sensors

Article
Kalman Filters in Geotechnical Monitoring of
Ground Subsidence Using Data from MEMS Sensors
Cheng Li 1, *, Rafig Azzam 2 and Tomás M. Fernández-Steeger 3
1 Chengdu Engineering Corporation Limited, Chengdu 610072, China
2 Department of Engineering Geology and Hydrogeology, RWTH Aachen University,
52064 Aachen, Germany; azzam@lih.rwth-aachen.de
3 Department of Applied Geosciences, TU Berlin University, 10587 Berlin, Germany;
fernandez-steeger@lih.rwth-aachen.de
* Correspondence: cheng.li@rwth-aachen.de

Academic Editor: Jörg F. Wagner

Received: 25 March 2016; Accepted: 1 June 2016; Published: 19 July 2016

Abstract: The fast development of wireless sensor networks and MEMS make it possible to set
up today real-time wireless geotechnical monitoring. To handle interferences and noises from the
output data, Kalman filter can be selected as a method to achieve a more realistic estimate of the
observations. In this paper, a one-day wireless measurement using accelerometers and inclinometers
was deployed on top of a tunnel section under construction in order to monitor ground subsidence.
The normal vectors of the sensors were firstly obtained with the help of rotation matrices, and then be
projected to the plane of longitudinal section, by which the dip angles over time would be obtained
via a trigonometric function. Finally, a centralized Kalman filter was applied to estimate the tilt
angles of the sensor nodes based on the data from the embedded accelerometer and the inclinometer.
Comparing the results from two sensor nodes deployed away and on the track respectively, the
passing of the tunnel boring machine can be identified from unusual performances. Using this
method, the ground settlement due to excavation can be measured and a real-time monitoring of
ground subsidence can be realized.

Keywords: Kalman filter; ground subsidence; rotation matrices; accelerometer; inclinometer

1. Introduction
Due to the fast development of Micro Electro Mechanical Systems (MEMS) and the optimization
of sensor cost, size and energy consumption in the last two decades, Wireless Sensor Networks
(WSNs) has been progressively entered many areas such as disaster monitoring [1–3] and industrial
sensing [4,5]. A WSN consists of spatially distributed sensor nodes to monitor various conditions such
as temperature, pressure, humidity, sound, vibration, pollutants and motion. In this sense they allow
us to enter Internet of Things (IoT) and provide physical information about our environment to users
where ever needed. Nowadays, real-time remote surveillance of environmental conditions has been
achieved and a much more comprehensive understanding of the variations of natural environments
over time can be obtained.
From 2007 to 2011, the Department of Engineering Geology and Hydrogeology at RWTH Aachen
University and its partners carried out a joint project namely SLEWS (a Sensor based Landslide
Early Warning System), aiming at developing a prototype of an alarm and early warning system
for landslides. SLEWS was funded by GEOTECHNOLOGIEN, which is a geoscientific research
and development program supported by the German Federal Ministry for Education and Research
(BMBF) and the German Research Foundation (DFG). A self-organizing and multi-hop wireless
monitoring network was created thereby [6,7]. This wireless monitoring system and its predecessor

Sensors 2016, 16, 1109; doi:10.3390/s16071109 www.mdpi.com/journal/sensors

Sensors 2016, 16, 1109 2 of 15

Sensors 2016, 16, 1109 2 of 14

system X-SLEWS [8] have been applied to monitor several kinds of natural hazards and geotechnical
geotechnical
activities suchactivities
as tilting ofsuch
rockastowers
tiltingdeformation
of rock towers deformation
[9], ground subsidence[9], due
ground subsidence
to ground due to
improvement
ground improvement
measures [10], tunneling measures [10], [11],
construction tunneling construction
and landslide [11], and
monitoring [12].landslide monitoring
Figure 1 exhibits [12].
the main
Figure 1 exhibits
components of the the main components
X-SLEWS network. The of the X-SLEWS
node capsulenetwork.
comprises Thea node capsule
base board forcomprises
processing a base
and
boardand
radio for an
processing
add-on board and radio and an
for sensing andadd-on
backupboard for sensing
data storage. and backup
The add-on boarddata storage.
is easily The
replaced
add-on
by board
another oneis with
easilydifferent
replacedsensors
by another one with
to allow fastdifferent
adoptionsensors to allow
to distinct fastThe
tasks. adoption
battery tocapsule
distinct
tasks. Thewith
allocated battery
threecapsule allocatedprovides
3.6-V batteries with three 3.6-V
the powerbatteries
supply provides the power
to the sensor nodesupply
capsule, toseparating
the sensor
node and
node capsule,
energy separating nodealso
source allows andtoenergy
exchange source allows
energy alsoortoadd
sources exchange
modulesenergy sources
for energy or add
harvesting
modules
like solar for energy
panels. Theharvesting like solar
gateway receives andpanels.
stores The gateway data
the retrieved receives
fromand stores
all the the in
nodes retrieved data
the network
frombroadcasts
and all the nodes the in the network
updated and broadcasts
clock periodically the updated
to connected clockItperiodically
nodes. bears also a to connected
raspberry PI nodes.
with a
It bears
server to also a raspberry
control PI withand
communication a server to control
data storage and cancommunication
be additional and data storage
equipped with a dataand up-link
can be
additional
via 3G modem equipped with operation
for remote a data up-link via 3G modem
and monitoring for remote operation and monitoring [8,11].
[8,11].

Figure 1.
1. Components
Components of X-SLEWS (From [11]). The node capsule comprises a base board for basic basic
operation and
and an
an add-on
add-onboard
boardfor
forsensing
sensingand storing
and storingfunction. The The
function. battery capsule
battery allocated
capsule with
allocated
threethree
with 3.6-V3.6-V
batteries provides
batteries the the
provides power
powersupply
supplyto to
the
thenode
nodecapsule
capsulewhile
while measuring
measuring and data
data
transmitting. The gateway
transmitting. gateway receives
receives and
and saves
saves the
the retrieved
retrieved data
data from
from all
all the
the nodes
nodes in the network and
broadcasts the
broadcasts the updated
updated clock
clock periodically
periodically to
to connected
connected nodes.
nodes.

The X-SLEWS
The X-SLEWS wireless
wireless monitoring
monitoring system
system was applied in
was applied in aa one-day
one-day field
field test
test on
on top
top of
of the
the
construction site of the South Hongmei Road super highway tunnel in Shanghai,
construction site of the South Hongmei Road super highway tunnel in Shanghai, and positional and positional
information ofofthe
information thesensor
sensor nodes
nodes hashas been
been obtained
obtained based
based on theonmeasurements
the measurements from embedded
from embedded MEMS
MEMS accelerometers
accelerometers and inclinometers.
and inclinometers. Both ofBoth
the of the derived
derived normalnormal vectors
vectors basedbased onoutput
on the the output of
of the
the MEMS sensors contain interferences and noises and thus a data filter was needed
MEMS sensors contain interferences and noises and thus a data filter was needed for a good estimate. for a good
estimate.
In In the following
the following we will
we will briefly briefly introduce
introduce the sensorsthe sensors
and and measuring
measuring principleThen
principle applied. applied.
the
Then the processing of sensor data using rotation matrices will be described. Finally, the
processing of sensor data using rotation matrices will be described. Finally, the usage of the Kalman usage of the
Kalman filter for data estimation is described in detail and the results will
filter for data estimation is described in detail and the results will be discussed. be discussed.

