Sei sulla pagina 1di 19

Computers and Chemical Engineering 128 (2019) 538–556

Contents lists available at ScienceDirect

Computers and Chemical Engineering


journal homepage: www.elsevier.com/locate/compchemeng

Process control practice and education: Past, present and future


B. Wayne Bequette
Rensselaer Polytechnic Institute, United States

a r t i c l e i n f o a b s t r a c t

Article history: Process control is a relatively young field, with commercial PID-type controllers developed largely in the
Received 4 February 2019 1930 s. Initial publications were applications-focused and authored primarily by instrumentation and con-
Revised 4 June 2019
trol vendors. While some chemical engineering departments had instrumentation and control courses by
Accepted 8 June 2019
the 1940 s, process control textbooks with a theoretical basis did not appear until the mid-1950 s, yet
Available online 18 June 2019
by 1950 there was already a concern about a theory vs. practice “gap.” Evolving views on this gap are
Keywords: discussed, and course content, delivery and textbooks used during the past 75 years are reviewed.
Process control The focus then turns to a flipped classroom, where students view screencasts, read textbook material
Flipped classroom and take an on-line quiz before class, which has been implemented in a process dynamics and control
Industrial process control course. The class periods involve brief lectures summarizing what they have studied, and include discus-
Chemical process control conferences sions and advanced problem solving using MATLAB and Simulink.
© 2019 Elsevier Ltd. All rights reserved.

1. Introduction of a key textbook, thus I also review the numerous textbooks that
have been used during the past seven decades, noting changes in
This is an extension and update to a paper presented at PSE content and presentation style.
2018 (Bequette, 2018), which was nominally on the flipped class- In reviewing this history I realized that the way that I teach,
room approach that I have been using at RPI. Since a keynote talk and many of the motivating examples and anecdotes that I use
should not solely cover one’s own work, I expanded my presenta- during my lectures, are based on my experience as a process en-
tion to include other course delivery methods, topics covered, etc. gineer in the refining industry, which I describe in the appendix.
While I had briefly summarized a “history of automatic control” There is nothing like handling actual manufacturing equipment to
in other talks, primarily to biomedical audiences, for this presen- make you appreciate systems and control, and I think that this ex-
tation I became more interested in when process control began to perience brings important credence to the material covered. A sab-
be considered unique enough to be adopted as a chemical engi- batical in the Pharma Industry leads to discussions of batch pro-
neering course – rather than a more general automatic control or cesses, scale-up, and the impact of scale on operability. Also, our
similar course taken by most engineers, particularly electrical and efforts to develop a closed-loop automated insulin dosing system,
mechanical. Further, I became interested in the history of process which has been tested in clinical trials (this opens up a discussion
control system development and application in the process indus- on performance metrics), is a motivating example for students and
tries. While the “retiring generation” of process control educators is a clear demonstration of the direct impact of control.
thinks of the “gap” between control theory and process control Process control is a core course in the chemical engineer-
practice as a concern raised in the late 1960’s (e.g. Foss, 1973), we ing undergraduate curriculum, yet it sometimes suffers from an
found evidence of these concerns much earlier. Sparked, however, over-emphasis on analytical mathematics without proper motiva-
by some key perspective articles in the 1970’s, there was a seri- tion from real process challenges. Our teaching has evolved over
ous attempt to understand and reduce this gap through a number the years from the traditional lecture, combined with computer-
of activities, including a series of conferences on chemical process based homework solutions, to a studio classroom where lec-
control (CPC) held at five-year intervals. Thus, one of my objec- tures and interactive simulations are combined in the same class-
tives is to review these conference outcomes and trace progress in room, to a flipped classroom where students read the material,
industrial application, process control theory to meet the industrial watch screencasts, and answer quizzes on the reading and view-
needs, and process control education (focussing on the undergrad- ing material before attending class. However, before presenting the
uate process control course). Most courses revolve around the use flipped/studio/interactive classroom in Section 6, we first provide a
review of process control practice (Section 2), theory (Section 3),
curricula (Section 4), and textbooks (Section 5) used during the
E-mail address: bequette@rpi.edu past 75 years.

https://doi.org/10.1016/j.compchemeng.2019.06.011
0098-1354/© 2019 Elsevier Ltd. All rights reserved.
B.W. Bequette / Computers and Chemical Engineering 128 (2019) 538–556 539

