VOL. 10, NO.

13, JULY 2015 ISSN 1819-6608

ARPN Journal of Engineering and Applied Sciences
©2006-2015 Asian Research Publishing Network (ARPN). All rights reserved.

www.arpnjournals.com

OPTIMIZATION OF NOZZLE: CONVERGENCE USING ANSYS WITH

RSM, MOGA
S. Jerin Winne and M. Chandrasekaran
Department of Mechanical Engineering, Sathyabama University, Chennai, India
E-Mail: jerinwinnes@gmail.com

ABSTRACT
In this paper, a multi objective optimization based on Genetic Algorithm, Response Surface Method and Design
of Experiments are adopted in order to calculate the optimal CFD model parameters, retaining the convergence. Objective
functions to be optimized simultaneously in such a real world complex multi-objective optimization problem. These
objective functions are either obtained from experiments or computed CFD approaches, unless a simple but effective meta-
model is constructed over the response surface from the numerical or experimental. So that modeling and optimization of
the parameters is investigated by using ANSYS. An ANSYS Fluent software package is utilized to simulate the viscous gas
flow-field in the nozzle; the Response Surface Methodology (RSM) applied for the reason of performing a multi-objective
optimization.

Keywords: optimization, DOE, RSM, MOGA, convergence.

INTRODUCTION of the process. Generally, these mathematical models are

The development of high-speed computers has polynomials with an unknown structure, so the
revolutionized the world. They have been changed our corresponding experiments are designed only for every
ways of thinking and have generated an impact in every particular problem. The choice of the design of
facet of our daily lives. Problems hitherto unsolvable have experiments can have a large influence on the accuracy of
come under the purview of computer solution. Over the the approximation and the cost of constructing the
past half-century, we have witnessed the rise in the new response surface.
methodology for attacking complex problems. In a traditional DoE, screening experiments are
performed in the early stages of the process, when it is
likely that many of the design variables initially
Geometry Optimized considered have little or no effect on the response. The
Design purpose is to identify the design variables that have large
Initial CAD
effects for further investigation.
update

Optimization RESPONSE SURFACE METHODOLOGY (RSM)

Response surface methodology (RSM) is a
Design Modeler update DesignXplorer
statistical technique in which smooth functions, typically
polynomials, are used to model an objective function.
update

Throughout this work, ANSYS program is used to

generate response surfaces. For response surface analysis,
WB solver you can choose from three sampling methods: Central
Mesh
composite design, Box Behnken matrix, Optimal Space
ANSYS Meshing ANSYS Fluent
Filling and user-defined. In this work the central
composite design method is used.
ANSYS Workbench
CENTRAL COMPOSITE DESIGN (CCD)
Central Composite Design (CCD) is the default
Figure-1. Workflow for optimization. DOE type. It provides a screening set to determine the
overall trends of the meta-model to better guide the choice
DESIGN OF EXPERIMENTS (DOE) of options in Optimal Space-Filling Design. The generated
An important aspect of RSM is the design of design point’s location for the method is based on the
experiments (Box and Draper, 1987), usually abbreviated central composite design. N is the number of input
as DoE. These strategies were originally developed for the parameters and f is Factorial number. In this, no of design
model fitting of physical experiments, but can also be variables are N=2, the total no of design points is = 1+
applied to numerical experiments. In Design of 2×N+2(N-f) =1+4+4 =9
Experiments the target is the selection of points with that
the response should be evaluated.
Most of the criteria for optimal design of
experiments are associated with the mathematical model

5486
 VOL. 10, NO. 13, JULY 2015 ISSN 1819-6608
ARPN Journal of Engineering and Applied Sciences
©2006-2015 Asian Research Publishing Network (ARPN). All rights reserved.