2. Applied
AppliedMEMS
MEMSSensors
Sensors
been equipped
Each sensor node has been equipped with aa 3-axis
3-axis accelerometer
accelerometer and
and two
two orthogonal
orthogonal 1-axis
1-axis
inclinometers,both
inclinometers, bothof which
of which are produced
are produced by Electronics
by Murata Murata Electronics Oycalled
Oy (previous (previous called VTI
VTI Technologies,
Technologies,
Vantaa, Vantaa,
Finland) and areFinland)
mounted and
on are mounted
an add-on on an
board. Inadd-on
Table 1 board. In data
a generic Table 1 a generic
sheet datasensor
for the two sheet
forshown.
is the two sensor is shown.
Basically, MEMS Accelerometers and inclinometers share the same principle to observe
acceleration. A sketch of a MEMS accelerometer is shown below in Figure 2. The movable proof
mass is suspended by the restoring springs, while the sensing plates are fixed onto the sensor board.
Sensors 2016, 16, 1109 3 of 15
Sensors 2016, 16, 1109 3 of 14
Sensors 2016, 16, 1109 3 of 14
The accelerometer is Table
sensitive to the
1. Generic linear
Data Sheetacceleration of the
of the two Sensors sensor
Used in the and the earth gravitational field.
Deployment.
When the sensing plate deflects
Table 1. from the original
Generic Data SheetAccelerometer position, the change in capacitance (C1 and C2)
of the two Sensors Used in theInclinometer
Deployment.
Sensor Type
between the fingers of Product
the proof mass and sensing plates can be converted
SCA3100-D04 into a voltage signal, which
SCA830-D07
Sensor Type Accelerometer Inclinometer
is subsequently digitized
Size(w to
× h
Product
digital
× l) value, from
7.0 × which
3.3 × 8.6
SCA3100-D04
the
mm 3 acceleration
7.6 ×along
3.3 themm
× 8.6
SCA830-D07
axis
3 can be expressed
Measurement
through a given formula. Due Axis
to
Size(w × h × l) different 3-axis
precision of the
7.0 × 3.3 × 8.6 mm 3 sensors, the 1-axis
output of the
7.6 × 3.3 × 8.6 mm 3 accelerometer and
Range
inclinometer is substituted
Measurement into respective formulas
Axis ±2 g
3-axisto obtain the acceleration ±1 g
1-axis value [13].
Sensitivity
Range(LSB/g) 900 (0.0637°/count)
±2 g 32,000 (0.00179°/count)
±1 g
Sensitivity (LSB/g) 900 (0.0637°/count) 32,000 (0.00179°/count)
Table 1. Generic Data Sheet of the two Sensors Used in the Deployment.
Basically, MEMS Accelerometers and inclinometers share the same principle to observe
acceleration. A sketch
Basically, MEMSofAccelerometers
a MEMS accelerometer is shown below
and inclinometers in Figure
share 2. Theprinciple
the same movable proof mass
to observe
Sensor Type Accelerometer Inclinometer
is suspendedA
acceleration. bysketch
the restoring
of a MEMS springs, while theissensing
accelerometer shown plates
below are fixed onto
in Figure 2. Thethe sensor board.
movable The
proof mass
accelerometer
is suspended by isProduct
sensitive
the to the
restoring linearwhile
springs, SCA3100-D04
acceleration of the
the sensing sensor
plates areand
fixedtheSCA830-D07
earth
onto thegravitational
sensor board.field.
The
Size(w ˆhˆ l)to the from 7.0original
ˆ 3.3 ˆ position,
8.6 3 3
When the sensing
accelerometer plate
is sensitive deflects linearthe
acceleration of mm and7.6
the change
the sensor in
the 3.3 ˆ gravitational
earth 8.6 mm
ˆ capacitance (C1 andfield.
C2)
between Measurement
the fingersplate Axis
of thedeflects
proof mass 3-axis can be converted into1-axis
When the sensing fromand
thesensing
originalplates a voltage signal,
position, the change in capacitance (C1 andwhich
C2)
is subsequently Range to digital value, from which
digitized ˘2 g the acceleration along the˘1 g can be expressed
axis
between the fingers of the proof mass and sensing plates can be converted into a voltage signal, which
Sensitivity (LSB/g) 900 (0.0637˝ /count) 32,000 (0.00179˝ /count)
through a givendigitized
is subsequently formula.to Due to different
digital precision
value, from which of thethe sensors, the
acceleration output
along of the
the axis canaccelerometer
be expressed
and inclinometer
through is substituted
a given formula. Due to into respective
different formulas
precision to obtain
of the thethe
sensors, acceleration
output of value [13].
the accelerometer
and inclinometer is substituted into respective formulas to obtain the acceleration value [13].

C1
C2
C1
Fixed C2 Movable
Sensing Proof
Fixed Movable
Plate Mass
Sensing Proof
Plate Mass

Figure 2. Sketch of a MEMS Accelerometer (Modified from [14]). The sensing plate is fixed onto the
Figure 2. Sketch of a MEMS Accelerometer (Modified from [14]). The sensing plate is fixed onto
sensor
Figure board andofthe
2. Sketch movable
a MEMS proof mass is(Modified
Accelerometer suspendedfrom
by the restoring
[14]). spring. The isproof
The sensing fixedmass will
the sensor board and the movable proof mass is suspended by the restoringplate
spring. Theonto the
proof mass
drift downward when a linear acceleration points upwards or the Earth’s gravitational field
sensor board and the movable proof mass is suspended by the restoring spring. The proof mass will points
will drift downward
downwards.
when a linear acceleration points upwards or the Earth’s gravitational field
drift downward when a linear acceleration points upwards or the Earth’s gravitational field points
points downwards.
downwards.
When the accelerometer or the inclinometer is used for measuring the gravity, as the output
represents
When thethe
When theweight of the earth
accelerometer
accelerometer ororthegravitational
the fieldis
inclinometer
inclinometer isalong
usedthe
used foraxis
for (Figure the
measuring
measuring 3), the
the tilt
gravity, angle
gravity, of
asthe
as the theaxis
output
output
with respect
represents toweight
the the horizontal plane
of the earth can be obtained
gravitational easily the
field along via axis
a sine function
(Figure [11].tilt angle of the axis
3), the
represents the weight of the earth gravitational field along the axis (Figure 3), the tilt angle of the axis
with respect
with respect to horizontal
to the the horizontal plane
plane canbebeobtained
can obtained easily
easily via
viaaasine
sinefunction [11].
function [11].
Horizontal Plane
α
Horizontal Plane
α
g'
g'
Axis
Axis
α
α
g
g
Figure 3. Measurement of Inclination from the Accelerometer. The accelerometer measures the value
Figureof 3.′,Measurement
Figure which of for
represents
3. Measurement Inclination
the weightfrom
of Inclination from the
of the Accelerometer.
earth’s
the gravitational
Accelerometer. The accelerometer
acceleration
The accelerometer themeasures
inmeasures
axial the the value
direction
value
1 on
of g , of the
which sensor board.
represents
′, which The tilt
for the
represents angle
weight
for the of the
of of
weight axis
thethe ∝ with
earth’s respect to
earth’sgravitationalthe horizontal plane
acceleration g in
gravitational acceleration can
in the be calculated
theaxial
axialdirection
direction on
'
via an board.
the sensor
on the inverse trigonometric
The
sensor board.tiltThe
angle function:
tilt angle ∝=
of theofaxis
the 9arcsin
∝ with
with
axis . respect
respect to the horizontal
to the horizontalplane
plane can be calculated
can be calculated via
g1 '
an inverse
via an trigonometric function:
inverse trigonometric function: ∝= arcsin
9 “ arcsin g . .
Sensors 2016, 16, 1109 4 of 15