2. Process control practice larly interesting is the 8-page discussion with comments from 8
contributors; one discussant mentions the method of “prevision”
While there is evidence feedback control applications date which appears similar to feedforward control.
back centuries to the water clocks of Ancient Greece, an early In the 1930’s the ASME had an active Process Industries Di-
use of feedback control for manufacturing purposes was by vision. Motivated by a lack of common terminology and anal-
Watt to regulate stream turbines in the late 18th century ysis methods, Smith (1936) issued a challenge to the process
(Bennett, 1979). Here we focus specifically on process control ap- control community to develop standards through a collabora-
plications. Thompson (1917) summarizes a Chemical Engineering tive ASME report. He provided equations for a few common
Exposition and notes that there are a number of pieces of equip- mechanical/hydraulic based controllers and includes on-off, pro-
ment for the purpose of controlling temperature, and to some ex- portional, floating (I) and proportional-floating (PI). Mason and
tent, pressure. Brown (1918) describes a thermocouple-solenoid ar- Philbrick (1940) present a third-order process with several dif-
rangement that allows the manipulation of fuel gas or oil valves ferent types of control algorithms, including “two-position” (on-
to control furnace temperature; since the valve can easily go all off), proportional + “floating” (integral); they develop analytical so-
the way open or closed, it is placed in a by-pass line so that some lutions for the resulting 3rd and 4th order differential equations
moderate flow is always going to the furnace without danger of and discuss the effect of tuning parameters on the closed-loop re-
flame extinction. Indeed, for the next 15 years we find that, be- sponses. Mason and Philbrick (1941) analyze surge vessels, includ-
cause of the high controller gains used, many controllers were ing a level cascaded to flow control system and study the follow-
effectively on-off. Clark (1922) presents a pneumatically operated ing inlet flowrate disturbances: (i) step, (ii) high frequency, and (iii)
temperature chart/controller and notes applications to rubber and low frequency.
milk manufacturing. Peters (1923) describes a fractionating column The publication by Ziegler and Nichols (1942) on PID controller
with constant reflux ratio, where bottoms level is controlled by tuning lead to rapid application and verification in the process in-
manipulating the feed to the column, and a tray temperature is dustries. For example, Allen (1943) discusses problems associated
controlled by steam flow to the bottom of the column. with controlling a distillation tray temperature by manipulating a
Swan (1928) presents two-position (on-off) pressure and tem- steam control valve, using a Foxboro temperature controller. He
perature controllers based on the adjustment of the “differential” finds that the original tuning procedure and parameters selected
(deadband). Ginsberg (1929) summarizes progress in process con- did not provide good control, but the Ziegler–Nichols parameters
trol applications, including examples of temperature, pressure and (developed based on a Taylor Instruments controller) yielded good
humidity control, and stresses the importance of recording as well results. Allen (1944) further compares the continuous-cycling and
as controlling. He also describes a “differential temperature” con- response-based methods. More et al. (1945) discuss heat exchanger
troller that appears identical to modern split-range controllers that temperature control and use the Ziegler–Nichols response-based
can manipulate both heating and cooling media to regulate tem- method.
perature. Wilson (1951) in a summary of an instrumentation sympo-
Riddle and Royal (1930) detail the evolution of the role of sium held in Cleveland as part of the American Chemical Soci-
measurements and control over a 10-year period in a ceramic ety, Division of the Industrial and Engineering Chemistry, expresses
manufacturing plant. Dow (1930) discusses labor-savers in chem- a view that Instrumentation should be taught to undergraduate
ical plant operations, including larger scale equipment, automatic chemical engineers. Five papers (with only one from academia)
analysis, automatic operation (control), and the use of contin- from the symposium appear in an Ind. Eng. Chem. special section.
uous processes that are automatically controlled and operated. Dobson (1951) provides many examples of strategies to control
Grebe et al. (1933) describe an effort at Dow that started in product properties, such as concentration. An isobutane/n-butane
1924, largely because of a lack of satisfactory commercial con- column, for example, uses a differential vapor pressure sensor to
trollers for a pH problem. After discussing limitations to on-off change the distillation column pressure, while a tray temperature
control through a series of graphs, the authors develop a propor- controller is cascaded to a reboiler hot oil flow controller; other
tional + “anticipatory” (derivative) + damping (integral) controller; loops include feed and reflux flowrates, and reflux drum and bot-
they then show how this can implemented in pneumatic and elec- tom levels. Other examples shown include superheat, water vapor,
trical devices. In addition the authors mention ratio control and pH, specific gravity, conductivity, and viscosity control.
conclude that it is important to design plants that are easy to con- During the period of 1945-1950 the frequency response tech-
trol – and that control must be considered at the design stage. niques of Nyquist and Bode began to be applied to feedback con-
Thomas (1937), in summarizing a symposium on automatic trol system design and analysis (see section 3), particularly in
control held at an ACS meeting, refers to automatic control as servomechanism applications. Tustin (1951) in a paper in Nature,
a special branch of chemical science. deFlorez (1937) discusses summarizes an automatic control conference held at Cranfield Uni-
the limitations to on-off control and indicates that controllers versity, and indicates that one day of the conference was devoted
now have other modes of action; he also notes that over a to extending control system theory to the chemical and process in-
10–12 year period refining processes have moved from the or- dustries. There was a vigorous discussion among the “process en-
der of 10 0 0 Bbl/day to 10,0 0 0 Bbl/day, largely due to automatic gineers” and the “servo engineers”, who apparently used different
control systems. Velton (1937) shows a number of ratio con- terms, drew diagrams in different directions and used inverse vari-
trol applications, including air-fuel ratio on furnace combustion. ables; he proposed more joint meetings to convince both parties
McMahon (1937) describes a cascade control system where the to develop a standard nomenclature. Tustin indicated that the pro-
output of a level controller is the setpoint to a flow controller. cess engineering papers used transfer functions for the pneumatic
Grebe (1937) uses a neutralization example to illustrate the princi- controllers and plant elements in a form that permits use in the
ples of ratio, proportional, damping (integral) and rate (derivative) “familiar frequency response” type of analysis.
control, similar to the Grebe et al. (1933) paper. There were a series of articles in a 1952 special issue in
All of the previously cited papers are descriptive in nature and Scientific American devoted to Automation. An ad from Taylor
are lacking quantitative details. Ivanoff (1933) is the first to ap- Instruments notes that, in 1905, they were among the first com-
ply rigorous mathematical techniques, specifically to a tempera- panies to offer an automatic feedback temperature controller.
ture control problem. He analyzes on-off, floating (integral), pro- Brown and Campbell (1952) present a number of industrial control
portional and proportional+ floating methods. What is particu- applications, and discuss the impact of automation and control on
540 B.W. Bequette / Computers and Chemical Engineering 128 (2019) 538–556

