GPS WIRELESS SENSOR NETWORK FOR MONITORING QUASI-STATIC DISPLACEMENT

Transcription

1 GPS WIRELESS SENSOR NETWORK FOR MONITORING QUASI-STATIC DISPLACEMENT Maayuki SAEKI 1, Mai SAWADA 2, Yukio SHIBA 3 and Kenji OGUNI 4 1 Member of JSCE, Aociate Profeor, Dept. of Civil Eng., Tokyo Univerity of Science (Yamazaki 2641, Noda-hi, Chiba , Japan) aeki@r.noda.tu.ac.jp 2 Member of JSCE, Civil Eng. Re. Int., Technology Center, Taiei Corp. (Nae-cho 344-1, Totuka-ku, Yokohama-hi , Japan) wdmi-00@pub.taiei.co.jp 3 Fellow Member of JSCE, Civil Eng. Re. Int., Technology Center, Taiei Corp. (Nae-cho 344-1, Totuka-ku, Yokohama-hi , Japan) hiba@ce.taiei.co.jp 4 Member of JSCE, Aociate Profeor, Dept. of Sytem Deign Eng., Keio Univerity ( Hiyohi, Minato-ku, Yokohama-hi , Japan) oguni@d.keio.ac.jp Thi paper preent a ytem for monitoring quai-tatic diplacement, named GWSN for GPS wirele enor network. Thi ytem conit of the technologie of a low-cot L1 GPS (Global Poitioning Sytem) and wirele enor network. The low-cot enor node and cable-le ytem enable u to deploy enor in dene number over a vat area. In general, diplacement monitoring uing GPS ha a trade-off relationhip between energy conumption and accuracy of etimation. Meauring continuou data and applying a trend analyi or Kalman filter output a highly accurate diplacement etimation but alo need much energy diipation. To olve the trade-off relationhip, a data compreion method and a tatic GPS poitioning analyi method were developed and implemented in the ytem. In thee method, the aumption of quai-tatic diplacement wa taken into account. A prototype ytem wa developed and experiment were conducted in a quarry ite and in an ideal condition uing the developed prototype. The experimental reult how that the prototype ytem work without eriou ytem trouble in outdoor environment and ha a great potential to meaure diplacement with ub-centimeter accuracy. Key Word : GPS, wirele enor network, quai-tatic diplacement, monitoring 1. INTRODUCTION (1) Background and objective Monitoring diplacement of embankment, lope or reclamation land i important in contruction ite for controlling quality and enuring afety. In recent year, diplacement monitoring ha been automated rather than done through traditional urvey work. In the exiting ytem, a high-performance GPS (Global Poitioning Sytem), a total tation and/or a laer diplacement meter are ued a a diplacement enor. The enor are generally connected to cable and remotely controlled. The exiting ytem i capable of meauring diplacement very accurately. However, it i unrealitic to deploy the enor in dene number over a vat area becaue the cot of intallation and maintenance i quite expenive. Therefore, we have been developing a diplacement monitoring ytem which ue a wirele enor network and an affordable L1 GPS receiver that i connected to a mall patch antenna generally ued for mobile navigation ytem. We call thi ytem GWSN (GPS Wirele Senor Network). Thi inexpenive wirele enor ytem allow u to monitor diplacement with high patial reolution. Application of wirele enor network i one of the hot topic in the reearch field of civil engineering. For example, reearch on an application for meauring vibration of civil infratructure and on an effective data collection method in maintenance work ha been conducted 1), 2). A enor node in a 56

2 wirele enor network conit of phyical enor, a mall wirele communication device, a micro-controller and a battery. Senor node organize a local network autonomouly, collect meaurement and analyze them 3). So the GWSN might be a typical application of a wirele enor network. The general merit of wirele enor network are ummarized a follow: 1) Becaue it i cable-le, the intallation cot can be minimized and there i no poibility of data lo due to cable cut. 2) Since enor node are eparately controlled, malfunction of a enor node doe not influence the whole ytem. 3) When a enor node break down, the ytem i eaily recovered by replacing the broken node with another one. In order to develop a cable-le ytem, a decreae in energy conumption i required, becaue high energy conumption require uage of power upply cable or a large olar battery. Intermittent action i the bet olution for minimizing energy conumption. But it lead to hortage of data and diminihe the accuracy of diplacement monitoring. Therefore, the ytem ha a trade-off relationhip between accuracy and energy conumption. To olve the trade-off relationhip, we have developed a data compreion method and a tatic GPS poitioning analyi method and implemented them in the prototype ytem. We alo conducted experiment in an open quarry ite and in an ideal condition to invetigate the performance of the prototype ytem. (2) Related reearch Diplacement monitoring ytem uing a lowcot L1 GPS receiver have been developed by many reearcher for monitoring landlide or volcanoe. For example, Shimizu et al. developed a diplacement monitoring ytem in which L1 GPS receiver were networked with cable, and applied their ytem to a landlide ite. In their ytem, the trend model wa adopted to improve the accuracy of the meaurement reult and achieve ub-centimeter accuracy 4), 5), 6). Seynat et al. developed a ytem coniting of a central erver and enor equipped with a radio -link, data torage and L1 GPS receiver 7). In thoe reearche, demontration were conducted by ditributing enor everal kilometer away from a bae tation. The ytem preented in thi paper differ from thoe reearche in our ue of the wirele enor network, which enable u to develop a cable-le, low energy conumption, mall in ize and affordable ytem. Data compreion i one of the important tak in a wirele enor network, becaue a low-power