LIP6 2002/002
- Thesis
Découverte automatique de circuits en électronique de puissance - B. Robisson
- 211 pages - 09/21/2001- document en - http://www.lip6.fr/lip6/reports/2002/lip6.2002.002.pdf - 1,255 Ko
- Contact : Bruno.Robisson (at) nulllip6.fr
- Ancien Thème : APA
- Keywords : Scientific discovery, abstraction, heuristic search, power electronics, qualitative simulation, bond graphs
- Publisher : Ghislaine.Mary (at) nulllip6.fr
The discovery of power electronic circuits is a task which involves large domain knowledge and which requires ability and creativity. This thesis aims to create a computer system which simulates this task. A system composed of two modules has been implemented.
The first module generates the circuits without rough design defect. This module realizes a heuristic search in a hierarchy of abstractions. This knowledge organization is used in order to focus, as soon as possible, the search towards the more interesting circuits. To generate circuits without redundancy, we have used the bond graph representation. We have also proposed a new algorithm which, contrary to existing ones, transforms a circuit into a bond graph fully adapted to the formalism.
The second module analyses qualitatively the functioning of the generated circuits. It describes this functioning in the same terms than those which are used by engineers. So, it can be used to express given specification easily and then to select the circuits which verify this specification.
This general approach is successful for discovering circuits because the system generated the representatives of important classes of circuits. It has proposed new circuits which seems to be as efficient as circuits that are already known. The proposed system is an original tool for the rational study of power electronics circuits. More generally, it proves the efficiency of artificial intelligence methods on a technical domain and points out the interest of abstraction for problem solving.
The first module generates the circuits without rough design defect. This module realizes a heuristic search in a hierarchy of abstractions. This knowledge organization is used in order to focus, as soon as possible, the search towards the more interesting circuits. To generate circuits without redundancy, we have used the bond graph representation. We have also proposed a new algorithm which, contrary to existing ones, transforms a circuit into a bond graph fully adapted to the formalism.
The second module analyses qualitatively the functioning of the generated circuits. It describes this functioning in the same terms than those which are used by engineers. So, it can be used to express given specification easily and then to select the circuits which verify this specification.
This general approach is successful for discovering circuits because the system generated the representatives of important classes of circuits. It has proposed new circuits which seems to be as efficient as circuits that are already known. The proposed system is an original tool for the rational study of power electronics circuits. More generally, it proves the efficiency of artificial intelligence methods on a technical domain and points out the interest of abstraction for problem solving.