An Optimization of the Chlorine Gas Wireless Sensor Network Coverage and Leakage Model

Transcription

1 Sensors & Transduers, Vol. 54, Issue 7, July 03, pp Sensors & Transduers 03 by IFSA An Optmzaton of the Chlorne Gas Wreless Sensor Network Coverage and Leakage Model Yun Bo Sh, Fang Du, Mengmeng Wang, Rongxn Wang, Debng Xu The Hgher Eduatonal Key Laboratory for Measurng & Control Tehnology and Instrumentaton of Helongang Provne, Harbn Unversty of Sene and Tehnology, No. 5 Xuefu Road, Harbn, 50080, Chna Tel: , fax: E-mal: owboy8877@sna.om Reeved: 0 May 03 /Aepted: 9 July 03 /Publshed: 3 July 03 Abstrat: In ths paper, we fous on buldng a system of Chlorne gas montorng wreless network, whh the overage optmzed by artfal fsh swarm algorthm. Based on ths system, Gaussan leakage model was mproved to estmate the loaton and damage of a leakage adent. The measurement resoluton of the system was ppm. Maxmum measurng error was less than 5 %. The overage of the system was km. And the system power onsumpton redued 40 % as well as the overage of network was not less than 95 % by usng artfal fsh swarm algorthm. The mproved model an aurately foreast the damage and dffuson path n a spef regon. Ths researh provded a method for desgnng wreless sensor network and foreastng leakage damage. Copyrght 03 IFSA. Keywords: Chlorne gas montorng, Wreless sensor network, Artfal fsh swarm algorthm, Coverage, Leakage model.. Introduton Adents of tox gas leakages, whh often ourred at the perod of produton, transport, storage and usng, aused great human damage and property losses [, ]. Chlorne, whh s a manufatured materal of 60 % hemals, s one of the hghest rates of hazardous hemals. If the leakage of Chlorne s more than 5 ppm, the damage wll threaten human lfe. Chlorne gas montorng deves urrently was dvded nto two types, snglehandheld and on-lne deves. Usng sngle-handheld deves, the operators dretly touhed the tox gas and annot ontnuous montor. On-lne montorng deves had weak ant-ammng apablty, omplex layng lnes, poor salablty and hgh ost [3]. Wreless sensor network (WSN) provded a new tehnal support for leakng montor, and t an be the manly method of tox gas leakages montorng [4, 5]. How to mprove deteton effeny, whh s assgned a mnmum of deteton nodes to get a mum overage, s the prmary ssue of WSN desgn [6]. At present, some researhers have appled the hybrd partle swarm optmzaton and the genet algorthm to get the optmal soluton. The average tme of both analyses was too long and the rate of onvergene was slow [7]. The artfal fsh swarm algorthm had fast onvergene rate and an solve real-tme problem [8]. In ths paper, the hardware and software of a WSN system for montorng Chlorne was desgned and expermentally testfed. What s more, we got an optmal soluton between the overage and power 6 Artle number P_5

