Luigino Benetazzo, Matteo Bertocco, Claudio Narduzzi
AD-HOC MEASUREMENT NETWORKS

This discussion paper presents issues in the design of ad-hoc, networked measurement systems. Key requisites are summarized at first. Then, architectural features will be analyzed in order to let a designer be aware of advantages and efforts of presented choices.

Marja Lähdekorpi, Heimo Ihalainen, Risto Ritala
USING IMAGE REGISTRATION AND ALIGNMENT TO COMPARE ALTERNATIVE 2D MEASUREMENTS

We present in this paper a novel method for comparing alternative 2D measurements. The method is based on a new image registration algorithm developed for the automatic registration and alignment of randomly textured image data. Our aim in the algorithm development has been to enable fast registration of the measured 2D property maps without the need for special registration marks. To improve robustness, the maps are registered in two steps; the first step exhibits plain translation and the second phase iteratively refines the transformation estimate. Sub-pixel registration accuracy is achieved. Several experiments have been conducted showing that the algorithm is able to register various 2D property maps successfully. After the registration we align the alternative 2D measurements. This enables the comparison and statistical joint analysis of several 2D property maps measured from the same target area. The high amount of independently measured data points in the property maps provides a firm statistical ground for conclusions. We have used the new registration algorithm to align various 2D surface profile measurements of paper and board. The analysis of the aligned measurements has confirmed the feasibility of our registration method and revealed fundamental differences between the measurement devices.

Silvio B. Melo, Carlos C. Dantas, Emerson A. de O. Lima, Francisco P. M. Simões, Eric F. de Oliveira,Valdemir. A. dos Santos
RECONSTRUCTION OF RADIAL CATALYST CONCENTRATION DISTRIBUTION IN AN EXPERIMENTAL TYPE FCC RISER

A gamma ray tomographic reconstruction of catalyst concentration distribution in laboratory experiments is presented. The images are generated by gamma transition measurements in an experimental riser simulating flow experiments. The manual scanner is adequate for mapping radial catalyst concentration distribution under spatial resolution of 0.002 m and density resolution of 10 kg/m³.

Karl H. Ruhm
MODELLING OF NON-IDEAL MEASURING PROCESSES – PREREQUISITE FOR ERROR AND UNCERTAINTY ANALYSIS

The determination of the uncertainty of measurement results is one of the most demanding but unpopular tasks in metrology. One of several reasons for this fact is the lack of well-structured models of non-ideal measurement processes on one hand and of qualification processes, which determine measurement uncertainties, on the other hand. Easily available and universally valid models would greatly help to improve measurement quality. But since process modelling is very laborious, practitioners are additionally confronted with cost-benefit questions. This paper starts with the most popular but rather incomplete error model and, on the basis of Signal and System Theory, derives a quantitative structure of the non-ideal measurement process. This structure is comprehensive and open for further and more challenging extensions.

I. M. Barbosa, E. del Moral Hernandez, M. L. C. C. Reis, O. A. F. Mello
MEASUREMENT UNCERTAINTY CONTRIBUTION TO THE CALIBRATION CURVE FITTING OF AN AERODYNAMIC EXTERNAL BALANCE USING MLP ARTIFICIAL NEURAL NETWORK

The aim of this study is to fit a calibration curve to a multivariate system. The experimental data are generated from the calibration of the aerodynamic external balance of the subsonic wind n.º 2, the TA-2, of the Brazilian Aerospace Institute, IAE. Multilayer Perceptrons (MLPs) Artificial Neural Networks are employed. To fit the calibration curve, the MLPs are submitted to the learning process. The measurement uncertainties are taken into consideration, through the modification of the MLP learning algorithm, which in its classical approach, considers the data points free from error sources. The results of both methodologies, learning algorithm endowed or without uncertainties, are compared.