A machine learning approach for the inverse quantification of set-theoretical uncertainty

This paper introduces a machine learning approach for the inverse quantification of set-theoretical uncertainty. Inverse uncertainty quantification (e.g., following Bayesian or interval methodologies) is usually obtained following a process where a distance metric between a set of predicted and measured model responses is iteratively minimized. Consequently, the corresponding computational effort is large and usually unpredictable. Furthermore, often only a limited dataset is available, further complicating the inverse procedure. To overcome these issues, a machine learning approach is proposed to predict the uncertainty in selected model parameters given a limited dataset comprising measured responses. To achieve this, machine learning is applied to train a Neural Network that is able to predict model parameter uncertainty, presented a limited set of measured responses, following a set-theoretical approach. This Neural Network is trained by means of a numerically generated data set that captures typical uncertainty in the model parameters. Also, the application of dimension-reduction techniques to aid this inverse quantification are studied. The developed method is applied to the well-known DLR AIRMOD test structure and the results are compared to literature data.

Jaar van publicatie:2019