Upload
others
View
6
Download
0
Embed Size (px)
Citation preview
Cognitive Analytics and
Next-Gen Prognostics for
the Industrial Internet
Usman Shuja - @kshujaSumant Kawale - @sumantkawale
@SparkCognition
Focused on large growing markets
5
IoT Security Software Market in Billions ($) Machine Condition Monitoring Market in Billions ($)
10.5
8.47.1
2018
6.5
20192014
7.7
+8%
9.79.0
201720162015 2020
14.4 18.2 22.9 28.4 34.8 42.1 50.1
Estimate of number of connected devices in billions
SparkCognition is targeting a $10B
Internet of Things Security market
20182014 2015 2016 2017
+6%
2013
$240B is lost in the US due to bearing failures
$150B in waste across major industries that the Industrial Internet can eliminate
Predictive maintenance within the
Industrial Internet is a sizeable market
1.61.8
2.12.2 2.4 2.5
Source: IDC, Gartner and SV Biz Journal; 30% SW of the tech market. 16% of this Security, per current “Global SW : Sec SW” ratio;
Machine Condition Monitoring market IBIS and Global Strategic Business
SparkCognition’s Focus, Customers & Partners
3
Manufacturing Safety & Machine Failure
Energy
Utilities
Oil & Gas
Rail, automotive and
aerospaceTransportation
Devices, Servers, VMsCloud
Industrial Internet Segments
Devices, fitness appsHealthcare
Consumer
M2M (e.g. Smart House)
Retail
Smart CitiesPublic
Ou
r P
rim
ary
Fo
cu
s
Customers
Partners
Select Relationships
• 20+ customers
• 9 Fortune 500 Companies
o 3 Utilities
o 3 Oil & Gas
o 1 Aviation
o 2 Financial Services
Proven Use Case: Industrial Internet
4
Value Addition
• Sophisticated failure prediction extended forewarning
(from hours to days)
Insights
• Client - Top-5 Power company in the US
• Decisions on replacing expensive capital assets (Six
large boiler feed pumps)
Solution
• Asset-agnostic prediction to cost-effectively support a
large fleet of diverse assets
• Automated model building to augment and expand the
capabilities of human data scientists
• Automated model tuning over time to adapt to
changes in operating conditions and environment
• Automated anomaly identification in historic data
• Powerful analytics, visualization and alerting
4
Client Case Study
Business Problem and Current Limitations
Business Problem
Flowserve, world-leading supplier of industrial and environmental machinery such as pumps, valves
sought cognitive technologies for better failure detection, forewarning and maintenance insight for its
customers
Limitations
Current signature library based approaches, and threshold systems could only identify failures a few
hours prior to them occurring
Additional issues with the current approach
• Insufficient time to respond effectively
• Hard to maintain prognostic models
• Inability of prognostic models to adapt to unique conditions of each individual pump
6
Results
7
Application 1 Application 2
Objective
• Recognize known operating modes
• Detect anomalies with wide
variations
• Advanced warning of possible
problem earlier than what simple
threshold detection methods provide
Results
• >99% accuracy in identifying desired
operating modes
• Application meets the four criteria set
by Flowserve
• Potential failures predicted 5 to 6
days in advance
• Minimal false positives
Identify operating modes, provide advance warning of failures
Data CleansingDimension
Reduction
Feature Space
NavigationOptimization Prediction
• Aggregated data
from 3 files based
on common time
stamps
• Analyzed data
over time to study
trends and to
reduce
dimensions
• Created new
feature abstract
spaces using
Machine Learning
• Analyzed the
derivative
features to
identify trends
• Applied
optimization
methods to
reduce false
positives and
maximize time to
prediction
• Used a prediction
algorithm to
predict the next
failure once
higher order
features were
extracted and
optimized
In Summary: Cognitive approach enabled us to
predict failures early
Enabled Flowserve to predict failure ~6 days ahead of time, minimal false positives
8
Valuable Insights
9
Asset State
• Is there a problem?
• If so, what kind of problem?
• Is there a problem we’ve never seen before? (signature DB
approaches don’t work well here)
Fleet State• Is the entire system operating well?
• Is the entire fleet optimized?
Failure Prediction• When will the problem occur?
• What problem will likely occur next?
Forensics • What factors were most responsible for a failure?
• What factors were most responsible for a sub-optimal state?
Other Cognitive Analytics
Applications
Next Generation Analytics & Prognostics
• Accurate failure prediction and anomaly detection
• Automated model building, selection & management
• Insights through deeper-order analyses
• Flexible and scalable architecture
• In-context technical advisory with IBM Watson
Improve safety and reduce remediation cost through intelligent prognostics
Proven Use Case for Energy
11
Potential vectors for traditional security approaches
Many potential vectors of attack even in an air-gapped facility
Proven Use Case for Security
12
Improve efficiency and reduce failures
Goal Decisions
Improve well
efficiency by
reducing stuck
pipe
• Recommend design based on formation type, well trajectory such as
type & placement of stabilizer in BHA
• Predict requirements for hole clean outs/ wiper trips
• Optimize drilling & operating parameters such as mud weight, mud
type & maximum connection time
Improve well
efficiency by
improving ROP
(rate of
penetration)
• Recommend BHA design such as Bit type, Downhole motor type
based on formation
• Optimize drilling parameters such as Weight on Bit, torque, RPM
Reduce
failures
(downtime)
• Predict when downhole equipment (e.g. mod motor) might fail
1
2
3
Potential Use Cases for O&G
13
Contacts
Usman Shuja, VP Market Development
Sumant Kawale, Sr. Director Business Development
www.sparkcognition.com
@sparkcognition
6034 W. Courtyard Drive, Suite 100
Austin TX 78730