and spent 50 million compute hours testing the car.
One reason for all the compute time is that the CT6 is the first sedan they've designed with mostly aluminum parts. Only with computer modeling were they able to see how the steel, aluminum, plastic and other parts would co-exist.
Modeling for what automakers call the "load paths" -- essentially, the force of impact and energy on a car -- is critically important as a way to test the materials used in the car.
Fifty million hours of compute time might seem excessive, but the cost to build just one physical prototype (or "mule") runs into six figures, explains Dave Sullivan, an automotive analyst with AutoPacific. It can cost an automaker 10 times as much to make a one-off prototype as it will eventually cost to make that same car on the assembly line, he adds. (Cadillac declined to give any specific figures about costs.)
"Cadillac has an extensive mixed-metal use in the CT6 that would have been too costly to tool up without computer simulations. Being able to perform structural tests in a virtual environment allowed Cadillac to design every rib in the aluminum castings used throughout the car," Sullivan says, adding that automotive simulations will become even more common this year for most car makers.
Mark Boyadjis, an IHS automotive analyst, concurs. "A vehicle's core differentiators are now more dependent upon software engineering than they are on mechanical engineering, so it's not surprising that vehicle design is becoming more dependent upon software as well," he says. And in light of how some automakers have struggled to even stay in business, Boyadjis adds, the cost reductions that come with computer modeling over physical prototyping are key.
A learning process
No other car from General Motors (Cadillac's parent company) has relied as much on computer modeling as the CT6, says Lyndon Lie, chief engineer for the CT6. The designers wanted to hit on three major sweet spots: zippy performance, exceptional safety and a high miles-per-gallon rating (which is mostly related to vehicle weight and aerodynamics). The door handles and roof antenna had to arc a certain way; the front air vents had to allow air to flow in a way that increased fuel economy.
The CT6 is 63% aluminum, which is much higher than in previous sedans, Lie says. Cadillac started by making sure the occupants are safe, so the main frame around the seats is made from reinforced steel. Everything else was up in the air.
Engineers used modeling to reduce the aluminum and metal parts in the front body door pillar down to only two from the 25 used in the Cadillac ATS.
One example: The CT6 side mirrors, which were originally intended to be remarkably similar to those of the Cadillac CTS and ATS models, caused a slight drag in a crosswind model. Because of those results, the team adjusted the size so the handles don't stick out quite as much, long before the car reached the prototype stage. And on the safety side, as you can well imagine, the engineers created many simulations for crashes before they ever drove a car into a cement wall with a dummy inside.
Of course, as with any major change in an industry, there were lessons to learn in the CT6 modeling process. Although the simulations helped reduce costs and speed up production schedules, Lie says, early prototypes of the CT6 revealed a few interesting shortcomings in the software models.
One had to do with how the robots would assemble the final car in the factory. There are 250 meters (273 yards) of structural adhesive inside the car. Slight variations in how the robots adhere the adhesive, which weren't anticipated in the simulations, meant Cadillac had to retool a few minor parts.
Another example had to do with the wiper blades. The simulation told engineers that the blades were positioned perfectly, but in a physical model in a wind tunnel, the blades emitted an annoying whispering soda-bottle sound. The problem was that the model didn't account for how wind would move up over the hood and hit the blades at a certain angle.
The crash-test simulations weren't perfect either. "We found the car hit a barrier in certain places in the real crash test that we did not see in the simulation," says Lie.
More importantly, there are two big areas where simulations fall short. One is in explaining the design and engineering of the car to partners and managers. Cadillac used augmented reality to show software simulations in a darkened room, giving employees a chance to open a virtual door or check a virtual rear-view mirror. Although the engineers were confident that the final product would match the software models, says Lie, they realized that people still needed to see and touch the design in a way that a simulation can't replace.
The second area is in the general "vibe" or personal sense the car provides. Every car has an overall "feel" due to the suspension, steering, turning and handling that you can realize only by driving the physical vehicle on a real road. "For now, you still have to get into a prototype and just drive it," says Lie.
Present and future
Despite the shortfalls, computer modeling is helping in other ways as well, says Sullivan. When an automaker makes a more accurate software model of a part, the supplier can provide a more accurate quote. Travel costs are lower, because the more software automakers use for design, the less often they have to meet suppliers in person.
And modeling is important as safety requirements change -- it makes automakers more responsive and nimble, and might even reduce the number of recalls. "Computers have helped drive accountability and improve communication, quality and efficiency at the manufacturing level," says AutoPacific's Sullivan.
For his part, IHS' Boyadjis predicts that software modeling will become even more prevalent as cars increasingly rely on information technology. Today's automatic braking and self-parking will lead to a not-too-distant future in which road infrastructure communicates directly with cars -- for instance, when a stop sign causes a car to brake -- and cars drive themselves.
Like other automakers, Cadillac is clearly headed in this direction. The company is developing a technology called Super Cruise for semi-autonomous steering and cruise control on the highway that's expected in 2017. And GM recently acquired Cruise Automation, a startup that makes technology designed for fully autonomous driving for longer periods.
Like the simulations for materials and safety, autonomous driving simulations use highly complex mathematical equations. For example, the AI in self-driving cars has to determine how far away you are from other vehicles, your speed, the angle of the road and even subtle variations in wind conditions. Today, a Tesla Model S comes about as close as possible to robotic driving in a commercial vehicle, because it can adjust your steering and speed automatically and even knows when you are on a two-lane road versus a major highway.
Sullivan says computer modeling will play a pivotal role in designing the autonomous car. There are a multitude of scenarios, such as materials in the car, conditions on the road and even subtleties in human perception that will all have to be modeled thoroughly in order to create a completely safe and desirable self-driving car.
Indeed, as computer modeling plays an ever more important role in car design and manufacturing, Boyadjis muses, it could come to have a major impact on sales. As part of the purchase decision, customers might start asking "how many million hours was this vehicle tested on a computer?" right alongside "what colors does the leather come in?"