The Design and Evaluation of a Wireless Sensor Network for Mine Safety Monitoring

Transcription

1 The Design and Evaluaion of a Wireless Sensor Nework for Mine Safey Monioring Xiaoguang Niu 1 (Member IEEE, Xi Huang 1, Ze Zhao 1, Yuhe Zhang 1, Changcheng Huang 1 (Senior Member IEEE, Li Cui 1* (Member IEEE 1 Insiue of Compuing Technology, Chinese Academy of Sciences, Beijing, China Graduae Universiy of Chinese Academy of Sciences, Beijing, China {niuxiaoguang, huangxi, zhaoze, zhangyuhe, huangchangcheng, lcui }@ic.ac.cn * Corresponding Auhor Absrac This paper describes a wireless sensor nework for mine safey monioring. Based on he characerisics of underground mine gallery and he requiremens for mine safey monioring, we proposed a disribued heerogeneous hierarchical mine safey monioring prooype sysem, namely HHMSM. This sysem is capable of monioring mehane concenraion, and locaing miner. We proposed a novel overhearing-based adapive daa collecing scheme which explois he redundancy and correlaion of he sampling readings in boh ime and space o reduce raffic and conrol overhead wih a well-bounded offse error for large-scale sensor neworks. This mechanism is easy o implemen and low-cos compared o oher more heoreically based mechanisms such as Kalman filer. Experimenal resuls show ha HHMSM achieves beer performance on flexibiliy, correcness, coverage, and lifeime compared wih oher exising wireless mine safey monioring sysems. I. INTRODUCTION Underground mining is inherenly dangerous and is safey monioring is a complex, inerdisciplinary research area which spans several decades. In China, almos half of he local mine faaliies are caused by mehane explosions [1]. Therefore, he essenial funcion of mine safey monioring sysem is o monior and locae he mehane leakage imely. Furhermore i should mee he requiremens of low-laency, high daa fideliy, fully and reliable coverage on he underground mining unnels. The curren underground mine safey monioring sysems are of wo kinds: wired sysems and wireless ones. The wired safey monioring sysems have some inheren problems (e.g. lack of monioring in he mos hazard working area. Wireless sensor neworks have he poenial o effecively handle hese problems []. In previous work, Lauri e al proposed a wireless soluion which provides wireless and real-ime connecion wih full coverage from underground mine o he enerprise informaion sysem [3]. However his sysem requires high bandwidh and a grea amoun of power consumpion. The goal of his work is o develop a large scale, energy-saving, low cos, reusable, and effecive wireless sensor nework o mee he pracical requiremens for he mine safey monioring. I is capable of monioring he change of mehane concenraion, and locaing miners. To achieve his challenging goal we sared by sudying he exising underground mine safey monioring sysems and idenifying heir requiremens and consrains. We hen se up a hierarchical prooype sysem for underground mine safey monioring, namely HHMSM. The major conribuions in his paper consis of he following aspecs: We developed a heerogeneous hierarchical mine safey monioring nework prooype which is divided ino hree iers o achieve sysem flexibiliy and coverage. We proposed a novel overhearing-based adapive daa collecing scheme which explois he rais of he lower-ier sensor nework--redundancy and correlaion of he sampling readings in boh ime and space o selec local sampling and ransmission rules for he sensor nodes so ha he performance of he sysem is opimized, including a beer correcness and lifeime. We implemened he prooype sysem and conduced experimens wih i. The remainder of he paper is organized as follows: Secion II discusses he relaed works in sensor nework applicaions and underground mine safey monioring sysems. In Secion III, we describe he overall archiecure and design conribuions of HHMSM sysem, and provide a high-level descripion of is funcionaliy. Secion IV describes he proposed efficien daa and ransmission scheme which drives our sysem s requiremens and design. Then