Void Fraction Estimation in Vertical Gas-Liquid Flow by Plural Long Short-Term Memory With Sparse Model Implemented in Multiple Current-Voltage System

Tanaka, Koji; Prayitno, Yosephus Ardean Kurnianto; Sejati, Prima Asmara; Kawashima, Daisuke; Takei, Masahiro

doi:10.1615/multscientechn.2021039801

Cited by 6 publications

(5 citation statements)

References 16 publications

Supporting

Mentioning

Contrasting

Unclassified

Order By: Relevance

“…実験装置を図 3 に示す (Tanaka et. al., 2021b)．多電極インピーダンス計(MIM)はマイクロコントローラ m/s)，気体流入速度u G 6 種類(0, 0.02, 0.12, 0.24, 0.47, 0.71 m/s)の条件下で，テスト用データとして液体流入速度 u L 1 種類(0.6 m/s)，気体流入速度u G 6 種類(0, 0.02, 0.12, 0.24, 0.47, 0.71 m/s)の条件下で，電圧ベクトル V を測定した．流動様式の真値は流動様式線図 (Mishima and Ishii, 1984)を用い，ボイド率の真値αはドリフトフラックスモデル (Bonnecaze et al, 1971)…”

Section: 実験装置，実験方法，及び実験条件unclassified

“…and Borst， 1988) ．筆者らは，電気計測法の一種である多電極インピーダンス計(Multielectrode Impedance Meter: MIM)によるボイド率推定法を提案した(Jeon et al, 2019a) forward neural network, FFNN)，サポートベクターマシーン，遺伝的プログラミングを搭載し，液体質量流量とボイド率の推定法が提案された(Wang et al, 2016)．筆者らは，前述の問題(1)を解決する手段として，MIM に FFNN を搭載し，ボイド率推定の精度を向上させた(Jeon et al, 2019b) ．ところが，気泡分布やボイド率変動の激しい非一様性や非定常性の高い流れ場の場合，気液二相流の時間依存性を無視される FFNN では推定ボイド率が不安定かつ低精度となる (Tanaka et al, 2021a) ．したがって前述の問題(2)を解決するには至らなかった．ごく最近になり，その問題(2)，すなわち，気液二相流の時間依存性を考慮するために，畳み込みニューラルネットワーク( Convolutional neural network, CNN)や長短期記憶(Long short-term memory, LSTM)が，様々な計測機器に搭載されつつある．例えば，多変量時系列データの時間軸とデータ次元軸両方に対して畳み込み処理を行う分割畳み込みブロック(Decomposed convolutional block) を 4 セクター・コンダクタンスセンサに搭載したボイド率推定システム(Gao et al, 2020)や，加重回帰 LSTM を流動接触分解装置内の時系列気液二相流温度の予測に適用し，流動接触分解装置からガソリンの生産量推定システムが提案された(Zhang et al, 2019) ．そして，筆者らは LSTM を多段階とした多重長短期記憶(plural Long shortterm memory, pLSTM)を MIM に搭載したボイド率推定システムを提案し(Tanaka et al, 2021a) ，さらに，入力次元数を減らすために入力のスパース性に注目して，ラッソ回帰 (least absolute shrinkage and selection operator, Lasso)(Tibshirani, 1996)を搭載したボイド率推定方法をも提案した(Tanaka et al, 2021b) …”

unclassified

See 1 more Smart Citation

ヒルベルトシュミット独立基準ラッソと多重長短期記憶による多電極インピーダンス・ボイド率推定法

Tanaka

Prayitno

Sejati

et al. 2022

Transactions of the JSME (In Japanese)

Self Cite

View full text Add to dashboard Cite

Hilbert-Schmidt independence criterion least absolute shrinkage and selection operator (HSIC Lasso) and plural long shortterm memory (pLSTM) has been implemented in a multi-electrode impedance meter (MIM) to estimate void fraction α accurately in vertical gas-liquid flows. The MIM measures voltage vector V reflecting the spatial distribution of bubbles in the gas-liquid flows with multiple pairs of electrodes on the inner wall of the pipe. HSIC Lasso selects the electrode pairs which are strongly dependent on the void fraction and flow regime based on the nonlinear assumption to select the voltage vector V without any hyperparameters. pLSTM estimates α after identifying the time-dependent flow regime. In the experiments, V was measured by MIM under the conditions of 25 points of liquid and gas flow rate in gas-liquid upward flows for training, on the other hand, five points for testing, respectively. As the result, HSIC Lasso successfully selects highly dependent electrode pairs with regression coefficient β over 0.03. pLSTM achieves a 100% accuracy of flow regime identification and time mean relative error 𝜂̅ with less than 0.0157 of void fraction estimation in bubble flow and slug flow, which decreases total mean relative error <η > by 69% compared with the existing method.