and narrow-bandwidth wirele communication device i employed 3). Data compreion ha many merit, uch a hort communication time, low packet-lo rate and low energy conumption. Therefore, there are many reearche on data compreion 8). For example, a tructural ening ytem ha been tudied in which ditributed enor node locally communicate to each other and hare their collected data. After analyzing the collected data uing the proceor mounted on the enor node, the enor node end only the analyzed reult to the central erver 2). In GWSN, the data cannot be analyzed on the enor node becaue the etimation of diplacement need both data of enor node located at a meauring point and data located at a reference point. So all the data received by the enor node hould be tranmitted to the central erver. In the propoed method, the data compreion i achieved by extracting the eential data from raw GPS data. Required data for etimating the relative poition of enor with accuracy of a few centimeter i decribed in the compact data tranmiion tandard 9). The CMR (Compact Meaurement Record) format i commonly ued 10). In the newly developed data compreion method implemented in GWSN, mot of the eential data which i truly needed for etimating quai-tatic diplacement i till extracted from the CMR format. (3) Content The author developed the required elemental technologie, demontrated the performance in cold nowy condition 12), implemented the prototype ytem and conducted an experiment on diplacement monitoring in an open quarry ite. Currently, we are improving the ytem for practical application. Thi paper decribe the developed elemental technologie and how the reult obtained from the experiment conducted in an open quarry ite and in an ideal condition. In the econd ection, an overview of the GWSN and the ytem requirement are explained. The elemental technologie, i.e., diplacement etimation analyi method and data compreion technique, are preented in the third and fourth ection, repectively. In the fifth ection, the reult of the experiment conducted uing the prototype ytem are decribed. Finally, we ummarize the outcome obtained o far and point out the remaining problem to be olved. 57

3 Senor Node Contruction ite Office Uer B Uer A Uer C Wirele communication device Small patch antenna GPS atellite Command Data Fig.1 Schematic view of GWSN. Control PC (data acquiition) Data tranmiion Server (analyi) 2. GPS WIRELESS SENSOR NETWORK (1) Configuration of the ytem Fig.1 illutrate a chematic view of the GWSN. At a target field, the control PC and many enor node are deployed. The prototype of the enor node i hown in Photo 1. The enor node receive raw GPS data, make data packet and end them to the control PC according to the command from the control PC. The control PC tranmit the data et to the Micro-controller Battery Photo 1 Prototype of a enor node. 10 cm central erver through internet or a network having trong communication. Upon receiving the data et, the central erver analyze them and create a databae of etimated diplacement. Uer are allowed to acce the monitoring reult via internet. At preent, the prototype ytem hown in Fig.1 ha been developed and demontrated to work with no eriou problem. In thi prototype, the tar topology i applied in the wirele communication between the control PC and enor node to make it eay to develop a imple yet robut ytem. Recently, we have been trying to implement the tree topology for expanding the area of diplacement monitoring. The enor node, hown in Photo 1, ha a pecified low-power radio module, a micro-controller, a low-cot L1 GPS receiver connected to a mall patch antenna and a mall battery. MU1 (Circuit Deign Co. Ltd.) i ued in the prototype a a wirele communication module. The bandwidth i only 6800bp and i much lower than that of IEEE (Zigbee). However, it ha a relatively long communication range, over 600 meter. (Becaue of the recent growing technology, Zigbee ha alo achieved a long communication range. So we are planning to apply Zigbee to be compatible with our ytem.) A a low-cot L1 GPS receiver, Furuno GT8032 i ued on the prototype. A mall patch antenna which i generally ued for car navigation and never for accurate poitioning i connected to the receiver. The wirele communication module and an L1 GPS receiver are controlled by a micro-controller. In the prototype, PIC16F877A (a middle-range erie of Microchip Technology, Inc.) i ued a the controller. The micro-controller alo control the power upply to the L1 GPS receiver and wirele communication module through the level converter. The level converter output both 3.3V and 5.0V electric power a neceary. For the mall battery, a lithium-ion battery pack or a rechargeable nickel-metal hybrid battery i employed. Thi i commonly ued for a portable game player or a muic playback device. (2) Sequence of operation The operation flow of the enor node i decribed in Fig.2. The control PC broadcat a command every 100 milliecond. Mot of the time the control PC keep broadcating the leep command. When the enor node receive the leep command, it turn all device mounted on the circuit board off and goe into leep mode to minimize energy conumption. The leep mode continue for T (=180) econd. In the leep mode, the enor node run with the internal clock, which depend on the temperature, and the 58

