Void Fraction Measurement of Gas-Liquid Two-Phase Flow Based on Empirical Mode Decomposition and Artificial Neural Networks

Abstract

Phone Number: 0 (+44) 131 650 4868, Fax Number: 0 (+44) 131 650 6551 2 ABSTRACT A new void fraction estimation method for gas-liquid two-phase flow combining two differential pressure (DP) signals acquired from a single Venturi tube and based on Empirical Mode Decomposition (EMD) and Artificial Neural Networks (ANN) was experimentally investigated. In order to study gas-liquid distribution in horizontal pipes, two DP signals from the top and bottom sections of the Venturi tube are acquired and EMD is adopted to extract stable and fluctuating components of the DP signals. Experimental data revealed that fluctuating index increases nearly linearly with increasing void fraction when void fraction is less than 0.4. When void fraction is larger than 0.4, this near-linearity ceases. A combination of ANN method and the fluctuating index of DP signals is developed to estimate void fraction. Experimental results show that void fraction based on DP signal from top section of Venturi tube is overestimated because of clustering of bubbles and the scarcity of liquid information when gas-liquid mixture velocity is low. Void fraction is underestimated when the mixture velocity is high. A high gas-liquid slip ratio results in void fraction underestimation. Void fraction prediction performance is satisfactory when void fraction is less than 0.4 and fluctuating index of DP signal less than 1.2. ll f ll kk p            (1)

Data-driven control of discrete-phase concentration in a vortex involved two-phase flow with optimized energy consumption

“…As shown in Figure 3.9 and 3.10, it is observed that the particles are removed much more difficultly out from the corner in the zero-control condition, if the particles are initially placed closer to the vertex of the corner. The similar experimental phenomenon reveals that usually the indoor particle concentration at the corner is higher than that in the other area, and it is influenced significantly by the airflow disturbance [135,159]. Moreover, the similar phenomenon is…”

“…found in the open literature [135], in which the indoor particles were monitored experimentally in a naturally ventilated meeting room, and the indoor particle concentration peaked at the corner due to vertexes capturing the particles, generated by local airflow disturbance. Hence, a desired residual particle concentration may be achieved by control of the vortex trajectory with a properly designed control input.…”

“…Among all the controller design algorithms proposed based on ML model, model predictive control is one of the most popular algorithms for engineering process, due to good adaptation on nonlinear dynamics model [137]. The design of MPC based on NN was formulated and valid by Ławryńczuk [133,135,136] and Janakiraman et al [134] via the simulations of different processes, in which NN is linearized by Taylor approximation and MPC then developed by solving a quadratic optimization problem. The proposed MPC-NN controller presented good control performance [138].…”

“…For example, the link between two-fluid dynamics and socalled "closure terms" was established by Buist et al [134] via NN, performing better predictive ability than the conventional methods. The void fraction was estimated by a well-trained NN for gas-liquid flow in a Venturi tube, via the data experimentally collected by two differential pressure signals, with good predictive accuracy for high the mixture velocity [135]. An SVM based image processing techniques were implemented by Zhou et al [136] to excellently classify the different flowing regimes within a short process time, for the twophase flow in a horizontal tube, with the image experimentally collected by digital high-speed video.…”

Dynamic measurement of gas volume fraction in a CO2 pipeline through capacitive sensing and data driven modelling

“…In recent years, data driven modeling techniques were proposed for the GVF and flowrate measurement of gas-liquid two-phase flow (Wang et al, 2019;Peyvandi and Rad, 2017). Among these data driven models, the BPNN, RBFNN and LS-SVM have been widely used as alternatives to physical-based and conceptual models.…”