sensors

Article
A Self-Powered Wireless Water Quality Sensing
Network Enabling Smart Monitoring of Biological
and Chemical Stability in Supply Systems
Marco Carminati 1, * , Andrea Turolla 2, * , Lorenzo Mezzera 1 , Michele Di Mauro 2 ,
Marco Tizzoni 2 , Gaia Pani 2 , Francesco Zanetto 1 , Jacopo Foschi 2 and Manuela Antonelli 2
1 Dipartimento di Elettronica, Informazione e Bioingegneria (DEIB), Politecnico di Milano, Milano 20133, Italy
2 Department of Civil and Environmental Engineering (DICA)—Environmental Section, Politecnico di Milano,
Milano 20133, Italy
* Correspondence: [email protected] (M.C.); [email protected] (A.T.)




Received: 19 January 2020; Accepted: 14 February 2020; Published: 19 February 2020


Abstract: A smart, safe, and efficient management of water is fundamental for both developed and
developing countries. Several wireless sensor networks have been proposed for real-time monitoring
of drinking water quantity and quality, both in the environment and in pipelines. However, surface
fouling significantly affects the long-term reliability of pipes and sensors installed in-line. To address
this relevant issue, we presented a multi-parameter sensing node embedding a miniaturized slime
monitor able to estimate the micrometric thickness and type of slime. The measurement of thin
deposits in pipes is descriptive of water biological and chemical stability and enables early warning
functions, predictive maintenance, and more efficient management processes. After the description of
the sensing node, the related electronics, and the data processing strategies, we presented the results
of a two-month validation in the field of a three-node pilot network. Furthermore, self-powering
by means of direct energy harvesting from the water flowing through the sensing node was also
demonstrated. The robustness and low cost of this solution enable its upscaling to larger monitoring
networks, paving the way to water monitoring with unprecedented spatio-temporal resolution.

Keywords: wireless sensor network; smart pipe; biofilm; scaling; impedance; interdigitated
microelectrodes; energy harvesting

1. Introduction
Water is irrefutably a critical resource for mankind, with major sanitary and economical relevance.
Water availability and quality directly affect human health, nutrition, agriculture, and the environment.
Correct handling of the full cycle of water has an impact on marine ecosystems, hydrogeological
safety, and it is pivotal within the paradigm of smart cities, especially in terms of sustainability. In fact,
safer and more efficient management of water is fundamental for densely-inhabited metropolitan
contexts, along with other distributed infrastructures, monitoring air pollution, traffic and mobility,
waste management, etc.
Conventionally, water quality is monitored by periodic sampling and laboratory chemical
analysis. This approach grants excellent sensitivity and selectivity in detection, at the price of limited
spatio-temporal resolution [1]. In-line instruments, monitoring mostly electrochemical parameters, are
being introduced as well in a few locations. The current trends of miniaturization and integration of
functions enabled by the fast-paced development of consumer electronics and wireless technologies
provide suitable infrastructures for pervasive sensorization of the environment. Several wireless sensor
networks have been proposed to monitor water quality in real-time, both in urban [2] and remote

Sensors 2020, 20, 1125; doi:10.3390/s20041125 www.mdpi.com/journal/sensors

Sensors 2020, 20, 1125 2 of 16

scenarios [3], as well as in oceans [4–6], rivers [7–9], and water reservoirs [10,11]. Although several
types of network topologies, hierarchies, radio protocols, and data elaboration approaches [12–17]
have been proposed in the last decade, no consolidated solution has emerged so far.
In this work, we focused on the drinking water supply infrastructure, and we addressed one of
the main related issues: the biological and chemical stability of water in distribution systems [18,19].
These properties are of paramount importance for safe and reliable water supply. In fact, they must
be guaranteed in order to avoid the occurrence of many related problems, as worsening of water
quality resulting in microbiological risk [20], system failure due to pipe clogging [21], water loss due
to leakages in pipes [22], or sensors malfunctioning due to fouling [23,24]. In this view, real-time
monitoring of water biological and chemical stability could enable early warning functions, predictive
maintenance, and more efficient management processes in the distribution system. We proposed a
sensing node, combining conventional chemo-physical sensors with a novel impedance monitor of
the surface fouling, able to estimate its thickness and discriminate its origin (biological vs. inorganic).
It offers faster response and better sensitivity than low-cost optical approaches [25].
Along with above-cited problems, the increasing aging of the drinking water distribution
infrastructure, in several cases one-century old, poses severe issues of safety and maintenance. For
instance, in Italy, the fraction of water reaching domestic users dropped from 62.6% in 2012 to 58.6%
in 2015 [26], meaning that the average leakage loss of the network exceeds 40%, and the trend is not
improving. This situation produces side effects, such as a widespread and unmotivated suspicion
towards the quality of drinking water (in Italy, 29% of families avoid drinking tap water, as reported
by the national statistics institute [26]). Thus, one of the main motivations of this work, along with
providing the network manager with real-time monitoring instruments and maps, is to invert this
trend and improve the confidence of citizens towards the quality of tap water.
A key aspect of the pervasive diffusion of sensing nodes is their energetic autonomy. Water
and energy are strictly entangled: energy is used to treat and distribute water, while energy can
be also harvested from water flows, especially in order to monitor its quantity and quality [27].
A demonstration of this nexus is the flourishing of self-powered devices and networks for smart
metering, leveraging Internet-of-Things (IoT) electronics for monitoring water quality and consumption,
especially aiming at leakage reduction [28,29]. Energy can be harvested from the flow of fluids in several
ways [30,31], improving the efficiency, stability, and compactness commonly achieved, for instance,
by solar panels [8]. Here, we demonstrated the possibility to self-power the proposed sensing node by
means of an energy harvesting unit that enables its installation in remote sites, off the grid.
Finally, it must be underlined that an additional challenge in this context is posed by a limited
budget. This aspect has been already highlighted by other works [2,11]: clearly, a large-scale diffusion
of water wireless sensing networks can only be achieved if the unitary cost of the nodes is affordable.
A similar paradigm, in which the spatio-temporal resolution of environmental monitoring maps
is significantly improved by the diffusion of low-cost measuring units and participatory sensing,
based on emerging miniaturized technologies and complementary to the few (often fixed) governmental
reference monitoring stations, has been observed in monitoring air quality [32], radioactivity [33,34],
and local forecasting of rainfalls [35].

2. System Architecture
The architecture of the wireless sensor network that we realized is pictured in Figure 1. Each
sensing node measures in real-time six parameters of the water flowing inside the distribution pipe:
temperature, conductivity, pH, pressure, flow rate, and the thickness of slime depositing on the inner
pipe surface. Additional sensors, in particular optical and electrochemical ones, measuring, for instance,
transmittance, turbidity, dissolved oxygen, heavy metals, or residual chlorine can be easily added since
the sensing node is endowed with auxiliary analog inputs and outputs. A microcontroller manages the
operation of the sensing node and performs basic calibration and fault detection of sensors. Data are
periodically acquired and transmitted by means of a radio link to a Cloud server. For this application,
Sensors 2020, 20, 1125 3 of 16
Sensors 2020, 20, x FOR PEER REVIEW 3 of 16

detection of sensors. Data are periodically acquired and transmitted by means of a radio link to a
we selected the GSM/GPRS cellular network, given its ubiquitous coverage of the territory (98% of the
Cloud server. For this application, we selected the GSM/GPRS cellular network, given its ubiquitous
Italian territory
coverage byterritory
of the Telecom(98%Italia
ofprovider). Other low-power
the Italian territory by TelecomradioItaliainfrastructures
provider). Other oriented to IoT
low-power
applications, such as a LoRaWAN
radio infrastructures andapplications,
oriented to IoT Sigfox, which areasina the
such processand
LoRaWAN of consolidation,
Sigfox, which arecaninreplace
the
theprocess
GSM when they reach the same coverage, especially required for rural areas
of consolidation, can replace the GSM when they reach the same coverage, especially crossed by the water
required for
distribution rural areas
network. crossed
Security bycourse,
is, of the water distribution
another network.
key aspect to beSecurity is, ofaccount
taken into course,when
another key
selecting
theaspect
propertoradio
be taken into account when selecting the proper radio protocol.
protocol.

