Adaptive Experimental Design Applied to Ergonomics Testing Procedure

Abstract

Nonlinear constrained optimization algorithms are widely utilized in artifact design. Certain algorithms also lend themselves well to design of experiments (DOE). Adaptive design refers to experimental design where determining where to sample next is influenced by information from previous experiments. We present a constrained optimization algorithm known as superEGO (a variant of the EGO algorithm of Schonlau, Welch and Jones) that is able to create adaptive designs effectively. Its ability to allow easily for a variety of sampling criteria and to incorporate constraint information accurately makes it well suited to the needs of adaptive design. The approach is demonstrated on a human reach experiment where the selection of sampling points adapts successfully to the stature and perception of the individual test subject. Results from the initial study indicate that superEGO is able to create experimental designs that yield more accurate models using fewer points than the original testing procedure.

Noisy kriging-based optimization methods: A unified implementation within the DiceOptim package

“…Adapting the efficient metamodel-assisted strategies from the deterministic case to the noisy one is not a straightforward task, and several authors from different disciplines have proposed various solutions over the past decade (Forrester et al, 2006;Huang et al, 2006;Osborne et al, 2009;Picheny et al, 2012a;Srinivas et al, 2010;Scott et al, 2011;Sasena et al, 2002;Sakata et al, 2007). All these strategies basically share the same algorithmic principles and surrogate management framework; their differences mainly appear in the "infill sampling" criteria used for choosing new points sequentially.…”

Generic Sequential Sampling for Metamodel Approximations

“…Table 1 is a summary of the criteria reviewed in these surveys, with the search scope of each criteria defined as appropriate. I;, = aqi) r, = -&z) Sasena [2002a found that the most consistently well performing technique was the switching criterion (Switch in Table 1). For specific problems, other criteria might outperform this switching criterion, but only the switching criterion was consistently amongst the best.…”

“…[Isaaks, 19891 In contrast, engineering metamodels (called Design and Analysis of Computer Experiments, DACE, Models) commonly employ a SCF similar to Equation 6. [Sasena, 2002a, andSasena, 2002bl where Bdefines the range of influence of the data (Or 0), and p defines the smoothness of the model (0 < p < 2) where increasing values of p lead to a smoother model. Both models interpolate given data points.…”

“…Sasena [2002a and 2002bl and Jin [ZOO21 include surveys of these criteria. These criteria can be classified as either global or local criteria, based on their search scope.…”

Qualitative and Quantitative Sequential Sampling

“…The Sasena [2002a; found that the most consistently and best performing technique was the switching criterion (Switch in …”