Sie sind auf Seite 1von 7

18

th
European Symposium on Computer Aided Process Engineering ESCAPE 18
Bertrand Braunschweig and Xavier Joulia Editors!
" #$$8 Elsevier B%&%'(td% All rights reserved%
Integration of Engineering Process Control and
Statistical Control in Pulp and Paper Industry
Ana S% )atos* Jos+ ,% -e.uei/o* 0ulema (% Pereira
Dept. Mec & Ind Eng and UNIDEMI, Faculty of Science and Tecnology, New
University of is!on, "#"$%&'( )aparica, *ortugal
Abstract
1he main goal o2 this article is to present a methodology and a 2ramewor3 that is a4le to
4ring together two important concepts5 Engineering Process Control EPC!* which was
developed 4y process engineers to achieve short time control* and Statistical Process
Control SPC!* conceived and implemented 4y statisticians and .uality managers 2or
attaining medium and long term control% 1he integration o2 4oth concepts can represent
a 4rea3through in the 2inal product per2ormance* 4y creating the necessary conditions to
decrease the varia4ility o2 .uality characteristics 4oth in the short and long term% 1he
integrated methodology was designed 2or the pulp and paper industry and was
esta4lished in several phases% 6irst* a mathematical model was developed to represent as
much as possi4le the process dynamic 4ehaviour% 1he trans2er 2unction o4tained was
then used to implement two components o2 the a4ove mentioned concepts* namely
controllers* 4ased on the minimum variance criterion* and statistical control charts% At
last* the two components were integrated into the process* which was su4mitted to
several distur4ances to ascertain the control achieved with the integration% 1he
methodology was tested in a real industrial process o2 one o2 the most important pulp
producers in the world and considered several scenarios% 1o illustrate the methodology*
we present one o2 the scenarios that shows the 4ene2its o2 EPC'SPC integration%
1hrough the application o2 the developed methodology to real data* process engineers at
the company are now a4le to use a valua4le decision ma3ing tool when the production
process is a22ected 4y certain disruptions* with o4vious conse.uences on product
.uality* productivity and competitiveness%
Keywords5 Engineering Process Control* Statistical Process Control* Process
)odelling%
1. Introduction
Continuous improvement o2 any process re.uires reduction in the varia4ility around the
target value o2 its parameters% 1raditionally* two di22erent approaches have 4een used to
accomplish this goal5 Engineering Process Control EPC! developed and employed 4y
process and control engineers and Statistical Process Control SPC!* used 4y
statisticians and .uality engineers% 7ntil recently* the main reason 2or 3eeping these two
concepts separate was the di22erent view each o2 them had a4out an industrial process%
8hile SPC monitoring procedures see3 to reduce the output varia4ility 4y detecting and
eliminating assigna4le causes o2 variation* EPC is usually applied to minimi9e the
output varia4ility 4y ma3ing online ad/ustments o2 one or more process inputs on a
regular 4asis%
# +.Matos et al.
1he 2irst attempts to integrate EPC and SPC appeared long ago* with the wor3 o2
Barnard 1:;:!% 7sing the machine<tool case study* the author demonstrated that
automatic control and statistical control can 4e used in parallel%
1he 2irst approach o2 integration presented to the statistical community was developed
4y )ac,regor 1:88!* who suggested the use o2 control charts 2or monitoring the
4ehaviour o2 a process under EPC% =nspired 4y the wor3 o2 )ac,regor 1:88!* several
