Standard validation methods do not take into account the variability of either the marginal conditions (climate) or the material parameters; they also greatly simplify the actual transportation and storage mechanisms. Although more precise procedures already exist, based on the finite elements or finite differences methods, the consideration of varying marginal conditions and the variable material parameters for statistical forecasts necessitates a high processing outlay. A new method of approach in which important functions are automated has therefore been adopted here in order to simplify or, indeed, first enable examination of the areas of variation created by material parameters, marginal conditions and transport mechanisms during the assessment of the reliability of wooden construction components.
Using TUN, a programme developed at the WKI, numerical simulations were carried out in order to calculate multi-dimensional transient temperature and moisture-transportation processes in components; this was validated through multiple comparisons with weathering experiments. During the course of the project, auxiliary programmes (tools) were developed, which extensively enable the automated planning and evaluation of parameter variations. At the stage of development attained by the tools during the project, a maximum of 20 parameters in up to 7 discrete steps respectively, or as a normal distribution, could be varied simultaneously. The histogram of the automatically-generated normal distribution of a parameter is shown in Fig. 1. In this way, the natural distribution of any parameter can be modelled through a discrete or normally-distributed scatter. The examination of an area of variation, however, necessitates the calculation of a very large number of variants. For example, the variation of just 7 parameters in 5 steps respectively results in 78,125 variants needing to be calculated. For variation with normal distribution, the number of possibilities is theoretically infinite. The limits of feasibility are determined here by the necessary computing effort. The quantity of parameter combinations to be calculated must therefore be limited; this can be effected through concentrating on the decisive parameters and implementing statistical test planning (DoE). Sensitivity analysis proved itself to be particularly suitable for filtering the parameters; it not only shows the strength of a parameter influence, but also its function curve (see Fig. 2). Test planning is carried out using either the random functions, which are integrated into the tools, or through DoE programmes. Comparisons between full-factorial and reduced test plans demonstrate a high level of congruity. The results of a full-factorial computation are shown in Fig. 3. When compared to the results from reduced test plans, it becomes obvious that the full-factorial range of results is comprehensively covered by the reduced plans. All the calculation results can be further evaluated using the tools and the DoE programmes.