3. Acquiring Sensor Motion Using Inertial Navigation Algorithm

During the monitoring, a 3-axis accelerometer returns three values in each measurement,
representing
Sensors 2016,the respective weights of the earth’s gravitational acceleration along the three
16, 1109 4 of 14 axial

directions. In this sense each axis can be seen and handled equally to an independent sensor, like
3. Acquiring Sensor Motion Using Inertial Navigation Algorithm
the two orthogonal 1-axis inclinometers which provide from each sensor a perpendicular inclination.
Processing During
output, the the
monitoring,
tilt anglea of
3-axis
eachaccelerometer returns three values
axis of the accelerometers and in
theeach measurement,
inclinometers can be
representing the respective weights of the earth’s gravitational acceleration along
deduced. However, we cannot gain an understanding of the spatial variations of the sensors by only the three axial
directions. In this sense each axis can be seen and handled equally to an independent sensor, like the
comparing the three values from accelerometers and the two values from inclinometers over time.
two orthogonal 1-axis inclinometers which provide from each sensor a perpendicular inclination.
Indeed, there is lack of a unified spatial parameter that enables a comparison of the results from
Processing output, the tilt angle of each axis of the accelerometers and the inclinometers can be
distinct sensors. Hence, before the study of the spatial variations of the sensors, a normal vector should
deduced. However, we cannot gain an understanding of the spatial variations of the sensors by only
be calculated
comparingasthe a standard
three valuesexpression, representing
from accelerometers and the
the position
two values offrom
a sensor in a three
inclinometers dimensional
over time.
Cartesian
Indeed, coordinate
there is lacksystem. The deviation
of a unified of the normal
spatial parameter vectora can
that enables be divided
comparison into
of the threefrom
results gradual
stepsdistinct
[15] and will beHence,
sensors. brieflybefore
explained as follows.
the study of the spatial variations of the sensors, a normal vector
should be calculated as a standard expression, representing the position of a sensor in a three
3.1. Expression of Cartesian
dimensional g’ coordinate system. The deviation of the normal vector can be divided into
three gradual steps [15] and will be briefly explained as follows.
We define a coordinate system OX0 Y0 Z0 , which is transformed to a new coordinate system
OX1Y1Z1 by rotating
3.1. Expression of g’ a roll angle φ, a pitch angle θ and a yaw angle ψ around x-, y- and z-axis
respectively (Figure 4). For each step of rotation, there is a corresponding rotation matrix R that
We define a coordinate system , which is transformed to a new coordinate system
transfers the expression of a vector based on the old coordinate system to an expression based on the
′ ′ ′ by rotating a roll angle ϕ, a pitch angle θ and a yaw angle ψ around x-, y- and z-axis
new one. There are six possible orders for the rotation of the coordinate system but only the orders
respectively (Figure 4). For each step of rotation, there is a corresponding rotation matrix that
of
Z-Y-Xtransfers
or Z-X-Y thecan assure aofsolution,
expression since aonspatial
a vector based the oldvector has two
coordinate degrees
system of freedombased
to an expression [14]. on
Taking
the the
Z-Y-Xnew
order
one.for instance,
There are sixafter being
possible rotated
orders for successively
the rotation ofaround z-, y- and
the coordinate x-axis,
system butthe rotated
only vector of
the orders
earthofgravitational
Z-Y-X or Z-X-Y acceleration
can assure ap0, 0, ´1q since
solution, 1 can abe expressed
spatial vector as:
has two degrees of freedom [14]. Taking
the Z-Y-X order for instance, after being rotated ¨ successively
˛ ¨ around z-, y- and
˛ x-axis, the rotated
vector of earth gravitational acceleration (0,0, −1)′ can0 be expressed as:
sinθ
g1xyz “ R x pϕq Ry pθq Rz pψq ˝ 00 ‚ “ ˝ ´cosθsinϕ ‚ (1)
˚ ‹ ˚ ‹
′ = ( ) ( ) ( ) ´10 = − ´cosθcosϕ (1)
−1 −

Z’ Z0
Yaw
ψ

Y’

O Y0
Pitch ϴ
Roll φ

X0
X’

Figure 4. Coordinate System Transforming Through the Order of X-Y-Z (modified from [11]). The
Figure 4. Coordinate System Transforming Through the Order of X-Y-Z (modified from [11]). The initial
initial coordinate system is transformed to considering the roll, pitch and yaw
coordinate system OX0 Y0 Z0 is transformed to OX 1 Y 1 Z1 considering the roll, pitch and yaw angles
angles around the x-, y- and z-axis respectively.
around the x-, y- and z-axis respectively.
3.2. Acquirement of Rotation Angles
3.2. Acquirement of Rotation Angles
In this step, based on the equivalent relations between the sensor output value and the
corresponding
In this step, components
based on the of the ′ vector,relations
equivalent the roll between
and pitchthecan be resolved
sensor outputand rotation
value and the
matrices arecomponents
corresponding hereby obtained. As ag1result
of the of the
vector, themechanism
roll ϕ andofpitch
an accelerometer, the vector
θ can be resolved andof rotation
the
earth’s gravitational acceleration is exported as (0,0,1) instead of (0,0, −1)
matrices are hereby obtained. As a result of the mechanism of an accelerometer, the vector [11]. Similarly, theof the
components of ′ are correspondingly equal to the magnitude
T of the normalized
T accelerometer
earth’s gravitational acceleration is exported as p0, 0, 1q instead of p0, 0, ´1q [11]. Similarly, the
reading ( , , ) after multiplying with −1, which can be presented as:
Sensors 2016, 16, 1109 5 of 15

components of g1xyz are correspondingly equal to the magnitude of the normalized accelerometer
` ˘T
reading G p G px , G py , G pz after multiplying with ´1, which can be presented as:
¨ ˛
sinθ
Gp
“ ´g1 xyz (2)
˚ ‹
“ ´ ˝ ´cosθsinϕ ‚
k Gp k
´cosθcosϕ

Solving the equation, the roll ϕ and the pitch θ can be expressed as:
Accelerometer: ˆ ˙
G py
ϕ “ arctan (3)
G pz
G px
θ “ ´arcsin b (4)
G2px ` G2py ` G2pz

Inclinometer:
G py
ϕ “ arcsin b (5)
1 ´ G2px

θ “ ´arcsinG px (6)

Since current monitoring system is not equipped with a magnetometer or a gyroscope, the value
of yaw ψ cannot be solved. Nevertheless, taking account of the relatively simple deformation mode
in geotechnical projects and the slight relevance to construction safety, the yaw ψ that represents the
rotation around z-axis could be considered as zero.

3.3. Derivation of the Normal Vector

After retrieving the expression of rotation matrices, we operate the matrices inversely and intend
to convert the normal vector from the transformed coordinate system into a description according
to the old one. Defining the upward direction that is perpendicular to the sensor board plane as the
normal direction; following the Z-Y-X order, the equation that describes the correlation between initial
normal vector Vn and the converted normal vector p0, 0, 1qT in the rotated coordinate system can be
expressed as: ¨ ˛
0
R x pϕq Ry pθq Rz pψq Vn “ ˝ 0 ‚ (7)
˚ ‹
1

As the rotation matrices Rz pψq, Ry pθq and R x pϕq are unit orthogonal matrices (Unit Orthogonal
Matrix: is a unitized square matrix, the transpose of which equals to its inverse, which can be expressed
as Q T “ Q´1 , QQ T “ Q T Q “ E, where E is the identity matrix), they satisfy the principle that the
transpose equals to their inverse. Hence Vn can be solved by:
¨ ˛
0
Vn “ RzT pψq RyT pθq R Tx pϕq ˝ 0 ‚ (8)
˚ ‹
1
¨ ˛
cosϕsinθ
ñ Vn “ ˝ ´sinϕ ‚ (9)
˚ ‹
cosϕcosθ
The normal vector represents the tilt direction and the tilt angle of the sensor board, which moves
along with the monitored objects during geotechnical events.
Sensors 2016, 16, 1109 6 of 15

4. Positional Estimate Using Kalman Filter

The normal vectors calculated from raw data of the MEMS sensor have inherited interferences and
noises from different sources and thus should not be used directly to represent the position. Therefore,
Kalman filter is used subsequently to estimate the state of the observations. In particular, this allows
also to apply the concept of sensor fusion if more than one sensor is used e.g., to monitor inclination
or acceleration.
Kalman filter goes back on 1960, when R. E. Kalman proposed a recursive solution to the
discrete-data linear filtering problem [16]. As one of the primary developers of the linear quadratic
estimation (LQE), this method is named after his name. Kalman filter is a set of mathematical equations
that uses a recursive means to estimate the underlying system state by minimizing the mean square
error [17]. Nowadays, Kalman filter and its extended version have been widely used in target tracking,
navigation, and other relevant fields in data processing [18,19]. In the following a brief introduction of
Kalman filter followed by his application to our problem of noisy positional data is given.

