This article has been accepted for inclusion in a future issue of this journal. Content is final as presented, with the exception of pagination. IEEE TRANSACTIONS ON INSTRUMENTATION AND MEASUREMENT

1

Design of a Multifunctional Wireless Sensor for In-Situ Monitoring of Debris Flows Huang-Chen Lee, Student Member, IEEE, Amit Banerjee, Yao-Min Fang, Bing-Jean Lee, and Chung-Ta King, Senior Member, IEEE

Abstract—Debris flows carrying saturated solid materials in water flowing downslopes often cause severe damage to the lives and properties in their path. Close monitoring and early warning are imperative to save lives and reduce damage. Current debris-flow-monitoring systems usually install sensor equipment along the riverbanks and mountain slopes to detect debris flows and track their data. Unfortunately, most of this equipment indirectly collects data only from a distance. So far, there is no way to understand what is happening inside a debris flow and to collect its internal parameters, not to mention doing this in real time. To answer this challenge, this paper presents a novel multifunctional wireless sensor for monitoring debris flows. The core idea is to let these sensors drift with the debris flow, to collect flow information as they move along, and to wirelessly transmit the collected data to base stations in real time. The design of such a sensor needs to address many challenging issues, including cost, deployment efforts, long-term standby, and fast reaction. This paper addresses these issues and reports our evaluation results. Index Terms—Debris flow, disaster monitoring, energy saving, radio communication, wireless sensors.

I. I NTRODUCTION

T

AIWAN is located on the collision boundary of the Philippine sea plate and the Eurasian plate. The mountain terrain is precipitous, and fragile rocks and frequent seismic activities characterize the region. Such land conditions were further deteriorated after excessive developments in the mountain areas and the devastating 921 earthquake in 1999. The land collapses after an earthquake, and many crevices are formed in the rocks and soil. Any concentrated torrential rainfalls, e.g., those brought by typhoons, cause the groundwater level to increase and the surface runoff to concentrate. As rock debris is saturated with water, debris flows occur. The high concentration of solid materials carried in a debris flow destroys everything in its path and causes severe damages to the inhabitants and their Manuscript received September 6, 2009; revised October 29, 2009; accepted November 19, 2009. This work was supported in part by the National Science Council, Taiwan, under Grant NSC 97-2218-E-007-001, by the Ministry of Economic Affairs, Taiwan, under Grant 96-EC-17-A-04-S1-044, and by Microsoft Research (SensorMap: Browsing the Physical World in Real-Time 2007 Awards). The Associate Editor coordinating the review process for this paper was Dr. Jesús Ureña. H.-C. Lee and C.-T. King are with the Department of Computer Science, National Tsing Hua University, Hsinchu 30013, Taiwan (e-mail: huclee@ mx.nthu.edu.tw). A. Banerjee is with the SoC Technology Center, Industrial Technology Research Institute, Hsinchu 31040, Taiwan. Y.-M. Fang and B.-J. Lee are with Feng Chia University, Taichung 40724, Taiwan. Color versions of one or more of the figures in this paper are available online at http://ieeexplore.ieee.org. Digital Object Identifier 10.1109/TIM.2010.2046361

Fig. 1. Overview of the proposed debris-flow-monitoring system.

properties. Close monitoring and early warning of debris flows become imperative to save life and reduce damage. Current debris-flow-monitoring stations are typically equipped with a comprehensive set of sensors, including a rain gauge, geophone, ultrasonic water level transmitter, pore pressure transducer gauge, wired sensor, charge-coupled device (CCD) camera, and infrared spotlight [1]. Most sensors are installed along the riverbanks or mountain slopes, and provide indirect measurements of the physics of the debris flows. So far, there is no way to tell what is happening inside a debris flow and to collect its internal parameters, not to mention doing these in real time. One viable solution to address this challenge is to leverage the vast advances in wireless technologies to develop wireless sensors. In this paper, we present a novel multifunctional wireless sensor for monitoring debris flows, which can move along with the debris flow while collecting data in situ and transmit the collected data to the base stations in real time. Wireless sensor network has extensively been studied in the past few years. The technology has been applied to many application domains, including cold chain [2], agriculture and irrigation [3], bridge structure [4], volcano [5], and ocean [6]. However, in-situ debris flow monitoring needs to address a multitude of issues, including cost, deployment efforts, longterm standby, and fast reaction. In this paper, we discuss how the proposed design addresses these issues to develop a lowcost wireless sensor that can provide long-term monitoring. An overview of the proposed system is shown in Fig. 1. It is composed of two types of devices, i.e., INSIDERs and COORDINATORs. The INSIDER is a wireless sensor designed

0018-9456/$26.00 © 2010 IEEE

This article has been accepted for inclusion in a future issue of this journal. Content is final as presented, with the exception of pagination. 2

IEEE TRANSACTIONS ON INSTRUMENTATION AND MEASUREMENT

Fig. 2. Overview of the Shen-Mu debris-flow-monitoring station at Shen-Mu Village, Nantou County, Taiwan.

