CFD Online Logo CFD Online URL
www.cfd-online.com
[Sponsors]
Home > Wiki > Shape Design Optimization

Shape Design Optimization

From CFD-Wiki

(Difference between revisions)
Jump to: navigation, search
(New page: Aerodynamic shape design optimization can be classified as two categories: inverse design and direct design. Inverse design requires specification of target pressure or velocity distributi...)
 
(2 intermediate revisions not shown)
Line 1: Line 1:
-
Aerodynamic shape design optimization can be classified as two categories: inverse design and direct design.
+
Aerodynamic shape design optimization can be classified as two categories: [[inverse design]] and [[direct design]].
-
Inverse design requires specification of target pressure or velocity distribution on the surface of an body, of which the success highly depends on the experience of the designer.
+
[[Inverse design]] requires specification of target pressure or velocity distribution on the surface of an body, of which the success highly depends on the experience of the designer.
-
Direct design can be further classified into gradient-based methods and global search methods.
+
[[Direct design]] can be further classified into [[gradient-based methods]] and [[global search methods]].
Gradient-based methods will reach a local optimum design, while global search methods can in theory reach a global optimum. But the cost of global search methods is prohibitively huge with a large number of design variables.
Gradient-based methods will reach a local optimum design, while global search methods can in theory reach a global optimum. But the cost of global search methods is prohibitively huge with a large number of design variables.
 +
 +
 +
[[Indirect Methods]]
 +
 +
[http://en.wikipedia.org/wiki/IOSO IOSO] (Indirect Optimization on the basis of Self-Organization) Technology is based on the [http://en.wikipedia.org/wiki/Response_surface_methodology response surface methodology] approach.
 +
At each IOSO iteration the internally constructed response surface model for the objective is being optimized within the current search region. This step is followed by a direct call to the actual mathematical model of the system for the candidate optimal point obtained from optimizing internal response surface model. During IOSO operation, the information about the system behavior is stored for the points in the neighborhood of the extremum, so that the response surface model becomes more accurate for this search area. The following steps are internally taken while proceeding from one IOSO iteration to another:
 +
*the modification of the experiment plan;
 +
*the adaptive adjustment of the current search area;
 +
*the function type choice (global or middle-range) for the response surface model;
 +
*the adjustment of the response surface model;
 +
*the modification of both parameters and structure of the optimization algorithms; if necessary, the selection of the new promising points within the search area.
 +
 +
IOSO Technology algorithms have good global properties and are able to find the global optimum with very high probability. IOSO Technology was designed to solve very complex optimization tasks.
 +
 +
 +
----
 +
'''ISSUES in Direct Design Shape Optimization for CFD'''
 +
 +
1) Huge Processing power is pre-requisite.  This factor is becoming less an obstacle as computing power is going cheaper as we speak.
 +
 +
2) Automatic surface or shell modification in unison with automatic volumesh adjustment.  As the surface changes and develops according to a preset lower bound and upper bound in an optimization code, the volumesh domain is affected unprecedentedly.  Refinement in the volumesh and quality aspects of the volumesh have to built into the CFD solver to ensure smooth flow in the auto-optimization process.

Latest revision as of 14:55, 7 January 2009

Aerodynamic shape design optimization can be classified as two categories: inverse design and direct design. Inverse design requires specification of target pressure or velocity distribution on the surface of an body, of which the success highly depends on the experience of the designer. Direct design can be further classified into gradient-based methods and global search methods. Gradient-based methods will reach a local optimum design, while global search methods can in theory reach a global optimum. But the cost of global search methods is prohibitively huge with a large number of design variables.


Indirect Methods

IOSO (Indirect Optimization on the basis of Self-Organization) Technology is based on the response surface methodology approach. At each IOSO iteration the internally constructed response surface model for the objective is being optimized within the current search region. This step is followed by a direct call to the actual mathematical model of the system for the candidate optimal point obtained from optimizing internal response surface model. During IOSO operation, the information about the system behavior is stored for the points in the neighborhood of the extremum, so that the response surface model becomes more accurate for this search area. The following steps are internally taken while proceeding from one IOSO iteration to another:

  • the modification of the experiment plan;
  • the adaptive adjustment of the current search area;
  • the function type choice (global or middle-range) for the response surface model;
  • the adjustment of the response surface model;
  • the modification of both parameters and structure of the optimization algorithms; if necessary, the selection of the new promising points within the search area.

IOSO Technology algorithms have good global properties and are able to find the global optimum with very high probability. IOSO Technology was designed to solve very complex optimization tasks.



ISSUES in Direct Design Shape Optimization for CFD

1) Huge Processing power is pre-requisite. This factor is becoming less an obstacle as computing power is going cheaper as we speak.

2) Automatic surface or shell modification in unison with automatic volumesh adjustment. As the surface changes and develops according to a preset lower bound and upper bound in an optimization code, the volumesh domain is affected unprecedentedly. Refinement in the volumesh and quality aspects of the volumesh have to built into the CFD solver to ensure smooth flow in the auto-optimization process.

My wiki