other authors 4ecame notorious in the 2ield* leading to di22erent approaches that reveal
two great concerns associated with this type o2 integration5
=denti2ication o2 the varia4les that must 4e monitored5 i2 only output varia4les
.uality characteristics!* input varia4les ad/usta4le varia4les! or 4oth o2 them>
?ecision on whether to use automatic or manual controllers* the latter 4eing or not
constrained 4y statistical control> such decision would depend on ad/ustment costs
and type o2 ad/ustment%
1he 2irst approach that e@plicitly com4ines SPC with EPC was proposed 4y &ander
8iel et al. 1::#! under the name o2 Algorithmic Statistical Process Control ASPC!%
6ollowing the same philosophy o2 ASPC* other re2erence studies emerged* such as
)ontgomery et al. #$$$! and Auang and (in #$$#!* among many others% An
innovation introduced to the ASPC approach was the use o2 control charts applied to
ad/usta4le varia4les> the /oint monitoring o2 input and output varia4les using di22erent
control charts! was presented 4y )essina et al. 1::B! and was 2ollowed 4y 1sung and
1sui #$$C!%
8ithin a slightly di22erent conte@t* a third approach appears which considered the
e@istence o2 ad/ustment costs% 1he implementation o2 control charts acting as a
DsupervisorE o2 the control actions was the way 2ound 4y several authors to minimi9e
ad/ustment costs e%g% Bo@ and (uceFo* 1::G and -uhhal et al.* #$$$!%
-ecent years have witnessed the appearance o2 several research studies in this 2ield%
Aowever* there have 4een scarce pu4lications using real production data and
demonstrating the practical application o2 the integration% 1his article tries to 2ill this
gap* as it gives an e@ample o2 integrated EPC'SPC applied to a real and comple@
continuous process% 1his new development was carried out in cooperation with a large
paper and pulp production plant in Portugal* which is one o2 the most important
producers in the world%
2. Developent of an EPC!SPC Integration "et#odology
1his article summarises an integrated EPC'SPC methodology developed 4y )atos
#$$B! as a doctoral research pro/ect% 1he main goal was the development o2 a
methodology that could 4e tested and applied to a real case study% Additionally* it
should also 4e ad/usta4le to other industrial processes with similar technical
re.uirements%
Hnce the pulp production process is well understood* the 2irst step is to characteri9e the
methodological DhardwareE and Dso2twareE elements that will 4e part o2 the integrated
EPC'SPC% 8ithin this wor3* the term hardware is used to descri4e the more physical
elements* such as the mathematical model* the controllers and the control charts% Hn the
other hand* the so2tware is related to the rather intangi4le elements* namely the
intervention criteria* the type o2 distur4ances 2or testing purposes! and the per2ormance
measures% 1he intervention criteria will constrain the rule to 4e applied 4y the controller
and the control chart* as well as the way they interact in the integration% Altogether* the
so2tware elements allow the evaluation and comparison o2 di22erent integration
scenarios%
Integration of E*) and S*) in *ulp and *aper Industry
C
As 6igure 1 shows* the main stages o2 the proposed methodology are the preliminary
data analysis* the identi2ication o2 all process varia4les outputs* inputs and distur4ance
varia4les!* the trans2er 2unction model i%e% the mathematical model that descri4es the
process 4ehaviour!* the controllers 4ased on the minimum variance criterion! and*
2inally* the univariate and multivariate control charts designed to 4e applied to auto<
correlated data!%

