closing the loop: from analytics & machine learning to ... · the hidden data of the iot sensor...
TRANSCRIPT
Closing the Loop:
From Analytics & Machine
Learning to Value Creation
JOHN CANOSA
TANDEM RUNNING
Who I Am
John Canosa - Over 30 years of experience in the electronics, software and device networking industries and over 20 years connecting things to the Internet. Recognized as an ‘M2M Pioneer’ and holder of several patents for IoT technology. Chief Strategist for Connected Products and General Manager of M2M Solutions at ThingWorx/PTC. Founder and CEO of Palantiri Systems. CTO of Questra Corporation. BSEE from Clarkson and a MSEE from Rochester Institute of Technology.
The Fallacy of the Common Mental Model of the IoT
Intelligence and Data are Everywhere
Not just in the Cloud or in the
Fog, it extends out to the mist
and the dew on the leaves
It’s not just in the form of water
vapor either – sometimes it’s a
torrential downpour and other
times its as hard and opaque
as hailstones
Creating Value Requires Taking Action,Taking Action Requires Closing the Loop
Feedback is essential, whether
it is in machine control,
creating more efficient
business processes, or
increasing revenue through
new services.
Analytics alone do NOT close
the loop.
But Take Action Where and When?
Machine Time – local closed loop control, single device
E.g safety critical shutdown
Network Time – Peer-to-peer comms, latency
tolerant setting adjustments, multiple devices
E.g re-routing supply truck
Human Time – Machine learning events,
predictive maintenance scheduling, manual
intervention, single device or fleet based analytics
E.g. SW Update, on-site maintenance
Business Time – deep analytics on user behavior for
marketing & sales, design defects detection for
engineering, component failure tracking, fleet, business,
and external data based analytics
MachineTime
NetworkTime
HumanTime
BusinessTime
day-wks
min-hrs
seconds
millisec
Analytics Requires Data, But Data Takes Many Forms
Highly structured databases
Time series – sensors
Semi-structured (e.g. syslog)
Unstructured
The Hidden Data of the IoT
SENSOR DATA
Tip of the iceberg, most visible, structured,
deterministic, pre-configured set of known
attributes, relatively easy to report and analyze
LOG DATA
Massive amounts of data, hidden from
normal view, unstructured, complex & messy
formats, ideal for machine learning and
predictions, but requires specialized tools for analytics
Extracting the Value of Unstructured Data Requires Transformation
Log files are typically formatted for human
consumption, so once
again we must remove the
human from the loop
IoT Analytics Challenge:
Over 60% of time is spent in
machine data
transformation &
preparation before any
analytics can be performed
Transformation Tools: Glassbeam
Real World Value of Mining “Dark Data”
Summary
Data is a cost center
Taking action on the meaning of that data is what provides value
Analytics and Action are two sides of a coin
Choosing where to close the loop is critical
Latencies, data transport & storage costs, data needs from single
devices or the entire population all factor into correct placement
Shining light on “Dark Data” can create tremendous value