Automated generation and analysis of dynamic system designs

Abstract

This research uses Genetic Algorithms (GA) to suggest new dynamic systems based on topological remapping of system constituents. The bondgraph representation of the dynamic system behavior is evolved by the operators encapsulated in the genetic algorithms to meet the specified design criteria. The resultant evolved graph is assembled by designers with schemes to produce design variants. Behavioral transformation and structural transformation are adopted as strategies to generate design variants that extend beyond the scope of parametric design into innovative design. Behavioral transformation involves changes in the structure of the representation graphs, while maintaining the functions. Structural transformation involves changes in the components and the subsystems represented by the graph fragments. GAs are used to implement the operators of the transformation to search the problem-solution space because GAs are very robust search routines. Further, since the goal is to generate many solutions, genetic speciation is used to diverge the search so as to uncover other desirable solutions. The dynamic systems are modeled using bond graphs. Bond graphs provide a unified approach to the analysis, synthesis and evaluation of dynamic engineering systems. Though the scope of this investigation is limited to systems represented by bond graphs, the domain is wide enough to include many interesting applications like pump systems and vibration isolation systems.