Visualization for a Multi-Sensor Data Analysis
Sang Ok Koo, Hyuk Don Kwon, Chang Geol Yoon, Won Seok Seo, Soon Ki Jung
Virtual Reality Laboratory
Kyungpook Nat’l University, South Korea
{sokoo,hdkwon,ycglove,wonseok}@vr.knu.ac.kr, skjung@knu.ac.kr
Abstract
This work describes our efforts in creating the soft-
ware in order to analyze the multi-sensor data for gas
transmission pipeline inspection. The amount of data
is usually considerable because the hardware system
that consists of multiple heterogeneous sensors records
multi-sensor values for long-distance inspection. It
imposes a heavy burden on the operators who should
sieve the huge and complex data, detect features of the
pipeline and decide a feature as a significant defect.
In our system, the virtual 3D pipeline helps the user
to examine the inside of pipeline intuitively by navi-
gating according to the realistic pipeline trajectory. We
mapped the geographical data of the pipeline and hetero-
geneous sensor data on the virtual 3D pipeline. More-
over, our system offer the various feature detail views
to help the users rapid and precise decision. Users can
switch the navigation mode and the feature detail mode
easily. Consequently, the virtual pipeline plays a role
as an intuitive interaction metaphor for pipeline in-
spection.
1 Introduction
Nowadays, modern industrial systems have evolved
towards complex structures that consist of multiple
heterogeneous sensors. At the same time, there is an
urgent need to develop methods for integrating multi-
sensor data. In many cases, the parameters affecting
the characteristics of multi-sensor data are numerous
and the dependencies between them are ambiguous.
For this reason, efficient data-level integration, such
as data fusion, needs to be sufficiently pre-anlayzed by
humans. On the other hand, visual integration using
multivariate visualization techniques can help humans
understand data easily without a thorough data anal-
ysis. An admittedly efficient form of visual integration
is the platform in which humans are immersed into a
point of view or an alternate real world.
The amount of data of inspection systems is usually
considerable because successive multiple sensory data
are recorded in a very short time, such as a millisec-
ond or nanosecond, over a long period. Sensor data
have some noise which is caused by the sensor itself or
the sensing environment whereby an analyzer has diffi-
culties in detecting suspicious sections through out the
entire inspected region. For this reason, many analyz-
ers employ signal processing techniques by automatic
filtering or calibration and AI(Artifitial Intelligence)
techniques are applied in order to classify specific fea-
tures automatically. The performance of an automatic
classifier, however, is not entirely reliable although it
can be quite accurate according to the type of data.
In order to satisfy industrial requirements for analysis
accuracy, feature identification and verification by hu-
mans is essential. Efficient visualization provides clues
reagrding the meaning of the data.
Our research investigated the creation of software
that can visualize and analyze multivariate time-series
data, that is acquired by pipeline inline inspection in-
struments which consist of multiple heterogeneous sen-
sors. A pipeline is a large pipe which is used to carry oil
or natural gas over a long distance, often underground.
A pipeline inspection gauge or PIG is a device which
is inserted into a pipeline. The PIG is propelled by
the pressure of the medium flow to do a specific task
within the pipeline. Pipelines represent a considerable
investment and can often prove strategic to countries
and governments. In order to protect these valuable
investments, maintenance must be done and pigging is
one such maintenance tool. Inline inspection PIGs are
designed to assess the pipeline condition on a periodic
basis.
In order to obtain accurate information of a pipeline,
it is necessary for the operator to examine closely the
multi-sensor data for the entire region that is inspected.
Although the operator has expert technical knowledge
about signal patterns, it is difficult and very time con-
Proceedings of the International Conference on Computer Graphics, Imaging and Visualisation (CGIV'06)
0-7695-2606-3/06 $20.00 © 2006