Model Predictive

Control and analytics

Presenters: Juliano de Goes Arantes
PUBLIC Giancarlo Scaturchio
Agenda

1 2 3 4
Intro Leveraging Model Predictive MPC in Calciner
Digital Control
Technologies for
process
optimization

5
Delivery
approach

PUBLIC | Copy right ©2020 Rockwell Automation, Inc. | 2

Mining and Mineral Processing
Mining manufacturing is a complex process

Process characteristics Process Objectives

Unreliable, infrequent or lack of Reduce Cost / Ton
critical measurements
Lower plant costs
The inertia of the process often
Lower energy intensity
obscures the understanding the
impacts of changes made by Lower emissions
operators
Continuing to meet regulatory
Multivariable characteristics of the requirements
process
Lack of experienced and trained
High degree of disturbances instrumentation technicians and
operators

How to achieve the process objectives in such

complex operations?

PUBLIC | Copy right ©2020 Rockwell Automation, Inc. | 3

 Leveraging Digital
Technologies for process
optimization

PUBLIC | Copy right ©2020 Rockwell Automation, Inc. | 4

Although regulatory control provides adequate control in terms of plant safety, it rarely achieves
optimal control in terms of quality, nor does it operate in the most economical fashion

Mining plant processes have multiple interactions not easily handled

Relationship between variables are non-linear
Still requires operator intervention with lots of experience

Process Limit Approximately 95% of the

closed-loop operations of
the industrial automation
sector use PID controllers

Model Model
Traditional
Predictive Predictive
Operation
Stabilization Optimization

PUBLIC | Copy right ©2020 Rockwell Automation, Inc. | 5

Machine Learning
Enables a System to Learn from Data

Goal example: Predict next value

Supervised learning
Creates a model on known input Soft Sensor/
and output data to predict future
problems
Model Predictive The categories of machine
learning applications can be
Machine Control deployed in production
today

Learning Unsupervised learning

Understanding each and
their unique benefit, can
Infers normal patterns from a help you determine which
data set without reference to
known outcomes and
Anomaly Detection are right for the use

identifies deviations

Goal example: Find some structure

A key value in the data
of machine
learning is
the ability to
quickly build
process or
data models
PUBLIC | Copy right ©2020 Rockwell Automation, Inc. | 6
 Model Predictive Control

PUBLIC | Copy right ©2020 Rockwell Automation, Inc. | 7

 Model Predictive Control Horizon
Manipulated
Variable (MV) Use first step of horizon
MV prediction

MV measured value

MV set point

MPC optimizes
future trajectories
using an internal
simulation
Controlled CV measured value
Variable (CV)

CV prediction

current time indicator CV desired value

PUBLIC | Copy right ©2020 Rockwell Automation, Inc. | 8

How MPC Generates Benefits

104
REDUCES Variability
102
Process/Specification Limit
100 ACHIEVES

98
M ANAGES the process
96
Benefits within constraints
Push process
94 tow ard limits
Variability under Variability reduction Maintain quality just ACHIEVES UPLIFT operate
operator control w ith Pavilion MPC w ithin specification
92 closer to specifications and
Before During After performance limits while
TIME maintaining safety margins

PUBLIC | Copy right ©2020 Rockwell Automation, Inc. | 9

Model Predictive Control Solution
MPC is a proven technology that improves plant and asset
performance by reducing process variability
( (
+ ( )

Controller based or software based Historical Plant Test Process

Data Data Knowledge
Hybrid modeling functionality allows the
user to take advantage of all of the known
information (empirical data, equations, ML / AI
equipment specifications, etc.) Modeling, Identification, Analysis

Console
Handle slow or complex processing lags
Virtual Online Analyzer Controller / Optimizer
Virtual Online Analyzers replace sparse,
or infrequent, feedback from Lab or
analyzer

run-time visualization
Modeling
Supervised
MPC uses a model of the process to predict how Machine Learning
Model Predictive
Control
the process output variables will respond to
changes in the process input variables and
disturbances