2 Sensors & Transduers, Vol. 54, Issue 7, July 03, pp onsumpton by artfal fsh swarm algorthm. The desgned system was appled n ndustral feld to buld a leakage model of the small area. Thus, the model provded theoretal bass of adent preventon and resue.. A System of Chlorne Wreless Sensor Network Fg. shows the system was onsttuted by three levels: montorng level, transport level and management level. The montorng level was the seton of perepton and nformaton proessng. It determned the auray of system. The deve was desgned for measurng envronmental parameters, suh as onentraton of Chlorne, temperature and humdty. Consderng of overage and salablty, a W-F ndependent network was bult by a hgh power wreless brdge. The management level was a frendly nterfae of the system whh operators an observe data hanges dretly n PC or Androd ellphone. Fg.. Struture of Chlorne WSN system. The montorng deve was dvded nto sx parts as follows: MCU, sensors, alarm, dsplay, power and wreless ommunaton, as shown n Fg.. The seton n the dashed lne was power. MSP430 was adopted as the mroprogram ontrol unt. Fg.. Struture of the deteton deve. The seton of sensor (n dash-dotted lne) measured gas onentraton of Chlorne, temperature and humdty of envronment, and transformed the sensor output (urrent sgnal) to standard voltage sgnal (-3 V). Measurement of temperature and humdty was n order to ompensate the Chlorne gas sensor. The Chlorne gas sensor was a three eletrodes eletrohemal sensor. Two stages of amplfer transmsson rut were desgned for ths based on sngle power supply [9]. As shown n Fg. 3, the rut adopted.5 V as zero potental. The urrent output of sensor I turned nto postve voltage sgnal V by low pass flter rut. V and zero potental V nput gan adustable operatonal amplfer whh was made up by A, B, C part of LM4. After two stages amplfatons, MCU an read the sgnal V o. R4 s used to adust the gan, and R 9 s used to regulate the zero output voltage. The relatonshp between nput and output s shown as formula (). V o ( R4 R5 R6 R7 ) R.5 ( R8 R9 ), (.5 IR3 ) ' ( R R ) R R R () where o V s the output sgnal and I s the nput sgnal. We adopted W-F as the ommunaton mode whh onsumpton was very hgh (>00 ma) [0]. In ths paper, ths problem was mproved from two aspets. On one hand, the operatng mode of wreless ommunaton was hanged from ontnuous to dsontnuous whh leaded to onsumpton derease 80 %. On the other hand, the system used DS438 to montor the battery state parameters, suh as voltage, urrent, workng tme and remanng apaty and to alarm. What s more, a rut of hargng battery and power supply automat swthng was desgned. External power an harge battery f t swthed on, at the same tme, battery dd not supply the system. The rut s shown as Fg

3 Sensors & Transduers, Vol. 54, Issue 7, July 03, pp Fg. 3. Transmsson rut of Chlorne sensor. LTC 4358 LTC4358 Fg. 4. Crut of hargng and automat swthng. The man funtons of software platform were ontrollng workng node, dsplayng measurement data, queryng hstoral data, loatng leakage poston, buldng adent model and et. 3. Experment 3.. Experment Detals System test was arred out n a seal self-made test box (50L). There was a small fan n the box for aeleratng the dffuson of gas. There was also a pump for exhaustng. The test was ondut aordng to the followng steps. - Gas dstrbuton: 50 ml % Chlorne gas wth Ntrogen gas as balane gas was prepared. - Temperature ompensaton: humdty of 30 % RH n the test box was set. After the deve was eletrally stable (about mnute), neted 0.5 ml (0 ppm) gas nto the test box, then, adusted the temperature from -30 0C to 50 0 C. - Conentraton albraton: Set humdty of 30 % and temperature of 0 0 C. n the test box. After the deve was stable, neted 0. ml ( ppm) gas nto the test box. The measured value and extrated the gas was reorded. we tested 0 tmes from ppm to 0 ppm. 64

4 Sensors & Transduers, Vol. 54, Issue 7, July 03, pp Temperature Compensaton Beause eletrohemal sensor was easly nfluened by temperature, the deve should be ompensaton at dfferent temperature. Nonompensaton test result shows n Table. The mum relatve error was 8 % whh was out of requrement. We normalzed the results measured under dfferent temperature wth the result under 0. We used quadrat polynomal to ft the relatonshp between normalzaton oeffent and temperature, as expresson (). Fg. 5 shows the fttng urve. The sum of squares due to error (SSE) of the fttng was 0.009, oeffent of determnaton ( R ) was , Root mean squared error (RMSE) was y x x , () where x s the temperature and y s the oeffent. The ompensated onentraton was the atual measurng result dvded by normalzaton oeffent y. The mum relatve error after ompensated was desended to 4 % Correton of Gas Conentraton After the temperature ompensaton of the system, we tested the response of Chlorne aordng to experment steps 3. As shown n the result of measurement (see Fg. 6 (a)), the urves of all testes showed the same trend. The response tme of dfferent onentraton testes was all about 30s approxmately. But the reovery tme of hgh onentraton was longer the low one. It was beause the power of pump was onstant and gas neted n the box of hgh onentraton was more than the low one. The reason of urves tter was that the fan n the test box blew the gas (densty of Chlorne s greater than the ar s) result n the gas whh the sensor ontat wth was unstable. When we neted 0. ml ( ppm) and 0. ml (4 ppm) of gas, the result was.5 ppm and 3.6 ppm. The relatve error was 5 %, 0 % whh exeeded the system auray requrement (5 %). We used quadrat polynomal to orret the result. We an get the exat value by as formula (3), and SSE of the fttng was 0.360, R was , RMSE was x.00908x 0.797x , (3) 0 where s x exat value and x s measurement result. The orreted value s shown n Fg. 6 (b). From the Fg. 6 and Table, t s observed that the mum relatve error desended from 5 % to 5 % after the orreton and the measure data was more lose to the real value. Table. Comparson result under dfferent temperature (ppm). Temperature (C) Atual Measurng Relatve Error (%) Compensaton Measurng Relatve Error (%) Fg. 5. Fttng of temperature ompensaton. 65

