You are on page 1of 44

1

Fault Detection in Wireless Sensor Network Using Distributed Approach


A Thesis submitted in partial fulfillment of the requirements for the degree of

Bachelor of Technolog
in

!o"puter Science # $ngineering


by

A"lan %u"ar Na ak 'oll No( )*+!S*,-

Bibhas &ishra 'oll No( )*+!S*,*

Depart"ent of !o"puter Science # $ngineering National .nstitute of Technolog 'ourkela 'ourkela/ 0rissa/ 1-2 **+/ .ndia &a ,*),
2

Depart"ent of !o"puter Science # $ngineering National .nstitute of Technolog 'ourkela/ 1-2**+

!$'T.F.!AT$
This is to certify that the thesis entitled, Distributed Fault Detection .n Wireless Sensor Networks by &r3 A"lan %u"ar Na ak # &r3 Bibhas &ishra in partial fulfillment of the requirements for the award of Bachelor of Technology Degree in Computer Science & Engineering at the National nstitute of Technology, !our"ela is an authentic wor" carried out by them under my super#ision and guidance$ To the best of my "nowledge the matter embodied in the thesis has not been submitted to any other %ni#ersity& nstitute for the award of any Degree or Diploma$

Date'

4rof3 &3N Sahoo


Dept$ of (omputer )cience * +ngineering National nstitute of Technology !our"ela,-./001

3(4N567+D8+9+NT
6e would li"e to e:press our heartfelt than"s and gratitude to ;rofessor 9$N$ )ahoo, Department of (omputer )cience +ngineering, N T !our"ela, our guide and our mentor, who was with us during e#ery stage of the wor"< and whose guidance and #aluable suggestions has made this wor" possible$ =urther, we would also li"e to than" all the faculty members and staff of Department of (omputer )cience +ngineering, N T !our"ela for their in#aluable support and help during the entire pro>ect wor"$ 6e would lo#e to than" our family members for encouraging us at e#ery stage of this pro>ect wor"$ 7ast but not least, our sincere than"s to all our friends who ha#e patiently e:tended all sorts of help for accomplishing this underta"ing$

SUB&.TT$D B5(

A"lan %u"ar Na ak )*+!S*,!o"puter Science # $ngineering

Bibhas&ishra )*+!S*,*

National .nstitute of Technolog 'ourkela

3B)T!3(T
n recent days, 6ireless )ensor Networ"s are emerging as a promising and interesting area$ 6ireless )ensor Networ" consists of a large number of heterogeneous&homogeneous sensor nodes which communicates through wireless medium and wor"s cooperati#ely to sense or monitor the en#ironment$ The number of sensor nodes in a networ" can #ary from hundreds to thousands$ The node senses data from en#ironment and sends these data to the gateway node$ 9ostly 6)Ns are used for applications such as military sur#eillance and disaster monitoring$ 6e propose a distributed locali@ed faulty sensor detection algorithm where each sensor identifies its own status to be either good or faulty which is then supported by its neighbors as they also chec" the node beha#ior$ =inally, the algorithm is tested under different number of faulty sensors in the same area$ 5ur )imulation results demonstrate that the time consumed to find out the faulty nodes in our proposed algorithm is relati#ely less with a large number of faulty sensors e:isting in the networ"$

!ontents
1$ NT!5D%(T 5N $$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$ / 1$1 6hat is 6)NB$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$ / 1$2 ntroduction to 6)N $$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$ / 1$2 6)N application $$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$ 10 1$? )ensor Networ" application classes $$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$ 11 1$?$1 +n#ironmental data collection$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$ 11 1$?$2 )ecurity 9onitoring $$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$ 12 1$?$2 Node trac"ing scenarios $$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$ 12 1$A 6)N )ystem 3rchitecture $$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$ 12 1$A$1 Networ"ing Topologies$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$ 12 2$ 7 T+!3T%!+ !+C +6 $$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$ 1. 2$1 3 )elf,9anaging =ault 9anagement 9echanism for 6)N$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$ 1. 2$1$1 =ault Detection and Diagnosis $$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$ 1. 2$1$2 =ault !eco#ery $$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$ 11 2$2 Distributed =ault Detection in 6)N $$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$ 20 2$2$1 Definition $$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$ 20 2$2$2 3n +:ample $$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$ 22 2 ;!5;5)+D 9+(D3N )9$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$ 22$1 Networ" model and =ault model$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$ 22$2 ?$ ssues in the +:isting 3lgorithm $$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$ 21 2$? ;roposed 3lgorithm $$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$ 2/ ) 9%73T 5N 3ND !+)%7T)$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$ 22 ?$1 +:isting 3lgorithm$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$ 22 ?$2 mpro#ed 3lgorithm $$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$ 2? A$ (5N(7%) 5N $$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$ 21 A$1 (onclusion $$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$ 21 A$2 =uture wor" $$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$ 21 REFERENCES 39

7 )T 5= = 8%!+)
=igure 1$1 6)N 3pplication 3reasEEEEEEEEEEEEEEEE$ E$$$ 11 =igure 1$2 6)N Networ" TopologiesEEEEEEEEEEEEEEE$ $E$$ 1? =igure 2$1 =ault Detection and Diagnosis ;rocessEEEEEEEEEEEE$$ 1=igure 2$2 Cirtual 8rid of NodesEEEEEEEEEEEEEEEEEEE$ 11 =igure 2$2 3 partial set of sensor nodes in a 6)N with faulty sensorsEEEE$$$ 22 =igure 2$1$)ensor nodes randomly deployed o#er an areaEEEEEEEEE$$ 2=igure ?$1$ (omparison of No$ of Nodes Cs Time +lapsedEEEEEEEEE 2.