to sit on the riverbed. The COORDINATOR is a wireless data receiver installed on the riverbank to receive data from the INSIDERs. Under normal conditions, the INSIDERs run in STANDBY mode and sparingly communicate to save energy. When a debris flow occurs and approaches the INSIDERs, any INSIDER detecting the low-frequency vibration generated by the debris flow will send alarm messages to the COORDINATORs. The leader COORDINATOR decides whether to wake all the INSIDERs. The INSIDERs then enter ACTIVE mode and send all the collected raw data to the COORDINATORs. When the debris flow hits the INSIDERs, the INSIDERs will be moved and drift with the flow. As long as the INSIDERs are operational, they keep measuring the debris flow’s internal parameters and transmitting data. The data sent by the INSIDERs are gathered by the COORDINATORs. These data are fused and analyzed in the back-end data center; thus, early warnings may be issued, and archived data may be analyzed afterward. In summary, the proposed system has five salient features. 1) The INSIDER can carry multiple sensing modules to continuously measure the internal parameters of debris flows, closeup and in situ, and transmit the data in real time to base stations. As far as we know, this is the first complete development of such a system with test results. 2) The INSIDER is designed with robustness in mind to resist clashes and motions. 3) A novel communication scheme is designed in the INSIDER to allow long standby with fast wakeup, which are two conflicting goals that are hard to achieve at the same time. 4) The costs of components, deployment labor, and the maintenance of the proposed system are low.

5) The dual-power-source design enables the INSIDER to use the internal battery as the major power source and the solar panel to harvest additional energy in the daytime. Even using only the internal D cell batteries, an INSIDER can continuously operate for five months. The INSIDER is intended to serve multiple functions: 1) detecting the coming of a debris flow; 2) verifying the occurrence of a debris flow; and 3) measuring the internal parameters (i.e., velocity and moving direction) of a debris flow. So far, different equipment such as a geophone and trip-wire sensor perform these functions. The multifunctional INSIDER integrates these functions into a single device. The rest of this paper is organized as follows: In Section II, we briefly describe related works. Section III lists the requirements of our system design. Section IV describes the details of our system. Section V shows the evaluation of the proposed system, and Section VI gives the conclusions. II. R ELATED W ORKS In this section, we briefly describe existing debris-flowmonitoring devices and discuss their deficiencies. A debris flow can be viewed as a large volume of fast-moving sand, gravel, and/or cobblestones that are saturated with water. To monitor debris flows, various sensors are installed on the riverbank and hillside to collect data of relevance. Fig. 2 shows the Shen-Mu debris-flow-monitoring station at Shen-Mu Village, Nantou County, Taiwan. Its missions are given as follows: 1) to detect the arrival of debris flows and 2) to measure the debris flow. Debris flow information that is important to scientists includes rainfalls, flow velocity, volume, peak height of surge, impact force, ground vibration, fluid pore

This article has been accepted for inclusion in a future issue of this journal. Content is final as presented, with the exception of pagination. LEE et al.: DESIGN OF MULTIFUNCTIONAL WIRELESS SENSOR FOR IN-SITU MONITORING OF DEBRIS FLOW

Fig. 3. Snapshot of the debris flow in Ai-Yu-Zi River at Shen-Mu Village, Nantou County, Taiwan, on August 8, 2009, during Typhoon Morakot.

pressure, and boulder size. With current technologies, not every aforementioned parameter can be detected and collected. Some of the commonly used sensors are reviewed here. A trip-wire sensor is commonly used to detect the arrival of moving debris in the channel. The stretch metal wires connected to electric switches will be open-circuited when the peak of the debris flow breaks the wires. The wires are placed at different heights, so that the height of the debris flow can be detected according to the level of the broken wires. It can also measure the speed of flowing debris by calculating the time difference between two consecutive broken wires and the distance between them. However, the trip-wire sensors cannot provide internal parameters of debris flows, such as the vibration frequency and volume of debris flows. Moreover, it becomes completely useless once the wires break. The moving of debris flows generates low-frequency vibrations, which can be used to detect the arrival of debris flow and estimate its parameters [7]. Geophone can measure lowfrequency vibrations (2–60 Hz) propagating underground [8]. One problem is that the debris flow must be within a certain range for the geophone to pick up strong enough signals. In addition, the stationary geophone can only track the moving debris flow as it passes by—not all the way through. Therefore, it cannot provide continuous data such as velocity changes and moving directions that are critical for analyzing the dynamics of the debris flow. The CCD cameras are used to capture visual information in the field. Various image recognition techniques can be applied to analyze the visual data and determine the occurrence and characteristics of debris flow in real time. Although the CCD cameras can present very intuitive visual information, filtering information from pictures captured during foggy weather, rainfall, or in darkness can be very difficult. A real debris flow picture taken by a CCD camera is shown in Fig. 3. Although the picture was taken in the daytime, the image is vague and distorted by raindrops and moisture. Human interpretation is usually needed to avoid false alarm. Therefore, CCD cameras are normally used only for recording but not detection. Since rainfall is the primary factor that triggers debris flows, most debris-flow-monitoring stations also install rain gauge

