| Autor: |
Chhantyal, Khim, Jondahl, Morten Hansen, Viumdal, Hakon, Mylvaganam, Saba |
| Zdroj: |
IEEE Sensors Journal; 6/15/2018, Vol. 18 Issue 12, p5002-5013, 12p |
| Abstrakt: |
Reliable flow measurements of drilling fluid entering and returning from the wellbore can improve safety of people and assets by avoiding kicks (blowouts) and fluid loss. The return flow with cuttings and entrained gas in the drilling fluid pose a big challenge to the rig operators. An indirect way of measuring the volumetric flow is by measuring the flow level in an already existing open channel in the flow loop, as this level changes with changing volumetric flow. In this paper, different mechanistic and machine learning models, based on one to three fluid levels at specific locations along the custom designed open Venturi channel, are presented. The mechanistic models involve tuning of different correction factors, related to fluid rheology. As rheological properties vary with time, and real-time rheological measurements are not available, these models are valid only for fluids with known rheology. With real-time density as an extra input, fluid specific machine learning models for mass flow, can be applied to any fluids. In contrast, the proposed machine learning models for volumetric flow are robust and not dependent on the rheological properties of the fluids. These models have mean absolute percentage error between 2.05% and 4.76%. Soft sensing of volumetric flow based on non-invasive level measurements presented here has an additional advantage for applications in harsh environments. [ABSTRACT FROM PUBLISHER] |
| Databáze: |
Complementary Index |
| Externí odkaz: |
|
