Chemical Process Control: Present Status 
and Future Needs - The View from 
European Industry 
Hans Schuler 
BASF Aktiengesellschaft, D-6700 Ludwigshafen, F.R.G. 
Frank Allgower, Emst-Dieter Gilles 
Inltitut fiir Systemdynamik und Regelungstechnik, Universitit Stuttgart, 
D-7000 Stuttgart 80, F.R.G. 
Ab.trad 
Not only in Europe, chemical process control iI characterized by a broad 
invuion of distributed control Iy.tems into chemical plantl. The information 
integration from procell control up to bwineSl management is a great challenge 
of today which follows from the overall computerization of production. MOlt 
of the recent progresl in process automation re.ults from the application of 
computer lcience paradigma to control Iy.tema, and of advanced developments 
in field inltrumentation. Despite these advancel and the considerable progress 
made in procell control theory, there i. only limited acceptance and application 
of modem advanced proce •• control methodologies in indultrial practice. Thi. 
paper iI an attempt to lummarize the European di.cussion on the reasons for 
these fact •. 
Keywords: Chemical process control, computer integrated production, advanced con-
trol, quality control, distributed control systems. 
29 
30 
THE VIEW FROM INDUSTRY 
1 Introduction 
With respect to chemical proceaa control, the lut decade, the eighties, can be char-
acterized by the broad invuion of distributed control .y.tem. (DeS) into chemical 
planll. For illu.lration, Fig. 1 .how .... mplary the numb.r or in.tall.d DCS in BASF 
AG, Ludwig.har.n. Th. Ludwig,haC.n compl.x or BASF conlain. aboul 360 chemic&! 
production plants wilh ,&! •• or aboul 22 Billion DM p.r y.ar. Th •• xl.nd.d runction· 
,so ,--- -,----, 
"'" I-----r---;ff-i 
sof----~~ 
" 
os 
Figur. 1: Numb.r of in.lall.d DCS, al BASF Ludwig,haC •• 
ality of computeria:ed procet. control equipment led to &Il increuing importa.nce of 
prace .. automation especially in newly installed plub. A. an eX&mplc, Fis:. 2 ,hOWl 
the el.ctric&! and conlrol'quipment COlli r.lated 10 Ih. co.1I of machinery and planll 
ov.r Ih.l .. ly.a", al BASF AG. Thi. groph upr ..... Ih •• trong Ir.nd 10 inot&!l im-
portant plant functions in the automation part of equipment. Thi, trend will probably 
continue due to upidly increasins computer power &D.d Itorage capacities, u well u 
the improving possibilities of computer integration at all fundionallevela of procesl 
control and plant ma.nagement. 
This rapid technical evolution is Dot only a consequence of the exponential srowth or 
microelectronici and or digital computers. It is stimulated by Itrons trend. in chemical 
engineering and pla.nt operation u well: 
The View from European Industry 31 
% 
10 
_ r-
-
60 ..... ,j 
'V 
40 ~ 
v'V 
20 
o 
77 81 85 89 
Figure 2: Ratio of electrical and control equipment costs to those of machinery and 
plants, at BASF Ludwigshafen 
• Plants are designed to be highly integrated in the mass, energy and information 
flows. 
• Plants should allow a flexible production of different products at different 
throughputs. 
• Production must be tracked to market demands without delays, just in time. 
• Information becomes an important production factor. 
• Product quality is superior to mass production. Quality requires transparent 
and reproducible plant operation. 
• Safety and environmental requirements give new constraints to plant design and 
operation. 
• Administrative regulations require an exhaustive documentation of plant opera-
tion principles and production runs. 
• The global competition requires an increase of productivity, and thus a contin-
uous cost reduction of production and of maintenance. 
32 
THE VIEW FROM INDUSTRY 
It is • common opinion that advanced control technique. will help to Collow these 
trend •. 
[n the (ollowing, the present indultrialstate of the increuingly important disci-
pline of procesl control will be analysed. Especially the impact of advanced process 
conlrollheoryon Ih. pr ••• nl .Ial. will be deall wilh in del ail. 
2 Present status of industrial process control ap-
plications 
The Hierarchical Layer Model of Plant Management and Control 
The organization of the following statements will be oriented to the hierarchical Ilruc-
ture of plant man_sement and control functions . Thi. layer model i, depicted in Fig. 3. 
It help. 10 clas.ify Ihe I .... of planl manasement and control funclion. due 10 Ihe 
Dumber of specified operational details, and due to tbe horizon of operative planning. 
The Ipecific (unctionl of the I.yeu will be discussed ill the followins with respect to 
their present state in chemical industty. The degree of computerized information pro-
_d_ 
--
........... p.". 'Ii 
.... .,. 
Fisure 3: Hierarchical layer model of plant ma.na.gement and control (unction. 
cessing differJ substantially in the different layers. There are deficiencies in the plant 
The View from European Industry 33 
management layer, whereas business administration and process control make use of 
almost complete information processing capabilities. 
