Integrating Diverse Approaches to the Reliability of Engineering Structures: Low Temperature Performance of Pressure Piping in the as-Welded Condition
TWI Industrial Member Report 1168-2023 [pdf / 813 KB]
By Isabel Hadley, Charles Schneider & Yin Jin Janin
Industrial Need
The probability of failure, POF (the converse of which is ‘reliability’) of engineering components and structures depends on numerous factors, including design, manufacture, inspection, operation and human factors. Reliability may be calculated on the basis of direct observation (what proportion of components fail prematurely?) or theoretical methods (what failure mechanisms are possible, and how likely is it that the conditions for failure will be present during operation?), or some mixture of the two. In this study, two methods are used to estimate probability of failure, using approaches drawn from a range of standards, and comparing the outcomes with situations related to other standards. Relevant documents include BS 7910 (BSI, 2019), R6 (EDF Energy, 2001), API 581 (API, 2016), PD 5500 (BSI, 2018) and EN 13445 (BSI, 2014), and ASME B31.3 (ASME, 2018):
- A ‘top-down’ method based on a mixture of failure statistics and engineering judgement.
- A ‘bottom-up’ method based on fracture mechanics/engineering critical assessment (ECA). Both ‘deterministic’ (using a single characteristic value of each variable) and probabilistic approaches were used, the latter allowing direct estimation of probability of failure.
Key Findings
- As expected, POF falls when progressing from Level 1 treatment of residual stress to Level 2, reflecting the difference between these two assumptions. Level 2 residual stresses in the outer quarter of the wall (corresponding to a 3mm high flaw in a section thickness of 12.7mm) are tensile but relatively low (an average of around 0.28σY over the face of the flaw) compared with 1.0σY using the Level 1 approach. This leads to a reduction in POF by a factor of around 24.
- Third assessment reflects both a distribution of flaw sizes and the assumption that there will be on average only 0.01 flaws per weld. Consequently, POF has fallen by about two orders of magnitude relative to previous case.
- When flaw distribution is taken into account, the estimate POF appears close to that predicted using the RBI approach.
- As more probabilistic variables are added to the PFM model, so the POF falls, reflecting increasingly realistic treatment of the variables.
Welding residual stress profiles (transverse to weld) for pipe girth welds (from Annex Q of BS 7910). The flaw location for Case 3 is also indicated. HI=heat input
Results of PFM and RBI calculations.