5 Sensors & Transduers, Vol. 54, Issue 7, July 03, pp a) The result of measurement. b) The result of ompensaton. Fg.6. Comparson orreton result under dfferent Chlorne onentraton. Table. Comparson of orreton results under dfferent Chlorne onentraton (ppm). Theoretal Value Measured Value Relatve Error (%) Correted Value Relatve Error (%) Coverage Optmzaton Based on Artfal Fsh Swarm Algorthm The overage optmzaton of Chlorne WSN was to get the greatest overage usng least sensor nodes n the premse of keepng the network flow. Thus, t redued the energy onsumpton of the system and nreased the workng tme. The prmary target of optmzaton was to keep the overage as large as possble. Eah sensor node had montorng range, and Chlorne onentraton an be measured n ths area probably. Montorng regon A was assumed to be two-dmensonal plane, and dvded nto m n parts whh the area of eah was. There were N same sensor nodes n regon A. The probablty ps (, m ) whh the target mx ( m, ym) was measured by sensor node s( x, y) an be alulated by formula (4)., ds (, m) dd e e e p( s, m) e, d d d( s, m) d d, (4) 0, otherwse where ds (, ms ) dstane between sensor node s and target m, d s theoretally montorng range of sensor, d s relably montorng range of sensor, e 66

6 Sensors & Transduers, Vol. 54, Issue 7, July 03, pp. 6-70,,, are all spef parameters of the sensor node. The ont probablty whh the target was measured by the set of sensor nodes was alulated by formula (5). n p( s, m) ( p( s, m)), (5) all where ps ( all, m) s the ont probablty. The network overage of the system f was alulated by formula (6). f m n x y p( s all m n, m) (6) The other optmzaton target was usng sensor nodes as few as possble. The less sensor nodes there are, the less onsumpton of system there wll be. We defned dormany f to show the onsumpton. If there were totally S montorng ponts, where S0 nodes was workng, the dormany f s shown as formula (7). S S (7) f 0 So, the optmzaton model of Chlorne WSN was establshed, onsderng the overage f and dormany f. In the formula (8), we onsdered that was 0.9 and was 0.. f (8) f f Artfal fsh swarm algorthm s an algorthm that an qukly fnd the optmal soluton by smulatng behavor of a fsh []. A fsh got food by foragng, bunhng, followershp and omparson. The poston of the fsh was x rght now. The status of the fsh was X and the food n the poston was N.. Foragng. The fsh moved to an area for a fxed dstane Step whh the food was more than where t was n the vsble regon. After the movement the status of the fsh was X. If the fsh annot meet the ondton after several seletons, t wll move to a random area. N was the food at the status X. X X ( Step)( x x ) x x X X ( Step) N N N N (9). Bunhng. The number of partners of the fsh was m n the vsble regon, and the entral loaton of them was x. If the food n x was more than x and the rowd densty n x was less than x, the fsh wll move to the partners. After the movement the status of the fsh was X. X X N x x N & N ( Step)( x x ) / m N (0) 3. Followershp. There was a fsh n x where the food was mum n the vsble regon, and the rowd densty n x was less than x. The fsh wll move to the fsh n x. After the movement the status of the fsh was X. X X x x ( Step)( x N N & N N x ) () 4. Comparson. The fsh ompared the food and the rowd densty to hoose an area to move and reorded. If the new seleton was better than the reord, t wll be replaed. Aordng to the food densty, the fsh hose bunhng or followershp and refresh the reord. The optmzaton soluton was obtaned by repeat ths. Usng ths algorthm, we adopt the wreless sensor node n the system as the artfal fsh and set them random n area square area of 00 m 00 m. The number of artfal fsh was 30. The vsble dstane was 0 m, the fxed dsplaement dstane was 3m. When the overage rate reahed 95 %, the average tme of teraton was 63 s operated n MATLAB 009. Coverage was lfted from the random dstrbuton of 63. % to 96.4 %, and the work of sensor nodes was dereased to 3 whh the energy onsumpton redues by 56.7 %. Fg. 7 ompares the ntal random dstrbuton and optmzaton of sensor node overage dstrbuton. 4. Feld Test and Chlorne Leakage Model 4.. Feld Test Detals Aordng to the terran and artfal fsh swarm algorthm, we ftted the system of Chlorne WSN n a waterworks for testng. The waterworks was loated n the suburbs of open area and t overed 5000 square meters. Fg. 8 shows a shemat drawng of waterworks and montorng pont. In the perod of montorng, the leakage of Chlorne happened adentally due to staff s mstakes of onveyng waste. The adent detals are shown as Table 3. 67