Hutput
varia4les

?istur4ance
varia4les

=nput
varia4les




Process
SPC
Preliinary data analysis
$ransfer%&unction "odel
Integrated "et#odology EPC ! SPC

EPC

Iuality
Characteristics

6igure 1 1he main stages o2 the integrated EPC'SPC methodology
".'. )ase Study , *ulp and *aper Industry
1he case study presented here deals with a pulp and paper production process% 1he plant
produces Bleached Eucalyptus Jra2t Pulp* using the EC6 process Elemental Chlorine
6ree!% 1he Jra2t pulping process is per2ormed in two di22erent phases* which in2luence
the 2inal pulp .uality5 the coo3ing process o2 wood chips eucalyptus glo!ules- 2ollowed
4y the pulp 4leaching% 1he coo3ing process is the phase that most contri4utes to the
preservation o2 the main pulp characteristics* which* in turn* will ensure high .uality
paper%
1he viscosity o2 the 4leached pulp* among other .uality characteristics* constitutes one
o2 the most important control parameters> the viscosity value depends* to a great e@tent*
on the coo3ing process carried out in two continuous digesters wor3ing in parallel%
A2ter understanding the )H .odus operandis! o2 the 4leached pulp process* the main
input varia4les which are measured in the digesters! were identi2ied 1a4le 1!%
1a4le 1 =nput varia4les and sym4ols
1emperature &aria4les Sym4ol Concentration &aria4les Sym4ol
1emperature in CK 9one
1emperature in C; 9one
1emperature in CB 9one
1emperature in C8 9one
Te.)/
Te.)&
Te.)(
Te.)#
Active<al3ali
Sulphur inde@
1op 4lac3 li.uor
8hite li.uor CK
8hite li.uor C8
++
SI
T0
1)/
1)#
1he present study considered 2our production periods with sta4ili9ed operational
conditions C temporal data windows 2or estimation and 1 2or validation!% 1he samples
were collected every K hours 2or 4oth input 1a4le 1! and output viscosity! varia4les%
K +.Matos et al.
".". Model Fitting
A2ter the preliminary data analysis* a satis2actory Bo@<Jen3ins trans2er 2unction model
was developed to descri4e* as much as possi4le* the dynamic 4ehaviour o2 the 4leached
pulp process% 1he methodology used to o4tain the trans2er 2unction model was carried
out in three phases* as 2ollows5
1
st
phase5 2itting o2 several single<input single<output S=SH! trans2er 2unction models
to identi2y possi4le relationships 4etween input < output varia4les%
#
nd
phase5 merging o2 the three data windows with the goal o2 o4taining a 4etter
pro2ile o2 the 4leached pulp process> the 1
st
phase was then repeated%
C
rd
phase5 development o2 a multiple<input single<output )=SH! trans2er 2unction
model using the results o2 the #
nd
phase%
1he o4tained model* which was developed using the 1ool4o@ System =denti2ication
2rom )A1(AB
L
so2tware* e@plained K#M o2 the total data variation within the
o4servation period5
( )
( )
( )
1 # # #
# 1 # 1
# #
1 #
#* 8B G* KB
8 ;* :$ K* $1 K ;* #C K
1
#* :$ ; 1* BB
1 $* ;B
D D D D
t t t t t
D D
t t t
y 1) SI 1) Te.)
Te.) ++



