IEEE SENSORS JOURNAL, VOL. 14, NO. 8, AUGUST 2014 2765

A Low-Cost Sensor Network for Real-TimeMonitoring and Contamination Detectionin Drinking Water Distribution Systems

Theofanis P. Lambrou, Christos C. Anastasiou, Christos G. Panayiotou, and Marios M. Polycarpou

Abstract— This paper presents a low cost and holistic approachto the water quality monitoring problem for drinking waterdistribution systems as well as for consumer sites. Our approachis based on the development of low cost sensor nodes for real timeand in-pipe monitoring and assessment of water quality on the fly.The main sensor node consists of several in-pipe electrochemicaland optical sensors and emphasis is given on low cost, light-weight implementation, and reliable long time operation. Suchimplementation is suitable for large scale deployments enablinga sensor network approach for providing spatiotemporally richdata to water consumers, water companies, and authorities.Extensive literature and market research are performed to iden-tify low cost sensors that can reliably monitor several parameters,which can be used to infer the water quality. Based on selectedparameters, a sensor array is developed along with severalmicrosystems for analog signal conditioning, processing, logging,and remote presentation of data. Finally, algorithms for fusingonline multisensor measurements at local level are developed toassess the water contamination risk. Experiments are performedto evaluate and validate these algorithms on intentional con-tamination events of various concentrations of escherichia colibacteria and heavy metals (arsenic). Experimental results indicatethat this inexpensive system is capable of detecting these highimpact contaminants at fairly low concentrations. The resultsdemonstrate that this system satisfies the online, in-pipe, lowdeployment-operation cost, and good detection accuracy criteriaof an ideal early warning system.

Index Terms— Water quality monitoring, flat surface sensors,turbidity sensor, multi-sensor system, sensor networks, arsenic& bacteria contamination detection.

I. INTRODUCTION

CLEAN drinking water is a critical resource, important forthe health and well-being of all humans. Drinking water

utilities are facing new challenges in their real-time operationbecause of limited water resources, intensive budget require-ments, growing population, ageing infrastructure, increasingly

Manuscript received December 19, 2013; revised March 7, 2014; acceptedMarch 20, 2014. Date of publication April 10, 2014; date of current versionJuly 1, 2014. This work was supported in part by the European ResearchCouncil through the ERC Advanced Grant Fault-Adaptive and in part by theEuropean Project Effinet under Grant FP7-ICT-2011-8-31855. The associateeditor coordinating the review of this paper and approving it for publicationwas Dr. M. R. Yuce.

T. P. Lambrou, C. G. Panayiotou, and M. M. Polycarpou are with the KIOSResearch Center for Intelligent Systems and Networks, Department of Elec-trical and Computer Engineering, University of Cyprus, Nicosia 2102, Cyprus(e-mail: faniseng@ucy.ac.cy; christosp@ucy.ac.cy; mpolycar@ucy.ac.cy).

C. C. Anastasiou is with the Department of Civil Engineering, FrederickUniversity, Nicosia 1036, Cyprus (e-mail: c.anastasiou@frederick.ac.cy).

Color versions of one or more of the figures in this paper are availableonline at http://ieeexplore.ieee.org.

Digital Object Identifier 10.1109/JSEN.2014.2316414

stringent regulations and increased attention towards safe-guarding water supplies from accidental or deliberate conta-mination. There is a need for better on-line water monitoringsystems given that existing laboratory-based methods are tooslow to develop operational response and do not provide alevel of public health protection in real time. Rapid detection(and response) to instances of contamination is critical due tothe potentially severe consequences to human health.

Traditional methods of water quality control involve themanual collection of water samples at various locations and atdifferent times, followed by laboratory analytical techniquesin order to characterize the water quality. Such approachesare no longer considered efficient [1]–[5]. Although, thecurrent methodology allows a thorough analysis includingchemical and biological agents, it has several drawbacks:a) the lack of real-time water quality information to enablecritical decisions for public health protection (long time gapsbetween sampling and detection of contamination) b) poorspatiotemporal coverage (small number locations are sampled)c) it is labor intensive and has relatively high costs (labor,operation and equipment). Therefore, there is a clear needfor continuous on-line water quality monitoring with effi-cient spatio-temporal resolution. US Environmental ProtectionAgency (USEPA) has carried out an extensive experimentalevaluation [6] of water quality sensors to assess their perfor-mance on several contaminations. The main conclusion wasthat many of the chemical and biological contaminants usedhave an effect on many water parameters monitored includingTurbidity (TU), Oxidation Reduction Potential (ORP),Electrical Conductivity (EC) and pH. Thus, it is feasible tomonitor and infer the water quality by detecting changes insuch parameters.

Given the absence of reliable, in-line, continuous and inex-pensive sensors for monitoring all possible biological andchemical contaminants, our approach is to measure physico-chemical water parameters that can be reliably monitored withlow cost sensors and develop low cost networked embeddedsystems (sensor nodes) as well as contamination detectionalgorithms to fuse these multi-sensor data in order to inferpossible contamination events. Even though this approachmay suffer from some false alarms, it can be compen-sated/eliminated by the large scale deployment and the possi-bility of correlating the decisions from various sensor nodeswhich is the topic of our future work.