Figure 1. The architecture of the proposed network composed of sensing nodes (monitoring 6 chemo-
Figure 1. The architecture of the proposed network composed of sensing nodes (monitoring
physical water parameters) powered by kinetic energy harvesting and connected through long-range
6 chemo-physical water parameters) powered by kinetic energy harvesting and connected through
radio links to a Cloud server, providing visualization, feedback control, and analytics.
long-range radio links to a Cloud server, providing visualization, feedback control, and analytics.
ThingSpeak™ (Mathworks) was selected as a Cloud server since it offers enough resources for
ThingSpeak™ (Mathworks) was selected as a Cloud server since it offers enough resources for this
this pilot demonstration in its free version. It enables to display data (both raw and processed) on a
pilot demonstration in its free version. It enables to display data (both raw and processed) on a website
website (Figure 2a) as well as on portable devices through an app (Figure 2b). Data on the server can
(Figure 2a) as well as on portable devices through an app (Figure 2b). Data on the server can be: (i) sent
be: (i) sent to the user, typically the utility or agency managing the water distribution system, for
to the user, typically
continuous the utility
monitoring, (ii) usedorbyagency managing
automatic the water
controllers distribution
to activate actuators system,
in order for continuous
to operate in
monitoring, (ii) used by automatic controllers to activate actuators in order to operate
closed-loop sections of the network and act, for instance, by closing electro-valves in response to in closed-loop
sudden
sections of leakages
the networkor alterations
and act, forof the water quality
instance, parameters,
by closing due to in
electro-valves pipe degradation
response or pollution
to sudden leakages
events, (iii) of
or alterations sent
thetowater
a dataquality
analytics unit. Data due
parameters, can be
to processed by statistical
pipe degradation tools in multiple
or pollution ways
events, (iii) sent
to aand at several levels. They can be merged, correlated, and compared with additional heterogeneous
data analytics unit. Data can be processed by statistical tools in multiple ways and at several
dataThey
levels. provided
can bebymerged,
other complementary
correlated, andsources.
comparedTheywithcan be used to
additional initialize anddata
heterogeneous fit models
provided
describing,
by other for instance,
complementary the growth
sources. of bacteria
They can be used or to train machine
to initialize learning
and fit models models tofor
describing, achieve
instance,
thepredictive
growth ofmaintenance
bacteria or toof the
trainpipes and tolearning
machine optimizemodels
various to
treatment
achieveprocesses,
predictiveoften involving the
maintenance of the
injection of chemicals, such as disinfection by chlorine-based compounds.
pipes and to optimize various treatment processes, often involving the injection of chemicals, such as
disinfection by chlorine-based compounds.
The sensing node can be installed along the main pipeline (Figure 2c) or in a branch (Figure 2d),
depending on installation and materials constraints. Since our prototypes have not been certified in
terms of food compliance, they have been installed in branches for the pilot validation in the field. The
unit also includes a digital signal to control an external electro-valve that can be open only during
the measurement time, to minimize water loss. In our demonstration, a standard motorized ball
electro-valve for 12 ” pipes (SEV3000 operated at +5 V DC and consuming 25 mWh in a full open-close
cycle) has been employed.
The node can be powered by a +5 V DC power supply by a battery. It also comprises an energy
harvesting system, extracting power from a turbine placed along the same pipe and, thus, enabling the
energetic autonomy of the node. The turbine should be placed downstream of the sensors, in order to
minimize the perturbation of the hydrodynamics in the measurement sites.
 ThingSpeak™ (Mathworks) was selected as a Cloud server since it offers enough resources for
this pilot demonstration in its free version. It enables to display data (both raw and processed) on a
website (Figure 2a) as well as on portable devices through an app (Figure 2b). Data on the server can
be: (i) sent to the user, typically the utility or agency managing the water distribution system, for
continuous
Sensors monitoring, (ii) used by automatic controllers to activate actuators in order to operate 4inof 16
2020, 20, 1125
closed-loop sections of the network and act, for instance, by closing electro-valves in response to
sudden leakages or alterations of the water quality parameters, due to pipe degradation or pollution
The choice
events, of the
(iii) sent to asampling period
data analytics is strictly
unit. Data canrelated to the specific
be processed needs tools
by statistical of theinapplication and to
multiple ways
theand
energyat several levels. They can be merged, correlated, and compared with additional heterogeneous in
available for each node. The minimum sampling period is 15 s. The typical period used
thedata provided
validation hasby other
been complementary
1 min. sources.
In case real-time They can
reactivity is notberequired,
used to the
initialize and period
sampling fit models
can be
describing,
increased for instance,
to hours themonitoring,
for routine growth of thus
bacteria or to train
significantly machine
reducing thelearning
averagemodels to achieve
energy consumption,
predictive
since maintenance
the sensing of the pipes
node switches andmode
to sleep to optimize
whenvarious treatment processes, often involving the
not acquiring.
injection of chemicals, such as disinfection by chlorine-based compounds.

Figure 2. User interface of the ThingSpeak™ Cloud displaying in real-time acquired data on a
password-protected website (a) and on smartphone app (b). Possible installation approaches of the
sensing node: (c) in the main pipe or (d) in a derivation branch, whose flow can be controlled.

3. Sensors

3.1. Slime Sensor

The deposition of a micrometric layer of slime on the surface of the pipe is measured by means
of an impedance sensor. The sensor is based on closely-spaced microelectrodes, whose working
principle is depicted in Figure 3. In detail, impedance is measured across two combs of interdigitated
microelectrodes whose fingers (of width W and thickness te ) are separated by a spacing D. In order
to maximize sensitivity, D should be matched with the thickness H of the deposit to be monitored.
The equivalent small-signal impedance model (Figure 3b) is composed of the capacitance of the
electrochemical double layer (CDL ) in series to the ionic resistance (Rion ), scaling with the liquid bulk
conductivity (drinking water has usually a conductivity between 500 and 1000 µS/cm). No faradaic
component is present since the voltage applied across the interface (50 mV) does not activate any redox
reaction. The value of CDL scales with the area of the electrodes, while D sets Rion and the vertical
sensitivity [36]. Given the coplanar geometry, conformal mapping expressions for a finite layer are
used to estimate the value of Rion [37]. In [38], we compared measured values of Rion with the value
provided by conformal mapping and finite-element simulations, finding a good agreement.
When a layer of slime deposits, it alters both CDL and Rion . The displacement of ions due to the
presence of an insulating body on the electrodes typically produces a decrease of CDL and an increase
of Rion . This technique is extensively used, for instance, to electrically monitor in-vitro the growth of
colonies of biological cells [39]. Since the interfacial CDL is extremely sensitive to surface events, at the
sub-nanometer level, it is also very unstable over time; thus, we chose to monitor Rion . By measuring
impedance at a frequency fs larger than the corner frequency 1/(π·Rion ·CDL ), it is possible to shunt CDL
and directly probe Rion .
Experimental tests evidenced that the change of Rion , ∆R is linearly related to the thickness
H of the slime, both in the case of biological (biofilm) and inorganic (limestone scaling) deposits.
Interestingly, the slope of the response is the opposite: a decrease in the case of biological deposits
(where the conductivity of the colony extracellular matrix is dominant) and an increase in case of
inorganic deposits. After the proof-of-concept of the sensor [40], a significant amount of additional
development has been carried out to improve the robustness and integration of the device under
relevant operating conditions in the view of its application. Different materials and geometries have
 consumption, since the sensing node switches to sleep mode when not acquiring.

3. Sensors

3.1. Slime Sensor

Sensors 2020, 20, 1125 5 of 16
The deposition of a micrometric layer of slime on the surface of the pipe is measured by means
of an impedance sensor. The sensor is based on closely-spaced microelectrodes, whose working
been systematically
principle is depicted tested (Figure
in Figure 3. In3c). Theimpedance
detail, commercial microelectrodes
is measured (DropSense)
across two with D of 5 and
combs of interdigitated
10 microelectrodes
µm (fabricated by lithography) used in preliminary tests have been replaced with
whose fingers (of width W and thickness te) are separated by a spacing D. In order custom-designed
to maximize
electrodes sensitivity,
produced withDthe should be matched
conventional with the thickness
technology used forHprinted
of the deposit
circuit to be monitored.
boards (PCB). This
The equivalent
manufacturing small-signal
process impedance
has allowed model reduction
a significant (Figure 3b) in is composed
costs and moreof versatility
the capacitance
in the of the
designed
electrochemical double layer (CDL) in series to the ionic resistance (Rion), scaling with the liquid bulk on
geometries. These components are small, circular, gold-plated elements with electrical contacts
theconductivity
back of the(drinking
rigid FR-4 water has usually
substrate. a conductivity
Moreover, betweenof500
the geometries and 1000 μS/cm).
components No faradaic
have been defined so
as component
to allow theis housing
present since
at thethe voltage
inner applied
surface acrosswith
of pipes the interface
diameter(50 mV)to
down does
1”, not
andactivate any of
the spacing
redox reaction. The value of C DL scales with the area of the electrodes, while D sets
microelectrodes has been set so as to target the expected deposit thickness H. Two different spacingRion and the vertical
sensitivity
values [36]. chosen,
have been Given the coplanar geometry,
respectively of 125 and conformal
250 µm, in mapping
the viewexpressions
of dealingfor a finite
with layer are
the characteristic
used to estimate the
thickness of slime in pipes.value of Rion [37]. In [38], we compared measured values of Rion with the value

provided by conformal mapping and finite-element simulations, finding a good agreement.

Figure 3. Slime thickness monitor based on planar interdigitated microelectrodes (a), of thickness te ,
width W, and spacing D (b), fabricated in different technologies and geometries (c). Impedance is
measured at a single frequency fs .

Flexible electrodes, fabricated on Kapton substrates with the same geometries, have been also
successfully realized (Figure 3c). If bending takes place along the longitudinal orientation of the metal
fingers, the measured sensitivity is the same as of the rigid version [41]. Since they can be conformally
applied inside ducts with bending radii below 1 cm, they represent a viable solution for pipes of small
diameters, such as in industrial and domestic appliances. Their long-term stability in water has not yet
been assessed, but the chemical robustness of Kapton allows to envision no issues.
The calibration of the sensor takes place in two phases. First, the cell constant, relating conductivity
to the measured Rion and affected by geometrical tolerances, is extracted once after fabrication for
every single sensor. Three saline solutions at known conductivities are used for a simple linear
calibration. Then, in order to obtain the value of the slime thickness H from Rion , a calibration response
curve obtained in the laboratory, under controlled conditions, is used. Since Rion is a function of
the water salinity and temperature, and since the calibration curve is obtained at 25 ◦ C for a specific
conductivity (750 µS/cm), temperature and conductivity are simultaneously measured to correct the
impedance value.