4.1. A Brief Introduction of Kalman Filter

The first step of the Kalman filter is to build a model that represents the series of data. The state
variable x is addressed as an expression of discrete linear stochastic difference equation and the
measurement value z is described as a linear function of x:

xk “ Axk´1 ` Buk´1 ` wk´1 (10)

zk “ Hxk ` vk (11)

where:

A—State transition matrix relating the state at the previous time step k ´ 1 to that at the current
step k;
u—Optional control input;
B—Control matrix that relates u to the state x;
w—Process noise vector, w(i) is normally # distributed denoted by N(0, Q) and the covariance
0 if i ‰ j
cov rw piq , w pjqs “ Q ¨ δij , where δij “ ;
1 if i “ j
H—Meaurement transition matrix relating the state x to the measurement z;
v—Measurement noise vector, v piq is normally distributed # denoted by N p0, Rq and the
0 if i ‰ j
covariance cov rv piq , v pjqs “ R ¨ δij , where δij “ , and w and v are
1 if i “ j
statistically independent.

Once the Kalman filter model is established, the estimating process is applied by a
predictor-corrector cycle. As is shown in Figure 5, when the initial state value is established, the
iteration can be immediately started. Specifically, time update equations and measurement update
equations are respectively grouped that are responsible for the priori estimate for the next time step
and the feedback of the priori estimate. Table 2 provides the update equations from groups of time
and measurement. A detailed explanation of Kalman filter can be found in [17,20]. There are several
formulations Kalman filter update equations; in this paper, a specific form of Kalman gain K and a
posteriori estimate error covariance P are selected on purpose of a concise expression of Kalman filter
algorithms (Tables 2 and 3) for both single and multi-sensor systems. More information can be found
in [21].
 feedback of the priori estimate. Table 2 provides the update equations from groups of time and
measurement. A detailed explanation of Kalman filter can be found in [17,20]. There are several
formulations Kalman filter update equations; in this paper, a specific form of Kalman gain and a
posteriori estimate error covariance are selected on purpose of a concise expression of Kalman
filter algorithms (Tables 2 and 3) for both single and multi-sensor systems. More information can be
Sensors 2016, 16, 1109 7 of 15
found in [21].

Time Update Measurement Update

(Predictor) (Corrector)

Figure 5. Kalman
Figure 5. KalmanFilter Predictor-Corrector
Filter Predictor-Corrector Cycle.
Cycle. Theupdate
The time time projects
update the
projects
currentthe
statecurrent
estimatestate
estimate ahead,
ahead, and then
and then the measurement
the measurement update update adjusts
adjusts the the projected
projected estimate
estimate based based
on the on the
actual
actualmeasurement.
measurement.

Table 2. Kalman Filter Update Equations (x̂k´ —A priori state estimate at step k, Pk´ —A priori estimate
error covariance at step k, x̂k —A posteriori state estimate at step k, Pk —A posteriori estimate error
Sensors 2016, 16, 1109 7 of 14
covariance at step k, Kk —Kalman Gain that is deduced by minimizing Pk ).
Table 2. Kalman Filter Update Equations ( —A priori state estimate at step , —A priori
Time Updateerror
estimate Equations
covariance at step , Measurement
—A posteriori Update
state estimate , —A posteriori
Equations
at step
x̂k´estimate error
“ A x̂k´1 ` Bucovariance
k´1
at step ,
(12) —Kalman Gain that is deduced
K k “ P H
k `
by
T Rminimizing
´1
k
). (14)
Pk´ “ APk´1 A T ` Qk´1
˘
(13) x̂k “ x̂k´ ` Kk zk ´ H x̂k´ (15)
Time Update Equations Measurement
´ Update¯´1
Equations
− P “ P ´ ´1
` H T R´1 H −1 (16)
= −1 + −1 (12) k k = k (14)
− − −
= −1 + −1
(13) = + ( − ) (15)
− −1 −1 −1
In a multi-sensor system, the main fusion methods include=centralized ( + and)distributed means (16)
[22],
the sketches of which are illustrated in Figure 6. Specifically, for centralized fusion methods, Willner,
In a multi-sensor system, the main fusion methods include centralized and distributed means
Chang [23]
[22], has introduced
the sketches threeare
of which main linear Kalman
illustrated in Figurefilter algorithms
6. Specifically, forfor a multi-sensor
centralized fusion system,
methods,namely
parallelWillner,
filter, sequentially
Chang [23] has filter and datathree
introduced compression
main linearfilter,
Kalmanandfilter
all the three filters
algorithms for a are equivalent and
multi-sensor
optimum. For namely
system, distributed fusion
parallel filter, methods,
sequentially Carson hasdata
filter and proposed a federated
compression Kalman
filter, and filterfilters
all the three as a typical
are equivalent
representation [24,25].and optimum.speaking,
Generally For distributed fusion methods,
the centralized fusionCarson
consumes has proposed
intensivearesources
federated during
Kalman
calculating butfilter as a typical
assures representation
a limited [24,25]. Generally
loss of information, speaking,
while the centralized
the distributed fusioncalculates
fusion consumes faster
intensive resources during calculating but assures a limited loss of information, while the distributed
but has a less accuracy. In the following application, the centralized fusion method will be applied
fusion calculates faster but has a less accuracy. In the following application, the centralized fusion
considering
method the relatively
will be applied small amount
considering theofrelatively
processed data
small at the
amount of same time.
processed data at the same time.

Fusion Center

S. 1 S. 2 S. 3

Sensors
Centralized Fusion

Fusion Center

Local Local ... Local

Center 1 Center 2 Center N

S. 1 S. 2 S S S S. m

Sensors
Distributed Fusion

Figure Figure 6. Centralized

6. Centralized Fusion
Fusion andand DistributedFusion.
Distributed Fusion. In
Incentralized
centralizedfusion, all measurements
fusion, are sentare sent
all measurements
to the fusion center; while in distributed fusion, the sensors supply data to a set of local processors
to the fusion center; while in distributed fusion, the sensors supply data to a set of local processors to to
conduct pretreatment firstly, and then the result will be sent to the fusion center for a global estimate.
conduct pretreatment firstly, and then the result will be sent to the fusion center for a global estimate.
4.2. Application of Kalman Filter to Monitor Tunneling-Induced Ground Subsidence from Sensor Motion
A one-day field test of the X-SLEWS wireless sensor network was conducted from 10 a.m. to
5 p.m. on 27 January in 2013 at a tunneling construction site in Shanghai. The sensor nodes were
deployed at the surface above the underground super highway tunnel at South Hongmei Road, while
the shield machine would pass below the area at a depth of 41.65 m. Due to the subsidence related to
SensorsSensors 16, 1109
2016, 2016, 16, 1109
8 of 15
8 of 14