7 )T 5= T3B7+)
Table 2$1 3nalysis of =aultsEEEEEEEEEEEEEEEEEEEEE$ 2A

!6A4T$' )

.NT'0DU!T.0N

)3 .NT'0DU!T.0N )3) What is WSN7 6ireless sensor networ"s ha#e seen tremendous ad#ances and utili@ation in the past two decades$ )tarting from petroleum e:ploration, mining, weather and e#en battle operations, all of these require sensor applications$ 5ne reason behind the growing popularity of wireless sensors is that they can wor" in remote areas without manual inter#ention$ 3ll the user needs to do is to gather the data sent by the sensors, and with certain analysis e:tract meaningful information from them$ %sually sensor applications in#ol#e many sensors deployed together$ These sensors form a networ" and collaborate with each other to gather data and send it to the base station$ The base station acts as the control centre where the data from the sensors are gathered for further analysis and processing$ n a nutshell, a wireless sensor networ" F6)NG is a wireless networ" consisting of spatially distributed nodes which use sensors to monitor physical or en#ironmental conditions$ These nodes combine with routers and gateways to create a 6)N system$ )3, .ntroduction to WSN The 6)N is made of nodes from a few to se#eral hundred, where each node is connected to one or se#eral sensors$ The basic components of a node are o )ensor and actuator , an interface to the physical world designed to sense the en#ironmental parameters li"e pressure and temperature$ o (ontroller , is to control different modes of operation for processing of data o 9emory , storage for programming data$ o (ommunication , a de#ice li"e antenna for sending and recei#ing data o#er a wireless channel$ o ;ower )upply, supply of energy for smooth operation of a node li"e battery$ The topology of the 6)Ns can #ary from a simple star networ" to an ad#anced wireless mesh networ"$ The propagation technique among the nodes of the networ" could be routing or flooding$ The power of the wireless sensor networ"s lies in the capability to deploy large numbers of small nodes that assemble and configure themsel#es$ n addition to drastically decreasing the installation costs, wireless sensor networ"s ha#e the capability to dynamically adapt to changing en#ironments$ 3daptation mechanisms can lead to changes in networ" topologies or can cause the networ" to shift between different modes of operation$ The characteristics of sensor nodes are as follows' o !esource (onstraint o %n"nown topology before deployment o %nattended and unprotected once deployed o %nreliable wireless communication 10

Due to the abo#e characteristics, 6)N are easily #ulnerable to attac"s$ ;ro#iding security solutions to these networ"s is difficult due to its characteristics such as tiny nature and constraints in resources$ )38 WSN application ? 3rea 9onitoring' t is a common application of 6)Ns$ Dere the 6)N is deployed o#er a region where some e#ent is to be monitored$ 3 military e:ample is the use of sensors to diagnosis enemy intrusion H.I$ 6hen the sensors detect the e#ent being monitored, the e#ent is reported to one of the base stations, which then ta"es rele#ant action$ )imilarly, wireless sensor networ"s may use a range of sensors to detect the presence&absence of #ehicles ranging from motorcycles to train cars$ +n#ironmental 9onitoring' 6ireless sensor networ"s ha#e been deployed in se#eral cities to monitor the concentration of dangerous gases for citi@ens$ 6ireless sensor networ"s can also be used to reduce the temperature and humidity le#els inside greenhouses H.I$ 9edical 3pplication' )ensor networ"s may also be broadly used in health care centres$ n some modern hospital sensor networ"s are designed to super#ise patient physiological data, to reduce the drug administration trac" and monitor patients and doctors inside the hospital$ )tructural monitoring( 6ireless sensors are used to monitor the mo#ement within large buildings and infrastructure such as bridges, flyo#ers, emban"ments, tunnels etc$ Traffic 9onitoring' The sensor node has a built,in magneto,resisti#e sensor that measures changes in the +arthJs magnetic field caused by the e:istence or passing of a #ehicle in the pro:imity of the node H-I$ By placing two nodes a few metres apart in the direction of traffic, accurate indi#idual #ehicle speeds can be calculated and reported$ Dabitat 9onitoring' The intimate connection with its immediate physical en#ironment allows each sensor to pro#ide locali@ed measurements and detailed report which is hard to obtain through traditional instrumentation$

? ?

11

!emote 9onitoring

sustainability

ndustrial 9easurements

3ir&(limate 6ater&)oil

ndoor 9onitoring

;ower system

)olar

6ind =arm

12

)tructural 9onitoring 9achine 9onitoring ;rocess 9onitoring =igure 1$1$6)N 3pplication 3reas )39 Sensor Network application classes The three application classes we ha#e selected are' en#ironmental data collection, security monitoring, and sensor node trac"ing$ The ma>ority of wireless sensor networ" deployments will be classified into one of these following class templates$ 1$?$1 +n#ironmental data collection 3t the networ" le#el, the en#ironmental data collection application is distinguished by ha#ing a large number of nodes continuously sensing and transmitting information bac" to a set of connected base stations which store the information using traditional methods H10I$ These networ"s generally need #ery low data rates and intensely long lifetimes$ n typical usage scenario, the nodes would be e#enly distributed o#er an outdoor en#ironment$ This distance between neighbor nodes will be minimal yet the distance across the entire networ" will be con#incing$