we presen he experimen resuls and evaluae he performance of HHMSM in Secion V. Finally, Secion VI concludes he paper. II. RELATED WORKS Numerous sensor nework applicaions have been proposed for he areas like habia monioring, srucure monioring [4, 5] and so on. As o he mine safey monioring sysem, many curren underground sysems are deployed by means of wireless communicaion echnology. Ndoh e al. proposed a wireless propagaion model [] o validae ha he realizaion of he wideband wireless LAN in a complex and diffracing rough mine is possible. Some researchers have implemened neworks and opologies based on wireless LAN for underground mine communicaions [3]. Srinivasan e al. proposed a hybrid wireless nework opologies along wih heerogeneous communicaion proocols o suppor high and low bandwidh applicaions [6]. However, he wired basesaions are only deployed in he relaively fixed mine corridors, and he wireless devices in he profile of mine working need high ransmission power o communicae wih basesaions. So he key disadvanage of hese sysems is he high power consumpion and cos of he wireless devices. Recenly, many researchers have invesigaed how o make he coninuous monioring in sensor neworks more energyefficien by exploiing he correlaion in boh ime and space of

2 he readings. Chu e al. proposed he Ken [7] framework which uses a join probabiliy model o reduce he communicaion overhead. The Ken explois spaial correlaions across nodes o improve compression of he daa communicaed o he basesaion wihou imposing undue communicaion burdens o mainain he models. However emporal correlaion canno be fully exploied in he Ken. By monioring a combinaion of individual values and relaion- ships beween neighboring values, CONCH [8] explois spaio-emporal correlaions in daa much more effecively. However, mos of hese mechanisms canno guaranee a bounded error exponen while aiming a minimizing he oal communicaion consumpion. One of our conribuions is o fully exploi he poenial of spaio-emporal correlaion in minimizing he energy cos wih a well-bounded error exponen. III. MINE SAFETY MONITORING USING SENSOR NETWORKS In his secion, we discuss he design of HHMSM, he underground mine safey monioring sysem, which is moivaed by he requiremens of a ypical underground mine surveillance applicaion. A. Hardware To mee he design requiremens of nodes in hierarchical archiecure, we used he EASINET series hardware developed in house [9] o build HHMSM prooype sysem. The wired basesaion, which is composed of he communicaion board EZ10 and he gaeway board EZ511, access he CAN field bus via a CAN adapive module. The EASINET moe includes EZ10 which is compaible wih TinyOS and a sensor board EZ310 which suppors analog gas sensor (LXK-3. The moe uses an 8-bi micro-processor running a 8MHz, and a daa rae of 38.4 Kbps operaing a 433MHz radio. B. Sysem Archiecure We now describe he sysem archiecure, funcionaliy of individual componens and how hey operae ogeher. The main unnels in he underground mine are kilomeers long and well equipped wih wired or fiber cables. Conneced wih main unnel are small branch unnels called drilling unnels. HHMSM is designed o be implemened a he profile of mine working and o be inegraed wih main cables. The heerogeneous iered archiecure is depiced in Fig.1. Fig. 1. HHMSM sysem archiecure and hardware plaform The HHMSM hierarchical archiecure consiss of hree layers. The lower ier is a nework of moe-class wireless sensor nodes ha can be densely deployed. This ier has he lowes compuing capabiliy and significan limiaions on radio bandwidh and baery capaciy. Each moe, which is equipped wih a sensor board ha has mehane and acousic sensors on i, may be aached o a miner, and herefore is mobile in naure. The middle ier forms he backbone of he enire sensor nework. As shown in Fig.1, i is composed of individual basesaions which have significanly more communicaion and power capaciy han hose moes in he lower ier. These basesaions which are deployed in each drilling unnel offload he