FactoryTalk® Analytics™ LogixAI™

Overview | April 2019

*This product continues to evolve based on Agile Methodology

Production
is down

PUBLIC | Copyright ©2019 Rockwell Automation, Inc. | 2

 Who’s the What’s the This has What tools
Reactive thinking during expert? problem? got to stop! to use?
a downtime instance

PUBLIC | Copyright ©2019 Rockwell Automation, Inc. | 3

 ANALYTICS
The discovery, interpretation, and communication
of meaningful patterns in data. Analytics relies on the
application of statistics, computer programming and
operations research to quantify performance.

PUBLIC | Copyright ©2019 Rockwell Automation, Inc. | 4

 Scalable analytics
The key differentiator for Rockwell Automation

DESCRIPTIVE DIAGNOSTIC PREDICTIVE

PRESCRIPTIVE
ENTERPRISE

What plant Why is site A throughput Will I meet plan today? How can I change operations
performed the best? below plan? Tomorrow? to improve profitability? Yield? Quality?
SYSTEM

Is Line 1 running ok? Why is Line 1 I predict that Line 1 quality is What action helps the operator to
quality poor? moving out of tolerance. avoid poor quality?
DEVICE

Am I running ok? Why did a fault happen? I predict a fault will What action helps to avoid
happen soon. the fault?

PUBLIC | Copyright ©2019 Rockwell Automation, Inc. | 5

Building predictive models and change management

3% 5%
4%
Groups spend this percentage of time on
9% each phase in a predictive analytics project
Based on 166 responses, adapted from Eckerson (2007)
What
data scientists
spend the most Project definition 13%
60%
time doing 19%
Data exploration 18%
Data preparation 25%
Model creation, testing & validation 23%
Scoring & deployment 12%
Model management 9%
Other 8%
Building training sets Other
Refining algorithms Mining data for patterns
Collecting data sets Cleaning and organizing data

PUBLIC | Copyright ©2019 Rockwell Automation, Inc. | 6

Building predictive models and change management

Exploratory Process the data

Data Analysis
(Hadoop)
(R)
Build model in dev/
modeling environment
PMML

Lifecycle of a ANALYTIC MODELING

OPERATIONS Log

predictive model
files

Deploy model in
operational systems with
scoring application Refine model
(Hadoop, streams & Augustus) Retire model and deploy
improved model

PUBLIC | Copyright ©2019 Rockwell Automation, Inc. | 7

Scalable analytics & decision-making
At the right time, at the right level

JUST BIG DATA

HISTORIANS ANALYTICS VENDORS

LEVEL 5 Enterprise

LEVEL 4 Info / MES

LEVEL 3 SCADA

LEVEL 2 Supervisory

LEVEL 1 Control

LEVEL 0 Devices

Produce, analyze and react to information as close to the source as possible

PUBLIC | Copyright ©2019 Rockwell Automation, Inc. | 8

Developing and using analytics

DEVELOP / DEPLOY USE

L4 Executives /
IT / Data scientists ERP
Business planning
Finance

WAN
L3 Manufacturing Operations /
IT / Process engineers MES operations Quality management

LAN
L2 Supervisory control
FactoryTalk® Analytics™ LogixAI™
Operators
L1 Process control
Automation / maintenance
Process engineers
L0 Physical plant

Control networks

PUBLIC | Copyright ©2019 Rockwell Automation, Inc. | 9

Value
Engineered execution
Project

Commercialized offerings
Product

ENTERPRISE
ANALYTICS

SYSTEMS
ANALYTICS
EMBEDDED
ANALYTICS
DEVICE
ANALYTICS

Complexity

PUBLIC | Copyright ©2019 Rockwell Automation, Inc. | 10

 Scalable analytics
Where FactoryTalk® Analytics™ LogixAI™ fits into the ecosystem
ENTERPRISE

What plant Why is site A throughput Will I meet plan today? How can I change operations
performed the best? below plan? Tomorrow? to improve profitability? Yield? Quality?

+
SYSTEM

Is Line 1 running ok? Why is Line 1 I predict that Line 1 quality is What action helps the operator to
quality poor? moving out of tolerance. avoid poor quality?
DEVICE

Am I running ok? Why did a fault happen? I predict a fault will What action helps to avoid
happen soon. the fault?

PUBLIC | Copyright ©2019 Rockwell Automation, Inc. | 11

Two approaches to advanced analytics

01101011001
Expert driven analysis 010100101101001011011 Automated data analysis
100101101000101111010010110
0101010010110100101101001010110
01101000111010010101010110100101010
• Tools like SaS, Pavilion®, SAP HANA 01011010010110101011010010110101000111 • Systems like Cortana, Watson and
0101001011010010110100101101001110101001
now FactoryTalk® Analytics™ LogixAI™
• Leverages human expertise – both 010100101101001010001010101010111001011001

DATA
0101101000101010100101101011010010110100100

in understanding the problem and 01011010001010100111101001011010010100101101

10100101101000101001001001011011011010100101 • Leverages AI and a foundation of
the tools 0101001011010010101010111010101001011010010
101010100101101001011010010111011101010001 universally applicable physics-based
• More powerful and flexible…
10010011101010011101001101000101001001010
101001010010111010100110110101011101001 modeling
1011101010010110101101001011010101101