The Field Layer: Sensors and Actuators 
Standardized measurement instrumentation in chemical process industries is mostly 
restricted to a limited number of important process variables, like temperature, pres-
sure, flow, level, etc .. The field instrumentation for these variables is designed to be 
robust and highly reliable. With respect to measurement range sensitivity and abso-
lute accuracy it is normally not as precise, as required in most of advanced process 
control concepts. 
Many other procesl variables, especially those repre~enting product properties 
and product quality, are measured continuously only in rare cases. That means that 
only a small part of the plant state vector is available as sensorial information. When 
process or product properties are measured, the signals represent not the state variables 
separately, but, in general, a nonlinear interference of many states. A typical example 
of these indirect measurements is the density which is a function of temperature, 
pressure and composition of the process mixture. 
On-line-measurements of composition and quality are provided (e.g. by chro-
matographs) in discrete times, and are normally corrupted by significant time lags. 
These instruments often need intensive maintenance. 
Field instrumentation covers the major part of investments in automation equip-
ment. According to the requirements of quality, operability, environment and safety, 
the number of measurements still increases continuously in chemical plants. Especially 
quality control will require additional process-analytical measurements not yet avail-
able for production conditions. There is a trend to apply new measurement principles 
providing additional process variables at least quasi-continuously. 
Another trend is the fusion of sensor elements with microcomputer components. 
These sensors called "smart" or "intelligent" are often based on digital principles. They 
allow data processing, check of plausibility and functionality, automatic calibration, 
and other features. These functions of decentralized intelligence provide a qualitative 
improvement of sensor data quality. 
The digital realization supports the mutual communication of individual sen-
34 
THE VIEW FROM INDUSTRY 
10rs, and the bidirectional communication with automation systems o{ the superim-
posed layer. There are intensive developments and standardization efforts on field bus 
communication protocols and {unctionalities. The communication o{ instruments of 
different suppliers i. mostly impossible at the moment. 
Actuators are the second fundament of the field layer. Despite o{ large progress 
in instrumentation, the actuator {unction. are blind spote in control theory intere.ts. 
The Process Management and Control Layer 
Process controls installed in production environment have a number of specific char-
acteristics: 
• Controllers are operated using predefined blocks in D\J:)-software. This tech-
nique results in a user-friendly interface to operators and t.o maintenance. 
• A characteristic feature of DCS-implementation is the great flexibility to make 
structural variations in the automation strategy. 
• The implementation of controls is mostly restricted to the capabilities of prede-
fined DCS-software. An implementation of advanced control concepts is difficult 
when special information processing structure. are required which are not avail-
able in the DCS toolbox. 
• Some DCS offer integrated specialized computer modulel, or programmable lOft-
ware modules, in order to allow the user to do procesl-Ipecific calculationl. 
• Single-loop controls are used in the vast majority of applicationl. Loop-pairing 
is based on physical, intuitive principles, and on trial and error during plant 
startup. Controller design packagel are used only in exceptional cases. 
• Simple loop nestling .trategies like cascading etc. are frequently used in DCS-
realizations. The same holds true for different forms of .tructure-selective con-
trols. 
• Multivariable control is done sometimes in the form of decoupling control (e.g. {or 
distillation columns (Trilling and Kaibel, 1980». Another multi variable struc-
The View from European Industry 
ture sometimes used is sensor data fusion to calculate control variables which 
are controlled in single-loop. 
• Controllers based on transfer function models are used for dead time compensa-
tion. 
• Controllers based on physicochemical models are used in few applications. 
• Adaptive controllers often work satisfactorily when based on heuristic principles 
(Krahl et al., 1986). 
• Predictive control schemes are implemented at some plants (Froisy and Richalet, 
1986). 
• Knowledge-based controls are far away from industrial acceptance. Only a few 
research-based applications are known (Soltysiak, 1989), many attempts failed 
in the past years (Ahrens, 1990). 
• Neural network approaches are in the state of academic research. No industrial 
application to chemical processes is known to the authors. 
• Maintenance of advanced controls is normally done by· special groups of highly 
qualified control experts. Personal continuity is not always provided in these 
groups. 
• In the past, unreliable computer hardware (e.g. micro-computers) in conven-
tionally instrumented plants led to loss of credibility of advanced methods in 
general. 
The predominance of continuous control in control theory should not divert from the 
fact that logic control is very important even in chemical process control. The applica-
tion of programmable logic controllers (PLC) in the process industries is in the same 
order as that of DCS. Especially in batch process control, there are broad applications 
including logic, sequential and recipe control in a variety of instrumentation types. 
Each potent vendor of DCS offers such batch control packages. The academic control 
community has ignored this development substantially. There are fundamental ques-
tions of hierarchical structuring of these logic controls, and the inclusion of continuous 
35 
36 THE VIEW FROM INDUSTRY 
controls with Iwitching model. These questions are treated in practice heuristically, 
on an intuitive level. 
The process control level doe. Dot only contain fundion. of automatic process 
control. Other specific {unctions are also included, th.t provide better insight into the 
actual process behaviour. and will help to operate procesles in a smut way ("smart 
operating"). Some of these fundions will be quoted in the following. 
