An extended latin hypercube sampling approach for CAD model generation

In this paper, extended version of Latin hypercube sampling (ELHS) is proposed to obtain different design variations of a CAD model. The model is first represented by design parameters. Design constraints that are relationships between the parameters are then determined. After assigning value ranges for the design parameters, design space is formed. Each design parameter represents a dimension of this design space. Design is a point in the design space and is feasible if it satisfies the predefined design constraints. Otherwise, it is infeasible. ELHS utilizes an input design in order to obtain feasible designs.

All dimensions of the design space are divided into equal number of intervals. ELHS perform trials in design space to find feasible designs. In each trial, all the candidate designs are enumerated and one of them is selected based on a cost function. Value of the cost function is zero if all design constraints of the design are satisfied. A similarity constraint is introduced in order to eliminate designs with similar geometries. Three different CAD models are utilized for this study’s experiments in order to show the results of the ELHS algorithm.