education, business and operating personnel. Ridenour visory and direct digital control (DDC). There continued to be de-
(1952) compares and contrasts analog and digital computing and bate of direct digital control vs. supervisory control for many years,
presents a vision that includes the ability of computers to respond and computer control continued to be implemented in process
to faults much faster than any human can. Ayers (1952) provides plants. The development of the Honeywell TDC 20 0 0, announced
an overview of the importance of instrumentation and control in in 1975, had perhaps the most profound impact on the industry.
the petroleum refining industry. One picture shows a control panel There were many benefits of computer control, but the pri-
with the instrumentation displayed similar to a flowsheet; I should mary motivation was improved real-time economic return using
note that 30 years later, most of the process units at American a hierarchical strategy, with plant-wide steady-state optimization
Petrofina in Port Arthur, Texas, where I was a process engineer in (roughly daily) cascading to more frequent operating unit opti-
1980-82, were still operating with the same instrumentation and mization. The unit operations layer still cascaded to lower-level PID
control panels! controllers, with override protections, anti-reset-windup, etc. in or-
Cohen and Coon (1953) developed a set of PID controller tuning der to enforce constraints (important since the optimum operat-
rules based on a first-order + deadtime model. Their performance ing point usually occurs at one or more constraints). Model pre-
criterion is based on rejection of step input load disturbances, and dictive control (MPC) techniques, largely developed in the 1970’s
they also refer to the “well-known” frequency response techniques. in the refining industry in both the US and France, enabled the
One thing particularly striking, after reading many articles from simultaneous control of multiple outputs using multiple, possibly
this era using the pronoun “he”, is to find that Geraldine Cohen constrained, inputs. Dynamic matrix control (DMC) was first ap-
was a “she.” Also in 1953 we find the first published process con- plied at Shell in 1973, but the US academic world largely learned
trol article to use frequency response techniques. Ceaglske and Eck- of these approaches from the presentations by Cutler and Ramaker
man (1953) provide a detailed frequency response analysis of an (1980) and Prett and Gillette (1980). To this day MPC remains the
experimental pressure control process, using P-only control. They most widely applied advanced control technique, and the impact
suggest that a 45° phase margin yields more acceptable results of MPC in industrial applications is summarized in a number of
than the Ziegler–Nichols tuning parameters. They do not refer to articles in Section 3.
the amplitude and phase plots as Bode diagrams, but simply cite During the past four decades there have been continuous ad-
textbooks by Brown and Campbell (1948) and Ahrendt and Ta- vances in the use of process systems engineering techniques at all
plin (1951) for the methodologies. levels in the corporate hierarchy, including planning and schedul-
It is particularly striking that the early process control articles ing and supply chain management, the use of “big data” techniques
are written primarily by instrumentation and control vendors (e.g. for predictive maintenance, fault detection and related topics. A
Taylor Instruments, Foxboro, Leeds and Northrup) or manufactur- major challenge has been in the use of proprietary systems that
ing companies (e.g. Dow Chemical). Also, the papers often include often require a system-wide upgrade if major changes are desired.
a detailed discussion by one or more reviewers followed by closure Goals of recent activities in open process automation and “Smart
from the authors (see the Trans. ASME papers, for example). It is Manufacturing” include the development of communication and
interesting that the problem of reset (integral) windup is not ex- software standards that make it easier to “plug and play” hardware
plicitly discussed; Astrom and Hagglund (2005) note that integral and software from different vendors to implement and/or maintain
windup was well-known in the early analog controllers and that advanced automation and control systems. These developments are
manufacturers kept the solutions (called preloading, batch unit and discussed further in Section 3.
other terms) as trade secrets.
A 1956 special issue on control with 12 articles, based on a 3. Process control theory and the “Gap” with practice
1955 Symposium sponsored by the ACS Division of Industrial and
Engineering Chemistry, is summarized by Friend (1956), with com- Process control instrumentation and application to chemical
mentary from Vannah and Slater (1956). Papers included a tutorial, processes largely led the development of relevant control theory
focused on modeling and frequency response analysis by Ceaglske by over 30 years; thus, there was an initial “gap” where industrial
(1956a,b), distillation control by Williams et al. (1956), pulse test- process control practice was conducted with little control theory,
ing of heat exchangers by Lees and Hougen (1956), and analysis of or basic tuning methods, to support it. In this section we focus ini-
cascade control by Franks and Worley (1956), among other papers. tially on theories that are covered, to some extent, in undergrad-
Seven of the 12 papers were from industry, perhaps that last time uate process control courses (which are covered in Section 4) and
that an academic journal issue had more industrial than academic textbooks (covered in Section 5). We also address the “gap” that
process control papers. was based on the perception that control theory had moved far
While the control of biological systems has only recently been ahead of industrial practice. By reviewing highlights of the series
a focus in undergraduate control courses and textbooks, Fuld and of Chemical Process Control conferences, which occur at five-year
Dunn (1957) present applications of process control to fermenta- intervals, we track the progress of the development and applica-
tion. tion of advanced industrial process control techniques.
Industrial computer process control became an active area in While there was an evolution of techniques to find the char-
the 1950’s, with the efforts nicely reviewed by Astrom (1985), and acteristics of the roots of a polynomial (Bennett, 1979), which
Stout and Williams (1995). A Ramo-Wooldridge RW-300 computer can be used to analyze the stability of a closed-loop system, the
achieved closed-loop control in 1959 at the Texaco Refinery in Port general form using an array is based on the methods of Routh
Arthur Texas. The installation was on a small catalytic polymer- and Hurwitz (Hurwitz, 1895). The frequency response techniques
ization unit which was selected because of the economic bene- by Nyquist (1932) and Bode (1940) largely serve as the basis for
fit, availability of operating records to estimate performance im- control system design and analysis. Frequency response techniques
provement, and because the small unit was isolated from the re- so rapidly permeated the literature that Bode diagrams were often
finery mainstream (crude unit, etc.) so any unit problems would presented without attribution to the original Bode paper (Bode,
not impact the entire plant. 100 analog signals (pressure, flowrate, 1940) or book (Bode, 1945). Ferrell (1945) and Harris (1946) both
temperature) were digitized and sent to the computer, in addition note that the design techniques used by radio and communication
to two chromatographic measurements. The computer outputted engineers could be applied to servo-mechanisms.
5 direct digital signals to control valves, and 9 setpoint signals to The commercial controllers discussed in the previous section
analog controllers; thus, this first installation included both super- and developed largely in the 1930s were primarily PID controllers;
B.W. Bequette / Computers and Chemical Engineering 128 (2019) 538–556 541

Bennett (2001) notes that the derivative component became stan- sored education, (2) component study in universities, (3) promo-
dard in Taylor and Foxboro products in 1939, although they tion of economic incentives, and (4) publication policies (tutorial
had used derivative (called “pre-act” by Taylor) in special in- papers involving new theory or applications, and evaluation papers
stallations for a few years before that. The first publication ex- to highlight significant achievements). Clark (1965) indicated that
plicitly proposing PID was by Minorsky, on ship steering, in there was effort at the 1965 Joint Automatic Control Conference
1922 (Minorsky, 1922) (for an interesting history of this, see (JACC) to invite more applications papers to reduce the imbalance
Bennett, 1984). It is unlikely that process control engineers re- between theory and application. I should note that the 1965 JACC,
viewed that paper and, in any event, tuning was largely ad-hoc un- held on campus at RPI in Troy, NY, while referred to as “large”
til the Ziegler–Nichols methods were proposed in 1942. It should with four parallel sessions, was certainly much smaller than recent
be noted that pneumatic (this is also true of electrical analog) con- American Control Conferences, which have 22-24 parallel sessions.
trollers typically implemented a series or “interacting” form (or Ed Bristol, with Foxboro, in a 1.5 page “correspondence” pa-
something similar) of a PID controller and not the ideal (parallel per in IEEE Transactions on Automatic Control, in 1966 proposes
or “non-interacting”) form. That is, while an ideal (non-interacting) a “new measure of interaction for multivariable process control.”
PID controller has the following transfer function Greg Shinskey, also with Foxboro, apparently gave this measure the
 1
 name “relative gain” and included it in his Process-Control Systems
u ( s ) = kc 1 + + τD s e ( s ) book in 1967. Shinskey, in the preface to this book, refers to a com-
τI s
munication problem between the scientists “searching for more
the series (interacting) version is represented by advanced control concepts” and the plant personnel that must ap-
 