Display functions. The represcntation of procesl variables on Icreen display. 
has changed the operator monitoring and control interface. totally during the last 
yeall. These displays offer new opportunities to man-machine-interfaces. There i, & 
tendency to usc not on1y display. or recorder-like trend. for procesl visualization, but 
also additional displays,like profiles, phase plots, <tc. Further improvements of display 
functionality i, required for ,ituationl with abnormal plant behaviour, in which Icreen 
dilplaYI are not as ft.exible a.a il required in pla.nt operation. 
Sensor data processing. These {unctions art usually located in the sensor 
systems of the field layer. They include data compression Crom multi-sensor systems, 
the use of sensor models to provide more evident process in£orma.tion (Schuler, 1990), 
and others. More ambitious applications, like Kalman-filters, require more computer 
power, and are therefore implemented in the DeS environment. 
Model-based measurement tools. The application of Kalman·filters and 
Luenberger-observers solves many practical problems even in industrial-Icale plants 
(Schuler, 1986; Bachmann and vom Felde, 1988). These methods provide sensor data 
interpretation by procels models and allow & reconltruction of unmeasured process 
varia.bles Crom these models. By doing that, they 10Ive .. buic problem of chemical 
process control: they improve the insight to the actual procell behaviour. The ap-
plication of these model·based measurement tools i. limited by the lack of adequate 
process models. 
Plant diagnosis. Diagnosis i, normally done by checking meuured variables to 
predefined constant alarm limits. There are a few approaches to shift these alarm limits 
depending on operating conditions, e.g. during pla.nt startup and shutdown. Other 
approaches use characteristic variables (calculated due to mathematical principles, or 
to models) as a basis of diagnosis . The use of knowledge-based Iystems i, far away 
Crom industrial acceptance. 
The View from European Industry 
Plant safety. Safety functions are not allowed to be realized in software but , 
in highly reliable hardware. Advanced diagnostic methods may be applied only for 
monitoring and supervision functions. 
Statistical process control, statistical quality control. Initiated by buyers 
of chemical products from other industries, statistical methods of process and qual-
ity control command considerable attention in chemical industry. These statistical 
methods originally developed in manufacturing industries are not always adequate in 
chemical process control: when process variables are provided by sensors continuously 
and in real-time, and when process behaviour can be controlled by suitable manipu-
lated variables, conventional process control of variables influencing product quality 
is obvious. Elementary statistics in on-line process control will introduce dead-time 
and threshold-type nonlinearities to the control loop, and its design does not consider 
process dynamics. We believe that classical statistical process control (SPC) is re-
stricted to a very small part of process control. Statistical quality control (SQC), as 
a method of monitoring control product properties, is of great importance for process 
improvement, and for documentation of production results. 
Robotics. Chemical process automation covers applications which are similar to 
the automation of manufacturing processes. Some of these applications are: the charge 
of sacks or barrels on palettes, the handling of samples in analytical laboratories, the 
cleaning of vessels containing aggressive atmospheres, the filling of multi-tubular fixed 
bed reactors with catalysts, etc .. In such applications, robots are applied with great 
success. Robots will continue to invade chemical industries in some few special appli-
cations. There is no need for new types of robots, but a creative look to corresponding 
problems of plant operation and management. 
The Plant Management Layer 
The degree of information processing in the plant management layer is relatively low 
compared to the adjoining layers of process control and production management. The 
plant management functions are considered by many vendors of application software 
as an increasing market of the upcoming years. There is a strong trend to introduce 
computerized tools for these functions. Typical functions of the plant management 
level will be touched upon in the following. 
37 
38 THE VIEW FROM INDUSTRY 
Batch plant management. Satch processel arc controlled in a rueruchical 
sequence of controllayen. It i. based on the lundamental concept of unit operation •. 
Unit operation. are understood in this contcxt u logical operation units needed to 
describe the implementation of a recipe. The a.5lisnment of a recipe to the plant, and 
of the amounh of react anti to the ba.tch, is &D importa.nt tuk of plant management. 
Thi. task include. batch Icheduling and batch .equencing, production logi.tics, etc. 
Whercu hatch control tends to be fully automated, thele functioDs of plant manage-
ment ue ruely integrated in the DeS·environment. Some {unction ... re stiU per{ormed 
manually even in multi-product plants. 
Automation or plant startup and Ihutdown. At pre.ent, plant Itartup 
and .hutdown need a lot of operator intervention and mlLIlual action. These opera· 
tions have to be supervised by the plant mana&cr. Similu procedures are applied to 
throughput variationl and to handling of di.turbances. The automation of the.e oper· 
ations requires a complex intera.ctioo of lopc, sequential and continuous controls. The 
design of these cootrol strategies il so far hued on intuitive and phYlical principle •. 
Theoretical concepti ue Dot uled except of simulation studies. 
