Optimization and implementation of a software code for the automated analysis of fatigue data

Material fatigue is the most common cause of failures of components and must be properly considered during the design process. Fatigue is a highly structure-sensitive process, depending mainly on the material characteristics at sub microscopic and microscopic scale. If, in fact, samples of the same size and material are subjected to the same loading conditions, different results can be obtained. This induces a statistical variability of the material fatigue strength. Therefore, data obtained from fatigue tests should be analysed using statistical methodologies. The purpose of a statistical analysis of fatigue test results is to estimate the fatigue properties of the material, such as fatigue limit, or fatigue strength at a given number of cycles, S-N curve, etc., from the test data. The least squares method is generally used for assessing the parameters of the S-N curve, in the finite life range, while the Staircase method is used for assessing the fatigue strength in the infinite life range of the Wӧhler curve. However, with the least squares method run-outs data (i.e. tests that are interrupted before failure) are not considered. Furthermore, although the Staircase method considers both runout and failure data, it does not consider the actual number of cycles to failure but only whether the specimen fails or does not fail. Other methods for estimating the fatigue S-N curve permit to considers both failures and runout, e.g., methods based on the Maximum Likelihood Principle. The material parameters that maximize the Maximum Likelihood function are the best estimates for the considered S-N curves. This method permits the estimation of the parameters used in a unified model that is able to shape the complete SN curve with a single equation. The goal of this thesis is to develop and optimise a user-friendly software implementing a complete model, whose parameters are estimated with the principle of maximum likelihood, for the statistical analysis of fatigue data. This software has been validated on several experimental datasets related to different types of materials and the results obtained were compared with those obtained with traditional approaches to highlight the differences between the methods. In addition, since the principle of maximum likelihood maximises the information contained in the experimental dataset, the possibility of reducing the number of tests to be carried out and thus reducing the time and costs of a characterisation campaign was also investigated. In order to validate the obtained results, virtual experimental datasets have been randomly simulated. In this way it was possible to verify how much the estimated parameters, computed with both methods, deviate from the real ones, and thus investigate which method was more accurate in estimating the fatigue parameters.