change of the orientation of the normal vector and define it as the expected motion due to the tunnel
4.2. Application of Kalman
driving. Figure Filter
7 shows to Monitor Tunneling-Induced
the orientation of the MEMS sensorGround Subsidence
nodes in relation tofrom
the Sensor Motion
tunnel axis as
well
A as the expected
one-day stages
field test of motion
of the X-SLEWScaused by shield
wireless tunneling.
sensor Whenwas
network the excavation
conducted causes
fromground
10 a.m. to
subsidence, the sensor node will firstly incline to the shield head, secondly restore to a certain extent
5 p.m. on 27 January in 2013 at a tunneling construction site in Shanghai. The sensor nodes were
with the expansion of the settlement area, and then tilt to the shield head again following the
deployed at the surface above the underground super highway tunnel at South Hongmei Road, while
succeeding subsidence.
the shieldThe machine would pass
data retrieved frombelow the area
two sensor at ahave
nodes depth of processed
been 41.65 m. Due to paper
in this the subsidence
as an example.related to
the tunnel excavation, subsurface deformations was expected to monitor. Each
Firstly, the normal vectors over time have been derived as briefly described above and in detail node contains a 3-axis
accelerometer and twotheorthogonal
in [11]. To monitor 1-axis inclinometers,
ground deformation, and both of
the change in inclination them
of the are able
surface, to measure
respectively the the
tiltingsensor
of thenode,
sensor with a settable
as depicted in Figurefrequency to the
7, can be used as base station
a measure andvia
to avisualize
wirelessthenetwork
changes[8].overThetime.MEMS
sensorsTheofnormal vectorsnode
the sensor have been consequently
are orientated onprojected
a planeinto the longitudinal
defined as sensor section
plane,ofandthe we
tunnel, and the
observe
change theofdip
theangle values have
orientation been
of the herebyvector
normal solvedand
to express
define ittheasvertical the bestmotion
the expected tilting of
due thetosensor.
the tunnel
Figure 8 illustrates the derivation of the dip angle from the projected
driving. Figure 7 shows the orientation of the MEMS sensor nodes in relation to the normal vector _ in the axis
tunnel
longitudinal section, and Figure 9 shows the plots of dip angles of two sensor nodes located away
as well as the expected stages of motion caused by shield tunneling. When the excavation causes
from and on the track in the longitudinal section over the observing day. Particularly, Node 102 was
ground subsidence, the sensor node will firstly incline to the shield head, secondly restore to a certain
deployed on the surface on the track of the tunnel for the purpose of an observation of ground
extent with the expansion
subsidence, while Nodeof the
101 settlement
located area,from
16 m away and thethen tilt to thecenter
excavation shieldis head
used asagain following
a reference of the
succeeding
Node 102.subsidence.

Figure
Figure 7. Schematic
7. Schematic DiagramofofSpatial
Diagram SpatialRelation
Relation Between
BetweenSensors
Sensorsand
andthethe
Shield Machine
Shield (a) and
Machine (a) and
Expected Motion Stages of a Sensor Node Caused by Shield Tunneling (b) (modified from [11]).
Expected Motion Stages of a Sensor Node Caused by Shield Tunneling (b) (modified from [11]).

N
The data retrieved from two sensor nodesZhave been processed in this paper as an example.
Firstly, the normal vectors over time have been derived as briefly described above and in detail
in [11]. To monitor the ground deformation, the change in inclination of the surface, respectively
Vn_prj
the sensor node, as depicted in Figure 7, can be used as a measure and to visualize the changes
over time. The normal vectors have been consequently
Dip Angle projected into the longitudinal section of the
tunnel, and the dip angle values have been hereby solved to express the vertical the best tilting of
the sensor. Figure 8 illustrates the derivationOof the dip angle from the projected normal vector Vn_prj
in the longitudinal section, and Figure 9 shows the plots of dip anglesY of two sensor nodes located
away from and on the track in the longitudinal section over the observing day. Particularly, Node 102
Dip Angle
was deployed on the surface on the track of the tunnel for the purpose of an observation of ground
subsidence, while Node 101 located 16 m away from the excavation center is used as a reference of
Node 102. X

Figure 8. Derivation of the Dip Angle From the Projected Normal Vector _ in the Longitudinal
Section (modified from [15]). The dip is the angle between _ and z positive semi-axis, as well as
the angle between the sensor board and the horizontal plane.
 Figure 7. Schematic Diagram of Spatial Relation Between Sensors and the Shield Machine (a) and
Sensors 2016, 16, 1109 9 of 15
Expected Motion Stages of a Sensor Node Caused by Shield Tunneling (b) (modified from [11]).

Z N

Vn_prj

Dip Angle

O
Y

Dip Angle

FigureFigure 8. Derivation
8. Derivation of the
of the DipAngle
Dip AngleFrom
From thethe Projected
ProjectedNormal
NormalVector in the
Vector _Vn_prj Longitudinal
in the Longitudinal
Section
Section (modified
(modified from
from [15]).
[15]). TheThedip
dipisisthe
the angle between V_n_prjand
angle between andz positive semi-axis,
z positive as well
semi-axis, as as
as well
the
the angle angle between
between the sensor board and the horizontal
the sensor board and the horizontal plane. plane.
Sensors 2016, 16, 1109 9 of 14

Projected Dip Angle of Node 101 (away from Track)

4
Dip Angle in Longitudinal Section/ °

Dip Angle ACC

Dip Angle INC
3.5

2.5

2
10:54:25 11:55:51 12:57:18 13:58:45 15:00:11 16:01:38 17:03:05
Time

Projected Dip Angle of Node 102 (on the Track)

2
Dip Angle in Longitudinal Section/ °

Dip Angle ACC

Dip Angle INC
1.5

0.5

0
10:54:25 11:55:38 12:56:51 13:58:04 14:59:17 16:00:30 17:01:44
Time

Figure
Figure 9. 9. Plots
Plots ofofDip
DipAngles
AnglesofofSensor
SensorNode
Node 101
101 (away
(away from
from Track)
Track) and
andNode
Node102102(on
(onthe
theTrack)
Track)inin
the Longitudinal Section over the Observing Day. Blue and green lines stand for the performances of
the Longitudinal Section over the Observing Day. Blue and green lines stand for the performances of
the accelerometer and the inclinometer. Vertical bars can be recognized as outliers.
the accelerometer and the inclinometer. Vertical bars can be recognized as outliers.

The change of the inclination has been simplified as a uniform motion. Hereby, the movement
The of
model change of the inclination
the inclination has node
of the sensor been can
simplified as aas:
be defined uniform motion. Hereby, the movement
model of the inclination of the sensor
= node
+ can + be defined
, = as:
1,2, … (17)
= + , = 1,2, … (18)
In which and k “ tilt
areθthe 1 ` Tω
θk´angle ` wangular
k´1the
and k´1 , k “velocity
1, 2, . . . at time , is the measuring (17)
time interval, is the process noise vector at time .
ω “ ωk´1 ` wk´were
During the monitoring, the kmeasurements 1 , k “ conducted
1, 2, . . . every ten seconds, thus the
(18)
measuring
In whichtime interval
θk and is taken
ωk are the as 10.and
tilt angle Therefore, the Kalman
the angular velocity filter model
at time is is
k, T built
theas:
measuring time
=
interval, wk is the process noise vector at time k. + δ , = 1,2, … (19)
1 10 1
where = , = , δ= , is the process noise vector and it is normally distributed
0 1 1
0 ≠
denoted by (0, ) and the covariance [ ( ), ( )] = ⋅ , where = .
1 =
Because the state transition matrix has a high reliability, the result will not be influenced
much by , hence we define a relatively small value of covariance as 10−7 and thus =
0.0000001 0
.
0 0.0000001
Since there are two types of sensors (accelerometers and an inclinometers) have been embedded
Sensors 2016, 16, 1109 10 of 15

During the monitoring, the measurements were conducted every ten seconds, thus the measuring
time interval T is taken as 10. Therefore, the Kalman filter model is built as:

xk “ Axk´1 ` δwk´1 , k “ 1, 2, . . . (19)

« ff « ff « ff
θk 1 10 1
where xk “ , A “ , w is the process noise vector and it is normally
,δ “
ωk 0 1 1
#
0 if i ‰ j
distributed denoted by N p0, Qq and the covariance cov rw piq , w pjqs “ Q ¨ δij , where δij “ .
1 if i “ j
Because the state transition matrix A has a high reliability, the result will not be influenced
much« by w, hence we define ff a relatively small value of covariance as 10´7 and thus
0.0000001 0
Q“ .
0 0.0000001
Since there are two types of sensors (accelerometers and an inclinometers) have been embedded
in a sensor node, both of the measurements are going to be processed in the fusion center. Hereby the
measurement value z known as the tilt angle at step k is defined as:

zk “ Hxk ` vk , k “ 0, 1, 2, . . . (20)
« ff « ff « ff « ff
zk,1 H1 1 0 vk,1
In which zk “ ,H“ “ , vk “ . Assuming the measurement
zk,2 H2 1 0 vk,2
« ff « ff
Rk,1 1
noise vector v is white noise with a covariance of 1, and thus Rk “ “ .
Rk,2 1
« ff « ff
0 1 0
The initial values of the model are defined as x0 “ , P0 “ . These values are
0 0 1
substituted into Equation (21) to Equation (25) in Table 3, following the Kalman filter cycle as shown
in Figure 5.