1   ply them.
u(s ) = kc 1+  1 + τD e(s ) Foss (1973) presents process control challenges that are not par-
τI s
ticularly well addressed by the modern control approaches. These
and, in practice, an ideal derivative cannot be implemented and is include the types of disturbances that occur, uncertain and com-
therefore represented as a filtered derivative. Bennett (2002) notes plex models, limited measurements, multivariable systems, and
that Aikman and Rutherford (1951) identified five different types somewhat irrelevant objective functions for optimal control. He
of interaction in their analysis of commonly used controllers. suggested that the “gap” needed to be closed by theoreticians that
Cohen and Coon (1953) include tuning parameters for both the were not addressing the real challenges, rather than the common
parallel and series forms. view that people applying control strategies did not understand or
Rutherford in 1949, as cited by Dillon (2012), finds a wide gap were not properly using control theory. There is evidence that this
between theory and practice and states “It is not felt that the ma- paper influenced the research activities of many process control
jority of the articles listed will provide any useful reading for the faculty that have recently retired or are approaching retirement
average Process Control engineer, but it was felt worthwhile in- age; see Stephanopoulos (1983), for example.
cluding everything that had been seen, even if only as a warning to The Foss (1973) paper, an NSF process control workshop in
others not to waste time searching for articles and reading them.” 1973, and discussions at the 1974 Joint Automatic Control Con-
Block diagram and transfer function techniques make analy- ference (JACC), led to a proposal for a Chemical Process Control
sis and, more importantly, synthesis techniques relatively easy. (CPC) conference that was held at Asilomar in 1976. The 80 partici-
Truxal (1955) details what is now known as “direct synthesis” to pants at the conference, co-chaired by Alan Foss and Morton Denn,
design a controller based on the desired closed-loop response char- were from academia (50%), operating companies (25%), and control
acteristics. The review paper by Johnson (1958) on automatic con- system vendors (25%) (Foss and Denn, 1976). The “gap” weighed
trol, which clearly provided an outline for his textbook (1967), heavily in most of the presentations and the discussions that fol-
could be used as the basis for roughly 2/3 of a current undergrad- lowed; the 2–3 page discussion summaries that follow every ses-
uate course. sion of 3–4 papers provide important insights about concerns and
Optimal and state space based techniques that formed the ba- future research opportunities. Foss and Denn suggested that there
sis for “Modern Control Theory” developed at a rapid pace in the was a need for test or theme problems that could be used by re-
late 1950’s (Athans, 1976). Within less than a decade there were searchers to test advanced control theories. Their recommendation
concerns about a gap between theory and practice (Fuller, 1963). that a conference be held at roughly 5-year intervals was followed,
Indeed, Bode (1960) provides a review of the development of feed- with individual CPC’s held through CPC-7 (2006), followed by the
back control, including his contributions; at the time Bode was the joint FOCAPO-CPC conferences in 2012 and 2017.
Vice President of Bell Labs, and had not worked in the control field Lee and Weekman (1976) delineate many of the challenges with
for 20 years. He uses the analogy of visiting one’s hometown af- implementing advanced control on chemical processes, using a
ter having not been there for several decades. He expresses a few fluid catalytic cracking unit (FCCU) as a prime example. They point
misgivings he had “concerning the present logical structure of the out that the primary economic return is through better steady-
control field, which, like my grandparents’ hometown, seems to be- state optimization, with improved regulatory control having a rela-
come more sprawling as it grows…” While this does not directly tively minor impact. They do note the need to improve the integra-
refer to a “gap” it does indicate that there was not as clear of a tion of design and control, and improving the ease of implement-
focus of the field; part of his concern was also about the “mixed- ing new control strategies.
marriage” of communication and feedback control. CPC2, co-chaired by Dale Seborg and Tom Edgar, was held in
Axelby (1964) notes that there has been recent discussion about Sea Island Georgia in January 1981, with 113 attendees – 48 from
the gap between theory and application, but that the gap has not the process industries, 47 from academia and 17 from control sys-
been well-defined; he also notes that there had been significant tem and consulting firms. Perhaps the most interesting session was
applications of modern control theory, citing the Apollo space pro- on the Design of Control Systems for Integrated Chemical Plants,
gram as one example. Chestnut (1965) claims that the gap has including presentations by Morari (1982) and Douglas (1982).
been a concern for almost a decade. He summarizes a meeting in The toluene hydrodealkylation (HDA) plant studied by Douglas is
New York in March, 1964, on the subject “Bridging the Gap Be- now used as an example in many papers, and is included in the
tween Theory and Practice.” Three groups, Universities, Suppliers process design textbook by Douglas (1988). Mehra and Rouhani
of Equipment, and Users of Equipment all have different objectives (1982) reported on many industrial applications of Model Algo-
and constraints. Proposed solutions included: (1) Company spon- rithmic Control, a form of MPC. The conference summaries are, on
542 B.W. Bequette / Computers and Chemical Engineering 128 (2019) 538–556

the whole, quite positive. Seinfeld felt that the distinction between 23 Shell participants. The 3.5-day workshop structure was simi-
process engineers and control engineers should be less sharp and lar to a CPC, with lectures in the morning and evening and tu-
that the integration of design and control is important (Seinfeld, torial workshops during the first 2 afternoons; 3 of the 14 re-
1982). Foss also noted the heightened interest and unmistakable search presentations were by Shell personnel, and 11 by academics.
confluence in the opinion among industrial practitioners and aca- Morari provided a perspective on what had been accomplished
demic researchers that plant control was an important problem. in response to the three key perspective papers from the 1970s
Denn (like at the CPC in 1976) elaborated on the need for a Theme (Foss, 1973; Lee and Weekman, 1976; Kestenbaum et al., 1976) and
Problem for Chemical Process Control, and noted that significant found that significant progress had been made during the previ-
interest appeared to be forming to undertake the project; I should ous decade, with the theory-practice gap narrowed, with a better
note that it took almost another decade for this to develop, as part understanding of the effect of uncertainty, and with the success
of a session at the 1990 AIChE Annual Meeting (reported later) of model predictive control. During the first afternoon workshop
(Denn, 1982). participants were presented a challenge problem with a heavy oil
The effect of process design on the ability to operate/control fractionator with seven outputs and five inputs, multiple objectives
chemical process plants became an active research area in and constraints, and parametric uncertainty. The main limitation
the 1980’s (Douglas, 1982; Morari, 1983), although its impor- to this as a test case was that the input-output relationships were
tance had been stressed in papers nearly 50 years earlier. all first-order + deadtime transfer functions. Shell agreed to pub-
Grebe et al. (1933) mentioned designing processes that are easy lish the challenge problem and the academics agreed to work on
to control, while Peters (1941) stated that engineers should know the problem and present solutions at a later date (Morari, 1988).
how results obtained depend upon process characteristics. Sim- A second Shell Process Control workshop was held in 1988,
ilarly, van Antwerpen (1942) noted that process design is usu- growing substantially to 18 academic and 32 Shell employees; the
ally based on steady-state analysis, without consideration of dy- proceedings were published in Prett et al. (1989). 12 of the papers
namics or instrumentation and control. Arant (1948) questioned were related to the Shell challenge problem while “distinguished
what changes could be made to improve controllability, and sug- papers” covered a range of topics including statistical process con-
gested the inclusion of at least one semester of work (dynamics trol (MacGregor, Moore) and artificial intelligence (Stephanopou-
and control) in the curriculum of every progressive chemical engi- los). Morari (1988) provided an overview of past results as well
neering school. Young (1955) used the term “controllability” in his as a vision for the future. One suggested challenge “how can one
industrially-based process control book, largely referring to chal- detect deteriorating valves and sensors from on-line measurements
lenges induced by lags and time-delays. before these control elements have failed entirely?” remains a cur-
CPCIII, co-chaired by Manfred Morari and Tom McAvoy, was rent active area of research, involving machine learning and “big
held at Asilomar in January 1986, with 145 attendees – 78 from in- data” techniques.
dustry and 67 from academia. Model predictive control continued Like Shell, DuPont had an active corporate process control re-
to have a major industrial impact with presentations on IDCOM by search activity that even included neural physiologists, and a num-
Froisy and Richalet, DMC by Garcia and Prett (1986), and UDMC by ber of faculty spent time with this group. An example outcome
Morshedi; the UDMC paper outlined a general nonlinear approach, was a series of papers illustrating process control algorithms that
yet without specific applications. The “gap” remained an underly- mimicked the baroreflex mechanism, which uses multiple physio-
ing theme, and the conference closed with a presentation by Greg logical inputs to regulate blood pressure and cardiac output (blood
Shinskey, who delineated three types of process control engineers. flowrate); see Henson et al. (1994, 1995) for examples.
This was followed by “several hours” of comments from the audi- CPCIV, co-chaired by Yaman Arkun and Harmon Ray, was held
ence; in abbreviated and edited form the presentation and discus- in Padre Island, Texas in February 1991, with 158 participants –
sion covered 15 pages in the final published volume. As mentioned 80 from industry and 78 from academia. The conference, with the
earlier, Shinskey, included (indeed, named) the Bristol (1966) rel- theme “Future Needs and Challenges in Process Control”, had many
ative gain (array) technique in his 1967 book (Shinskey, 1967), model predictive control papers, with three from industry and
but the RGA did not appear in any academic textbook until 1984! five from academia. Industrial contributions were from Cutler and
Morari and McAvoy noted several take-home messages from the Yocum (1991) (DMC Corp. and GE), Caldwell and Dearwater (Profi-
conference: (i) interest in process control had increased dramat- matics), and Garcia et al. (1991) (Shell), while academic contribu-
ically, and roughly 100 more registrants could not be accommo- tions were from Ricker, Zafiriou, Clarke, (1991), Morari and Lee, and
dated in order to retain the conference format, (ii) academic re- Biegler and Rawlings (Morari and Lee, 1991; Ricker, 1991). As part
search has shifted to more practical problems, and (iii) future of their paper on “A View from North American Industry (Zafiriou,
progress will depend on intensive university/industry cooperation. 1991; Biegler and Rawlings, 1991)”, Downs and Doss (Tennessee
The proposed university/industry cooperation began to ramp- Eastman) critiqued current and future process control education.
up through several avenues –industrial challenge problems (Shell, They felt that process control was taught from too general of a per-
Tennessee Eastman and others), faculty spending time in indus- spective, applicable to any “system” but without a chemical pro-
try (DuPont), either as postdocs or on sabbatical, and Indus- cess focus. They proposed that a better understanding of process
try/University Cooperative Research Centers and related Process dynamics was important and that teaching unit operations control,
Systems and Control consortia. In the 1980s the NSF began to sup- with case studies, would be a better approach.
port Industry/University Cooperative Research Centers, and a num- A special session on industrial process control challenge prob-
ber of these involved process systems and control. Relevant cen- lems was held at the 1990 Annual AIChE meeting. Five of these
ters included the Lehigh Chemical Process Modeling and Control were published in a special issue of Computers and Chemical Engi-
consortium (1984), Washington Center for Process Analysis and neering in 1993 (Mellichamp, 1993). The Tennessee Eastman prob-
Control (1984), Massachusetts Center for Process Design and Con- lem (Downs and Vogel, 1993), a reactor/separator/recycle process
trol (1985), CMU Computer-Aided Process Design laboratory (1985), with 12 manipulated inputs, 41 measured outputs, has been the
McMaster Advanced Control Consortium (1988), and Texas Control most popular, and has been cited by at least 1981 other papers.
Consortium (1993) (eventually including Wisconsin and California), CPCV was held in Lake Tahoe in January 1996, and co-chaired
among others. by Jeff Kantor and Carlos Garcia (with Brice Carnahan co-editing
The Second Process Control workshop in 1986 (Prett and the published volume in 1997). Qin and Badgwell (1997) pro-
Morari, 1987; Morari, 1987; Morshedi, 1986) had 12 academic and vided a survey of industrial model predictive control applications.
B.W. Bequette / Computers and Chemical Engineering 128 (2019) 538–556 543