4 Control PC Data tran. command Send jth data Senor Node No.i wake up (initialization) while recv. packet ye check command GPS recv. command if T c < 10 no wait for ye loop end accuracy i very low. The control PC tart to broadcat the countdown command 10 minute before the obervation time. When the enor node receive the command, it check the value of countdown T c. If the value i greater than 10 econd, it goe into leep mode again. In thi cae, the leep time i determined by the following equation. T = max( T c / 2, 10) (1) Since the internal clock of the micro-controller i inaccurate, it i not able to count time correctly. So in thi ytem, the enor node leep only for half of T c to tart data collection at the expected time. At the tart of the obervation time, the enor node turn the L1 GPS receiver on and initialize it. Since the GPS receiver i off mot of the time in thi ytem, it tart in the cold tart mode. In thi cae, the receiver firt ha to earch for the atellite ignal and analyze the accurate GPS time and the approximate poition of the receiver. It take at leat 30 econd and generally need much more time to find out the carrier wave of all available atellite vehicle. In the current ytem, the initialization time i T c 1. GPS data acquiition after 60 econd for initialization. 2. Data compreion. Fig.2 Operation flow of enor node. no leep for T Energy conumption [W] Recv. command Collect GPS data Stand by Send data Sleep Accumulation time [ec] Fig.3 Temporal variation in energy conumption 16). et to be 60 econd and the GPS data received during the initialization i thrown away. After the initialization, the micro-controller receive data et from the GPS receiver and ave them in it memory in compreed data format. After the cheduled obervation period, the enor node goe into leep mode again. After obervation, the control PC broadcat the data tranmiion command, which requet the i-th enor node to end the j-th epoch data back to the control PC. If the control PC receive the j-th epoch data, it add one to the epoch number j+1 and broadcat the ame command again. After getting the total data from a enor node, the control PC add one to the ID number of the enor node. If the control PC cannot get the total data from the enor node in a certain period of time, it quit the enor node and trie to the next one. Thi communication protocol i able to gather all data et if no eriou communication problem occur. The control PC tranmit the data et to the central erver through the internet after gathering all data from enor node. The central erver etimate diplacement of enor node and create the databae. (3) Energy conumption of the enor node Fig.3 illutrate the temporal variation in energy conumption meaured uing the prototype of a enor node. A hown in the figure, the enor node need the larget electric current when it receive GPS data, which current i about 400mW. The next larget value i about 220mW, which happen when the wirele communication module end data to the control PC. Thee fact indicate that it i very important to decreae the work duration of the GPS and wirele module. If the GPS receiver i kept on, the lithium-ion battery hown in Photo 1 (12.48Wh) lat about 30 hour. 59

5 (4) Requirement of the ytem The GWSN i deigned to monitor quai-tatic diplacement with high patial reolution over an area of one kilometer quared. Thi retriction give ome important contraint to the development of thi ytem. The following contraint are taken into account: 1) The baeline between meauring and reference point i horter than that of the uual GPS relative poitioning. In thi cae, the effect of atmopheric delay and ionopheric delay are conidered to be negligible in the analyi of tatic GPS poitioning. 2) The GPS obervation period hould be retricted to a hort one to ave battery energy for long-term operation. 3) Becaue of the monitoring of quai-tatic diplacement, the daily change are conidered to be very mall, for example a few millimeter a day. (5) Component technologie to be developed The above condition 2) require a component technology which enable u to etimate the diplacement accurately uing data of hort length. In general, it eem effective for improving accuracy to analyze continuou data with Kalman filter or trend analyi to uppre noie influence. However, a longer period of data acquiition need much more energy to run the GPS receiver. To limit energy conumption, a hort obervation period i a neceary contraint. Therefore, an analyi method which i able to etimate the diplacement accurately with a hort data length i required. To minimize energy conumption, it i alo important to compre the data ize of wirele communication. Thu a data compreion method i alo required. From the above dicuion, we need to develop the following component technologie: 1) A diplacement analyi method which etimate the relative poition of enor accurately with hort data length. 2) A data compreion method for raw GPS data. 3. DISPLACEMENT ANALYSIS METHOD (1) Obervation equation and noie behavior The phae of the carrier wave omitted from a atellite vehicle k and received at GPS receiver i at time t i decribed a φ k i (t) in thi paper. Since the noie due to the tropophere and ionophere are negligible in the relative poitioning analyi, the Double-Difference (DD) obervation equation can be modeled like the following equation 13) : φ ( t) = ρ ( x, t) + λn + Δ + ε ( t) (2) where repreent the DD value of the GPS atellite k, l and the enor node i, j: k i k j l l ( ) = (3) The term ρ ( x, t) i the DD range between atellite and the node, x i the relative poition vector of a enor node, λ i the wavelength of the L1 carrier wave, N i the DD integer ambiguity, Δ i the DD antenna noie and ε (t) i the random noie. Becaue of the quai-tatic diplacement condition and the 20-centimeter wave-length of L1 carrier wave, once the initial relative poition i uccefully determined, the DD integer ambiguity on the right-hand ide of equation (2) can be eaily reolved. Therefore, in the following dicuion, we conider the DD integer