Optimization of large-Ieale planll. At large-scale plant., advanced con· 
trol and optimization .how sometimes significant economic incentives. These plants 
(e.g. refineries, ethylene and ammonia plants) are operated uling .imilar principles all 
over the world. That's why advanced control and optimization techniques have been 
worked out for theae plantl by different engineering companies. In mOlt applicationl, 
advanced controa work latisfactorily, where .. the on-line optimization tools are oCten 
at standstilL The optimization package. need often a lot of economic data. not entered 
automatically to the IYltems. Rigoroul procesl model. are used, and optimization is 
typically constrained by economical and technical limits. 
Operator training. The hiSh degree or automation, and the advanced Itate 
of procesa design, lead often to a very constant operation of many planb. In this 
lituation, operatofl are not alway. familiar with plant operation in the case of Ipo--
tadie disturba.n.cel. In order to traiD operatoll to abnormallituations, limulaton are 
operated using the s&me operator interface u the teal procell, and the limulation is 
run in real· time. There ate some verYluccenful applications of simulators mainly for 
di.tillation column. (Gilles et aI., 1990; Schuler et al., 1990). 
The View from European Industry 
Documentation of plant data. Administrative regulations and customer 
needs require an exhaustive documentation of plant data. The databases of plant 
runs are used for administrative purposes in quality and environmental management. 
It should be stimulated to use these data also for operational purposes, e.g. as a pat. 
tern store of plant situations, for plant optimization, diagnosis and model fitting. 
The Production Management Layer 
When coming up Crom process control to production management, the number of ad-
ministrative functions increases substantially. These administrative functions shall 
not be treated in this context. We will restrict to those functions which are directly 
connected to the plant management and process control layer. 
39 
Production planning. Production planning systems are sometimes used in 
practice and are implemented on digital computers. Data integration to plant man-
agement, with connections to batch scheduling and recipe control, is only realized in 
very few examples. These interfaces are needed with high priority. This deficiency is 
not only a question of computer communication, but also a basic problem of control 
system structuring. 
3 Present status of process control research in Eu-
rope 
To make general statements about specific European research interests and trends is 
& very difficult task if one wants to generalize for the whole of what is geographically 
called Europe. The diversity Crom Poland to England, Crom Greece to Norway is 
probably more striking than anything which might be in common. Therefore what 
follows can not be considered the status of European process control research or the 
European view. Most probably it is influenced by the ract that all authors &tem 
from Germany. Nevertheless, there are some differences between European and non-
European process control research orientation. These differences shall be pointed out 
in the following. 
Looking into the academic process control literature reveals some areas of research 
where most effort is put in worldwide at the moment. This is in particular the huge 
40 TIlE VIEW FROM INDUSTRY 
area of predictive control, nonlinear control with its various directions, lobuli linear 
multi variable control, and AI control, learning .ysteml, fuzzy control and the like. 
Looking only al Europe and whal i. published by European researchers, all Ihe above 
areu arc also present, but much less distinct. In addition there i. a .ttong inteteat in 
adaptive control and in the development of cootrolltrateKies for specific processes and 
problems. Generalizing, European process control rescarch il closer to the process, 
while e.g. American control research is mOIUy more abstrad and closer to control 
Iheory. This i. also .xpr •••• d in Ih. facl, Ihal in Europ. d.parlmenlal laboralories 
a.re usually very good equipped. A reuon for thi. orientation of European procesa 
control rese&rch can be seen in the different educalion for chemical engineers, and aliO 
in the difference in who i. doing process control research: if we look at the backsround 
of graduale chemical engineers (.quivalenllo Masle, or Ph.D.), Ih.n Ih. amounl of 
formal mathematical training i. much lell .tressed in many parh of Europe than e.g. 
in the USA. There i. also a greater emphasis on thermodynamical Cundamentals in 
European .tudent education and less effort is put into the special properties oC vanous 
units. 
From a operational view, processes can be divided into two main group.: con· 
tinuou. processes and batch processes. A third uea, the control of entire pla.nts, also 
reveals different operational properties. In difference to the problems leen by industry, 
academic rcse&rch ia mostly concerned with control of continuoua proceues and only 
little research ia conducted on batch and plant. wide control. 
It i. widely recognized, that there ia a very doae relationah.lp between deaign 
a.nd control of chemical procellcs. There i. no doubt that control atrongly depends 
on design, but the Ipecific needs and requirements of controllhould also be embodied 
in the design atep. Nevertheless, organization of science in Europe usually separates 
design and control to a great extend, disregarding this strong interference. This aeems 
to be different to other regions, .., e.g. the US, where design and control are very often 
represented by the same people within chemical engineering departments. 
On the other hand, there are very few research groups in European chemical engineer. 
ing, that only work on controller design. In order to be able to practically apply new 
and advanced control .trategiea, auitable procell models have to be available. Most 
probably out of this need, there are strong research efrorh within process control groups 
The View from European Industry 41 
towards modelling, identification and dynamic simulation of chemical processes (Holl 
et aI., 1988; Perkins and Sargent, 1982; Marquardt, 1991). Also control related areas 
like estimation, observer design or fault detection play an important role. 
Research efforts that concern the management layers can be observed, but do not play 
by far the same central role as in industrial practice. 
