Earlier this year, an experiment was conducted simultaneously at Lehigh, Illinois and Rensselaer Polytechnic Institute. Researchers simulated the effect of earthquakes on a freeway bridge structure, including several columns and the soil foundation. The columns were replicas of structures that were damaged during the 1994 Los Angeles earthquake.
The novel experiment, called a distributed hybrid test, was conducted to obtain an integrated picture of what happened systemically to the bridge, soil foundation and columns during the earthquake.
Until now, says Jim Ricles, director of Lehigh's RTMD facility, earthquake engineering experiments have focused on single structural members as opposed to entire systems.
"Before, one isolated specimen was tested without integrating the data to the overall bridge structure. But if you test only one member, you don't know how the rest of the structure will react.
"After we prove the worth and accuracy of distributed hybrid testing, we can use it to demonstrate alternative types of constructions and renovations. It will be a tool to demonstrate as realistically and accurately as possible the real-time performance of a structure during an earthquake."
The RTMD facility has five 20-foot-long, torpedo-shaped actuators, which can impose 500,000 pounds of force on structures from a variety of angles at the rate of 1 meter per second. The actuators are the largest of their kind for structural testing. An advanced high-speed data acquisition, coupled with sensors and an eight-channel digital control system, enables the RTMD facility to be operated remotely.
 |
Jeff Linderoth |
Investors predicting the stock market, farmers fretting over the weather, security officials anticipating a terrorist attack – the amount of uncertainty in the world can seem daunting.
Not to Jeff Linderoth.
Linderoth, assistant professor of industrial and systems engineering, marshals the world's most powerful computers to tackle problems with millions of variables and possible outcomes.
Last year he received an Early Career Award from the U.S. Department of Energy (DOE) to solve large-scale numerical optimization problems characterized by high levels of uncertainty.
Linderoth tackles these "stochastic optimization" problems by developing software programs that use computational grids – groups and networks of computers whose collective capacities are harnessed to solve large-scale problems that would overwhelm a single computer.
Stochastic optimization problems require engineers to take into account two classes of variables, says Linderoth. Decision variables can be controlled. Random variables cannot.
One example of a stochastic optimization problem, says Linderoth, is the effort by the federal government to intercept illegal drugs flowing into the U.S.
"The government has a fixed budget to make it more difficult for drug smugglers to travel through the network, or border," says Linderoth. "The uncertainty comes into play when, even if we choose to beef up a part of the network by installing sensors, we don't know if the sensors will detect the drugs. The goal is to design a system that is robust with respect to these failures and will, in the long run, catch as many smugglers as possible."
Linderoth will also help DOE with mixed-integer nonlinear programming problems (nonconvex optimization), which involve discrete, yes-or-no types of decisions as well as nonlinear elements that have a continuum of random values.
The design and installation of a gas pipeline network, says Linderoth, requires engineers to make both types of decisions. The location of pipes and compressors requires discrete decisions, while calculating the optimal flow of natural gas is a nonlinear problem.
All the software Linderoth develops will be made available via the Network-Enabled Optimization System (NEOS), a server established 10 years ago to help engineers, scientists, students and businesspeople solve optimization problems remotely over the Internet.
Linderoth was a member of the Argonne and University of Iowa team that solved the nug30 Quadratic Assignment Problem, a complex facility-location problem that had been unsolved for 30 years. The computation took one week of calendar time and 11 years of CPU time, as 653 machines participated.
