Reliability analysis of timber structures

As part of the European harmonisation of building codes, the determination of design values for loads and materials is important and is the motivation for this research. This report begins with a summary of the probabilistic basis of Eurocodes, analyses the strength distributions of wooden materials, demonstrates the effects of different distribution functions on the calculated safety level and shows some results of the applications of reliability analysis. When the number of experiments allows, determination of the 5% fractile of strength should be based on the function fitting on the lower tail of the strength values, for instance 10%. All smooth functions fitted to tail data gave good estimates of the 5% fractile. When the 5% fractile was determined from a function fitted to all data, up to 5% error occurred (in one case 9%) when compared to a non-parametric estimate. Three-parameter Weibull distribution gave, in all calculated cases, the 5% fractile within an accuracy of ±3%. The result of structural reliability analysis depends strongly on the load and strength distribution types used. When fitted functions are used in reliability analysis, it is essential that the fit is good in the lower tail area, the lowest values being most important. When fitted to the same data, a two-parametric Weibull distribution gives the most pessimistic prediction for the tail, with a normal distribution being next, and lognormal and three-parameter Weibull being the most optimistic. In an example, a two-parameter Weibull gave a failure probability 10 times higher than that of a three-parameter Weibull. The analysis suggests that gM = 1.2 to 1.3 is reasonable for timber structures when gG = 1.2 and gQ = 1.5.