3.2. Other Chemo-Physical Sensors

The custom-developed slime sensor has been complemented by five standard sensors to create the
multi-parameter sensing node. Temperature and pH are measured by a Sensorex S272CD probe. This
is the most expensive sensor and is critical in terms of long-term stability. The choice of a commercial
device is due to the higher reliability and precision with respect to non-consolidated miniaturized
pH technologies present in the literature. The pH probe is a flat glass membrane suitable for in-flow
Sensors 2020, 20, 1125 6 of 16

measurements, embedding also a thermistor. The instrument resolution is higher than 0.1 ◦ C between
0 and ◦ C and 0.1 between pH 0 and pH 14, respectively, for temperature and pH.
50 2020,
Sensors 20, x FOR PEER REVIEW 6 of 16
Conductivity is measured by a homemade low-cost miniaturized device. The development of a
dedicated Conductivity
instrumentishas measured by a homemade
been carried out in orderlow-cost
to improveminiaturized device. The
the characteristics development
of existing of a
commercial
dedicated
probes. instrument
Gold-coated has been
connector pins carried
are usedout for in order as
sensing, todescribed
improve in the[42].
characteristics
Fault detection of existing
of sensors
cancommercial
be achieved probes. Gold-coated
by physical connector
redundancy pins are usedapproaches
or model-based for sensing,(observers,
as described suchin [42]. Fault
as a Kalman
detection of sensors can be achieved by physical redundancy or model-based
Filter used in avionics [43]). In order to increase the self-diagnostic capacity of the system, multipleapproaches (observers,
such as a Kalman Filter used in avionics [43]). In order to increase the self-diagnostic capacity of the
identical conductivity sensors can be inserted. The geometry of the electrodes has been chosen to obtain
system, multiple identical conductivity sensors can be inserted. The geometry of the electrodes has
an fs below 100 kHz, allowing the use of a single-chip impedance detector (AD5933 by Analog Devices),
been chosen to obtain an fs below 100 kHz, allowing the use of a single-chip impedance detector
to save space and complexity. The electrodes are spaced by a few millimeters and are sensitive to the
(AD5933 by Analog Devices), to save space and complexity. The electrodes are spaced by a few
passage of mm-sized impurities and micro-bubble. In the initial prototype, a vibrating linear motor
millimeters and are sensitive to the passage of mm-sized impurities and micro-bubble. In the initial
hasprototype,
been inserted, allowing
a vibrating linear periodic
motor has shaking of the pins
been inserted, to remove
allowing micro-bubbles
periodic shaking of thefrompinsthe electrode
to remove
surface, whose detrimental presence on time tracking of conductivity
micro-bubbles from the electrode surface, whose detrimental presence on time tracking has been observed in the of
field
conductivity
tests in open tanks has been
[44]. observed
In the finalin the field teststhis
prototype, in open tankshas
feature [44]. In the
been final prototype,
removed since (i)this
thefeature
presence
has been removed since (i) the presence of bubbles is very limited in closed and pressurized pipes,
of bubbles is very limited in closed and pressurized pipes, (ii) an algorithm for discarding artifacts
(ii) on
based an algorithm for discarding
low-pass filtering artifacts based
and combining on low-pass
the signals filtering and
from multiple combining
probes has been theintroduced.
signals fromThe
multipleof
resolution probes has been introduced.
the conductivity probe isThe resolution
better than 20ofµS/cm
the conductivity
between 50 probe
and is better
2500 than 20
µS/cm, itsμS/cm
intrinsic
between
response 50 and
time 2500 μS/cm,
is sub-ms, and theits intrinsic
minimum response
samplingtimeperiod
is sub-ms,
is 10and
ms.the minimum sampling period
is The
10 ms.
flow rate is measured by a commercial device (FL-808 by Digiten), providing a signal, which is
The flow rate is using
frequency-modulated measured by aeffect
a Hall commercial
sensor. device (FL-808 byisDigiten),
The resolution providing
better than 0.1 L/sabetween
signal, which
2.5 and
is frequency-modulated using a Hall effect sensor. The resolution is better
80 L/s. The pressure is measured by a commercial device (3525VG1/40-1.2 by Elco), providing an analog than 0.1 L/s between 2.5
and 80 L/s. The pressure is measured by a commercial device (3525VG1/40-1.2 by
voltage signal for pressure in the range from 0 to 1.2 MPa. The resolution is 1.5% of the full-scale range. Elco), providing an
analog voltage signal for pressure in the range from 0 to 1.2 MPa. The resolution is 1.5% of the full-
3.3.scale range.Sensing Node
Complete

3.3.
A Complete
picture ofSensing Nodeassembled sensing node is visible in Figure 4. The placing of sensors is
the fully
modular: they are located in sequence, along a 200 commercial HDPE pipe, commonly used for drinking
A picture of the fully assembled sensing node is visible in Figure 4. The placing of sensors is
water applications. As visible, four probes are inserted transversally to the water flow, screwed in
modular: they are located in sequence, along a 2′′ commercial HDPE pipe, commonly used for
holders for 1 14 00 pipes for fast disassembly and inspection (sealing is granted by Teflon tape). The
drinking water applications. As visible, four probes are inserted transversally to the water flow,
design of the holder for the slime sensor is critical in terms of water tightness. The sensor PCB is glued
screwed in holders for 1 ¼′′ pipes for fast disassembly and inspection (sealing is granted by Teflon
on tape).
top ofThe
a plastic
designholder and sealed
of the holder with
for the resin.
slime The
sensor is adhesion of resin
critical in terms on thetightness.
of water holder isThe
crucial.
sensorThe
standard
PCB is plastic
glued onmaterial
top of aused forholder
plastic filament-based
and sealed3D printing
with (PLA)
resin. The has proved
adhesion to be
of resin on not suitableis for
the holder
thiscrucial. The standard plastic material used for filament-based 3D printing (PLA) has proved to be
application, and machined polypropylene has been used. The flow rate sensor, which not
can also
suitable
be used as for this application,
a harvesting and is
turbine, machined polypropylene Shielded
placed downstream. has been used.
cablesThe flow rate
connect thesensor, which
sensors to the
can alsounit,
electronic be used as ainharvesting
placed turbine,
a waterproof is placed
plastic downstream.
box. The GSM antenna Shielded cables connect
(ground-plane type)the sensors
is connected
to theaelectronic
through unit, and
coaxial cable placed in a waterproof
located above ground plastic
withbox. The GSMholder.
a magnetic antenna (ground-plane type) is
connected through a coaxial cable and located above ground with a magnetic holder.

Figure
Figure 4. Photograph
4. Photograph of of
thethe fully
fully assembled
assembled node,
node, showingthe
showing thein-line
in-linesensors,
sensors,the
theplastic
plasticholder
holderof
ofthe
the slime monitor, the electronic boards, and the GSM antenna.
slime monitor, the electronic boards, and the GSM antenna.

4. Electronics Design
A custom acquisition and processing electronic platform was designed. The design of the
electronics was led by the search for reliability and for the optimum balance between cost, simplicity,
Sensors 2020, 20, 1125 7 of 16

4. Electronics Design
A custom acquisition and processing electronic platform was designed. The design of the
electronics was
Sensors 2020, 20, xled
FORby theREVIEW
PEER search for reliability and for the optimum balance between cost, simplicity,
7 of 16
and compactness on one side and versatility and high-resolution performance on the other. In particular,
and compactness
special care was devoted on one to side and versatility
the design and high-resolution
of the analog performance
stages, which grants on the (by
the accuracy other. In
properly
particular,
choosing special
circuit care wasand
topologies devoted to the design
low-tolerance of the analog
components) stages,
and which grants
the resolution thelow
(i.e., accuracy
noise)(byof the
properly choosing circuit topologies and low-tolerance components) and the resolution
measurements. Three generations of the electronic platform were developed during the project. The (i.e., low
noise) of the measurements. Three generations of the electronic platform were developed during the
final system architecture is shown in Figure 5. The platform was composed of two units: the main unit
project. The final system architecture is shown in Figure 5. The platform was composed of two units:
performing conditioning, acquisition, processing, and transmission of signals from all the sensors and
the main unit performing conditioning, acquisition, processing, and transmission of signals from all
a separate power management unit taking care of the energy harvesting. Both units were based on
the sensors and a separate power management unit taking care of the energy harvesting. Both units
8-bit microcontrollers programmed in the Arduino development environment.
were based on 8-bit microcontrollers programmed in the Arduino development environment.

Figure
Figure 5. 5. Scheme
Scheme ofof the
the electronicplatform,
electronic platform,controlling
controllingthe
thenode:
node: ititisiscomposed
composed of
of the
the main
main board
board for
for conditioning, acquiring, and transmitting data and by a power management board for harvesting.
conditioning, acquiring, and transmitting data and by a power management board for harvesting.