ambiguity to be involved in the DD carrier phae φ (t). The antenna noie Δ, which i caued by the multi-path, behave almot in the ame manner a when the urrounding condition and the geometry of atellite and a enor are the ame. In addition, the GPS atellite go around almot the ame orbit every 23h56m. So if we oberve carrier phae every 23h56m, the DD carrier phae include almot the ame antenna noie. Thi fact indicate that the antenna noie can be canceled. Therefore, in the following dicuion, we conider that the antenna noie Δ are included in the random noie ε (t). Here, we expre the initial poition of enor node x 0 and diplacement Δx. The poition of the enor node can be expreed a i j x = x 0 + Δx (4) Subtituting equation (4) into equation (2) and applying Taylor expanion with repect to x 0 yield the following linearized equation. φ t) ρ ( x, t) = ρ ( x, t) Δx + ε ( ) (5) ( 0 0 t Gathering equation (5) for available atellite yield imultaneou equation. The diplacement vector Δx can be etimated by applying the leat-quare method to the imultaneou equation. (2) Linear approximation of the corrected DD carrier phae It i known that the term ρ ( x 0, t) on the right- 60

6 hand ide of equation (5) can be approximated a a linear function of time. In addition, in an actual ituation, the amount of diplacement Δx i at mot one meter and the diplacement are contant in a hort period of time. Therefore, the left-hand ide of equation (5) can be expreed a a linear function of time with a very mall tangent 14). u ( t) = φ ( t) ρ ( x 0, t) (6) In the propoed method, the diplacement Δx i not etimated by olving equation (5) directly. Firt, the approximate linear function, u ( t) = at + b, i etimated for all DD obervation equation, and the value of u (t) correponding to the firt and lat epoch are calculated 15). Gathering the equation calculated for available atellite, we obtain the following imultaneou equation: U ( t) = A( t) Δx + e( t) (7) where U (t) i the obervation vector whoe component are u (t) ; A (t) i the deign matrix which depend only on the atellite contellation and enor poition; and e (t) i the noie vector. Solving equation (7) correponding to the firt and lat epoch through the leat-quare method yield the following olution: Δ x = G G = A 1 T T 1 ( A ( t ) R ( t ) U( t ) T 1 + A ( t ) R ( t ) U( t (8) ( t e e 1 ) R ( t ) A( t ) T 1 + A ( t ) R ( t ) A( t ) (9) e where t and t e repreent the time of the firt and lat epoch, repectively. R(t) i the variance-covariance matrix of the obervation vector. The accuracy of diplacement etimated by the above method i relatively improved from the conventional method, epecially in cae of hort data length 15). (3) Segmentation of obervation period The accuracy of diplacement etimated by olving equation (8) depend on the regularity of coefficient matrix G and the noie included in the obervation vector U(t). Therefore, in order to improve the accuracy, there i no choice except to reduce the noie from U(t) or improve the regularity of coefficient matrix G. In order to reduce the amount of noie, a high-performance antenna and receiver are needed, which are alo expenive. On the other hand, e e e )) the improvement of regularity of coefficient matrix G can be achieved by electing a favorable obervation chedule. The coefficient matrix G i determined by the geometry of the enor node and atellite contellation. If the condition number of G i mall enough, the accuracy deterioration due to noie can be minimized. So, in GWSN, we divide the obervation time into egment o that the atellite contellation correponding to each egment become a different a poible 16). 4. DATA COMPRESSION In the prototype ytem, we ue the wirele communication device with very narrow bandwidth. The data tranmiion time T [m] of n byte data i etimated by the following equation 17) : T = ( n + 15) + ( n 4) + 34 (10) br where br i the baud rate of the erial communication between the micro-controller and wirele communication device and i et at 9800 bp in the prototype. The GPS receiver mounted on the prototype of the enor node output raw GPS data at 266 byte per econd in Furuno Binary Format. If the data i logged for 4 minute, the total amount of data i about 63.8 kbyte. To end the whole data to the control PC, it might need 98.7 econd, which can be etimated uing equation (10). In the actual communication, ome command a well a the data itelf need to be ent. So the total time of communication hould be longer. Thi reult in further energy conumption and horten the battery lifetime. And alo, if there are many enor node and the data i collected through multi-hop communication, a long period of time i needed for data collection. Thi lead to the limitation of the number of meauring point. Thee two fact motivate u to develop a new data compreion method. In the data compreion method preented in thi paper, the eential data, which i truly needed for etimating quai-tatic diplacement without accuracy deterioration, i extracted from the raw data. The amount of data of one epoch i compreed from 266 byte to 28 byte. Fig.4 how the format of the compreed data. The packet conit of the tart flag (2byte), acquiition time (4byte), atellite ID (1byte), carrier phae (1byte) and check um (2 byte). The number of the atellite ID and the correponding carrier phae i 61