3

and water level sensor. Other types of sensors are also possible, but the major hurdle is the cost of building such stations, e.g., installation labors, maintenance costs, and wiring costs for stable AC power supplies and data communication. New apparatuses for debris flow monitoring have also been proposed. In [9], the design of an intelligent sensor called DUMPLING is presented. The sensor can be thrown inside a debris flow for in-situ measurement of the acceleration and other internal parameters of debris flows. It records the data in its internal storage while it rolls along with the moving debris. After the debris flow is over, the sensors are manually retrieved from the debris, and data from their internal storage are downloaded. DUMPLING does not have sufficient storage to store the large volume of data generated during a real debris flow, and it cannot report the data in real time. It is best used in a controlled environment for experimental purposes. The same approach is also used in [10] and [11], where several novel apparatuses were designed to study sediment transportation in riverbeds and to monitor landslides. In [12], a mass flow sensor is designed to monitor debris flows, lahars, and flash floods. This sensor reports the warning events if it has been displaced by mass flows. However, this design cannot report the internal parameters of debris flows, and it is impossible to tell whether it is moved by a debris flow or floodwater. The design seems to be in the concept stage only while our system has been developed, and initial tests have been performed. III. D ESIGN R EQUIREMENTS The design requirements of a debris-flow-monitoring system, compiled together with the domain experts, are given here. 1) The system should be able to detect the occurrences of debris flows in advance by sensing the relevant signals. It should also measure the internal parameters of the debris flows and report the data in real time. 2) The sensors should sit on the riverbed and remain stationary under normal water flows. On the other hand, when debris flows occur, they should be moved and carried away. 3) The sensors should operate for at least five months without any physical maintenance, such as battery replacement or solar panel cleaning. 4) The system must be inexpensive and easy to install. This should provide incentives to deploy the system. It also allows us to deploy more sensors to improve the reliability of the data collected by the system. The requirements are converted into the system design and implementation, which are described in the next section. IV. S YSTEM A RCHITECTURE AND D ESIGN In this section, we discuss the architecture and design of the INSIDER and COORDINATOR, which are the major components in our debris-flow-monitoring system. The architecture of the debris-flow-monitoring system is shown in Fig. 4. The INSIDERs at the right side of the figure are wireless battery-powered sensors sitting in the riverbed to

This article has been accepted for inclusion in a future issue of this journal. Content is final as presented, with the exception of pagination. 4

IEEE TRANSACTIONS ON INSTRUMENTATION AND MEASUREMENT

Fig. 4. System architecture of the proposed debris-flow-monitoring system.

Fig. 5. Internal structure of the INSIDER.

detect debris flow. They report the sensed data through lowpower radio protocol 802.15.4 to the COORDINATORs on the riverbank. The data collected by several COORDINATORs is transferred through 802.3 Ethernet and gathered in the back-end computer. Scientists can use the collected real-time information to predict damage and take preventive measures. A. INSIDER The INSIDER, as shown in Fig. 5, is a mobile wireless sensor encapsulated in plastic packaging. It is designed to sit on the riverbed and will be moved only with a debris flow. It will wirelessly transmit collected data to the COORDINATORs deployed on the riverbank. 1) Hardware Platform: The INSIDER is based on Telos [13], using a TI MSP430F1611 [14] microprocessor and a TI CC2420 [15] 802.15.4 RF transceiver. Its software runs on top of TinyOS [16] to take advantage of the well-defined programming model and service components. A 3-dbi omnidirectional antenna is connected to improve radio performance. To detect the very low frequency vibrations generated by moving debris, a geophone or seismometer is conventionally used. However, their size, energy consumption, and cost make it infeasible to integrate within the INSIDER. On the other hand, MEMS accelerometers have the advantages of a small and lowprofile package and low power consumption. Moreover, they have been proven to be useful in detecting low-frequency vibrations generated by debris flows [7], [8], [11]. Therefore, the

INSIDER integrates an Analog Device ADXL330 [19], which is a low-power three-axis MEMS accelerometer to measure accelerations in the range of ±3g. To cover a border acceleration range, we can also integrate additional accelerometers such as ADXL78 that sense the range of ±70g. The energy source of the INSIDER is from two D cell alkaline batteries with a small solar panel (4 V/75 mAh) as the secondary energy source. In this prototype, we found that a pair of D cell alkaline batteries is sufficient for the wireless sensor to work for five months during the flood control period, which is from May to September in Taiwan. The system lifetime can further be extended using a solar panel as a secondary power source. However, this depends on the availability of goodquality solar panels and rechargeable batteries. 2) Working Mode and Communication Schedule: The INSIDER has three working modes: 1) INITIAL; 2) STANDBY; and 3) ACTIVE. In all three modes, the CPU constantly reads from the accelerometer ADXL330. The INSIDER starts in the INITIAL mode and broadcasts a MSG_JOIN request every 60 s to discover a neighboring COORDINATOR. The communication layer of the system is based on B-MAC [17] of TinyOS. BMAC uses carrier sense multiple access with collision avoidance (CSMA/CA) to reduce package collision. On receiving the MSG_JOIN message, the COORDINATOR assigns a new communication schedule, depending on the number of INSIDERs that it is currently handling, and sends out the CMD_NEWSCHEDULE command. The INSIDER synchronizes its local clock according to the time tag contained in the CMD_NEWSCHEDULE and CMD_TIMESYNC message that it receives from the COORDINATOR. Operating under a one-hop network configuration, the experiment shows that this simple method can synchronize the clocks of all INSIDERs with a time difference of less than 2 ms. Regarding the frequency to synchronize the sensors, we made the following design choices. We noted first that the original Telos used an inexpensive crystal oscillator to reduce cost. As a result, the internal clock of MSP430 badly drifted, i.e., ±40 ms/h in the worst case. In our application, we decided to use the COORDINATOR to synchronize all INSIDERs every 30 min to keep the clock drift less than ±20 ms. We believe this choice strikes a good balance between simplicity of time synchronization, energy saving, and precision of the clocks.

This article has been accepted for inclusion in a future issue of this journal. Content is final as presented, with the exception of pagination. LEE et al.: DESIGN OF MULTIFUNCTIONAL WIRELESS SENSOR FOR IN-SITU MONITORING OF DEBRIS FLOW

Fig. 6.

5

SCS of INSIDERs in the STANDBY mode.