It is also interesting to note, that the total number of people being involved 
in process control research is much smaller than the comparable number e.g. in the 
US. This i. not only true for universities, but also for research laboratories, whether 
they are sponsored by industry or by the respective government. There are even few 
research labs outside universities concerned with process control in Europe. 
On the other hand there is quite a number of people outside chemi<-al engineering 
departments who use chemical engineering control problems to test their theories. 
Altogether it is interesting to note that European research orientation in process 
control differs from other region., although not tremendously. This is most likely due 
to a different research organization and student education. 
4 Discrepancy between the status of industrial 
application and of process control research 
Modern control theory has been applied very successfully in aerospace, mechanical and 
electrical industries. Thi. luccessful development is still lasting, remember the trends 
in robotics, mechatronics, and others. The use of control theory has never been as 
spectacular in chemical process applications. Early expectations to transfer available 
methods to this discipline could not be completely fulfilled . Nevertheless, there is a 
broad academic literature on methods and on possible applications in chemical indus-
tries. At present, a considerable discrepancy can be observed between the research 
interests of academia and the implemented control concepts in industry. More impor-
tantly, industry sees the most urgent needs in fields other than the design of specific 
control algorithms. But it should be noted at this point, that the discrepancy in Eu-
rope has never been unbridgeably big. A considerable number of European research 
groups is working on many of the problems stressed by industry. The questions remain: 
42 THE VIEW FROM INDUSTRY 
Why is there not more use of advanced control concepts in industrial practice? Why is 
there not more rcscarch in directions where industry see. the most urgent problema? 
The following two subsections will summarize this longterm discussion in the view of 
the present European situ&tion. 
4.1 Industrial Perspective on Limited Acceptance of Ad-
vanced Control Techniques 
Advanced control techniquel are applied in chemical indultry only to a limited degree. 
The reasons for that situation can be found in the inherent properties of the procelSes, 
the practical requirements to procesl controls, and in the state of available theories as 
well . 
Properties of chemical processes. With rei peel to proce •• control., chemical 
processes are characterized by & number of specific featurel : 
• Plant behaviour is in mOlt cues not perfectly known. Even rigorous models are 
often not adequate to predict plant beha.viour with latilfying accuncy. 
• When modellins is bued on physicochemical principles, even simple models are 
nonlinear. Also, differential· algebraic Iyltem., distributed parameter .y.tem., 
differential-integral equation. are very common modelling result. . 
• When linearizing these equations, .. lot of physical information is 101t. 
• Plants &re generally of multi variable nature. The dimension of models is lome-
time. very high. Together with the nonlinearitie. , thi. results in .trongly inter-
acting dynamical systems of high complexity. 
• Ma.nipulated variables as well as controlled variables &re normally constrained. 
• A great part of procesl state va.ria.bles are not meuurcd continuously. Variables 
describing product quality are usually not mea.sured, and i f, thcse measurements 
arc intcrfered by other in8.uence., or are provided in discrete times a.nd with 
considerable time lags. 
The View from European Industry 
• Some of the sensor signals are corrupted by considerable noise. Sources of noise 
may be fluctuations due to pump movements, and others. 
• There is a natural variability in the process due to raw material quality fluctua-
tions, and due to unsteady environmental conditions. 
• Dynamics is important in most aspects of chemical process control and operation. 
• The occurrence of dead time is evident as a consequence of transportation lag, 
and sensor time lags, etc .. 
• The time constants of many processes are rather high. Dynamic transients are 
time consuming and expensive. 
• Plant behaviour is often time-variant due to catalyst activity variations, heat 
transfer decay, etc .. 
• The behaviour of a plant can change dramatically during operation. (e.g., when 
another mixture is separated in the same distillation wlumn, or another reaction 
is run in the same reactor). 
• The localization of disturbance sources in process and model structure is in most 
cases not trivial. 
• The configuration of plants is changed during their lifetime . 
• The variety of processes is very high, each process is a unique construction. Only 
a few processes are operated in a larger number of pieces. 
It is obvious from the above, that most chemical processes are very difficult to describe 
analytically. Nevertheless most of the process control problems that appear in indus-
trial practice can successfully be solved by applying very simple and intuitive control 
strategies. For these more than 95% of control problems, no significant improvement 
can be achieved using advanced control strategies. 
Practical requirements to chemical process control. Plants are operated 
under the principles of profitability, transparency, simplicity, maintainability, safety. 
These principles shall be explained in more detail with respect to process control: 
43 
44 
THE VIEW FROM INDUSTRY 
• Control structures must be robust to disturbances, malfunctions, and failures, 
and should not cause serious maintenance problems. 
• Often implementation oC highly sophisticated control strategies does not satisfy 
the requirements to safety, reliability and maintainability. 
• Controller structure must be made evident to responsible plant manageu not 
familiar with control theory. 
• Plant reconfiguration should be tolerated (or automatically detected and 
tracked) by the control system. 
• Maintenance must be possible by personnel without academic qualification. 
• The successful operation of advanced control strategies should not depend on 
the support of single individual persons. 
• Dynamic experiments (e.g. for recording transients and dynamic matrices) are 
tolerated rarely and only exceptionally in production plants. 
