-
TPA - Assignment 2 Korneel Melkebeke, Pieter-Jan Steenbeke
Assignment 2: RTLS
Total Plant Automation
1 Introduction
In order to automatically detect and track targeted people or
objects within a building or some
defined area in real time, a real-time location system (RTLS)
can be utilised. Fixed reference
points receive wireless signals from active RTLS tags attached
to people or objects to determine
their position. An RTLS differentiates itself from a GSP in
terms of local positioning due to its
better accuracy and has therefore other applications such as
asset management and patient track-
ing, predominately in manufactories, warehouses and
healthcare.
First of all, as described in Chapter 2, an RTLS was configured
for a defined area in the lab to
track three objects - an operator, a tool, and a car. Next,
spatial relationships were analysed. To
dispose the interference of the various metal objects in the
defined area, different filters were
designated to the tags. Those filters were then examined by
tracking a fixed trajectory. These
analyses are discussed in Chapter 3 whereas conclusions are
articulated in Chapter 4.
2 Setup
A Ubisense RTLS was used comprising four antennas and three
active tags, seen in Figure 1.
Moreover, three software applications of Ubisense are utilised
to configure the system: (i) Site
Manager; (ii) Map View; (iii) and Location Engine Configuration.
The first section will describe
the Site Manager, section two will further handle the Map View.
Section three will go over the
Location Engine Configuration.
(a) antenna
(b) active tag
Figure 1 - Ubisense RTLS system components
-
TPA - Assignment 2 Korneel Melkebeke, Pieter-Jan Steenbeke 2
2.1 Site Manager
Various characteristics, such as types, objects,
representations, areas, cells, object locations, and
geometry, must be determined in the Site Manager. Types refer to
categorical traits, objects repre-
sent the items - persons, machines, equipment, goods -
accompanied with tags one wants to
track, and representations can be assigned to the objects for
visualisation purposes, as seen on
Figure 2. The area in which one wants to track the items must be
defined in the Areas tab - Fig-
ure 3. The locations of the antennas are set in the Cells tab -
Figure 5. Figure 4 depicts the object
locations.
In order to examine whether objects are in close proximity or
enter an alarm zone, spatial rela-
tionships can be assigned between multiple objects and between
objects and areas in the Geome-
try tab - Figure 6. Here, roles are utilised which represent
what part the object takes in the rela-
tionship, such as object or zone. Each role is assigned with an
area, i.e. shape, referring to what
extent the object takes part in the relation - Figure 7. For
example, a person has a small area
whereas an alarm zone can be large. Dependent on the type of
role, the area is absolute or rela-
tive. Specifically, a persons area is relative whereas an alarm
zone has an absolute area. There are
two methods to designate the relation to the roles. A role can
either contain another role via a
contains relation, or can be contained by another role object
via a contained by relation. For
example, in case a person is not allowed to enter a particular
area (alarm zone), the alarm zone
has a contains relationship with the person whereas the person
has a contained by relation with
the alarm zone. The objective of this experiment was to
determine when an assembly operator
was adding value to the product. It was assumed that the
operator only adds value when he
stands at a product (i.e. car) with a tool (i.e. screwing
device). Therefore, in the configuration of
this experiment, the tool had a contained by relation with the
operator while the operator had a
contains relation with the tool. On the other hand, the operator
had a contained by relation
with the car while the car had a contains relation with the
operator.
2.2 Map View
The visualisation is rendered in Map View. Both the objects and
the shapes of the roles are de-
picted. The shapes of the roles colour green when the spatial
relationship holds, see Figure 8,
indicating a valid spatial relationship. In case of an invalid
spatial relationship, the shapes colour
red. Note that the shape of the containing object must be
completely inside the shape of con-
tained by object for the spatial relationship to be true.
-
TPA - Assignment 2 Korneel Melkebeke, Pieter-Jan Steenbeke 3
(a) Types tab
(b) Objects tab
(c) Representations tab [1]
Figure 2 - Types and Objects tab in Site Manager
Figure 3 - Areas tab in Site Manager
Figure 4 - Object locations
Figure 5 - Cells tab in Site Manager
-
TPA - Assignment 2 Korneel Melkebeke, Pieter-Jan Steenbeke 4
Figure 6 - Roles configuration in the Geometry tab of Site
Manager
Figure 7 - Shape configuration Figure 8 - Spatial Relationships
view in Map View
2.3 Location Engine Configuration
Tags must be assigned to objects in the Location Engine
Configuration. First tags are added. In
the software application, this is conducted via ownerships, i.e.
tags own an object. This is illus-
trated in Figure 9. The three objects that were considered in
this experiment were (i) a person
entitled Minion; (ii) a vehicle entitled Lamborghini; and (iii)
a tool entitled screwing device.