There is a clear need for a shift in the current monitoringparadigm and this paper proposes the idea of monitoring

1530-437X © 2014 IEEE. Personal use is permitted, but republication/redistribution requires IEEE permission.See http://www.ieee.org/publications_standards/publications/rights/index.html for more information.

2766 IEEE SENSORS JOURNAL, VOL. 14, NO. 8, AUGUST 2014

the quality of water delivered to consumers, using low cost,low power and tiny in-pipe sensors. The main contributionof this paper is the design and development of a low costsystem that can be used at the premises of consumers tocontinuously monitor qualitative water parameters and fusemulti-parametric sensor response in order to assess the waterconsumption risk. In particular, the contributions regardingthe low cost system is the design and development of lowcost networked embedded systems as well as optical sensors(turbidity) for water quality monitoring, the development ofevent detection algorithms using fusion techniques and theexperimental evaluation and validation of system performancein various concentrations of microbiologically (E.coli) andchemically (Arsenic) contaminated drinking water.

The remaining of this paper is organized as follows.Section II reviews related work. Section III presents themethodology and justification for the selection of water qualityparameters to be monitored. Section IV presents the systemdesign and the experimental implementation of the hardwareand software modules. Section V validates the performanceof the developed system and finally the paper concludeswith Section VI.

II. RELATED WORK

A preliminary version of this article has appeared in [2].In this article, we present an improved hardware platform,develop a new advanced contamination event detection algo-rithm and provide an experimental evaluation and validationof system and event detection algorithms in the presence ofreal microbiological and chemical contamination events.

A limited number of on-line, reagent-free water monitor-ing systems are commercially available [7] (e.g. Hach HSTGuardianBlue [8], J-MAR BioSentry [9], etc), but these sys-tems are bulky (sensors are installed in flow cells located incabinets) and remain cost prohibitive for large scale deploy-ments (cost tens of thousands of dollars per unit). It isworth mentioning that cost is mostly attributed not to sensingprobes but to instrumentation-automation controllers (analyz-ers) and panels. Such systems can take frequent samplesof the water quality at a very limited number of locations.However, substantial proportion of contamination problems isattributable to problems within distribution systems and dueto the limited spatio-temporal sampling, it is impossible forthe water companies and consumers to know the quality ofpotable water delivered to consumer households.

A number of bare multi-parametric sensor arrays have beendeveloped and presented in the literature based on varioussensor technologies. A recent review on multi-parametricsolid-state sensors for water quality is given [3]. A chemicalsensor array for water quality monitoring based on thick-film technology is presented in [24], [25], [26], and [27],these sensors are very low cost, though they have limited life-time (few months) and require a conventional glass referenceelectrode to operate accurately. Along similar lines, a multi-parametric sensor array based on semiconductor rutheniumoxide nanostructures is presented in [3] and [28].

In addition, several water monitoring microsystems (sensornodes) have been developed for large scale water monitoring

based on wireless sensor networks (WSNs) technology. In [29]a sensor node is developed for monitoring salinity in groundwaters as well as the water temperature in surface waters.In [33] and [38], the authors have developed a WSN and anenergy harvesting system (based on a solar panel) to monitornitrate, ammonium and chloride levels in rivers and lakes.Energy harvesting techniques along with hibernation methodsplay an important role in extending the lifetime of sensornodes. A survey on energy harvesting for WSNs is providedin [39] and [40]. Finally, in [34] an autonomous boat equippedwith water sensors is proposed to collect samples from lakesusing the A∗ search algorithm. More efficient navigationalgorithms for a group of boats with obstacle avoidance arepresented in [35]–[37].

Next, we provide a number of academic and commercialefforts aim to develop hardware and software platforms forreal-time monitoring of the water distribution systems. In [41]a WSN is proposed to monitor hydraulic parameters in order todetect events such as leaks, pipe bursts. A cost effective multi-sensor probe (Endetec KAPTA 3000-AC4) for monitoringchlorine, conductivity and pressure without any event detectionalgorithms has been proposed by Endetec [42] in 2012. Finally,in [43] an optical interferometric sensor along with an eventdetection algorithm to monitor refractive index aberrations inwater has been developed.

Apart from the on going research towards the design anddevelopment of sensors and microsystems another parallelresearch direction is that of the development of software andalgorithms for the detection of water quality anomalies andcontamination events. A thorough survey on recent advancesin this area is provided in [30]. A limited number of eventdetection software is commercially available (Hach EventMonitor [8], BlueBox [10]). A currently freely available toolis CANARY software [11] developed at Sandia NationalLaboratories in collaboration with the USEPA. CANARYindicates possible contamination events by using a range ofmathematical and statistical techniques to identify the onsetof anomalous water quality incidents from online raw sensordata. Other event detection and data validation methodologiesare given in [31] and references therein.

III. METHODS

Drinking water quality standards are determined accordingto World Health Organization (WHO) [12] guidelines fordrinking-water quality as well as other pertinent organizations(i.e. EU [13], USEPA [14]). These organizations set the stan-dards for drinking water quality parameters and indicate whichmicrobiological, chemical and indicator parameters must bemonitored and tested regularly in order to protect the healthof the consumers and to make sure the water is wholesomeand clean.

