Application of computers in fatigue analysis
The availability of minicomputers and microprocessors, at a reasonable cost, has provided a significant stimulus in a critical appraisal of fatigue testing and analysis methods. This thesis reviews and extends some of the recent fatigue analysis methods. Two major areas investigated in detail are cycle counting methods and methods for prediction of fatigue life to crack initiation. The three recent counting methods, range-pair, Wetzel's and rainflow, which avoid the distortion and inaccuracy from which the traditional cycle counting methods suffer, are described and compared with each other to find out the similarities and differences between them. It is shown that if a service loading history starts and ends at an extreme peak, then all the three methods give an identical count. All relevant methods for the description of measured service histories are reviewed critically in connection with fatigue life assessment, service history regeneration and simulation. Confidence in the rainflow method for better fatigue life predictions and increased use of analytical methods like Finite Element analysis offering frequency domain information about a component have initiated a search for a link between rainflow counting and the power spectral density of a stationary and ergodic random process. Using a Monte Carlo approach and digital simulation techniques, the thesis presents a link in the shape of a closed-form expression which defines the probability density function of rainflow counted ranges for any given power spectral density. A closed-form expression for the distribution of ordinary ranges is also presented. Methods of predicting fatigue crack initiation life under variable amplitude loading are reviewed. From the basic ingredients of the local-strain approach, various life prediction procedures are assembled methodically with regard to how the local stress and strain are determined for a given load level, how the local stress and strain are linked to the life, and how the mean stress effect is accounted for. Predictions made by these methods are compared with the published test data; however predictions are compared mostly within themselves in order to highlight the differences between methods. It is shown that under certain circumstances, some methods give very erroneous results. A sensitivity analysis is carried out to examine how sensitive various methods are to changes in the material properties. A new procedure of determining the material properties from the experimental data is proposed.