Details on the types of algorithms and models are provided for five vided an overview of the development and use of the TMODS sim-
vendors and six technologies (one vender had two methods); the ulator at DuPont; a key highlight was the ease of use in compari-
vendors cited over 2200 MPC applications to date. The largest ap- son to many commercial simulators. Similarly, Alford (2006) noted
plication, by DMC Corp., was 603 outputs and 283 inputs. Morari that the simulator at E.I. Lilly had many features, including integra-
(1997) provided an assessment of the progress in chemical process tion with a data historian, that were not available in commercial
control during the 20-year period beginning with the first CPC. By simulators. Results of an education panel discussion are presented
virtually all metrics the field had grown substantially – funding, in Section 4.
Ph.D.’s granted (tripling durng the past decade), and published pa- It was recognized by some members of the process opera-
pers. He noted that nine consortia, with an average of 13 member tions and process control communities that there was much over-
companies, raised about 30% more support than the NSF-CTS di- lap in their interests, particularly in actual industrial applications.
vision in 1995. He also noted that company hiring of Ph.D.s had Thus, the originally independent conferences on the Foundations of
shifted from corporate research centers to the production units. It Computer Aided Process Operations (FOCAPO) and CPC were com-
is interesting to note that, during the conference AspenTech an- bined for the first International FOCAPO-CPC conference in 2012,
nounced that it had purchased two companies, Setpoint and DMCC, held in January in Savannah, Georgia. A total of 182 participants
with different cultures and approaches to model predictive control. registered for the two conferences, with 76 people attending both
DMCC had evolved from the DMC approach developed by Cutler conferences; 52 participants were students or postdocs. FOCAPO
(Cutler and Ramaker, 1980), while Setpoint had evolved from ID- had 139 and CPCVIII had 119 registrants, thus the number of peo-
COM (Froisy and Richalet, 1986). ple only registering for FOCAPO was 63, while the CPCVIII-only
CPCVI was held in Tucson in January 2001, and co-chaired by registration was 43. The CPCVIII co-chairs were Don Bartusiak and
Jim Rawlings and Tunde Ogunnaike, with 135 attendees – 65 from Prodromos Daoutidis, with FOCAPO chaired by Nick Sahanidis and
industry and 70 from academia and government. The continuing Jose Pinto, while I served as a general chair to promote interactions
diversification of the industries served by the process control com- between the groups. Some of the proceedings papers were pub-
munity was noted. Also, the meeting discussions seemed to be lished in a special issue of Computers and Chemical Engineering
less acrimonious than the past. A number of papers focused on in (2013), with others appearing in 2012. Davis et al. (2012) pro-
the continuing application of model predictive control in indus- vided an overview of “Smart Manufacturing” and activities of the
try. Downs (2002) suggested than classical decentralized control Smart Manufacturing Leadership Coalition (SMLC). A Smart Manu-
techniques should be used during the first year of operation of facturing platform would enable (i) reduction of development and
a new plant; advanced control, including MPC, could be added deployment costs for manufacturing oriented modeling and simu-
later when the operating characteristics are better known. He also lation, (ii) reduced costs for IT infrastructure, (iii) access to “apps”
felt that Eastman had benefited by having a central group man- for a wide-variety of problems, (iv) an enterprise layer for applied
age all control-related activities. Vogel and Downs (2002) provided manufacturing intelligence and performance metrics, (v) test bed
an overview of how MPC was implemented at Eastman, by devel- demonstrations, and (vi) involvement of small, medium and large
oping their own software based on the approaches presented by enterprises.
Muske and Rawlings (1993). A parallel cracking furnace example Common themes expressed by an industrial panel included
included 33 controlled outputs, 40 manipulated inputs and 32 con- (Daoutidis and Bartusiak, 2013):
strained variables. Young et al. (2002) provided an overview of the
• improved integration of control, optimization, and planning;
development and application of nonlinear model based techniques
• MPC performance monitoring;
used at ExxonMobil. An early motivation was a new polymeriza-
• more-complete, more-accurate representation of process behav-
tion plant with a new catalyst system being started up in 1990.
ior via data-driven methods beyond currently available identifi-
Since they deemed linear MPC as insufficient they worked with
cation techniques;
academic collaborators on model development and state estima-
• more-extensive use of first principle models fit to the process,
tion techniques to incorporate into a nonlinear MPC-based strategy.
including dynamic simulations for rapid prototyping of control
The 4 input, 4 output process had roughly 50 states and 120 DAEs.
applications; greater interoperability of control applications, in-
They decided to commercialize the nonlinear control methodology,
cluding cooperative MPC and real time optimization;
with an award to DOT Products to develop a commercial version
• reducing the information technology complexity or barriers to
of the software.
technology transfer with current process control systems;
CPCVII was held in Banff, Canada in January 2006, co-chaired
• a significant increase in sensor data enabled by low-cost net-
by Mike Henson and Tom Badgwell (2006), with the theme “As-
working or wireless technologies; and
sessment of Core Technologies and Emerging Applications,” with
• more explicit assessment and measurement of the economic
70 academic (including 15 students), and 43 industrial participants.
benefits of the use of process control technology to motivate
Froisy (2006) provided a perspective overview of the development
research, attract investment and sustain implementation
of a new state space-based MPC product at AspenTech; the user in-
terface was particularly important, to shield the user from critical FOCAPO-CPC 2017 with the theme “Bringing Research to Prac-
underlying checks and balances that assure detectability, etc. The tice and Practice to Research” was held in Tucson in January 2017,
initial prototype was developed in 1999, based on the approach with 206 participants with 68 coming from industry, 6 from Gov-
presented by Muske and Rawlings (1993); several years later the ernment and National Labs, and 132 from Universities. Erik Ydstie
fourth phase of the prototype was applied to a chemical process and Larry Meagan were the CPCIX co-chairs, Christos Maravelias
with 8 inputs, 20 outputs and 109 states. Froisy showed exam- and John Wassick were the FOCAPO co-chairs, and I served as a Li-
ples to illustrate how additional measurements can improve un- aison to encourage discipline overlap. The conference included 56
measured disturbance rejection using a disturbance model in a invited talks and 86 contributed papers (55 in FOCAPO and 31 in
Kalman Filter framework, and to illustrate the effect of assump- CPCIX). The final day concluded with an ideation session to sug-
tions when implementing an infinite horizon strategy. He ended gest grand challenges and future directions. Thirty papers were re-
the paper with expected future developments including batch ap- vised/extended and published in a special issue of Computers and
plications, nonlinear models and better Kalman Filter tuning meth- Chemical Engineering in 2018.
ods. Vinson (2006) provided an overview of high performance pro- Two of the lasting contributions/ideas post-classical control,
cess control applied to Air Separation Units. Cox et al. (2006) pro- developed in the 1960s and 1970s, the relative gain array
544 B.W. Bequette / Computers and Chemical Engineering 128 (2019) 538–556