TheThe main
main unit
unit waswas controlled
controlled by ATMega2560
by an an ATMega2560 microcontroller
microcontroller andand interfaces
interfaces withwith
eighteight
probes:
probes: pH, temperature, flow rate, pressure, two pairs of pin electrodes for conductivity, and two
pH, temperature, flow rate, pressure, two pairs of pin electrodes for conductivity, and two pairs of
pairs of interdigitated planar microelectrode for slime. The signals from the pH probe (temperature,
interdigitated planar microelectrode for slime. The signals from the pH probe (temperature, pH) were
pH) were acquired by a 24-bit sigma-delta analog-to-digital converted (AD7793 by Analog Devices),
acquired by a 24-bit sigma-delta analog-to-digital converted (AD7793 by Analog Devices), setting a
setting a temperature-compensated reference and reading the value from the sensor, properly
temperature-compensated reference and reading the value from the sensor, properly buffered at high
buffered at high impedance. The output of the flowmeter was frequency-modulated by the water
impedance. The to
speed, thanks output
a Hallofsensor,
the flowmeter
and it is was
readfrequency-modulated
as a digital signal. The by the water
output of thespeed, thanks
manometer to aanHall
was
sensor,
analog voltage (0–5 V), directly converted by the ADC embedded in the microcontroller, since(0–5
and it is read as a digital signal. The output of the manometer was an analog voltage its V),
performances
directly converted (resolution,
by the ADC linearity,
embeddedsampling rate)
in the were adequate.
microcontroller, Impedance
since measurement
its performances was
(resolution,
used for
linearity, both the rate)
sampling conductivity and the slime
were adequate. deposit.measurement
Impedance Impedance was wasmeasured
used forbyboth
means
the of a single
conductivity
andintegrated
the slimecircuit (AD5933
deposit. by Analog
Impedance was Devices)
measuredthat by
simultaneously stimulated
means of a single the electrodes
integrated circuit with
(AD5933a
bysine waveDevices)
Analog (100 kHz)that
andsimultaneously
performed an on-chip 12-bit the
stimulated FFTelectrodes
analysis with
with1 ms time wave
a sine resolution.
(100 With
kHz)aand
low-parasitics
performed multiplexer,
an on-chip the microcontroller
12-bit FFT analysis with 1 selected
ms time the correct channel
resolution. and extracted the
With a low-parasitics value
multiplexer,
of impedance, which was related to the physical dimensions of the electrodes. In
the microcontroller selected the correct channel and extracted the value of impedance, which was a previous version
[18], a more sophisticated analog demodulation solution (synchronous lock-in demodulation) was
adopted to operate with fs = 2 MHz due to microelectrodes’ dimensions. The choice to unify the circuit
for conductivity and slime deposit greatly simplified the system with no performance degradation.
Sensors 2020, 20, 1125 8 of 16

related to the physical dimensions of the electrodes. In a previous version [18], a more sophisticated
analog demodulation solution (synchronous lock-in demodulation) was adopted to operate with
fs = 2 MHz due to microelectrodes’ dimensions. The choice to unify the circuit for conductivity
and slime deposit greatly simplified the system with no performance degradation. Furthermore, to
increase the versatility, an interface for an additional electrochemical probe was implemented. A 12-bit
SPI DAC (AD5721 by Analog Devices) was coupled with a 12-bit single-channel ADC (AD7091 by
Analog Devices).
Robustness to firmware faults (especially caused by issues in the GSM communication) was
achieved by means of a watchdog on board (MAX6746, by Maxim). External watch-dog boards could
be used for further robustness. Additionally, an SD memory and a real-time clock, providing time
stamps, offered a memory buffer for data logging and in case of delays in the radio communication.
Double wireless communication technology was implemented in the platform for long-range
communication. A very robust GSM module (M10 by Quectel) performed a connection through the
GSM (LTE) network. In addition, a low-power LoRaWAN module (mDot by Multitech) was included,
requiring, in this case, the presence of the LoRa radio infrastructure. The interface between the main
board and the modem was a standard serial digital bus, thus open to other types of protocols, such as
Sigfox. The communication was mostly directed from the node toward the Cloud server. However,
thanks to the “talk back” function of ThingSpeak™, it was possible to remotely set a few parameters of
the board.
The main board was powered at +5 V, regulated by a filtered low-drop linear regulator. The battery
chosen in the final prototype was a 6 V, 1200 mAh Sealed Lead-Acid battery (by RS Components).
This type of battery is well consolidated and suited for IoT and industrial applications. Low cost,
float charge for prolonged periods, and a large number of recharging cycles are three of the main
advantages, especially for remote installation and long-term operation. A modular approach was
chosen here, in order to adapt to different turbines and batteries (thus addressing different installation
conditions). The energy harvesting unit was thus realized on a separate board, managed by a dedicated
microcontroller (ATMega328p, an 8-bit very low-power and low-cost microcontroller). The battery
management electronics was based on a DC/DC converter required since the input voltage coming
from the turbine varied from 5 to 12 V, and the charging voltage switched from 2.4 V/cell (7.2 V in
this case), during the charging phase, to 2.13 V/cell (6.4 V), during the float charge phase, when the
battery was almost fully charged. A buck-boost DC/DC converter (by Haljia), switching at 180 kHz
with adjustable output (5.5–30 V) and LCD display, was selected. A sensing resistor, coupled with a
sense current amplifier (MAX4373 by Maxim), allowed the measurement of the value of the current
absorbed by the battery, and the unit adaptively set the correct value of the charging voltage among
the four current profiles (about 30, 45, 70, and 90 mA), using as input also the data from the flow rate
probe. The physical realization of the compact boards is visible in Figure 4.

5. Experimental Results and Discussion

5.1. Laboratory Characterization of the Slime Monitor

In order to extend the readiness level of the slime monitor from the initial laboratory proof of
feasibility [40] to a degree suitable for operation in the field, robustness had to be primarily addressed.
For application in the field, the slime sensor (fabricated in a standard rigid PCB technology with D = W
= 250 µm) was identified as the best option, given the expected thickness of deposits in the range of a
fraction of mm. Its robustness was assessed during long-term measurement, and stress tests carried
out in the laboratory under various conditions of temperature (up to 50 ◦ C), pressure (up to 5 bar),
and flow velocity on the surface of the sensor (up to 1.5 m/s) by a six-month continuous monitoring
test performed with two hydraulic loops: a low-pressure (4 bar, Figure 6a) and a high-pressure (9 bar,
Figure 6b) one. The former was used to characterize only slime sensors, while the latter for the
validation of the node, fully equipped with all probes. Different and independent operating conditions
Sensors 2020, 20, 1125 9 of 16

could be determined in each channel of the low-pressure loop: flow velocity of 1, 3, and 5 cm/s
and pressure from 0 to 4 bar. Water temperature could be controlled and reagents—to promote the
formation or the cleaning-up of slime—were dosed in the tank reactor. Slime sensors were mounted
on holders similar to the final ones (Figure 5) and could be easily inserted into dedicated sockets
Sensors 2020, 20, x FOR PEER REVIEW 9 of 16
along the channels and aligned with the inner surface of pipes. Each channel could host up to four
Sensors 2020,
holders, 20, x FOR
in could
which PEER REVIEW
polymeric or metallic to9simulate
of 16
channel host up to four holders, element
in whichcould be inserted
polymeric in the
or metallic place of
element sensors,
could be inserted in
thethe
behavior
place of sensors, to simulate the behavior of the inner surface of pipes of different materials. A be
of the inner surface of pipes of different materials. A total of 24 samples could
channel could host up to four holders, in which polymeric or metallic element could be inserted in
simultaneously subjected
total of 24 samples couldtobecontrolled fouling.
simultaneously subjected to controlled fouling.
the place of sensors, to simulate the behavior of the inner surface of pipes of different materials. A
total of 24 samples could be simultaneously subjected to controlled fouling.

Figure
Figure 6. Laboratory
6. Laboratory setups
setups for dynamic
for dynamic and continuous
and continuous characterization
characterization of theof the sensors
sensors in the
in the hydraulic
loophydraulic loop withconditions
with controlled controlledof
conditions of temperature,
temperature, flow,concentration,
flow, chemical chemical concentration, and pressure:
and pressure: up to 4 bar
Figure 6. Laboratory setups for dynamic and continuous characterization of the sensors in the
up to
in the 4 bar in the low-pressure
low-pressure loop (a) and loop
up to(a) andin
9 bar upthe
to 9high-pressure
bar in the high-pressure
one (b). one (b).
hydraulic loop with controlled conditions of temperature, flow, chemical concentration, and pressure:
up to 4 bar in the low-pressure loop (a) and up to 9 bar in the high-pressure one (b).
AsAs anan example,the
example, theresponse
response curve
curve of
of the
the sensor
sensor to
toinorganic
inorganicscaling
scalingis is
reported
reported in Figure
in Figure7. 7.
Here,
Here, aAsa wide
wide range
range of of limestone
limestone thicknesseswas
thicknesses wasexplored,
explored,from
fromaafew
fewmicrons
microns to to 300
300 µm,
μm, achieved
achieved by
an example, the response curve of the sensor to inorganic scaling is reported in Figure 7.
by dosing
dosing the the volume
volume of a CaCO 3-saturated solutionof
ofwater
water whose
whose evaporation was forced ononthethe
Here, a wide rangeofofalimestone
CaCO 3 -saturated
thicknessessolution
was explored, from a fewevaporation
microns to 300 wasμm, forced
achieved
electrodes’ area. For thicknesses below 8 μm, an Atomic Force Microscope [40] was used to measure
by dosingarea.
electrodes’ the volume of a CaCO
For thicknesses 3-saturated
below 8 µm, solution
an Atomicof water whose evaporation
Force Microscope [40] was was forced
used on the H,
to measure
H, while for larger thicknesses, a gravimetric technique (combined with optical microscopy to
electrodes’
while for largerarea. For thicknesses
thicknesses, below 8 μm,
a gravimetric an Atomic
technique Force Microscope
(combined [40] was
with optical used to measure
microscopy to estimate
estimate
theH,fraction the fraction of covered area) was adopted.
while50
forcovered
of larger thicknesses, a gravimetric technique (combined with optical microscopy to
area) was adopted.
estimate45the fraction
A of covered area)
B was adopted. C A H < te = 35
50
40
45 A B C A H < te = 35
35
ΔR [Ω]ΔR [Ω]

