|
|
Type of Document Dissertation Author Padmanabhan, Dhanesh Author's Email Address dpadmana@nd.edu URN etd-07022003-191128 Title Reliability-Based Optimization for Multidisciplinary System Design Degree Doctor of Philosophy Department Aerospace and Mechanical Engineering Advisory Committee
Advisor Name Title John E. Renaud Committee Member Michael M. Stanisic Committee Member Stephen M. Batill Committee Member Timothy C. Ovaert Committee Member Keywords
- Multidisciplinary Design Optimization
- Structural Reliability
- Reliability-Based Optimization
Date of Defense 2003-06-18 Availability unrestricted Abstract Reliability-Based Optimization (RBO) for engineering design deals mainly with twodesign attributes, namely the merit, for example cost, and the reliability of the design. In
this work the class of design problems which are considered, are designs characterized
by a minimum merit function and that satisfy certain reliability constraints. The reliability
constraints are typically constraints on the probabilities of failure due to component
failure events or a system failure event. These are obtained using standard reliability analysis
techniques such as First Order Reliability Method (FORM), Second Order Reliability
Methods (SORM) and Monte Carlo Simulation (MCS) techniques.
The reliability analysis and RBO are very expensive for multidisciplinary systems
consisting of various disciplines that are dependent on each other or coupled, for example,
an aeroelastic structure. Hence, the primary goal of the research is to develop
efficient methodologies that perform RBO for multidisciplinary systems. The methodologies
considered incorporates a Concurrent Subspace Optimization technique that allows
concurrent design optimization in each discipline. The methodologies also incorporate
approximation concepts to reduce the computational costs. There are essentially two
methodologies, one that uses a traditional reliability analysis method and the other that
uses a new reliability analysis method geared towards reduction of computational expenses
for coupled multidisciplinary problems. A new reliability analysis tool based on Trust Region methods was developed for the latter case. Both methodologies were applied
to multidisciplinary test problems and about 20%-30% computational savings were
observed.
A second goal of the research was to investigate the use of Monte Carlo Simulation
(MCS) techniques for reliability analysis in RBO, that are more accurate but more expensive
than FORM or SORM. In this work, conditional expectation MCS was selected
over indicator-based MCS techniques based on smoothness criteria and the availability
of analytic sensitivities. A MCS-based RBO methodology was developed and successfully
implemented to problems with both component and series failure events. It was
observed that designs with significantly lower merit functions were obtained for the application
problems considered, compared to a FORM-based RBO approach. It was also
observed that the computational costs were extremely high for one of the application problems.
Some suggestions for future research are made regarding development of efficient
methodologies for the MCS-based RBO.
Files
Filename Size Approximate Download Time (Hours:Minutes:Seconds)
28.8 Modem 56K Modem ISDN (64 Kb) ISDN (128 Kb) Higher-speed Access etd.pdf 665.56 Kb 00:03:04 00:01:35 00:01:23 00:00:41 00:00:03