1

Petrocontrol

Coker advanced control andinferential modeling at BPGelsenkirchen refinery

Volker Haseloff, BP GelsenkirchenSean Goodhart, AMT Y. Zak Friedman, Petrocontrol

2

Petrocontrol

Coker unit overview

Naphtha

FEEDFURNACE

FEED

COKE DRUMS

MAINFRACTIONATOR

LCGO

HCGO

COKERFURNACE

3

Petrocontrol

Coker APC challenge

• Batch reactors (coke drums), integrated with acontinuous distillation process.

• Two completely different modes of operation:low sulfur and high sulfur

• Drum switching every twelve hours or so• After switching the unit is short of heat and

light products. The challenge is to keep theunit under control during that time.

• Product separation optimization is a secondarychallenge. Large product price differencesprovide incentive for precise product qualitycontrol.

4

Petrocontrol

Project objectives

• Control drum switch procedure– Warm up– Cool down– Throughput and inventories

• Control product qualities– Keep within range during drum switches– Optimize during steady operation

• Maximize feed

5

Petrocontrol

Design features

• Foxboro IA instruments, FoxboroDMCplus Bridge interface

• DMC+ multi-variable predictivecontroller (MVPC)

• First principles inferential models (FPM)

• FPM’s are coded on Aspen IQ platform

• Performance monitoring by Aspenwatch

6

Petrocontrol

Project highlights• Implementation by AMT– Functional design– Step testing– Coding FPMs– Commissioning

• FPM by Petrocontrol– GCC model for the fractionator– Coke drum outage model (in open loop)– Petrocontrol also acted as a client representative for

the project• Heavy participation by ROG– Step testing– Operator training– Commissioning support– Performance testing

7

Petrocontrol

DMC sub-controllers

• Feed surge drum “FEED”• Feed furnace “BA160”• Coker furnace “BA101”• Main fractionator “FRAC”

• Total size: about 90CVs, 30MVs,14FFWs• One set of DMC dynamic models

satisfy both modes of operation

8

Petrocontrol

Feed management

HCO

SRTANK

L0106

F0133VISB RESID

F0135

F0134

SL RESID

Surge drum

L0173

BLOWDOWNDA-109

F0112

LCGOGA-177

F0165

P0173

9

Petrocontrol

Feed management

• Feed surge drum

• Many feeds– Feed surge drum LC slave can be selected to

be any one of the feeds

• Recipe to be controlled

• Hydraulic constraints

• Feed management through drumswitches and warm-ups

10

Petrocontrol

Coker Fractionator

FC

LCGO

HGCO

To coker furnace

FC

To HP drum

FC

TC To HP drum

PC

FC

PC

Feed

Steam

Drum effluent

LC

FC

Furnace

FC

FC

FC

TLFC

11

Petrocontrol

Fractionator control

• Product quality control• Responding to drum switches• Column level – a key CV• Column bottom temperature• Hydraulic constraints

12

Petrocontrol

Key inferences

• Naphtha 90% point

• Heavy naphtha 5% point

• LCGO flash and 90% point

• HCGO 90% point

• Coke drum outage (in open loop)

13

Petrocontrol

Coker furnace

• Many constraints

• Pass balancing by skin temperatures• Throughput maximized to– Furnace constraints in both furnaces– Inventory constraints– Fractionator constraints, when the

fractionator sub-controller cannothandle them

14

Petrocontrol

Heavy naphtha 5 and 90% inference

30

50

70

90

110

130

150

170

190

210

230

7/30/04 0:00 9/18/04 0:00 11/7/04 0:00 12/27/04 0:00 2/15/05 0:00

HN05_M HN90_L HN05_L HN90_M

15

Petrocontrol

LCGO and HCGO 90% inference

300

320

340

360

380

400

420

440

460

480

500

7/30/2004 0:00 9/18/2004 0:00 11/7/2004 0:00 12/27/2004 0:00 2/15/2005 0:00

HCG90_M LCGO90_M L90_L H90_L

16

Petrocontrol

LCGO flash inference

10

20

30

40

50

60

70

80

90

100

110

7/30/2004 0:00 9/18/2004 0:00 11/7/2004 0:00 12/27/2004 0:00 2/15/2005 0:00

LFL_L FLLCGO_M2

17

Petrocontrol

Commissioning issues

• Difficult to identify coker real constraints pertaining todownstream units.

• The noise pattern of certain flow readings impededinferential model commissioning.Implemented validation routines and/or alternativemeasurements.Also applied custom validation for all CVs.

• Secondary constraints caused DMC to move furnacefeed flow unnecessarily during drum switchesDMC parameters were set to avoid throughputreduction unless no other route is possible.

• The application is now almost fully commissioned, andwe will report performance data in another paper.