A View from the Bridge

1
A View from the Bridge Research Impact of
Bringing Together WSNs and Structural Health
Monitoring

• David E. Culler
• UC Berkeley
• NSF Bridge Condition Monitoring and
  Prognostication Workshop

2
GGB What Got Accomplished
SF (south)
Sausalito (north)
8 nodes
56 nodes _at_ 50-100 ft
246 ft
43 hops
1125 ft
4200 ft

• High-Fidelity wireless 2D x 2 accelerometer node
• earthquake 2G _at_ 2 mG, ambient 0.1G _at_ 30 uG
• High-frequency, low-jitter, time-synchronized
  sample-process-log
• Large-scale monitoring network in the Real World
• Routing, command dissemination, time synch
• Reliable bulk data transport on very deep network
• Off-line Modal Analysis
• 2 PhDs CS Sukun Kim, CE Shamim Pakzad

3
Bottom Line Spatio-Temporal Analysis

• First (and only?) large-scale, time-correlated,
  high-frequency wireless physical monitoring

4
GGB What didnt?

• Any deep epiphany on the design, maintenance, or
  modifications of the Golden Gate Bridge
• Continuous, near real time, long term,
  high-fidelity, vibration analysis service
• On-going fatigue, wear-and-tear, alteration,
  traffic,
• Intermittent critical events
• Earthquakes, Storms, World Series,
• Not even close
• But it is now a different world than in 2003

5
Today

• Live Monitoring of the Golden Gate Bridge
• Impact on WSN research
• High speed, high fidelity sampling and logging
• VERY deep networks
• Routing, Dissemination, Time Synchronization
• Reliable, Efficient bulk data transport
• Innumerable Real World Deployment Challenges
• Impact on Civil Engineering Research
• Systems error analysis
• Modal analyses
• Challenges on the Horizon
• In-network Analysis
• Analysis of Ambient Structures
• A Meaningful Role in NSFs World

6
Wireless Sensor Networks
SF (south)
Sausalito (north)
500 ft
1125 ft
4200 ft
246 ft

• Embedded local data acquisition, storage, and
  processing
• Radio broadcast (e.g, IEEE 802.15.4)
• Selective retransmission
• Everything is built from that
• Routing, Collection, Dissemination, Time
  Synchronization,

7
The Experimental Systems Challenge

• Computer Science
• Electrical Engineering
• Mechanical Engineering
• Civil Engineering

8
GGB Vibration Mote

• MicaZ 802.15.4 mote
• 2D Mems Accel for earthquakes
• 1 mg _at_ 25 Hz
• 2 x 1D SiDe high fidelity Accel for ambient (10
  ug)
• Multichannel 16bit ADC
• Low-pass 1G adjust
• Thermometer for correction
• Regulators,

ADXL 202E
Silicon Designs 1221L

• Signal processing on-the-fly averaging
• Calibration for manufacturing variation and
  temperature

9
Environment
Fog Strong and salty wind Rapidly changing ...
high and scary
10
Node Physical Attachment
Battery (4 X 6V Lantern Battery)
Bi-directional Patch Antenna - long linear
structure
Node (Mote Accelerometer Board) - bits of
precision in orientation?
11
Node Changing Physical Attachment
Zip-tie Antenna stay - blowin the wind
Rapid Rusting of C-clamp
12
Node Going Vertical
Antenna Cable To Base Station
Signal Splitter
13
Base Station
Laptop
Students At Work
14
Installation
Crawling and Installing
Hard Hat
Done!
Strong Wind
Harness
However
Ouch
Sharp Edge
15
Holistic Error Analysis Design
Quantization
Sampling
xx (µG/vHz)
f range 2 (n-k)
Sensor
ADC

• Lesson from Computational Science roughly equal
  contribution to the overall error

16
Protocol for Unprecedented Data Collection and
Analysis?

• Experimental protocol
• Not just network protocol
• Trust, Accuracy, Establishing Limits
• Price gtgt Cost when you want it to work
• Verify what we build (or buy) and what we do with
  it

17
Inventing Protocol Sensor Calibration
18
Verification of Jitter (6.67KHz)
10µs
0µs
0µs
10µs
Time Series
Histogram

• Jitter is within 10µs (6.67), 0.2 at 200Hz
• Real Time communication vs Real Time Sampling
• gt 1 KHz sampling, down sampled to 50 Hz with 25
  Hz low pass

19
Deployment at Footbridge
Using Sentri (structural health monitoring
toolkit)
20
Plots of calibrated data
21
Modal Properties
Match with SAP bridge model
First Vertical Mode of Vibration
22
Second Vertical Mode of Vibration
0.94
1.00
0.68
0.33
0.41
Frequency 1.78 Hz Damping Ratio 1
23
The Network is the Sensor

• The data at the nodes is only as good as what you
  get out
• Deploy, test, and validate network
• Establish time reference
• Issue start at time T for L
• Time-correlated high-frequency sampling into
  flash log
• Collect all the data over 46 hops!
• Analyze, analyze, analyze,
• Repeat