(Bristol, 1966), and model predictive control (e.g. Cutler and Ra- ter on Automation Technologies and Systems (CATS)), PhD degrees
maker, 1980), were ideas generated in industry. It took around are awarded by individual departments.
15 years for the Bristol paper to stimulate academic research Williams (1958) notes that AIChE had established a committee
on the topic, which led to more fundamental studies by to help Universities in setting up and presenting process control
McAvoy (1983) and Grosdidier et al. (1985), and widespread use in and process dynamics courses. He also provides an extensive list-
undergraduate control courses and textbooks. The period between ing of process instrumentation and control articles published in
the DMC presentations in 1979/1980 and the initial academic con- 1957. McKetta and Schechter (1961) list automatic control as an
tributions from Garcia and Morari (1982), Marchetti et al. (1983), important course in a flexible undergraduate chemical engineering
and Ricker (1985) was shorter, but sparked fundamental contribu- curriculum.
tions that continue to this day. The academic contributions from It is natural that process control faculty would be involved in
Muske and Rawlings (1993) have been incorporated into MPC early implementation of computers in the undergraduate curricu-
strategies at ASPEN and at Tennessee Eastman, as examples. lum. Stice and Swanson (1963) report the use of analog computing
It is clear that, while discussions about a “gap” will always be in an undergraduate process control course at IIT.
with us (after all, a “driving force” is needed for progress in any Eisen et al. (1975), in a survey with 101 chemical engineering
endeavor), process control research and industrial practice in ad- departments responding, found that all but six departments of-
vanced process control are in a relatively good state. Thru con- fered at least one undergraduate process control course. The fol-
sortia, collaborations, and interactions at major conferences, aca- lowing topics were covered: fundamental modeling (78%), empiri-
demics are working on applications and challenges of interest to cal modeling (33%), control (88%), instrumentation (27%); approx-
industry. Ph.D. graduates continue to be hired by industry, largely imately 1/3 use classroom demonstrations and audio-visual aids
into operating divisions since corporate research centers have typ- and 3/4 rely on laboratory work (including digital and analog sim-
ically been downsized. In the next section we focus on undergrad- ulation) to complement classroom lectures. Most of the survey re-
uate process control curricula and how it has evolved over time, sponses listed frequency response as a difficult concept.
partially in response to industrial needs. A 1978 survey of 158 chemical engineering departments in the
US and Canada, reported by Seborg (1980), yielded 143 responses,
with 136 departments reporting the teaching of an undergraduate
4. Process control curricula and course delivery control course, with 75% requiring such a course. Laboratory ex-
periments were reported by 70% of the departments. For example,
Van Antwerpen (1980) and Hougen (1967) suggest that process Morari and Ray (1980) report a series of experiments in real-time
control began to assume importance in the chemical engineering computing and control used at the University of Wisconsin.
curriculum during the 1935–1945 decade, yet no process control- Ray (1983), in a review of multivariable control, summarizes
oriented textbooks were available until at least 1945. Purdue Uni- topics that have entered the undergraduate curricula over the
versity (Wankat and Peppas, 2011) introduced an Engineering In- decades. In the 1940’s measurements and controller hardware, PID
strumentation course in 1937, which included “the measurement, control, linear systems, controller tuning, cascade and ratio ap-
recording, and controlling of various quantities, including pressure, pear. In the 1950’s step responses, transfer functions, stability,
temperature, flow, viscosity; applications of telemetering, electron- analog computation and frequency responses are added. In the
ics.” The Rensselaer Polytechnic Institute course catalog of 1946 1960’s nonlinear SISO analysis and pulse testing are included. In
lists a course on instrumentation that includes the principles of the 1970’s feedforward control, digital simulation, computer con-
automatic control. trol and multivariable systems are added.
Stout and Ceaglske (1945) present a process control labora- Eisen (1985), in a survey with 110 chemical engineering de-
tory at Washington University that is used in undergraduate and partments responding, found that all departments offered at least
graduate education. Johnson (1951), at Princeton University, used one undergraduate process control course. The following topics
a pneumatic multiple tank system for both undergraduate edu- were covered: fundamental modeling (82%), empirical modeling
cation and graduate research; its development was inspired by (49%), control (98%), instrumentation (30%); 40% use classroom
a demonstration unit used by Bristol Co. in summer courses for demonstrations and 41% audio-visual aids and 68% rely on labo-
teachers of instrumentation. At the undergraduate level the fo- ratory work (including digital and analog simulation) to comple-
cus is on the different types of valves and controllers, and no at- ment classroom lectures. While the 1985 survey did not indicate
tempt is made to present control theory in a quantitative fashion. what simulators were used, it is likely that many were “home
Munch (1951) notes that “chemical engineering departments in grown” with students and faculty contributing software programs.
some of our more progressive technical schools now have courses Holt et al. (1987) reported that the University of Wisconsin CON-
designed to teach their students the fundamentals of instrumenta- SYD interactive control system design and simulation package was
tion.” used by many universities and companies. Within a short period of
Brown and Campbell (1952) outline the need for interdisci- time MATLAB and Simulink from the MATHWORKS made CONSYD
plinary study in systems and control; except for a few masters de- obsolete.
gree programs, this idea did not take hold in the US. For a brief pe- While the majority of departments report a single 3-hr
riod in the 1970’s the RPI School of Engineering created a Division semester course in process control, Williams and Tarrar (1986), at
structure, with systems and control as a division, including faculty Auburn, present a two-quarter (4-credit hours each) sequence in
from many departments. Every decade or so a Dean will attempt a lecture-based course, with an additional 2-credit control labora-
to consolidate departmental courses into a general course, but this tory course with 5 or 6 experiments (some pneumatic, some dig-
always received pushback from the faculty; the common impres- ital). Koppel and Sullivan (1986) report the use of an industrial
sion is that students learn better when studying examples of di- process control system (IBM’s Advanced Control System) to con-
rect interest to them. Naturally, at the graduate level more general trol simulated industrial processes. In addition to Purdue and Wa-
systems and control courses are taught. For example, our Electri- terloo, similar facilities were available at LSU, Imperial College and
cal Computer and Systems Engineering (ECSE) department teaches Queensland.
the introductory graduate course on Systems Analysis Techniques, Edgar (1990) summarizes control courses ca. 1988, and presents
taken by graduate students in all departments. And, while some a vision for 20 0 0, which includes more discrete modeling and con-
research occurs as part of Center activities (for example, our Cen- trol, optimization-based methods (PID tuning and MPC) and expert
B.W. Bequette / Computers and Chemical Engineering 128 (2019) 538–556 545