• Economic incentives are oCten found only with difficulties for advanced control 
strategies. A prerequisite oC advanced process control is an investment in analyt-
ics, computer hardware and in control system development. These high set-up 
costs cannot always be compensated by the expected improvements. 
• Operator and maintenance interfaces oC advanced controls lack oC accepted stan-
dards. 
• Operators need not more, but better inCormation about the state of the process. 
• Continuously operated plants need good stationary control and disturbance re-
jection. Startup and shutdown are very infrequent. 
• The startup and shutdown oC batch processes is 50 Car a major problem of logic 
control I.nd sequential control, not of classical optimal control. 
Practically available control theory. The inherent deficiencies oC theories, or the 
fact that the theories are not appropriate Cor solving the problems often prevent the 
application of control theories. Some reasons are: 
The View from European Industry 
• The lack of proven process models prevents the successful application of most 
theoretical concepts. 
• Many common mathematical forms of physicochemical process models can 
hardly be handled by available control theory. 
• Linear theory is sometimes (often?) inadequate. Nonlinear dynamics is not 
incorporated in process control theory. 
• There is no proven theory for the design of robust nonlinear multi variable con-
trollers. 
• Some complex control strategies lack of robustness. 
• There is no fully developed theory of discrete, logic, sequential control. 
• In general, there are practically no sufficiently customized problem-solving meth-
ods relying on realistic process models. 
4.2 Perspective from Research Community on the Discrep-
ancy between Industrial Practice and Academic Re-
search 
In the last section it was stated that more than 95% of all control problems that 
appear in chemical plants are very good-natured, although the underlying process 
dynamics might be analytically very complex. In many cases the most fundamental 
knowledge about control suffices for adjusting the tuning parameters in the respective 
PI- or sometimes PID-controllers. Of course, these control problems are not the ones 
that make practicing control engineers worry, neither do academic researchers try to 
find new and advanced solutions for these control problems. The approximately 5% 
remaining demanding control problems are the ones we will direct our attention on. 
Considerable improvements in the quality of the products, in the reduction of energy 
consumption, in the safety and operability of the processes or in the diminution of the 
environmental impact can in some cases be achieved through application of "better" 
control algorithms. Despite the relatively small number of such cases, efficiency and 
45 
46 THE VIEW FROM INDUSTRY 
profitability o( chemical plants can thereby increased notably. 
If better control algorithms would exist, some design practices, that were necessary 
because o( quality requirements, could be changed. E.g. in distillation, a column usu-
ally consists o( more trays than needed (or achieving a desired stationary separation o( 
the ingoing compound. Because of possible disturbances additional trays are included. 
If those trays are included column design is not optimal, but the control problem is 
rather simple. If those additional trays are left out, the column is well designed, but 
difficult to control. This over-dimensioning is common practice in industry. 
As stated in Section 2, there is also a tendency to not only look at single units within 
a chemical plant, but to consider the plant as a whole. By explicit consideration of 
the interconnections, which are sometimes even material or energy "feedback"-loops, 
again a much higher efficiency of design and operation can be achieved. Of course this 
trend creates also new demands on the control concepts needed. 
Because o( these reasons industry is expected to (ace a considerably growing number 
o( difficult control problems in the years to come. There is a definite and growing need 
(or a process control theory which is able to handle the nontrivial cOlltrol problems 
addressed above. 
Because of the specific facts existing about chemical processes, that were ex-
plained in more detail in the previous section (like nonlinearity in process dynamics 
and sensors, uncertain dynamics, distributed parameters, underlying algebraic equa-
tions, non-measurable states, high order, ... ) it is a very difficult venture to find 
generally applicable design and analysis methods. It needs no great prophetic gift to 
(oretell that such a complete and perfect theory will not be developed in this mille-
mum. 
A fatal conclusion o( this realistic insight would be to condemn control theory at all. 
Not only the last decade has proved, that progress can be made. But of course progress 
will only be achieved one step after the other. Linear geometric control theory, known 
(or its limited realistic applicability (Shimizu and Matsubara, 1985), was necessary in 
order to be able to develop nonlinear differential geometric control theory, which is, 
due to its nonlinear nature, much closer to reality. 
There are two fields of activity desirable, where research should be conducted: 
the first is to provide practical solutions for the control problems that exist at present. 
The View from European Industry 
And the second, actually more important area, is the development of a realistic (i.e. 
nonlinear, robust, ... ) and applicable controller analysis and design theory (or the 
future. These two fields are not disjunctive, although the objective is different. For 
the first field, applicability is a short term objective, (or the second, applicability is 
necessarily a long term objective. 
In the preceding we often used the term proceu control theory. It is our strong 
believe, that special properties of chemical processes, that are usually seen as compli-
cations for purposes of control, can at least partially be exploited for a specific proceu 
control design and analysis theory. Of course a greater expenditure is necessary for 
understanding the respective process dynamics and for modelling these dynamics ap-
propriately. By means of the insight won about the process, the actual -:ontroller 
design step is then rather straightforward (Retzbach, 1986b). Another advantage of 
this approach is that control specifications, due to physically motivated quality re-
quirements, can be formulated and considered in a much simpler way. 
