Dimitri Papadimtriou "Analytical methods in system analysis"
18.9.2014, 2:00 pm, FMI 03.05.033 (MI-Building, Campus Garching), this talk will be held in English Abstract: Predictive simulations provide insight into the physics that drive complex systems enabling identification of key mechanisms and allows for rigorous and robust optimization/design. However, predictive simulation requires careful assessment of all sources of numerical errors (discretization, resolution, etc.) and uncertainty (sampling, model parameters and equations, etc.) in particular, when aleatory (inherent to the variability present in system and its surroundings) and irreducible. Uncertainty propagation methods quantifies uncertainties in system output from all sources of input uncertainty by means of computational model (forward uncertainty propagation), determines which uncertainties contribute the most to output uncertainties (sensitivity analysis) and characterizes uncertainty. In turn, uncertainty quantification assists in model prediction (interpolation, extrapolation, etc.) and thus decision making. In this context, validation (process of determining the degree to which a model is an accurate representation of the real world for) of computational models for establishing confidence in model predictions plays a fundamental role. While predictive methods are mostly based on sampling, this talk investigates the use of analytical methods (such as reliability methods) to compute the probability of failure of structural systems.