Information is presented on the magnitude of errors associated with various sampling simulation schemes of the distribution of three different populations, representing actual bending strength af dimension lumber. Errors were determined between the simulated and actual distributions. Graphical evaluations indicated good fits with the three-parameter form of the Weibull distribution for both original and simulated bending strength data, as well as with the resulting error terms. Error terms, based on the simulated versus actual distributions, were generated for the lower 5 % exclusion limit, for the 50 % exclusion limit and for the entire distribution curve. Simulations were carried out with the aid of Monte Carlo techniques using distribution functions fitted to actual test data for dimension lumber. The errors are expressed as functions of confidence levels. The comparison of the erro: obtained through the various sampling schemes could provide some initial directions to choose an economical sampling plan for the presently ongoing in-grade lumber testing program.
Objective
The objective of this work is to evaluate the relative merits of various sampling schemes in order to predict the bending strength distribution for possible use with the in-grade lumber testing program. This program is developed by the lumber industry in the United States to evaluate the mechanical properties of lumber within a grade. The use of actual test data, descriptions by three-parameter Weibull distributions and Monte Carlo simulations are emphasized over theoretical derivations. Relative magnitude of errors and efficiencies are emphasized over exact distribution types and errors associated with the various sampling schemes.