In other words, the proposed approach worked well in reaching the desired zone of the Pareto Set that was not mapped by the initial NSGA-II first phase optimization. It should be emphasized that the improvement has been obtained with only 19 additional CFD computations (DOE of 16 for the restricted design parameter space, plus 3 validations), and hence in less than one additional generation of the original NSGA-II optimization.
Figure 6 compares the temperature fields in the vertical symmetry plane for the original design (left), the NSGA-II improved design (center) and the final one (right). The decreased heat dispersion is due to a smaller warm air plume and to the more regular temperature distribution (generally lower) inside the device. Moreover, the proposed design is also quite original, with respect to the result of the first direct optimization, with regard to its location in the input parameter space. This means that the proposed approach found a new optimal ”island” in the design parameter space with a ”cost” of only 19 additional CFD computations following the initial 300–CFD computations of the NSGA-II first phase optimization.