In a previous paper, we discussed the margins of the design fatigue curve in ASME Boiler and Pressure Vessel Codes Section III using a probabilistic approach. Limit state functions of the design fatigue curve for carbon and low-alloy steels were proposed in order to investigate the physical implication of the design margin. In the limit state functions, four parameters, namely, fatigue data scatter, size effect, surface roughness, and applied stress were taken into consideration as random variables. Based on the limit state functions, reliability index contours of the design fatigue curve were obtained together with the partial safety factors (PSFs).
Among these partial safety factors, the fatigue strength (or life) and the applied stress were predominant. The other parameters, the PSFs for the size effect and the surface roughness were small enough and almost constant. The parameter sensitivity, however, remains unknown and whether these parameters should be treated as constants or as random variables is also not clear.
In this study, a probabilistic parameter sensitivity study on the design fatigue curve was conducted to discuss how the design fatigue curve should be. Two sensitivities were computed. One is the rate sensitivity indicating the contribution of the mean value of the parameter in the probability of failure. The other is the probabilistic sensitivity to parameter dispersion. The parameters for the size effect and the surface roughness showed lower probabilistic sensitivities. This result suggests that the parameters can be considered as constants. In general, a higher number of parameters in a probabilistic model leads to more uncertainty and the design concept tends to be more conservative. We, therefore, proposed to deal with the parameters of the size effect and the surface roughness separately in the design fatigue curve to eliminate conservativeness.