Integrating Mach Number Prediction with Outlier Detection for Wind Tunnel Systems

Abstract

Mach number prediction plays a crucial role in wind tunnel systems. Due to the complicated system behavior, many existing predictors cannot obtain the desired level of accuracy. In addition, the presence of outliers in databases further negatively influences predictive accuracy. In this paper, we address these two problems in one scheme. In contrast to robust regression models, in this paper the problems of prediction and outlier detection are considered separately but are solved by one paradigm. We propose an ensemble model as a predictor, in which a Gaussian process model is used as the base learner. The motivation for using the Gaussian process is its superiority in solving complex nonlinear regression problems. The objective of the ensemble model is to further improve the predictive accuracy of the Gaussian process model. Our outlier detection model is also based on a Gaussian process. It is composed of two complementary components; one is based on Gaussian process regression, and the other is based on Gaussian process classification. We verify our predictor and outlier detection model with three data sets from a real-world wind tunnel system. The results not only verify the model’s predictive performance but also underline the superiority of the detection model.