Abstract
The Simulink environment allows rapid prototyping of complex software systems. Because many of these systems are mission-critical, it is of utmost importance to determine their input and output constraints. Determining input constraints is a trivial matter, but the constraint determination of a system's output values is a serious and challenging problem that historically has entailed an exhaustive exploration of the system's input states. The work presented in this thesis recounts and extends a research project supported by NASA whose focus was to develop a strategy to constrain the outputs of a Simulink model. Simulink models are quite similar to mathematical functions and therefore optimization algorithms can be applied to constrain the outputs. Optimizations of simple mathematical functions paved the way for random functions and finally led to the development of two optimization algorithms. During the exploration of potential optimization algorithms, strategies such as Monte Carlo, the simplex method, simulated annealing, and evolution strategy were explored. In the end, a combined approach utilizing both simulated annealing and the simplex method was compared with evolution strategy for relative strengths and weaknesses. It was determined that the evolution strategy algorithm was more suited to optimization of Simulink models due to its more effective usage of model calls and to its higher success rate.
|
