An uncertainty quantification method for nanomaterial prediction models

This paper presents a new method for quantifying uncertainty in the predictions of a nanomaterial computational model to account for variability in the constituent nanostructure properties and characterization measurements. The stiffness of a buckypaper-polymer composite is predicted using a micromechanics model. The model requires from the user as inputs the nanostructure properties, including the diameter, length, and curvature distribution of the carbon nanotubes which shows large variability. The current characterization techniques used to describe these dimensions are subject to considerable measurement error. We propose a constrained nonlinear programming approach for quantification of raw material variability and its impact on the property prediction of buckypaper-polymer composites. The uncertainty quantification method is useful for decision making to predict probability that the quality characteristic of the final part will satisfy design constraints. A case study based on data from a real buckypaper manufacturing process was used to illustrate the approach. It is shown that modeling the correlation between nanostructure properties using a multivariate distribution rather than independent univariate distributions is important to accurately quantify the effect of these properties on the final-part property.