You are on page 1of 6

Application of NSGA II and optimization

• Variables considered are the inlet absolute flow angle of the rotor 𝛼 and flow coefficient 𝜑

𝛼 ∈ [400 , 850 ] 𝑎𝑛𝑑 𝜑 ∈ [0.1,0.5]

• Objective functions are the isentropic efficiency and the specific velocity. The objectives are the
maximize the isentropic efficiency and to minimize the specific velocity.

Initial selected population size is 240, 24 samples with their binary value are denoted below.

1 2 3 4 5 6
𝛼 80 56.1 43.2 82.9 77.3 68.9
𝜑 0.15 0.23 0.37 0.42 0.46 0.5

7 8 9 10 11 12
𝛼 73.4 66.1 48.4 68.2 52.8 78.5
𝜑 0.11 0.30 0.47 0.22 0.34 0.25

13 14 15 16 17 18
𝛼 44 45.4 63.9 84.7 74.2 55.5
𝜑 0.14 0.22 0.39 0.29 0.41 0.19

19 20 21 22 23 24
𝛼 70 44.9 33.8 52.9 63.3 67.0
𝜑 0.25 0.43 0.34 0.42 0.36 0.48

To represent 𝛼, 9 bits are used considering 10-1 significance and 6 bits are used for 𝜑
considering 10-2 significance.

So, our population will be, Cross over points

1 2 3 4 5 6
𝛼 1100 10000 0101 00001 0001 00000 1101 01101 1011 10101 1001 00001
𝜑 000 101 011 011 001 011 100 000 100 100 101 000

Cross over points

The cross over is denoted only for a sample of 6, but this was done for all the 240 samples
So we consider one point cross-over probability for the first 6,

1 2 3 4 5 6
𝛼 1100 00001 0101 10000 0001 01101 1101 00000 1011 00001 1001 10101
𝜑 000 011 011 101 011 000 100 011 100 000 101 100

1 2 3 4 5 6
𝛼 78.5 57.6 44.5 81.6 75.3 70.9
𝜑 0.13 0.39 0.34 0.45 0.42 0.54

After conducting mutation,

1 2 3 4 5 6
𝛼 0011 11110 1010 01111 1110 10010 1101 00000 1011 00001 1001 10101
𝜑 111 100 100 010 011 111 100 011 100 000 101 100

1 2 3 4 5 6
𝛼 68.3 47.4 54.6 83.7 72.5 75.7
𝜑 0.23 0.39 0.44 0.28 0.32 0.47

Considering the given figure 1 below, next parent population is selected considering the best
solutions of the pareto fonts shown (a fitness parameter driven by high isentropic efficiency and
specific velocity in the optimum range (0.7-0.85).
So, the new parent population of 24 selected and one point cross-over was done followed by
mutation.

Finally, after selecting 16 parents from the previous one, the same procedure was applied and
the solutions are converged to a single pareto font.

1 2 3 4 5 6
𝛼 79.8 52.6 73.5 86.6 49.5 60.2
𝜑 0.3 0.16 0.44 0.49 0.38 0.29
Figure 1: Pareto fronts of the children population (4th generation)

Isentropic efficiency vs specific speed


Ideal point
0.9

0.88
Specific velocity

0.86

0.84
Non
0.82 Ideal
point
0.8

0.78
74 76 78 80 82 84 86 88
Isentropic efficiency (%)

Figure 2: Pareto front of the converged population


The criterion is to obtain a higher isentropic efficiency and a higher specific velocity in the range.
When considering the solutions, the best non dominating solution is obtained. The priority
should be provided for the isentropic efficiency.

In order to pick the best solution, Technique for Order Preference by Similarity to Ideal Solution
(TOPSIS) was applied in this study, which gives the solution closer to the ideal solution and
farthest from the non-ideal solution. During the decision-making process, it was decided to
prioritize the i performance of the system by assigning weight of 0.7 to the isentropic efficiency
and weight 0.3 to the specific speed. For that weight distribution, the final optimal point was
indicated, and it is marked in the above figure as the TOPSIS point.

The weights for the two variables are assigned like below.
The priority given for each of the variables, isentropic efficiency and the specific speed are
collected from literature and a score is given for each between 0-1.
Then the scores are standardized by taking the root of the square of the sum of the both.

Table 1- The weights given by the literature

Variable Li et al Meitner Vivian L Jankowski Rahbar K. Rohlik


et al et al H.E

Isentropic 0.7 0.8 0.75 0.9 0.85 0.75


efficiency 1.79

Specific 0.45 0.2 0.33 0.25 0.4 0.35


speed 0.835

Table 2- Standardization of the weights given by the literature

Variable Li et al Meitner Vivian L Jankowski Rahbar K. Rohlik


et al et al H.E

Isentropic 0.97/1.49 0.8/1.49 0.85/1.49 0.9/1.49 0.85/1.49 0.95


efficiency =0.74 =0.65 =0.72 =0.7 =0.72 /1.49
=0.81
Specific 0.45/0.835 0.2/0.835 0.33/0.835 0.25/0.835 0.4/0.835 0.35/
speed =0.54 =0.24 =0.4 =0.3 =0.48 0.42
After obtaining the average weight of each of the variables, the parameter sets were evaluated
and the point closer to the ideal point is selected.

Table 3- Final pareto points

Variable Point 1 Point 2 Point 3 Point 4 Point 5 Point 6 Point 7 Point 8

Isentropic 87 84.3 71 76.7 82.4 87 78.2 79.5


efficiency

Specific 0.81 0.8131 0.859 0.884 0.8131 0.937 0.85 0.846


speed

𝛼 (0) 𝜑 Ƞ (%) Specific velocity


79.8 0.3 87 0.817014
73.5 0.44 84.3 0.813161
52.6 0.16 71 0.859618
86.6 0.49 76.7 0.884354
49.5 0.38 82.4 0.813161
60.2 0.29 87 0.937014
65.8 0.37 78.2 0.85
55.09 0.41 79.5 0.846

Isentropic efficiency vs specific speed


Ideal point
0.9

0.88
Specific velocity

0.86

0.84 TOPSIS point


Non
0.82 Ideal
point
0.8

0.78
74 76 78 80 82 84 86 88
Isentropic efficiency (%)

Figure 3: Best point of the pareto front of the converged population

You might also like