Total least squares versus RBF neural networks in static calibration of transducers

The problem of static calibration of measurement channels is considered under an assumption that the raw result of measurement depends both on a scalar measurand and on a scalar influence quantity. The methodology of calibration based on the use of cubic B-splines for total-least-squares approximation of the forward static characteristics of measurements channels and on the use of radial-basis-function neural networks for approximation of the inverse static characteristics is developed and examined using synthetic data representing a measurement channel with a fibre-optic sensor. Two algorithms of calibration are compared. The accuracy of measurements, based on the results of calibration, is used as the main criterion of comparison. Some conclusions are formulated concerning the properties of the compared algorithms of calibration