Session: DFMLC-02-01 Design for Supply Chain, end of Life Recovery, and Large Systems

Paper Number: 118241

118241 - On Holistically Optimizing the Spatial Configuration of Systems 

While disruptive design methods such as material distribution topology optimization have greatly improved components and structures over the past several decades, these methods and gains have yet to be realized for system-level applications. Specifically, the difference between good and bad system integrations can result in designs with key value metrics that vary by order of magnitude.

Many design decisions, such as not locating flammable materials directly adjacent to the hot exhaust piping, are obvious; however, this human intuition cannot scale for complex engineering systems with innumerable design decisions. Moreover, many of these decisions are nontrivial, and their tradeoffs must be carefully weighed. For instance, increasing power density is desirable but makes cooling and maintenance more difficult, which is undesirable. To compensate, current practice substitutes rapid innovation for incremental improvement, building upon proven designs. While these improvements accumulate over time, the prolonged design cycle comes at a great opportunity cost.

NSF recognized the need to improve system integration capabilities and funded the Engineering Research Center for Power Optimization of Electro-Thermal Systems (POETS). POETS laid the foundation for the SPI2 (spatial packaging of interconnected systems with physical interactions) design automation framework [1].

What sets SPI2 apart from other methods is how it holistically treats the various facets of system design optimization. Specifically, while existing design research has made significant progress toward effectively solving NP-hard (or similarly daunting complexity) problems such as component layout (3D bin packing) and interconnect routing (path planning) individually, SPI2 takes the first steps toward solving them simultaneously. Codesign is essential for this class of problem because systems and their environments are governed by strongly coupled phenomena (e.g., thermal, hydraulic, electromagnetic, mechanical).

SPI2 approaches this codesign through topology optimization methods. However, instead of representing the system as a continuously morphable, monolithic structure, it projects the rigid bodies of components and flexible splines of interconnects onto a mesh [2]. This enables gradient-based solvers to obtain the sensitivities of both spatial metrics (i.e., interference) and physics metrics from simulation [3, 4]. Since gradient-based methods exhibit local convergence, multistart analysis is employed. Naïve multistart sampling strategies include random or pseudo-random (e.g., Latin Hypercube). More informed strategies include decomposing the design space by unique spatial graphs using Yamada polynomial algebraic invariants [5].

Presenting Author: James Allison University of Illinois

Presenting Author Biography: James T. Allison is the director of the Engineering System Design Laboratory at the University of Illinois at Urbana-Champaign. His research interests include engineering system design (dynamic systems in particular), multidisciplinary design optimization, integrated physical and control system design, system architecture design, and design for energy efficiency. Design application domains include renewable energy system design, automotive and aerospace system design, robotic system design, and synthetic biology.

Prior to joining UIUC in 2011, he worked as a lecturer at Tufts University in the Department of Mechanical Engineering, as a senior engineer at MathWorks, Inc. in the area of dynamic system modeling and design, at GM in hybrid powertrain design, and at Ford Motor Company in the area of engine design optimization. Prof. Allison received his MSE degrees in mechanical engineering (2004) and industrial and operations engineering (2005), and a Ph.D. in mechanical engineering (2008), all from the University of Michigan.

Authors:

Chad Peterson University of Illinois
Satya Peddada University of Illinois
James Allison University of Illinois