If you've ever been lucky enough to walk away from a car accident without any serious injury then you may want to thank Hybrid III.
This pinkish-brown humanoid, with distinctive yellow and black patches on its temples, is the most common crash test dummy used to prove that most new vehicles can meet safety standards. But this long-standing guardian of health and safety may soon be forced to retire, and it’s thanks, in part, to our eating habits.
When Hybrid III was introduced in 1976 it was designed to represent the "average" American male: 1.8m (5ft 9in tall) and weighing 77kg (170lbs). Since then, the average American male has gained more than 11kg (25lbs) and is 2.5cm (1in) shorter.
"There is now a distinct mismatch between people and dummies," says Scott Gayzick, a scientist at the Wake Forest Center for Injury Biomechanics in Virginia.
Of course, that was always the case – an average is not representative of an entire population. But increasingly the standard models used for crash tests are becoming less and less like the world’s population of drivers.
For example, nations like China have seen a boom in car sales. In terms of testing vehicles, this is important because there is a significant difference between the average height of Chinese and American adults. “It doesn’t mean that crash tests are meaningless,” stresses Dr Michiel van Ratingen, secretary general at the European car safety programme Euro NCAP. He says that dummies have improved – and continue to improve - car safety.
For instance, Hybrid III has tried to diversify. It now comes in three adult sizes - a taller male, a shorter female and the original – and with a family of two children. And because it is only really suitable for simulating head-on collisions, a handful of cousins have been developed to test other types of common accident, such as rear-end crashes and side-impacts. Even then, Hybrid III and his extended family only cover a fraction of the number of crash scenarios in the real world.
And then there is a bigger problem - they are not really very much like humans at all. Anthropomorphic test devices (ATDs), as they are known, are made with steel and rubber so that they are robust enough to withstand the tests they are subjected to. More importantly, they don't have bones or vital organs like livers, hearts and brains. To get around this, they are equipped with a number of sensors that record things like motion and compression at certain points on the dummy during a crash. But, in reality ATDs can't accurately predict what happens to a real person’s bones and organs during a crash, let alone simulating the difference between the bone densities of a young or older person.
These limitations have driven researchers to come up with a different approach to crash tests that that can simulate the wide variety of people and body shapes found around the world. One of these projects is led by the Global Human Body Models Consortium (GHBMC), a group of car manufacturers and suppliers, including Chrysler and General Motors. Its idea is to create virtual humans: computer models of people that are far more human-like - or "biofidelic" in the terminology - than ATDs. "We want to be able to run tests with models of people who are obese, or with extra mass where real people have extra mass - around the abdomen - and we can't do that with an ATD," says Gayzick, a member of the consortium.
The virtual humans GHBMC scientists are creating are sophisticated affairs. To develop the models, they start by taking computed tomography (CT) scans of people in both lying and seated positions to build up a 3D representation of a human skeleton. Then they add magnetic resonance imaging (MRI) scans of people in a seated position using an upright MRI machine, to add the muscles, internal organs and brain.
"Once we have the bone component and the soft-tissue component we can build an accurate image of a person," says Joel Stitzel, one of the leaders of the GHBMC team working at Wake Forest. "That's like the canvas. Then we add the mechanics and the stiffnesses of the components to make it like a real human: we play Frankenstein."
So far the GHBMC has completed a model of a 50th percentile male, and despite its complexity it is only about 1Mb in size - small enough to fit on an old-fashioned floppy disk. The group is now working towards constructing a more sophisticated model that can easily be scaled up or down and adjusted to simulate people of different sizes, builds and ages.
A human model made like this can be placed in a computer model of a car that is under development (and since cars are designed on computer systems these are usually readily available.) The model can then subjected to a variety of simulated accidents.
A typical car accident – or even one using a crash test dummy - takes just one tenth of a second, or about half the time it takes to blink, Dr Stitzel says. But a virtual simulation can take up to ten hours, running on a powerful 64-bit processor computer system. That’s because rather than simply taking sensor readings at a few discrete points, they can show millions of points of information and create detailed models of exactly what is happening to different parts of the body.
"When you run a test with a dummy, all you can say is that the dummy gets hit in the head, here are the forces, here is the acceleration and here is the motion-time history of the head," says Guy Nusholtz, a manager at the Experimental and Computational Mechanics department at car maker Chrysler who is interested in brain injuries. "With a computer model we can attempt to duplicate the fundamental physics of the brain and see what happens to it."
Zombies and dragons
Whilst these models may provide important insights, they are not without problems for both manufactures and lawmakers, says Euro NCAP’s Dr van Ratingen.
“The more detail you add to a model the more it becomes like an individual person,” he says. “And of course when you develop a car, you don’t develop it for a single person; you develop it for a general population.”
So, manufacturers would have to run multiple tests with different virtual models, using an average of the result for the final design.
He also says that virtual models may have quality control problems which would be unacceptable to regulators. “If you use a different server, or computer you might not get the same answer,” he says, something that would be unworkable for certifying new cars.
As a result, both Dr van Ratingen and Dr Gayzick believe new cars will continue to prove that they meet safety regulations by passing tests conducted with ATDs for many years yet. “You can't argue with a physical test," says Dr Gayzick.
But he does suggest one way that virtual humans can help make safety tests using ATDs more stringent: running many different crash scenarios in a computer model can help identify a small number of scenarios that are most challenging to a vehicle frame and its occupants. Once those scenarios have been identified, they can be used for physical crash tests, he says.
And ironically, the same technology that is used to simulate crash tests with virtual humans is also being used by manufacturers to simulate tests with virtual crash test dummies.
The purpose is to ensure that a car design can pass a regulatory test first time and not run the risk of an expensive failure, according to Donal Mcnally, joint head of the Bioengineering Research Group at the University of Nottingham in the UK. "You could use software with models of crash test dummies to fine tune the time after an impact that an airbag is fired," he says. "That way you can find the optimal time to ensure you pass the regulations."
Ultimately these developments should make cars far safer. In addition, it raises the possibility of creating different versions of the same car for different markets. This would be expensive, but could result in cars sold in Japan being optimized for the safety of older people, while cars for the Chinese market could be designed for the safety of people who are, on average, smaller than Americans.
The work also may find its way into other parts of our lives, making an appearance on televisions or computer screens. "We are excited about the possibility of computer games companies using our human models," Dr Gayzick explains. "They have their own models already, but their movements are not as realistic, and they are not internally accurate."
That could be good news for gamers, but less so for the virtual humans themselves. After escaping repeated car crashes, they would be thrown into worlds of dragons, monsters or flesh eating zombies. Faced with that possibility, perhaps the idea of Hybrid III’s quiet retirement does not sound so bad after all.