The use of a probabilistic fracture mechanics analysis to estimate reliability of fabricated structures is shown to be a valid way of identifying and ranking various quality improvement factors for a quality control programme although the limitation of the analysis (the lack of data and the assumptions which must be made) are such that accurate reliability predictions are not yet possible by such a route.
An example is given of a steel production platform jacket for Northern North Sea Service where an important possible failure mode is by fracture. By analysing the actual flaw size distribution (which can be conservatively described by a three parameter Weibull distribution) and the associated toughness distribution (which will be a normal or a log normal distribution depending upon whether the steel is operating at the upper 'shelf' or in the 'transition' range) in the critical regions values of a time independent failure integral are calculated which describe the reliability in terms of failure probability per structure-year of service. Computed reliabilities defined in this way range from 4 × 10−1 to 10−7 for different circumstances. The importance of non-destructive testing and the specification of a material with a high mean level of toughness operating above its transition temperature are quantified as the important controlling factors reducing the failure integral to a low level.