Semantic sewer pipe failure detection:Linked data approaches for discovering events
CSIRO LAND AND WATER
Jonathan Yu | Research software engineerEnvironmental Information Systems, CLW Highett21 October 2013
Performing event detection over real-time sensor data using ontology-driven approaches | Jonathan Yu
Undetected sewer rising mains pipe failures...
Direct costs: water service providers ($ mil. per event)
Indirect costs: social, environmental ($10k - $1 mil. per event)
2 |
Performing event detection over real-time sensor data using ontology-driven approaches | Jonathan Yu
Sewer rising mains case study
3 |
0 2000 4000 6000 8000 10000 12000 14000 160000
20
40
60
80
100
120
140
160
Time (mins)
Flow
rate
(l/s
)
Pipe failure event = flow > 100 l/s
Example event: Flow rate > 100 l/s
Semantic Sensor Network
SSN Extensions - quantity values
Pipe features, observations, sensors defs
Water domain – flow, pressure, units of measure
RDF Triple Store contains ontologies
Event-detection
Pipe domain rules – MCA, risk levels, PVC pipes, asbestos concrete pipes
• These ontologies provide semantics and constructs to describe sensors and observations generally
• General model extended with domain semantics and knowledge
• Allows definitions to be captured explicitly and used consistently.
• Extensible – able to capture more domain knowledge/rules
• E.g. PVC pipe feature definitions and domain rules
SensorMiddleware
(GSN)
Web server
RDFTriple Store
Sensor Network
Real-time sensordata
SPARQL
Observation and
NotificationREST Web
service
Internal network
Event Notification Interaction
• Event rules deployed in GSN send notifications to web service
• Web service adds metadata to notification and sends to RDF Triple Store
• RDF Triple Store persists the sensor observations and event notifications like a semantic knowledge base
Event Rules
VirtualSensors
Web Server Web Server
Event Detection
Linked Data
RDFTriple Store
SPARQL
Internal networkPublic accessible network
Event Detection Linked Data API
• List notifications, list sensor observations, view semantic descriptions of pipes, pumps, observed properties
• allows users to browse contents of a RDF triple store via standard web browser• configured to view sensor observations, event notifications, semantic definitions,
domain knowledge base • Also enables software clients to retrieve JSON/XML/RDF/TXT formats of the same
information for mashups and data fusion activities
Web Server Web Server
Event Detection
Linked Data
Viz
RDFTriple Store
SPARQL
Internal networkPublic accessible network
Visualization client
• Example of a visualization client querying the RDF triple store for sensor observations and event notifications
• Identifiers from the RDF triple store resolve to metadata and semantic definitions delivered via the Event Detection Linked Data API
SensorMiddleware
(GSN)
Web Server Web Server
Event Detection
Linked Data
Viz
RDFTriple Store
Sensor Network
Real-time sensordata
SPARQL
Observation and NotificationREST Web
service
Internal networkPublic accessible network
Reverse Proxy/ auth
EventDashboard
Overall architecture schematic
Event Rules
VirtualSensors
Performing event detection over real-time sensor data using ontology-driven approaches | Jonathan Yu9 |
Performing event detection over real-time sensor data using ontology-driven approaches | Jonathan Yu
Summary
• Domain & event ontologies: • Defining and capturing sewer pipe event descriptions extending SSN ontology
and
• Linked Data APIs and RESTful services• Publish and discovery of sewer pipe event notifications and observations
• Demo visualization client
• Preliminary work to demo real-time events can be combined with domain knowledge for context sensitive event detection using ontologies
10 |
Performing event detection over real-time sensor data using ontology-driven approaches | Jonathan Yu
Event Detection Ontology def’s:Event Rule, Value Constraints, Units
11 |
Performing event detection over real-time sensor data using ontology-driven approaches | Jonathan Yu
Domain ontologies (uwda:) - Sensors
12 |
Performing event detection over real-time sensor data using ontology-driven approaches | Jonathan Yu
Sewer rising mains case study
13 |
0 2000 4000 6000 8000 10000 12000 14000 160000
20
40
60
80
100
120
140
160
Time (mins)
Flow
rate
(l/s
)
Pipe failure event = flow > 100 l/s
Example event: Flow rate > 100 l/s
Performing event detection over real-time sensor data using ontology-driven approaches | Jonathan Yu
Event rule definition instances
Rule ID Observed property
Value constraint
Feature of interest
Observed By (Sensor)
1 Flow
2 Flow > 100 l/s
3 Flow > 100 l/s Pipe A
4 Flow > 100 l/s Pipe Sensor A-1
5 Flow > 100 l/s Pipe A Pipe Sensor A-1
14 |
Performing event detection over real-time sensor data using ontology-driven approaches | Jonathan Yu
Fusing real-time events with domain knowledge
15 |
KnowledgeBase
Sensor NetworkReal-time data
Event of Interest
Query knowledge base(domain knowledge)
Notifications
e.g. Populate knowledge base with parameterised historical pipe failure data.
Infer likelihood of pipe failure based on physical attributes and known operating environment
Performing event detection over real-time sensor data using ontology-driven approaches | Jonathan Yu
Modelling the feature of interest – pipe materials
16 |
Performing event detection over real-time sensor data using ontology-driven approaches | Jonathan Yu
Event detections using dynamic and static info
18 |
> 200 PSI
+
Pipe material is PVCand
Risk level of pipe is A (good)
(Dynamic)
(Static)
Notification:
Location: Pipe XRisk of burst: LOW
Performing event detection over real-time sensor data using ontology-driven approaches | Jonathan Yu
Event detections using dynamic and static info
19 |
> 200 PSI
+
Pipe material is PVCand
Risk level of pipe is E (bad)
(Dynamic)
(Static)
Notification:
Location: Pipe XRisk of burst: HIGH
Performing event detection over real-time sensor data using ontology-driven approaches | Jonathan Yu20 |
Define event constraint
Map sensorsInitialise sensor
network and sensor middleware
Deploy event constraint
Query Notifications
Visualise Notifications
Integrate with Notification
systems
Event Dashboard Notification clientsSensor Infrastructure
Performing event detection over real-time sensor data using ontology-driven approaches | Jonathan Yu
Future work
Notification handling• Messaging queue systems• Attaching metadata based on
event rule semantics
More complex events• Event semantics • Incorporate processing-filters
User studies to evaluate the user interface
Deployments on actual sensor networks
21 |
A
BEventSmoothing
function
Email / SMS
Database
Execute workflow
Existing alert systems
Ontology-enabledUser Interface
Sensor Network
Performing event detection over real-time sensor data using ontology-driven approaches | Jonathan Yu
Questions?
22 |
Land and WaterScott GouldResearch Projects Officert +61 3 9252 6103e [email protected] www.csiro.au/clw
ICT CentreKerry TaylorPrincipal Research Scientistt +61 2 6216 7038e [email protected] www.csiro.au/ict
Land and WaterDonavan MarneyResearch team leadert +61 3 9252 6585e [email protected] www.csiro.au/clw
LAND AND WATER
Thank youLand and WaterJonathan YuResearch Software Engineert +61 3 9252 6440e [email protected] www.csiro.au/clw
Land and WaterPaul DavisResearch Scientistt +61 3 9252 6310e [email protected] www.csiro.au/clw
Land and WaterBrad ShermanResearch Scientist
e [email protected] www.csiro.au/clw