For the developed system, the selection of the physico-chemical parameters to be monitored was based on exten-sive scientific literature review [6], [16], [17], and [18] onthe relation between certain physicochemical parameters andchemical or biological contaminations that present in water.Table I enumerates the suggested parameters to be monitored

LAMBROU et al.: LOW-COST SENSOR NETWORK FOR REAL-TIME MONITORING AND CONTAMINATION DETECTION 2767

TABLE I

SUGGESTED PARAMETERS TO BE MONITORED

from high to low correlation significance when interpretingwater contaminations (assess hazard). Table I also presentsthe measurement cost (for purchase and maintenance) asso-ciated with these parameters based on recent review [19] ofmeasurement and instrumentation methods, compensation andcalibration procedures and probe lifetime concerning theseparameters. Therefore, the parameters selected to be moni-tored are the following: 1) Turbidity, 2) Oxidation ReductionPotential (instead of Free Chlorine), 3) Temperature 4) pH,and 5) Electrical Conductivity.

It is noted that Free Chlorine concentration (HOCl) canbe approximated based on the ORP, pH and temperaturemeasurements. Free chlorine monitoring is expensive becauseit is very sensitive in the pH, temperature, flow and pressureof the sample. Therefore accurate free chlorine measure-ments require a flow cell with additional pH and temperaturesensors for compensation. Nitrates, though considered as animportant parameter for human health is not selected becausemeasurement methods are subjected to failures (Ion-Selectiveelectrodes) or are cost prohibitive (UV spectrophotometricmethod). In [15], a new promising method is presented basedon a PCB planar electromagnetic sensor. Finally, dissolvedoxygen is not selected due to several compensations andfrequent membrane replacements needed.

Convectional combined electrodes (for ORP and pH) havebeen widely used due to their good sensitivity, selectivity,stability and long lifetime. However, convectional pH glasselectrodes have several disadvantages due to the intrinsicnature of the glass membrane. For example, they have limitedpressure tolerance, exhibit a sluggish response, require a highinput impedance signal conditioning circuits and it is difficultto miniaturize based on current manufacturing technologies.Therefore, a number of emerging-alternative sensor technolo-gies in various stages of research and development have beenproposed in the literature.

Thick film chemical sensor arrays developments show that itis possible to develop a single miniaturized multi-parametricsensor probe in a cost effective manner, however thick filmchemical sensors have limited lifetime (few months), sufferfrom electrode drift (due to salt loss) and the developmentof a stable reference electrode is not possible so far [25].ISFET based microsensors (developed using MOSFET semi-conductor technology) offer advantages such small size (massfabrication and compact probes), robustness (no glass mem-branes), low output impedance and rapid response, however

they have several limitations as they require a glass referenceelectrode (REFET) to operate robustly and encapsulation isdifficult, which increases dramatically the final cost of thesensors [32]. Nano-sensors based on nanostructures of noblemetals and their oxides (like Pt, Ru, Ir) is a recent promisingconcept however developments so far suffer from severaldrawbacks like temperature dependent delay response and non-deterministic potential drift (electrolysis of water on oxidesurfaces and unpredictable temperature dependence) [28].Therefore, despite the recent advances in sensor developmenttechnologies, the reliability and performance of convectionalglass electrodes is still unsurpassed for continuous waterquality monitoring [7]. Therefore, convectional (pH,ORP)glass electrodes and solid-state sensors (TU, EC, T) are usedin this work as they provide the most reliable technology.

In-line water sensors illustrate the need for efficient andperiodic probe cleaning to maintain reliable measurements.Cleaning mechanisms constitute an important cost parameterwhich can consume as high as 50% of the operational budgets.Conventionally, ultrasonic, brush, water-jet, or chemical typeof automatic cleaners [20] are used to remove coatings fromthe sensor probes. Recently, several alternative cost effectivemethods have been proposed that can either actively removefouling (e.g. electrolysis) or passively prevent fouling (coppermesh or CuO2 doped materials). In this work, flat measuringsurface probe method [23] is used because is the most costeffective, passive self-cleaning method and is based on themechanical package and design of the probe. When theelectrode’s flat measuring surface is exposed to turbulent flow,the resulting scrubbing action provides a self-cleaning effect inmost applications under medium range flows. The flat sensingsurface virtually eliminates deposits that can foul the electrodeand significantly reduces necessary maintenance. This simple,but effective method has no moving parts, requires no powerand also prolongs electrode life and eliminates breakage. Addi-tional antifouling technologies have be proposed for solid-stateand optical sensors based on nano-scale materials possessingsuper-hydrophobic properties [21], [22].

IV. PLATFORM DESIGN

A. System and Sensors Development and Integration

The overall system architecture under discussion in pre-sented in Fig. 1 and is comprised of the following threesubsystems: a central measurement node (PIC32 MCU basedboard) that collects water quality measurements from sensors,implements the algorithm to assess water quality and transmitsdata to other nodes, a control node (ARM/Linux based plat-form) that stores measurement data received from the centralmeasurement node in a local database and provides gateway tothe internet, visualize data (charts), and sends email/sms alertsand finally a tiny notification node(s) (PIC MCU based board)that receives information from the central measurement nodethrough an interconnected ZigBee RF transceiver and provideslocal near-tap notifications to the user (water consumer) viaseveral interfaced peripherals (LED, LCD, Buzzer).