As shown in Fig. 1, an interested region is cooperatively monitored by a group of INSIDERs controlled by the COORDINATORs. As the INSIDERs share the same radio channel, each INSIDER must have an exclusive communication schedule to reduce the transmission conflicts and maximize the throughput. In our design, an INSIDER has two unique communication schedules assigned by the COORDINATOR: standby communication schedule (SCS) for STANDBY mode and active communication schedule (ACS) for ACTIVE mode. After synchronization, the INSIDERs enter STANDBY mode and follow SCS to turn on and off their RF transceiver. Fig. 6 shows an example of SCS with six INSIDERs in the same group; the cycle length is 40 timeslots. The cycle length and time slot are adjustable, and the tradeoff is described in Section IV-A3. Considering the clock drift, our current design uses a larger timeslot, i.e., at least 50 ms. In Fig. 6, each INSIDER (denoted “IS”) is assigned a time slot (e.g., INSIDER 1 exclusively uses the first time slot of the cycle), in which it can report its status and statistics of sensed data, such as battery voltage, power-on time, and average and variance of measured acceleration. Based on this information, the COORDINATOR (denoted “CO”) can decide whether it needs to wake up the INSIDERs under its control by broadcasting CMD_ACTIVE. The 38th and 39th time slots in SCS are reserved for the COORDINATOR to broadcast commands to the INSIDERs, such as synchronizing clock, updating configuration, and setting working state. During these two timeslots, all INSIDERs turn on CC2420 to receive messages from the COORDINATOR. From the figure, we can see that the INSIDER turns off its RF transceiver most of the time, while, at the same time, the COORDINATOR can wake up the INSIDER just in time if an interesting event occurs. On receiving the CMD_ACTIVE message, the INSIDERs enter ACTIVE mode and use ACS to transfer collected raw data back to the COORDINATOR. In this mode, INSIDERs try to transmit as much raw data as possible before damaged by the debris flow. To avoid data collision, ACS uses time-division multiple access to schedule the INSIDERs. In Fig. 7, each INSIDER in a group of six INSIDERs is assigned a time slot for sending its raw data to the COORDINATOR. The seventh timeslot is reserved for the COORDINATOR to dynamically adjust the number of INSIDERs. This is required, because, during debris flow, some INSIDERs may be malfunctioned or

Fig. 7. ACS of INSIDERs in the ACTIVE mode.

move out of transmission range. The COORDINATOR can reclaim unused time slots and broadcast a new ACS to the reachable INSIDERs. In ACTIVE mode, the INSIDERs sense and collect data from the three-axe accelerometer at high frequency, so it can accumulate huge data in a very short time. To avoid buffer overflow, we need to send the data out as soon as possible. On the other hand, to increase the throughput, we need to set the length of the time slots as large as possible. This yields the following inequality: Rgen ≤ (Rtx /(Npresent + 1))

(1)

where Rtx and Rgen are the data transmission rate and the data generation rate of an INSIDER, respectively, and Npresent is the number of INSIDERs currently in a group. In the inequality, it is assumed for simplicity that the length of the time slots is the same. In practice, we can decrease the time slot of the COORDINATOR to increase the time of INSIDERs in an ACS cycle. From our experiments of the sensor platform, to be discussed later, the maximum package-sending rate Rtx was found to be about 120 packet/s. The platform runs TinyOS with a package payload length of 28 B and BMAC CSMA/CA enabled. The first 3 B of each package are reserved for TinyOS header data, so only 25 B are available in a package. Suppose Npresent = 6 and Rtx = 120. We have Rgen = (120/(6 + 1)) ≈ 17 package/s. As the ADC of MSP430 takes samples of the accelerometer at 12 bit/axis/sample, a package can carry 25 × 8/12/3 = 5 samples. Therefore, the INSIDERs can sample at a frequency of 5 × 17 = 85 Hz, which is sufficient for the debris-flow-monitoring application. Notice that we can further improve the sampling frequency and send more data if a certain data compression scheme, e.g., adaptive differential pulse code modulation, is used. The length of a time slot T Slength is bounded by the size of the internal data buffer Sintbuf . If the time slot is set for too long, the buffer may overflow. This is given by the following inequality: T Slength ≤ Sintbuf / (Rgen × (Npresent + 1)) .

(2)

This article has been accepted for inclusion in a future issue of this journal. Content is final as presented, with the exception of pagination. 6

IEEE TRANSACTIONS ON INSTRUMENTATION AND MEASUREMENT

TABLE I MRT VERSUS TIME SLOT LENGTH AND CYCLE LENGTH

In our platform, the available memory of an INSIDER for data buffering is about 8 kB or 8×1024/28 = 292 packages. Considering the case of six INSIDERs and one COORDINATOR, as shown in Fig. 7, the upper bound of T Slength is 292/(17 × (6 + 1)) ≈ 2.45 s. Let Tsendout be the maximum duration in an INSIDER from the time data is generated to the time it is sent out. It is affected by the time slot T Slength as follows: Tsendout = T Slength × Npresent .

