Surrogate Modeling Based Uncertainty Quantification In Subject Matter Expert Systems

Abstract

The future performance of yet-to-be designed technologies is inherently uncertain. Absence of any historical data to capture the impacts of these technologies on the system necessitates the elicitation of subject matter expert knowledge, application of statistical methods and selection of suitable techniques for quantification of uncertainties in the process. Estimates of the appropriate parameter settings often come from disciplinary experts, who may disagree with each other because of varying experience and backgrounds. It is observed that in case of such lack of knowledge about the problem, epistemic uncertainty is most suitable representation of the process. Epistemic uncertainty can be quantified by many techniques. In present study it is proposed that interval analysis and DempsterShafer theory of evidence are better suited for quantification of epistemic uncertainty in technology valuation process. One of the problems faced in applying these non-deterministic methods is the heavy computational effort that is required to perform such analyses. Surrogate models can help alleviate this issue. The surrogate models enable fast execution of disciplinary codes and hence act as an enabler for uncertainty quantification techniques that are inherently expensive as far computational efforts are concerned. Proposed technique seeks to offset some of the problems faced by using deterministic or traditional probabilistic approaches for uncertainty propagation. A detailed modeling and simulation environment for gas turbine is setup. Latin hypercube DoE was used to sample the design space of a gas turbine. A surrogate model was created using a neural network. Sensitivity analysis is then carried out. The inputs for the M&S environment are selected based on this sensitivity analysis. The results generated from proposed technique provide a comprehensive understanding on the effects of uncertainty on the problem. NomenclatureBPA = basic probability assignment DOE = design of experiments DR = Dempster's rule ET = Evidence Theory M&S = modeling and simulation RSE = response surface equation TCM = technology compatibility matrix TIF = technology impact factor TIM = technology impact matrix TSM = technology synergy matrix 1 Associate Professor, Downloaded by KUNGLIGA TEKNISKA HOGSKOLEN KTH on July 29, 2015 | http://arc.aiaa.org |