MPC algorithms making use of ML, where the

machine learn engine learns/updates the
Process Model MPC
mathematical model using data while the MPC
Set Points
then uses it for control
Closed-loop technology allows processes
to run closer to constraints
Target-specific KPIs like product quality, specific Data
energy costs and tons of product produced.
Then use those KPIs to achieve a targeted Supervised machine learning technology that
balance within equipment and quality limits assesses current and predicted operational data,
compares the data to desired results, then computes
and updates the process online set point targets

PUBLIC | Copy right ©2020 Rockwell Automation, Inc. | 11

Virtual Online Analyzers - VOA® / Soft Sensors
Soft sensors are widely used to estimate process variables that are difficult to measure
online

Clinker quality Mills Kiln Burning Zone Combined Oxygen

Freelime Blaine/Residue Temperature Crystal Water

LSF, AM, SM Feed recipe (%) Kiln torque Feed Total Energy
Fuel mix Separator Feed Nox Feed Quality Fuel mix
Feed rate Feed rate Heaty Duty Feed rate
O2, CO
Kiln Energy Kiln Speed
Feed rate Fan speed
Precalciner Energy Fan Speed
Kiln speed Temperature
O2, CO Oxygen Measurement
BZT Condition Clinker freelime Tertiary Air Temperature
Mill load Secondary Air Temperature
Returns/Reject Fan Speed
Back End Temperature
Pyrometer

PUBLIC | Copy right ©2020 Rockwell Automation, Inc. | 12

Virtual Online Analyzers - VOA® / Soft Sensors
Advanced Continuous Emissions Monitoring

Patented, model-based, Predictive

Emissions Monitoring System
Engine to provide highly accurate
measurements of NOX , CO, CO2 , Fuel flows
green house gas and other emissions Fuel Quality
as an alternative to costly hardware Air Flow
Process O2 NO X
based CEMS Temperatures O2
Ambient CO
Humidity
Inputs Data Qualification Outputs Real-time
Faster deployment reduces costs and time to achieve and Sensor Alerts
and maintain compliance Validation

Minimizes compliance monitoring, record-keeping

and reporting costs
Model Predictive Control in cement processing
Basic Concepts
CV
Kiln
Controlled Variables (CVs) NOx

Process variables to maintain at a target or

within a range (it can be considered outputs)
MV
Constrain Variable (CCVs) Kiln
Speed
The state is forbidden to penetrate or may have
physical limitations
Manipulated Variables (MVs)
A manipulated input is one that can be adjusted by
the control system (or process operator)
Disturbance Variables (DVs)
Disturbance variables - these are also called
"load" variables and represent input variables that DV CCV

can cause the controlled variables to deviate Roller Mill Power

On

PUBLIC | Copy right ©2020 Rockwell Automation, Inc. | 14

Model Predictive Control Solution: From model to execution
Example of a fictitious Kiln optimization (Simplified)

Model Matrix: It shows dynamic relationships Console: Optimization /

Data Collection
between process outputs and inputs Visualization

Process Trends

Variation Example

Production Rate
Example

Kiln Speed

Graphical Modeling Control Utilization

Merely Illustrative Stage 1A Example
Temp

PUBLIC | Copy right ©2020 Rockwell Automation, Inc. | 15

So what makes MPC different?
PID MPC

Single variable in & single variable out Multivariable in & multivariable out
Set up a target and control process variable to the target. Control strategy based on a holistic comprehension
No awareness of how control changes impact other PID loops. of key process variables and their interdependencies

Feedback control Predictive control

Dynamic models developed through process step tests. Controller action
The controller will take no action unless PV deviates from target
based on current and anticipated future PV deviations from target.