40
30
35 35 < H < D = 250
Slope 0.12 Ω/µm B
Resistance

25
30
20 B 35 < H < D = 250
Slope 0.12 Ω/µm
Resistance

25
15
20
10
15 C H > 250
5
10
0 C H > 250
5 0 25 50 75 100 125 150 175 200 225 250 275 300 325
35 Limestone Thickness H [µm]
0
0 25 50 75 100 125 150 175 200 225 250 275 300 325
35 Limestone Thickness H [µm]
Figure 7. The response curve of the slime monitor for increasing thickness H of limestone, showing
three7.regions
The of sensitivity. of the slime monitor for increasing thickness H of limestone, showing
Figure
Figure 7. Theresponse
responsecurve
curve of the slime monitor for increasing thickness H of limestone, showing
three
three regions
regions ofofsensitivity.
sensitivity.
As expected, three different regions (A–C) could be identified in the response curve. In the first
region (A), when H was smaller than te (35 μm), slime only filled the gap between the electrodes, i.e.,
As expected, three different regions (A–C) could be identified in the response curve. In the first
in the region where the electric field was the highest (electric field lines were horizontal and parallel
region (A), when H was smaller than te (35 μm), slime only filled the gap between the electrodes, i.e.,
as in a parallel-plate geometry). When H exceeded 35 μm (B), the layer of slime covered the
in the region where the electric field was the highest (electric field lines were horizontal and parallel
electrodes, and the sensitivity was reduced to +0.12 Ω/μm. This number could be compared with +1.4
as in a parallel-plate geometry). When H exceeded 35 μm (B), the layer of slime covered the
Ω/μm obtained with the same CaCO3 deposition achieved with microelectrodes on glass: in that case,
electrodes, and the sensitivity was reduced to +0.12 Ω/μm. This number could be compared with +1.4
sensitivity was an order of magnitude larger, thanks to the smaller D = 10 μm [40]. However, that
Ω/μm obtained with the same CaCO3 deposition achieved with microelectrodes on glass: in that case,
Sensors 2020, 20, 1125 10 of 16

As expected, three different regions (A–C) could be identified in the response curve. In the first
region (A), when H was smaller than te (35 µm), slime only filled the gap between the electrodes,
i.e., in the region where the electric field was the highest (electric field lines were horizontal and
parallel as in a parallel-plate geometry). When H exceeded 35 µm (B), the layer of slime covered the
electrodes, and the sensitivity was reduced to +0.12 Ω/µm. This number could be compared with
+1.4 Ω/µm obtained with the same CaCO3 deposition achieved with microelectrodes on glass: in that
case, sensitivity was an order of magnitude larger, thanks to the smaller D = 10 µm [40]. However,
that sensitivity saturated for very thin layers above 10 µm. In region C, when H exceeded D (250 µm),
the sensitivity was further reduced, since 95% of the electric field lines extended up to a thickness
equal to D. Consequently, the recommended operating regions were A and B, confirming the optimal
sizing in matching D with the maximum H to be measured. In the reference application of drinking
water pipes, the typical values of H for scaling were in the order of hundreds of microns. Assuming
a measurement accuracy better than 1 Ω, the resolution and accuracy in slime thickness estimation
were better than 10 µm. This piece-wise linear and compressing response curve could be leveraged to
accommodate a large dynamic range: for very thin H, a larger “gain” was provided by region A, while,
for thicker H, the gain was reduced in region B. In other words, the monitor was typically operated in
region B, but, for very thin layers, it could “zoom in” when H was in region A.

5.2. Field Validation of a Pilot Network

The validation in the field of the proposed network took place in a drinking water distribution
system in Emilia Romagna region, thanks to the support of Romagna Acque—Società delle Fonti,
one of the major Italian drinking water utility, supplying water to 1.1 million citizens through a water
distribution network of 604 km. Three nodes were installed to form a pilot network for demonstration
purposes. Their location is shown in Figure 8: it was carefully chosen together with the network
managers to be representative of a relevant case study, despite the limited number of nodes. Node 1
was installed on the pipeline transporting water from Capaccio plant, treating water from the dam
of Ridracoli. Node 2 was installed on the pipeline transporting water from the Standiana plant,
treating water from a surface canal connected to the river Po. Node 3 was installed on the junction line
of the two pipelines, in which water was mixed and transported to the distribution network. Since the
installations were conducted on branches of main pipelines characterized by consistent flow rates and
high pressure (up to 12 bar), pressure regulators were interposed, setting a pressure of 2 bar and a flow
rate of 5 L/min in each node. All nodes were installed inside stations (for inspection and monitoring)
with ease of access. They were all powered by the power supply, and some tests of energy harvesting
were performed on Node 3 (the turbine is visible in Figure 8).
Data were recorded along different time frames: here, we reported data continuously acquired
for 50 days from 7 July to 26 August 2018. Due to the sensitive nature of monitoring data from field
validation, the complete set of acquired data could not be reported in the present work.
Figure 9 reports the monitoring of water temperature variations in the three nodes. From the time
tracking of Figure 9a, it is evident that the temperature in the three points was significantly different:
10 ◦ C in Node 1, 28 ◦ C in Node 2, and around an average of 22 ◦ C in Node 3. The latter showed
significant variations with daily periodicity, as highlighted in Figure 9b. Since the water in the pipeline
monitored by Node 3 was a mixture of water from the other two pipelines, continuous modifications
in the values of temperature were due to the changing ratio of blending flow rates. It was possible
to observe that the temperature increased during the night. Such behavior was confirmed by the
water utility staff, who reported the common practice of increasing the flow rate of water from the
Standiana plant during the night to preserve water in the Ridracoli dam. Furthermore, the practice of
water pumping in the basin was sometimes adopted to maximize hydroelectric energy production
during the day, when electricity demand increased. A strong correlation between the temperature of
input Node 1 and output Node 3 confirmed the mixing in the junction. This example demonstrated
 managers to be representative of a relevant case study, despite the limited number of nodes. Node 1
was installed on the pipeline transporting water from Capaccio plant, treating water from the dam
of Ridracoli. Node 2 was installed on the pipeline transporting water from the Standiana plant,
treating water from a surface canal connected to the river Po. Node 3 was installed on the junction
line of the two pipelines, in which water was mixed and transported to the distribution network.
Sensors 2020, 20, 1125 11 of 16
Since the installations were conducted on branches of main pipelines characterized by consistent flow
rates and high pressure (up to 12 bar), pressure regulators were interposed, setting a pressure of 2
thatbar and a flow
real-time rate of 5 monitoring
continuous L/min in each
of node. All nodes
even basic were installed
parameters inside statistical
(and simple stations (for inspection
analysis) could
and monitoring) with ease of access. They were all powered by the power supply, and some tests of
provide relevant information about the operation of the distribution network.
energy harvesting were performed on Node 3 (the turbine is visible in Figure 8).

Sensors 2020, 20, x FOR PEER REVIEW 11 of 16

the pipeline monitored by Node 3 was a mixture of water from the other two pipelines, continuous
modifications in the values of temperature were due to the changing ratio of blending flow rates. It
was possible to observe that the temperature increased during the night. Such behavior was
confirmed by the water utility staff, who reported the common practice of increasing the flow rate of
water from the Standiana plant during the night to preserve water in the Ridracoli dam. Furthermore,
the practice of water pumping in the basin was sometimes adopted to maximize hydroelectric energy
production during the day, when electricity demand increased. A strong correlation between the
temperature
Figure
Figure of
8. 8.Map
Mapinput
ofoftheNode
the 1 and output
installation
installation Node
sites of
sites of 3 confirmed
the three
the three nodes the mixing
nodes during
during the in the
thefield junction.
validation
field validationin This
a example
a water
in water
demonstrated
distribution that real-time
network in continuous
Romagna, Italy. monitoring
They are of
located even
in basic parameters
Y-junction.
distribution network in Romagna, Italy. They are located in Y-junction. (and simple statistical
analysis) could provide relevant information about the operation of the distribution network.
Data were recorded along different time frames: here, we reported data continuously acquired
for 50 days(a)
from357 July to 26 August 2018. Due to the sensitive nature of monitoring data from field
validation, the complete
30 set of acquired data could not be reported in the present work.
Temperature [°C]

Figure 9 reports
25 the monitoring of water temperature variations in the three nodes. From the
time tracking of20Figure 9a, it is evident that the temperature in the three points was significantly
different: 10 °C 15in Node 1, 28 °C in Node 2, and around an average of 22 °C in Node 3. The latter
showed significant10 variations with daily periodicity, as highlighted in Figure 9b. Since the water in
5 Node1 Node2 Node3
0
09/07/2018 18/07/2018 27/07/2018 05/08/2018 14/08/2018 23/08/2018
Day
(b) 35 (c) 24.5
30 24.0
Temperature [°C]

23.5
25 23.0
T Node 3 [°C]

20 22.5
22.0
15 21.5
10 21.0
20.5
5 Node1 Node2 Node3 20.0
0 19.5
0 4 8 12 16 20 24 10.6 10.8 11.0
Hour of the day T Node1 [°C]

