Allocation of Samples Between Exploration and Replication for Open-Hole-Tension Test

Abstract

It is possible to test structural elements for a matrix of dimensions, stacking sequences and processing parameters, and then to interpolate or extrapolate using surrogate models to predict the failure at untested points. Extrapolation suffers larger uncertainty than interpolation especially when failure mode changes beyond the window of experiments. With a given budget for tests, there is a tradeoff between exploration of different configurations and replication to minimize the adverse effects of noise. For simulated tests, exploration has been shown to be more effective based on the assumption that noise in data is normally distributed and uncorrelated. This paper studies experimental data from a series of Open-Hole-Tension tests to check whether the conclusion stands with real experimental data. The tests are intended to investigate the impact of the ratio between coupon width and hole diameter on failure strength of a composite panel. The failure mode and strength are evaluated at four structural configurations. We first examine the statistical distribution of failure strengths and correlation among them from same pre-preg batch. Experimental results show that the noise distribution is closer to Gaussian than to Weibull or uniform distributions. The failure strengths from same batch tend to show systematic bias. The surrogate model selection is justified based on the experimental data, and the performance of two sample allocation strategies, exploration or replication, is compared. Sample selections emphasizing exploration proved to be more favorable for prediction accuracy.