DETERMINATION OF THE POSSIBLE ERROR IN FATIGUE DAMAGE FOR A DISCRETE STRAIN HISTORY

Abstract

Abstract— The Markov matrix, which is a statistical representation of a service load history, is generally derived using 32 or more strain levels, since this is considered an optimum level between computational effort and accuracy of fatigue life prediction. A recently developed model which predicts the distribution of closed loops and fatigue life from the Markov matrix, is limited in the number of levels which can be analysed because of computational restrictions. This investigation was undertaken to determine the change in the accuracy of fatigue life predictions as the number of strain levels used to represent the strain history is reduced. Since each discretized level is used to represent a range of strain values, a given discretized history will represent a family of histories. A technique is developed which gives the distribution of damage for the family of histories given the discretized history and the distribution of peaks and valleys within each strain interval. By comparing the predicted damage caused by the measured history to a conservative estimate of the distribution of damage for the family of histories, the range of possible error is calculated and used to determine the relationship between the number of strain levels used in the discrete history and the possible error in predicted fatigue damage.