Infrastructure for Smart Cities Bridging research and production Sandy Taylor | Software Engineer Ben Barnes | Software Engineer September 2016 www.data61.csiro.au
Section 1: Background The need for a solution
Bridges Are Needy The Sydney Harbour Bridge needs: • $15,000,000 maintenance p/a • Almost 100 full-time workers • 15 full-time painters Harbour Bridge Paint • 30,000 L of paint per coat Where do we come in? • Provide early warning of maintenance needs for the road deck 3 | Infrastructure for Smart Cities | Ben Barnes, Sandy Taylor
The Road Deck • 800 jack arches support the road deck • Aging infrastructure, needs to be replaced eventually • Need to extend the life of the deck without a significant increase in maintenance costs 4 | Infrastructure for Smart Cities | Ben Barnes, Sandy Taylor
Inspection & Maintenance • Two year visual inspection cycle • Inspection manual is exhaustive • Similar for metalwork, paintwork … • Access is difficult. Crawl spaces, gantry crane + scaffold required in places 5 | Infrastructure for Smart Cities | Ben Barnes, Sandy Taylor
Inspection & Maintenance • Road surface cracks often the first indication of damage • But too late, cracks are a result of underlying structural damage • Need to detect damage (fractures) in jack arch before it becomes serious 6 | Infrastructure for Smart Cities | Ben Barnes, Sandy Taylor
Failure Is Costly Mortal risk, but also economic: • A crash in March shut down three lanes for around two hours • Estimated cost to the economy: $16,000,000 Jan 2016 Sept 2016 Mar 2016 7 | Infrastructure for Smart Cities | Ben Barnes, Sandy Taylor
Section 2: The Solution How did we get there?
Can We Do Better With Data? • Continuously monitor 800 structural joints? • Provide early warning of maintenance needs? • Augment and direct inspection routines? • Extend service life of bridge deck? • Without a significant increase in maintenance costs? 9 | Infrastructure for Smart Cities | Ben Barnes, Sandy Taylor
Structural Health Monitoring 2400 sensors installed I I I I I I 10 | Infrastructure for Smart Cities | Ben Barnes, Sandy Taylor
Structural Health Monitoring • A problem in three parts: Presentation, Data Data Alerting & Collection Processing Aggregation 11 | Infrastructure for Smart Cities | Ben Barnes, Sandy Taylor
Data Challenges & Constraints • A production, real time system that also caters for future research and analysis techniques • Large data volumes (100s of Gb per day) • Limited uplink capacity from site – reduce network storage and load (store and forward approach) • Real time decisions and visualisation – immediate notification of structural defects • Timeframe: milliseconds to decades 12 | Infrastructure for Smart Cities | Ben Barnes, Sandy Taylor
Data Collection Server ADSL Modem Gateway Gateway Sensor Sensor Sensor Sensor … … Node Node Node Node 13 | Infrastructure for Smart Cities | Ben Barnes, Sandy Taylor
Designed For Research • The SHM field changes rapidly • Structures don’t (few chances to modify the system) • Long-running project, have to adapt Data Processor Data Data Storage Acquisition Processor Data Processor 14 | Infrastructure for Smart Cities | Ben Barnes, Sandy Taylor
Research To Production Collaboration MATLAB/ C++ Data Packaged Live data to Raw data with R Processor by Ops frontend Engineering 15 | Infrastructure for Smart Cities | Ben Barnes, Sandy Taylor
Data Analysis Methods Several methods are either in the process of being developed or have been deployed on the bridge: • Defect/Damage Detection • Comparative Vibration • SVM classifiers • Time Series Analysis • Population clustering analysis • Condition Assessment • Multi-scale (localised vs global) • Operational modal analysis • Classifiers derived from dynamic FEA models • Vehicle Tracking • Tracking and classification • Impact force reconstruction from sensor data • Predictive Analysis • Predicting relative likelihood of failure based on multiple factors • Data fusion technique We will cover a few of these damage detection methods 16 | Infrastructure for Smart Cities | Ben Barnes, Sandy Taylor
First Step -Detecting Events • Only keep data based on traffic events • Threshold based detection • Reduces data volume to 1/30 th • Better signal to noise ratio, more consistent samples • Filter out smaller vehicles 17 | Infrastructure for Smart Cities | Ben Barnes, Sandy Taylor
Heuristic Approach • Intuition: structural member should behave as a rigid body • Sensors should show same response to disturbance • Cracks allow sections to move independently • Can use cross-correlation to measure similarity • Scaled based on vibration energy: more vibration, more damage Joint 5 Joint 1 Joint 2 Joint 3 Joint 4 Joint 6 18 | Infrastructure for Smart Cities | Ben Barnes, Sandy Taylor
Heuristic Approach Problems • Vibration varies significantly over the bridge • Proof-of- concept nodes weren’t representative of this • Could tune parameters per sensor, but not a scalable solution • There must be a better way! A C B 19 | Infrastructure for Smart Cities | Ben Barnes, Sandy Taylor
SVM & FFT traffic “events” • Unsupervised single class 0.5 Support Vector Machine (SVM) Joint 5 Joint 4 • Model-free, not tied to Normal 0 structural details Decision values • Looks for any change, not just -0.5 damage -1 • Training is a challenge Degraded -1.5 • Still some threshold adjustment -2 0 50 100 150 200 250 300 350 400 450 Test event index 20 | Infrastructure for Smart Cities | Ben Barnes, Sandy Taylor
Dynamic Event Threshold • Looks for a change in variance • Threshold is now a ratio: event amplitude noise amplitude • Event detection is decoupled from units of measurement • No more manual tweaking 21 | Infrastructure for Smart Cities | Ben Barnes, Sandy Taylor
Sensor Faults • Sudden, unexplained health warnings • How do we reduce false positives? Consider environmental factors: • Sensor detachment • Water ingress • Temperature sensitivity Missing potting Build in measures of confidence 22 | Infrastructure for Smart Cities | Ben Barnes, Sandy Taylor
Presentation & Aggregation Visual Presentation • How to clearly convey status? • Historical analysis & aggregation • Simple up front, more information if desired 23 | Infrastructure for Smart Cities | Ben Barnes, Sandy Taylor
Grouped Aggregation & Alerting Grouped Aggregation Alerting Bay (Group) Overall Jack Arch Health Heuristic H M M M ML Decision Score L L L Values Sensors S S S 1 2 3 Structural Component System Component Virtual Sensor 24 | Infrastructure for Smart Cities | Ben Barnes, Sandy Taylor
Validation Of Our Method • Whilst there hasn’t been any actual damage: 25 | Infrastructure for Smart Cities | Ben Barnes, Sandy Taylor
Section 3: Future work, takeaways
What have we learned? • Encourage collaboration between research and engineering throughout the project • Divide the problem into collection, processing and presentation • Separate logical processing topology from physical network • Build in confidence metrics from the beginning 27 | Infrastructure for Smart Cities | Ben Barnes, Sandy Taylor
Future Development • Decouple processing from platform – node, gateway, cloud • Improve confidence estimates • Strain gauges - fatigue / life cycle • FEA model updating 28 | Infrastructure for Smart Cities | Ben Barnes, Sandy Taylor
Recommend
More recommend
