Model Identification for Engineering Variables

Presented herein are objective methods for determining the preferred statistical model to represent an engineering design variable. Given test data, two different criteria are proposed to indicate which of several competing statistical models, e.g., normal, lognormal, Weibull, EVD, etc., provides the best description of the variable. Using a Monte Carlo scheme, it was demonstrated that the proposed criteria is capable of a relatively high degree of discrimination with the power increasing with large samples. The results are useful in those applications of probabilistic design procedures where it is necessary to establish the best statistical description of stress or strength.