Field Testing of Wireless Interactive Sensor Nodes

Transcription

1 Field Testing of Wireless Interactive Sensor Nodes Judith Mitrani, Jan Goethals, Steven Glaser University of California, Berkeley

2 Introduction/Purpose This report describes the University of California at Berkeley s contribution to a PARI full-scale experiment using controlled blasting to induce lateral spreading. Our purpose in participating in this experiment at the Port of Tokachi was to utilize new wireless sensor nodes to measure ground accelerations and the accelerations of surcharge weights representing buildings. The sensors were placed in two zones of the experiment area: (1) the improved ground zone, including the boxes acting as surcharge weights, and (2) the liquefaction area behind the seismically resistant sheet pile wall (PARI study area). The sensor motes have been used extensively at UC Berkeley to measure accelerations on structures during induced earthquakes of varying magnitudes. Many of the tests have been carried out on a full-scale 3-story wood frame building mounted on the EERC (Earthquake Engineering Research Center) 6 m 2 shaking table capable of subjecting specimens weighing up to 50,000 kg to three translational components of dynamic excitation. In these experiments, the sensors have measured valuable data describing structural behavior during major dynamic events. Direct comparison with results from traditional Wilcoxen piezo-electric accelerometers digitized to 16-bits show that the Berkeley mote records exactly the same accelerations. Further analysis of the recorded data will indicate where local areas of damage occur. The wireless sensor modules have proven to be successful in controlled environments for full-scale experiments of structures. We seek to demonstrate that their use may be extended to numerous other engineering applications, such as this effort at the Port of Tokachi.

3 Background on Sensors/Accelerometers The motes are devices developed at the University of California Berkeley by a large interdisciplinary team lead by David Culler (Computer Science), Kristofer S. J. Pister (Electrical Engineering), and Steven Glaser (Civil Engineering). Crossbow Technology, Inc is currently putting the devices into commercial production. The Crossbow CN4000 Preliminary Series wireless sensor node, shown in Figure 1, was used in the experiment at the Port of Tokachi. The sensor s dimensions are 8 cm 6 cm 3.5 cm (L W D). The sensor platform consists of a two-board sandwich: the main board with the CPU and radio communication shown on the lefthand side of Figure 2, and the secondary sensor board shown on the right-hand side of Figure 2. The main board, or the motherboard consists of the ATMEL 4Mhz processor (8bit microcontroller unit), 512 bytes of RAM, 8K program flash memory, a RFM 900Mhz radio with up to 100-ft range, radio signal strength control and sensing, I 2 C EEPROM, three LED debugging indicators and a stackable expansion connector. The sensor board we used carries a dual-axis accelerometer chip, the ADXL202E. The chip is a low-powered, complete 2-axis accelerometer with digital and analog output, on a monolithic integrated circuit. We use the analog output to improve resolution. The accelerometer measures dynamic accelerations (vibrations) and static accelerations (tilt). The accelerometer has a range of +/- 2g and a typical precision of 2mg. The sensor nodes are quite versatile and can support various types of sensors (e.g. accelerometers, magnetometers, thermistors, strain, pressure) at once by stacking one on top of the other on the motherboard.

4 Motherboard Sensor Board Figure 1 Figure 2 The motes operate using an operating system (TinyOS) designed for use with embedded network sensors, developed by Professor Culler (EECS). This operating system allows the user of the sensor nodes to easily write software for a variety of custom applications, without worrying much about the low-level implementations. These devices have been developed to explore the uses of distributed computation, sensing, and communication modules. To further this goal of distributed computation the motes have been designed to generate a so-called ad-hoc network. The motes can determine optimal network configuration and message passing based solely on information sensed at run-time. The philosophy that has governed the developments of these devices is that the true power of a distributed network comes not from the individual capabilities of the nodes in the network, but rather from the intelligent integration of the networked sensors. The sensor modules shown and described above are programmed to communicate amongst themselves as well as with a base-station. In our implementation, the base-station that is

5 attached to a programming board (as shown in Figure 3) is connected with a serial cable to a laptop, which serves as the center for data acquisition. The wireless sensor nodes remain in communication with the base-station, which provides a time-stamp used to synchronize the network of sensors. The sensor nodes were programmed to sample accelerations at a rate of 64 samples per second. The base-station triggered the nodes to begin saving the values of the ground accelerations in the EEPROM. Once commanded to save data, the sensors re-transmit this command out to neighboring nodes in case the others do not have a direct line of communication with the base-station. Having the nodes re-transmit messages sent by the basestation allows for a larger radius of monitoring. The sensors store their data in the EEPROM until queried by the base-station to transmit the data wirelessly back to be saved on the laptop. The Crossbow software bundled with the CN4000 Series sensor node helps to manage the data acquisition process. Parallel Port for Programming Serial Port for Base- Station Mote Topography Figure 3 A total of twenty sensor motes were used in the two zones of the Port of Tokachi experiment area. Eight sensors were distributed in the improved ground zone. From these eight, four were placed immediately in front of the surcharge weights to measure ground accelerations. The other

6 four sensors were placed on the side of the actual surcharge weights (boxes) to measure their accelerations relative to the ground. The remaining twelve nodes were distributed on the liquefaction area behind the seismically resistant sheet pile wall (PARI study area) to measure ground accelerations. Some of these nodes were placed alongside PARI s sensors (for redundancy), as shown in Table 1. The entire topography of the wireless sensor nodes is shown in Figure 4. PARI Sensor UC Berkeley Sensor AA 3 13 AB 2 31 AB 3 33 AB 4 15 AB 6 17 AC 4 24 AC 6 19 AD 4 11 Table 1 Experiment Data Of the 20 experimental wireless sensor nodes used in this experiment, nine motes gave useful data. From these nine data streams, one was disrupted halfway through the experiment (sensor node #23) and one started to record about one-third of the way through the experiment (sensor node #22). The reasons for this phenomenon are addressed in the Results section of this report.

