1. Derivative-free Methods for Structural Optimization
Guilherme Ilunga, António Leitão
INESC-ID/Instituto Superior Técnico, Universidade de Lisboa

2. What is Optimization?

3. What is Optimization?
Finding the parameters which give the best value of a given metric

4. What is Optimization?
Finding the parameters which give the best value of a given metric
Example:
Parameters: position and height of each pillar

5. What is Optimization?
Finding the parameters which give the best value of a given metric
Example:
Parameters: position and height of each pillar
Metric: maximum displacement of the roof

6. How can it be done?

7. How can it be done?
Algorithmic Design

8. How can it be done?
Algorithmic Design
Analysis

9. How can it be done?
Algorithmic Design
Analysis
Optimization

10. How can it be done?
Algorithmic Design
Analysis
Optimization

11. How can it be done?
Algorithmic Design
Analysis
Optimization

12. The Genetic Algorithm

13. The Genetic Algorithm
Create Population

14. The Genetic Algorithm
Create Population
Evaluate Population

15. The Genetic Algorithm Create Population Evaluate Population
Select Parents

16. The Genetic Algorithm Create Population Evaluate Population
Select Parents
Create Children

17. The Genetic Algorithm Create Population Evaluate Population
Select Parents
Create Children
Mutate Population

18. The Genetic Algorithm Create Population Evaluate Population
Select Parents
Create Children
Mutate Population

19. The Genetic Algorithm Create Population Evaluate Population
Select Parents
Create Children
Mutate Population
Size? Random? Gaussian? Uniform?

20. The Genetic Algorithm Create Population Evaluate Population
Select Parents
Create Children
Mutate Population
Probability? Random? Best? Tournament?

21. The Genetic Algorithm Create Population Evaluate Population
Select Parents
Create Children
Mutate Population
Probability? One Point? Two Point? Uniform?

22. The Genetic Algorithm Create Population Evaluate Population
Select Parents
Create Children
Mutate Population
Probability? Shuffle? Gaussian? Uniform?

23. Evolutionary Principles applied to Problem Solving
using
David Rutten

24. Evolutionary Principles applied to Problem Solving
using
David Rutten
Evolutionary Algorithms are slow. Dead slow.

25. Alternatives

26. Alternatives
Direct-search

27. Alternatives
Direct-search
Model-based

28. Case Study

30. Results

35. Conclusions

36. Conclusions
Using only Genetic Algorithms is not a good choice

37. Conclusions Using only Genetic Algorithms is not a good choice
Direct-search and Model-based methods achieve better results

38. Conclusions Using only Genetic Algorithms is not a good choice
Direct-search and Model-based methods achieve better results
Try to use several algorithms, instead of just one

39. Future Work

40. Future Work Other optimization algorithms Other case studies
Optimization framework
Optimization plugin

41. Thank You! Questions?