Chinese Journal of Chemical Engineering, 20(2) 400ü405 (2012)
Voidage Measurement of Air-Water Two-phase Flow Based on ERT
Sensor and Data Mining Technology
*
WANG Baoliang (ฆ)
**
, MENG Zhenzhen (ჲჲ), HUANG Zhiyao (ܻᄝྉ), JI Haifeng
(ސںע) and LI Haiqing (ں)
State Key Laboratory of Industrial Control Technology, Department of Control Science and Engineering, Zhejiang
University, Hangzhou 310027, China
Abstract Based on an electrical resistance tomography (ERT) sensor and the data mining technology, a new voi-
dage measurement method is proposed for air-water two-phase flow. The data mining technology used in this work
is a least squares support vector machine (LS-SVM) algorithm together with the feature extraction method, and
three feature extraction methods are tested: principal component analysis (PCA), partial least squares (PLS) and in-
dependent component analysis (ICA). In the practical voidage measurement process, the flow pattern is firstly iden-
tified directly from the conductance values obtained by the ERT sensor. Then, the appropriate voidage measurement
model is selected according to the flow pattern identification result. Finally, the voidage is calculated. Experimental
results show that the proposed method can measure the voidage effectively, and the measurement accuracy and
speed are satisfactory. Compared with the conventional voidage measurement methods based on ERT, the proposed
method doesn’t need any image reconstruction process, so it has the advantage of good real-time performance. Due
to the introduction of flow pattern identification, the influence of flow pattern on the voidage measurement is over-
come. Besides, it is demonstrated that the LS-SVM method with PLS feature extraction presents the best measure-
ment performance among the tested methods.
Keywords two-phase flow, voidage measurement, electrical resistance tomography sensor, data mining, feature
extraction
1 INTRODUCTION
Voidage, which is the ratio of gas area to the total
area of pipeline cross section, is an important parame-
ter in gas-liquid two-phase flow. Its measurement is
significant for status monitoring, quality control,
safety assurance and flowrate measurement of the
two-phase flow system
[1]. Although many voidage
measurement methods have been proposed, few of
them can meet the practical requirements because of
the inherent complexity of two-phase flow.
Electrical resistance tomography (ERT) technique
has proved to be an attractive and promising method for
voidage measurement of gas-liquid two-phase flow
[24].
However, most of the conventional voidage measure-
ment methods based on ERT are to construct the phase
distribution image of two-phase flow in the pipe and
calculate the voidage value based on the image
[2, 4].
In these methods, the complex and time-consuming
image reconstruction process will lead to bad real-time
performance, and the measurement accuracy is not
satisfactory because of the influence of “soft-field”
characteristics of ERT. Therefore, their practical ap-
plications are limited, and more research work should
be undertaken in this area.
The conductance data measured by the ERT sensor
reflects the information of the phase distribution and
fraction of gas-liquid flow. It is possible to identify the
flow pattern and measure the voidage directly from
the conductance data. The aim of this work is to pro-
pose a new voidage measurement method for air-water
two-phase flow based on a 16-electrode ERT sensor
and the data mining technology. The data mining tech-
nology is used to implement the flow pattern identifi-
cation using the least squares support vector machine
(LS-SVM) classification algorithm
[5], and the voidage
determination from the conductance data using the
LS-SVM regression algorithm
[6]. Thus, the image re-
construction process in conventional voidage measure-
ment methods is by-passed, and the real-time perform-
ance is improved. Meanwhile, flow pattern identifica-
tion is introduced before voidage measurement and
different measurement models are used for different
flow patterns, so the influence of flow pattern on the
voidage measurement is overcome. Three feature ex-
traction methods are applied to pre-process the con-
ductance data and reduce the input dimensionality of
LS-SVM: principal component analysis (PCA), partial
least squares (PLS) and independent component analy-
sis (ICA). These three methods are tested and compared
in their performance for the voidage measurement.
Experiments showed that the proposed method
could measure the voidage effectively for air-water
two-phase flow and the LS-SVM method with PLS
feature extraction presented the best measurement
performance among the tested methods.
2 VOIDAGE MEASUREMENT SYSTEM
2.1 Structure of measurement system
The voidage measurement system consists of an
Received 2011-11-17, accepted 2012-01-17.
* Supported by the National Natural Science Foundation of China (60972138).
** To whom correspondence should be addressed. E-mail: blwang@iipc.zju.edu.cn