A Novel Latin hypercube algorithm via translational propagation.

Area of Science:

• Engineering Design
• Computational Science
• Statistical Modeling

Background:

• Metamodels are crucial for analyzing complex engineering systems with expensive simulations.
• Metamodel accuracy depends heavily on experimental design, with Optimal Latin Hypercube Designs (OLHD) being common.
• The computational cost of constructing OLHD limits their widespread application.

Purpose of the Study:

• To develop a novel, computationally efficient methodology for creating Latin hypercube designs.
• To introduce the translational propagation and successive local enumeration (TPSLE) algorithm for design construction.
• To evaluate the TPSLE algorithm's performance against existing methods.

Main Methods:

• A new algorithm, TPSLE, constructs Latin hypercube designs without formal optimization.
• TPSLE utilizes an initial block and points generated by the Successive Local Enumeration (SLE) algorithm.
• The proposed algorithm's efficiency and sampling properties are compared to two established algorithms.

Main Results:

• The TPSLE algorithm demonstrates significantly reduced computation time compared to existing methods.
• TPSLE achieves acceptable space-filling and projective properties for experimental designs.
• The algorithm provides a balanced trade-off between computational efficiency and sampling performance.

Conclusions:

• The TPSLE algorithm offers a more efficient approach to constructing Latin hypercube designs.
• This method facilitates the use of metamodels in engineering design by lowering the cost of experimental design.
• TPSLE presents a practical alternative for generating high-quality experimental designs for complex simulations.