12

3fter deployment, the nodes must first disco#er the topology of the networ" and e#aluate optimal routing strategies$ The routing strategy may then be used to route the data to a central collection points$ n en#ironmental monitoring applications, it is not essential that the nodes establish the optimal routing strategies on their own$ nstead, it may be possible to compute the optimal routing topology outside of the networ" and then communicate the necessary information to the nodes as required which is possible because the physical topology of the networ" is relati#ely constant H11I$ 1$?$2 )ecurity 9onitoring 5ur second class of sensor networ" application is security monitoring$ )ecurity monitoring networ"s are built of nodes which are placed at fi:ed locations throughout an en#ironment that continuously control one or more sensors to detect an anomaly$ 3 ma>or difference between security monitoring and en#ironmental monitoring is that security networ"s do not collect any data$ This leads to a significant impact on the optimal networ" architecture $+ach node frequently chec" the status of its sensors but it only transmit a data report when there will a security #iolation$ The immediate and reliable communication among the alarm messages is the systemKs primary requirement$ These are the report generated by e:ception networ"s$ 3dditionally, it is essential that it is #alidated that each node is still present and wor"ing$ f a node is disabled or fail, it will represent a security #iolation that must be reported$ =or security monitoring applications, the networ" should be configured so that nodes are responsible for finding the status of each other$ 6e ha#e one approach where each node is assigned to peer that will report if a node does not function$ The optimal topology of a security monitoring networ" will loo" quite different from that of a data collection networ"$ 5nce detected, a security #iolation should be communicated to the connected station immediately$ The latency of the information communication across the networ" to the base station has a se#ere impact on application performance$ %sers demand that alarm situations should be reported within seconds of detection$ This means that networ" nodes should be able to respond readily to requests from their neighbors to forward data H12I$ 1$?$2 Node trac"ing scenarios 3 third usage scenario commonly analy@ed for wireless sensor networ"s is the tracing of a tagged ob>ect through a region of space controlled by a sensor networ"$ There are #arious situations where one would li"e to trace the location of #aluable assets or personnel$ (urrent control systems attempt to trac" ob>ects by recording the last chec"point which an ob>ect passed through$ Dowe#er, with these traditional systems it is not easily possible to determine the ob>ectKs current location$ =or e:ample, %;) trac"s e#ery shipment by scanning it with a barcode whene#er it passes through a routing center$ The system brea"s down when ob>ects do not flow from chec"point to chec"point$ n typical wor" en#ironments it is not practical to e:pect ob>ects to be continuously passed through chec"points$ 1?

%sing wireless sensor networ"s, ob>ects can be traced by simply tagging them with a tiny sensor node$ The sensor node will be traced as it mo#es through an area of sensor nodes which are deployed in the en#ironment at "nown locations$ nstead of sensing the en#ironmental data, these nodes will be deployed to sense the messages of the nodes which are attached to #arious ob>ects$ The nodes may be used as acti#e tags which announce the presence of a de#ice$ 3 database may be used to record the location of traced ob>ects relati#e to the large set of nodes at "nown locations$ )3: WSN S ste" Architecture n a common 6ireless )ensor Networ" architecture, the measurement nodes are deployed to calculate measurements such as temperature, #oltage, heat, or e#en dissol#ed o:ygen$ The nodes are part of a wireless sensor networ" administered by the gateway that go#erns networ" aspects such as client authentication and data security H1?I$ The gateway collects the measurement data from each and e#ery node and sends it through a wired connection, typically +thernet, to a host controller$ )3:3) Networking Topologies 6e can use se#eral networ" topologies to coordinate the 6ireless sensor networ" gateway, end nodes, and router nodes$ !outer nodes are much similar to end nodes in that they can store measurement data, but they also can be used to pass along measurement data from other nodes H1AI$ The first, and most basic topology, is the star topology, in which each node maintains a single, direct communication lin" with the gateway$ This topology is #ery simple but restricts the o#erall distance that our networ" can achie#e$ To increase the distance that a networ" can co#er, you could implement a cluster, or tree, topology$ n this more comple: architecture scenario, each node still uses only one communication path to the gateway but can use other nodes to route its information along that path$ This topology suffers from a typical problem, howe#er$ f a router node goes down, all the nodes which depend on that router node also lose their communication lin"s to the gateway$ The mesh networ" topology reduces this issue by e:tensi#ely using redundant communication paths to increase reliability of the system$ n a mesh networ", nodes maintain multiple communication lin"s bac" to the gateway, so that if a router node goes down or does not wor" properly, the networ" automatically reroutes the data through a different sets of path$ The mesh topology, although #ery reliable, suffers from an wide increase in networ" latency because data must ma"e multiples of hops before successfully arri#ing at the gateway H1.I$

