1. IPSN 2012 Ted Tsung-Te Lai, Wei-Ju Chen, Kuei-Han Li, Polly Huang, Hao-Hua Chu NSLab study group 2012/03/26 Reporter: Yuting 1

2.  PipeProbe: A Mobile Sensor Droplet for Mapping Hidden Pipeline  Jeffery reported it at study group last year ◦ http://nslab.ee.ntu.edu.tw/NetworkSeminar/index. php?action=schedule&year=2010_Fall&pattern= http://nslab.ee.ntu.edu.tw/NetworkSeminar/index. php?action=schedule&year=2010_Fall&pattern=  http://mll.csie.ntu.edu.tw/papers/PipeProbe_ Sensys10.pdf http://mll.csie.ntu.edu.tw/papers/PipeProbe_ Sensys10.pdf  http://mll.csie.ntu.edu.tw/papers/PipeProbe_ Sensys10.pptx (Best Presentation Award) http://mll.csie.ntu.edu.tw/papers/PipeProbe_ Sensys10.pptx 2

3.  Sensor network data is wirelessly transmitted to nearby gateway nodes  The gateway is a (laptop) computer wired to a Kmote node 3

4.  Abstract  Introduction  System Overview, Assumptions and Limitations  Hardware Design  System Design  Experiment  Discussion  Conclusion 4

5.  Abstract  Introduction  System Overview, Assumptions and Limitations  Hardware Design  System Design  Experiment  Discussion  Conclusion 5

6.  Autonomous sensor deployment ◦ For pipeline monitoring  Centralized repository at pipeline’s source ◦ Automatically releasing nodes  Placement: ◦ Nodes will latch itself in pipeline  Replacement: ◦ Source will send new nodes to replace failed one, ex: low battery level; experiences a fault 6

7.  Evaluated on testbed  Advantage: ◦ Less sensor nodes to cover a sensing area ◦ High data collection rate ◦ Recover from the network disconnection 7

8.  Abstract  Introduction  System Overview, Assumptions and Limitations  Hardware Design  System Design  Experiment  Discussion  Conclusion 8

9.  Flow assurance ◦ A major safety concern ◦ Ex: clean and uncontaminated water  Traditional method: ◦ Manually placing, but it’s hard and waste time  TriopusNet ◦ Automated ◦ Scalable ◦ Human effort strictly needed only at the start to deposit mobile sensors 9

10.  Sensor deployment algorithm depends on: ◦ Sensing coverage ◦ Network connectivity ◦ Deployment location  Upon arrival at its deployment location, a traveling sensor activates its latching mechanism 10

11.  Upon detection of low battery level (or a fault), the sensor node retracts its mechanical arms to detach itself ◦ Flow in the pipes carries it out ◦ System releases a fresh sensor node and runs the sensor replacement algorithm ◦ And adjust the locations of existing ones 11

12.  Automates sensor deployment and replacement by leveraging natural water propulsion to carry sensor nodes throughout pipes  Real prototype and pipeline testbed show that this quality deployment using no more sensor nodes  Successfully replaced a battery-depleted sensor node with a fresh sensor node while recovering data collection rate from the departure of a battery-depleted sensor node 12

13.  Abstract  Introduction  System Overview, Assumptions and Limitations  Hardware Design  System Design  Experiment  Discussion  Conclusion 13

14.  Pipelines interconnect a set of vertical and horizontal pipes, starting with a single water inlet and ending at multiple water outlets  Pipelines form a virtual tree!  The inlet also serves as the storage point where sensor nodes are deposited into a dispatch queue at the start of deployment 14

15.  A significant size reduction in 2 nd type - 6 cm in diameter ◦ May still get stuck in some pipes  (a~d): gyro, water pressure sensors, relays, Kmote (TelosB-like w/o USB) ◦ In water, sonar and light are better than radio -> they leave the choice in future  One customized motor drives three arms in 2 nd type 15

16.  Preparation Step  Sensor Deployment Step  Sensor Latching Step  Sensor Replacement Step 16

17.  Pipeline spatial topology must be measured a-priori as an input for automated sensor deployment ◦ PipeProbe system (their previous work)  Inlet must be filled with sensor nodes  Faucets in the pipeline are turned on ◦ Manually or automatically (by installing a remote- control actuation device)  One-time manual effort 17

18.  Runs the sensor deployment algorithm prior to releasing  Then sends the “release” message including the deployment position, to the head sensor node 18

19.  Sensor node continuously computes its current location as it travels  When the node approaches its deployment position: ◦ Latch itself ◦ Report the completion ◦ Triopusnet releases the next (repeat step2) 19

20.  Some sensor node may report low-battery to the system ◦ Detach itself, carried out by the water ◦ Triopusnet releases fresh one 20

21.  Must be installed prior to any sensor node deployment inside the pipelines  Must have wireless communication with at least one in-pipe sensor node  Must also have a network connection to a computer for: ◦ Remote control ◦ Data logging ◦ Automated sensor deployment and replacement algorithms 21

22.  Abstract  Introduction  System Overview, Assumptions and Limitations  Hardware Design  System Design  Experiment  Discussion  Conclusion 22

23.  Linear actuator controls a mechanical arm ◦ Push: SW1&4, pull: SW2&3  Motor calibration was achieved by adding a spiral gear that connects and pushes three separate gears 23

24.  Abstract  Introduction  System Overview, Assumptions and Limitations  Hardware Design  System Design  Experiment  Discussion  Conclusion 24

25.  Placing nodes close to the releasing point early may result in blockage ◦ Transforms the layout of the pipelines into a tree ◦ Subsequently runs a post-order traversal of the tree ◦ Deploying nodes in the above sequence will: assure covering all pipes without blocking others 25