Table 1 depicts the ownerships.
-
TPA - Assignment 2 Korneel Melkebeke, Pieter-Jan Steenbeke 5
Figure 9 - Ownership tab in Location Engine Configuration
Table 1 - Ownerships of the tags
Tag Owner
020-000-118-230 screwing device
020-000-118-224 Lamborghini
020-000-118-228 Minion
The Sensor and Cells tab illustrates the positions of the tags
in the area, see Figure 10. Here, the
trail can be tracked as well, see Figure 11.
Figure 10 - Real-time positioning of the tags in the Sensor and
Cells tab
-
TPA - Assignment 2 Korneel Melkebeke, Pieter-Jan Steenbeke 6
Figure 11 - Example tracked trail of a tag in real-time
In the area, a lot of metal objects were present and interfere
with the localisation system. There-
fore, filters have been applied, see Figure 12, to examine the
accuracy of the system by tracking
the trajectory of a straight line with and without the use of
filters. For this experiment, the trajec-
tory was first tracked using no filter. Then, the four
predefined filters as displayed in Figure 12
were applied. Thereafter, the parameters of the resulting
preeminent filter were compared with
the parameters of the inferior filters. By altering a few
parameters of the preeminent filter and
visually inspecting the tracked trail in the Sensor and Cells
tab, a best practices filter was defined.
The tag entitled with Minion was put into a railway car to
travel in a straight line and on a flat
surface, so as to control the variations in both the x- and
z-direction. The trajectory was tracked
explicitly by the Location Engine Configuration application and
implicitly by an additional soft-
ware tool CoordinatenLoggen.exe. The latter provides the
coordinates of the trail which will be
used to establish a linear regression model per filter. Then,
the coefficient of determination, R-
squared, will be utilised to define the accuracy of the filter
as it represents the proportion of vari-
ability in the observed response variable that is explained by
the linear regression model. Im-
portant to note, here, is that in the data sets of information
filtering and best practices filter
several lines of data representing either impracticable or
erroneous coordinates at the beginning
of the tests, were deleted but are still perceptible on the
tracked trajectories.
3 Analysis
This chapter outlines and discusses the results. The first
section presents the visualisation of the
spatial relationships between roles. Section 2 examines the
different effects of the filters on a
tracked trajectory.
-
TPA - Assignment 2 Korneel Melkebeke, Pieter-Jan Steenbeke 7
Figure 12 - Available filters in the Location Engine
Configuration
3.1 Spatial relationships
Figure 13 and Figure 14 respectively depict the situation where
the operator is inside and outside
the range of the car. The shades of both the operator and the
car did not colour red, representing
an invalid spatial relationship. It appeared that the vertical
height difference was not violated in
this activity, by which the system did not perceive the relation
as invalid.
Figure 13 - Operation inside the range of car
Figure 14 - Operator outside the range of car
3.2 Accuracy and Filters
In this section, the trail and the linear regression model is
presented per filter. Table 2 gives an
overview of the effects of the various filters in terms of the
linear model and the corresponding
R-square coefficient. Following filters present a vertical
near-linear path: (i) fixed height infor-
-
TPA - Assignment 2 Korneel Melkebeke, Pieter-Jan Steenbeke 8
mation filtering; (ii) information filtering; (iii) static
information; and (iv) a best practices1 combi-
nation of the filters.
Remarkably, here, is that the linear regression model computed
for the situation where fixed
height information filtering was conducted, indicates a large
variance whereas the observed tra-
jectory shows a primarily linear line. The plot of the tracked
x- and y-coordinates and the small
R-squared coefficient fosters this large variance.. A reasonable
cause for this bias is probably a
malfunction of the additional software tool. In contrast, the
tracked trajectories where infor-
mation filtering and where the best practices filter were
utilised, provide preeminent results.