burden of communicaions from he lower ier. To guaranee he reliabiliy and reusabiliy of his sysem, he basesaions a he mine corridor are linked by a CAN field bus while he basesaions a he profile of mine working communicae wih each oher by radio. The upper ier is he sensor nework s inerface o he operaor. The console of he upper ier can generae exernal alarms o indicae mehane leakage, and show he locaions of all he miners. The profiles of mine working are harsh wireless environmens where conneciviy varies wih ime. Miners wih sensor moes move around while working. Engineering a basesaion placemen o guaranee each sensor node wihin one hop of a base is difficul and cosly. So differen from he exising soluions, we propose o adop muli-hop wireless sensor neworks for communicaions a he middle ier. This srucure can accommodae he mobiliy of miners and he relocaion of he profile of mine working efficienly. If a sensor node finds i can no communicae wih any oher node, i will enhance he ransmission power emporarily in order o deliver daa o basesaion successfully. C. Funcionaliy design and implemenaion C.1 Daa sampling and collecion The key funcion of he HHMSM sysem is o collec he daa of mehane concenraion. The moes equipped wih gas sensor sample he mehane concenraion in he underground mine periodically and ransmi hem o he monioring server. There are a large number of moes in he sysem and hey are randomly disribued in he mine, which will cause large redundan raffic. This raffic may exploi mos of he limied bandwidh, which will inerrup oher applicaions (e.g. miner localizaion. In order o solve his problem, we proposed some measures o reduce he unnecessary raffic a a minimized cos. Specifically, we proposed a disribued overhearing-based adapive daa collecing scheme described in secion IV o reduce raffic significanly. C. Miner localizaion In he underground mine, he basesaions are deployed linearly along he drilling unnel. Based on his characerisic, we proposed a lighweigh scheme ha locaes moes based on he basesaion i is close o. A moe sends is locaion informaion o he miner localizaion server on he ground periodically. The informaion sen o he server consiss of he miner s idenificaion, hops o he basesaion(s i is associaed wih and radio signal inensiy of received packe,

3 which allow he server o learn relaively fine-grained locaion of he miner. As he locaions of he basesaion nodes are known o he server, he locaion of he moe can be pinpoined. TABLE I NOTATIONS Noaion Descripion C(x, The funcion of concenraion diffusion process C ( x, + The observaion of C(x, a he server X The disance of he diffusion process in C(x, The ime of he diffusion process in C(x, D The diffusion coefficien The duraion of sampling resul delivering Δ The offse error of sampling resul a he server The sampling inerval Φ The maximal accepable offse error IV. OVERHEARING-BASED ADAPTIVE DATA COLLECTING SCHEME In he underground mine, he mehane diffusion process follows Fick s second law, as shown in equaion 1. From ha law we can see ha here exiss a srong correlaion in boh ime and space in erms of he concenraion of mehane, and he correlaion changes dynamically along wih he move of sensor nodes and he changing speed of gas concenraion. Based on his characerisic, we proposed an overhearingbased daa collecing scheme via inerpolaion mehod. The gis of he scheme is o selec and acivae sensors o saisfy he desired deecion performance wih he minimum amoun of sensor daa since he number of acivaed sensors is direcly relaed o he energy consumpion of he enire nework. The scheme includes wo pars: a lighweigh overhearing-based daa collecing algorihm which explois daa redundancy and correlaion in space, and an adapive daa sampling rae adjusing mechanism ha explois correlaion in ime. C( x, C( x, = D (1 x A. Lighweigh overhearing-based daa collecing algorihm In our algorihm, he region of he enire sensor nework, which can be reaed as roughly linear along he profile of mine working, is pariioned ino several overlapping