show abstract

Section: 実験装置，実験方法，及び実験条件unclassified

unclassified

ヒルベルトシュミット独立基準ラッソと多重長短期記憶による多電極インピーダンス・ボイド率推定法

Tanaka

Prayitno

Sejati

et al. 2022

Transactions of the JSME (In Japanese)

Self Cite

View full text Add to dashboard Cite

show abstract

“…In the multiphase flow community, machine learning techniques, especially the artificial neural network (ANN) (Beale and Jackson, 1990), have been applied to multiphase flow researches have been developed, e.g. flow pattern identification from video images (Liu and Bai, 2019;Shibata et al, 2021), subgridscale modeling of interfacial mass transfer in DNS of bubbles (Weiner et al, 2019), modeling of heat transfer coefficients (Enoki et al, 2017), predictions of void fractions in upward bubbly flows (Tanaka et al, 2022) and modeling of hydrodynamics in bubble column reactors (Behkish, 2004;Behkish et al, 2005;Tanaka, 2010;Tanaka et al, 2010). Behkish (2004) developed an ANN for predicting the total gas holdup and the gas holdup of large bubbles in bubble column and slurry bubble column reactors (Behkish et al, 2005).…”

Section: Introductionmentioning

confidence: 99%

An Ann Correlation of Lift Coefficients of Bubbles in Linear Shear Flows

et al. 2022

View full text Add to dashboard Cite

An artificial neural network (ANN) correlation of the lift coefficients of spherical and deformed bubbles in liner shear flows is developed. The ANN has three hidden layers and the number of neurons in each hidden layer is 10. Lift coefficient data available in literature are used to train and test the ANN correlation. The data include bubbles in both viscous force dominant and surface tensioninertial force dominant regimes. Good agreements between the training data and model predictions are obtained. Comparing the model with the test data confirms that the correlation has a good generalization performance.

show abstract

“…This work is dedicated to support the theoretical model development and validation of CFD codes [12], [13] by high-quality temporal and spatial experimental data of air-water slug flow sub-regimes. Moreover, the proposed model is valuable for the development of advanced machine learning for two-phase flow application, such as a comprehensive input model in plural long short-time memory with sparse model implemented in multiple current-voltage system (pLSTM-SM-MCV) [14].…”

Section: Introductionmentioning

confidence: 99%

Sub-Regimes of Air-Water Slug Flow Characteristics in Horizontal Pipes by Liquid Hold-up Model Correlated to Bubble Behaviours (LHmBb)

Prayitno

Dinaryanto²,

Deendarlianto

et al. 2022

FREE

View full text Add to dashboard Cite

Sub-regimes of air-water slug flow in horizontal pipes have been characterized by a liquid hold-up model correlated to bubble behaviors (LHmBb). The LHmBB is composed of two sections which are the liquid hold-up model and bubble behavior investigations. The liquid hold-up model is developed based on the statistical analysis of the probability density function (PDF) to quantify the bubble distribution. The bubble behaviors are qualitatively investigated based on the high-speed camera and correlated to the quantified LH for the characterization of sub-regime of air-water slug flow. The LHmBB characterizes the sub-regime of air-water slug flow in a horizontal transparent acrylic pipe with an inner diameter of underwater and air superficial velocities of and, respectively. As a result, four sub-regimes are determined as Initially dispersed Bubbles (IdB), Low dispersed Bubbles (LdB), High dispersed Bubbles (HdB), and Dominantly dispersed Bubbles (DdB). The presence of bubbles determines the type of sub-regime by decreasing the number of bubbles and dispersing the bubbles mechanism. Moreover, the proposed LHmBb includes the correlation function to ease the prediction of the sub-regimes of air-water slug flow characteristics which leads to the enhancements of the two-phase flow pattern map in horizontal pipes.

show abstract

Void Fraction Estimation in Vertical Gas-Liquid Flow by Plural Long Short-Term Memory With Sparse Model Implemented in Multiple Current-Voltage System

Cited by 6 publications

References 16 publications

ヒルベルトシュミット独立基準ラッソと多重長短期記憶による多電極インピーダンス・ボイド率推定法

ヒルベルトシュミット独立基準ラッソと多重長短期記憶による多電極インピーダンス・ボイド率推定法

An Ann Correlation of Lift Coefficients of Bubbles in Linear Shear Flows

Sub-Regimes of Air-Water Slug Flow Characteristics in Horizontal Pipes by Liquid Hold-up Model Correlated to Bubble Behaviours (LHmBb)

Contact Info

Product

Resources

About