1A

=igure 1$2$6)N Networ" Topologies

14

!6A4T$' ,

;.T$'ATU'$ '$<.$W

1-

,3 ;.T$'ATU'$ '$<.$W n this section, we li"e to gi#e a brief re#iew of the different schemes that we ha#e studied for wireless sensor networ"s$ 2$1 3 )elf,9anaging =ault 9anagement 9echanism for 6)N n this approach a new fault management mechanism was proposed to deal with fault detection and reco#ery$ t proposes a hierarchical structure to properly distribute fault management tas"s among sensor nodes by hea#ily introducing more self,managing functions$ The proposed failure detection and reco#ery algorithms ha#e been compared with some e:isting related algorithm and pro#en to be more energy efficient$ The proposed fault management mechanism can be di#ided into two phases' o =ault detection and diagnosis o =ault reco#ery 2$1$1 =ault Detection and Diagnosis Detection of faulty sensor nodes can be achie#ed by two mechanisms i$e$ self,detection For passi#e,detectionG and acti#e,detection$ n self,detection, sensor nodes are required to periodically monitor their residual energy, and identify the potential failure$ n this scheme, we consider the battery depletion as a main cause of node sudden death$ 3 node is termed as failing when its energy drops below the threshold #alue$ 6hen a common node is failing due to energy depletion, it sends a message to its cell manager that it is going to sleep mode due to energy below the threshold #alue H1-I$ This requires no reco#ery steps$ )elf,detection is considered as a local computational process of sensor nodes, and requires less in,networ" communication to conser#e the node energy$ n addition, it also reduces the response delay of the management system towards the potential failure of sensor nodes H11I$ To efficiently detect the node sudden death, our fault management system employed an acti#e detection mode$ n this approach, the message of updating the node residual battery is applied to trac" the e:istence of sensor nodes$ n acti#e detection, cell manager as"s its cell members on regular basis to send their updates$ )uch as the cell manager sends get messages to the associated common nodes on regular basis and in return nodes send their updates$ This is called in,cell update cycle$ The updateLmsg consists of node D, energy and location information$ 3s shown in figure 2$1, e:change of update messages ta"es place between cell manager and its cell members$ f the cell manager does not recei#e an update from any node then it sends an instant message to the node acquiring about its status H1/I$ f cell manager does not recei#e the ac"nowledgement in a gi#en time, it then declares the node faulty and passes this information to the remaining nodes in the cell$ (ell managers only concentrate on its cell members and only inform the group manager for further assistant if the networ" performance of its small region has been in a critical le#el$ 11

=igure 2$1$=ault Detection and Diagnosis ;rocess 3 cell manager also employs the self,detection approach and regularly monitors its residual energy status$ 3ll sensor nodes start with the same residual energy$ 3fter going through #arious transmissions, the node energy decreases$ f the node energy becomes less than or equal to 20M of battery life, the node is ran"ed as low energy node and becomes liable to put to sleep$ f the node energy is greater or equal to A0M of the battery life, it is ran"ed as high and becomes the promising candidate for the cell manager$ Thus, if a cell manager residual energy becomes less than or equal to 20M of battery life, it then triggers the alarm and notifies its cell members and the group manager of its low energy status and appoints a new cell manager to replace it$ +#ery cell manager sends health status information to its group manager$ This is called out,cell update cycle and are less frequent than in,cell update cycle$ f a group manager does not hear from a particular cell manager during out,cell update cycle, it then sends a quic" reminder to the cell manager and enquires about its status$ f the group manager does not hear from the same cell manager again during second update cycle, it then declares the cell manager faulty and informs its cell members H20I$ This approach is used to detect the sudden death of a cell manager$ 8roup manager also monitor its health status regularly and respond when its residual energy drops below the threshold #alue$ t notifies its cell members and neighboring group managers of its low energy status and an indication to appoint a new group manager$ )udden death of a group manager can be detected by the base station$ f the bases station does not recei#e any traffic from a particular group manager, it then consults the group manager and as"s for its current status$ f the base station does not recei#e any ac"nowledgement, it then considers the group manager faulty Fsudden deathG and propagates this information to its cell managers$ The base station 1/

primarily focuses on the e:istence of the group managers from their sudden death$ 9eanwhile, the group managers and cell managers ta"e most parts in passi#e and acti#e detection in the networ"$ 2$1$2 =ault !eco#ery 3fter nodes failure detection Fas a result of self,detection or acti#e detectionG, sleeping nodes can be awa"ed to co#er the required cell density or mobile nodes can be mo#ed to fill the co#erage hole$ 3 cell manager also appoints a secondary cell manager within its cell to acts as a bac"up cell manager$ (ell manager and secondary cell manager are "nown to their cell members$ f the cell manager energy drops below the threshold #alue Fi$e$ less than or equal to 20M of battery lifeG, it then sends a message to its cell members including secondary cell manager$ t also informs its group manager of its residual energy status and about the candidate secondary cell manager$ This is an indication for secondary cell manager to stand up as a new cell manager and the e:isting cell manager becomes common node and goes to a low computational mode$ (ommon nodes will automatically start treating the secondary cell manager as their new cell manager and the new cell manager upon recei#ing updates from its cell members< choose a new secondary cell manager H20I$ The failure reco#ery mechanisms are performed locally by each cell$ n =igure 2$2, let us assume that cell 1 cell manager is failing due to energy depletion and node 2 is chosen as secondary cell manager$ (ell manager will send a message to node 1, 2, 2 and ? and this will initiate the reco#ery mechanism by in#o"ing node 2 to stand up as a new cell manager$

