The Application of Uncertainty Quantification (UQ) and Sensitivity Analysis (SA) Methodologies to Engineering Models and Mechanical Experiments

Understanding the effects of uncertainty on modeling has seen an increased focus as engineering disciplines rely more heavily on computational modeling of complex physical processes to predict system performance and make informed engineering decisions. These computational methods often use simplified models and assumptions with models calibrated using uncertain, averaged experimental data. This commonplace method ignores the effects of uncertainty on the variation of modeling output. Qualitatively, uncertainty is the possibility of error existing from experiment to experiment, from model to model, or from experiment to model. Quantitatively, uncertainty quantification (UQ) methodologies seek to determine the how variable an engineering system is when subjected to variation in the factors that control it. Often performed in conjunction, sensitivity analysis (SA) methods seek to describe what model factor contributes the most to variation in model output. UQ and SA methodologies were employed in the analysis of the Modified Embedded Atom Method (MEAM) model for a pure aluminum, a microstructure sensitive fatigue crack growth model for polycarbonate, and the MultiStage Fatigue (MSF) model for AZ31 magnesium alloy. For the MEAM model, local uncertainty and sensitivity measures were investigated for the purpose of improving model calibrations. In polycarbonate fatigue crack growth, a Monte Carlo method is implemented in code and employed to investigate how variations in model input factors effect fatigue crack growth predictions. Lastly, in the analysis of fatigue life predictions with the MSF model for AZ31, the expected fatigue performance range due to variation in experimental parameters is investigated using both Monte Carlo Simple Random Sampling (MCSRS) methods and the estimation of first order effects indices using the Fourier Amplitude Sensitivity Test (FAST) method