System Uncertainty Propagation Using Automatic Differentiation

In general, the behavior of science and engineering is predicted based on nonlinear math models. Imprecise knowledge of the model parameters alters the system response from the assumed nominal model data. We propose an algorithm for generating insights into the range of variability that can be the expected due to model uncertainty. An Automatic differentiation tool builds exact partial derivative models to develop State Transition Tensor Series-based (STTS) solution for mapping initial uncertainty models into instantaneous uncertainty models. Development of nonlinear transformations for mapping an initial probability distribution function into a current probability distribution function for computing fully nonlinear statistical system properties. This also demands the inverse mapping of the series. The resulting nonlinear probability distribution function (pdf) represents a Liouiville approximation for the stochastic Fokker Planck equation. Numerical examples are presented that demonstrate the effectiveness of the proposed methodology.