47 
On the other hand, when applying "general" control theory, much less effort has to be 
put into modelling the process dynamics, although, here too, better modelling always 
pays off. The control specifications, motivated from physical quality requirements, 
have to be "translated" into a suitable form for the controller design methodology to 
be applied. For example, when designing an Hoc-controller for a distillation column, 
suitable weighting matrices (frequency domain!) have to be found to reflect the qual-
ity requirements on the separation products. There is a lot of practical knowledge 
and experience to do this, and in many cases this "translation" is possible without 
causing too much trouble. In many other cases, however, a lot of design iterations 
with repeated adjustment of design specifications is necessary, sometimes even leading 
to unnecessarily conservative controllers. The smaller effort for modelling often faces 
a greater effort for the controller design step. The great advantage of this approach is 
the general applicability to practically all control problems. A phenomenon-oriented 
process control design methodology is usually only applicable to the restricted class of 
processes for which it was developed. 
But no matter how we approach the control problems, there is always a number 
of prerequisites that have to be fulfilled in order to be able to develop a practically 
meaningful control theory: 
48 
TIlE VIEW FROM INDUSTRY 
• For the processel to be controlled, there must be a sufficient understanding of 
the physical and chemical principles that govern the stationary and dynamic 
behaviour. 
• There must be a good mathematical model for the process. "Good" meamng, 
for the purpose to use. 
• There must be a powerful and comfortably usable dynamic simulation tool for 
chemical processes for reasons of identification, controller test a.s.o .. 
• Finally, there should be the possibility to test controllers at pilot plants and to 
verify model assumptions made about the process. 
Especially this last point plays a very important role. 
Our experience .hows, that good and meaningful resulu may be achieved when con· 
troller design methodologies are applied with care. Applicationl to binary and multi· 
component distillation, as well as to fixed-bed reactors show that especially nonlinear 
design philosophies (e.g. exact linearization) (Allgower et al., 1989), linear multivari-
able designs (Allgower and Raisch, 1988) (e.g. Hac- or H2-optimization) and phe-
nomenon oriented approaches (Retzbach, 1986a) lead to very satisfying performance 
of the closed loop, with reasonable effort needed to get the respective controllers. Some 
control schemes were implemented at a distillation column in pilot plant scale, sat-
isfying industrial standards. Due to the very realistic application, the good results 
achieved there are also meaningful for industry and show that considerable advances 
can be achieved using the, admittedly incomplete, control theory of today. 
A promising field of research for the future, that is stressed by industry, is con-
trol of batch processes and startup/shutdown automation. Recent developments in 
discrete event control theory and related areas show very appealing results. There is 
for example a rather well developed theory in the framework or state machines and 
formal languages (Ramadge and Wonham, 1987). Other notable approaches are Petri-
Net theory (Peterson, 1981), branching-time temporal logics (Emerson and Srinivasan, 
1989), Boolean differential equations (Bochmann and Post hoff, 1981), and several ap-
proaches using algebras of concurrent processes (Heyman, 1990; de Bakker et al., 1989; 
Brooks et al., 1984). 
The View from European Industry 49 
For all the important advancements, that we can expect for the following years 
and decades, especially academia has to be cautious not to let a second gap open, a 
gap in information, knowledge and interests between universities and industry. 
5 Conclusions 
Presenting an European view on the status of process control in industry and its 
relations to the research community is a difficult venture due to the diversity that 
makes Europe what it is. In 1992 there will be the start of the European Common 
Market which might lead to a more homogeneous picture for the future. Despite 
this diversity practically all regions have one thing in comm?n: the different view 
from industry and from research community on the relevant topics in process control. 
Industry is interested in easy-to-handle support for improving process operation and 
plant management. Academia is mostly concerned with theories for control system 
analysis and design. Short-term goals are different, but many long-term goals are very 
similar: 
• Process design and process control should be treated as an entirety. This is a 
very important perspective of both industry and academia. 
• Integrated software tools for modelling, simulation and control system analysis 
and design are needed. 
• Control system design methods are required for special types of model equations 
(large-scale, nonlinear, differential-algebraic, differential-integral, distributed pa-
rameter systems). Design methods for discrete event and logic controls are of 
great interest. 
• Design principles should refer to physical principles. Practical aspects, like main-
tenance, simplicity, transparency, should be included in the design process. 
• Problems like quality control, plant safety, flexibility and operability need theo-
retical concepts that can be applied in industrial applications. 
• New results should always be tested under realistic conditions, e.g. at pilot plants. 
50 THE VIEW FROM INDUSTRY 
In the last years, advances in industrial process control were mostly determined by 
application of informatics principles and less by the introduction of new advanced 
control concepts. From this, a definite gap between academic research and industrial 
practice can be seen. Often this discrepancy is considered unwanted. But there are 
also positive effects: the gap acts as a driving force for progress in academic research as 
well as advances in industrial application. Some methods, that are developed today, 
will be applied in industrial practice in the future, and some practical problems of 
today will be significant topics of research in the years to come. 