7 check um (2 byte) acquiition time (4 byte) ployment in the open quarry ite and the intalled enor node, repectively. The exiting GPS diplacement monitoring ytem 18) had already been intalled at the ite. The exiting ytem wa upplied with electric power uing cable. The wirele enor node, developed by the author, were fixed next to the exiting enor to compare the reult. The ymbol K-1 in Photo 2 repreent the reference point, and G-1 to G-7 the meauring point, whoe relative poition from K-1 were monitored. In the data collection, raw GPS data wa logged for 5 minute at a 1 Hz ampling rate 4 time a day. Every enor node had a nickel-hydride battery (22.8Wh). Conidering thi meaurement chedule and the energy conumption of the enor node, it wa predicted that the enor node could run for about 96 day without battery change. The control PC wa et at the bae tation in which the commercial AC power upply wa available. The bae tation wa viible from every meauring point. The ditance in a traight line between the bae tatart flag (2byte) data (2*10 byte) carrier phae (1byte) + atellite ID (1byte) = 2 byte Fig.4 Data format of the preented compreion method 11). fixed at 10 becaue a GPS receiver rarely oberve more than 10 atellite at once. The CMR format contain other data than that hown in Fig.4, uch a code peudo-range and approximation poition. The mot characteritic part of thi data compreion method i compreion of the carrier phae to only 1 byte. The carrier phae i generally treated a double-preciion data. The GPS receiver ued in the prototype output the carrier phae a 4-byte data. The three mot ignificant byte repreent the integer part of the carrier phae and the remaining 1 byte the fraction part. We extract only 1 byte correponding to the decimal fraction. A mentioned in ection 5, our target i quai-tatic diplacement. In thi cae, we can eaily reolve the DD integer ambiguity. Thi mean that the integer part of the carrier phae i not necearily needed. Then, we can uccefully compre the amount of data without diminihing the accuracy of diplacement etimation 11). 5. PERFORMANCE ASSESSMENT Experiment were conducted both in an open quarry ite and in an ideal condition to invetigate the performance of the prototype of the GWSN. The objective of the experiment in the open quarry ite were mainly focued on confirming that the ytem worked without eriou ytem trouble and finding bug hidden in the ytem. The experiment in the ideal condition wa carried out to ae the accuracy of diplacement monitoring. Photo 2 Aerial view of open quarry ite. Senor node ground plane GPS antenna Thermometer Battery Photo 3 Senor node intalled at the ite. (1) Demontration in an open quarry ite The experiment wa conducted in the open quarry ite (Ohhu-hi, Iwate-prefecture) over 4 month from July to November The objective were 1. to confirm that the ytem worked without any eriou ytem trouble; 2. to check the performance of the wirele communication (ditance and tranmiion rate); 3. to compare the reultant diplacement between the exiting ytem and the preented one; 4. to find out problem inherent in the ytem. Photo 2 and 3 how an aerial view of enor de- 62

8 09/7/15,3:0 09/7/15,13:0 09/7/15,15:30 09/7/15,23:30 SA22 SA12 SA32 SA17 SA18 SA28SA14 SA26 SA4 SA11 SA2 SA31 SA17 SA27 SA21 SA30 SA9 SA8 SA22 SA28 SA24 SA18 SA20 SA12 SA20 SA15 SA12 SA26 SA32SA23 SA13 SA29 SA28 SA14 SA32 SA9 SA27 SA19 SA11 SA10 Fig.5 Satellite contellation in the obervation period. SA3 tion and the enor node G-7, which correponded to the longet bae line, wa about 504 meter. On the other hand, the hortet bae line wa 45 meter, between the bae tation and G-6. The obervation chedule wa determined o that the atellite contellation uniformly covered the ite. The atellite contellation of thi experiment i hown in Fig.5. The circle repreent the elevation angle. The center of the circle correpond to a 90-degree elevation angle (the right overhead direction) and the larget circle repreent 0 degree (the horizon). The upward direction in Fig.5 correpond to the northern direction. In thi figure, there are no GPS atellite in the northern direction. Thi i becaue the orbit of the GPS atellite i deigned to cover low latitude. The arrow repreent the orbit of the atellite in the duration of obervation. The arrow are colored according to the egment. In thi experiment, the atellite flying at the low elevation angle in the northern and wetern direction were unavailable due to the lope of the quarry. a) Running tate and battery lifetime The lifetime of the enor node are lited in Table 1. The obervation were carried out two time uing fully charged batterie. The firt experiment wa conducted from 23 July to 13 October 2009 and the econd from 13 October to 26 November The end date in Table 1 repreent the final date of the data file aved on the central erver. A hown in Table 1, mot of the enor node ran for more than 1 month, but the lifetime were very different from each other. The average wa about 50 day, which i almot half of the prediction (96 day). And alo, the lifetime in the econd experiment became horter than that in the firt one. Table 1 Working day of enor node in the quarry ite. The firt trial (7/23-) The econd trial (10/13-) ID End date Day End date Day K-1 9/ /21 40 G-1 10/ /26 45 G-2 9/ /21 40 G-3 9/ /31 19 G-4 9/ /21 40 G-5 9/ /26 45 G-6 8/ /17 36 G-7 8/ /23 42 Table 2 Data tranmiion rate in the firt experiment. Senor ID Ditance [m] Number of trial Number of data Rate [%] K G G G G G G G The main caue of the lifetime hortage i conidered to be the high humidity condition. In thi experiment, a thermohygrometer wa placed in the box of the enor node, and the temperature and humidity were ampled every 10 minute. According to the meaurement data, the humidity wa over 90% mot of the time. So the high humidity condition may have moitened the circuit board of the enor node and decreaed the electric reitance. Thi may have caued the hortage of the battery lifetime. In order to examine the above hypothei, we conducted a imple experiment in our laboratory uing the ame enor node, which had been ufficiently dried after the experiment at the open quarry ite. In thi experiment, two enor node were placed in boxe in dry condition and the other two into other boxe in wet condition. The control PC collected GPS data from the enor node according to the ame chedule a in the quarry ite. Reult how that the two enor node in dry condition of le than 10% humidity worked for 106 and 110 day, repectively. On the other hand, the two enor node in wet condition of 100% humidity worked 63

