Hasitha Jayaneththi Koralalage
Title: Three methods of generating space-filling designs: a comparison
Date: July 21st, 2025
Time: 10:30am
Location: LIB 2020
Supervised by: Boxin Tang
Abstract:
The construction of space-filling designs has introduced a new direction in design research. In this study, we evaluate the performance of three design construction methods. While one method simply uses random balanced designs (RBDs), the other two methods are based on orthogonal arrays (OAs) and strong orthogonal arrays (SOAs), respectively. Our focus is on 18-run and 36-run designs with six levels. The results demonstrate that the OA-based approach consistently produces evenly dispersed design points, outperforming the other two methods notably under the criteria of maximin distance and centered L2 discrepancy, especially in higher dimensional designs. The SOA approach consistently delivers low variance in inter-point distances, indicating its strength in achieving uniform coverage of the design space, particularly advantageous in higher dimensional settings. In contrast, the RBD approach consistently yields inferior performance, with relatively higher correlation values and discrepancies, and lower maximin distances. These findings highlight the advantages of orthogonal arrays and strong orthogonal arrays in maintaining space-filling and uniformity, particularly important in high dimensional experimental settings.
Keywords: space-filling design, OA-based design, strong orthogonal array, design comparison, design selection criteria