zones as shown in Fig.. A zone consiss of a zone leader and all is direc neighbors. The sensor node whose sampling daa value is he highes among is direc neighbors is he zone leader, which is he cener of he zone. Wih he change of sampling values of each sensor, he boundary of a zone and is leader change dynamically. Sensor nodes which are no a leader can be wihin muliple overlapping zones. In our algorihm, a sensor does no ransmi is sampled daa unless he value of is sample is higher han any of is direc neighbor s daa i overheard during he las period, i.e. a sensor will become he leader auomaically if is sampled value is higher han he values i overheard from is neighbors. Unlike general cluser formaion and mainenance algorihms [10], our algorihm, in which each sensor node decide wheher o be a zone leader jus by overhearing he messages derived from is direc neighbors, does no incur any addiional raffic and conrol overhead. This overhearing echnique explois correlaion in space in a very simple and efficien way. Zone Leader Node Ordinary Node Fig.. An overhearing-based nework for HHMSM B. Adapive daa sampling rae adjusing machanism As he nework being pariioned ino zones, he raffic in a given area is a funcion of he updaing frequency of zone, which equals o he sampling inerval of he nodes in ha zone. Therefore, we proposed a disribued algorihm which employs inerpolaion mehod and spaio-emporal correlaion of mehane concenraion iniiaed by Fick s second law o obain an adapive daa sampling rae adjusing sraegy. In his algorihm, each node can calculae he opimized sampling inerval periodically based on he local informaion o minimize he raffic hroughou he nework wihin an accepable offse error. Firs, we define some noaions ha are used in he formulaion as shown in TABLE 1. In mehane monioring, he main cause of server observaion s offse error is he duraion of sampling resul delivering. If a node sends he sampling reading a ime, he server will receive ha observaion a ime +. To reduce he offse error, we add an esimae value o he observaion based on he laes observaions: C C Δ C ( ( ( + = C( + ( Δ Then, he observaion offse error δ can be calculaed hrough Lagrange inerpolaing polynomial as follows: δ = C ( + C( + Δ = C ( ξ1 C ( ξ C ( ξ 3 ( Δ + (3 where ξ1 ( Δ,, ξ (,, ξ3 ( ξ1, ξ. By using equaion 3, we can exend he offse error consrain o he following: δ C ( ξ3 ( Δ + ϕ ϕ Δ (4 C ( ξ 3 and he opimized sampling inerval is: ϕ Δ = (5 C ( ξ3 here, C ( ξ 3 is second-order parial derivaive of concenraion diffusion funcion wih respec o ime. From he equaion 1 we can induce ha each node can calculae he approximae value of C ( ξ 3 periodically jus based on he overheard sampling resuls sen by is adjacen neighbor nodes, i.e. calculae he second-order parial derivaive wih respec o ime using second-order parial derivaion wih respec o space x. This will make he esimae of he second-order parial derivaive wih respec o ime faser and more accurae because he mehane concenraion change

4 in one area is caused by he concenraion gradien a is adjacen areas. And he nodes can hen ge he opimized sampling inerval. Our approach is clearly much simpler han approaches such as Kalman filer which is oo complex o be implemened in a moe. Normally, all values of sampling daa are under he dangerous hreshold, and in each zone only he highes value of sampling daa (i.e. he daa from he zone leader, which can be used o predic he mehane leakage imely, is ransmied o he gaeway in he drilling unnel. However, in case of mehane leakage, more sampling daa beyond he dangerous hreshold will be ransmied o he monioring server direcly. Wih enough sample daa in a dangerous area, he server can apply large deviaion echnique o accuraely predic and locae he mehane concenraion and leakage in a mining zone [11, 1]. V. PERFORMANCE EVALUATION A. Curren Resuls We have experimened wih he EASINET moes for several imes in Dayan Coal Mine, Inner Mongolia province. As shown in Fig.3(a, sensor nodes operaed a differen radio frequencies were esed