(3)

Consider Fig. 7 again. INSIDER 1 samples the accelerometer at time t1 , which is at the start of the second time slot of cycle m. This INSIDER has to wait until its time slot (the first time slot of cycle m + 1) at time t2 to send the sampled data out. Thus, Tsendout = t2 − t1 . According to (3), if we set T Slength = 2.45 s, then Tsendout is 2.45 × 6 = 14.7 s. Unfortunately, this is too long for debris flow monitoring, because INSIDERs might have already been damaged, and the buffered data are lost. Considering a reasonable delay, we define T Slength = 0.2 s, so Tsendout = 0.2 × 6 = 1.2 s. 3) Energy-Saving and Wake-Up Response Time: In the current wireless sensor platform, the RF transceiver still dominates the energy consumption. We used a precise current meter to measure the INSIDER in the ACTIVE mode, i.e., MSP430, CC2420, and ADXL330 are all on. The RF transceiver, CC2420, and related components drew about 27.5 mA at 3 V, whereas the remaining components used only 0.5 mA. It follows that the best strategy for extending the lifetime of the INSIDER is simply to turn off the RF transceiver for as long as possible. When a debris flow occurs, e.g., detected by some upstream INSIDERs, the COORDINATORs on the riverbank will be notified. They will then wake up the INSIDERs downstream to prepare for the coming debris flow. The next time the INSIDERs turn on their RF transceiver, they will receive the notification and switch to ACTIVE mode. They will also start transmitting data to the COORDINATORs. All these operations have to be completed before the debris flow arrives. If the standby interval, i.e., the maximum duration for an INSIDER to turn off its RF transceiver, is too long, they may not be able to receive the notification in time. If the standby interval is too short, the INSIDERs will more frequently turn on, even if events do not occur, which wastes energy. There is, thus, a critical tradeoff between the standby interval and the wakeup latency. We define the duration when a COORDINATOR decides to wake up an INSIDER until the INSIDER is turned on as the maximal response time (MRT). MRT is mainly affected by the SCS of INSIDERs. We use the configuration in Fig. 6 as an example. In this example, we set the cycle length to 40 time slots and the slot length to 50 ms. From Fig. 6, the MRT occurs on INSIDER 1. This INSIDER will turn on its RF transceiver only in the 1st, 38th, and 39th time slots. If the event occurs at time slot 2, the COORDINATOR has to wait until slot 38 to notify INSIDER 1. The following equation shows how MRT is calculated: MRT = (SCScycle_length − 2 − 2) × T Slength .

(4)

Fig. 8.

INSIDER’s MRT and current under different cycle lengths.

MRT for this case is thus ((40-2-2) × 50 ms) = 1800 ms. Table I shows MRT under different cycle lengths and time slot lengths. Obviously, MRT increases proportional to the time slot length. In this table, MRT in a gray-colored cell is more than 1 s, and in a brown-colored cell, it is more than 2 s. It seems that, to maintain a longer cycle length while keeping MRT low, we have to choose the smallest time slot length, i.e., 50 ms. We have conducted an experiment to evaluate the energysaving performance of INSIDERs. We set an INSIDER’s cycle length to ten time slots and the time slot length to 50 ms. The INSIDERs used a pair of D alkaline batteries as the primary energy source, and the built-in solar panel was not used. The experiment shows the INSIDER’s voltage drops to 2 V after it ran in the STANDBY mode for 47 days and shut down. The average current drawn in the aforementioned setting is ((27.5 mA × 3) + (0.5 mA × (10 − 3)))/10 = 8.6 mA. We can derive the capacity of the alkaline D cell as 8.6 mA × 24 h × 47 days ≈ 9700 mAh, which is close to the typical capacity of a D cell alkaline battery. In a similar way, we can derive the expected standby time under other cycle lengths. Fig. 8 shows the relation between MRT and current consumption under different cycle lengths. The time slot length was set to 50 ms. From Fig. 8, we can see that current consumption drops when cycle length and MRT increase. The curve of current consumption becomes flat beyond the cycle length of 26 time slots. This shows that a cycle length larger than 26 time slots does not significantly reduce the current consumption. Because debris flows travel very fast (2–20 m/s), it is critical to define a reasonable MRT to ensure proper operations of the wireless sensor system. According to the requirements of debris flow monitoring, the upper bound of MRT is set to 1 s in

This article has been accepted for inclusion in a future issue of this journal. Content is final as presented, with the exception of pagination. LEE et al.: DESIGN OF MULTIFUNCTIONAL WIRELESS SENSOR FOR IN-SITU MONITORING OF DEBRIS FLOW

7

TABLE II ANTENNA USED IN THE EVALUATION

B. Coordinator

Fig. 9. INSIDER stands in floodwater on the riverbed to test its waterproof encapsulation.

our design. For other applications that do not require very fast response time, we may set MRT to 1 min or even 1 h to extend the lifetime of the wireless sensor system. Based on the preceding settings, we can set the cycle length to 22 time slots, resulting in a current consumption of 4.18 mA. The estimated INSIDER standby time is about 97 days. If we further extend MRT to 2 s and cycle length to 40 time slots, we can further reduce the average current consumption to 2.53 mA, where the standby time is extended to about 160 days. To satisfy the requirement that INSIDERs continuously work for five months, we can set MRT to 1 s on rainy days and 2 s on sunny days. This is under the observation that debris flow occurrence is highly correlated to rainfall. According to the data provided by the Central Weather Bureau of Taiwan [23], there is an average of 105 rainy days per year or 105/365 ≈ 29% of the days. The INSIDER lifetime is given by 9700 mAh/ (4.18 mA × 29% + 2.53 mA × 71%)/24 h = 136 days, which is about 4.5 months. The requirement is thus almost satisfied. Note that, if we use solar panels to harvest solar energy, the lifetime could be much longer than just using the energy from the internal battery. 4) Packaging: The INSIDER is tightly encapsulated. Its shape is specially designed to make it streamlined to lessen water resistance and to prevent it from moving in fresh water flows. The package is waterproof, weather resistant, and inexpensive. It is modified from the off-the-shelf ABS 4-in cleanout bushing and cap. The internal electronics can easily be maintained by simply loosing the cleanout bushing. A metal plate is attached to the bottom of the package to keep the center of gravity low and prevent it from overturn when carried along by debris flows. To prevent INSIDERs from floating away with fresh water flows, the weight of this metal plate helps to adjust their density. In general, the density of debris flows is in the range of 1.5 g/cm3 and 2.4 g/cm3 , and the density of a bare INSIDER is 0.86 g/cm3 . After a 1.8-kg metal plate is installed, the INSIDER density becomes 1.5 g/cm3 , making it denser than clear water and less dense than debris flow. As shown in Fig. 9, the INSIDER tested in the field proved that this design works. If floodwater is very fast, we may apply additional temporary fastening to ensure that the INSIDER can securely stand on the riverbed.

