MEASUREMENT UNCERTAINTY CONTRIBUTION TO THE MLP NEURAL NETWORK LEARNING ALGORITHM APPLIED TO AERODYNAMIC EXTERNAL BALANCE CALIBRATION CURVE FITTING

Author(s)

Barbosa, I. M. ; del Moral Hernandez, E. ; Reis, M. L. C. C. ; Mello, O. A. F.

• Document type: Journal article
• Source: Metrology and Measurement Systems; v. 14, n. 1, p. 12-pg., 2007-01-01.

Abstract

The aim of this study is to fit a calibration curve to a multivariate measurement system considering uncertainty in load measurements. The experimental data are generated from the calibration of the aerodynamic external balance of the subsonic wind tunnel n.degrees 2, at the Brazilian Institute of Aeronautics and Space. To fit the calibration curve, Multilayer Perceptron Artificial Neural Networks are submitted to the learning process. Studies employing several different architectures were carried out in order to improve the MLP convergence. The uncertainty in measurements is taken into consideration, through the modification of the learning algorithm, which in its classical approach, considers the data points free from error sources. The results of both methodologies, learning algorithm endowed with and without uncertainties, are compared. The artificial neural network performance in predicting future calibration data is explored through the simulation process. (AU)

• FAPESP's process: 00/13769-0 - Development of technology for two- and three- dimensional aerodynamic trials for high performance aircraft project
• Grantee: Olympio Achilles de Faria Mello
• Support Opportunities: Research Grants - Research Partnership for Technological Innovation - PITE