Maximizing Alternative Fuels

Using ML & MPC in a Cement Kiln

Juliano de Goes Arantes – MPC Sales Manager, EMEA
Jon Ander Bengoechea Lopetegui – MPC Solution Consultant

PUBLIC — Copyright © 2024 Kalypso: A Rockwell Automation Busi ness

Agenda

Company Introduction

Customer & Use Case Intro

MPC, Machine Learning and AI

Use Case Results

Key Takeaways

PU BLI C — Copyr ight © 2024 Ka l ypso: A Roc kw el l A utoma ti on Busi ness 2

Company Introduction Rockwell Automation

PU BLI C — Copyr ight © 2024 Ka l ypso: A Roc kw el l A utoma ti on Busi ness 3

OUR PROMIS E

As the world’s largest company dedicated to

industrial automation and digital transformation,
we are creating the future
of industrial operations.
We connect the imaginations of people with
the potential of technology to expand what is
humanly possible, making the world more
productive and more sustainable.

PU BLI C — Copyr ight © 2024 Ka l ypso: A Roc kw el l A utoma ti on Busi ness 4

Working with you to achieve business outcomes

OPTIMIZE EMPOWER BUILD DRIVE ACCELERATE

PRODUCTION PEOPLE RESILIENCE SUSTAINABILITY TRANSFORMATION

Improve productivity Drive efficient upskilling Maintain Leverage data Accelerate

with better & provide support to compliance to improve speed to market,
visibility across your frontline & business productivity respond quickly
operations continuity while achieving to market changes,
sustainability goals and drive business growth

PU BLI C — Copyr ight © 2024 Ka l ypso: A Roc kw el l A utoma ti on Busi ness 5

Rockwell Automation at a Glance
6% $2.1B
~25% $4.1B
Latin America 21% Lifecycle
EMEA
Discrete ~35% Services Intelligent
2023 SALES Process Devices

9.1
BILLION
58%
North
America
BY REGION 15%
Asia
Pacific
BY INDUSTRY
$2.9B
Software and
BY SEGMENT

U SD Control

~40%
Hybrid

Total shareowner return

10 00
Rockwell Automation
S&P 500 Index

100+
Number of countries
29k
Employees: more than World’s Most Ethical The 50 best places to work
half outside the U.S. Companies for innovators
10 0

2008 2023

Serving customers for 120 years

PU BLI C — Copyr ight © 2024 Ka l ypso: A Roc kw el l A utoma ti on Busi ness 6

 INDUST RY FO CUS Life Consumer Industrial
Sciences High Tech
Kalypso delivers digital Digital Thread Strategy Supply Chain
transformation for Product & Portfolio Optimization

CO MPR EHE NSIVE CA PABILIT IES

Management (PPM) Sustainability
CO NS UL TI NG + Operations +
Rockwell Automation. Performance
Improvement
Organizational Change
Management

Product Lifecycle Model Predictive

We help our clients discover, create, D AT A SC I EN CE Intelligence Control
+ Autonomous + Predictive Service &
make and sell better products with digital. & AI Manufacturing Maintenance

Digital Twin Enterprise Resource

Product & Data Planning (ERP)
TECHNO LOG Y + Lifecycle Management + Internet of Things (IoT)
Recognized Digital Leader Supply Chain Extended Reality
Management (SCM) (AR/VR/MR)

Application Cloud Infrastructure

MANAGED Management & Hosting
SER VICES + Continuous +
Improvement Solution Services

PU BLI C — Copyr ight © 2024 Ka l ypso: A Roc kw el l A utoma ti on Busi ness 7

 ValueFirst PROJECT
METHODOLOGY

ASSESS DELIVER AUDIT SUSTAIN

Design, develop,
Propose & plan Commissioning Support
deploy

Confirm business Data gathering & Knowledge

KPI configuration
value validation transfer

Model Performance
Set expectations Training
development metrics
Determine
Application Performance
benchmark Ongoing support
deployment validation
metrics

VALUE-BASED PROPOSAL VALUE-BASED SOLUTION MEASURE VALUE ON-GOING VALUE

PU BLI C — Copyr ight © 2024 Ka l ypso: A Roc kw el l A utoma ti on Busi ness 8

Customer & Use Case Intro Votorantim Cimentos

PU BLI C — Copyr ight © 2024 Ka l ypso: A Roc kw el l A utoma ti on Busi ness 9

Votorantim Group
Votorantim Cimentos is part of Votorantim Group
Votorantim Group is a permanently capitalized
investment holding company, with a long-term
approach, which seeks superior financial returns
with positive socio-environmental impacts.