COORDINATORs are responsible for managing the schedule of INSIDERs and gathering the data from INSIDERs. We may install multiple COORDINATORs on the riverbank to increase the probability of receiving data from the INSIDERs. However, only one COORDINATOR will be assigned as the leader that can use the time slots reserved in SCS and ACS to broadcast the commands. In the prototype implementation, the COORDINATOR integrates Tmote connect [20], using Ethernet to communicate with other devices and back-end computers. As COORDINATORs are responsible for collecting data from INSIDERs, their antenna will affect the wireless communication performance. We tested several types of antenna, and the results are discussed in the next section. V. S YSTEM E VALUATION AND D ISCUSSION In the previous sections, we have evaluated several system performance factors, including the tradeoff between long standby time and short reaction time, estimating the expected system lifetime and analyzing the bound of sampling frequency and data throughput. In this section, we evaluate the performance of our debris-flow-monitoring platform, emphasizing the effects of different antenna on communication performances. The experimental environment was set as follows: An INSIDER was placed on the ground 20 to 120 m from the COORDINATOR with an interval of 20. The radio path was unobstructed. The COORDINATOR was installed at a 3-m height above the ground to simulate the scenario in which the COORDINATOR is on the riverbank and the INSIDER is on the riverbed. The INSIDER used an omnidirectional 3-dBi antenna, which cannot be replaced due to the limited space inside the INSIDER. Four types of 2.4-GHz antennas from [21] for the COORDINATOR were tested. The details of these antennas are listed in Table II. We use the notations in the first column of the table to identify the antenna in the following descriptions: 15-O indicates the omnidirectional 15-dBi antenna, and 20-P indicates the panel 20-dBi antenna. Factors affecting the performance of the antenna include the gain of the antenna, which affects the communication distance, and horizontal polarization and vertical polarization, which affect communication range. The panel-type antenna is a directional antenna, so, in the experiment, we must align the direction of the antenna’s maximum horizontal and vertical polarizations to the INSIDER on the ground. On the other hand, omnidirectional antenna does not require alignment. The test procedure is given as follows: The COORDINATOR sends a message to query the internal parameters of the

This article has been accepted for inclusion in a future issue of this journal. Content is final as presented, with the exception of pagination. 8

Fig. 10. IS2CO PDR in different antenna and distance settings.

Fig. 11. IS2CO PDR in different IS2CO LQI and IS2CO RSSI.

INSIDER, and the INSIDER immediately replies. This query is repeated 1000 times with an interval of 50 ms. At the end, the COORDINATOR computes the average link quality indicator (LQI), receive signal strength indicator (RSSI), and package delivery rate (PDR). The INSIDER’s reply message contains RSSI and LQI from both sides. We denote IS2CO for the signal quality detected at the COORDINATOR’s end for the data received from the INSIDER and CO2IS at the INSIDER’s end for the data received from the COORDINATOR. RSSI and LQI are two signalquality measurements provided by CC2420 for radio performance estimation [15]. RSSI indicates the signal strength, and the higher the RSSI, the better the receiving performance. LQI is the quality characteristics of the incoming data provided by the MAC layer. Its value ranges from 0 to 255. The lowest LQI for CC2420 to successfully detect a data frame is about 50. In this evaluation, we focus on IS2CO for the COORDINATOR’s communication performance. Fig. 10 shows the PDR under different distances and antenna. Except that 8-O badly performs, the PDRs of 20-P, 18-P, and 15-O can all reach about 100% when the distance is less than 80 m. For distances more than 80 m, the PDR significantly varies, and the link is not as stable. Note that 20-P requires extra effort to align the antenna to achieve such a performance, because the horizontal polarization of 20-P is only 15◦ wide. On the other hand, the horizontal polarization of 18-P is 35◦ wide, which makes it easy to align to the INSIDER at a distance. Fig. 11 shows how LQI and RSSI are related to PDR in terms of IS2CO. When RSSI is higher than −89 dBm and LQI is higher than 105, the PDRs are all at 100%. This means that, in real deployment, we can measure the IS2CO’s LQI and RSSI to ensure PDR performance. This approach provides a simple