www.arpnjournals.com

Table-1. CCD Design Points for the specified two input functions and Lagrange multipliers are not needed because
design parameters. the constraint handling uses the same non-dominance
principle as the objectives. This ensures the feasible
solutions are ranked higher than the infeasible solutions.
First Pareto front solutions are archived by
separate internal sample sets and this is different from the
developing sample set. This ensures Pareto front patterns
already available from earlier iterations minimally
disrupted. The selection pressure can be (and,
consequently, the elitism of the process) to avoid
premature convergence by altering the parameter Percent
Pareto.
The concept of Pareto dominance is importance
in multi-objective optimization, objectives and constraints
are mutually conflicting particularly where some or all. In
such case, no single point yields the "greatest" value for all
objectives and constraints. The greatest solutions, often
called a Pareto set, are group of solutions such that choose
any one of them in a position of another will constantly
MULTI-OBJECTIVE OPTIMIZATION give up quality for at least one objective or constraint,
Multi-objective Optimization Mathematically, a while improving at least one other.
multi-objective problem consists of optimizing (i.e., Unfortunately, the Pareto optimum almost always
minimizing or maximizing) several objectives gives not a single solution, but a set of solutions. Usually
simultaneously, with a number of inequality or equality Pareto optimality is spoken of as being global or local
constraints. The Multi-objective optimization, which is depending on the neighborhood of the solutions X, and in
also called multi criteria optimization or vector this case, almost all traditional algorithms can at best
optimization, has been defined as verdict a vector of guarantee a local Pareto optimality. However, this
decision variables fulfilling constraints to reach adequate MOGA-based system, which incorporates global Pareto
values to all objective functions. The general Multi filters, yields the global Pareto front. The Maximum
objective Optimization Problem (MOP) can be formally Allowable Pareto Percentage criterion looks for a
defined as: Find the vector x ⃗ ∗ = [x∗ 1, x∗ 2, . . . , x∗ n] T percentage that represents a specified ratio of Pareto points
which will satisfy the m inequality constraints: gi(x ⃗ ) ≥ 0 per Number of Samples per Iteration. When this
i = 1, 2, . . . , m, The p equality constraints h i(x ⃗) = 0 i = percentage is reached, the optimization is converged.
1, 2, . . . , p and will optimize the vector function f ⃗ (x ⃗) =
[f1(x ⃗), f2(x ⃗), . . . , fk(x ⃗)]T. In these problems, there are CONVERGENCE CRITERIA
several objectives (a vector of objectives) to be optimized The iterative process is repeated until the change
(minimized or maximized) simultaneously. These in the variable from the one iteration to the next becomes
objectives often conflict with each other so that improving so small that the solution can be considered converged. At
one of them will deteriorate another. Therefore, there is no convergence: Discrete conservation equations like
single optimal solution as the best with respect to all the momentum, energy, etc. are maintained to be a specified
objective functions. Instead, there is a set of optimal tolerance in all cells. The results are no longer altered with
solutions, known as Pareto optimal solutions or Pareto additional iterations & Mass, momentum, energy and
front (Pareto, 1896) for multi-objective optimization scalar balances are obtained. Residuals measure imbalance
problems. The concept of Pareto front or set of optimal (or error) in conservation equations the convergence of the
solutions in the space of objective functions in multi- simulations is said to be achieved when all the residuals
objective optimization problems (MOPs) reach the required convergence criteria. These
convergence criteria are found by monitoring. The
convergence criterion for the continuity equation,
MULTI-OBJECTIVE GENETIC ALGORITHM
momentum, k and epsilon equations are 1E-4 and it is set to
(MOGA)
To attain the optimal parameters of nozzle, an 1E-6 for Energy equation.
integrated method that combine genetic algorithm with
CFD simulation analysis is set forward. The integrated
method not only shortens the system design, it also
extends optimization technique to realize the potential of
computer based design automation.
The Pareto ranking done by a fast, non-dominated
sorting method and this is an order of magnitude faster
than traditional Pareto ranking methods. The penalty

5487
 VOL. 10, NO. 13, JULY 2015 ISSN 1819-6608
ARPN Journal of Engineering and Applied Sciences
©2006-2015 Asian Research Publishing Network (ARPN). All rights reserved.

www.arpnjournals.com

RESULT AND DISCUSSIONS

Figure-3. Outlet and inlet angle convergence.

CONCLUSIONS
Multi objective Genetic algorithms have been
successfully used for optimization of nozzle and the
convergence of the simulation for Pareto based
optimization. This paper has been presented to multi-
objective optimization, aims to help the user to speed up
the choice of correct parameters and ensuring
Figure-2. Multi-objective genetic algorithm. simultaneously convergence of the CFD model. The
purpose is to recognize the optimal designs of supersonic
The MOGA method (Multi-Objective Genetic nozzles that perform utmost equality of thermodynamic
Algorithm) is a variant of the popular NSGA-II (Non- and flow-field properties respect to their average values at
dominated Sorted Genetic Algorithm-II) and based on nozzle exit. This work has established the effectiveness of
controlled elitism concepts. It supports multiple objectives Multi-Objective Optimization techniques in convergence
and constraints and aims at finding the global optimum. It and optimization of nozzle.
is limited to continuous input parameters. Initially
generate 10000 samples, 100 samples per iteration and REFERENCE
find 3 candidates in a maximum of 20 iterations. It
converged after 10719 evaluations. [1] Afsaneh Morshedi and Mina Akbarian. 2014. “Indian
Journal of Fundamental and Applied Life Sciences”
ISSN: 2231– 6345 (Online) An Open Access, Online
International Journal Available at
www.cibtech.org/sp.ed/jls/2014/04/jls.htm Vol. 4,
(S4), pp. 2434-2439/Morshedi and Akbarian.

[2] Ram Krishna Rathore, Amit Sarda and Rituraj

Chandrakar. 2012. “International Journal of Soft
Computing and Engineering”, (IJSCE) ISSN: 2231-
2307, Vol. 2, No.1, March.

[3] Y. Yu, M. Shademan, R.M. Barron and R.

Balachandar. 2012. “Engineering Applications of
Computational Fluid Mechanics” Vol. 6, No. 3, pp.
412-425.

[4] Design Xplorer user guide, ANSYS help

documentation, ANSYS version 14.0 2012.

[5] Montgomery, D.C. 1997, “Design and analysis of

experiments”, New York, John Wiley & Sons.

5488
 VOL. 10, NO. 13, JULY 2015 ISSN 1819-6608
ARPN Journal of Engineering and Applied Sciences
©2006-2015 Asian Research Publishing Network (ARPN). All rights reserved.

www.arpnjournals.com

APPENDICES

Appendix-A. Pressure velocity contours.

5489