106 22 + 61 k
years of
Countries Employes
history

US 2.2 US 0.4
billions billions
EBITDA Net Profit
Yacuces cement plant, Bolivia

PU BLI C — Copyr ight © 2024 Ka l ypso: A Roc kw el l A utoma ti on Busi ness 10

Votorantim Cimentos
Operations & activities around the world

Alconera cement plant, Spain

PU BLI C — Copyr ight © 2024 Ka l ypso: A Roc kw el l A utoma ti on Busi ness 11

Our strategy for emissions reduction

Malaga cement plant, Spain

PU BLI C — Copyr ight © 2024 Ka l ypso: A Roc kw el l A utoma ti on Busi ness 12

MPC, Machine Learning & AI Causal Modelling in MPC

PU BLI C — Copyr ight © 2024 Ka l ypso: A Roc kw el l A utoma ti on Busi ness 13

Ways to turn data into action
Model predictive control is a closed-loop prescriptive analytic

BASIC ANALYTICS ADVANCED ANALYTICS

ACTION

HUMAN INPUT

DESCRIPTIVE DIAGNOSTICS PREDICTIVE PRESCRIPTIVE

Historical Data Historical Future Future Action

WHAT HAPPENED? WHY DID IT HAPPEN? WHAT IS ABOUT TO HAPPEN? WHAT CAN I DO TO AVOID IT?

Motor fault – Motor fault – Motor will fault,

Motor will fault
high current high current decrease load

PU BLI C — Copyr ight © 2024 Ka l ypso: A Roc kw el l A utoma ti on Busi ness 14

Causality as a Key Differentiator in AI

3. COUNTER-FACTUALS
• Necessary for Prescription: closed-loop control & optimization.
What would have happened with
output (CV) if I had moved inputs (MV) • Questions cannot be answered by passively collected (historical) data.
differently? (Optimization)
• Best way to predict the result is to experiment under controlled conditions: step testing.
2. INTERVENTION
• Requires a causal model. Pavilion8 uses control matrix to encapsulate SME knowledge.
How can I reach my target (CV) by
manipulating the inputs (MV)?

1. ASSOCIATION • This is what most ML/AI tools do, with different algorithms & tools.

How are the inputs & outputs related • Focus is on improving mathematical accuracy of prediction. Process gains not relevant.
in my system? (Correlation)
• Bad at generalizing. They fit the past well, but no causality.
What will the output be, given inputs X,
Y and Z? (e.g., VOA) • Sufficient for Prediction. Should not be used for control.

‘The Ladder of Causation’, illustration adapted from ‘The Book of Why’ “No machine can derive explanations from raw data. It needs a push”
The Book of Why, The New Science of Cause and Effect, J. Pearl & D. Mackenzie

PU BLI C — Copyr ight © 2024 Ka l ypso: A Roc kw el l A utoma ti on Busi ness 15

Our Approach to Causal Modelling

Data-driven Machine Learning models

• Trained with Design of Experiment / Step tests
Physics-based Models
• Expressions with known physical relationships
• Coefficients fit with plant data
Flexible Hybrid Composite Models
• Independent, parsimonious inputs, backed by
understanding of underlying process causality
• Deal with time-influence, dynamics
• Gain-constrained training as needed

“Do machine learning like the great engineer you are, not like the great ML expert you aren’t.” FLEXIBLE COMPOSITE
Rules of Machine Learning, Best Practices for ML Engineering, Martin Zinkevich, Google
HYBRI D MODELS

