by Aries Keck for NCSA

Dirt. It seems simple. It’s, well, dirt. You step on it; plant bulbs in it; wash it off your hands. But when an explosive landmine is buried in it, dirt suddenly gets more complicated.

Carey Rappaport, a professor at Northeastern University’s Center for Electromagnetic Research, spends a lot of time thinking about the complications of dirt. He’s leading a five-year, $5-million Army Research Office initiative to develop new technologies for detecting landmines.

“The key to finding a buried mine is really the dirt it’s buried in,” he says. “And that’s a real challenge, because a handful of dirt could be anything. It could be clay. It could be sand. If it’s raining, it could be mud. If it’s cold, it could be frozen solid. A handful of dirt. What is that, really?”

Getting down to the nitty-gritty is essential, he says, because you can’t detect a mine if your sensors can’t separate it from the soil around it. Accordingly, the centerpiece of Rappaport’s project is a complicated, three-dimensional computer model of dirt. Magda El-Shenawee is developing the model on NCSA’s SGI Origin2000 supercomputer. She is a rough-surface computational scientist who has postdoctoral training at the University of Illinois at Urbana-Champaign and is a visiting scholar at Northeastern.

Because no single system can find each of the more than 600 types of mines, Rappaport plans to use a family of sensing systems — radar, sound, infrared, and electromagnetic — in conjunction with the model. Along with this family of systems comes a family of specialists: Anthony Devany in diffraction tomography, Harold Raemer in radar, and Charles DiMarzio and Steve McKnight in acoustics. All these systems operate essentially as sonar does in the sea; send out a signal, wait for that signal to bounce off an object back to a receiver, and interpret the return signal. Running different types of simulations with the dirt model at their core will show the best way to integrate and set up an overall sensing system. The simulations will also show the best way to lay out minesweeping equipment that makes up that system, such as what sort of antenna to use and whether to put the receiver above or below ground.

“But the model has to happen before anything else,” says El-Shenawee. “We can’t do these things in practice.” It’s too dangerous to experiment with the real thing.

El-Shenawee developed the three-dimensional model of dirt using a unique technique called the steepest descent fast multilevel multipole method, or SDFMM, an algorithm that analyzes how electromagnetic waves scatter as they bounce off rough surfaces. SDFMM is essentially a combination of mathematical equations that calculate the electric and magnetic currents on the surface of an object. Originally developed by Vikram Jandhyala, Eric Michielssen, and Weng Cho Chew at U of I’s Center for Computational Electromagnetics, SDFMM was first used by El-Shenawee to study radar scattering on the ocean’s surface.

“When I came to Northeastern, we thought, ‘This is perfect. Why don’t we apply this technique to mine detection?'” she says. Working with Rappaport and signal processing specialist Eric Miller, El-Shenawee repurposed SDFMM to look beneath the soil’s surface.

“All the previous working models were two-dimensional,” she says. “There are three-dimensional models, but they’re not fast enough. It would take days and days to get an answer.” Using SDFMM, a simulation takes only hours. SDFMM runs a linear system of equations and considers a host of inputs important in creating the model all at once. These data include statistics that represent the behavior of electromagnetic scattered waves, receiver positions and angles, the boundary conditions of electric and magnetic fields, and interactions between the rough ground and smooth mine. Altering these inputs slightly and looking at the different outcomes will ultimately allow researchers to design the best system for detecting a mine.

Before any mine detection simulations could be run, however, El-Shenawee had to cajole the simulated sensory waves to get under the surface. The first computational tangle in this challenge was modeling the soil’s surface itself. Looking at the surface of soil, she found that the character of the dirt was not uniform. “Here it could be smoother, there rougher,” she says. To accommodate this natural occurrence, El-Shenawee adapted her ocean surface simulations so the model could take in all the possible combinations of soil’s hills and valleys.

Compounding the problem was the fact that getting this true picture of the soil’s character required running the same simulation over and over again. “If we run it once, the result means little,” she says. “We have to run it many times and take the average. Then what you get is useful information.”

Once the surface was modeled, she then had to extend the representation deep into the ground, essentially building a three-dimensional column of dirt. Next, she modeled the different types of sensory waves as they entered and bounced off of the soil’s surface. By extending those simulations down the column, much as she created the column in the first place, she developed a model that showed how those waves bounced along deep inside the soil.

Getting a good three-dimensional picture of dirt and seeing how different kinds of waves penetrate the surface and bounce through and out of the column let El-Shenawee move to the next step, simulating buried targets.

Large, metal anti-tank mines are no problem, Rappaport says. Finding them doesn’t even require a sophisticated sensing process. “That’s a solved problem,” he says. “Go down to Radio Shack, buy a metal detector, and you’re in business.” In contrast El-Shenawee’s model is designed to help the team detect cheaper and more prevalent plastic anti-personnel mines. These mines can be quite small, as little as three inches in diameter, and they’re often buried unmarked and forgotten.

“These are the ones that are tough to find,” Rappaport says. “The plastic and the soil background look the same. Electromagnetically, the explosive in the mine looks like the soil.” Often these mines contain only as much metal as a BB, rendering useless the conventional, Radio Shack methods.

To start finding these tiny targets, El-Shenawee used NCSA’s Origin2000 system to complete hundreds of runs that held just soil and hundreds that contained a buried mine. Side by side, images from the two types of runs looked remarkably similar. But subtracting the dirt from the model with a mine and using the model of just the ground as a reference caused the hidden explosive to suddenly pop into view. This simple subtraction began to give El-Shenawee a perfect picture of the mine itself. Using this information, she’s started measuring and cataloging the distinguishing features of the waves that scatter when they collide with a mine.

The idea, Rappaport says, is that this will ultimately help develop sensors that can find the little mines and unveil some clues about where to look for the mines in the first place. But there’s still a fly in the ointment. Unlike modeled dirt, real dirt isn’t very clean.

“The natural state of soil is so complex,” Rappaport said. “There’re lots of rocks and roots and vegetation, all sorts of junk is thrown in.”

The team is beginning to combat this problem as well by modeling earth that has other buried objects in it. Simulations like these will help the team discern between a buried mine and the false positive of a buried rock.

Even with such a model, however, there’s a long way to go between these still-in-development models and the current state of the art. The U.S. Army, which uses the best available detection equipment, is still using systems based on 50-year-old metal-detector technology, according to Rapapport. Further, the world’s most common detection method, used by the cash-poor countries that are riddled with mines, is still crawling belly-down on the ground, poking a sharp stick into the dirt, feeling for resistance.

In such a world, the smallest advance would be great step forward. The Northeastern team is already on its way to finding a new, sophisticated way to dig through the dirt-and save lives.

This research is supported by the Army Research Office and the Department of Defense.

============================================================