Share |

Supercomputers help reveal dinosaur mechanics

The adult weight of animals that live on the land varies from a few grams to many tonnes. Still, they are all made up of the same materials. As animals get larger they also get stronger but not as quickly as they get heavier, which has a major influence on their mechanics. The BigDino project contributes to reveal the limits at the larger end of the spectrum by combining knowledge from different fields and simulating dinosaur locomotion.

The BigDino project is a collaboration between palaeontologists, biologists and computational scientists. The goal is simply to understand more about how dinosaurs moved.

“People are used to seeing the skeletons in museums and they have seen Jurassic Park, but there is really no biological basis behind explaining their movement. We do not really know the details, especially when it comes to the very big dinosaurs that weighed at least 70 tonnes, possibly even 100 tonnes. It is hard to imagine how they moved, and due to their massive size there must have been some mechanical problems,” explains William Sellers from University of Manchester’s faculty of life sciences, UK.

Sellers and his colleagues started making computer simulations of fossils about ten years ago. Initially their interest has focused on human evolution and why humans walk on two legs, which was the earliest difference between the human and the chimpanzee lineage and started long before people developed the large brains that we have today.

“We became interested in dinosaurs because they are so extreme; bigger than any other land animal that is alive now. Looking at the extremes reveals the limits of the whole system,” Sellers says.

Voilà, a virtual dinosaur

Making a dinosaur simulation begins with the skeleton. “Some amazing, very complete skeletons have been found, so assembling them is not very difficult. We go into a museum and use a laser scanner to produce accurate 3D-pictures of the dinosaur. We put the scans into the computer, and start to add muscles around the skeleton,” says Sellers.

The muscle structure is based on closely related animals like crocodiles and birds. In fact, as Sellers points out, muscles have stayed remarkably consistent throughout vertebrate evolution and scientists know how they work. The dinosaur bones, like any other bones, have lumps on them, which reveals where the muscles must have been – the anatomical fundamentals are all the same, whether it is humans, birds or crocodiles.

To finish off the virtual robot, the researchers add joints, and voilà, they have created a dinosaur on the computer using modern animal physiology. But this is only where the project gets complicated.

The adult weight of animals that live on the land varies from a few grams to many tonnes. Still, they are all made up of the same materials. As animals get larger they also get stronger but not as quickly as they get heavier, which has a major influence on their mechanics. The BigDino project contributes to reveal the limits at the larger end of the spectrum by combining knowledge from different fields and simulating dinosaur locomotion.

“We teach the robot to walk or run, using genetic algorithms and activation patterns of the muscles that would lead to, for example, maximum running speed,” Sellers says.

On the scale

One key thing that Sellers and his colleagues want to find out is how much the dinosaurs weighed. Determining the weight is very important because it is such a fundamental biological property. Everything depends on body mass: power, speed, strength, and so on. People often argue about how much dinosaurs weighed, and the information can vary greatly between different sources.

“We have recently come up with a new technique of calculating the body mass, and it seems reliable. It is calibrated on big modern animals like elephants, rhinos, giraffes and polar bears. The computer wraps the surface around the skeleton – just like the minimum amount of paper you need to cover any given object, it can be mathematically calculated. One of my colleagues compared it to wrapping a tea pot,” Sellers adds.

Interestingly, the calculations continue to produce fairly low values of body mass. That means that the dinosaurs could have been faster and more athletic than previously thought. “If we make them very heavy in the simulation, it is difficult to make them fast,” he says.

At the faster end of creatures

So far, the results from the BigDino project show that dinosaurs were quite fast but not super fast. “The simulations never show a big dinosaur that is really fast, like a racehorse, nor really slow. They seem to be more like the modern elephant in speed, which is surprising because they were a lot bigger”.

The dinosaurs seem to have been able to accelerate to speeds around 5–10 m/s, which is faster than the average human. A racehorse can run 20 m/s, which is impressive for a half a ton animal.

“With dinosaurs, there is always a vast range of predictions, but our results are based on sounder biological and mechanical evidence. They are based on an actual robot that uses everything we know about anatomy and physiology together, which makes it as accurate as possible,” Sellers concludes.

Supercomputing

Making the actual simulation is relatively quick, but teaching it to move is complex and requires the development of so called machine learning algorithms. That means that the scientist can collect data that would have affected how the dinosaurs moved from the computer simulations and can try to find patterns that reveal the underlying mechanisms.

The simulations have to be repeated many times, which raises the dimensionality of the problem to a very high level. The more realistic they want the simulation to be the larger the search base becomes. Sellers explains that there are two criteria for the genetic algorithm: If looking for efficiency in motion and lower speeds, you have to use the maximum economy optimization criteria. For maximum speed you need the maximum speed criteria.

What is PRACE?

The mission of PRACE (Partnership for Advanced Computing in Europe) is to enable high impact scientific discovery and engineering research and development across all disciplines to enhance European competitiveness for the benefit of society. PRACE seeks to realize this mission by offering world class computing and data management resources and services through a peer review process. Find out more on the PRACE website.

The machine-learning algorithms are also the reason why supercomputing is required in the BigDino project. According to Sellers, it is a difficult problem with a very few good and many bad solutions. There are a couple of areas of the simulation that Sellers is hoping to improve in the future.

“Currently we are using a simple way of measuring contact between the feet and the ground, but we would like to be able to do it properly. That means that we will need to combine two different methods.”

Those two methods are MBDA, which refers to MultiBody Dynamic Analysis, and FEA, which stands for Finite Element Analysis. Combining the two would be much more accurate than the current method, but it will also slow down the simulation process. The calculation would take about 10 to 100 times longer, requiring an even bigger computer in the future.

BigDino-project used primarily the IBM Power6-system located at Italy’s largest computing centre CINECA, with some assist from Hector-supercomputer in EPCC (Edinburgh Parallel Computing Centre).

This article was originally published in the 2013 PRACE digest.

Your rating: None Average: 4.3 (3 votes)

Comments

Post new comment

By submitting this form, you accept the Mollom privacy policy.