Extreme value risk analysis in the structural design of reactor components

Abstract

The tendency to confuse “uncertainties” associated with design assumptions and parameters and compensated by the safety factor with objective ‘risks of failure’ implicit in the design, has been characteristic of the approach to probability-based structural design on all levels. However, a clear differentiation between uncertainty and risk is required to remove the lack of correlation between design safety analysis and risk analysis implicit in the present approach to the design of major structural and mechanical components of nuclear reactors as well as other structures. In a recent paper [5] the author has used the definition of the safety factor as a random variable (distribution of a quotient) to construct a probability model that justifies the introduction of the asymptotic distributions of extreme values as the physically relevant distributions of the design parameters governing ultimate load failure on which a realistic risk assessment can be based. Realistic reliability and risk assessment of reactor components subject to fatigue and creep, i.e. design conditions that exceed in practical importance that of ultimate load failure, can be based on the use of the third asymptotic distribution of smallest values. In the case of structural components working under complicated conditions it becomes necessary to perform full-scale tests reproducing, as closely as possible, the anticipated operational and, whenever necessary, critical limiting conditions to be provided in the design as well as in the associated reliability and risk assessment. The economic necessity of keeping the number of such full-scale tests to a minimum which, in the case of larger components, is usually a single or very small number of tests, raises the problem of integration of the test results into the framework of a reliability and risk assessment.