Standards

Certification

Education & Training

Publishing

Conferences & Exhibits

Real Time Optimization of Air Separation Plants

Tong Li, Thierry Roba, Marc Bastid, and Amogh Prabhu

2

Presenter

• Amogh Vishwanath Prabhu entered graduate school at the University of Texas at Austin in August, 2004 and worked as a graduate research assistant

• During his study at the University of Texas at Austin, he also served as a Teaching Assistant in the fall of 2005 and a graduate level co-op at Advanced Micro Devices, Inc. during the spring and the summer of 2006, and the spring of 2007

• He received his PhD on Performance Monitoring of Run-to-Run Control Systems Used in Semiconductor Manufacturing in the summer of 2008

• Mr. Prabhu is currently employed at Air Liquide R&D, North America since the fall of 2008 at the Delaware Research and Technology Center in the Process Control & Logistics group

3

Cryogenic Air Separation

4

Motivations

• Energy Intensive– Air Liquide consumed more than 0.1% of the world’s electricity in

2010

• Dynamic Operating Environment– Energy price– Customer demands– Plant and Ambient Conditions

Compressor LimitsP

ower L

imits

Pip

elin

e P

ress

ure

Liquid Demand

GO

X D

eman

d

GA

N D

eman

d

Operator’s Preferred

Operating Region

“Sweet Spot”Optimized operating point considering all constraints and maximizing throughput

5

Plant Control System

RTO Technical Solution

Optimal setpoints

Problem to solve

Implement best setpoints and ramp the plant

Actual process and pipeline values

Target and schedule setup

Real Time Information

Customer demand

Energy price

Process Model

Predefined

6

DCS

RTO IT Structure

Optimal Set Points

Ramp the PlantActual Process Values

Real Time Information

Process Model

Predefined

Air Separation

Unit

Real Time Optimizer

OPC Server APC

Expert System

Energy Price

Real Time Value Customer

Demand

Real Time Value

7

RTO Project Workflow

• Step 1: Plant Evaluation and Project Justification – KPI– Operating Environment

• Step 2: Scope Definition– Degrees of Freedom– Identifying Manipulated Variables

• Step 3: Plant Modeling– Controlled Variables, Objective Function, Constraints

• Step 4: Offline Optimization– Selection of Optimization Solvers

• Step 5: Online Implementation– Configuring Sampling Time, Solving Frequency, etc.– Designing Expert System– Connecting to DCS through OPC

8

Collaboration of a Cross-function Team

• Sponsor/Management– Project Justification

• Process Expert– Scope Definition– Process Modeling

• Operations– Expert System– Online Implementation

• Optimization Expert– Selection of Optimization Solvers– Model Configuration and Debugging

9

Case Study

C 2

D 2

C 1

ASU 1

ASU 2

K 1K-2

LOX Storage Tank

LAR Storage Tank

LIN Storage Tank

GOX

GAN

Compressed Air

P-6MV1

MV2

MV3

MV4

MV5 MV6

MV7

MV8

10

Plant Model

• Manipulated Variables– Air flow rate to the ASU I (MV1)

– GOX production rate of ASU I (MV2)

– Compressed air production rate of ASU I (MV3)

– LIN production rate of ASU I (MV4)

– Air flow rate to the ASU II (MV5)

– LIN production rate of ASU II (MV6)

– The status of the turbine (MV7)

– The flow rate through the turbine if it is on (MV8)

• Objective Function( ) ( )

( ) ( ) ( )

⋅+−⋅++⋅++

⋅+⋅++⋅+

eIIIairIIairIairGOXIIGOXIGOX

LARILARLINIILINILINLOXIILOXILOX

PkkPQQPQQ

PQPQQPQQ

,,,,

,,,,,max

11

Model Equations

• Controlled Variables (CV)– Mass Balance

– Regression from Historical Operation Data

IaircustomerairIIair

IGOXcustomerGOXIIGOX

QQQ

QQQ

,,,

,,,

−=−=

( )( )

( )( )( )IIairII

I

ILAR

IIGOXIILOX

ILOX

QMVfk

MVMVfk

MVMVfQ

MVMVMVQMVfQ

MVMVMVfQ

,55

314

213,

876,52,

4211,

,

,

,

,,,,

,,

==

===

12

Optimization Features and Online Implementation

• Optimization Features– Mixed Integer Nonlinear Programming (MINLP)

– Solver: AOA of AIMMS

– Nonconvex– Multi-start technique of AIMMS for global optimization

• Online Implementation– Model Configured in OnOpt– Connected to DCS through MatrikonOPC Data Calculator as the

expert system

• Performance– Both solver and communication are robust– Savings have been observed and are being evaluated

13

Conclusions

• Real time optimization can increase an air separation plant’s gross margin in a dynamic environment

• Investment is mainly software license and manpower• Cross functional team is needed. • The methodology can be easily applied to other process

plants