7 Two axes of acceleration were collected from each node. These accelerations, in g s, of both axes are simultaneously plotted with respect to time in Figures Figures 5-13 (a and b) are plots of the individual axes with respect to time. Please refer to the map, Figure 4, for spatial reference of the sensor nodes on the experiment site. The results of the sensor nodes that collected useful data are summarized in Table 2 shown below. Mote ID Packet Loss Spikes Location Notes 16 No No Box 1 Antenna in vertical position. 14 No No Box 2 Antenna in vertical position. 23 Yes Yes Box 2 Ground Antenna in vertical position; sensor stopped recording data midway, possibly due to a strong shake or coming undone inside the box. 22 Yes Yes Box 3 Antenna in vertical position; roughly 38% of data at the beginning of the experiment is lost for an unknown reason; could have started saving in eeprom mid-way or could have saved data over its original good data. 29 No No Box 4 Antenna in vertical position. 21 Yes Yes 18 Yes Yes 24 Yes Yes 33 Yes Yes Box 4 Ground PARI Ground PARI Ground PARI Ground Antenna in vertical position; this mote was submerged in water about 4 inches, after the explosion. This mote was used as a message hopper; its antenna was in a vertical position; was not redundant with PARI's sensors. Antenna in horizontal position and a redundant sensor, with one of PARI's sensors. Antenna in horizontal position and a redundant sensor, with one of PARI's sensors. Table 2

8

9 From Top to Bottom: Figure 5, Figure 5a, Figure 5b

10 From Top to Bottom: Figure 6, Figure 6a, Figure 6b

11 From Top to Bottom: Figure 7, Figure 7a, Figure 7b

12 From Top to Bottom: Figure 8, Figure 8a, Figure 8b

13 From Top to Bottom: Figure 9, Figure 9a, Figure 9b

14 From Top to Bottom: Figure 10, Figure 10a, Figure 10b

15 From Top to Bottom: Figure 11, Figure 11a, Figure 11b

16 From Top to Bottom: Figure 12, Figure 12a, Figure 12b

17 From Top to Bottom: Figure 13, Figure 13a, Figure 13b

18 PARI Sensors vs. UCB Sensors Two UC Berkeley sensors (#24, #33), redundant in position with PARI s sensors (#AC4, #AB3), were compared using basic signal processing methods (provided by the Matlab Signal Processing Toolbox). Both UCB s and PARI s sensors were resampled at 40 samples per second. They were resampled using a lowpass filter to the input sequence to prevent aliasing during resampling. Then, the cross-correlation sequence was evaluated to find the correlation lag (time lag) between the UCB and PARI signals. The cross-correlation was calculated with an efficient FFT (Fast Fourier Transform)-based algorithm, to measure the similarities between the two different sets of data. The correlation coefficient of the two signals is 0.78 in both cases (UCB sensor #24 vs. PARI sensor AC4, and UCB sensor 33 compared vs. PARI sensor AB3). The comparison of the signals is plotted in Figures Also, a window of zoomed-in data is presented in Figures (a), to better demonstrate the similarities in the signals. The power spectral density, PSD, of the two signals was computed for the cases described above. This generated three indistinguishable plots, one of which is shown in Figure 17. The PSD describes how the power (or variance) of a time series is distributed with frequency. Mathematically, it is defined as the Fourier Transform of the autocorrelation sequence of the time series. The PSD was estimated using the modified covariance method. This method fits an autoregressive (AR) model to the signal by minimizing the forward and backward prediction errors in the least-squares sense. It is clear from the plots presented in Figures and from the calculated cross-correlation coefficient that the signals are very similar, despite differences in mounting methods and differences in the type of accelerometers used.

19 From Top to Bottom: Figure 14, Figure 14a

20 From Top to Bottom: Figure 15, Figure 15a

21 From Top to Bottom: Figure 16, Figure 16a

22 PARI UC Berkeley Figure 17 Results The experimental motes worked well, considering they were not specifically designed for use in an extreme environment, such as a controlled blasting experiment. This experiment provided much insight into the inadequacies of the motes. Both the motherboard and the secondary sensor board are dependent on a consistent power source. It is known that battery operation is temperature dependent; the cold, wet weather of the Port of Tokachi may have played a role in the motes behaving inconsistently. Also, these experimental motes are still susceptible to corruption due to frequencies of similar bandwidth to that of the mote s radio. From this and

23 subsequent experiments at UC Berkeley, we conclude that the motes behavior worsens while their antennas are parallel to the ground (horizontal position). This explains the relative success of motes in the improved ground zone, where all antennas were in a vertical position, to the ones in PARI s study area. Conclusion The motes behavior can be drastically improved with small packaging and hardware changes. A new model of these versatile sensing platforms, known as the micha, will be available shortly. The micha s will have more memory, a better processor, a much-improved radio, and a voltage regulator, which will decrease erratic behavior. Simple packaging changes, such as a tougher exterior for experiments where harsh conditions are expected, can drastically improve the utility of these devices. Also, having all antennas in a vertical position and above the ground will increase communication amongst the motes. These sensors have proven to measure data as accurately as conventional wired sensors and with further development, will provide versatile and robust wireless networked sensors for many engineering applications. References 1. htttp://tinyos.millennium.berkeley.edu