systems. Edgar also mentions the coverage of safety, particularly response analysis (2.4), expert systems and artificial intelligence
when considering plantwide control, but surprisingly, it does not (1.9).
appear as a separate topic in any survey until 2015. Griffith, in a Silverstein et al. (2016) in a 2015 survey titled “How we teach
1993 survey with 124 departments replying, found that the lecture process control” found substantial variability in course content; the
in % time covered was process dynamics and modeling (28.1%), survey was much more specific about the course content, with
feedback control and tuning (22.1%), stability and frequency anal- 30 topics listed. For example, the following topics were covered:
ysis (14.2%), computer simulation (8.9%), advanced control tech- multiple loops (43%), frequency response (50%), ratio (51%), safety
niques (8.4%), control system hardware (7.7%), computer control (63%), modeling in time domain (68%), cascade (77%), feedfor-
systems (4.8%), and other (5.7%). Also, MATLAB was the most ward (80%), transfer functions (88%), feedback tuning (92%), and
widely used (30 departments) software package, followed by PI- PID control (96%). Topics receiving little coverage included discrete
CLES (16), TUTSIM (13), CC (12), and DIRA (8). control (7%), plantwide (11%), real-time optimization (11%), batch
In the early 1990’s Doug Cooper at UConn developed PI- (12%) and process monitoring/statistical process control (14%), and
CLES to simulate realistic process dynamics and control prob- model predictive control (28%). So, two of the major post-classical
lems (Cooper, 1996); this evolved into ControlStation, which techniques, the RGA for multiple SISO loop synthesis (pairing),
is also used as a training tool for a number of companies. and model predictive control, are covered in less than 50% of un-
Doyle et al. (1998) developed a set of process control modules dergraduate classes. The survey indicates that 70% of courses use
based on MATAB; these were published as a workbook, including project or problem-based learning and 45% of courses use com-
electronic files, in 20 0 0 (Doyle et al., 20 0 0). Bequette (1998) re- puters in the classroom. Note the low adoption of topics proposed
ported the use of MATLAB and Simulink-based learning modules in the Edgar (1990) vision for 20 0 0: discrete modeling and con-
for process dynamics and control, as well as an initiative at RPI to trol (7%), optimization-based methods (PID tuning (11%) and MPC
develop internet-based learning materials to integrate mathematics (28%) and expert systems (not surveyed, but likely near 0%) has
throughout the science and engineering curriculum. not come to pass in most courses. A comparison of the industry
One of the goals of the Computer Aids for Chemical Engineer- needs with the current academic coverage is in Table 1. Also, the
ing (CACHE) corporation (a non-profit) is to create learning mod- 2015 survey results are compared with the 1988 actual and 20 0 0
ules and software to undergraduate education. As part of a 25- proposed (Edgar, 1990) in Table 2; topics for the 2015 survey were
year history of CACHE, Arkun and Garcia (1996) summarize process only included if they were covered in more than 50% of the re-
control education activities, including the development of a Model sponses. Note that little has changed in most courses during the
Predictive Control Toolbox, which is now distributed by the Math- 25-year period!
Works as part of MATLAB. Two of the primary future education In a 2015 NSF-sponsored AIChE study on the Chemical
challenges outlined are the integration of control with process de- Engineering Industry-Academia Alignment (Luo and Westmore-
sign, and plantwide control. Certainly the methods of Douglas, par- land, 2015), the common industrial view was that a better under-
ticularly as presented in his textbook (Douglas, 1988) have helped standing of process control and of dynamics was needed. A con-
integrate design and operability. The Seider et al. (2016) textbook cern of academics was that it was hard to justify a faculty po-
considers operability and control directly in the design process. sition in control because of limited research funding in the area.
The ASPEN-based books by Luyben devoted to chemical reactors There was also a concern that most faculty hired during the past
(Luyben, 2007) and distillation (Luyben, 2013) can be used as sup- 30 years have no industrial experience. Perhaps of greater concern
plemental material in process design, or advanced process control, is that more and more science-oriented faculty, particularly in the
courses to assist students in implanting steady-state and dynamic areas of bio- and nano-technology, have been hired, with no ex-
simulations. posure to core ChE topics in their education or research. The 2016
Shinskey (2002), while misrepresenting much of the coverage survey by Silverstein et al. indicated that 22% of process control in-
of process control courses, points out that these courses tend to structors had no industrial experience. Of those that had industrial
overemphasize setpoint changes, and that IMC-based design tends experience, the average time was 4.1 years, compared to 11.3 years
to result in inferior load disturbance rejection. Perkins (2002) as- (omitting those with no experience) for those teaching process de-
sesses the past, present and future of process systems education, sign.
including process control and process design. He notes that early It is likely that faculty without a background in control, that are
process control textbooks focused on analysis, whereas more re- teaching the control course, will focus on the mathematical anal-
cent (pre-2002) texts included more synthesis (but perhaps not as ysis. While the students can benefit from linear systems theory,
much as he desired). Laplace transforms, etc., it is possible that they walk away assum-
Edgar et al. (2006) summarize the content of typical process ing that control problems are clearly defined and that there is a
control courses, which differs little from 1988. The following single correct solution to the problem. When control-oriented fac-
changes are suggested: (i) incorporation of biological content ulty teach the course they can bring a broader perspective that ties
(particularly important, given that many departments had changed in safety, economics and other important issues into the discus-
their name to chemical and biological engineering, or something sion. Rather than simply designing a control strategy for a steam
similar), (ii) development of non-traditional examples, (iii) batch drum – which is interesting because it incorporates feed-forward,
vs. continuous processing emphasis, and (iv) reducing the em- feedback and cascade (three mode) control – they can tie-in safety.
phasis on Laplace transforms, frequency response and controller Should the boiler feedwater to the drum be fail-open or fail-
tuning. With coverage of additional topics, such as safety, it is closed? If the water overflows into the superheating section it can
recognized that the control course is getting squeezed. They blow out those tubes. If the level drops into the furnace tubes, they
also report a survey of people from industry, who provide the can overheat and fail, or have a reduced lifetime. Similarly, when
following rankings (scale of 1 to 10, with 10 most important): the load to a furnace changes, should the combustion air “lead” or
Process or operation optimization (8.6), Data statistical analysis “lag” the fuel gas flowrate? Naturally this depends on whether the
and experimental design (7.2), physical dynamic process mod- fuel gas flow is increasing or decreasing, and the strategy requires
eling (7.0), statistical/empirical dynamic process modeling (6.9), a combination of high/low selectors and ratio controllers; the point
multivariable interactions (6.6), statistical process control and is, an instructor with a strong control background is much more
process monitoring (5.3), PID loop design and tuning (5.1), non- likely to cover this material than one that is only comfortable with
linear dynamics and nonlinear system analysis (3.9), frequency the mathematics related to modeling and control.
546 B.W. Bequette / Computers and Chemical Engineering 128 (2019) 538–556