=igure 2$2$Cirual 8rid of Nodes n a scenario, where the residual battery energy of a particular cell manager is not sufficient enough to support its management role, and the secondary cell manager also does not ha#e sufficient energy to replace its cell manager$ Thus, common nodes e:change energy messages 20

within the cell to appoint a new cell manager with residual energy greater or equal to A0M of battery life$ n addition, if there is no candidate node within the cell that has sufficient energy to replace the cell manager$ The e#ent cell manager sends a request to its group manager to merge the remaining nodes with the neighboring cells$ 6hen a group manager detects the sudden death of a cell manager, it then informs the cell members of that faulty cell manager Fincluding the secondary cell managerG$ This is an indication for the secondary cell manager to start acting as a new cell manager$ 3 group manager also maintains a bac"up node within the group to replace it when required$ f the group manager residual energy drops below the threshold #alue Fi$e$ greater or equal to A0M of battery lifeG, it may downgrade itself to a common node or enter into a sleep mode, and notify its bac"up node to replace it$ The information of this change is propagated to neighboring group managers and cell managers within the group$ 3s a result of group manager sudden death, the bac"up node will recei#e a message from the base station to start acting as the new group manager$ f the bac"up node does not ha#e enough energy to replace the group manager, cell managers within a group co,ordinate to appoint a new group manager for themsel#es based on residual energy$ +ach cell maintains its health status in terms of energy$ t can be Digh, 9edium or 7ow$ These health statuses are then sent out to their associate group managers periodically during out, cell update cycle$ %pon recei#ing these health statuses, group manager predict and a#oid future faults$ =or e:ample< if a cell has health status high then group manager always recommends that cell for any operation or routing but if the health status is medium then group manager will occasionally recommend it for any operation H21I$ Dealth status 7ow means that the cell has insufficient energy and should be a#oided for any operation$ Therefore, a group manager can easily a#oid using cells with low health status or alternati#ely, instruct the low health status cell to >oin the neighboring cell$ (onsider =igure 2$2, let cell ? manager is a group manager and it recei#es health status updates from cell 1, 2 and 2$ (ell 2 sends a health status low to its group manager, which alert group manager about the energy status of cell 2 H22I$

21

,3, Distributed Fault Detection in WSN 2$2$1 Definition n: total n umber of sensors; p: probability of fa ilure of a sensor; k: number of neighbo r sensors; S: set of all the se nsors; N(Si): set of the neighb ors of Si; xi: measurement of S i; ij d t : measurement differ en e bet!e en Si and Sj at time t "

d ij ? x ? i x ;
#tl ? tl ? 1 ? tl; ij

#d #tl : measurement differ en e bet!e en Si and Sj from


#tl

ij ij i ij: test bet !een Si and Sj" ij ? $ 0" 1%" ij ? ji; &1 and & 2: t!o predefined thr eshold 'alues;

time tl to tl ? 1" ij #

?d

tl ?1

? d ? (x ? x ) ? (x ? x );
j

tl

tl ? 1

tl ?1

tl

tl

Ti: tenden y 'alue of a sensor" Ti ? $()" (*" )+" *T%;


2.2.2 Existing Algorithm

ij

)T+; 1

,a h sensor Si" set ij ? 0 and ompute d t ; ij ? &1 T/,N -* .d t . ; 0al ulate #l #tl ij -* .#d #tl. ? & 2 T/,N ji ? 1;

ij

22

)T+; 2 -*
Sj ? N(Si)

the ? number of the S i2s neighboring nodes Ti ? (); ,(S, Ti ? (*; 0ommuni ate Ti to neighb ors;

ij

? .N(Si).1 2 " !here .N(Si). is ? ? T/,N

)T+; 2 -*

?
( Sj ? N(Si) a n d Tj ? () T/,N

1 ? 2 ij )

? ?

2 .N(Si).1

Ti ? )+; 0ommuni ate Ti to neighb ors; )T+; ? *34 i ? 1 to n -* Ti ? () or Ti ? (* T/,N -* Tj ? )+ ?Sj ? N(Si) T/,N -* ? 0 T/,N Ti ? )+;
ij

,(S, Ti ? *T; ,(S, repeat 0ommuni ate Ti to neighb ors; )T+; A *34 ea h Si" -* Tj ? Th ? )+ ?Sj" Sh ? N(Si)" !here j ? h" and -* ji ? hi T/,N -* Ti ? () (or (*) T/,N Ti ? )+ (or *T)

)ensors are considered as neighboring sensors if they are within the transmission range of each other$ +ach node regularly sends its measured #alue to all its neighbors$ 6e are interested in the history data if more than half of the sensorKs neighbors ha#e a significantly different #alue from Nd N to find if the current measurement is different from pre#ious it$ 6e can use this tl ij measurement$ f the measurements change o#er the time significantly, it is more li"ely the sensor is faulty$ 3 test result t ij is generated by sensor Si based on its neighbor Sj Ks measurements using two Ntl ij two predefined threshold #alue ?1and? 2 $ f a sensor is faulty, it can #ariables, d ij and Nd , and generate arbitrary measurements$ f
ij

is 0, most li"ely either both Si and Sj are good or both

are faulty$ 5therwise, if ij is 1, Si and Sj are most li"ely in different status$ )ensors can be either 78 or 7=, determined by using test #alue from its neighboring sensors$ +ach sensor sends its tendency #alue to all its neighbors$ The number of the 78 sensors with coincident test results determines whether the sensors are 8D or =T$ f a 8D sensor is found in the networ", its test result can be used to diagnose other sensorsK status$ The information can be propagated through the whole networ" to diagnose all other sensors as good or faulty$ f the diagnosis is consistent with the test results, the diagnosis is #alid$ f thereKs no sensor being diagnosed, all its neighbors are either not diagnosed or are diagnosed ? F1 ? 20ij G as faulty$ That is ? N FSiG and Tj ? 78 , 1 ? 0ij O 0ij must be greater or ?Sj ? ? ? equal to P N FSiG P & to claim Si is good$ n other words, a Si will be diagnosed as 8D in 2 ? good the first round if it has less than "&? bad neighbors$ f a 8D sensor is found in the networ", its test result can be used to diagnose other sensorsK status$ The information can be propagated through the whole networ" to diagnose all other sensors as good or faulty$ f the diagnosis is consistent with the test results, the diagnosis is #alid$ f thereKs no sensor being diagnosed, all its neighbors are either not diagnosed or are diagnosed as faulty$