Table 3. Kalman Filter Equations for Centralized Fusion.

Time Update Equations Measurement Update Equations

x̂k´ “ A x̂k´1 (21) Kk,i “ Pk HiT R´1
k,i (23)
2
AT
“ ‰
Pk´ x̂k “ x̂k´ ` Kk,i zk,i ´ Hi x̂k´
ř
“ APk´1 ` Qk´1 (22) (24)
i“1
2 ´1
Pk “ pPk´ HiT R´1 (25)
´1 ř
` k,i Hi q
i“1

Before the application of Kalman filter, the values of dip angles of Nodes 101 and 102 were
subtracted by their initial values individually, as there was an initial offset for each sensor due to the
soldering. In this way the variation of dip angles of both sensors of Nodes 101 and 102 can be plotted,
which are presented in Figure 10 together with the filtered result. In general, the filtered dip angles
show much lower variances compared to the values from the accelerometer and the inclinometer.
A discussion and evaluation is given in the following.
 Before the application of Kalman filter, the values of dip angles of Nodes 101 and 102 were
subtracted by their initial values individually, as there was an initial offset for each sensor due to the
soldering. In this way the variation of dip angles of both sensors of Nodes 101 and 102 can be plotted,
which are presented in Figure 10 together with the filtered result. In general, the filtered dip angles
show 2016,
Sensors much
16,lower
1109 variances compared to the values from the accelerometer and the inclinometer. A
11 of 15
discussion and evaluation is given in the following.

Variations of Projected Dip Angle of Node 101 (away from Track)

1.5
Dip Angle in Longitudinal Section/ °

Dip Angle INC

1 Dip Angle ACC
Dip Angle KF
0.5

-0.5

-1

-1.5
10:54:25 11:55:51 12:57:18 13:58:45 15:00:11 16:01:38 17:03:05
Time

Variations of Projected Dip Angle of Node 102 (on the Track)

1.5
Dip Angle in Longitudinal Section/ °

Dip Angle INC

1 Dip Angle ACC
Dip Angle KF
0.5

-0.5

-1

-1.5
10:54:25 11:55:38 12:56:51 13:58:04 14:59:17 16:00:30 17:01:44
Time

Figure 10.
Figure 10. Plots
Plots of
of Dip
Dip Angle
Angle Variations
Variations of
of Sensor
Sensor Node
Node 101
101 (away
(away from
from Track) and Node
Track) and Node 102
102 (on
(on the
the
Track) in
Track) inthe
theLongitudinal
LongitudinalSection
Sectionover
overthe
theObserving
Observing Day.
Day. TheThe absolute
absolute values
values were
were subtracted
subtracted by
by the
the initial value to obtain variations of dip angles, and the changes detected from the
initial value to obtain variations of dip angles, and the changes detected from the accelerometer andaccelerometer
and inclinometer
inclinometer can thus
can thus be compared
be compared directly.
directly. BlueBlue
and and
greengreen
lineslines stand
stand for performances
for the the performances of
of the
the accelerometer and the inclinometer, and the red line is the result after the application of
accelerometer and the inclinometer, and the red line is the result after the application of Kalman filter. Kalman
filter. Vertical
Vertical bars canbars
becan be recognized
recognized as outliers.
as outliers.

4.3. Evaluation
4.3. Evaluation and
and Discussion
Discussion of
of the
the Filtering
Filtering
As is
As is shown
shown in in Figure
Figure 10,
10, vertical
vertical bars
bars inin blue
blue and
and green
green indicate
indicate outliers
outliers caused
caused byby noise
noise
interferences from electrical fields and ground vibration when the tunnel boring
interferences from electrical fields and ground vibration when the tunnel boring machine and machine and other
construction
other measures
construction approach,
measures where where
approach, the sensor node would
the sensor be affected
node would and thus
be affected andwould obtain
thus would
inaccurate
obtain measurements.
inaccurate Regarding
measurements. the accelerometer,
Regarding the accelerometer, the the
combination
combination of ofthe
thethree 1-axis
three 1-axis
accelerometer has enabled a normalization in calculating the normal vector,
accelerometer has enabled a normalization in calculating the normal vector, but the relatively but the relatively low
sensitivity
low of 0.0637°/count
sensitivity has has
of 0.0637˝ /count reduced
reducedthe thereliability
reliabilityof of
thethededuced
deduceddip dipangle.
angle. AsAs for the
for the
inclinometer, in spite of the high sensitivity of 0.00179°/count
˝ compared to the
inclinometer, in spite of the high sensitivity of 0.00179 /count compared to the accelerometer, the accelerometer, the
output from the inclinometer could not be normalized because only two orthogonal
output from the inclinometer could not be normalized because only two orthogonal axes were deployed axes were
deployed
with with inclinometers.
inclinometers. Therefore, Therefore,
to carry outtoa data
carryfilter
out to
a data filter
identify thetooutliers
identifyonthe
theoutliers
observed ondata
the
before they are used for an assessment is necessary and valuable. In Figure 10, the scale of the outliers
for the inclinometer is evidently much bigger than that of the accelerometer for both nodes. Hence, it
can be inferred that with a strong effect of noise interferences and ground vibration, the difference of
sensitivity would not be the main factor of malfunction any more. Instead, the defect of normalization
of the inclinometer has resulted in higher ratio of outliers and an unstable performance.
To explore more information hidden from the variations of dip angles, standard deviations of the
filtered dip angles were calculated every hour for both sensor nodes, and amplified filtered variations
of dip angles with sectional standard deviation values is presented in Figure 11. The standard deviation
values are mostly between 0.0172 and 0.0266, but two exceptions appear in the first two hours for Node
102 with a value of 0.0481 and 0.0373 respectively. On the other hand, Node 101 that was deployed
16 meters away performed regularly in the first two hours compared to its latter performance. Hereby,
the abnormality of Node 102 indicates a disturbance of the data when the tunnel boring machine
passed through the area underground.
variations of dip angles with sectional standard deviation values is presented in Figure 11. The
standard deviation values are mostly between 0.0172 and 0.0266, but two exceptions appear in the
first two hours for Node 102 with a value of 0.0481 and 0.0373 respectively. On the other hand, Node
101 that was deployed 16 meters away performed regularly in the first two hours compared to its
latter 2016,
Sensors performance.
16, 1109 Hereby, the abnormality of Node 102 indicates a disturbance of the data 12when
of 15
the tunnel boring machine passed through the area underground.

