Estimation of the characteristic tensile strength of the wood in the parallel direction to the grains through of probability models

ABSTRACT For a structural design of wood within of the safety requirements, the strength for a certain mechanical request is calculated based on its characteristic value, corresponding to the 5% percentile of a given probability distribution model. The Brazilian standard NBR 7190 [1] establishes, for small samples, relations to estimate the characteristic strength of the wood, which can result in a significant difference of the values obtained when compared to a suitable probabilistic model. Considering the results of the tensile strength (fto,k) in the direction parallel to the grain for 40 different hardwood species, with 480 experimental determinations with standard moisture content (≈ 12%), were used distribution density functions (Normal, LogNormal, Weibull and Exponential) for each species, in order to determine the highest adhesion model. Once determined the values and respective models for each species, a multivariable linear regression model, based on analysis of variance (ANOVA) and dependent on the mean value (x), coefficient of variation (CV%) and lowest (min) and the highest (max) strength value for each species, is adopted to estimate the fto,kadjusted with the most significant terms, in order to infer the quality of the estimator and, consequently, the reliability of such mechanical property. Finally, the regression model proposed here is compared to the proposition of the Brazilian standard, in order to evaluate the reliability of the model and its adequacy in the estimation of the tensile strength fto,kin the direction parallel to the grain.