2$2$2 3n +:ample

=igure 2$2$3 partial set of sensor nodes in a 6)N with faulty sensors

n this section, we present an e:ample to illustrate our algorithm$ =ig$2$2 shows a partial set of S1 O S / inside the sensor nodes in a wireless sensor networ" with some faulty nodes$ Nodes circle area are the nodes which we are interested in$ f the two nodes are neighbors, they are connected by dotted line$ (ommunication between nodes outside the circle is not shown in the figure$ +ach node inside the interested area is tested by its neighbors$ Test results are either 0 or 1 depending upon the measurement difference and threshold #alue ? $ Tendency #alue Ti is finali@ed at the third iteration$ Table 2$1 lists the analysis results obtained by applying the 7ocali@ed =ault Detection 3lgorithm$ =our out of nine sensor nodes in the area are faulty$ The other fi#e nodes are good and there is no ambiguity occurring in this e:ample$ +ach nodeKs neighbors with 8D tendency #alue generate the same testing results when they determine the nodeKs status$ =irst, each of
S1 O S / generates
ij

test results for all their neighbors in the way as specified in

step 1 of our algorithm$ The results are shown under the 2nd and the 2rd columns of Table 2$1$

)econdly, S1 O S / decide their own tendency #alue, T 1 ? T / $ f the summation of test results is less than half of the number of its neighbors, the sensor is li"ely good$ 5therwise, it is li"ely faulty$ ? ? P N FS1G P &2 ? 2 1j 1 ? ? =or Sj ? N F S 1G ?? ? $ The same test is ? T 78 1 e:ampl e, for S1 done for all ,
? ? P N FS 2G P &2 other S 2 2j 2 ? ? j? N F S 2G S nodes$ ? 2 ? ?? T 2 ? 7= $ 6e assume that =or sensors

outside the circle can decide their tendency #alue in the same way$ Then, we need to find 8D sensors from all the sensors$ 7oo" at )1, as specified in step 2 of our locali@ed fault detection algorithm

Sj ? N F S 1G and Tj ?78

F1? 201 j G ? 2 ? P N FS1G P

& 2 ?? T 1 ? 8D $

6e obtained all the #alues under the teration 1 column in Table 2$1 from this step$ =inally, by using the 8D sensors, we can test other non, 8D sensors to find out their status base upon the test results$ The #alues under teration 2 column in Table 2$1 are generated from this step$ The last step is to chec" if there is any ambiguity between any neighbours test results$ 3ll test results are consistent in this e:ample$

Table 2$1 $3nalysis of =aults

Ti in terations S
i

Sj with 0ij Q0

Sj with 0ij Q1 0 1
8D 7= 8D 7= 8D 7= 8D 7= 8D

2
8D =T 8D =T 8D =T 8D =T 8D

1 2 2 ? A . 1 /

2,A,11,12 ? 1 , 2 1, 1A 1 2,/,1? .,1-,11,1/

1 0 2,12, 12 2 , -,1?, 20 . , 2,A,,/ ?,., 1. A,/, 11 .,1, 1.

78 7= 78 7= 78 7= 78 7= 78

25

!6A4T$' 8

;!5;5)+D 9+(D3N )9

21

4'040S$D &$!6AN.S&

2$1 Networ" model and =ault model


6e assume that sensors are randomly deployed in the interested area which is #ery dense and all the sensors ha#e a common transmission range$ The dar" circles in the figure represent faulty sensors and the gray circles are good sensors$ There might be a failure occurring in a certain area as illustrated in the figure 2$1$ 3ll sensors in this area go out of ser#iceH1I$ 3s we are depending on ma>ority #oting among the sensors, we assume that each sensor node has at least 2 neighboring nodes$ Because a large amount of sensors are deployed into the interested area to form a wireless networ", this condition can be easily obtained$ +ach sensor node is able to locate its neighbors within its transmission range #ia a broadcast& ac"nowledge protocol$ =aults can occur at different le#els of the sensor networ" H1I, such as system software, hardware, physical layer, and middleware$ n this mechanism, we focus on hardware le#el faults by assuming all system software as well as the application software is always fault tolerant$ 6e can categori@e the hardware components of sensor nodes into two groups$ The first group of hardware le#el components consists of a storage subsystem, computation engine and power supply infrastructure$ The second groups of components are sensors and actuators$ The second group is most prone to malfunctioning$ 6e only consider the sensor faults which occur in the second group H1I$ )ensor nodes are still capable of recei#ing, sending, and processing when they are faulty in the algorithm$

=ailure Node

6or"ing Node

=igure 2$1$)ensor nodes randomly deployed o#er an area


2/

83, Definition

ij

n: total n umber of sensors; p: probability of fa ilure of a sensor; k: number of neighbo r sensors; S: set of all the se nsors; N(Si): set of the neighb ors of Si; xi: measurement of S i; d t : measurement differ en e bet!e en Si and Sj at time t " d ij ? x ? i x ; #tl ? tl ? 1 ? tl;
t t t

ij

#d #tl : measurement differ en e bet!e en Si and Sj from #t t t t t t t


l l ?1 l l ?1 l ?1 l

? d ? d ? (x ? x ) ? (x ? x ); time tl to tl ? 1" # ij ij ij i j ij: test bet !een Si and Sj" ij ?$ 0" 1%" ij ? ji;
&1 and & 2: t!o predefined thr eshold 'alues;

