Qualitative and quantitative airfoil design optimisation using interactive genetic algorithms
This paper introduces the necessity of qualitative views on the evolutionary optimisation of airfoil shape design and applies multi-objective and parallel interactive genetic algorithms to combine qualitative evaluation with quantitative shape optimisation. The interactive evaluation enables user to embed domain specific knowledge which is frequently hard to describe. A comparison of the multi-objective and parallel IGA results is made using the fitness convergence, diversity and user preference performance metrics. Although the multi-objective IGA provides more diverse results the parallel IGA obtains significantly better fitness convergence and user preference. It is reported that the ability to vary population sizes and number of generations on separate population islands in parallel IGA is an ideal method to combine user guidance with computationally intensive search.
| Attachment | Size |
|---|---|
| A_Brintrup_iPROMS.wmv | 9.26 MB |
Dear authors,
Unfortunately the quality of sound is very poor and there are lots of noises. It seems it is because of AC or another machine, would you please re-record your nice presentation and re-upload it in case all can get benefit from your presentation.
Many thanks,
Afshin
Dear Afshin,
Unfortunately I am unable to re-record the presentation at the moment. A possible solution which I tried is to download the file, open the file with Windows Media player, open the graphic equalizer and reduce the following frequencies to zero: 500 Hz, 1 and 2 KHz.
This seems to give a much clearer sound. Many thanks for bringing this to my attention.
Alexandra
Dear D Pham,
Thank you for your message. You are right a significant hinder of IGA is human fatigue at the moment. There is research on the development of various strategies to support the human evaluator and reduce human fatigue. Some of these include [1]:
-using neural networks that learn from user evaluation in the first few generations,
-using clustering of solutions such that designs with similar characteristics are assigned the same fitness value or a value proportional to their difference,
-active user intervention where user modifies designs real time
These strategies all have pros and cons associated with them. For example using active user intervention in a multi-objective environment where the other objective is quantitative leads to fast convergence but may create sub optimal designs in the quantitative objective space [2].
Many other algorithms get around this problem by using the interactive element at every set number of generations, however the effectiveness of this solution depends on the qualitative objective. With a fuzzy objective this yields better results than a qualitative objective such as aesthetics [3].
I hope this helps; best regards,
Alexandra
[1] H. Takagi. Interactive Evolutionary Computation: Fusion of the capabilities of EC Computation and Human Evaluation, in Proc. IEEE, 89:9, 2001, pp.1275-1296.
[2] A. Brintrup, H. Takagi. The Affect of User Interaction Mechanisms in Multi-objective IGA, Proc. of Genetic and Evolutionary Computation Conference (GECCO 2007), 7-11 July 2007
[3] A Brintrup, J.Ramsden, and A.Tiwari. Integrated qualitativeness in design by multi-objective optimization and interactive evolutionary computation, IEEE Cong. Evolutionary Computation (CEC2005), Edinburgh, September 2005, pp. 2154-2160.
Dear authors,
Thank you for your well-written and interesting paper.
My question concerns the feasibility of using IGAs in practice when there are large numbers of solutions to evaluate and many generations to evolve.
Best wishes.
D Pham.