There has been an explosion of data digitising our physical world – from cameras, environmental sensors and embedded devices, right down to the phones in our pockets. Which means that, now, companies have new ways to transform their businesses – both operationally, and through their products and services – by leveraging this data and applying fresh analytical techniques to make sense of it. But are they ready? The answer is “no” in most cases.
In this session, we’ll be discussing the challenges facing companies trying to embrace the Analytics of Things, and how Teradata has helped customers work through and turn those challenges to their advantage.
5. 5
‘Things’ are listeningSpeech, location,
places, music,
browsing, app data,
health, commute,
social network, device
logs
PEOPLE: Faces, gait,
demographics
ENVIRONMENT:
sound, weather, air
quality, traffic,
gunshots, speeders,
live video
ME: Speech,
location, places,
driving behavior
CAR:
operational &
performance
logs
Speech, location,
places, music,
browsing, app data,
health, commute,
social network, device
logsPEOPLE: Faces, gait,
demographics, device,
network traffic
ENVIRONMENT: sound,
air quality, mechanicals,
live video
PEOPLE: employee
actions and behaviors
SHIP: operational &
performance logs,
geolocation
6. 6
This Changes Everything
PRODUCT
One-time product
design
Continuous design/
feedback
MANUFACTURING
Silos, manual
operations
Automated
Smart factories
SUPPORT
Unplanned
downtime
Fixed before
it breaks
ORGANIZATION
Independent
business units
Cross-functional
coordination
BUSINESS GOALS
Product
centered
New digital
business revenue
9. 9
Use cases
Automotive
conglomerate
Use-case: Smart
navigation, object
recognition & smart
parking assistance
Lead-time: Real-
time
# Data Sources: 6+
Value: Reducing
accidents, road traffic,
improving driving
experience
ML Accuracy: 98%
Savings: on-going
(with potential to reduce
accidents & traffic)
Storage
technologies
Use-case: Reduce
analyze time failures
& defect patterns
Lead-time: near
Real-time
# Data Sources: 10+
Value: Reduce dev.
time, improved
manufacturing yield,
customer satisfaction
ML Accuracy: n/a
Savings: $m/year
Shipping
company
Use-case: Predicting
Piston Ring Failure
Lead-time: 10 days
# Data Sources: 8
Value: Reducing
maintenance costs
with repairs &
replacement (R&R)
ML Accuracy: 63%
Savings: $m/year
Railway
company
Use-case: Predicting
Engine failures
Lead-time: 7-30 days
# Data Sources: 4
Value: Improving
downtime, customer
expe-rience &
optimize R&R costs
ML Accuracy: 87%
Savings: $m/year
Network
rail
Use-case: Predicting
switches failures
Lead-time: 7-30 days
# Data Sources: 6
Value: Reducing
delays, improve
service & optimize
R&R costs
ML Accuracy: 85%
Savings: $m/year
10. 10
Use cases
Shipping
company
Use-case: Predicting
Piston Ring Failure
Lead-time: 10 days
# Data Sources: 8
Value: Reducing
maintenance costs
with repairs &
replacement (R&R)
ML Accuracy: 63%
Savings: $m/year
Length 400 m
Beam 59 m
Draft 16 m
Capacity 18000+ TEU
Price 185M USD
Dual engines Ultra long stroke,
2-stroke diesel, 86000 hp
Engine room
(one of the
engines)
11. 11
2
3
4
1
2
Cylinder and piston on a two-stroke engine
Use-case: Predicting Piston Ring Failure
1
2
3
4
Rings changed due to failure on ring no.2
Rings changed due to failure on ring no.2
12. 12
Approach
A
Data capturing
• Determining sources of data and
making data available
• Processing before data can be
used
Readability of the data, not just
access to the data was a key
learning
B
Data understanding
• Discussions with SME to
understand the data
• Determination of data focus
• Data Analysis
• Learnings
Significant learnings while
understanding the raw data
available from various sources
C
Developing the model
• Model development
• Model fine-tuning
• Model validation
• Promotion to Production
The deteriorating general piston
ring condition can be predicted
with a lead time of 10 days
D
Implementing the model
• Run in Production
• Catalog Model
• Promote to Scoring
• Design and Implement Alerting
The model supports engineers to
plan maintenance of the engine
by alarming about potential
failures
Activities Activities Activities
Take-away Take-away Take-away
Activities
Take-away
13. 13
High-level process conducted
Learnings
• Understanding engineering processes
and technology of the engine and its
components is a major facilitator in
determining the right focus on the data
Sensor Data
Sensor
Descriptions
Maintenance
Data
Inspections
…
• Cleansing
• Outlier
removal
• Removal of
inactive
sensors
• Time series
alignment
In conjunction with
customers SME:
• Sensor hierarchy
creation
• Identification of
sensors relevant for
engines
• Sensor prioritization
based on engineering
knowledge
• Identification of
failures
• Inspection of various
data sources for
usability
Data to Model
• Feature
Engineer
ing
• Statistical
Analysis
• Correlation
Analysis
• Machine
Learning Model
• Cross-Validation
• Business
Validation
14. 14
An example to put “Things” in context
1 month
High
Low
Abnormalcylinderbehaviour*
Lead
time
Port stays
PROB1
0.0
0.2
0.4
0.6
0.8
20xx-xx-xx 20xx-xx-xx 20xx-xx-xx 20xx-xx-xx
Piston ring change
Cylinders
Piston ring(s) changed Threshold
Abnormal behaviour:
Everything above
the threshold triggers
an alarm
Other cylinders below
threshold
Worn piston ring was
changed
Each point is a
combination of selected
sensor data for a
specific cylinder
15. 15
Parameters Target
Level of detail
Prediction of piston
ring condition
Coverage
Accuracy
Lead time
60-80%
50-65%
10 days
Success criteria
Prediction piston condition on
engine level
75%
63%
>10 days
Results Success
Outcome
Results
19. 19
Lessons Learned and key takeaways
Include All the
data,
not just sensor
data and it
changes
Combine
Science/Engine
ering with SME
Knowledge
Validate the
results &
measure the
uplift/savings
Measure the
cost of putting
the solution in
production
Measure the
Lead-time
required upfront
Don’t assume
your data is not
good enough
but measure the
quality upfront
Share the
results in a
business
language with
relevant
stakeholders
Manage the
expectations
all the time