|Volume 7, Issue 1 -
Research Focus: Optimization Group Explores New Areas in Multidisciplinary Design
Participants: John Dennis, Robert Michael Lewis, Virginia Torczon, and Karen Williamson, Department of Computational and Applied Mathematics, Rice University
Industry researchers are increasingly realizing the need to integrate into the design process considerations formerly kept separate. For example, the processes for manufacturing and maintaining the product should be considered at the same stage of design as the performance of the product. This is necessary to reduce product development costs that skyrocket when myopic early design decisions must be changed. A 1992 article in Business Week notes that although five percent of the cost of most products is spent on design, product design locks in about 70% of the total cost. In addition, the current climate of global competition requires manufacturers to decrease the "time-to-market" for new products.
To approach these issues, CRPC researchers are investigating the use of high-performance computer networks to develop innovative methods in multidisciplinary design optimization (MDO). Initial research has focused on developing an MDO system template into which various applications modules can be placed to implement a specific design system. Researchers will initially test this template with applications modules on the optimal design of physical- chemical processes for water treatment and of catalytic converters for air pollutant control. Research from this project may ultimately lead to computational tools that influence the way design engineers develop new products.
More generally, "multidisciplinary design optimization problems arise whenever product engineers attempt to optimize, control, design, or identify parameterized processes," said John Dennis, a CRPC researcher at Rice University who leads the CRPC optimization group. "There are instances of MDO problems in aircraft design, automotive design, environmental process design, and chemical process control, to mention just a few examples."
The mathematical and computational formulation of an MDO problem frequently involves the manipulation of conceivably millions of variables and constraints that are involved with the design. What Dennis and CRPC colleagues Virginia Torczon and Robert M. Lewis have proposed is to develop an optimization core for a tool that will help engineers design better products and engineering systems. The tool would be a computing environment that rapidly reconfigures individual engineering simulation codes and couples them with advanced optimization methods and distributed computation tools.
An important but generally neglected aspect of an MDO system is modeling the optimization criterion or objective itself, a process that becomes more complex in a multidisciplinary environment. Designers who are used to working with a single discipline have a simplified view of what constitutes an optimal design. To address this problem, the group is investigating the use of optimization methods to synthesize the various single design objectives into a total product design objective. Torczon says that, "we are working to develop tools that put the human decision maker in the loop. We are not attempting to replace engineering judgment with push-button design."
Advances in computational science have led to a surge in single- discipline simulators. This makes a renewed effort in MDO timely. The group is also addressing other issues involved with MDO systems that contribute to the difficulty of the problems. For instance, how will explicit design constraints and discrete variables be managed? Also, when working with a number of competing goals, what is the measure of design quality that should be optimized?
The current project goal is to use a testbed system consisting of a network of scientific workstations to measure the real-world feasibility of the venture. By using this configuration, the potential of parallel computing will be brought closer to practicing engineers. However, the computational model being employed can be adapted to all forms of parallel processing, from advanced parallel supercomputers to networks of personal computers. Preliminary work has already been conducted on an Intel iPSC/860 parallel supercomputer. Collaborations with researchers from Boeing have resulted in new approaches to MDO that yield efficient optimization methods with structures exhibiting coarse-grained parallelism.
The researchers plan to expand their collaborations, working with a team of engineers and computer scientists from various areas of industry and academia. Researchers from Boeing and Shell Development Company have expressed interest in continuing their involvement with this project. The formation of MADIC, the Multidisciplinary Analysis and Design Industrial Consortium, may result in further collaborations for the project. John Dennis and Ken Kennedy of the CRPC are currently principal investigators on a NASA cooperative agreement to fund several MADIC activities.
"We think that multidisciplinary design optimization could prove to be the defining technology in the emerging field of computational engineering," said Dennis. "The 'grand challenge' that needs to be addressed is how to take MDO from being an engineer's dream to being an engineer's standard design tool."
Editor's note: The CRPC has five major research thrusts. Each issue of Parallel Computing Research will highlight one of these thrusts through a "Research Focus" article.
Table of Contents