It should be noted that the central measurement node servesas the sensor node. The idea is to install these sensor nodes

2768 IEEE SENSORS JOURNAL, VOL. 14, NO. 8, AUGUST 2014

Fig. 1. System architecture.

Fig. 2. Complete system photo.

in many consumer sites in a spatially-distributed manner toform a WSN that will monitor the drinking water qualityin the water distribution system from the source to the tap.The central measurement node is interfaced to multi-parametersensor array comprised of Turbidity(TU), ORP, pH, ElectricalConductivity(EC) and Temperature(T) sensors. The in-pipeTurbidity sensor is constructed from scratch based on ourprevious work [1] while the other sensor probes obtainedfrom SensoreX Corp�. The pH sensor embeds an RTDsensor which is used for temperature sensing and temperaturecompensation of pH and EC measurements. TU, ORP, pHand toroidal EC sensors have flat measuring surfaces for costeffective self-cleaning. The complete system photo, with TU,ORP, pH, EC and T sensors as well as a rotor-flow sensormounted in a plastic pipe, is shown in Fig 2.

Turbidity Sensor Development: Although there is plenty ofturbidity measuring instruments available on the market atthe moment, most of them are expensive and not directlycompatible with in-pipe, in-line requirements as well as WSNstechnology. Therefore, the goal is to develop a low cost, easyto use and accurate enough turbidity sensor for continuous inpipe turbidity monitoring in water distribution systems using

Fig. 3. Turbidity sensor. (a) Measurement principle. (b) Probe board.(c) Flat surface PTFE housing. (d) Inline Tee fitting.

commercial off-the self-components. The turbidity sensordevelopment was based on the ratio turbidimeter design (seeFig. 3) where both transmitted and scattered light intensitiesare measured to eliminate errors (interferences) due to IRemitter intensity drift and sample absorption characteristics.An infrared (860nm) narrow beam LED emits light throughan optical gap to the water sample and two IR photodiodesseparated around 1cm from the emitter receive simultaneouslythe 90o scattered and 0o transmitted light. The photodiodesspectral sensitivity are selected to fit with that of the IR lightsource. The instrumentation and analog signal conditioning ofthe sensor is as follows: The IR emitter is pulsed at 1kHzwith a square wave signal and the photodiodes convert thelight directly into electrical current, then a high-gain, low-noiseCMOS (Complementary metal-oxide-semiconductor) transim-pedance amplifier with background light rejection is used toconvert the each photocurrent to voltage output. The ac outputof each transimpedance amplifier is then converted to a dcsignal using a precision active peak detector. Finally the 90o

scattered dc signal is further conditioned by an instrumentationamplifier for 0 NTU offset nulling and additional amplifica-tion. The conditioned voltage outputs are then sampled bya 10 bit A/D converter with reference voltage of 1.1V andthe sensor output voltage V = V90o

c.V0ois given as the signal

ratio of the scattered V90o to the transmitted V0o voltage, c iscalibration coefficient.

An indirect method for the sensor calibration was employed,in order to avoid the use of the carcinogen and expensivechemical formazin solutions. Therefore, a number of sampleswere created and the turbidity of each sample is measured bothby the turbidity sensor under calibration and by a laboratoryturbidimeter (Lutron TU-2016) used as reference. Then therelationship between turbidity (in NTU) and the voltage output(in mV) of the turbidity sensor is extracted and given byT U = 0.1035V − 0.292. The sensor generates an outputvoltage proportional to the turbidity or suspended particles andhas a linear response in the range of 0-100 NTU with 0.1 NTUresolution. Finally, as shown in Fig. 3 the turbidity sensorprobe was mounted in a flat surface PTFE (teflon) housingand sealed in a hydraulic Tee fitting for inline installation.

Apart from TU sensor, analog signal conditioning circuits,calibration and compensation procedures were developed forpH, OPR, RTD and conductive/inductive EC sensors. Con-siderable attention is given to acquire linear response, reducenoise and attain high resolution and accuracy. A dedicatedPIC based microsystem is developed for each parameter to

LAMBROU et al.: LOW-COST SENSOR NETWORK FOR REAL-TIME MONITORING AND CONTAMINATION DETECTION 2769

Fig. 4. The first stage of analog signal conditioning circuitry. (a) Turbiditypreamplifier. (b) Conductivity preamplifier. (c) ORP preamplifier. (d) pHpreamplifier. (e) Temp.

Fig. 5. Software platform.

accomplish this task. The first stage of analog signal condi-tioning circuitry for each parameter is presented in Fig. 4 whileTable II shows the results regarding laboratory evaluation(using standard buffer solutions and reference instruments)of each parameter along with the quality range suggested byWHO guidelines and EU standards. The overall power con-sumption of the central measurement sensor node with the onboard LEDs off and the RF Xbee transceiver module sendingwater quality data every 5s is about 50mA at 5V operatingvoltage, however further improvements are planned to min-imize the power consumption using hibernation schemes.It worth mentioning that wireless communication is by far thelargest consumer of the energy of the sensor node, comparedto other functions such as sensing and computation.