9 only for 46 and 59 day. Thi ugget that a high humidity condition horten the battery lifetime. b) Wirele communication ditance and data tranmiion rate Table 2 preent the data tranmiion rate of each enor node in the firt experiment. The number of trial can be approximately calculated by multiplying the working day by four, ince raw GPS data wa collected 4 time a day. The number of data in Table 2 i the number of data file aved in the control PC. The rate repreent the ratio of the number of aved data file to trial. In thi experiment, the data et were almot 100% tranmitted to the control PC. In the prototype ytem, the control PC i deigned to keep ending data tranmiion command to the enor node up to time out. Therefore, once the enor node receive a packet from the control PC, it can tranmit the whole data et of a egment to the control PC. On the other hand, if the enor node doe not receive a packet until time out, the control PC doe not collect the correponding data et. Thi i one of the iue to be improved in future work. c) Comparion with the exiting diplacement monitoring ytem In thi experiment, we fixed our enor node next to the exiting GPS enor to compare the reult of diplacement monitoring. Fig 6. preent the time erie of diplacement meaured at the enor node G-1. The figure correpond to the NS (North-South), EW (Eat-Wet) and UD (Up-Down) component from top to bottom, repectively. The horizontal and vertical axe repreent the accumulation time [day] and diplacement [cm]. The dot and broken line repreent the reult of the exiting ytem. The exiting ytem output diplacement every hour (dot) and etimate the moothed curve by applying trend analyi (broken line). The triangle and olid line are the reult obtained from the preented ytem. The triangle are plotted with a 1-day interval ince the diplacement of the preented ytem i etimated uing 4 egment together, while the olid line repreent their moving average over 5 day. In the graph, there are miing data for the exiting ytem from 6 to 16 day. In thi period, the diplacement of all meauring point were not monitored due to a power upply fault. Comparion of the reult of the exiting and the preented ytem indicate that the trend of etimated diplacement are imilar to each other. So the preented ytem eem to work well in thi experiment. However, ome mall dicrepancie appear in the NS and UD direction. For example, the data for the 11th day in the UD direction include a relatively large etimation error of about 1.5 cm. The main Dip. in NS [cm] Dip. in EW [cm] Dip. in UD [cm] Fig. 6 Comparion between the diplacement monitored by the exiting and the preented ytem at the enor G-1. caue of thi error i the antenna noie. In the preent ytem, a mall patch antenna i ued to minimize the cot of the enor node. However, uch a mall patch antenna i generally never ued for diplacement monitoring becaue it i very weak againt the multi-path noie. It i known that the multi-path noie behave almot the ame way every day becaue the geometrical condition between GPS atellite and a enor i almot the ame every day too. Therefore, in the analyi, the multi-path noie wa firt etimated uing the data collected on the firt day and the multi-path noie included in the following meaurement were ubtracted uing the etimated noie. However, the multi-path noie included in the raw data ometime behaved in a very different manner. That i why the multi-path noie wa unuccefully canceled and the etimated diplacement include a relatively large error. d) Summary of demontration in the quarry ite We conducted an experiment in the open quarry ite and the reult how that the preented ytem run without eriou ytem trouble and i able to monitor diplacement. On the other hand, thi experiment clarifie ome problem to be olved: 1) The need to develop a box for protecting the enor node from a high humidity environment and to verify the battery lifetime. 2) The need to implement a more reliable data tranmiion method. 64

10 3) The need to develop an etimation method of multi-path noie and invetigate the term of validity of the etimated noie. (2) Demontration in an ideal condition In the experiment conducted in the open quarry ite, we could not ae the accuracy of diplacement monitored by the preented ytem becaue nobody knew the actual diplacement. Therefore, we carried out an experiment in the well-known condition, which mean that we intentionally moved the GPS antenna uing an accurate movable tage which i able to control 3D diplacement with an accuracy of a few micrometer. a) Outline of experiment In thi experiment, one of the GPS antenna wa fixed on top of the movable tage, a hown in Photo 4. The other antenna wa placed on a platic box a a reference point. Both antenna were attached to the ground plane to uppre multi-path noie. Thee two antenna were fixed at different height o that the data might include the different multi-path noie. The experiment wa carried out in an ideal condition, where there were no obtacle over the ite. The obervation period wa about 40 day from 26 December 2009 to 3 February In the experiment, the movable tage kept it poition during the firt 2 day. After that, it vertically went down by 1 cm in 10 day at the contant peed. In the next 10 day, it moved to the northern direction by 1cm and after that to the wetern direction by 1cm in the following 10 day. After the 32 day mentioned above, the movable tage tood till for 9 day. The raw GPS data (carrier phae) wa oberved for 5 minute with a 1 Hz ampling rate every hour. However, the data acquiition tarted 4 minute earlier every day to equalize the atellite contellation. In the experiment, 23 egment were collected every day. b) Analyi method The diplacement were etimated by picking out ome egment from the 23 and analyzing them together, a mentioned in ection 3.