Figure 9. Temperature values recorded for the three nodes for 50 days (a), and elaboration, showing
dailyFigure 9. Temperature values recorded for the three nodes for 50 days (a), and elaboration, showing
variations (b) and correlation (c) between Node 1 and 3.
daily variations (b) and correlation (c) between Node 1 and 3.
As an example of the validation of the self-diagnostic function of the node in the case of physically
As an example of the validation of the self-diagnostic function of the node in the case of
redundant sensors, three identical conductivity probes were installed in Node 3. The temporal
physically redundant sensors, three identical conductivity probes were installed in Node 3. The
evolution of water conductivity, normalized to the initial value for the three probes, is shown in
temporal evolution of water conductivity, normalized to the initial value for the three probes, is
Figure 10. It was possible to notice the strong agreement of the three couples of pins in the first part of
shown in Figure 10. It was possible to notice the strong agreement of the three couples of pins in the
the data series, whose averaged value was slightly higher with respect to the raw data as a consequence
first part of the data series, whose averaged value was slightly higher with respect to the raw data as
of the correction for
a consequence of temperature,
the correctionalso resulting in an
for temperature, alsoattenuation
resulting inofanthe characteristic
attenuation of thedaily trend. The
characteristic
second part of the data series was characterized by an accentuated drift in values of conductivity
daily trend. The second part of the data series was characterized by an accentuated drift in values of
conductivity
measured measured
by a specific by a specific
couple of pins couple
(Probe of
1).pins
Such(Probe 1). Such
behavior was behavior
due to awas due to a malfunction
malfunction induced on
induced on purpose, since a non-gold-coated couple of pins was used in this case, inducing the of
purpose, since a non-gold-coated couple of pins was used in this case, inducing the occurrence
occurrence of surface modification phenomena. It was interesting to note that the algorithms
automatically excluded the diverging signal and applied noise reduction functions, until performing
an automatic recalibration procedure of the diverged couple of pins using the measured data from
the other couples, once its signal was stabilized. No significant variations in the values of Rion were
observed during the field validation, indicating that water was biologically and chemically stable,
Sensors 2020, 20, 1125 12 of 16

surface modification phenomena. It was interesting to note that the algorithms automatically excluded
the diverging signal and applied noise reduction functions, until performing an automatic recalibration
procedure of the diverged couple of pins using the measured data from the other couples, once its
signal was
Sensors stabilized.
2020, NoREVIEW
20, x FOR PEER significant variations in the values of Rion were observed during12the field
of 16
Sensors 2020,
validation, 20, x FOR PEER
indicating thatREVIEW 12 formed.
water was biologically and chemically stable, and no deposits were of 16
3
3 Probe1
Conductivity

2.5 Probe1
Conductivity

2.5 Probe2
2 Probe2
Probe3
2
1.5 Probe3
Corrected average
Normalized

1.5 Corrected average

1
Normalized

1
0.5
0.5
0
09/07/2018
0 19/07/2018 29/07/2018 08/08/2018 18/08/2018 28/08/2018
09/07/2018 19/07/2018 29/07/2018Day 08/08/2018 18/08/2018 28/08/2018
Day
Figure 10. Normalized conductivity tracking for 3 identical probes in Node 3, showing fault detection
Figure
and
Figure 10.10. Normalized
isolation 1conductivity
of Probeconductivity
Normalized tracking
(whose faulttracking for 33 identical
was stimulated
for probes
probesininNode
on purpose).
identical Node3, 3,
showing fault
showing detection
fault detection
and isolation of Probe 1 (whose fault was stimulated on purpose).
and isolation of Probe 1 (whose fault was stimulated on purpose).
5.3. Energy Harvesting
5.3.5.3. Energy
Energy Harvesting
Harvesting
One of the main differences between the final version of the electronics here presented, and
One One
previous of the
of ones
the main main
[38,44] differences
was power
differences between
consumption,
between thewhich
the final final version
version theofelectronics
hasofbeen the electronics
significantly here (more
reduced
here presented,
presented, and
thanprevious
and 30%
previous ones [38,44] was power consumption, which has been significantly reduced
from the initial 1.5 W). Figure 11 reports the current consumption measured with a 7-digit digital the
ones [38,44] was power consumption, which has been significantly reduced (more (more
than 30%than 30%
from
from1.5
initial the
multimeter initial
W). 1.5 11
(DMM7510
Figure W).reports
Figure
by 11 reports
Keithley,
the the current
USA, sampling
current consumption at consumption
1 MSa/s)
measuredalongmeasured with digital
a fulla operation
with 7-digit a cycle
7-digitof digital
about
multimeter
multimeter
1 min. In the
(DMM7510 (DMM7510
by sleep by
phase,USA,
Keithley, Keithley,
the average USA,
sampling sampling
current at
at 1consumption 1
MSa/s) along MSa/s)
was along
a around a full operation
3 mA. When
full operation cyclethecycle
ofnode of about
about woke
1 min.
1 min.
In up,
the the In
sleep the sleepthe
acquisition
phase, phase,
of the signals
sensors
average average current
took
current consumption
10 s at
consumption120 mA. was was
Then, around
the
around GSM
3 mA.3modem
mA. When
When wasthethenode
node
turned on:woke
wokethis up,
up,
thewas the
theacquisition
dominant
acquisition of sensors
source
of sensors signals
of power
signals took 10120
tookdissipation.
10 s at s atConnection
120 mA.
mA. Then, Then,
to
thethethe
GSM GSM
network
modemmodem
andwas was turned
transmission
turned on:
ofthis
on: thiswas
data
was the
thetodominant dominant
the ThingSpeak™ source
source ofserver of power
powertook dissipation.
15 s (at 250
dissipation. Connection
mA), followed
Connection to
to theby the network
additional
network and10 and transmission
s for disconnection
transmission of data
of data (at to the
to the
120 mA).ThingSpeak™
If we server atook
considered 15 s between
period (at 250 mA), followed by
transmissions of additional
100 min, 10 s for
which disconnection
was reasonable (at
for
ThingSpeak™ server took 15 s (at 250 mA), followed by additional 10 s for disconnection (at 120 mA).
120 mA). If these
monitoring we considered
parameters, a period between
the average powertransmissions
consumption of of100the
min,
nodewhich
was was
42 mWhreasonable for
If we considered a period between transmissions of 100 min, which was reasonable for(30 mWh
monitoring
monitoring
during thesemode,
the sleep parameters, the average
and 12 mWh duringpower consumption
the active time). Theof thecompact
very node was 42 mWh (97 ×(30
24 ×mWh
these parameters, the average power consumption of the node was 42 mWh (30battery mWh during the51.5
sleep
during
mm the sleep
3) inserted in mode,
the node andhad12 mWh during
a capacity of the
7200 active
mWh time).
that The
would very compact
allow battery (97
the operation of3×the
24 ×node
51.5
mode, and
mm73)days
12 mWh
inserted
during the active time). The very compact battery (97 × 24 × 51.5 mm ) inserted in
inh).
the node had a capacity of 7200 mWh that would allow the operation of the node
for (171.4
thefor
node
7 days (171.4 h). of 7200 mWh that would allow the operation of the node for 7 days (171.4 h).
had a capacity

Figure
Figure 11.Recording
11. Recordingof
ofthe
the node
node current
current consumption
consumptionduring
duringanan
operation cycle.
operation cycle.
Figure 11. Recording of the node current consumption during an operation cycle.
Electric power could be extracted from the kinetic energy of water flowing the pipe to
Electric or
continuously power could be
periodically extracted
recharge from the
the battery. kinetic commercial
A low-cost energy of water
device flowing
was testedtheto pipe to
harvest
continuously
energy: or periodically
a Gaoxing Tech F-50recharge thenominal
turbine (a battery. 12
A low-cost commercial
V-10 W turbine). Thedevice
turbinewas tested
was to harvest
preliminarily
energy: a Gaoxing
characterized in theTech F-50 turbine
laboratory: (a nominal
it provided 12 V-10 W
a maximum turbine).
current ThemA
of 110 turbine
with was preliminarily
a flow rate of 20
characterized in the laboratory: it provided a maximum current of 110 mA with a flow rate of 20
Sensors 2020, 20, 1125 13 of 16

Electric power could be extracted from the kinetic energy of water flowing the pipe to continuously
or periodically recharge the battery. A low-cost commercial device was tested to harvest energy:
Sensors 2020,
a Gaoxing Tech 20,F-50
x FOR PEER REVIEW
turbine (a nominal 12 V-10 W turbine). The turbine was preliminarily characterized 13 of 16

in the laboratory: it provided a maximum current of 110 mA with a flow rate of 20 L/min. Figure 12
L/min. Figure 12 shows the display of the DC/DC converter with the values of its voltage and current
shows the display of the DC/DC converter with the values of its voltage and current outputs for two
outputs for two values of water flow rate in the node. In both cases, the power management board
values of water flow rate in the node. In both cases, the power management board correctly set the
correctly set the output voltage to 6 V and automatically selected the proper current value to recharge
output voltage The
the battery. to 6 efficiency
V and automatically
of the DC-DC selected the proper
converter is 90%, current
while the value to recharge
efficiency of the the battery.
turbine in theThe
efficiency of the
conversion fromDC-DC converter
kinetic is 90%,
to electrical power while
provedthetoefficiency
be 50% for ofthis
therange
turbine in the
of flow conversion
rates. During thefrom
kinetic to electrical power proved to be 50% for this range of flow rates. During
100-min cycle, the battery discharged of about 1% that could be easily recovered in 1 min of water the 100-min cycle,
the flow at rates
battery below 10
discharged ofL/min.
about This demonstrated
1% that the full
could be easily self-powering
recovered operating
in 1 min of water mode
flowofat
the sensing
rates below
node. This demonstrated the full self-powering operating mode of the sensing node.
10 L/min.

Figure
Figure 12. 12. Operation
Operation of of
thethe energyharvesting
energy harvestingunit,
unit, showing
showing the
the display
displayofofthe
theDC/DC
DC/DCconverter forfor
converter
two values of the water flow rate Q in the pipe. The power manager sets an output voltage
two values of the water flow rate Q in the pipe. The power manager sets an output voltage 6 V and 6 V and
selects
selects thethe optimal
optimal battery
battery chargingcurrent.
charging current.