The components for the complete system prototype costapproximatelye400 (e300 for sensors-mounts-enclosures ande100 for microsystems and electronic components) which is atleast an order of magnitude less expensive than commerciallyavailable multi-parameter instruments. It worth mentioningthat site preparation costs will be also minimized if suchlow cost and lightweight systems are deployed in consumerpremises instead of buried main supply pipes.

The software platform developed for the control node isillustrated in Fig. 5. This platform enables real time measure-ment charts of monitored parameters, real time assessmentof water quality and sensor calibration instructions through

Fig. 6. On-line web platform.

a Graphical User Interface (GUI). It also logs sensor data ina local database and posts data to web using Pachube opensource web platform. Using Pachube scripts the user can setupvarious thresholds for sending notifications via sms or email.Fig. 6 illustrates the main window of the internet platform.

B. Contamination Event Detection Algorithms

Two event detection algorithms were developed to fuseon-line multi-sensor measurements in order to assess the watercontamination risk when anomalies are detected. An eventdetection algorithm enables the system to act as an “earlywarning system" for possible potable water quality deteriora-tion at the point of installation (e.g. homes). Both algorithmsare based on normalized sensor outputs given by

Ni = |Si − μi |τiσi

(1)

where Si is the current measurement of parameter i ∈{T U, O RP, pH, EC}, μi , σi are the mean and standarddeviation over a moving time window w and τi is a sensorbased parameter associated with measurement accuracy ofeach parameter i . Normalized sensor outputs Ni are used tofilter baseline (i.e mean) fluctuations.

The objective of the event detection algorithms is to activatean alarm when normalized sensor outputs exhibit sudden andsignificant changes, given that these changes are boundedwithin the quality ranges suggested by drinking water qualitystandards (see Table II, quality range). The detection of waterquality changes that are outside the expected quality ranges(min/max violations) is easier and can be done by a weightedmulti-parameter cost function in the form of RO = ∑

i wOi Ji ,where Ji are binary variables that indicate whether parameteri has been violated and wOi are non-negative weights whichimply the significance of the violation of each parameter i .If RO = 0 no violation is assumed, however as RO > 0increases the water contamination risk is also increases.

2770 IEEE SENSORS JOURNAL, VOL. 14, NO. 8, AUGUST 2014

TABLE II

SPECIFICATIONS AND ACCOMPLISHED PERFORMANCE FOR EACH MONITORED PARAMETER

As previously indicated, the objective in this paper isto detect anomalies when water quality changes are insidethe expected quality ranges by fusing the multi-sensor data.Therefore a risk indicator RI function is defined that takes avalue RI = 1 if a contamination event is detected or RI = 0otherwise.

The first event detection algorithm is denoted as VectorDistance Algorithm (VDA) and the risk indicator RV D A

I func-tion used in this algorithm is estimated based on the Euclidiandistance between the normalized sensor signal vector N andthe normalized control signal vector N0 of pure (clean) water.Therefore, the risk indicator RV D A

I is given by

RV D AI =

{1 if ||N − N0||2 > d0 otherwise

(2)

Note that VDA algorithm requires the normalized controlsignal vector N0 as well as a calibration threshold d (obtainedfrom a learning phase) to execute.

The second event detection algorithm is denoted as PolygonArea Algorithm (PAA) and the risk indicator R P AA

I functionused in this algorithm is estimated based on the ratio of thepolygon area AN formed by the N vector components (whenprojected (displayed) on a two-dimensional spider graph withfour (TU, ORP, pH, EC) axes starting from the same point) tothe polygon area A1 formed by the 1 ones vector components(i.e 1 = [1111]T ). Therefore, the risk indicator R P AA

I is givenby

R P AAI =

⎧⎨

⎩

1 ifAN

A1> 1

0 otherwise(3)

Note that PAA algorithm does not require any further infor-mation to execute.

V. EXPERIMENTAL VALIDATION

In this section we present the results of the experimen-tal trials performed to validate the behavior and evaluatethe performance of the developed hardware and algorithmson intentional contamination events. The experimental setupconsists of the sensor node (central measurement node) thattakes samples every 5s from potable water flowing througha flow cell. Intentional contamination of two important con-taminants (escherichia coli bacteria and arsenic) of variousconcentrations was injected at discrete time intervals and theperformance of the event detection algorithms is evaluated onreal time. Escherichia coli bacteria and arsenic contaminationin drinking water is very severe problem causing seriouspoisoning to large numbers of people all over the world [18].

Fig. 7. Experiments with E.coli bacteria contaminated water. (a) Sensorsresponses to E.coli bacteria. (b) E.Coli bacteria contamination detection.

1) Microbiologically (E.coli) Contaminated DrinkingWater: The first experiment considers the case ofmicrobiologically (E.coli) contaminated drinking water.Most E. coli strains are in general harmless to humans, butsome types can cause serious food and water poisoning.However, the presence of E.coli is used to indicate thatother pathogenic organisms may be present (often of faecalorigin). According to WHO guidelines & EU Drinking WaterDirective E.coli parametric value is 0 CFU/100mL.