IEEE TRANSACTIONS ON INSTRUMENTATION AND MEASUREMENT

yet practical solution to keeping PDR high during deployment in the field. From the preceding discussions, we can see that the COORDINATOR’s antenna indeed considerably affect PDR, LQI, and RSSI. Since the INSIDERs may be moved by debris flows, if we need to receive data from a group of INSIDERs scattered in a small region, 18-P is a better choice than 20-P. 18-P provides a broadened region of sensitivity and better RSSI than 15-O. If the INSIDERs are scattered over a large region, then the omnidirectional antenna like 15-O is a good choice, because it provides 360◦ horizontal polarization. The experiments also show that the distance between the COORDINATOR and INSIDER should be less than 80 m to keep PDR high. In fact, this communication distance is sufficient if we want to deploy the system in the field such as the Shen-Mu debrisflow-monitoring station, where the typical riverbed width is about 25 m. In addition, we can always install additional COORDINATORs in different locations of the riverbank to increase the PDR and the amount of data received. To evaluate the effects of weather on communication performance, we deployed the INSIDER 40 m away from the COORDINATOR (equipped with 15-O) and let the COORDINATOR query the status of the INSIDER every 6 s. The experiment ran through the Prama Typhoon during October 3–6, 2009, which brought heavy rain. The results show that the communication performance, i.e., RSSI and LQI, is not degraded by rainfall. During the rainfall period, the average RSSI (−83 dBm) is even better than that on sunny days (−86 dBm). A similar result is also reported in [25], which concludes that 2.4 GHz is not degraded by rainfall. Another issue is how communication is affected if the INSIDER is immersed in water. We performed another experiment that sets the INSIDER on the riverbed and the COORDINATOR (equipped with 15-O) 20 m away from the INSIDER. If the lower portion of the INSIDER is soaked in water, RSSI and LQI will fluctuate. If the INSIDER is completely under water, the RSSI dramatically drops, and communication becomes very unstable, primarily due to the low-power radio platform of the INSIDER, which uses 2.4-GHz 802.15.4 at 0-dBm output power level. This issue may be alleviated by integrating a power amplifier, such as TI CC2591 [27], into the INSIDER to increase the output power to 22 dBm. On the other hand, the intended use of the INSIDER provides other possibilities. Most rivers in Taiwan are short. They are easily flooded under heavy rain or typhoons but otherwise have large portions of dry riverbed (see Fig. 12). INSIDERs will be deployed on the dry riverbed under normal conditions. If a debris flow occurs, the debris must fill up the whole riverbed, and the INSIDER will directly be hit. In the worst case, they will stop transmitting packets, which also means something. Furthermore, each INSIDER is equipped with flash memory, which can store the collected information during a debris flow. The devices may be retrieved afterward. VI. C ONCLUSION AND F UTURE W ORK In this paper, we have presented the design and implementation of a prototype multifunctional wireless sensor for debris

This article has been accepted for inclusion in a future issue of this journal. Content is final as presented, with the exception of pagination. LEE et al.: DESIGN OF MULTIFUNCTIONAL WIRELESS SENSOR FOR IN-SITU MONITORING OF DEBRIS FLOW

Fig. 12. Typical riverbed in the Shen-Mu debris-flow-monitoring station. Note that the regions of riverbed closed to both riverbanks are dry and suitable for setting INSIDERs.

flow monitoring. Its key feature is to be moved along with the debris flows to collect internal parameters of the flows in situ. As far as we know, no such device has previously been developed and tested. We have described the design and related issues, including the hardware and software design, energysaving mechanism, communication protocols, packaging, deployment strategy, and communication performance. We envision that this design can also be applied to similar natural disaster monitoring applications, such as snow avalanche and landslides. We plan to deploy our proposed system in one of the 14 existing debris-flow-monitoring stations in Taiwan to help scientists to more efficiently monitor debris flows and provide more detailed information.

9

[8] M. Jakob and O. Hungr, Debris-Flow Hazards and Related Phenomena. Berlin, Germany: Springer-Verlag, 2005. [9] J. Hanisch, P. Ergenzinger, and M. Bonte, “Dumpling—An “intelligent” boulder for studying internal processes of debris flows,” in Proc. 3rd Int. Conf. Debris-Flow Hazard Mitigation—Mechanics, Prediction, Assessment, Davos, Switzerland, Sep. 2003. [10] M. Spazzapan, J. Petrovèiè, and M. Mikoš, “A new tracer for monitoring dynamics of sediment transport in turbulent flows,” Acta Hydrotechnica, vol. 22, no. 37, pp. 135–148, 2004. [11] Y. Itakura, T. Kitajima, K. Endo, and T. Sawada, “A new double dualaxes accelerometer debris flow detection system,” in Proc. 2nd Int. Conf. Debris-Flow Hazard Mitigation—Mechanics, Prediction, Assessment, Taipei, Taiwan, Aug. 2000. [12] [Online]. Available: http://www.zostrich.com/index.html [13] J. Polastre, R. Szewczyk, and D. Culler, “Telos: Enabling ultra-low power wireless research,” in Proc. 4th Int. Symp. Inf. Process. Sensor Netw., Los Angeles, CA, 2005, p. 48. [14] [Online]. Available: http://www.ti.com/msp430 [15] [Online]. Available: http://focus.ti.com/docs/prod/folders/print/ cc2420.html [16] [Online]. Available: http://www.tinyos.net/ [17] J. Polastre, J. Hill, and D. Culler, “Versatile low power media access for wireless sensor networks,” in Proc. 2nd Int. Conf. Embedded Netw. Sensor Syst., Baltimore, MD, 2004, pp. 95–107. [18] Soil and Water Conservation Bureau of Taiwan, ROC. [Online]. Available: http://en.swcb.gov.tw/ [19] [Online]. Available: www.analog.com/en/ADXL330/productsearch.html [20] [Online]. Available: http://www.sentilla.com/pdf/eol/tmote-connectdatasheet.pdf [21] RF Castle Electronics Co., Ltd. [Online]. Available: http://www. rfcastle.com [22] [Online]. Available: http://www.geospacelp.com/ [23] Central Weather Bureau of Taiwan, ROC. [Online]. Available: http:// www.cwb.gov.tw/eng/index.htm [24] Crossbow Technology. [Online]. Available: http://www.xbow.com [25] eKo: Wireless Crop and Environmental Monitoring System. [Online]. Available: http://www.xbow.com/eko/index.aspx [26] C. A. Bonao, J. Brown, Z. He, R. Utz, and V. Thiemo, “Low-power radio communication in industrial outdoor deployments: The impact of weather conditions and ATEX-compliance,” in Proc. 1st Int. Conf. Sensor Netw. Appl., Experimentation Logist., Athens, Greece, 2009, pp. 159–176. [27] [Online]. Available: http://focus.ti.com/docs/prod/folders/print/ cc2591.htm