6. Conclusions
6. Conclusions
WeWe presented
presented a multi-parameterwireless
a multi-parameter wirelesssmart
smart sensing
sensing node
node forforreal-time
real-timemonitoring
monitoring of of
water
water
quantity
quantity andand quality.The
quality. Thetwotwoprincipal
principal features
features of of this
this system
system are:are: (i)(i)self-diagnostic
self-diagnostic capability,
capability,
thanks
thanks to physical
to physical redundancy
redundancy andandmodel-based
model-basedfault fault detection,
detection, alongalongwith
withaanovel
noveland andquantitative
quantitative
slime measuring device, (ii) energetic autonomy, thanks to an efficient energy harvesting solution,
slime measuring device, (ii) energetic autonomy, thanks to an efficient energy harvesting solution,
applicable to compact commercial turbines. Furthermore, the absence of expensive components in
applicable to compact commercial turbines. Furthermore, the absence of expensive components in the
the system would enable a pervasive diffusion of sensing nodes. The total cost of the node hardware,
system would enable a pervasive diffusion of sensing nodes. The total cost of the node hardware, at a
at a prototypical scale, is below 1000 €. This could be further reduced by the medium-scale
prototypical scale, is below 1000 €. This could be further reduced by the medium-scale manufacture of
manufacture of an engineered product. This cost was comparable with other standard IoT nodes for
an engineered
environmental product. This cost
monitoring. was comparable
The additional hardware withcostother
of thestandard IoT nodes
slime monitor for environmental
is negligible, while the
monitoring. The additional hardware cost of the slime monitor is negligible,
potential savings enabled by predictive and smart maintenance of pipelines are significant. while the potentialIt savings
must
be underlined that the dominant term in the total cost of each node was clearly the installation. From
enabled by predictive and smart maintenance of pipelines are significant. It must be underlined that
the this point ofterm
dominant view,in in
thethe most
total costcommon
of eachcase
node ofwasexisting infrastructures,
clearly the installation. the installation of nodes
From this point in
of view,
branches
in the already present
most common in treatment
case of existing stations and
infrastructures, theinspection
installation wells
of nodes appeared as the already
in branches most
straightforward
present in treatmentsolution.
stations and inspection wells appeared as the most straightforward solution.
Sensing
Sensing performance
performance waswas alignedwith
aligned withstate
stateof of the
the art:
art: conductivity
conductivityresolution
resolutionofof 1313ppm
ppm (from
(from
50 2500
50 to to 2500 μS/cm)and
µS/cm) andpH pHresolution
resolution of of0.01
0.01(from
(from 0 to
0 14) could
to 14) be compared
could be compared with 500
withppm500and
ppm 0.05,
and
respectively, demonstrated in [2]. The resolution in the estimation of the
0.05, respectively, demonstrated in [2]. The resolution in the estimation of the deposit thickness wasdeposit thickness was about
10 μm
about 10 µm (from 10 to
(from 10300 μm).µm).
to 300 No cross-talk among
No cross-talk amongthe signals at the at
the signals level
theof the of
level electronic board was
the electronic board
observed. Instead, correlations among chemo-physical parameters
was observed. Instead, correlations among chemo-physical parameters were observed. For instance, were observed. For instance, in
addition to the temperature/conductivity dependence already highlighted in paragraph 3.1, in the
in addition to the temperature/conductivity dependence already highlighted in paragraph 3.1, in the
case of chemical contamination, both pH and conductivity levels (which are typically very stable in
case of chemical contamination, both pH and conductivity levels (which are typically very stable
water distribution networks) could suddenly change. In perspective, machine learning models,
in water distribution networks) could suddenly change. In perspective, machine learning models,
leveraging the correlations among parameters, could be trained to quickly and automatically identify
leveraging the correlations among parameters, could be trained to quickly and automatically identify
alterations and faults.
alterations
Theand faults.
results of an extensive characterization, both in the laboratory (up to 50 °C temperature, 4
The results of an extensive characterization, ◦ C temperature, 4 bar
bar pressure, and 1.5 m/s water velocity) and in boththe fieldin the
withlaboratory (up to of
a pilot network 50three nodes installed
pressure, andwater
in a real 1.5 m/s water velocity)
distribution network andand
in the field with operating
continuously a pilot network of three demonstrated
for 2 months, nodes installed thein a
realrobustness of the proposed system. Thanks to these results, in 2019, this solution was selected as a
water distribution network and continuously operating for 2 months, demonstrated the robustness
finalist
of the in the system.
proposed Zero Power Water
Thanks toMonitoring Horizon
these results, in 2019, Prize
thisbysolution
the European Commission.
was selected as a finalist in the
Zero Power Water Monitoring Horizon Prize by the European Commission.
Author Contributions: Conceptualization, M.C., A.T., J.F., and M.A.; electronics development, L.M., M.C., and
F.Z.; validation, L.M., M.D.M., M.T., and G.P.; writing, M.C. and A.T.; supervision and funding acquisition,
M.A., M.C., and A.T. All authors have read and agreed to the published version of the manuscript.
Sensors 2020, 20, 1125 14 of 16

Author Contributions: Conceptualization, M.C., A.T., J.F., and M.A.; electronics development, L.M., M.C., and
F.Z.; validation, L.M., M.D.M., M.T., and G.P.; writing, M.C. and A.T.; supervision and funding acquisition, M.A.,
M.C., and A.T. All authors have read and agreed to the published version of the manuscript.
Funding: This work has been funded by Politecnico di Milano by means of the Swith2Product 2017 award,
and partially by Italian MIUR under “giovani ricercatori” funds and by the Camera di Commercio special prize at
StartCup Lombardia 2018 competition.
Acknowledgments: The authors would like to warmly thank Romagna Acque-Società delle Fonti (Andrea Gambi,
Davide Lolli, and Ivo Vasumini, in particular) for the support in the field validation, Vittoria Roiati and the
Technology Transfer Office of Politecnico di Milano for support of the whole “Dirty Sensing” initiative. Marco
Mariani (Giakova srl) is thanked for providing the DMM7510. Giorgio Ferrari and Marco Sampietro (Politecnico
di Milano) are also thanked for logistical support in the electronics development.
Conflicts of Interest: The authors declare no conflict of interest. The funders had no role in the design of the
study; in the collection, analyses, or interpretation of data; in the writing of the manuscript, or in the decision to
publish the results.

References
1. Zulkifli, S.N.; Rahim, H.A.; Lau, W.-J. Detection of contaminants in water supply: A review on state-of-the-art
monitoring technologies and their applications. Sens. Actuators B Chem. 2018, 255, 2657–2689. [CrossRef]
2. Lambrou, T.P.; Anastasiou, C.C.; Panayiotou, C.G.; Polycarpou, M.M. A Low-Cost Sensor Network for
Real-Time Monitoring and Contamination Detection in Drinking Water Distribution Systems. IEEE Sens. J.
2014, 14, 2765–2772. [CrossRef]
3. Almazyad, A.; Seddiq, Y.; Alotaibi, A.; Al-Nasheri, A.; BenSaleh, M.; Obeid, A.; Qasim, S. A Proposed
Scalable Design and Simulation of Wireless Sensor Network-Based Long-Distance Water Pipeline Leakage
Monitoring System. Sensors 2014, 14, 3557–3577. [CrossRef] [PubMed]
4. Xu, G.; Shen, W.; Wang, X. Applications of Wireless Sensor Networks in Marine Environment Monitoring:
A Survey. Sensors 2014, 14, 16932–16954. [CrossRef] [PubMed]
5. Albaladejo, C.; Soto, F.; Torres, R.; Sánchez, P.; López, J.A. A Low-Cost Sensor Buoy System for Monitoring
Shallow Marine Environments. Sensors 2012, 12, 9613–9634. [CrossRef] [PubMed]
6. Albaladejo, C.; Sánchez, P.; Iborra, A.; Soto, F.; López, J.A.; Torres, R. Wireless Sensor Networks for
Oceanographic Monitoring: A Systematic Review. Sensors 2010, 10, 6948–6968. [CrossRef] [PubMed]
7. Kotamäki, N.; Thessler, S.; Koskiaho, J.; Hannukkala, A.; Huitu, H.; Huttula, T.; Havento, J.; Järvenpää, M.
Wireless in-situ Sensor Network for Agriculture and Water Monitoring on a River Basin Scale in Southern
Finland: Evaluation from a Data User’s Perspective. Sensors 2009, 9, 2862–2883. [CrossRef]
8. Kadir, E.A.; Siswanto, A.; Rosa, S.L.; Syukur, A.; Irie, H.; Othman, M. Smart Sensor Node of WSNs for River
Water Pollution Monitoring System. In Proceedings of the 2019 International Conference on Advanced
Communication Technologies and Networking (CommNet), Rabat, Morocco, 12–14 April 2019; IEEE: Rabat,
Morocco, 2019; pp. 1–5.
9. Postolache, O.A.; Girao, P.M.B.S.; Pereira, J.M.D.; Ramos, H.M.G. Self-Organizing Maps Application in a
Remote Water Quality Monitoring System. IEEE Trans. Instrum. Meas. 2005, 54, 322–329. [CrossRef]
10. Jiang, P.; Xia, H.; He, Z.; Wang, Z. Design of a Water Environment Monitoring System Based on Wireless
Sensor Networks. Sensors 2009, 9, 6411–6434. [CrossRef]
11. Tuna, G.; Arkoc, O.; Gulez, K. Continuous Monitoring of Water Quality Using Portable and Low-Cost
Approaches. Int. J. Distrib. Sens. Netw. 2013, 9, 249598. [CrossRef]
12. Dong, J.; Wang, G.; Yan, H.; Xu, J.; Zhang, X. A survey of smart water quality monitoring system. Environ.
Sci. Pollut. Res. 2015, 22, 4893–4906. [CrossRef] [PubMed]
13. Geetha, S.; Gouthami, S. Internet of things enabled real time water quality monitoring system. Smart Water
2016, 2, 1. [CrossRef]
14. Japitana, M.V.; Palconit, E.V.; Demetillo, A.T.; Burce, M.E.C.; Taboada, E.B.; Abundo, M.L.S. Integrated
Technologies for Low Cost Environmental Monitoring in the Water Bodies of the Philippines: A Review.
Nat. Environ. Pollut. Technol. 2018, 17, 13.
15. Metje, N.; Chapman, D.N.; Cheneler, D.; Ward, M.; Thomas, A.M. Smart Pipes—Instrumented Water Pipes,
Can This Be Made a Reality? Sensors 2011, 11, 7455–7475. [CrossRef]
Sensors 2020, 20, 1125 15 of 16