Indirect control of lab measurements Direct control of lab measurements

Controller predictions of lab measurements used for control
Control property variables through proxy (temperature, pressure, etc.).
updated as available

Poor ability to handle process delays Explicit dynamic models

During complex dynamic interactions Full understanding of process dynamics and interactions

Constraints Constraints
Only internal awareness of loops limits on setpoint and output Predict and monitor future values of constraints

PUBLIC | Copy right ©2020 Rockwell Automation, Inc. | 16

How Model Predictive Control generates benefits
Reduce variability, increase recovery and save consumables

104
REDUCES Variability
102
Process/Specification Limit
100
ACHIEVES Plant Stability

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

PUBLIC | Copy right ©2020 Rockwell Automation, Inc. | 17

 Model Predictive Control
Application in calciner

PUBLIC | Copy right ©2020 Rockwell Automation, Inc. | 18

MPC in calciner P

DP
T

Manipulated Controlled
Variables (MVs) Variables (CVs) Process Filter Fan
Process variables you can variables that need to be Blaine
adjust that affect the CVs maintained at a target or
(typically PID set-points) within a set range
Combined
H2O F
T

T
Recycle
Disturbance RPM Stucco
Variables (DVs)
Measured process variables F
Gas Fan

that affect the CVs that are F

not MVs Feed

CV MV DV P
P

Crystal water as a function of T DP

Heater Fan
calciner DT per ton of feed Gear Box

Vib
kW

PUBLIC | Copy right ©2020 Rockwell Automation, Inc. | 19

 Capabilities and Benefits

PUBLIC | Copy right ©2020 Rockwell Automation, Inc. | 20

Blending
Unstable chemistry in raw materials may lead to unstable kiln operation

Process Challenges MPC delivers

Raw materials variability Reduce blend variability
Chemistry variations in the Reduce blend costs
overall material blending
Improve raw material resource usage
process
Improve kiln performance
Blending process controlled
manually in several operations Raw material cost reduction
Major impact of kiln stability

Potential benefits:

LSF typical deviation reduction up to 40%

PUBLIC | Copy right ©2020 Rockwell Automation, Inc. | 21

Rotary Kiln
The reactions involved in rotary kilns are still not well understood because of their
complexity
Process Challenges MPC delivers
Required quality of clinker with Increase throughput
minimum fuel consumption
Decrease product variability
Stable Operation
Reduce energy costs
Stabilize a temperature profile
Improve product grade
along the rotary kiln
Reduce emissions
Uncertainty of raw material
characteristics Improve heat recovery
Complex dynamics Longer campaign run and refractory life

Potential benefits:

Increases throughput: 2% - 3%
Fuel consumption recurion:1% 2%
Power consumption reduction: 1% - 3%
Deviation free lime: -50%
Reducing NOx emissions by 20%
Reducing CO2 emissions by 5% - 10%

PUBLIC | Copy right ©2020 Rockwell Automation, Inc. | 22

Cement Grinding
Almost 70% of this electrical energy is used for size reduction, which includes crushing and
grinding of cement raw materials and clinker grinding

Process Challenges
MPC delivers
Maximize the productivity,
Reduce energy cost
variability within the products Increase throughput
Reduce fineness variability
Energy consumption
Reduced maintenance costs
Lack of reliable measurements
of some key variables

Potential benefits (Cement Mill):

Increases throughput: 2-10%

Reduced specific power consumption: 2-10%
Reduced transition time for type changes up to 40%
Reducing of off-spec product: 20-30%
Reduced standard deviation for product residue/Blaine: 20-30%

PUBLIC | Copy right ©2020 Rockwell Automation, Inc. | 23

Maximize Performance within Constraints

Fresh feed is maximized up to return flow

constraint.
Blaine is controlled to target as throughput
is maximized.

PUBLIC | Copy right ©2020 Rockwell Automation, Inc. | 24

User Case: Aalborg Portland
white cement improves energy and
environmental efficiency of its rotary
kiln

Challenges Solution Results

Minimize specific fuel consumption for MPC solutions for calciner, rotary-kiln Total of 12.3 GWh / year of energy
required calcination and clinker free lime and cooler controllers savings
Minimize electrical energy consumption Split solution into multiple subsystems 5.6% power reduction in specific ID
from ID fans to allow more operational flexibility and fans
robustness, while still generating
Minimize excess O2 in the system while 0.96% reduction in stack oxygen and
benefits where only a part of the unit
helping prevent high CO excursions
may be excluded 5.4-degree Celsius reduction in exit gas
Maximize alternative fuel usage in the temperature
precalciner within allowed constraints
6.1% reduction in main burner fuel
Reduce calciner temperature variability
36% reduction in quality variability
Reduce bed depth fluctuation in grate
Dramatically eased operation through
cooler
all typical changes caused by feed
Maintain clinker quality within chemistry, coal quality and operational
specifications and reduce variability changes.