24
WSN research that worked!

• Mint Route
• Self-organized tree-based topology formation and
  routing
• Trickle Drip
• Density aware, adaptive topology free,
  dissemination
• Flooding Time Synchronization Protocol
• Dissemination tree establishes time reference
• Learn local correction factor for offset and
  drift
• Pushed far far beyond their development, testing,
  or evaluation

25
Self-Organized Routing - nutshell
Retrans. rate lt 1/density Neighbor table exceeds
RAM Dissem. rate lt change
0
26
Reliable, Efficient Bulk Transfer over Many Many
Lossly Hops
SF (south)
Sausalito (north)
500 ft
1125 ft
4200 ft
246 ft

• Lossy links 90 PRR is good!
• Hop-by-hop vs end-to-end error detection and
  retransmission?
• Losses are bursty (correlated)
• Interference internal and external
• Costly to turn the link around
• Acks cause interference too!

27
Intra-flow Interference

• Scalar multihop forwarding
• Streaming
• Straw simple, static, source rate control
• 1 hop gt full, 2 hop gt ½, 3 hops gt 1/3

28
Straw Reliability Protocol
2, 4, 5
2
4
5
4, 9
4, 9
4
9
29
Straw Data Collection at GGB
Data is collected reliably over a 46-hop network,
with a bandwidth of 441B/s at the 46th hop
30
Experience Guiding Problem Focus

• Implemented, tested, deployed a very simple
  solution
• Also built on established work wherever possible
• Sophisticated Enhancements focused on improvement
• Interference and loss is not uniform nor constant
• Adaptive protocols detect loss hop-by-hop and
  adjust rate

31
Flush Adaptive Rate Control
1. At each node, Flush attempts to send as fast
as possible without causing interference at the
next hop along the flow 2. A nodes sending rate
cannot exceed the sending rate of its successor
8
6
5
4
All boils down to simple queue threshold
8
7
6
5
4
3
32
Implementation

• 16-deep Rate-limited Queue
• Enforces departure delay D(i)
• When a node becomes congested (depth 5), it
  doubles the delay advertised to its descendants
• But continues to drain its own queue with the
  smaller delay until it is no longer congested

33
Packet Throughput at Transfer Phase
Effective Goodput (pkt/s)
Effective goodput during the transfer
phase Flush provides comparable goodput as a
lower loss rate which reduces the time spent in
the expensive acknowledgment phase, which
increases the effective bandwidth
34
Packet Rate over Time for a Source

• Source is 7 hops away, Data is smoothed by
  averaging 16 values
• Flush approximates the best fixed rate with the
  least variance

35
Maximum Queue Occupancy across All Nodes for Each
Packet

• Flush exhibits more stable queue occupancies
  than Flush-e2e

36
Scalability Test
Effective bandwidth from the real-world
scalability test where 79 nodes formed 48 hop
network Flush closely tracks or exceeds the best
possible fixed rate across at all hop distances
that we tested
Effective Bandwidth (B/s)
37
Real Life Data
38
Really Real Real Life Data
39
How dense is enough?
40
Deeper SHM Challenges
Space
Accel(t)
Time

• Time Series and Spectral Analysis is Local
• If you know what you are looking for you can do
  it in place
• We KNOW this for machines, pumps,
• Structural Analyses (MIMO, ARX) deal with the
  spatial dimension
• How do you do this in place when there is very
  little communication bandwidth ???

41
The Much Deeper Issue

• If you dont know what you are looking for, you
  are not likely to find it
• Ambient Analysis (SHM) is not an easy fit with
  academic science an engineering
• Shake table for controlled phenomenon
• Destruction for critical response
• Health takes time, is unscientific, and is mostly
  uneventful
• Realism and Controlled Study are at odds
• The bridge is not going to fall down while you
  are measuring it.
• It is not going to age or fatigue either
• Maybe you will confirm design parameters
• Just maybe you will observe changes due to
  repairs, upgrades, or accidents
• But

42
in the Real World
43
InCheon Cable Stayed Bridge Measurement
Objective 1) Modal Tests at various
construction Stages 2) Wind Pressure
Distribution on the Bridge
Unit m
44
Courtesy of Samsung Construction
45
(No Transcript)
46
(No Transcript)
47
Free Standing Pylon
48
Estimated Mode Shapes
49
(No Transcript)
50
(No Transcript)
51
(No Transcript)
52
(No Transcript)
53
(No Transcript)
54
(No Transcript)
55
Estimated Mode Shapes
56
Towards a Structures Design Cycle

• In the Internet of Every Thing age
• Standard IPv6 over reliable LoWPANs in industry
  and research
• Epic platform expert modules, application PCB
  glue
• TinyOS 2 No longer research, so it works

57
Thanks

• Sukun Kim, Shamim Pakzad, James Demmel, Gregory
  Fenves
• NSF, CITRIS, Golden Gate Bridge Authority
• For more Design and Implementation of Scalable
  Wireless Sensor Network for Structural
  Monitoring, Shamim N. Pakzad, Gregory L. Fenves,
  Sukun Kim, and David E. Culler, In ASCE Journal
  of Infrastructure Engineering, March 2008, Volume
  14, Issue 1, pp. 89-101.