Neural network approach to ultrasonic flow measurements

Mathematical models and numerical methods offer a flexible tool to investigate flow disturbance effects on flowmeters of different types. In this paper a simple neural network based approach has been used to study the velocity profile dependence of ultrasonic flowmeters. Neural networks have been used in two ways: to interpolate the velocity profiles in the points needed for the modelling of ultrasonic flow measurement, and to compute the weights for different paths of multipath ultrasonic flowmeters. In the former case two types of neural networks, multilayer perceptron networks and radial basis function networks, have been investigated. In the latter case, a single layer neural network with linear neurons is first trained with known velocity profiles, and the weights determined by the network have then been used in the computation of the errors in other piping configurations. The results have been compared with the errors computed with the weights for different paths given in Pannel CN, Evans WAB, Jackson DA. A new integration technique for flowmeters with chordal paths, Flow Measurement and Instrumentation 1990;1:216–224.