Eventos Anais de eventos
COBEM 2023
27th International Congress of Mechanical Engineering
FLOW RATE MEASUREMENT OF TWO-PHASE FLOW IN DIFFERENT PATTERNS WITH A SINGLE THROTTLE DEVICE USING PgNN
Submission Author:
Tiago Francisconi Borges Camargo , SC
Co-Authors:
Tiago Francisconi Borges Camargo, Emilio Paladino
Presenter: Tiago Francisconi Borges Camargo
doi://10.26678/ABCM.COBEM2023.COB2023-1657
Abstract
The inline flow rate measurement of individual phases in gas-liquid two-phase flows is crucial for process monitoring and control in various industrial applications. Despite the availability of flow sensors capable of measuring the mixture mass or volume flow rate, another variable, such as phase volume fraction or slip velocity is necessary for the determination of individual phase flow rates. The sensors for the determination of phase fraction, as gamma-ray based or capacitive/resistive are complex, difficult to calibrate for large ranges of phase flow rates, and require specific protocols for their operation, which impedes their usage in harsh environments such as deep offshore production. Another option, in the context of differential pressure based flow meters, is the use of two throttle devices, which, with the use of cross-correlated parameters allow the measurement of two flow variables. However, this approach again lacks generalization for large ranges of gas and liquid flow rates. Typical flow rate correlations used in differential pressure flowmeters use the average Δ͞p to correlate flow rate. However, there is a huge quantity of information in the transient Δp signal, mainly when it comes to gas-liquid flow streams. The phase interactions as they flow through a constriction result in very specific characteristics of this signal. In this work, we propose a deep learning model to develop a virtual flow sensor capable of determining individual phase flow rates from the transient differential pressure signal of a single throttle device. Furthermore, although we use an orifice plate, the concept can be applied to any constriction available in the process that acts as a throttle device or a choke or flow control valve, since there is available data for model training. In addition, the proposed model considered some physical theories in learning. A Physic-guided Neural Network is proposed to ensure a more generic output model. The inputs of the model are extracted from the time series of the differential pressure signal from a large dataset obtained from an experimental test of water-air horizontal flow that ranges from 0.03 to 1.10 m/s for liquid superficial velocity jl and 0.03 to 15 m/s for gas superficial velocity jg in a 25.4 mm internal diameter tube. The models based on PgNN and DNN were evaluated on another unseen orifice area ratio dataset to compare the performance and generalization of different learning methods, mainly in terms of overfitting.
Keywords
Inline two-phase flow metering, Virtual Sensing, Pressure flutuations, physics-guided neural nework