ACKNOWLEDGMENT The authors would like to thank Y.-H. Huang, C.-H. Chen, and K.-M. Lo of the Geographic Information Systems Research Center, Feng Chia University, Taichung, Taiwan, and P.-Z. Kuo and P.-C. Lee for their excellent technical assistance and comments.

Huang-Chen Lee (S’09) received the M.S. degree in computer science from Soochow University, Taipei, Taiwan, in 2005. He is currently working toward the Ph.D. degree in computer science from the Department of Computer Science, National Tsing Hua University, Hsinchu, Taiwan. He has worked in the industry since 2000 and has a wide breadth of experience in designing personal digital assistant/cellular phones and low-power embedded systems. His research interests include distributed processing and networked embedded

R EFERENCES [1] M. Arattano and L. Marchi, “Systems and sensors for debris flow monitoring and warning,” Sensors, vol. 8, pp. 2436–2452, Apr. 2008. [2] A. Carullo, S. Corbellini, M. Parvis, and A. Vallan, “A wireless sensor network for cold-chain monitoring,” IEEE Trans. Instrum. Meas., vol. 58, no. 5, pp. 1405–1411, May 2009. [3] Y. K. Kim, R. G. Evans, and W. M. Iversen, “Remote sensing and control of an irrigation system using a distributed wireless sensor network,” IEEE Trans. Instrum. Meas., vol. 57, no. 7, pp. 1379–1387, Jul. 2008. [4] K. Chebrolu, B. Raman, N. Mishra, P. K. Valiveti, and R. Kumar, “Brimon: A sensor network system for railway bridge monitoring,” in Proc. 6th Int. Conf. Mobile Syst., Appl., Services, Breckenridge, CO, 2008, pp. 2–14. [5] W.-Z. Song and R. Huang, “Air-dropped sensor network for real-time high-fidelity volcano monitoring,” in Proc. 7th Int. Conf. Mobile Syst., Appl., Services, Kraków, Poland, 2009, pp. 305–318. [6] Z. Yang, M. Li, and Y. Liu, “Sea depth measurement with restricted floating sensors,” in Proc. 28th IEEE Int. Real-Time Syst. Symp., Tucson, AZ, 2007, pp. 469–478. [7] R. G. Lahusen, Detecting Debris Flows Using Ground Vibrations, U.S. Geological Survey Fact Sheet 236-96, 1996.

systems.

Amit Banerjee received the B.S. and Master of Computer Application (MCA) degrees from VisvaBharati University, West Bengal, India, in 1998 and 2002, respectively, and the Ph.D. degree in computer science from National Tsing-Hua University, Hsinchu, Taiwan, in 2009. He is currently working with the SoC Technology Center, Industrial Technology Research Institute, Taiwan. His research interests include mobile ad hoc and sensor networks, specifically scalable routing, reliable multicasting, scheduling, and resource management.

This article has been accepted for inclusion in a future issue of this journal. Content is final as presented, with the exception of pagination. 10

IEEE TRANSACTIONS ON INSTRUMENTATION AND MEASUREMENT

Yao-Min Fang received the Ph.D. degree from Feng Chia University (FCU), Taichung, Taiwan, in 2006. Before he joined the Department of Civil and Hydraulic Engineering, FCU, in 2006, he was a Postdoctoral Researcher with the Geographic Information Systems Research Center, FCU. He is currently an Assistant Professor and the Chief Researcher of the Geographic Information Systems Research Center. His research interests include disaster monitoring and bridge engineering.

Bing-Jean Lee received the Ph.D. degree from the University of Texas, Austin, in 1991. Before he joined the Department of Civil Engineering, Feng Chia University (FCU), Taichung, Taiwan, in 1994, he was a Postdoctoral Researcher with Massachusetts Institute of Technology, Cambridge, and the University of California, San Diego. He is currently the Vice President of FCU. Before that, he was the Chairman of the Department of Civil Engineering, Dean of the College of Construction and Development, and Dean of the Office of Academic Affairs. Dr. Lee serves as the President of the Chinese Information Literacy Association, Taiwan.

Chung-Ta King (M’88–SM’06) received the B.S. degree in electrical engineering from National Taiwan University, Taipei, Taiwan, in 1980 and the M.S. and Ph.D. degrees in computer science from Michigan State University, East Lansing, in 1985 and 1988, respectively. From 1988 to 1990, he was an Assistant Professor of computer and information science with New Jersey Institute of Technology, Newark, NJ. In 1990, he joined the faculty of the Department of Computer Science, National Tsing Hua University, Hsinchu, Taiwan, where he is currently a Professor and the Department Chair. His research interests include parallel and distributed processing, and networked embedded systems.