o obain he communicaion capaciy which helped o design a more feasible and efficien sysem. We have also deployed a 3-basesaion HHMSM prooype sysem in a lab corridor (see Fig. 3(b. which is similar o he drilling unnel in underground mine. B. Evaluaion of Mehane Concenraion Monioring In order o examine he effeciveness of HHMSM sysem in monioring he variaion of mehane concenraion, deecing he mehane leakage imely and sudying he impac of our proposed overhearing-based adapive daa collecing algorihm on he monioring performance, we compared HHMSM sysem wih a reference sysem named ORDINARY in which all he sampling resuls will be ransmied o he server. In our experimens, firsly we invesigae he performance of he wo sysems under he same condiions by changing he densiy of moes, and hen compare he wo sysems under differen condiions. B.1 Change densiy of moes In his monioring experimen, each moe sampled he mehane concenraion every 5 seconds. We varied he number of direc neighbors of each moe from o 11. Fig. 4 shows how he number of sampling packes received by he No.1 basesaion during a 100-second experimen varies wih he moe densiy. I demonsraes ha when he moe densiy increase from o 11, he number of sampling packes ransmied in ONDINARY sysem increases seadily because each moe may send a sampling packe periodically while he number in HHMSM sysem remains almos a a relaive fixed value. These resuls indicaed ha our proposed daa collecing algorihm, resuling from organizing he moes ino zones dynamically, significanly reduces he raffic volume during monioring. The sampling raffic volume in our HHMSM is virually independen from he densiy of moes in he sysem, which leads o much less energy consumpion in he daa ransmission especially in a dense case. (a EASINET moe experimen in Dayan Coal Mine Cluser (b HHMSM prooype sysem deployed in a lab corridor (c HHMSM prooype sysem archiecure Fig. 3. HHMSM prooype sysem experimens in a corridor Fig. 3(c shows he layou of our nework and also depics he opology in he experimens. There are 10 moe-clusers in he nework. Each cluser consiss of he same number of sensor moes. For differen experimens, he densiy of he nework can be varied by changing he number of moes in each cluser. One moe can communicae wih oher moes boh in is own cluser and is direc neighboring clusers. The daa from he lef four clusers will be ransmied o he No.1 basesaion. The daa from he righ four clusers will be ransmied o he No. basesaion. The daa from he middle wo clusers will be ransmied o boh basesaions. During experimens, we manually acuaed a device o release a proper amoun of mehane o change he mehane concenraion, which can be regarded as a mehane leakage. Fig. 4. Impac of nework densiy on ransmission overhead B. Change mehane concenraion In his experimen, each cluser consiss of 3 sensor moes. We varied he mehane concenraion. Our goal is o analyze he monioring accuracy and he corresponding ransmission overhead under differen condiions. Fig. 5 shows he mehane concenraion daa colleced from he No.1 basesaion. When hese values are under normal condiion, he value of he sampling resul in HHMSM sysem is slighly higher han ha value in ORDINARY sysem. When he value of sample resul is above he dangerous hreshold, boh sysems have he same sampling resuls, and

5 he sampling raes of boh sysems were increased adapively. In HHMSM, only he highes value of sampling daa in each zone is ransmied o he gaeway in he drilling unnel. Ye in he ORDINARY sysem all he sample daa will be ransmied. Thus in he HHMSM sysem he sample daa colleced by he monioring applicaion on he ground are higher han ha of he ORDINARY sysem, whose sampling resuls are he average values of all sample daa. However, our goal of monioring mehane concenraion is o predic he mehane leakage, which is only concerned wih hose high sampling daa, so ha he colleced sampling daa normally in HHMSM sysem is enough o predic a mehane leakage. When he sampling daa is beyond he dangerous hreshold, he sampling rae in our HHMSM sysem