= +
+ +

B
1+ 0, 99B
B
1!
=n the previous e.uation* yt is the deviation o2 viscosity 2rom target at time t* t is the
white noise se.uence and
B
de2ines the 4ac3shi2t operator% 1)8* SI* 1)K* Te.)K*
Te.); and ++ represent the varia4les o2 1a4le 1 in digester 1 ?1! and digester # ?#!%
1he 2itted model was success2ully validated utili9ing the data o2 the K
th
window
esta4lished 2or that purpose% Such mathematical model is the 2irst DhardwareE element
and is the 2oundation o2 the complete methodology developed in the research%
".2. Engineering *rocess )ontrol and Statistical *rocess )ontrol
According to 6igure 1* once the mathematical model has 4een de2ined* the study carries
on with the de2inition o2 the other hardware components5 controllers and control charts%
1he integrated control strategy used manual controllers constrained 4y the control chart
decisions%
1he -idge controller del Castillo* #$$#! 4ased on a minimum variance criterion was
2ound to have good characteristics to 4e adapted to the trans2er 2unction de2ined in 1!%
1his controller has a tuning parameter that 4alances the variances o2 the output with the
inputs%
1he development o2 an appropriate monitoring scheme through control charts leads to
several .uestions* such as5 which characteristics are to 4e monitored* where to apply the
charts and which control charts are appropriate% 1o monitor the viscosity, three control
charts were considered as good candidates* namely the E8)A with residuals* the
C7SCH-E chart and the E8)AS1 chart* since they revealed to 4e more e22ective in
detecting changes o2 small magnitude than some other charts% =t was considered e.ually
important to apply multivariate control charts to monitor the input varia4les o2 the
digesters% 1he multivariate study was per2ormed using two control charts o2 the same
type* 4ut applied with di22erent goals% 1he 2irst one was applied directly to the digesters
input varia4les* whereas the second one was applied to the di22erence 4etween each
input varia4le real value! and the theoretical value shown 4y the controller% ,iven the
large amount o2 varia4les and the auto<correlated structure e@hi4ited 4y the data* the
Integration of E*) and S*) in *ulp and *aper Industry
;
control charts were 4ased on pro/ection methods* namely the dyna.ic principal
co.ponents analysis ?PCA!* proposed 4y Ju et al. 1::;!%
=n 4oth cases* i%e% controllers and control charts* the tuning parameters were o4tained
through simulation models developed on a )A1(AB
L
plat2orm%
'. (esults and Discussion of Integration )utputs
A2ter the de2inition o2 the three DhardwareE elements previously descri4ed* one has to
esta4lish the Dso2twareE elements5 intervention criterion* per2ormance measures* type o2
distur4ances and simulation conditions%
1he approach used in the research is somehow in 4etween the second and the third
approaches mentioned in the =ntroduction> on one hand* the integrated EPC'SPC
2ocuses on control charts to monitor 4oth the input and the output varia4les and* on the
other* it uses a constrained manual controller% 1here2ore* the intervention criterion was
esta4lished as a constrained action o2 the controllers 4y 4oth control charts%
1he +sy.ptotic Mean S3uare Deviation A)S?! was used as the per2ormance measure
to compare di22erent scenarios* since it incorporates the deviations 2rom the target and
the varia4ility particularly use2ul when a process is su4mitted to a distur4ance!%
Besides the use o2 A)S?* the +verage 4un engt5 A-(! was applied to evaluate the
per2ormance o2 each control chart%
,iven the main characteristics o2 the pulp process* it was possi4le to list seven di22erent
types o2 distur4ance that can a22ect the dynamic 4ehaviour o2 the process5 the input and
output varia4les* the autoregressive parameter and the residuals o2 the model * Nt!%
As regards the simulation conditions* the running o2 1$ $$$ cycles* with #K8
o4servations each* was considered su22iciently credi4le%
Since the proposed EPC'SPC was designed to 4e applied in se.uential stages* the
construction o2 the di22erent scenarios starts with an open<loop process* 2ollowed 4y the
incorporation o2 the control charts and then the manual controller% 1he scenarios are5
1! open loop system operating 2reely!* #! entry o2 univariate control chart* C! entry
o2 controllers manual regime!* K! entry o2 multivariate control chart applied to
controllers* and ;! entry o2 multivariate charts to control the input varia4les%
6igure # compares the 2ive scenarios when two distur4ances were applied to the
process% =n the 2igure* is the si9e o2 the shi2t in the mean* measured in terms o2 the
standard deviation new mean N O!%
0 0.5 1 1.5 2 2.5 3 3.5 4 4.5
Scenario 1 Scenario # Scenario C Scenario K Scenario ;
?istur4ance in Alcali<Active ++!
#K$$
#;$$
#B$$
#G$$
#8$$
#:$$
C$$$
C1$$
C#$$
$ $%; 1 1%; # #%; C C%; K
?istur4ance in noise component N
t
!
#K$$
#;$$
#B$$
#G$$
#8$$
#:$$
C$$$
C1$$
C#$$
$ $%; 1 1%; # #%; C C%; K
?istur4ance in noise component N
t
! ?istur4ance in Alcali<Active ++!
6igure # A)S? 2or the 2ive scenarios with implementation o2 E8)AS1 chart
shi2t shi2t









A
)
S
?









