What We Do and Our Approach: Engineering Design Support, Technical Evaluations and Analysis

TR4

Technical Review 4 - Weld Durability Analysis

An evaluation of differing weld fatigue lifing methods applied to FE models of rail vehicles was carried out. Four methods were considered: nominal stress methods to BS7608 (Steel) and BS8118 (aluminium), the Hot Spot Method and Eurocode 9. An automated method was also assessed, which calculates cumulative damage from multiple loadcases (using Miner's Law) at each weld node in an FE model, whilst accounting for the effect of principal stress direction on weld class.

Various standards exist for assessment of weld life but these are not generally designed for use with FEA. The British Standards mentioned in particular were designed to be used in hand calculations with a nominal stress, i.e. the combined axial and bending stress in the section, ignoring the local stress concentration effects of the weld feature. These standards are thus difficult to apply to an FE model because it is not easy to determine where the nominal stresses can be taken from in the model. This ambiguity usually dictates that a stress at or near the weld joint is taken, which can give an overly conservative result.

Another method, called the Hot Spot Method, addresses these issues by prescribing what level of mesh refinement should be used around the weld in any FE model, as well as what details of the weld should be modelled, if any. This requires a set of new life curves but should give increased accuracy over the nominal methods described above.

Aside from the issue of determining the working stress from an FE model, these methods also require the weld to be classified, before an allowable fatigue stress can be obtained. Classifying welds from British Standards requires experience and is often contentious.

Weld classification is based on a consideration of both the weld geometry and whether the stress is parallel or normal to the weld. It is also dependant on which member is being considered at the welded joint. This can make finding the lowest life weld an iterative process when for example, the highest stress may occur in one member but a lower stress in another may have a lower class for the weld and therefore, a lower life.

Doing this entirely manually on large models often results in error or uncertainty and the pre-assignment of weld classes to parts of a standard model and the introduction of automatic processes are generally beneficial.