26.  Before sensor nodes can be released, the sensor deployment algorithm computes first the coarse- grain positions: ◦ The pipe segment ◦ The approximate latching point  Assume a simple coverage function (but not limited) ◦ Circle with radius R ◦ “Subtracting 2*R distance from the most recently deployed sensor node in segment S gives the position of the new one” ◦ “If segment S is not long enough to accommodate the new sensor node, the new sensor node is placed in the next segment” 26

27.  The sensor movement algorithm computes first the flow paths from the inlet to each outlet  Then selects a path intersecting the pipe segment the node is positioned to 27

28.  There are both vertical and horizontal pipes  Adopts the pipeline localization technique from the PipeProbe system [4]  Sensor node tracks its location by: ◦ Counting the number of turns with:  pressure and gyroscope sensors ◦ Segment offset distance from the last run:  Vertical: the change in water pressure  Horizontal: multiplying velocity by traveled time  Buoyancy? -> the sensor node was designed with its weight density equal to the water density 28

29.  Turning on radio after latching and measures the packet received rate for the link quality  Upon detecting a low packet received rate, the sensor node moves one increment closer to its downstream sensor node  Until a pre-defined link quality threshold is met, sends a “latching completion” packet ◦ May be tricky to ensure the first sensor node of an intermediate segment is connected to the sensor nodes of all downstream segments  May moves into one of the unreachable downstream segment  Repeats until full sensing and network coverage 29

30.  Collection tree protocol (CTP) implemented in TinyOS 2.1  Use anycast (provided by CTP) to multiple sinks(gateway nodes) in order to balance traffic load 30

31.  Battery-depleted node (determined simply by voltage) ◦ Informs the downstream gateway ◦ Faucet can be turned on ◦ Downstream nodes are also flushed out ◦ Fishing net is inserted at the ends of pipelines  Good nodes ◦ Each upstream node repeats: detachment, movement, localization, reattachment ◦ Until the uncovered area reaches the root location ◦ System then releases fresh nodes  With a smaller prototype in the future, it will be easier and save more energy! 31

32.  Abstract  Introduction  System Overview, Assumptions and Limitations  Hardware Design  System Design  Experiment  Discussion  Conclusion 32

33.  6 “transparent” pipe tubes (10 cm in diameter)  2 water valves control the volumetric flow rate on each flow path 33

34.  And: time to replacement energy consumption (2 cases) 34

35.  System parameter: ◦ PRR threshold = 95% ◦ Water flow velocity = 12.5 cm/sec ◦ Each node’s sensing range R >= radio range  4 scenarios * 5 runs/scenario = 20 test runs  Data was logged during both: ◦ node deployment and data collection  Replacement performance is measured in scenario #4 ◦ 20 test runs of node replacement 35

36.  Static deployment: a good baseline for performance comparison ◦ Nodes are 90 cm apart ( average radio range between two sensor nodes in a straight pipe ) ◦ Might have better DCR, but more redundant nodes (DCR: Dada Collection Rate) 36

37.  The radio range can reach up to 170 cm for nodes placed in different tubes  Benefits of using online deployment 37

38.  Indicate whether a network is well connected  80% of the sensor nodes show a data collection rate exceeding 99% ◦ And all are above 86.5%  Each sensor node sent 1000 data packets to a gateway node 38

39.  18,20,20,30 location estimates for scenario 1~4, respectively  Overall median: 7.14cm  90% of the errors are less than 20.45 cm  Sufficient for most pipeline applications, ex: pinpointing the location of pipe leakage 39

40.  Time to manually turn on/off faucets is not included here  If the flow velocity is set at 12.5 cm/sec, the average time to deploy nodes is less than 2.5 minutes 40

41.  Primary energy consumer in the sensor node is in the motor and relays that drive the three mechanical arms ◦ (Note: energy consumption: motors > radio > MCU)  A single act of latching: ◦ 1.01W * 2 sec < 1% * 600mAh = 2.16J  # of latching: ◦ average is: 2.35; 90% of nodes required less than 5 41

42.  DCR before a node reported low-battery level and after the node was replaced: ◦ 0.989 and 0.984 respectively ◦ Small difference, effective! ◦ [YT] But which node are they use? ( last or 2 nd last )  DCR without automated replacement: 0.81 42

43.  Depends on the location of the node and the size of the network 43

44.  Abstract  Introduction  System Overview, Assumptions and Limitations  Hardware Design  System Design  Experiment  Discussion  Conclusion 44

45.  Several assumptions and limitations require extensions before practical deployment ◦ Node is too big to be flushed out independently  [YT] If the size is reduced, there may be extra works on gryo measurement ◦ Node placement requires controlling or obtaining the direction of the water flow in the pipes  Automatical method: attaching a sensor-trigger node to activate/deactivate the infrared sensor in each automatic faucet 45

46.  Nodes are equipped with a water flow sensor  Can infer the current flow path  May Releases new nodes whose destinations must match the current water flow path 46

47.  Abstract  Introduction  System Overview, Assumptions and Limitations  Hardware Design  System Design  Experiment  Discussion  Conclusion 47

48.  Pipeline monitoring  Autonomous sensor deployment  Scales down human effort  Real pipeline testbed  No more nodes than non-automated static sensor deployment  Restore sensing and network coverage from the departure of a battery-depleted node 48

49.  Strength ◦ Save lots of nodes using online deployment method ◦ Successfully replaced a battery-depleted sensor node with a fresh one  Weakness ◦ Not adaptive with varying water flow rate now ◦ No automatically water faucet now ◦ Will the mechanical arms be reliable under strong water flow? ◦ For high traffic load, the deployment performance may not be as good as now ◦ Evaluation for DCR in replacement is not clearly enough 49

50. Thanks for your listening! 50