(3). Fig.7 how the diplacement time erie etimated uing all egment. The graph repreent the diplacement in the NS, EW and UD direction and the total number of atellite ued in the analyi from top to bottom, repectively. The olid and broken line repreent the analyzed reult and the actual diplacement controlled by the movable tage, repectively. The analyzed reult are miing on the third day becaue the correponding raw GPS data wa not collected due to wirele communication trouble. In the cae of analyzing 23 egment together, the 3-directional diplacement wa etimated accurately and the doubled tandard deviation from the actual controller of movable tage reference antenna movable tage and antenna Photo 4 Experimental etup uing the movable tage in an ideal condition. Dip. NS [cm] Dip. EW [cm] Dip. UD [cm] Total number of ued atellite Fig.7 Time erie of diplacement etimated uing 23 egment. diplacement wa about 2.8 and 2.2 [mm] in horizontal and vertical direction, repectively. Thi reult indicate that the preented ytem ha a potential ability of monitoring ground diplacement with ub-centimeter accuracy. However, in order to decreae the energy conumption and make the battery 65

11 lifetime longer, the obervation period in a day needed to be decreaed. So, in the following ubection, how to decreae the ue of egment without diminihing the accuracy i dicued. c) Effect of egment election on accuracy We invetigated the relationhip between the choice of egment and the accuracy of diplacement by electing 1 to 3 egment from the 23. Fig.8 how the reult for the cae of analyzing only 1 egment. The vertical axe of the graph repreent the accuracy (doubled tandard deviation), number of ued atellite, PDOP (Poition Dilution of Preciion) and condition number of the coefficient matrix of the normal equation, repectively. The horizontal axi repreent the egment ID number. In the analyi, a time erie of 3-directional diplacement wa calculated for each ingle egment. The accuracy of each egment wa evaluated by calculating the doubled tandard deviation from the actual diplacement. The number of atellite, PDOP and condition number were obtained by averaging their etimated value over 40 day. A hown in Fig.8, if the PDOP or condition number i relatively mall, the accuracy tend to be low. The accuracy in the cae of uing egment #3, #5, #9 and #12 i better than for the other. Since the value of PDOP can be predicted without collecting raw GPS data on-ite, we can poibly make a reaonable obervation chedule with the PDOP. However, the accuracy i only cm even if the bet egment i elected in the cae of uing a ingle egment. Therefore more egment mut be analyzed together to improve the accuracy. The time erie of diplacement etimated uing egment #5 i decribed in Fig.9 for reference. The etimated diplacement i relatively inaccurate compared with the reult when uing 23 egment, a hown in Fig.7. At the next tep, we conidered the cae of analyzing two egment together. In thi analyi, egment #5 wa alway ued and the other egment wa elected from the other 22 egment. The relationhip between the choice of egment and the accuracy in the cae of analyzing two egment i decribed in Fig.10. The accuracy tend to be good in the cae of a mall PDOP, but the correlation i not very clear. A maller value of PDOP or condition number doe not necearily yield better accuracy. For example, the combination of egment #5 and #11 yield a relatively large PDOP and condition number but give good accuracy. The main caue i conidered to be the influence of obervation noie. The accuracy depend on the obervation noie a well a the quality of the coefficient matrix. A hown in Fig.8 and Fig.10, the accuracie when uing egment #4, #7, #10, #15 and following are accuracy [cm] Number of atellite PDOP [cm] Condition number Fig.8 Relationhip between the choice of egment and the accuracy in the cae of analyzing a ingle egment. Dip. NS [cm] Dip. EW [cm] Dip. UD [cm] Fig.9 Time erie of diplacement etimated uing egment #5. relatively wore than the other. Thi fact how that thee egment do not improve the accuracy even if more than two egment are ued. Fig.11 decribe the time erie of diplacement etimated uing egment #5 and #8. Thi combination give the mallet PDOP value. In thi cae, the doubled tandard deviation i about [cm], which i an improvement from the reult of a ingle egment. In order to improve the accuracy further, three- 66

