A Mobile Sensor Droplet for Mapping Hidden Pipeline

1. PipeProbe Tsung-te (Ted) Lai Yu-han (Tiffany) Chen Polly Huang Hao-hua Chu National Taiwan University A Mobile Sensor Droplet for Mapping Hidden Pipeline

2. Outline • Motivation • Layout mapping algorithm • Design iterations • Testbed and evaluation • Limitations • Related work • Future work

3. Water scarcity

4. Residential water usage 35 liters/person/day Toilet Clothes Washer Shower Faucet Leakage Other Domestic Dish Washer Bath Source: Residential End Uses of Water, AWWA Research Foundation

5. Residential water usage x18 Shower Faucet

6. Pipes are often hidden behind walls or underneath floors hidden pipes Motivation

7. Leakage often occurs at the joints of tubes Motivation leaking leaking

8. PipeProbe system ‧Map 3D spatial topology of water pipelines ‧Mobile sensing approach ‧Leverage natural water flow for mobility

9. ECo wireless sensor mote (Pai Chou, UC Irvine) ‧ Low-power ‧ 13mm(L) x 11mm(W) x 7mm(H), 3 grams ‧ Radio ‧ 3-axis accelerometer

10. Pressure sensor ‧0 – 14 bars, resolution: mbar ‧< 5uA operating current

11. Gyroscope ‧yaw (z) axis rotation angle ‧ ±300 deg/second

12. Mapped topology 1. Drop PipeProbe into the main water inlet 2. Open a water outlet 3. Collect sensor readings from the pressure and gyro sensors 4. Analyze the pressure and rotation angle readings

13. Open another water outlet to map out the fork path

14. Gyroscope graph Pressure graph

15. Outline • Motivation • Layout mapping algorithm • Design iterations • Testbed and evaluation • Limitations • Related work • Future work

16. Vertical tube Horizontal layer • Problem formulation

17. Starting position Ending position • Problem formulation

18. Starting position Ending position • Problem formulation

19. Assumptions • Diameter of pipes is uniform • Turns are 90-degree

20. Assumptions • Diameter of pipes are uniform • Turns are 90-degree

21. Layout mapping algorithm (2) Conquer

22. Layout mapping algorithm (2) Conquer

23. Divide phase • partition pipes into vertical tubes and horizontal layers of tubes • use pressure graph to detect vertical-to-horizontal or horizontal-to-vertical turns. Time

24. Layout mapping algorithm (2) Conquer

25. Conquer phase • Estimate vertical tube length • Based on pressure principle to estimate vertical tube length ∆P = ∆height

26. Layout mapping algorithm (2) Conquer

27. Conquer phase • Map horizontal pipe layout • (1) Detect horizontal turns linking horizontal pipes • based on a change in rotation angles • (2) Estimate horizontal tube length Time

28. Conquer phase • Map horizontal pipe layout • (1) Detect horizontal turns linking horizontal pipes • based on a change in rotation angles • (2) Estimate horizontal tube length • -∆ length = time * water flow velocity • - water flow velocity (constant) • = volume of water outflow / pipe cross-section area • ~ capsule moving velocity ∆ length = ∆t * v Time ∆t = t2 –t1 t1 t2

29. Layout mapping algorithm (2) Conquer

30. Merge phase • Link vertical pipes to start/end points of each horizontal pipe layout • Problem: Vertical-to-horizontal turn angle (θ) is non-deterministic θ 360 degrees of freedom

31. Merge phase • How to determine θ? Starting position Θ Ending position

32. Outline • Motivation • Layout mapping algorithm • Design iterations • Testbed and evaluation • Limitations • Related work • Future work

33. 1 Prototype PressureSensor Mote Design: pressure sensor + Eco mote in a round and flat capsule Problem: unstable flow velocity

34. 2 Prototype • Design: • spherical capsule • capsule flow velocity ≈ water velocity • added weight such that PipeProbe’s density ≈ water density • Problem: arbitrary rotation caused unreliable sensor reading

35. 3 Prototype PressureSensor Gyro Bottom Design: heavy bottom half pressure sensor on the top, gyro sensor flat on bottom Problem: arbitrary horizontal spinning caused high noisy gyro reading

36. 4 Final Prototype Tail-like Fin Design: tail-like fin aligns capsule’s heading to the water flow direction

37. Gyro graph Pressure graph • 1. Pressure sensor on top and gyro sensor vertical to ground • 2. Flow velocity ≈ water velocity 3. Flow straight

38. Outline • Motivation • Layout mapping algorithm • Design iterations • Testbed and evaluation • Limitations • Related work • Future work

39. Evaluation metric #1: length error Length error = actual pipe length – estimated pipe length = L1 – L2 Actual length: L1 Estimated length:L2

40. Evaluation metric #2: positional error Positional error (of the pipe turning point) = Euclidean distance between the actual and estimated positions estimated position (x2, y2, z2) error (x1, y1, z1) actual position

41. Experimental testbed

42. Water pipeline testbed

43. Control valves to produce different flow paths

44. Create a flow path

45. Testbed spatial layout (unit: cm) inlet outlet outlet

46. Experimental Procedure (12 test scenarios) 1 2 3 4 5 6 7 8 9 10 11 12 flow path

47. pipe probe (2010) Test 11 (flow path in red)

48. Test 11 (actual flow path) flow path

49. Test 11 (1st mapping trip) flow path estimates

50. Test 11 (2nd mapping trip) flow path estimates