PUBLIC | Copy right ©2020 Rockwell Automation, Inc. | 25

 Capability

Model predictive control

Controller or software based

Crusher Challenges Grinder Challenges Flotation Challenges Thickener Challenges

Poor level control in the secondary and Oscillatory behavior and deviation from
set-
secondary and tertiary crushers choked of reagents control is inefficient (long residence time,
in order to maximize efficiency and
(steel-on-
Declining ore grades the
large disturbances and non-linear

secondary and tertiary crushers

maximum energy efficiency extraction processes and
all the must
production line run at

an optimum level. As consequence

Potential Benefits Potential Benefits Potential Benefits Potential
mining operations areBenefits
becoming
Increase Increase Maintain Increase Maintain concentrate grade to reduce Increase water recovery by 3% (typically
Decrease equipment Decrease energy product give-away or off- more complex
Reduce flocculants by
wear by 5% Decrease maintenance Increase Reduce Reduce possibility of emergency
cost by 2% reagents by 3% feed shutdown

PUBLIC | Copy right ©2020 Rockwell Automation, Inc. | 26

 Scalable Analytics
Architecture

INTERNAL | Copy right ©2020 Rockwell Automation, Inc. | 27

Minimum configuration
Bill of Material MPC

Solution Builder App Run Time

Console
(optional for EU) (required)
(required)
9529-PV8SBENE 9529-PV8MPCXENE
9529-PV8DBENE

Bill of Material:
- One (1) Solution Builder (optional)
- One (1) Console (Required)
- Run time based on #-of-MVs required (required)
Pavilion8® Toolset

1 2 3

Solution Builder Application Manager Console

Configure Execute Visualize

4
 Mine to Market Visibility
3rd party applications AR Solutions

Real-time Status Modernized

MES / Analytics Step-by-Step Instructions Workforce
Expert Capture
Knowledge-driven Asset / Process Performance Remote Collaboration Modern digital tools
Operations Asset Health
help deliver better
Inventory / Accounting
Downtime insights and
Solutions that Maintenance improve field
enable better Design & Simulation experience
Quality
decision making Energy

Intelligent Packaged Power Process Solutions

Model Predictive
Control / Anomaly
Detection
Device level Predictive
Capabilities

Controlled based AI
Operational Model Predictive
Efficiency Control

Modern
technologies
improve
performance of
process,
equipment, and
people

The Connected Mine Architecture

Our delivery capabilities
Analytic Solutions Group - How We Do It

Discover Demonstrate Deploy

Value Definition Value Testing Value Realization

Persona Interviews

Use Case Alignment Immersive Adoption

Objectives & Key Results Illustrate

User Validation &
Business Case Alignment

PUBLIC | Copy right ©2020 Rockwell Automation, Inc. | 32

High level project schedule

Order Commissioning Final

Acceptance FDS FAT
complete Acceptance

1 2 3 4 5 6 7

Plan
Design
FDS
Step testing
Develop

Deploy

Close
Audit Period

Customer Resources

PUBLIC | Copy right ©2020 Rockwell Automation, Inc. | 33

 Data science used in machine learning may be complex, but
Continuous Learning the result for you is simple: better decision-making or
greater automation to help optimize your plants. With the
right machine learning technique and access to the relevant

your operations and optimize production long term.

PUBLIC | Copy right ©2020 Rockwell Automation, Inc. | 34

 Thank you
Juliano de Goes Arantes
www.rockwellautomation.com
Email: [email protected]

INTERNAL | Copy right ©2020 Rockwell Automation, Inc. | 35