Development of metamodel-based robust simulation optimization for complex systems under uncertainty
Computer simulations can help a rapid investigation of various alternative designs to decrease the required time to improve the system. Because of the complexity for analyzing complex systems in way of mathematical formulation, a simulation optimization has been an interest in analyzing and study...
Saved in:
| Main Author: | |
|---|---|
| Format: | Thesis |
| Language: | English |
| Published: |
2018
|
| Subjects: | |
| Online Access: | http://psasir.upm.edu.my/id/eprint/77627/1/FK%202019%2023%20ir.pdf |
| Tags: |
Add Tag
No Tags, Be the first to tag this record!
|
| Summary: | Computer simulations can help a rapid investigation of various alternative
designs to decrease the required time to improve the system. Because of the
complexity for analyzing complex systems in way of mathematical formulation, a
simulation optimization has been an interest in analyzing and studying the
behavior of complex systems in the real world of engineering problems. One of
the main difficulties of existing model–based simulation optimization methods is
dealing with large number of required simulation evaluation (also called
simulation experiments or computer experiments) which causes of costly
computational time. In addition, in order to improve the validity of optimal results,
uncertainty as a source of variability in the model’s output(s) need to be
considered while this importance mostly has been ignored in designing of
existing simulation optimization models. Under uncertainty, simulation running
with stochastic output is complex in terms of computational time and/or cost,
therefore the limited number of simulations is desirable. However, the accuracy
of simulation result strongly depends on the reality of computer coding and
discrepancy between simulation model and actual physical system. Most
existing simulation optimization methods need to be improved in such a way to
handle conflicting of multiple responses and constraints. This research generally
aims to develop the black-box simulation optimization technique to be applicable
in stochastic complex systems under effect of uncertainty with the least
optimization computational burden (number of simulation experiments). This
research develops a new distribution-free method for uncertainty management
with unknown distribution of uncertainty. This research also aims to show the
applicability and validity of proposed metamodel-based robust simulation
optimization method in practical engineering design problems such as direct
speed control of DC motor and PID tuning under uncertainty. For this purpose,
metamodeling techniques are used for global approximation of complex
simulation model. The statistical terminology of Taguchi crossed array design is
replaced by global modern metamodels. A distribution-free method is suggested
to tackle the lack of information about possible probability distribution of uncertainty scenarios in the model. Results of this research confirmed the validity
and applicability of the proposed methodology dealing with practical stochastic
complex engineering design problems in three terms; reducing computational
time, enhancing flexibility, and improving the applicability. The proposed method
can reduce the number of function evaluations for PID tuning under uncertainty
to 50 simulation runs compared to more than 1000 function evaluations in
common model based method. Compared to classical Ziegler Nichols method,
the proposed method shows the better performance which is more than 10% for
PID tuning under uncertainty. The proposed distribution–free method applied in
economic order quantity problem shows the same accuracy compared to studies
in literature whereby this study does not need to estimate distribution of
uncertainty. |
|---|