Statistical distributions and their influence on the material property values of tropical timber: Case study of Gmelina arborea

Abstract

Previous studies as well as reference standards report specific statistical distributions to be used for determining the material property values of specific engineering materials. In many cases, data collected from the laboratory do not match these statistical distributions thus, leaving a discrepancy between experimental data and code recommendations. Although previous studies have investigated the characterization of engineering materials, those dealing with this issue are lacking. This study investigates the influence of various statistical distributions on the material properties of timber, using Gmelina wood as a case study. The aim of the study is to discuss from experimental data, the difference / similarities in the material properties of Gmelina wood obtained using different statistical distributions and to discuss a possible unification between the varying statistical distributions reported in the literature for determining the actual material properties of construction timber. To achieve these, experimental data were collected from mechanical tests conducted on Gmelina wood, and the statistical values of the material properties were determined in accordance with the normal, lognormal and the Weibull distributions. For each dataset, the 5th percentile, mean, standard deviation and the coefficient of variation were determined. The results showed that at a coefficient of variation of less than 22%, all the material property values computed from the different statistical distributions were at least 92% in agreement with each other. At a coefficient of variation of 28% however, the lognormal distribution over-estimated the 5th percentile value by an amount equal to 18% and 20% with respect to the normal and the Weibull distributions respectively. The 5th percentile values obtained from the Weibull and normal distributions however, stayed within a 2% difference and the mean values for all cases were uninfluenced. It was determined that below a coefficient of variation of 22%, any of the statistical distributions could be used to determine the material properties of Gmelina wood for structural applications with a 92–100% accuracy.