7 Sensors & Transduers, Vol. 54, Issue 7, July 03, pp a) Intal status b) Optmzaton Fg. 7. Comparson of Optmzaton result. 4.. Results and Analyss Fg. 8. Shemat drawng of waterworks. Table 3. Adent detals. Spllage 5 kg Leakage rate 50 g/s Heght of leakage pont.5 m Heght of montorng pont.5 m Temperature 7 C Humdty 5 % Wnd dreton(angle wth north wnd) 7 Wnd speed m/s Atmosphere stablty degree B Workers started to resue after 3 mnutes when the adent happened. In the perod of tme, the leakage of Chlorne was wthout manual nterventon. We adopted the montorng data durng ths perod to establsh the model. Fg 9 shows the montorng data of eah test pont. Whh not shown n Fg. 9 s that the onentraton of them s 0 all the tme Analyss of Loaton The measurng range of montorng pont was 0-0 ppm. Although the measurement of hgh onentraton was not prese, we stll an use the data to alarm and analyss of leakage. As shown n Fg. 9, the data of No. 3 node nreased fast, and stayed at mum range stable, the No. 8 node nreased soon afterwards. Aordng to weather ondtons at the tme of the adent, we onsdered that the adent happened n the west of No. 3 nodes. Fg. 9. Montorng data of eah testpont. 68

8 Sensors & Transduers, Vol. 54, Issue 7, July 03, pp Analyss of Leakage Model The expresson of Gaussan model [] s shown as formula (). If H s 0 then we get an expresson as formula (3). C( x, y, z, H ) Q y ( z H ) exp exp u z y y ( z H ), exp z () where Cx, y, z, Hs the onentraton degree of a ertan downwnd pont x, yz, whh the heght of leakage s H. Q s the leakage rate. u s the average wnd speed. x, y, z are respetve oeffent of dffuson on X, Y, Z three dreton. C( x, y,0,0) Q y exp u z y y (3) In the analyss, we onsdered that heght of leakage was 0 approxmately. Aordng to the ondton of leakage, we get the theoretally two dmenson equdensty urve based on Gaussan model as shown n Fg. 0. For onvenene of the analyss, we establshed two-dmensonal oordnate that the adent pont was orgn, the dreton of wnd was the dreton of X axs and the dreton perpendular to downwnd was Y axs. Fg. ompares the theoretally model and mproved model. We put the theoretally urve and ondton of waterworks on the oordnate (see Fg. (a)), and the Chlorne onentraton of A area was greater than or equal to 00 ppm, the B regon was greater than or equal to 80 ppm, the C regon was greater than or equal to 40 ppm, the D regon was greater than or equal to 0 ppm, the E regon was greater than or equal to 8 ppm. The theoretal Chlorne onentraton of No. 5 test pont was 7. ppm, but the measurng data was hanged from 0 to 0 ppm after mnute. For the No. 9 and No. test pont, the theoretal Chlorne onentratons were greater than 40 ppm and 8 ppm respetvely, but the stable measurng data were 6 ppm and 8 ppm ultmately. The reason for devaton between the atual and theoretal was that buldng prevented the gas dffuson. The buldng not only redued the onentraton of Chlorne gas, but also delned the dffuson rate. We onneted the ponts whh measurng onentraton was same, and made up the hgh onentraton measurement problem by Gaussan model. Then we an get a new model that more ft for the real stuaton. Fg. (b) shows the mproved model. If the adent happened, people should esape from the olor area. Fg. 0. Theoretally urve of Gaussan mode. a) Gaussan model b) Improved Gaussan model. Fg.. Model Comparson. 69

