Accurate Void Fraction Estimation by Plural Long Short Term Memory Applying to Multiple Voltage Current System in Gas-Liquid Flows.

Plural long short term memory (${p}$ LSTM) applying to a multiple voltage-current (mVC) system has been proposed in order to estimate the void fraction $\hat {\alpha }$ accurately in gas-liquid flows. The ${p}$ LSTM consists of two LSTMs, one for flow regime identification (fri-LSTM) and the other for void fraction estimation (vfe-LSTM). The fri-LSTM identifies a flow regime ${q}$ from current vectors ${i}$ , corresponding to gas distribution, measured by mVC system. Based on the identification result and ${i}$ , the customized vfe-LSTM to each ${q}$ estimates $\hat {\alpha }$. For training, ${i}$ are experimentally measured at 36 points of the true void fraction $\alpha $ , which is calculated by the drift flux model. On the other hand, ${i}$ for test data are measured under 12 points of $\alpha $. Two parameters of each LSTM, one is sequence length ${S}$ representing time dependence length considered within the LSTM and the other is the number of LSTM blocks ${M}$ related to the estimation performance, are optimized so that accurate void fraction estimation is achieved. As a result, ${p}$ LSTM applying to mVC system achieves a 100% accuracy of flow regime identification and less ±0.00034 standard error of void fraction estimation in liquid single-phase flow, bubble flow, slug flow, and churn flow. Accurate estimation is caused by the fact that ${p}$ LSTM can consider the time dependence of gas-liquid flows suitable for each flow regime and neglect the effect of the flow regime on ${i}$. [ABSTRACT FROM AUTHOR]

Copyright of IEEE Sensors Journal is the property of IEEE and its content may not be copied or emailed to multiple sites or posted to a listserv without the copyright holder's express written permission. However, users may print, download, or email articles for individual use. This abstract may be abridged. No warranty is given about the accuracy of the copy. Users should refer to the original published version of the material for the full abstract. (Copyright applies to all Abstracts.)