Fig. 7(a) presents the measurements received using thedeveloped sensor node when the following concentrations ofE.coli were injected: 5x10−2, 5x10−1, 5x100, 5x101, 5x103,5x104, 1x107 CFU/mL. It is evident that TU and EC sensorsresponded well when microbiological contaminants injectedin chlorinated potable water. ORP sensors has respondedwith delay and pH sensor has a spiky type of response.

LAMBROU et al.: LOW-COST SENSOR NETWORK FOR REAL-TIME MONITORING AND CONTAMINATION DETECTION 2771

Fig. 8. Experiments with Heavy metals (As) contaminated water.(a) Sensors responses to Heavy metal (As). (b) Heavy metals (As) conta-mination detection.

Fig. 7(b) presents the output signals of the Vector DistanceAlgorithm (VDA) and Polygon Area Algorithm (PAA). Theresults of Fig. 7(b) indicate that both algorithms miss thedetection of 5x10−2CFU/mL because sensors responses werevery close to background levels (no anomalies occurred). Itshould be noted that the performance of PAA algorithm isbetter and given that it utilizes less information, PAA algorithmis better than the VDA algorithm.

2) Chemically (Arsenic) Contaminated Drinking Water:The second experiment considers the case of chemically(Arsenic) contaminated drinking water. Water contaminationby toxic heavy metals and especially arsenic contaminationis a common problem encountered in many countries dueto undue deposition of mining, agricultural, industrial andurban wastes in water resources. Arsenic is known to affectnegatively the mental and central nervous system function, todamage the blood composition, lungs, kidneys, liver, and othervital organs, as well as it contributes to certain neurologicaldegenerative processes and causes skin cancer. According toWHO guidelines & EU Drinking Water Directive Arsenicparametric value is 10μg/L.

Fig. 8(a) presents the measurements received using thedeveloped sensor node when the following concentrations ofArsenic were injected: 5, 10, 25, 50, 125, 500, 1000 μg/L.Arsenic solutions created from a standard solution of1000mg/L As. Unfortunately, almost all sensors did notrespond at low arsenic contamination. However, at concen-

trations above 25 μg/L ORP and pH sensors have respondedand at higher concentrations (above 500 μg/L) all sensorsresponded well. Fig. 8(b) presents the output signals ofthe Vector Distance Algorithm (VDA) and Polygon AreaAlgorithm (PAA). The results of Fig. 8(b) indicate that bothalgorithms miss the detection of 5 and 10 μg/L becausesensors responses were very sluggish and close to backgroundlevels and that the VDA algorithm exhibits two false alarms.Therefore, the performance of PAA algorithm is again better(sharp response, no false alarms) than the VDA algorithm.

Finally, it should be noted that the signatures of normalizedsensor outputs can be further processed to minimize falsealarms and to identify the type of contaminants, given thata contamination library is available/developed.

VI. CONCLUSION

In this article, the design and development of a low costsensor node for real time monitoring of drinking water qualityat consumer sites is presented. The proposed sensor nodeconsist of several in-pipe water quality sensors with flatmeasuring probes. Unlike commercially available analyzers,the developed system is low cost, low power, lightweight andcapable to process, log, and remotely present data. Moreover,contamination event detection algorithms have been developedand validated to enable these sensor nodes to make deci-sions and trigger alarms when anomalies are detected. Suchimplementation is suitable for large deployments enabling asensor network approach for providing spatiotemporally richdata to water consumers, water companies and authorities.In the future, we plan to investigate the performance of theevent detection algorithms on other types of contaminants(e.g. nitrates) and install the system in several locations ofthe water distribution network to characterize system/sensorsresponse and wireless communication performance in real fielddeployments. Finally, we plan to investigate network-widefusion/correlation algorithms to assess water quality over theentire water distribution system.

ACKNOWLEDGMENT

The authors would like to thank their colleaguesDr. E. Kastanos and Dr. M. Stylianou for the preparation andtesting of E.coli and Arsenic contaminated water samples.

REFERENCES

[1] T. P. Lambrou, C. C. Anastasiou, and C. G. Panayiotou, “A nephelo-metric turbidity system for monitoring residential drinking water qual-ity,” in Sensor Networks Applications, Experimentation and Logistics.New York, NY, USA: Springer-Verlag, 2009, pp. 43–55.

[2] T. P. Lambrou, C. G. Panayiotou, and C. C. Anastasiou, “A low-costsystem for real time monitoring and assessment of potable water qualityat consumer sites,” in Proc. IEEE Sensors, Oct. 2012, pp. 1–4.

[3] S. Zhuiykov, “Solid-state sensors monitoring parameters of water qualityfor the next generation of wireless sensor networks,” Sens. Actuators B,Chem., vol. 161, no. 1, pp. 1–20, 2012.

[4] A. Aisopou, I. Stoianov, and N. Graham, “In-pipe water quality mon-itoring in water supply systems under steady and unsteady state flowconditions: A quantitative assessment,” Water Res., vol. 46, no. 1,pp. 235–246, 2012.