Variations of Projected Dip Angle of Node 101 After Kalman Filter (away from Track)
Dip Angle in Longitudinal Section Section/ °

0.1
Dip Angle KF

Sn = 0.0206 Sn = 0.0236 Sn = 0.0235 Sn = 0.0215 Sn = 0.0172 Sn = 0.0185

-0.1

-0.2

-0.3
11:04:53 12:04:35 13:04:17 14:03:59 15:03:41 16:03:23 17:03:05
Time

Variations of Projected Dip Angle of Node 102 After Kalman Filter (on the Track)
0.1
Dip Angle in Longitudinal Section/ °

Dip Angle KF

-0.1

Sn = 0.0481 Sn = 0.0373 Sn = 0.0220 Sn = 0.0235 Sn = 0.0266 Sn = 0.0229

-0.2

-0.3
11:04:53 12:04:21 13:03:50 14:03:18 15:02:47 16:02:15 17:01:44
Time

Figure 11.
Figure 11. Plots
Plots of
of Dip
Dip Angle
Angle Variations after Kalman
Variations after Kalman Filter
Filter of
of Sensor
Sensor Node
Node 101
101 (away
(away from
from Track)
Track)
and Node 102 (on the Track) in the Longitudinal Section over the Observing Day
and Node 102 (on the Track) in the Longitudinal Section over the Observing Day with Hourly with Hourly
Standard Deviations.
Standard Deviations.The
Thevalues
valuesofofstandard
standarddeviations
deviations
areare mostly
mostly between
between 0.0172
0.0172 andand 0.0266,
0.0266, andand
the
two exceptions appear in the first two hours for Node 102 indicating a disturbance of the data.data.
the two exceptions appear in the first two hours for Node 102 indicating a disturbance of the

Taking the second measuring hour for Node 102 as an example, a comparison was made
Taking the second measuring hour for Node 102 as an example, a comparison was made between
between the filtered dip angles based on different algorithms of Kalman filter. Figure 12 shows the
the filtered dip angles based on different algorithms of Kalman filter. Figure 12 shows the plots of
plots of dip angle variations over time after distinct filtering. The blue line corresponds to the filtered
dip angle variations over time after distinct filtering. The blue line corresponds to the filtered dip
dip angles based on measurements from the 3-axis accelerometer, while the green line represents the
angles based on measurements from the 3-axis accelerometer, while the green line represents the result
result corresponding to the two orthogonal 1-axis inclinometers. The red line shows centralized
corresponding to the two orthogonal 1-axis inclinometers. The red line shows centralized filtered dip
filtered dip angles on the basis of both the accelerometer and the inclinometer, and the average of
angles on the basis of both the accelerometer and the inclinometer, and the average of two individual
two individual filtering (blue and green line) is also indicated with purple dot line. It can be observed
filtering (blue and green line) is also indicated with purple dot line. It can be observed that the
that the centralized filtering is not always consistent with the curve of the average. The standard
centralized filtering is not always consistent with the curve of the average. The standard deviation of
deviation of the inclinometer line is much higher than the accelerometer line, which indicates that the
the inclinometer line is much higher than the accelerometer line, which indicates that the lack of z-axis
lack of z-axis of the inclinometer combination has led an instable performance. Nevertheless,
of the inclinometer combination has led an instable performance. Nevertheless, considering its higher
considering its higher sensitivity, the variations calculated by inclinometers should still be taken into
sensitivity, the variations calculated by inclinometers should still be taken into account, especially
account, especially when the variations of dip angle are not notable. The standard deviation of the
when the variations of dip angle are not notable. The standard deviation of the dip angle variations
dip angle variations based on both sensor values is higher than that from the accelerometer and lower
based on both sensor values is higher than that from the accelerometer and lower than that from the
inclinometer, which have combined the advantages from both sensors and reduced to the instability to
a certain level.
If the sensor node was deployed exactly on the ground above the tunneling center, it is not
possible to evaluate the ground settlement by just calculating the tilt angle from only one single node
located at the center along the longitudinal or the transversal direction. Nevertheless, when a series
of sensor nodes are closely located on the ground from a distance away to the excavating center, a
smooth subsidence map can be achieved. Figure 13 presents a sketch showing the way of deploying
sensor nodes in order to draw a settlement curve in the cross section. Simplifying the subsided ground
surfaces between two adjacent sensor nodes are even, the absolute ground subsidence in the section
can be calculated via the tangent function presented as a polyline. After that, a smooth subsidence
curve can be obtained by interpolating the polyline (Figure 13). Anyhow, it has to be considered
that modern tunnel driving in urban areas and especially in the Shanghai areas is carried out with
of sensor nodes are closely located on the ground from a distance away to the excavating center, a
smooth subsidence map can be achieved. Figure 13 presents a sketch showing the way of deploying
sensor nodes in order to draw a settlement curve in the cross section. Simplifying the subsided
ground surfaces between two adjacent sensor nodes are even, the absolute ground subsidence in the
section can be calculated via the tangent function presented as a polyline. After that, a smooth
Sensors 2016, 16, 1109 13 of 15
subsidence curve can be obtained by interpolating the polyline (Figure 13). Anyhow, it has to be
considered that modern tunnel driving in urban areas and especially in the Shanghai areas is carried
large
out withexperience and considering
large experience larger safety
and considering margins
larger safety to reducetosettlements.
margins Therefore,
reduce settlements. ground
Therefore,
displacements
ground displacementsand tilt motion
and tiltare always
motion areconsiderably low.
always considerably low.

Variations of Projected KF Dip Angle of Node 102 After Kalman Filter (on the Track)
0.15
KF Dip Angle ACC
KF Dip Angle INC
Sn = 0.0174
0.1 Centralized KF Dip Angle
Average of ACC and INC

0.05
KF Dip Angle in Longitudinal Section/ °

0 Sn = 0.0373

-0.05

-0.1

-0.15 Sn = 0.0643

-0.2

-0.25

-0.3
12:04:23 12:14:25 12:24:27 12:34:29 12:44:31 12:54:33 13:04:36
Time

Figure 12.
Figure 12. Plots
Plots ofof Dip
Dip Angle
Angle Variations
Variations after
after Kalman
Kalman Filter
Filter ofof Node
Node 102102 (on
(on the Track) in
the Track) in the
the
Longitudinal Section
Longitudinal Section During
During the
the Second
Second Measuring
Measuring Hour.Hour. Blue
Blue and
and green lines are
green lines are results
results after
after
Kalman filter
Kalman filter for
for the
the accelerometer
accelerometer andand inclinometer
inclinometer respectively,
respectively, and
and the
the red
red line
line shows
shows the
the result
result
solved by
solved by aa centralized
centralized Kalman
Kalman filter
filter based
based on
on the
the tilt
tilt values
values from
from two
two sensors.
sensors. The
The standard deviation
standard deviation
of the red line is higher than that of the blue line and lower than that of the green line. The
of the red line is higher than that of the blue line and lower than that of the green line. The purple dot purple dot
line displays
line displays the
the average
average of
of the
the blue
blue and
and green
green lines.
lines.

Sensors 2016, 16, 1109 13 of 14

Finally, as Kalman filter is a recursive solution that can be continually applied to estimate the
situation in the next time step and then be updated after obtaining the observation, the a priori
estimate made by Kalman filter can be taken into account into a Building Information Model (BIM)
to support the management and facility operation. Furthermore, it can be used as a kind of early
warning systems to indicate changes in the ground response to the excavation.

Figure 13. Planned Distribution of Sensor Nodes along the Transversal Direction and Drawing of a
Figure 13. Planned Distribution of Sensor Nodes along the Transversal Direction and Drawing of a
Settlement Curve. The settlement is gradually accumulated from the node deployed at a distance
Settlement Curve. The settlement is gradually accumulated from the node deployed at a distance
showing no variations to the one at the excavating center on the ground. Starting with a value of zero,
showing no variations to the one at the excavating center on the ground. Starting with a value
a smooth
of curve ofcurve
zero, a smooth vertical groundground
of vertical settlement can becan
settlement obtained usingusing
be obtained the tangent function
the tangent and
function
interpolation.
and interpolation.

5. Conclusions
Finally, as Kalman filter is a recursive solution that can be continually applied to estimate the
In this
situation paper,
in the nextatime
one-day fieldthen
step and testbewas deployed
updated afteron top of athe
obtaining tunnel section the
observation, under construction
a priori estimate
and MEMS sensor nodes equipped with both accelerometers and inclinometers were used to measure
the changes of the surface caused by excavation. Normal vectors of the sensors were deduced using
rotation matrices and consequently were projected to the longitudinal section for the purpose of an
assessment of ground settlement. Aiming at an improved data reliability and a better estimate of the
inclination, a centralized Kalman filter was conducted based on the results from accelerometers and
Sensors 2016, 16, 1109 14 of 15

made by Kalman filter can be taken into account into a Building Information Model (BIM) to support
the management and facility operation. Furthermore, it can be used as a kind of early warning systems
to indicate changes in the ground response to the excavation.