Their linear models and their corresponding R-squared
coefficient profoundly comply with their
tracked trajectories. Furthermore, in spite of the three outer
data points, the trajectory of the case
where static information filtering was applied is fairly
straight. Those outlying data points justify
the lower R-squared coefficient. Once more, a probable
malfunction of the additional software
tool can be assigned as a reasonable cause. Despite the
moderately smooth trajectory of the case
when no filter was used, the corresponding linear model has a
slope and unexplainable variance
comparable to the four aforementioned near-linear models - apart
from the R-squared coefficient
of the linear model associated with the fixed height information
filter. Finally, in case of static
fixed height information filtering, the trajectory shows a
rather curved line in the upper half of
the trajectory. The latter is also visible on the plot of the x-
and y-coordinates, and resulted in a
slope almost twice as large as any other regression coefficient.
The proportion of total variation
of outcomes explained by the model, however, relates to the
R-squared coefficients of the other
near-linear models - also apart from the R-squared coefficient
of the linear model associated with
the fixed height information filter.
Table 2 - Overview linear regression models of the trajectories
using various filters
Filter Linear model R Reference
No filtering x = 0.0253y - 942.62 0.520 Figure 15
Fixed height information filtering x = 0.0205y - 954.52 0.146
Figure 16
Information filtering x = 0.0164y - 942.44 0.549 Figure 17
Static fixed height information filtering x = 0.0488y - 961.11
0.460 Figure 18
Static information filtering x = 0.0210y - 945.71 0.386 Figure
19
Best practices combination x = 0.0271y - 950.66 0.659 Figure
20
1 Of the preeminent fixed height information filtering, the
vertical positioning std dev was set to zero, and the tag height
above cell floor was set to 0.7.
-
TPA - Assignment 2 Korneel Melkebeke, Pieter-Jan Steenbeke 9
(a) (b)
Figure 15 - Tracked trail (a) and linear model of tracked
coordinates(b) using no filter
(a) (b)
Figure 16 - Tracked trail (a) and linear model of tracked
coordinates(b) using fixed height information filtering
x = 0.0253y - 942.62R = 0.5196
300 500 700 900
-1,000
-975
-950
-925
-900
x co
ord
inat
e
y coordinate
x = 0.0205y - 954.52R = 0.1459
300 500 700 900
-1,000
-975
-950
-925
-900
x co
ord
inat
e
y coordinate
-
TPA - Assignment 2 Korneel Melkebeke, Pieter-Jan Steenbeke
10
(a) (b)
Figure 17 - Tracked trail (a) and linear model of tracked
coordinates(b) using information filtering
(a) (b)
Figure 18 - Tracked trail (a) and linear model of tracked
coordinates(b) using static fixed height information filtering
x = 0.0164y - 942.44R = 0.5493
300 500 700 900
-1,000
-975
-950
-925
-900
x co
ord
inat
e
y coordinate
x = 0.0448y - 961.11R = 0.46
300 500 700 900
-1,000
-975
-950
-925
-900
x co
ord
inat
e
y coordinate
-
TPA - Assignment 2 Korneel Melkebeke, Pieter-Jan Steenbeke
11
(a) (b)
Figure 19 - Tracked trail (a) and linear model of tracked
coordinates(b) using static information filtering
(a) (b)
Figure 20 - Tracked trail (a) and linear model of tracked
coordinates(b) using a combination of best practises filter
4 Conclusions
An RTLS was first configured for a defined area in the lab to
get acquainted with the basic opera-
tions. Three objects - an operator, a tool, and a car - were
tracked and their spatial relationships
were analysed. Due to insufficient vertical height difference,
the shades of the objects did not
depict an invalid spatial relationship. Next, different filters
were administered to the tags to dis-
x = 0.021y - 945.71R = 0.3855
300 500 700 900
-1,000
-975
-950
-925
-900
x co
ord
inat
e
y coordinate
x = 0.0271y - 950.66R = 0.659
300 500 700 900
-1,000
-975
-950
-925
-900
x co
ord
inat
e
y coordinate
-
TPA - Assignment 2 Korneel Melkebeke, Pieter-Jan Steenbeke
12
card the interference of the various metal objects in the
defined area. Finally, the filters were ex-
amined by tracking a fixed trajectory by means of linear
regression models and corresponding
coefficients of determination. It was discovered that using
filters showed a predominately linear
path, despite some ambiguous results such as a high
unexplainable variance in the response vari-
able or a high slope of the linear model. Still, when no filter
was used, a fairly preeminent linear
regression model with a rather small unexplainable variance was
attained. Therefore, it can be
stipulated that filters enhance the performance of the system
when used properly.
-
References
[1] How to create a representation using Site Manager, Ubisense,
[Online]. Available:
https://download.ubisense.net/howto/SiteManagerRep_article/SiteManagerRep.html.
[Accessed 22 March 2015].