References 
Ahrens, W. (1990). Expertensysteme fiir die ProzeBfiihrung - Erfahrungen aus dem 
Verbundprojekt TEX-!. Chem. lng. Techn., 62(8):635-644. 
Allgower, F. and Raisch, J. (1988). Multivariable controller design for an industrial 
distillation column. In 5'" IMA International Conference on Control Theory, 
Southend-on-Sea, Essex. The Institute of Mathematics and its Applications. 
Allgower, F., Sax, A., and Gilles, E. D. (1989). Nonlinear controller design for a binary 
distillation column by exact input/output linearization. AIChE Annual Meeting, 
San Francisco. 
Bachmann, R. and vom Felde, M. (1988). BeobachterverCahren zur Regelung 
und Uberwachung chemischer Prozesse. Automatuierung.dechnuche Prcu:i.s, 
30( 12):581-585. 
Bochmann, D. and Posthoff, C. (1981). Binare dynami!che Sy!teme. Oldenbourg, 
M iinc1:.en. 
Brooks, S. D., Hoare, C. A. R., and Roscoe, A. W. (1984). A theory for communicating 
sequential processes. J. A!!oc. Comput. Mach., 31:560-599. 
de Bakker, J. W., de Roever, W.-P., and Rozenberg, G., editors (1989). Linear time, 
branching time and partial order in logic! and mode,", for concurrency, volume 
354. Lect. Notes Comput. Sci. 
The View from European Industry 
Emerson, E.-A. and Srinivasan, J. (1989). Branching time temporal logic, volume 354, 
pages 123-172. Leet. Notes Comput. Sci. 
Froisy, J. B. and Richalet, J. (1986). Industrial application of IDCOM. In Morari, M. 
and McAvoy, T. J., editors, Chemical proce!! control- CPC III, pages 233-244. 
Elsevier, CACHE. 
Gilles, E. D., Holl, P., Marquardt, W., Schneider, H., Mahler, R., Brinkmann, K., and 
Will, K.-H. (1990). Ein Trainingssimulator zur Ausbildung von Betriebspersonal 
in der Chemischen Industrie. Automatuierung!technuche Pran,,32(7):261-268. 
Heyman, M. (1990). Concurrency and discrete event control. IEEE Control Sy!tem! 
Magazine, 10(4):109-112. 
Holl, P., Marquardt, W., and Gilles, E. D. (1988). DIVA - A powerful tool for dynamic 
process simulation. Computer! in Chemical Engineering, 12:421-425. 
Krahl, F., Pallaske, U., and Rhiel, F. F. (1986). Einsatz der adaptiven 
MehrgroBenregelung an Destillationskolonnen. 
58(7):608-609. 
Chemie Ingenieur Technik, 
Marquardt, W. (1991). Dynamic process simulation - recent progress and future 
challenges. In Preprinu of the fourth conference on Chemical Proce!! Control. to 
appear. 
Perkins, J. D. and Sargent, R. W. H. (1982). SPEEDUP: A computer program for 
steady state and dynamic simulation and design of chemical processes. In Mah, 
R. S. H. and Reklaitis, G. V., editors, AIChE Sympo!ium !erie! 21-1, Vol 78. 
Peterson, J. L. (1981). Petri Neu and Modelling of System!. Prentice-Hall, Englewood 
Cliffs, NJ. 
Ramadge, P. J. and Wonham, W. M. (1987). Supervisory control of a class of discrete-
event processes. SIAM J. Control and Optimization, 25:206-230. 
51 
Retzbach, B. (1986a). Control of an extractive distillation plant. In Preprinu of the 
1FAC Sympo$ium Dynamic! and Control of Chemical Reactor$ and Dutillation 
Columru, pages 225-230, Bournemouth. IFAC. 
52 THE VIEW FROM INDUSTRY 
Retzbach, B. (1986b). Mathematuche Modelle von De!tillatiorukolonnen zur Synthe!e 
von Regelung!konupten. VOl Verlag, Dusseldorf. 
Schuler, H. (1986). Modellgestutzte meBtechnische Uberwachung chemischer Reak-
toren. Technuche! Me!!en, 53(7/8):299-304. 
Schuler, H. (1990). Signalverarbeitung mit Sensormodellen. VDI Berichte, 85:405-415. 
Schuler, H., Molter, M., and Schremser, U. (1990). Schulungssimulator fur Chargen-
prozesse. Chem. Ing. Techn., 62:559-562. 
Shimizu, K. and Matsubara, M. (1985). Triangular decoupling for a binary distillation 
column under pressure control. IEEE 7hlru.Autom.Contr., AC-30(5):474-477. 
Soltysiak, R. (1989). Wu!eru6tJ!ierte Prozepregdung, volume 11.2 of Hand6uch for 
Informatik. Oldenbourg Verlag. 
Trilling, U. and Kaibel, G. (1980). Experiences with the design and the application of 
multivariable control for distillation and reaction columns. Proceu Automation, 
pages 24-31.