Optimization and Computational Fluid Dynamics - Department of ...
Optimization and Computational Fluid Dynamics - Department of ...
Optimization and Computational Fluid Dynamics - Department of ...
Create successful ePaper yourself
Turn your PDF publications into a flip-book with our unique Google optimized e-Paper software.
8 Multi-objective <strong>Optimization</strong> in Convective Heat Transfer 237<br />
classified in several ways that depend on different aspects <strong>of</strong> the problems<br />
themselves. In [45], the following classifications are highlighted.<br />
A first classification can be based on:<br />
1. Existence <strong>of</strong> constraints. As stressed earlier, problems can be classified<br />
constrained or unconstrained. Constraint h<strong>and</strong>ling is not a trivial task for<br />
most optimization techniques.<br />
2. Nature <strong>of</strong> design variables. f can be a function <strong>of</strong> a primitive set <strong>of</strong> variables<br />
depending on further parameters, thus becoming trajectory optimization<br />
problems [24].<br />
3. Physical structure <strong>of</strong> the problem. Depending on the structure <strong>of</strong> the problem,<br />
optimal control theory can be applied, where a global cost functional<br />
is minimized to obtain the desired solution.<br />
4. Nature <strong>of</strong> the relations involved. When known or at least well guessed, the<br />
nature <strong>of</strong> the equations governing the model <strong>of</strong> the system under study<br />
can address the choice to the most efficient among a set <strong>of</strong> optimization<br />
methods. Linear, quadratic geometric <strong>and</strong> nonlinear programming are examples.<br />
5. Permissible values <strong>of</strong> design variables. Design variables can be real value<br />
or discrete.<br />
6. Deterministic nature <strong>of</strong> the variables. The deterministic or stochastic nature<br />
<strong>of</strong> the parameters is a criterion to classify optimization problems.<br />
In particular, the concept <strong>of</strong> Robust Design or Robust <strong>Optimization</strong> has<br />
recently gained popularity [32].<br />
7. Separability <strong>of</strong> the functions. A problem is considered separable if f functions<br />
can be considered a combination <strong>of</strong> functions <strong>of</strong> single design variables<br />
f1(X1), f2(X2),..., f2(Xn) <strong>and</strong>f becomes:<br />
f(X)=<br />
n�<br />
fi(Xi) .<br />
The advantage <strong>of</strong> such a feature is that in nonlinear problems, nonlinearities<br />
are mathematically independent [22].<br />
8. Number <strong>of</strong> objective functions. Depending on the number <strong>of</strong> objective functions,<br />
the problem can be single- or multi-objective. This is an outst<strong>and</strong>ing<br />
distinction, since in multi-objective optimizations, objectives are usually<br />
conflicting. No single optimum exists, but rather a set <strong>of</strong> designs the decision<br />
maker has to choose from. This is one <strong>of</strong> the motivations that have led<br />
to the birth <strong>of</strong> Evolutionary Multi-Objective <strong>Optimization</strong> (EMOO) [8].<br />
Depending on the characteristics <strong>of</strong> optimization problems, many techniques<br />
have been developed to solve them. These can be roughly divided into two<br />
categories.<br />
i=1<br />
1. Traditional mathematical programming techniques. They require a certain<br />
knowledge <strong>of</strong> the relation between objectives <strong>and</strong> design variables, <strong>and</strong><br />
they are usually best suited for single-objective optimizations.