[5] S. Panguluri, G. Meiners, J. Hall, and J. G. Szabo, “Distribution systemwater quality monitoring: Sensor technology evaluation methodologyand results,” U.S. Environ. Protection Agency, Washington, DC, USA,Tech. Rep. EPA/600/R-09/076, 2009.

2772 IEEE SENSORS JOURNAL, VOL. 14, NO. 8, AUGUST 2014

[6] J. Hall et al., “On-line water quality parameters as indicators ofdistribution system contamination,” J. Amer. Water Works Assoc., vol. 99,no. 1, pp. 66–77, 2007.

[7] M. V. Storey, B. Gaag, and B. P. Burns, “Advances in on-line drinkingwater quality monitoring and early warning systems,” Water Res.,vol. 45, no. 2, pp. 741–747, Jan. 2011.

[8] Hach HST, GuardianBlue Early Warning System Brochure. Loveland,CO, USA: Hach Company, 2008.

[9] BioSentry Contamination Warning System Overview, JMAR, Wyoming,MI, USA, 2006.

[10] BlueBox Intelligent Water Analytics System, Whitewater Technologies,West Jordan, UT, USA, 2012.

[11] D. Hart, S. McKenna, K. Klise, V. Cruz, and M. Wilson, “CANARY:A water quality event detection algorithm development tool,” in Proc.World Environ. Water Resour. Congr., 2007.

[12] Guidelines for Drinking-Water Quality, World Health Organization,Geneva, Switzerland, 2011.

[13] Drinking Water Regulations (No.2), European Communities, Europe,2007.

[14] U.S. Environmental Protection Agency, “Drinking water standards andhealth advisories,” Tech. Rep. EPA 822-S-12-001, 2012.

[15] M. Yunus and S. Mukhopadhyay, “Novel planar electromagnetic sensorsfor detection of nitrates and contamination in natural water sources,”IEEE Sensors J., vol. 11, no. 6, pp. 1440–1447, Jun. 2011.

[16] K. N. Power and L. A. Nagy, “Relationship between bacterial regrowthand some physical and chemical parameters within Sydney’s drinkingwater distribution system,” Water Res., vol. 33, no. 3, pp. 741–750, 1999.

[17] C. C. Anastasiou, P. Grafias, T. P. Lambrou, A. Kalli, andC. Onisiphorou, “Use of holding tanks and their effect on the qualityof potable water in households; The case study of Cyprus,” in Proc.Protection Restoration Environ. XI, 2012, pp. 1–10.

[18] J. De Zuane, Handbook of Drinking Water Quality. New York, NY,USA: Wiley, 1997.

[19] T. P. Lambrou, A Technical Report for Water Qualitative and Quantita-tive Sensors. Nicosia, Cyprus: Univ. Cyprus, 2011.

[20] A. Whelan and F. Regan, “Antifouling strategies for marine and riverinesensors,” J. Environ. Monitor., vol. 8, no. 9, pp. 880–886, 2006.

[21] J. Genser and K. Efimenko, “Recent development in superhydrophobicsurfaces and their relevance to marine fouling: A review,” Biofouling,vol. 22, nos. 5–6, pp. 339–360, 2006.

[22] A. J. Scardino, H. Zhang, D. J. Cookson, R. N. Lamb, and R. de Nys,“The role of nano-roughness in antifouling,” Biofouling, vol. 25, no. 8,pp. 757–767, 2009.

[23] Flat Surface Operating Principles, Sensorex Corporation, Garden Grove,CA, USA, 2013.

[24] M. Sophocleous, M. Glanc, Monika, J. Atkinson, and E. Garcia-Breijo,“The effect on performance of fabrication parameter variations ofthick-film screen printed silver/silver chloride potentiometric referenceelectrodes,” Sens. Actuators A, Phys., vol. 197, pp. 1–8, Aug. 2013.

[25] J. Atkinson, J. M. Glanc, M. Prakorbjanya, M. Sophocleous, R. Sion,and E. Garcia-Breijo, “Thick film screen printed environmental andchemical sensor array reference electrodes suitable for subterranean andsubaqueous deployments,” Microelectron. Int., vol. 30, no. 2, pp. 92–98,2013.

[26] R. M. Manez, J. Soto, E. G. Breijo, J. I. Civera, and E. G. Morant,“System for determining water quality with thick-film multisensor,” inProc. Spanish Conf. Electron Devices, Feb. 2005, pp. 607–610.

[27] R. Martinez-Manez, J. Soto, E. García-Breijo, L. Gil, J. Ibanez, andE. Gadea, “A multisensor in thick-film technology for water qualitycontrol,” Sens. Actuators A, Phys., vol. 120, no. 2, pp. 589–595, 2005.

[28] S. Zhuiykov, D. O’Brien, and M. Best, “Water quality assessment byintegrated multi-sensor based on semiconductor RuO2 nanostructures,”Meas. Sci. Technol., vol. 20, no. 9, p. 095201, 2009.

[29] P. Corke, T. Wark, R. Jurdak, H. Wen, P. Valencia, and D. Moore,“Environmental wireless sensor networks,” Proc. IEEE, vol. 98, no. 11,pp. 1903–1917, Nov. 2010.