• But varied approaches mean

0110101001110101001010101010010101
010101011010010110100101101010 • Still less intelligent than a human…
1010111001010010010110101
different results each time you try 010010101101001010
101001110
• Repeatable results each time

PUBLIC | Copyright ©2019 Rockwell Automation, Inc. | 12

 Supervised learning Unsupervised learning
Given datasets and labels, predict Draw inferences from datasets
a label for a new dataset without human-labeled responses

EXAMPLE EXAMPLE
Recommender system by Facebook or Google Finding hidden patterns in clustering applications

Machine learning
Paradigms Reinforcement learning Active learning
Learn how to take actions in an Learn by interactively obtaining the
environment to maximize some desired labels at selected new
notion of cumulative reward data points

EXAMPLE EXAMPLE
Games, self-driving cars Medical applications

PUBLIC | Copyright ©2019 Rockwell Automation, Inc. | 13

FactoryTalk® Analytics™ LogixAI™
Descriptive | Diagnostic | Predictive | Prescriptive

Configure Identify data Model Monitor Integrate

NO DATA SCIENTIST REQUIRED

PUBLIC | Copyright ©2019 Rockwell Automation, Inc. | 14

How FactoryTalk® Analytics™ LogixAI™ works

F1 Select controller output(s) you care about

PUBLIC | Copyright ©2019 Rockwell Automation, Inc. | 15

How FactoryTalk® Analytics™ LogixAI™ works

F1

Select potential inputs or variables from controller tags

L1 L2 L3 L4 L5 L6 … LN

PUBLIC | Copyright ©2019 Rockwell Automation, Inc. | 16

How FactoryTalk® Analytics™ LogixAI™ works

F1

FactoryTalk® Analytics™
LogixAI™ determines
inputs that relate to
S2 your output S3

L1 L2 L3 L4 L5 L6 L7 L8 L9 … LN

PUBLIC | Copyright ©2019 Rockwell Automation, Inc. | 17

How FactoryTalk® Analytics™ LogixAI™ works

F1

S1 S2 S3

L3 L4 L5 L6 L7 L8 L9 … LN

PUBLIC | Copyright ©2019 Rockwell Automation, Inc. | 18

How FactoryTalk® Analytics™ LogixAI™ works

! FactoryTalk® Analytics™ LogixAI™

F1 monitors the process and sets
a bit in the controller when it
predicts a problem

S1 S2 S3

L3 L4 L5 L6 L7 L8 L9 … LN

PUBLIC | Copyright ©2019 Rockwell Automation, Inc. | 19

FactoryTalk® Analytics™ LogixAI™
Modes of operation

Operational monitor Value estimation

• “Anomaly detection” • “Soft Sensor®”
• Create a model of • Create a model
normal operation, from existing data
detect anomalies to estimate another
value

PUBLIC | Copyright ©2019 Rockwell Automation, Inc. | 20

 Fill in the blanks with this example statement:

[Process X] is known as our bottleneck. Every

time key [controlled variable Y] of [process X]
goes unstable, it relates to [$Z of lost production,
scrapped product, etc.]
Analytics is
a word problem Several key manipulated variables are known to
What’s your problem statement? control [Y]; like [A, B, C, D, E…ZZ]. PID control
works ok, but MPC is overkill.

If I could predictably monitor the control variable

setpoint, when instability occurs; I would be in a
better position to take action by doing [1, 2, 3].

PUBLIC | Copyright ©2019 Rockwell Automation, Inc. | 21

FactoryTalk® Analytics™ LogixAI™

Project charter concepts

• Operations / System focused improvements!
• Targeting one key source of data / the controller code
• Target key control variables / stability issues
• Definition of key metrics – like Quality / Scrap %
• Capture baseline for measuring success
• Hypothesize the benefit
• Processes to leverage tool based on
• Anomaly Detection / Soft Sensing
• What would you do with a prediction?
• Process to assign financial impact – track ROI for
future efforts.
• Process to share success internally (and with RA)

PUBLIC | Copyright ©2019 Rockwell Automation, Inc. | 22

FactoryTalk® Analytics™ LogixAI™

Technical considerations
• Data Scientist is not needed but an automation
expert is
• ControlLogix® L7 / L8 controllers
• Operations obey first unit operations principles – no
human subjectivity
• One application code file contains all necessary data
(phase 1)
• If used as a Soft Sensor® – where’s the data?
• What to do with a prediction?
• Static modeling versus dynamic modeling?
• When is PID not enough?

PUBLIC | Copyright ©2019 Rockwell Automation, Inc. | 23

Future of FactoryTalk® Analytics™ LogixAI™

Pumping station monitoring & energy optimal operation

Analytics solutions

Performance Multi-engine Generator Digital oilfield

monitoring optimization set control diagnostics

Phase 1 COTS Product Piloted / deployed through SSB until incorporated into COTS product

Analytics engines

Data-driven Data-driven Data-driven Data-driven adaptive

modeling engine classification engine optimization engine control engine

PUBLIC | Copyright ©2019 Rockwell Automation, Inc. | 24

Talk to us today about how we can
qualify your use case!
www.rockwellautomation.com