The work of this thesis was to investigate the probability distributions of test data taken for aluminum 7050-T7451, and to attempt to develop a probability based model from the variation of the 4 fatigue life constants ( , f ̤, ' f ,Ì b,c). The goal was to create strain-life-probability curves that would more accurately describe the likelihood of failure at a given strain amplitude. The investigator determined that the test data did not demonstrate any consistent known probability density function. The investigator cautioned against assuming a normal distribution before it could be completely established as the predominate probability density function. Possible consequences of invalid assumptions were presented. Attempts were made to explain the disparity of sample data between two different laboratories testing of the same material. Assuming random behavior within an established range, probability based models were developed using the 4 strain- life constants. It was determined that in order to create a complete probability based model, an accurate regression of the test data must fit all strain levels to include the intermediate strain level's \"knee\". In an attempt to solve that problem, 8 parameter equations were explored. Methods to predict the 8 parameters included random number simulation combined with non- least squares curve fits, evolutionary algorithms and genetic algorithms.

Subjects: Metals; Fatigue; Strain hardening; Metal fatigue; Variable strain-life; Coffin-Manson; Monte Carlo simulation; Service life extension; Evolutionary algorithms; Genetic algorithms; Aluminum 7050-T7451; Probabilistic investigation