is increased and all he dangerous sampling daa are ransmied o he monioring applicaion on he ground allowing imely and accurae predicion for he mehane leakage. predic mehane overproof imely wih less bandwidh and energy consumpion han exising wireless soluions. VI. CONCLUSIONS AND FUTURE WORK This paper presens a heerogeneous hierarchical prooype sysem for underground mine safey monioring, namely HHMSM. This sysem can mee muliple requiremens for he applicaions such as monioring he mehane concenraion, and miner localizaion. To realize he hybrid archiecure, we also designed and implemened an overhearing-based adapive daa collecing scheme which explois he redundancy and correlaion of he sampling readings in boh ime and space o reduce raffic volume and conrol overhead wihin a bounded offse error. We have experimened wih HHMSM sysem in corridor environmen wih dozens of sensor moes o evaluae he performance of his prooype sysem. The HHMSM sysem has several advanages comparing o he exising sysems, such as full coverage o underground mine, faul olerance and reliabiliy, energy and bandwidh efficiency, low cos, and mainainabiliy. ACKNOWLEDGEMENTS This work was parially suppored by he Naional Key Basic Research Program of China 973 Projec (grans No. 006CB and NSFC (grans No and No The auhors wish o hank Dong Li and Chunli Hui for heir useful commens and help o his work. REFERENCES Fig. 5. The mehane concenraion daa colleced from he No.1 basesaion Fig. 6. The ransmission overhead from he No.1 basesaion Fig. 6 shows he ransmission overhead during his experimen. The wo sysems have he same ransmission overhead only when he sampling daa is beyond he dangerous hreshold. While for mos par of he experimen, he ransmission overhead of ORDINARY sysem is much larger han ha of he HHMSM sysem. Tha is o say, he HHMSM sysem is more efficien in power consumpion han ORDINARY sysem. The above discussion demonsraes ha he HHMSM sysem which uses differen daa collecing measures under differen condiions can monior mehane concenraion and [1] Major causes of coalmine faaliies in China in 001, China Coal Indusry Yearbook 00, pp. 58, June 003. [] M. Ndoh and G.Y. Delisle, Underground mines wireless propagaion modeling, Proc. of he 60h IEEE Vehicular Technology Conference (VTC, Vol. 5, pp , Sep [3] L. Sydanheimo, M. Keskilammi, and M. Kivikoski, Reliable Mobile Compuing o Underground Mine, Proc. of IEEE Inernaional Conference on Communicaions (ICC, Vol., pp , Jun [4] G. Tolle, J. Polasre, D. Culler, N. Turner, K. Tu, S. Burgess, T. Dawson, P. Buonadonna, D. Gay, and W. Hong, A macroscope in he redwoods, Proc. of he 3rd ACM Conference on Embedded Neworked Sensor Sysems (SenSys, pp , Nov [5] W. Hu, V.N. Tran, N. Bulusu, C.T. Chou, S. Jha, and A. Tayor, The design and evaluaion of a hybrid sensor newowrk for Cane-Toad monioring, Proc. of he 4h inernaional Symposium on informaion Processing in Sensor Neworks (IPSN, April 005. [6] K. Srinivasan, M. Ndoh, and K. Kaluri, Advanced Wireless Neworks for Underground Mine Communicaions, Proc. of IWWCUCA, pp , June 005. [7] D. Chu, A. Deshpande, J.M. Hellersein, and W. Hong, Approximae Daa Collecion in Sensor Neworks using Probabilisic Models, Proc. of he nd Inernaional Conference on Daa Engineering (ICDE, pp , April 006. [8] A. Silbersein, R. Braynard, and J. Yang, Consrain Chaining: On Energy-Efficien Coninuous Monioring in Sensor Neworks, Proc. of ACM Inernaional Conference on Managemen of Daa (SIGMOD, pp , June 006 [9] EASINET produc series. hp:// [10] J.Y. Yu and P.H.J. Chong, A Survey of Clusering Schemes for Mobile Ad Hoc Neworks, IEEE Communicaions Surveys and Tuorials, Vol. 7, No. 1, pp. 3-48, 005. [11] C.Y. Chang, K.P. Shih, and S.C. Lee, ZBP: A Zone-Based Broadcasing Proocol for Wireless Sensor Neworks, Wireless Personal Communicaions, Vol. 33, no. 1, pp , April 005. [1] J.F. Chamberland and V.V. Veeravalli, Design of sensor neworks for deecion applicaions via large-deviaion heory, Proc. of IEEE Informaion Theory Workshop, pp , Oc. 004.