A
)
S
?
B +.Matos et al.
As can 4e seen in 6igure #* the scenarios present a di22erent 4ehaviour when the same
control chart E8)AS1! was used% =t is visi4le that the proposed methodology
increased the process per2ormance scenarios K and ;!* when compared with scenario 1
no control! or with the tradicional approaches scenarios # and C!%
*. Conclusions
1he success inherent to an integrated methodology o2 this nature is closely associated
with the a4ility o2 creating di22erent scenarios and the s3ill in comparing them% Hther
re.uirements e.ually important are the .uality o2 the mathematical model and the
selection o2 4oth the controllers and the control charts% Although this study has used real
production data to create the model* the employment o2 computational simulation
revealed to 4e an essential tool in accomplishing the aim o2 the research% 1he developed
simulation models were used 2or studying the sensitivity and ro4ustness o2 controllers
and control charts% Additionally* the simulation e@ercise is the only way o2 testing
di22erent scenarios when the process is su4mitted to several types o2 distur4ance%
As 2ar as the literature review revealed* the pulp and paper industry has not applied
integrated methodologies 4ased on D4lac3 4o@esE% According with the 2indings o2
present research* and once the 4est integrated scenario is o4tained and appropriate
inter2ace so2tware is developed* the process engineers can use the methodology as a
decision ma3ing tool when the production process is a22ected 4y certain disruptions*
with valua4le conse.uences on .uality* productivity and competitiveness%
Conse.uently* one e@pects that the company where the research too3 place will* in the
near 2uture* 4ene2it 2rom the implementation o2 the proposed appoach%
At last* it is also important to highlight the 2le@i4ility and adapta4ility o2 the
methodology to any other type o2 production system when the production sta22 can use
the availa4le data to 4uild a mathematical trans2er 2unction that models the dynamic
4ehaviour o2 the process%
(eferences
Barnard* ,% A%* 1:;:* Control Charts and Stochastic Processes* 6ournal of t5e 4oyal Statistical
Society 0, #1#!* #C:<#G1%
Bo@* ,% and (uceFo* A%* 1::G* Statistical )ontrol !y Monitoring and Feed!ac7 +d8ust.ent* John
8iley P Sons* QR%
del Castillo* E%* #$$#* Statistical *rocess +d8ust.ent for 9uality )ontrol, John 8iley P Sons*
QR%
Auang* B% and (in* R% (%* #$$#* ?ecision -ule o2 Assigna4le Causes -emoval under an SPC<EPC
=ntegration System* International 6ournal of Syste.s Science, CC1$!* 8;;<8BG%
Ju* 8%* Storer* -% A% and ,eorga3is* C%* 1::;* ?istur4ance ?etection and =solation 4y ?ynamic
Principal Component Analysis% )5e.o.etrics and Intelligent a!oratory Syste.s* C$*
1G:<1:B%
)ac,regor* 6%* 1:88* Hn<(ine Statistical Process Control* )5e.ical Engineering *rocess, Hct5
#1<C1%
)atos* A% S%* #$$B* Engen5aria de )ontrolo do *rocesso e )ontrolo Estat:stico da 9ualidade;
Metodologia de Integra<=o +plicada na Ind>stria da *asta de *apel* Ph? thesis* 6C1'7Q(*
(is4oa* Portugal%
)essina* 8% S%* )ontgomery* ?% C% and Jeats* J% B%* 1::B* Strategies for Statistical Monitoring
of Integral )ontrol for t5e )ontinuous *rocess Industries* Statistical Applications in Process
Control* Qew Ror3* )arcel ?e33er* KG* 1:C<#1K%
)ontgomery* ?% C%* Rats3ievitch* )% and )essina* 8% S%* #$$$* =ntegrating Statistical Process
)onitoring with 6eed2orward Control* 9uality and 4elia!ility Engineering International*
1BB!* ;1;<;#;%
Integration of E*) and S*) in *ulp and *aper Industry
G
-uhhal* Q% A%* -unger* ,% C% and ?umitrescu* )%* #$$$* Control Charts and 6eed4ac3
Ad/ustments 2or a Jump ?istur4ance )odel% 6ournal of 9uality Tec5nology, C#K!* CG:<C:K%
1sung* 6% and 1sui* J% (%* #$$C* A )ean<Shi2t Pattern Study on =ntegration o2 SPC and APC 2or
Process )onitoring* IIE Transactions, C;* #C1<#K#%
&ander 8iel* S% A%* 1uc3er* 8% 1%* 6altin* 6% 8% and ?ogana3soy* Q%* 1::#* Algorithmic
Statistical Process Control5 Concepts and an Application* Tec5no.etrics, CKC!* #8B<#:G%

Das könnte Ihnen auch gefallen