Skip to Main Content
This paper presents an artificial neural network (ANN) based data integration method for a four-path ultrasonic flowmeter to improve its performance under complex flow profiles. Computational fluid dynamics (CFD) has been used to obtain the flow profiles inside a single elbow and an out-plane double-elbow pipeline, respectively, to extract the flow velocities on different individual sound paths and the corresponding mean flow velocities on the cross-sections located at 5 and 10 times pipe diameter downstream the elbow. The results from the CFD simulation for Reynolds number in the range from 3.25×103 to 3.25×105 were used to construct the data set. A three-layer ANN was designed, in which the flow velocities on individual sound paths and the cross-sectional mean flow velocity are taken as the input and output, respectively. Part of the data set is used to train the ANN. The other part of the data set is used to test the feasibility of the ANN. It was found that the error of the estimated cross-sectional mean flow velocity based on the ANN is within ±0.3% without the requirement of any flow conditioner for Reynolds number in the range from 3.25×103 to 3.25×105, which is significantly better than the results from the traditional weighted integration method. The proposed ANN based data integration method is of extending the limitation of straight pipe length for the installation of multi-path ultrasonic flowmeter and promoting its practical applications under complex flow profiles.