Ti: tenden y 'alue of a sensor" Ti ?$()" (*" )+" *T%;


)ensors are considered as neighboring sensors if they are within the transmission range of each other$ +ach node regularly sends its measured #alue to all its neighbors$ 6e are interested in the history data if more than half of the sensorKs neighbors ha#e a significantly different #alue from it$ 6e can find the current measurement is different from pre#ious measurement$ f the measurements change o#er the time significantly, it is more li"ely the sensor is faulty H2I$ 3 test result 0ij is generated by sensor Si based on its neighbor Sj Ks measurements using two #ariables and two predefined threshold #alue$ f a sensor is faulty, it can generate arbitrary measurements$ f 0ij is 0, most li"ely either both Si and Sj are good or both are faulty$ 5therwise, if 0ij is 1, Si and Sj are most li"ely in different status$ 2$2 ssues in the +:isting 3lgorithm , if the =rom the reali@ation of D=D node fault detection scheme, for a normal node Snormal number of its neighbor nodes ha#ing initial detection status of 78 is less than P N FSnormalG P & 2 , then Snormal is misdiagnosed as faulty, thus reducing the fault detection accuracy$ The conditions of detecting the normal node as normal are too harsh in D=D node fault detection 20

scheme$ Besides, the node fault accuracy of D=D scheme will decrease rapidly when there are not many neighbors of the nodes to be detected or the nodeKs failure ratio of networ" is high H?I$

21

The impro#ed D=D node fault detection scheme proposed in this pro>ect changes the detection criterion of D=D scheme as follows' =or any node Si and the nodes in N FSiG whose initial detection status is 78, if the nodes whose test result with Si is 0 are not less than the nodes whose test result is 1, then the status of Si is normal F8DG, otherwise, the status of Si is faulty F=TG$ 2$? ;roposed 3lgorithm )T+; 1 ij ,a h sensor Si and any sensor Sj ? N(Si) set ij ? 0 and ompute d t ij ? &1 T/,N ij ? 1and turn to th e next nod e in N(Si); ij -* .d t ? &1 . -* .d
t

. 0al ulate #a #tl ; ? & 2 T/,N ji ? 1 and turn to the nex t node in N(Si) ; -* .#d #tl.

ij ij

!e peat abo'e steps until the tes t results of ea h no de in N(Si) !ith S i are all obtained; )T+; 2 -*

Sj ? N(Si)

ij ? .N(Si).1 2 " !here .N(Si).

is the number of the S i2s neighboring nodes T/,N Ti ? (); ,(S, Ti ? (*; 0ommuni ate Ti to neighb ors; )T+; 2 -*

Sj ? N(Si) a n d Tj ? ()

0ij ? .N(Si)Tj ? ().1 2

T/,N Ti ? )+; ,(S, T/,N Ti ? *T; 0ommuni ate Ti to neighb ors;

)T+; ? -f there are no neighbor nodes of Si 5hose initial dete tion status is ()" and if the initial dete tion status Ti of Si is ()" then set the status of S as normal ()+)" other!ise as fault(*T);
i

)T+; A 0he k !hether dete tion of the status of all nodes in net!ork is ompleted or not6 -f it has been ompleted" then exit6 3ther!ise" repeat steps of (7)" (8)" (9) and (:)6

!6A4T$' 9

S.&U;AT.0N AND '$SU;TS

?$ S.&U;AT.0N AND '$SU;TS The simulation set up used is 39D Dual (ore ;rocessor with 1$.8D@ (loc" speed and 9emory of 2 8B$ The de#elopment en#ironment is D+C (RR Cersion ?$/$ 93) $=isting Algorith" ?$1$1 =or A Nodes

?$1$2 =or 1A Nodes

33

93, ."pro>ed Algorith" ?$2$1 =or A nodes

34

?$2$2 =or 1A nodes

35

?$2 (omparison

=igure ?$1$(omparison of No$ of Nodes Cs Time +lapsed

2/

!6A4T$' :

!0N!;US.0N

?0

:3 !0N!;US.0N A$1 (onclusion 6e proposed a distributed locali@ed faulty sensor detection algorithm where each sensor identifies its own status to be either good or faulty and the claim is then supported or re#erted by its neighbors as they also e#aluate the node beha#ior$ 6e ha#e shown the simulation results in the form of graphs$ By the )imulation results, we conclude that the time consumed by our approach to find out the faulty node is relati#ely less than the time consumed by the e:isting scheme$ A$2 =uture wor" n future we intend to calculate the detection accuracy for the nodes in the 6ireless )ensor Networ" where detection accuracy depicts the ratio of the number of faulty sensors detected to the total number of faulty sensors in the networ"$ The time consumed by our approach to find out the faulty node is relati#ely less$ )o we want to #erify it for larger number of nodes$

