Autonomous	Intelligence	for	the	
Industrial	Internet
Marco Laucelli
Founder & CEO,
marco@novelti.io
Librecon 16, Bilbao
November 22th, 2016 www.novelti.io
www.microduino.cc
IoT
www.amazon.com/Dash-Button
People don’t want
gadgets anymore.
They want services
that improve over
time.
Personalized
Context-aware
Interactive
Real-time
Service
=Data
Usage-driven
product design
Energy saving
recommendation
service
Usage-driven
product design
Predictive
maintenance &
sales
Energy saving
recommendation
service
Usage-driven
product design
Performance Monitoring
Predictive Maintenance
Quality Control
Equipment as a service
%
Manual
Expensive
Today’s analytics process requires a lot of manual efforts
Outdated
Models require frequent retraining andold data is useless
High CAPEX and non-scalable tool-based approach
New data
sources
Unknown
environments
New
interactions
New business
models
Predictive Maintenance Industrial Assets
Physical inputs: vibration, temperature and pressure...
Used supervised off-line trained models
What is normal for a machine is environment-dependent
Machines are continuously evolving
Can we provide an continuous
behavior learning system?
Can we monitor the real-time
behavior against the repository?
Can we use behavior monitoring
as an input for maintenance
procedures?
*Morales et al: Big Data
Stream Mining Tutorial
2014
Autonomous· Behavior
monitoring · Anomaly
detection · Pattern discovery ·
Real time profiling· SaaS
Plug
Learn
Metrics & Alerts
The production goal of this system is to
maintain the gas concentrations.
The capacity is determined by the main
compressor and the water pump flow.
Fault aboutMay. Maintenance
intervention.
Continued to operate the machine for
months until production loss was too high.
0
5
10
15
20
25
30
31/7/13 31/8/13 30/9/13 31/10/13 30/11/13 31/12/13 31/1/14 28/2/14 31/3/14 30/4/14 31/5/14 30/6/14 31/7/14
Caution
0
5
10
15
20
25
30
31/7/13 31/8/13 30/9/13 31/10/13 30/11/13 31/12/13 31/1/14 28/2/14 31/3/14 30/4/14 31/5/14 30/6/14 31/7/14
BH7
Maintenance
Symptom
evidence
Final Failure
6 months
On-line
Machine
Learning
Real-time
metrics and
alerts
Collective
Intelligence*
Anomalies
and
behaviours
Connected
Cars
Smart
Manufacturing
Home
equipment
Real time
efficiency
Connected
Cars and M2M
Industrial
IoT
Smart Home &
Appliances
Smart Metering and
supply
Asset PerformanceMonitoring Predictive Maintenance Quality Control
Real Time Autonomous
analytics for Internet of Things
Technical
Challenges
Streaming
Machine Learning
What, When, How
Long
Devices expected to fail
Limited computing power
Limited connectivity
Power limitations
Expect failures
Periodical synchronization
Out of order delivery
Globalclock
Different data
transfer patterns
Streaming
Time windows
ML steps synchronization
State-full or State-less
Update and recalculate reports
Concluding remarks
1. IoT is about everything becoming a web business.
Flexible and fast business model innovation.
2. Analytics complexity will not stop growing. Adaptive
learning is needed for IoT.
3. IoT data capture and quality will an issue and we need
resilient ML approach.
Edge
intelligence
Digital Twin
integration
Collaborative
Capturedomain
expertise
Future trends
THANKS!Any questions?
www.novelti.io
info@novelti.io

Autonomous intelligence for the Industrial Internet - LibreCon 2016