5. Conclusions
In this paper, a one-day field test was deployed on top of a tunnel section under construction
and MEMS sensor nodes equipped with both accelerometers and inclinometers were used to measure
the changes of the surface caused by excavation. Normal vectors of the sensors were deduced using
rotation matrices and consequently were projected to the longitudinal section for the purpose of an
assessment of ground settlement. Aiming at an improved data reliability and a better estimate of
the inclination, a centralized Kalman filter was conducted based on the results from accelerometers
and inclinometers. Furthermore, a discussion based on the filtered result was provided afterwards to
evaluate the performance of the sensors and the process of tunneling.

Author Contributions: Cheng Li analyzed the data and wrote the paper, Tomás Fernández-Steeger provided the
data and contributed in writing and interpretation, Rafig Azzam took part in the final discussion.
Conflicts of Interest: The authors declare no conflict of interest.

References
1. Padhy, P.; Martinez, K.; Riddoch, A.; Ong, H.L.R.; Hart, J.K. Glacial environment monitoring using sensor
networks. In Real-World Wireless Sensor Networks; Springer: Stockholm, Sweden, 2005.
2. Cho, C.Y.; Chou, P.H.; Chung, Y.C.; King, C.T.; Tsai, M.J.; Lee, B.J.; Chou, T.Y. Wireless Sensor Networks for
Debris Flow Observation. In Proceedings of the 5th Annual IEEE Communications Society Conference on
Sensor, Mesh and Ad Hoc Communications and Networks, San Francisco, CA, USA, 16–20 June 2008.
3. Ramesh, M.V. Real-time Wireless Sensor Network for Landslide Detection. In Proceedings of the 3rd
International Conference on Sensor Technologies and Applications (Sensorcomm 2009), Athens, Greece,
18–23 June 2009; pp. 405–409.
4. Li, M.; Liu, Y.H. Underground Coal Mine Monitoring with Wireless Sensor Networks. ACM Trans.
Sensor Netw. 2009, 5. [CrossRef]
5. Roy, P.; Bhattacharjee, S.; Ghosh, S.; Misra, S.; Obaidat, M.S. Fire monitoring in coal mines using Wireless
Sensor Networks. In Proceedings of the 2011 International Symposium on Performance Evaluation of
Computer & Telecommunication Systems (SPECTS), Hague, The Netherlands, 27–30 June 2011.
6. Fernandez-Steeger, T.M.; Arnhardt, C.; Walter, K.; Niemeyer, F.; Nakaten, B.; Homfeld, S.D.; Asch, K.;
Azzam, R.; Bill, R.; Ritter, H.; et al. SLEWS—A Prototype System for Flexible Real Time Monitoring of
Landslides Using an Open Spatial Data Infrastructure and Wireless Sensor Networks. Geotechnol. Sci. Rep.
2009, 13, 3–15.
7. Azzam, R.; Arnhardt, C.; Fernandez-Steeger, T.M. Monitoring and Early Warning of Slope Instabilities
and Deformations by Sensor Fusion in Self-Organized Wireless ad-hoc Sensor Networks. In Proceedings
of the International Symposium and the 2nd AUN/Seed-Net Regional Conference on Geo-Disaster
Mitigation in ASEAN-Protecting Life from Geo-Disaster and Environmental Hazards, Yogyakarta, Indonesia,
25–26 February 2010.
8. May, M. X-SLEWS: Developing a Modular Wireless Monitoring System for Geoengineering Wide Area
Monitoring. Master Thesis, RWTH Aachen University, Aachen, Germany, March 2013.
9. Arnhardt, C. Monitoring of Surface Movements in Landslide Areas with a Self Organizing Wireless Sensor
Network (WSN). Ph.D. Thesis, RWTH Aachen University, Aachen, German, 2011.
10. Fernández-Steeger, T.; Ceriotti, M.; Link, J.Á.B.; May, M.; Hentschel, K.; Wehrle, K. “And Then, the Weekend
Started”: Story of a WSN Deployment on a Construction Site. J. Sensor Actuator Netw. 2013, 2, 156–171.
[CrossRef]
11. Li, C.; Tomás Manuel, F.-S.; Jó Ágila, B.L.; Matthias, M.; Rafig, A. Use of Mems Accelerometers/Inclinometers
as a Geotechnical Monitoring Method for Ground Subsidence. Acta Geodyn. Geomater. 2014, 11, 337–348.
[CrossRef]
Sensors 2016, 16, 1109 15 of 15

12. Canli, E.; Thiebes, B.; Engels, A.; Glade, T.; Schweigl, J.; Bertagnoli, M. Multi-parameter monitoring of a slow
moving landslide in Gresten (Austria). In Proceedings of the European Geoscience Union General Assembly,
Vienna, Austria, 12–17 April 2015.
13. Murata. SCA8X0 21X0 3100 Product Family Specification. Available online: http://www.murata.com/~/
media/webrenewal/products/sensor/accel/sca2100/sca8x0_21x0_3100_product_family_specification_
82_694_00f.ashx?la=en (accessed on 2 June 2016).
14. Pedley, M. Tilt Sensing Using a Three-Axis Accelerometer; Freescale Semiconductor Inc.: Austin, TX, USA, 2013.
15. Li, C. Multi-Sensor Data Fusion for Geohazards Early Warning System—An Adapted Process Model.
Ph.D. Thesis, RWTH Aachen University, Aachen, Germany, 2015.
16. Kalman, R.E. A new approach to linear filtering and prediction problems. J. Fluids Eng. 1960, 82, 35–45.
[CrossRef]
17. Welch, G.; Bishop, G. An Introduction to the Kalman Filter; University of North Carolina: Chapel Hill, NC,
USA, 1995.
18. Willner, D.; Chang, C.B.; Dunn, K.P. Kalman Filter Configurations for Multiple Radar Systems; NASA:
Washington, DC, USA, 1976.
19. Brown, C.M.; Durrant-Whyte, H.; Leonard, J.J.; Rao, B. Centralized and decentralized Kalman filter
techniques for navigation, and control. In Proceedings of the a workshop on Image understanding workshop,
Morgan Kaufmann Publishers Inc., Palo Alto, CA, USA, 23–26 May 1989; pp. 651–675.
20. Grewal, M.S.; Andrews, A.P. Kalman Filtering: Theory and Practice Using MATLAB, 3rd ed.; John Wiley & Sons:
Chichester, UK, 2008.
21. Brown, R.G.; Hwang, P.Y.C. Introduction to Random Signals and Applied Kalman Filtering, 3rd ed.;
John Wiley & Sons: New York, NY, USA, 1997.
22. Fourati, H. Multisensor Data Fusion: From Algorithms and Architectural Design to Applications; CRC Press:
Boca Raton, FL, USA, 2015.
23. Willner, D.; Chang, C.B.; Dunn, K.P. Kalman filter algorithms for a multi-sensor system. In Proceedings
of the IEEE Conference on Decision and Control including the 15th Symposium on Adaptive Processes,
Clearwater, FL, USA, 1–3 December 1976; pp. 570–574.
24. Carlson, N.A. Federated square root filter for decentralized parallel processors. IEEE Trans. Aerosp. Electron. Syst.
1990, 26, 517–525. [CrossRef]
25. Carlson, N.A.; Berarducci, M.P. Federated Kalman Filter Simulation Results. Navigation 1994, 41, 297–322.
[CrossRef]

© 2016 by the authors; licensee MDPI, Basel, Switzerland. This article is an open access
article distributed under the terms and conditions of the Creative Commons Attribution
(CC-BY) license (http://creativecommons.org/licenses/by/4.0/).