12 accuracy [cm] Number of atellite PDOP [cm] Condition number Fig.10 Relationhip between the choice of egment and the accuracy in the cae of analyzing two egment. Dip. NS [cm] Dip. EW [cm] Dip. UD [cm] Fig.11 Time erie of diplacement etimated uing egment #5 and #8. egment combination were teted. In thi analyi, egment #5 and #8 were alway ued and the other egment wa added from the other 21. The reult are hown in Fig.12. In thi analyi, the accuracy of all combination eem to be improved due to the additional egment. Therefore it can be aid that the analyi with multiple egment i able to prevent eriou accuracy deterioration. The addition of egment #2, #9, #15 or #17 epecially improve the accuracy. However, the PDOP and condition number accuracy [cm] Number of atellite PDOP [cm] Condition number Fig.12 Relationhip between the choice of egment and the accuracy in the cae of analyzing three egment Fig.13 Time erie of diplacement etimated uing egment #5, #8 and #9. Dip. NS [cm] Dip. EW [cm] Dip. UD [cm] of the reult of #15 and #17 are not a mall, o that it i difficult to predict the accuracy improvement from before the analyi. Fig.13 decribe the time erie of diplacement in the cae of analyzing egment #5, #8 and #9 together. Comparing the reult hown in Fig.9, Fig.11 and Fig.13 how that the behavior of the etimated diplacement i gradually getting cloe to the actual diplacement. 67

13 d) Summary of demontration in an ideal condition We conducted an experiment in an ideal condition to evaluate the accuracy of the ytem preented in thi paper. In the experiment, one GPS antenna wa fixed on top of a movable tage, which wa able to control the diplacement with an accuracy of a few micrometer. We alo teted whether different egment combination improved the accuracy. The reult of the experiment indicate the following: 1) The analyi uing 23 egment ha the potential of monitoring diplacement with ub-centimeter accuracy in an ideal condition. 2) The accuracy in the cae of uing three egment can be comparable to the cae of uing 23 egment if the bet combination i elected. 3) The index of the PDOP or condition number wa helpful in electing the bet egment when we analyzed diplacement with a ingle egment. But it i difficult to predict the bet egment combination uing thoe indexe. Thi fact implie that the accuracy depend on the obervation noie rather than the quality of the coefficient matrix in the cae of multiple egment uage. 6. SUMMARY The author have developed a diplacement monitoring ytem uing the technologie of a low-cot L1 GPS and wirele enor network. The ue of thee technologie enable reduction of the total cot of diplacement monitoring becaue of the low-cot enor node and the workability of the ytem. In thi reearch, we developed a prototype ytem and demontrated it in an open quarry ite. The reult of thi experiment indicate that the ytem run without eriou ytem trouble in an outdoor environment. Area for future work were found. We alo teted the accuracy of diplacement by conducting an experiment in an ideal condition uing a movable tage. The reult how that the accuracy can be improved to the ub-centimeter level. The future work are to implement protection againt high humidity, to recontruct the network protocol of collecting data and to improve the diplacement analyi method, uch a the noie etimation method and the choice of the bet egment combination. REFERENCES 1) Lynch, J. P.: Decentralization of wirele monitoring and control technologie for mart civil tructure, Ph.D. thei, Department of Civil and Environmental Engineering, Stanford Univerity, Stanford, CA., ) Nagayama, T., Spencer, Jr. B. F. and Fujino, Y.: Smart enor middleware development for dene tructural vibration meaurement, Journal of JSCE A, Vol. 65, No. 2, pp , (in Japanee) 3) Sakata, S.: Ubiquitou Technology, Senor Network, Ohmha, (in Japanee) 4) Kondo, H., Cannon, E., Shimizu, N. and Nakagawa, K.: Development of a ground diplacement monitoring ytem by uing the global poitioning ytem, Journal of JSCE, No.546/VI-32, pp , (in Japanee) 5) Matuda, H., Adachi, H., Nihimura, Y. and Shimizu, N.: Applicability of the trend model for moothing meaured diplacement by uing global poitioning ytem and method for predicting diplacement behavior, Journal of JSCE, No.715/III-60, pp , (in Japanee) 6) Shimizu, N.: Continuou Diplacement Monitoring uing Global Poitioning Sytem for Aement of Slope Stability, the 2nd Southeat Aian Workhop on Rock Engineering, Hanoi, pp , ) Seynat, C., Hooper, G., Robert, C. and Rizo, C.: Low-cot deformation meaurement ytem for volcano monitoring, Proceeding of the 2004 International Sympoium on GNSS/GPS, ) Sadler, C. M. and Martonoi, M.: Data Compreion Algorithm for Energy-Contrained Device in Delay Tolerant Network, Proc., the ACM Conference on Embedded Networked Senor Sytem (SenSy'06), November 1-3, ) Talbot, N. C.: Compact data tranmiion tandard for high-preciion GPS, 9th Int. Tech. Meeting of the Sat. Div. of the U.S. Int. of Navigation, Kana City, Miouri, 17-20, Sept., pp , ) Heo, Y., Yan, T., Lim, S. and Rizo, C.: International Standard GNSS Real-Time Data Format and Protocol, Proceeding of IGNSS Sympoium, Holiday Inn Surfer Paradie, Autralia, Dec., ) Koaka, T. and Saeki, M.: Data reduction method for L1 GPS receiver of wirele enor network ytem, Journal of Applied Mechanic, Vol.9, JSCE, pp , (in Japanee) 12) Sawada, M., Shiba, Y. and Saeki, M.: Development of diplacement monitoring ytem and GPS wirele enor, Report of Taiei Technology Center, Vol.43, (in Japanee) 13) Hofmann-Wellenhof, B., Lichtenegger, H. and Collin, J.: GPS, Theory and Practice, SpringerWienNewYork, ) Ganer, G., Wieer, A. and Bruvver, F.: GPS Software development for monitoring of landlide, Proc., FIG XXII Congre (CD-ROM), ) Saeki, M., Kaneko, S. and Inoue, T.: Development of GPS poitioning algorithm baed on tatic and dene deployment condition, Journal of Applied Mechanic, Vol.10, JSCE, pp , (in Japanee) 16) Saeki, M., Inoue, T., Sawada, M. and Shiba, Y.: Fundamental tudy on the accuracy of diplacement monitoring of GPS wirele enor network, Journal of Applied Mechanic, Vol.12, JSCE, pp , (in Japanee) 17) MU-1 manual (Webite of circuit deign INC.), 0.pdf (in Japanee) 18) Webite of hamen-net.com, hamen01.html (Received September 18, 2012) 68