Rapid Design Space Exploration with Constraint Programming

Abstract

Sample-efficient design space exploration (DSE) of complex CPS architectures remains a key challenge for identifying optimal configurations of components, design parameters and architectural choices. Detailed executable models require significant investment to build and are typically slow to evaluate. On the other hand, high-level conceptual models may lack the exactness or accuracy to evaluate and compare. In this paper we propose a constraint-based approach for capturing the design space and a vectorized, iterative solver for rapidly discovering Pareto-optimal design points. The paper describes the constraint-based modeling approach and developed tools through a concrete design optimization problem of unmanned underwater vehicles.