?1

4,*,4,N0,S: H1I Sinran (hen, )hubha 4her and 3run )omani$ Distributed =ault Detection of 6ireless )ensor networ"s, Dependable (omputing and Networ"ing 7ab owa )tate %ni#ersity3mes, owa $pages-A,1?,(omputer Networ",200.$ H2I D$ Blough, 8$ )ulli#an, and 8$ 9asson$ =aultdiagnosis for sparsely interconnected multiprocessor systems$ n n 1/th nt$ +++ )ymp$ on =aultTolerant(omputing, pages .2T./$ +++ (omputer )ociety, 1/1/$ H2I =$ 4oushanfar, 9$ ;ot"on>a", and 3$ )angio#anni,Cincentelli$ 5n,line fault detection of sensor measurements$ n )ensors, 2002$ ;roceedings of +++, pages /-?T/-/ Col$2, 2002$ H?I U$ 7uo, 9$ Dong, and V$ Duang$ 5n distributed fault,tolerant detection in wireless sensor networ"s$ +++ Transactions on (omputers, San200.$ HAI 3"yildi@, $=$< )u, 6$< )an"arasubramaniam, V< (ayirci, +$ 6ireless sensor networ"s' a sur#ey$0omputer Net!orks ,**,$ H.I 9$ Ding, D$ (hen, 4$ Uing, and U$ (heng$ 7ocali@ed fault,tolerant e#ent boundary detection in sensor networ"s$ n ;roceedings of +++ N=5(59 200A,200A$ H-I =$ 4oushanfar, 9$ ;ot"on>a", and 3$ )angio#anni,Cincentelli$ =ault,tolerance in sensor networ"s$ n Dandboo" of )ensor Networ"s$ (!(press, 200?$ H1I B$ 4rishnamachari and )$ yengar$ Distributed bayesian algorithms for fault,tolerant e#ent region detection in wireless sensor networ"s$ +++ Transactions on (omputers, A2F2G'2?1T2A0, 9arch 200?$ H/I 9ahmood and +$ S$ 9c(lus"ey$ (oncurrent error detection using watchdog processors, a sur#ey$(omputers, +++ Transactions on, 2-F2G'1.0T1-?, =eb 1/11$ H10I )$ 9arti, T$ S$ 8iuli, 4$ 7ai, and 9$ Ba"er$ 9itigating routing misbeha#iour in mobile ad hoc networ"s$ n 9obi(om K00' ;roceedings of the .th annual international conference on 9obile computing and networ"ing, pages 2AAT2.A, New Vor", NV, %)3, 2000$ 3(9 ;ress$ H11I ;errig, !$ )@ewc@y", C$ 6en, D$ +$ (uller, andS$ D$ Tygar$ ); N)' security protocols for sensor netowr"s$ n 9obile (omputing and Networ"ing, pages 11/T 1//, 2001$ ?2

H12I 7$ B$ !ui@, $ 8$ )iqueira, 7$ B$ e 5li#eira, D$ ($ 6ong, S$ 9$ )$ Nogueira, and 3$ 3$ =$ 7oureiro$ =ault management in e#ent,dri#en wireless sensornetwor"s$ n9)6i9K0?' H12I ;roceedings of the -th 3(9 international symposium on 9odeling, analysis and simulation of wireless and mobile systems, pages1?/T1A., New Vor", NV, %)3, 200?$ 3(9 ;ress$ H1?I 4$ )omani and C$ 4$ 3garwal$ Distributed diagnosis algorithms for regular interconnected structures$ (omputers, +++ Transactions on,?1F-G'1//T/0., Suly 1//2$

H1AI

S$ )taddon, D$ Balfan@, and 8$ Durfee$ +fficient tracing of failed nodes in sensor networ"s$ n 6)N3 K02' ;roceedings of the 1st 3(9 internationalwor"shop on 6ireless sensor networ"s and applications, pages 122T120, New Vor", NV, %)3, 2002$ 3(9 ;ress$

H1.I )$ Tanachaiwiwat, ;$ Da#e, !$ Bhindwale, and 3$ Delmy$ )ecure locations' routing on trust and isolating compromised sensors in location,aware sensor networ"s$ n )en)ys, pages 22?T22A, 2002$ H1-I 9$ Vu, D$ 9o"htar, and 9$ 9erabti, W3 sur#ey on =ault 9anagement in wireless sensor networ",W in;ro eedings of the <th Annual ;ost)raduate Symposium on The 0on'ergen e of Tele ommuni ations" Net!orking and =road asting 7i#erpool, %4, 200-$ H11I )$ 9arti, T$ S$ 8iuli, 4$7ai, and 9$ Ba"er, W9itigating routing misbeha#iour in mobile ad hoc networ"s,W in A0> >obi om, 2000, pp$ 2AA,2.A$ H1/I =$ 4oushanfar, 9$ ;ot"on>a", and 3$ )angio#anniCincentelli, W=ault tolerance techniques in wireless ad,hoc sensor networ"s,W %( Ber"eley technical reports 2002$ H20I 6$ 7$ 7ee, 3$ Datta, and !$ (ardell,5li#er, W6in9)' 6ireless )ensor Networ", 9anagement )ystem, 3n 3dapti#e ;olicy,Based 9anagement for 6ireless )ensor Networ"s,W )chool of nternational Sournal of 6ireless * 9obile Networ"s F S69NG Col$2, No$?, No#ember 2010$

?2

H21I T$ (louqueur, 4$)alu>a, and ;$ !amanathan, W=ault Tolerance in (ollaborati#e )ensor Networ"s for Target Detection,W in -,,, Transa tions on 0omputers, 200?, pp$ 220,222$ H22I 8$ 8upta and 9$ Vounis, W=ault,Tolerant (lustering of 6ireless )ensor Networ"s,W in ;ro eedings of the -,,, 50N0 8??9 New 5rleans, 7ouisiana, 2002$

??

You might also like