Structural Health MonitoringUsing Wireless Sensor NetworksSukun Kim492 Soda HallUniversity of California atBerkeleyBerkeley, CA [email protected] Culler627 Soda HallUniversity of California atBerkeleyBerkeley, CA [email protected] Demmel737 Soda HallUniversity of California atBerkeleyBerkeley, CA [email protected] monitoring brings new challenges to wireless sen-sor network: high-fidelity sampling, collecting large volumeof data, and sophisticated signal processing. New accelerom-eter board measures tens of G acceleration. High frequencysampling is enabled by new component of David Gay. Withnew component, up to 6.67KHz sampling is possible withjitter less than 10s. Large-scale Reliable Transfer (LRX)component collects data at the expense of 15% penalty ofchannel utilization for no data loss. To overcome low signal-to-noise ratio, analog low-pass filter is used, and multipledigital data are averaged. Structure monitoring is a drivingforce for extending capability of wireless sensor networkssystem.Keywordswireless sensor networks, structure monitoring1. INTRODUCTIONWireless sensor network enables low-cost sensing of envi-ronment. Many applications using wireless sensor networkshave low duty cycle and low power consumption. Howeverthe ability of wireless sensor networks can be extended in re-verse way. Enhanced TinyOS, and new components openedpossibility for more aggressive applications. Structure mon-itoring is one example of such applications. To monitor astructure (e.g. bridge, building), we measure behavior (e.g.vibration, displacement) of structure, and analyze healthof the structure based on measured data. Figure 1 showsoverall system. Each component can have multiple subcom-ponents. In our case, sensor is accelerometer which will bediscussed in Section 2, and analog processing has low-passfilter (Section 6.) Digital processing includes averaging (Sec-tion 6), data collection (Section 5), and system identification(Section 6). Low-jitter control contains high-frequency sam-pling (Section 4). There are more sub-components to beAnalogProcessingSensorDigitalProcessingAnalog to DigitalConverter (ADC)FeedbackLow-JitterControlAnalogProcessingSensorDigitalProcessingAnalog to DigitalConverter (ADC)FeedbackLow-JitterControl Figure 1: Overall Systemadded in the future: time synchronization in low-jitter con-trol, calibration and digital filtering in digital processing.Here we present challenges, findings, and our experience instructure monitoring using wireless sensor networks. Ratherthan focusing on one single component, this paper overviewoverall system and issues in each component.2. RELATED WORKHabitat monitoring is a leading application of wireless sen-sor network. And it is an example application with low dutycycle. ZebraNet[6] uses PDA-level device with 802.11b wire-less network. Great Duck Island[8] uses Berkeley mote, andwatch ducks without disturbing them at low cost. For struc-ture monitoring, there are tremendous amount of researchusing conventional wired way. GPS was used combined withwired data collection[10, 3], however at a high cost. Thereis an approach using wireless network for data collection[2],which has great advantage over wired network. However,it uses large hardware platform (in terms of size, power,and cost) which diminishes benefit of wireless approach. [7]uses low-cost device and wireless network, but it is more likeconceptual test, and fidelity is not sufficient for real deploy-ment. We begins with high fidelity sampling in the followingsection.3. DATA ACQUISITIONData acquisition is composed of mainly two parts: data sam-pling, and data collection. Structure monitoring requireshigh fidelity data sampling. Accurate, high frequency sam-pling, and low jitter are main requirement for high qualitysample. Accuracy is discussed in this section, and high fre-quency sampling with low jitter will be covered in Section 4.And data collection will be discussed in Section 5. In struc-ture monitoring, acceleration signal is very week. Detectingeven moderate earthquake requires to measure 500G accel-eration. Sensitivity and accuracy of accelerometer is cru-Figure 2: Accelerometer BoardTable 1: Two Accelerometers Combined with Sys-temADXL 202E Silicon Designs 1221LType MEMS MEMSNumber of axis 2 1Range -2G to 2G -0.1G to 0.1GSystem noise floor 200(G/√Hz) 30(G/√Hz)Price $10 $150cial, so we put significant portion of effort to accelerometerboard. New accelerometer board was designed by as shownin Figure 2.3.1 AccelerometersIt has two kinds of accelerometers: ADXL 202E, SiliconDesigns 1221L. Table 1 shows characteristics of each ac-celerometer combined with entire system. Accelerometerboard contains 1 of ADXL 202E, and 2 of Silicon Designs1221L, and 4 16bit analog to digital converter (ADC). Thereare two channels for ADXL 202E, and two channels for Sil-icon Designs 1221L with same orientation. One is parallelto gravity, and the other is vertical to gravity. Initially bothaccelerometers had range of -2G 2G, but for better sen-sitivity, range of Silicon Designs 1221L is change to -0.1G0.1G. Channel with axis parallel to gravity has 1G offsetto compensate for offset by gravity. It also contains onetemperature sensor (reason will be explained later). Newversion of Berkeley mote, named as Mica2 [1], is used forcontrol and communication.3.2 Noise Floor Test and Shaking Table TestTo see static characteristic of accelerometers, accelerome-ter board was put to quiet place (from vibration and sound)with constant temperature. This test shows noise floor whichis shown in Table 1. For Silicon Designs 1221L, range was-0.1G 0.3G. Then to see dynamic behavior of accelerome-ters, we performed shaking table test with constant temper-ature. Even though test site was not completely free fromvibration and sound noise, it was quiet enough for a dynamicrange of shaking table to dominate noise. Results are shownin Figure 3. Left figure is result of ADXL 202E, right oneis result of Silicon Designs 1221L, and driving frequency is0.5Hz. Data are read from both channels at the same time.For this test, channel for Silicon Designs 1221L had rangeof -2G 2G.We can see Silicon Designs 1221L shows cleaner shape in0 2 4 6 8 10 12 14 16 18-0.5-0.4-0.3-0.2-0.100.1Low resolution Sensor, Test1, 0.5HzTime
View Full Document
Unlocking...