A model for multiphase flow velocity calculation in pipelines based on ultrasonic sensors

Abstract

In the petroleum and natural gas industry, a wide variety of multiphase fluids are prevalent, and precise measurement of their flow velocity in pipelines holds significant importance for different stages of well drilling and construction. However, due to the presence of large solid particles and the corrosive nature of the liquid phase in multiphase fluids within the petroleum industry, invasive measurement methods struggle to maintain long-term acceptable detection accuracy. Therefore, the non-contact fluid flow velocity measurement method based on ultrasonic sensors exhibits substantial research value. Nonetheless, when employing this approach for pipeline multiphase fluid flow velocity measurement, the abundance of background interference noise at the site poses challenges in Doppler echo signal reconstruction and results in lower precision for frequency shift extraction, leading to considerable errors in flow velocity calculation results. To address this issue, the present study utilizes a transmit-receive separated continuous wave ultrasonic sensor. First, a mathematical model is developed for the superimposed signal of ultrasonic Doppler echoes within the pipeline. Next, a novel signal reconstruction method is proposed by employing Chebyshev polynomials for interpolation computation of the sampled discrete signals. Subsequently, a Doppler shift model is introduced, leading to the formulation of a new model for multiphase flow velocity calculation in pipelines based on ultrasonic sensors. Finally, a comparison experiment for full-pipe multiphase flow velocity detection is conducted to validate the computational performance of the new model. The experimental results show that, compared with the FFT model and the conventional cross correlation model, the comprehensive meter factor of the ultrasonic flow measurement system with the new model is reduced by 0.024 445, the accuracy is reduced by 2.98%, the nonlinear error is reduced by 2.4405%, the average relative error is reduced by 0.646%, the standard deviation is reduced by 0.045 175, and the root mean squared error is reduced by 0.029 615.