Support Vector Regression-Based Data Integration Method for Multipath Ultrasonic Flowmeter

This paper presents a support vector regression (SVR)-based data integration method for a 4-path ultrasonic flowmeter, which is able to estimate accurately the mean cross-sectional flow velocity under complex flow profiles. While installed in the pipeline with complex configurations, such as single-elbow or out-plane double-elbow, the performance of multipath ultrasonic flowmeter will degenerate due to the strong nonlinear relationships between the flow velocities on different individual sound paths and the mean flow velocity on the cross section, particularly when the straight pipe length is not guaranteed. The presented SVR-based method is of an excellent nonlinear mapping and generalization ability. The cases while the Reynolds number in the range of 3.25 × 10³ - 3.25 × 10⁵ were simulated using computational fluid dynamics and the flow profiles located on the cross sections of 5 and 10 times pipe diameter downstream a single elbow and an out-plane doubleelbow were extracted to construct the data set for SVR training and test. It is found that the error of the estimated crosssectional mean flow velocity obtained by the SVR-based data integration method is within ±0.5% without the requirement of a flow conditioner, which is significantly better than the results from the traditional integration method with constant weights. The presented SVR-based data integration method is helpful to extend the limitation of straight pipe length for the installation of multipath ultrasonic flowmeter, which is attractive for the practical applications of multipath ultrasonic flowmeter.