Use of random sampling and Bayesian methods in the prediction of chestnut timber beams bending stiffness

Abstract

The assessment of global bending stiffness of three old timber beams is predicted by use of random sampling and Bayesian Network approaches. The information for the prediction models derive from the visual grading and localized mechanical testing of 20 beams retrieved from the same building, which have been scope of analysis in previous studies. Although the results indicate moderate correlations between predicted and experimental results, significant percentage errors are also found. To minimize the percentage error between experimental and predicted values, coefficients corresponding to the scale effect relation between existing chestnut timber elements with round cross section and sawn beams of smaller dimensions with rectangular cross section were proposed.