Abstract - An enhanced augmented Lagrangian
coordination(ALC) method based on Kriging model is
proposed .The classic speed reducer problem is designed as
an example to verify the enhanced augmented Lagrangian
coordination method, numerical results show that the
enhanced ALC method can not only obtain good
optimization results, but also greatly reduce the
computational cost and improve the efficiency of
optimization.
Keywords - Multidisciplinary design optimization,
augmented Lagrangian coordination, Kriging model, Latin
hypercube sampling.
I. INTRODUCTION
Multidisciplinary design optimization (MDO) is
used for the design of large and complex engineering
systems which are normally composed of a set of linked
subsystems. From last century, MDO has been widely
used in many areas with complex engineering systems,
including aviation, aerospace, automotive manufacturing.
Due to the large scale of the complex system to be
designed, it is often decomposed into smaller subsystems
to perform discrete optimization and management. The
design coordination methods are often applied to ensure
the coupling among subsystems. Typical MDO methods
include concurrent subspace optimization (CSSO)[1], bi-
level integrated system synthesis (BLISS)[2],
collaborative optimization (CO)[3], analytical target
cascading( ATC)[4]and augmented Lagrangian
coordination(ALC)
[5].Among them, CSSO and BLISS
are two-level optimization methods, which involve
unavoidable system analysis after each round of iteration,
and thus consume large computing efforts. As a result,
they are only applied to the design problem containing
continuous variables. In addition, their convergence
properties have not been theoretically proved. CO is
strictly limited to two-level optimization to ensure its
convergence and cannot be applied to most of the
complex systems which involve multiple levels. ATC is
based on hierarchical decomposition of targets, which is
mainly used for non-centralized problems with
hierarchical structures and belongs to a subclass of ALC.
ALC method overcomes the limitations of existing
methods. It has the advantages of flexibility and wide
application, which is widely concerned by scholars[6].
However, there exist the following deficiencies during the
solving process of ALC. The couplings between
subsystems make the number of their interaction
increased significantly, especially with the optimization
variables, objective function and the number of
constraints, the number of iterations is significantly
increased. Thus, it takes a lot of time to complete the
whole optimization process. In order to solve the above
problems, this paper intends to introduce the Kriging
approximation model[7]
in the optimization process,
without reducing the accuracy, to construct a calculation,
calculation simulation results and the actual results similar
to the mathematical model to replace the actual simulation
program, and in the iterative process of updating the
model, improve the accuracy and the optimization
efficiency of ALC method.
The approximate model technology is an important
method to solve complex engineering systems
multidisciplinary design optimization. At present, the
approximate models of the application of the engineering
optimization field are polynomial response surface model
[8], artificial neural network approximation model [9] and
Kriging model[7]
.
. Polynomial response surface model
although there are the advantages of computation quantity
is small, easy to use, but for engineering tend to have
multiple local optima and nonlinear degree higher, the
model fitting precision is poor[10]. Artificial neural
network approximation model, although it has a good
global approximation ability, but the approximate the
improvement of model accuracy in largely dependent on
more design samples and the neural network training
iteration, so a large amount of calculation[11]. The
Kriging model as an effective approximation techniques
in recent years has been widely used in structure
optimization[12,13], multidisciplinary design
optimization[14], aircraft[15] and vehicles [16]et al in the
field of structural optimization, is an estimate of the
minimum variance of estimation model, which has
features of global approximation and local random error
estimation. The Kriging model not only has a better fitting
effect on the problem of high nonlinear and local response,
but the parameters of its can be determined according to
the design sample. Therefore, the problem of large
amount of computation in the process of ALC is studied
by using the Kriging approximation model, so as to
improve the computational efficiency of ALC. The
remainder of this paper is organized as follows. The
Kriging approximation model is introduced in the second
section. The third section is the enhanced ALC method
based on Kriging model. The example analysis is in the
fourth section. Conclusions are in the last section
II. KRIGING MODEL
2.1 Mechanism of Kriging model
The model is derived from the spatial statistics[7], and it
is an unbiased estimation model with the smallest
An Enhanced ALC Based on Kriging Model for Multidisciplinary Design
Optimization
NIE Du-xian
1,2
, QU Ting
1,3*
, CHEN Xin
1
, WANG Mei-lin
1
, HUANG Guo-quan
1,3
1
Guangdong CIMS Provincial Key Lab, Guangdong University of Technology , Guangzhou, China
2
College of Mathematics and Informatics, South China Agricultural University, Guangzhou, China
3
Department of Industrial and Manufacturing Systems Engineering, The University of Hong Kong, Hong Kong ,China
(
*
quting@gdut.edu.cn)