[30] S. McKenna, W. Mark, and K. Katherine, “Detecting changes in waterquality data,” J. Amer. Water Works Assoc., vol. 100, no. 1, pp. 74–85,2008.

[31] A. Jonathan, M. Housh, L. Perelman, and A. Ostfeld, “A dynamicthresholds scheme for contaminant event detection in water distributionsystems,” Water Res., vol. 47, no. 5, pp. 1899–1908, 2013.

[32] C. Jimenez-Jorquera, J. Orozco, and A. Baldi, “ISFET based microsen-sors for environmental monitoring,” Sensors, vol. 10, no. 1, pp. 61–83,2010.

[33] J. V. Capella, A. Bonastre, R. Ors, and M. Peris, “A wireless sensornetwork approach for distributed in-line chemical analysis of water,”Talanta, vol. 80, no. 5, pp. 1789–1798, 2010.

[34] G. Tuna, O. Arkoc, and K. Gulez, “Continuous monitoring of waterquality using portable and low-cost approaches,” Int. J. Distrib. SensorNetw., vol. 2013, pp. 1–11, May 2013.

[35] T. P. Lambrou and C. G. Panayiotou, “A testbed for coverage controlusing mixed wireless sensor networks,” J. Netw. Comput. Appl., vol. 35,no. 2, pp. 527–537, 2012.

[36] T. P. Lambrou and C. G. Panayiotou, “Collaborative path planning forevent search and exploration in mixed sensor networks,” Int. J. Robot.Res., vol. 32, pp. 1424–1437, Sep. 2013.

[37] T. P. Lambrou and C. G. Panayiotou, “Improving area coverage usingmobility in sensor networks,” in Proc. Int. Conf. Intell. Syst. Comput.,2006.

[38] J. V. Capella, A. Bonastre, R. Ors, and M. Peris, “In line river moni-toring of nitrate concentration by means of a wireless sensor networkwith energy harvesting,” Sens. Actuators B, Chem., vol. 177, no. 1,pp. 419–427, 2013.

[39] S. Sudevalayam and P. Kulkarni, “Energy harvesting sensor nodes:Survey and implications,” IEEE Commun. Surveys Tuts., vol. 13, no. 3,pp. 443–461, Jul. 2013.

[40] G. Tuna and K. Gulez, “Energy harvesting techniques for industrialwireless sensor networks,” in Industrial Wireless Sensor Networks:Applications, Protocols, Standards, and Products. Cleveland, OH, USA:CRC Press, 2013.

[41] A. Whittle, M. Allen, A. Preis, and M. Iqbal, “Sensor networks formonitoring and control of water distribution systems,” in Proc. 6th Conf.Structural Health Monitor. Intell. Infrastruct., 2013.

[42] KAPTA 3000-AC4 Product Technical Datasheet, KAPTA, Boulder, CO,USA, 2014.

[43] B. H. Tangena et al., “A novel approach for early warning of drinkingwater contamination events,” in Proc. Water Contaminat. EmergenciesConf., 2011, pp. 13–31.

Theofanis P. Lambrou (S’05–M’12) is a Post-Doctoral Researcher at KIOSResearch Center, University of Cyprus, Nicosia, Cyprus. He received theDiploma degree in electrical and computer engineering from the NationalTechnical University of Athens, Athens, Greece, in 2004, and the Ph.D.degree in electrical engineering from the University of Cyprus in 2011.His research interests include wireless sensor networks, sensor technologiesand signal conditioning, sensors, instrumentation and embedded networkedsystems design, water monitoring, decision algorithms, cooperative controlfor distributed robotic systems, and motion planning for multirobot systems.

Christos C. Anastasiou received the B.S. and M.E. degrees in environmentalengineering from the University of Florida, Gainesville, FL, USA, and thePh.D. degree in civil engineering (environmental systems analysis) from NorthCarolina State University, Raleigh, NC, USA. He is currently the Chair-man of the Department of Civil Engineering, Frederick University, Nicosia,Cyprus. His research focuses on mathematical programming/optimization andthe development of decision support systems, especially as these apply towastewater treatment and reuse, and agricultural waste management.

Christos G. Panayiotou (M’94–SM’06) is the Deputy Director of the KIOSResearch Center and an Associate Professor of Electrical and ComputerEngineering with the University of Cyprus. He received the B.Sc. andPh.D. degrees in electrical and computer engineering from the Universityof Massachusetts at Amherst, in 1994 and 1999, respectively, and the M.B.A.degree from the Isenberg School of Management, University of Massachusetts,in 1999. His research interests include distributed control systems, wireless,ad hoc and sensor networks, computer communication networks, optimizationand control of discrete-event systems, and fault tolerant systems.

Marios M. Polycarpou (M’92–SM’98–F’06) is the Director of the KIOSResearch Center and a Professor of Electrical and Computer Engineering withthe University of Cyprus. He received the B.A. degree in computer scienceand the B.Sc. degree in electrical engineering from Rice University, USA,and the M.S. and Ph.D. degrees in electrical engineering from the Universityof Southern California, USA. His teaching and research interests includeintelligent systems and control, adaptive and cooperative control systems,computational intelligence, fault diagnosis, and distributed agents.