9 Sensors & Transduers, Vol. 54, Issue 7, July 03, pp Conlusons Ths paper ompleted the desgn of the wreless montorng network of Chlorne leakage based on W-F. Its haraterst was that wreless selforganzng network was desgned based W-F. Thus, t nreased the ommunaton range and had good salablty. It wll be the dreton of ndustral automaton n future. The mum error of the system was less than 5 %, the overage radus was larger than km. The workng tme of the system was prolonged at the same tme the overage was not dereased by usng artfal fsh-swarm algorthm. Aordng to mproved Gaussan model, we an know the loaton of adent and the evauaton routes after the adent. Ths system an be wdely used n the feld of data aquston and montorng system. Aknowledgements Ths study was supported by Innovaton Fund for Small Tehnology Based Frms of State Commsson of Sene Tehnology of Chna (No. C ), the Key Sene and Tehnology Plan Program of Helongang Provne of Chna (No. ZD07977) and the Program of Innovatve talent of ty of Haerbn (No.00RFXXG009). Referenes []. Gao Z. X., L H. F., Lu J., et al., A smple mroflud hlorne gas sensor based on gas-lqud hemlumnesene of lumnal-hlorne system, Analyta Chma Ata, Vol. 6, Issue, 008, pp []. Ruth Strnger, Paul Johnston, Chlorne and the Envronment: An Overvew of the Chlorne Industry, Kluwer Aadem Publshers, Dordreht, 00. [3]. Jang Lurong, Jang Peng, et al. Desgn of Chlorne Gas Onlne Detet Instrument Based on MSP430, n Proeedngs of the Internatonal Conferene on Embedded Software and Systems Symposa (ICESS 08), Chengdu, Chna, January 008, pp [4]. Chelakara Subramanan, Gabrel Lapll, et al., Expermental and Computatonal Performane Analyss of a Mult-Sensor Wreless Network System for Hurrane Montorng, Sensors & Transduers, Vol. 0, Speal Issue, 0, pp [5]. Ch-Tsun Cheng, Ch K. Tse, et al., A Delay-Aware Data Colleton Network Struture for Wreless Sensor Networks, Sensors Journal, Vol., Issue 3, 0, pp [6]. Steven La, Jannong Cao, et al., TED: Effent typebased omposte event deteton for wreless sensor network, n Proeedngs of the Dstrbuted Computng n Sensor Systems and Workshops (DCOSS), Barelona, Span, 7-9 June 0, pp [7]. Zhong Yuan-Chang, Zhao Zhen-Zhen, et al., Coverage Optmzaton Algorthm In Wreless Sensor Network, Computer Engneerng, Vol. 38, Issue 8, 0, pp [8]. Huang Yu Yue, L Ke Qng, A overage optmzaton of wreless sensor networks based on artfal fsh swarm algorthm, Applaton Researh of Computers, Vol., Issue, 0, pp [9]. Hu Qan, Ge Sbo, et al., Prnple and Applaton of Eletro-hemal Gas Sensor, Instrument Tehnque and Sensor, Vol. 5, Issue 4, 007, pp [0]. Kang Yuemng, Hu Yuhao, et al., Applaton of WI-FI n m n emergeny WSN, Journal of Henan Polytehn Unversty (Natural Sene), Vol. 9, Issue 5, 00, pp []. Wang Yyue, Lao Hongme, et al., Wreless Sensor Network Deployment Usng an Optmzed Artfal Fsh Swarm Algorthm, n Proeedngs of the Internatonal Conferene on Computer Sene and Eletrons Engneerng, Nannng, Chna, 5-7 June 0, pp []. Dehong Zhou, Computer smulaton of the Chlorne leakage based on MATLAB, n Proeedngs of the Computer Sene and Automaton Engneerng (CSAE), Zhangae, Chna, 5-7 May 0, pp Copyrght, Internatonal Frequeny Sensor Assoaton (IFSA). All rghts reserved. ( 70