PU BLI C — Copyr ight © 2024 Ka l ypso: A Roc kw el l A utoma ti on Busi ness 16

Elevating Operators and Continuously Driving Value

FactoryTalk® Analytics Pavilion8® Soft Sensor® MPC

Server or PlantPAx® MPC

SP SP SP

PID PID PID PLC/DCS

Maximize customer value, reduce
process variability and improve product
consistency leveraging industry leading:
• Model Predictive Control Lab PROCESS
• Hybrid modeling
• Robust, reliable, closed loop
run-time solutions

PU BLI C — Copyr ight © 2024 Ka l ypso: A Roc kw el l A utoma ti on Busi ness 17

MPC is a closed-loop predictive analytic
Multivariable coordinating, dynamic predicting, constraint enforcing and disturbance rejecting control

CV 1 CV 2 CV 3
Predict future values
MV 1 of the controlled variables (CVs)
by movement of all the manipulated
variables (MVs) and disturbance
variables (DVs
MV 2 No model No model

MV 3
Proactive control
to coordinate MV setpoints to
minimize CV deviations from targets,
DV 1 hence reducing variability

DV 2 No model

TIME
PU BLI C — Copyr ight © 2024 Ka l ypso: A Roc kw el l A utoma ti on Busi ness 18
How MPC Generates Benefits

104
REDUCES variability
102
Process/Specification Limit
100
ACHIEVES plant stability

98
MANAGES the process
96 within constraints
Benefits
94
Push process
toward limits ACHIEVES UPLIFT – operate
92 Variability under
operator control
Variability reduction
with MPC
Maintain quality just
within specification
closer to specifications and
Before During After
performance limits while
TIME maintaining safety margins

PU BLI C — Copyr ight © 2024 Ka l ypso: A Roc kw el l A utoma ti on Busi ness 19

Use Case Results Precalciners & Rotary Kilns

PU BLI C — Copyr ight © 2024 Ka l ypso: A Roc kw el l A utoma ti on Busi ness 20

Use Case 1: Laranjeiras
▪ Controller manages pet coke continuously to
accommodate changes in AFR availability, while
increasing stability of the temperature at pre-
calciner outlet. KPIs
• - 4,6% avg. specific energy
▪ The more stable temperature improves quality
control in the clinker. • - 4,8% avg. liter weight
• - 31% std. specific energy
▪ Tighter quality control allows to shift the
average, saving energy and reducing CO2 • - 54% std. liter weight
emissions. • - 42% std. calciner temp.

PU BLI C — Copyr ight © 2024 Ka l ypso: A Roc kw el l A utoma ti on Busi ness 21

Use Case 2: Itau de Minas
▪ Controller allows the operator to maximize AFR
usage in the calciner, resulting in cost savings and
CO2 emissions reduction.
KPIs
▪ Despite the challenge in increasing AFR usage, the
• + 10% AFR usage
controller is able to improve the calciner outlet
temperature control, making it more stable. • - 2% avg. liter weight
• - 10% std. liter weight
▪ Tighter quality control allows to shift the average,
saving energy. • - 20% std. calciner temp.

PU BLI C — Copyr ight © 2024 Ka l ypso: A Roc kw el l A utoma ti on Busi ness 22

Key Takeaways

PU BLI C — Copyr ight © 2024 Ka l ypso: A Roc kw el l A utoma ti on Busi ness 23

Key Takeaways

▪ Rockwell has a long history of applying ML, AI & MPC to cement plants.
▪ The ValueFirst ® approach & state-of-the-art technology ensure project success.
▪ FactoryTalk® Analytics Pavilion8® users in cement kilns achieve:
Improved plant obedience & stability
Maximum alternative fuel usage
Reduced energy usage and CO2 emissions
Reduced giveaway of clinker quality

PU BLI C — Copyr ight © 2024 Ka l ypso: A Roc kw el l A utoma ti on Busi ness 24

 Juliano de Goes Arantes [email protected]

Thank you
Jon Ander Bengoechea Lopetegui [email protected]

PU BLI C — Copyr ight © 2024 Ka l ypso: A Roc kw el l A utoma ti on Busi ness