16. Ali, H.; Choi, J. A Review of Underground Pipeline Leakage and Sinkhole Monitoring Methods Based on
Wireless Sensor Networking. Sustainability 2019, 11, 4007. [CrossRef]
17. Stoianov, I.; Nachman, L.; Madden, S.; Tokmouline, T. PIPENET: A Wireless Sensor Network for Pipeline
Monitoring. In Proceedings of the 2007 6th International Symposium on Information Processing in Sensor
Networks, Cambridge, MA, USA, 25–27 April 2007; IEEE: Cambridge, MA, USA, 2007; pp. 264–273.
18. Prest, E.I.; Hammes, F.; van Loosdrecht, M.C.M.; Vrouwenvelder, J.S. Biological Stability of Drinking Water:
Controlling Factors, Methods, and Challenges. Front. Microbiol. 2016, 7, 45. [CrossRef]
19. Banna, M.H.; Imran, S.; Francisque, A.; Najjaran, H.; Sadiq, R.; Rodriguez, M.; Hoorfar, M. Online Drinking
Water Quality Monitoring: Review on Available and Emerging Technologies. Crit. Rev. Environ. Sci. Technol.
2014, 44, 1370–1421. [CrossRef]
20. Wingender, J.; Flemming, H.-C. Biofilms in drinking water and their role as reservoir for pathogens. Int. J.
Hyg. Environ. Health 2011, 214, 417–423. [CrossRef]
21. Richards, C.S.; Wang, F.; Becker, W.C.; Edwards, M.A. A 21st-Century Perspective on Calcium Carbonate
Formation in Potable Water Systems. Environ. Eng. Sci. 2018, 35, 143–158. [CrossRef]
22. Chung, W.S.; Yu, M.J.; Lee, H.D. Prediction of corrosion rates of water distribution pipelines according to
aggressive corrosive water in Korea. Water Sci. Technol. 2004, 49, 19–26. [CrossRef]
23. Delauney, L.; Compère, C.; Lehaitre, M. Biofouling protection for marine environmental sensors. Ocean Sci.
2010, 6, 503–511. [CrossRef]
24. Aisopou, A.; Stoianov, I.; Graham, N.J.D. In-pipe water quality monitoring in water supply systems under
steady and unsteady state flow conditions: A quantitative assessment. Water Res. 2012, 46, 235–246.
[CrossRef] [PubMed]
25. Kim, K.; Myung, H. Sensor Node for Remote Monitoring of Waterborne Disease-Causing Bacteria. Sensors
2015, 15, 10569–10579. [CrossRef] [PubMed]
26. Available online: https://www.istat.it/it/files/2019/03/Testo-integrale_Report_Acqua_2019.pdf (accessed on
19 January 2020).
27. Olatinwo, S.O.; Joubert, T.-H. Energy Efficient Solutions in Wireless Sensor Systems for Water Quality
Monitoring: A Review. IEEE Sens. J. 2019, 19, 1596–1625. [CrossRef]
28. Becker, P.; Folkmer, B.; Goepfert, R.; Hoffmann, D.; Willmann, A.; Manoli, Y. Energy Autonomous Wireless
Water Meter with Integrated Turbine Driven Energy Harvester. J. Phys. Conf. Ser. 2013, 476, 012046.
[CrossRef]
29. Li, X.J.; Chong, P.H.J. Design and Implementation of a Self-Powered Smart Water Meter. Sensors 2019, 19,
4177. [CrossRef]
30. Malavasi, S.; Rossi, M.M.A.; Ferrarese, G. GreenValve: Hydrodynamics and applications of the control valve
for energy harvesting. Urban Water J. 2018, 15, 200–209. [CrossRef]
31. Boccalero, G.; Boragno, C.; Caviglia, D.; Morasso, R. FLEHAP: A Wind Powered Supply for Autonomous
Sensor Nodes. JSAN 2016, 5, 15. [CrossRef]
32. Carminati, M.; Ferrari, G.; Sampietro, M. Emerging miniaturized technologies for airborne particulate matter
pervasive monitoring. Measurement 2017, 101, 250–256. [CrossRef]
33. Montagnani, G.L.; Carminati, M.; Lavelli, E.; Morandi, G.; Rizzacasa, P.; Fiorini, C. SiPM-Based Scrap
Metal Radioactivity Detector Embeddable in Lifting Electromagnets. In Proceedings of the 2018 IEEE
Nuclear Science Symposium and Medical Imaging Conference Proceedings (NSS/MIC), Sydney, Australia,
10–17 November 2018; IEEE: Sydney, Australia, 2018; pp. 1–3.
34. Keller, O.; Benoit, M.; Müller, A.; Schmeling, S. Smartphone and Tablet-Based Sensing of Environmental
Radioactivity: Mobile Low-Cost Measurements for Monitoring, Citizen Science, and Educational Purposes.
Sensors 2019, 19, 4264. [CrossRef]
35. Colli, M.; Stagnaro, M.; Caridi, A.; Lanza, L.G.; Randazzo, A.; Pastorino, M.; Caviglia, D.D.; Delucchi, A. A
Field Assessment of a Rain Estimation System Based on Satellite-to-Earth Microwave Links. IEEE Trans.
Geosci. Remote Sens. 2019, 57, 2864–2875. [CrossRef]
36. Carminati, M.; Ferrari, G.; Vahey, M.D.; Voldman, J.; Sampietro, M. Miniaturized Impedance Flow Cytometer:
Design Rules and Integrated Readout. IEEE Trans. Biomed. Circuits Syst. 2017, 11, 1438–1449. [CrossRef]
[PubMed]
37. Igreja, R.; Dias, C.J. Analytical evaluation of the interdigital electrodes capacitance for a multi-layered
structure. Sens. Actuators A Phys. 2004, 112, 291–301. [CrossRef]
Sensors 2020, 20, 1125 16 of 16

38. Mezzera, L.; Carminati, M.; Di Mauro, M.; Turolla, A.; Tizzoni, M.; Antonelli, M. A 7-Parameter Platform
for Smart and Wireless Networks Monitoring On-Line Water Quality. In Proceedings of the 2018 25th IEEE
International Conference on Electronics, Circuits and Systems (ICECS), Bordeaux, France, 9–12 December
2018; IEEE: Bordeaux, France, 2018; pp. 709–712.
39. Vergani, M.; Carminati, M.; Ferrari, G.; Landini, E.; Caviglia, C.; Heiskanen, A.; Comminges, C.; Zor, K.;
Sabourin, D.; Dufva, M.; et al. Multichannel Bipotentiostat Integrated With a Microfluidic Platform for
Electrochemical Real-Time Monitoring of Cell Cultures. IEEE Trans. Biomed. Circuits Syst. 2012, 6, 498–507.
[CrossRef]
40. Turolla, A.; Di Mauro, M.; Mezzera, L.; Antonelli, M.; Carminati, M. Development of a Miniaturized and
Selective Impedance Sensor for Real-Time Slime Monitoring in Pipes and Tanks. Sens. Actuators B Chem.
2019, 281, 288–295. [CrossRef]
41. Carminati, M.; Mezzera, L.; Turolla, A.; Pani, G.; Tizzoni, M.; Di Mauro, M.; Antonelli, M. Flexible Impedance
Sensor for In-Line Monitoring of Water and Beverages. In Proceedings of the 2019 IEEE International
Symposium on Circuits and Systems (ISCAS), Sapporo, Japan, 26–29 May 2019; IEEE: Sapporo, Japan, 2019;
pp. 1–4.
42. Carminati, M.; Luzzatto-Fegiz, P. Conduino: Affordable and high-resolution multichannel water conductivity
sensor using micro USB connectors. Sens. Actuators B Chem. 2017, 251, 1034–1041. [CrossRef]
43. Carminati, M.; Ferrari, G.; Grassetti, R.; Sampietro, M. Real-Time Data Fusion and MEMS Sensors Fault
Detection in an Aircraft Emergency Attitude Unit Based on Kalman Filtering. IEEE Sens. J. 2012, 12,
2984–2992. [CrossRef]
44. Carminati, M.; Mezzera, L.; Ferrari, G.; Sampietro, M.; Turolla, A.; Di Mauro, M.; Antonelli, M. A Smart
Sensing Node for Pervasive Water Quality Monitoring with Anti-Fouling Self-Diagnostics. In Proceedings of
the 2018 IEEE International Symposium on Circuits and Systems (ISCAS), Florence, Italy, 27–30 May 2018;
IEEE: Florence, Italy, 2018; pp. 1–5.

© 2020 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/).