Table 1
Comparison of industry needs (2006) vs. academic coverage (2015).

Table 2
Comparison of 1988, predicted future 20 0 0, and estimated average 2015 courses.

Topic 1988 20 0 0 (proposed) 2015

Intro, feedback/feedforward 1 week 1 week


Math modeling 1 week 1 week
Linear, Laplace Transforms 2 weeks 2 weeks
Dynamic simulation 2 weeks 1 week
Response characteristics 1 week 1 week 2 weeks
Discrete-time models 1 week
Conventional & predictive structures 2 weeks
Controller hardware, Inst. 1 week 1 week
Closed-loop analysis, stability 1 week 1 week
Optimization methods for cont. design 2 weeks
Tuning PID 2 weeks 1 week 1 week
Frequency response 1-2 weeks 1-2 weeks
Adv. FF, cascade, MV, adaptive, etc. 3-4 weeks 2 weeks 2 weeks
Digital hardware/implementation 1 week
Plant control, case studies 1 week
Expert Systems 1 week
Misc. topics 1 week (safety)

While it would be great to have better integration and control off), “floating” (integral only), proportional, proportional-reset (PI),
across the curriculum, this is difficult with the “decentralized” ap- and proportional reset + rate (PID); thus, in 1945 the common PID
proach to teaching assignments – most departments give signifi- terms used today were not in common use.
cant autonomy to the individual instructors for the content cov- An Introduction to Process Control System Design (379 pages)
ered. So, while certain key overall learning objectives are common by Young at ICI in 1955 used frequency response plots that were
between instructors, there is potentially a wide variability in fo- in the “opposite direction” with period rather than frequency as
cus. In the next section we review undergraduate process control the x-coordinate, and attenuation, defined as the ratio of the input
textbooks, including the increasing amount of material covered, in to the output (rather than the amplitude ratio of output/input), as
part, to meet the needs of individual instructors. the y-coordinate. Cascade, ratio, averaging and feedforward (called
disturbance-feedback) control are covered. While his definition of
5. Process control textbooks controllability is somewhat vague (generally related to lags and
time-delays), he stresses the importance of designing processes for
It is interesting to note that there was much control-related controllability. Two detailed chapters cover PID algorithms in com-
textbook publishing activity immediately after World War II, par- mercial pneumatic and electro-pneumatic controllers, in compari-
ticularly related to “servomechanisms” (e.g. James et al., 1947; son with the ideal parallel PID algorithm.
Brown and Campbell, 1948), yet the earliest textbook involved pro- Automatic Process Control for Chemical Engineers (228 pages)
cess control. The Principles of Industrial Process Control book (235 by Ceaglske at Minnesota in 1956 (Ceaglske, 1956a,b) provides de-
pages) by Eckman at Case (with the support of the Brown instru- tails on instrumentation, and introduces block diagrams and trans-
ment company), published in 1945 (Eckman, 1945), is largely fo- fer function analysis. It does not cover Routh stability, but includes
cused on chemical process examples. In the preface he states that frequency response analysis. It does not present more complex
the book is intended for the student in chemical, metallurgical, control strategies such as feedforward, cascade or ratio. The open-
mechanical or electrical engineering. The are chapters providing loop Ziegler–Nichols procedure is covered.
reasonably detailed explanations of measuring and control instru- Automatic Process Control (368 pages) by Eckman at Case, pub-
mentation, and the majority of the examples are those that we use lished in 1958 (Eckman, 1958), is an extensive update of his previ-
in process control courses today: interacting tanks, stirred tanks, ous (1945) book. While he suggests that the textbook is for engi-
stirred tank heaters. He discusses step response-based modeling, neers in general, there is definitely a process focus appropriate for
including how these models are used in the Ziegler–Nichols open- undergraduate chemical engineers. Block diagrams, frequency re-
loop procedure. While differential equation models are shown, sponse, Routh stability, cascade and ratio control are covered. The
there is no block diagram analysis, and no frequency response open-loop Ziegler–Nichols method is covered. Feedforward control
analysis. Cascade, ratio and split-range control strategies are pre- is not covered. It is interesting that Eckman also describes a form
sented. His chapter on “control theory” covers “two-position” (on- of extremum-seeking control.
B.W. Bequette / Computers and Chemical Engineering 128 (2019) 538–556 547

Techniques of Process Control (303 pages) by Buckley at DuPont Frequency response is introduced early, with the notion of gain and
in 1964 was targeted more towards practicing engineers than un- phase margins presented by page 22!
dergraduate students; indeed, it assumes a relatively sophisticated Process Engineering Control (466 pages) by Tyner and May at
mathematical background as Laplace Transforms, frequency re- Florida in 1968 (Tyner and May, 1968) compares ideal (parallel)
sponse, block diagrams and z-transforms are all reviewed in the and actual (series for pneumatic) PID controllers, includes dead-
first 44 pages! It is the first process control book to present digital time and inverse response compensation (using a Smith predictor),
control in detail. Nonlinear valve and sensor characteristics, valve and appendices for a Fortran program for pulse testing and ana-
stiction, feedforward, cascade, and ratio control are all covered. log computing. Feedforward and cascade, but not ratio, control are
There is much discussion about the difference in time scales be- covered; the only discussion of multiloop systems refers to cascade
tween “material balance” and “quality control” loops. Models and control.
control strategies are presented for a large number of unit opera- Process Dynamics and Control (Vol. 1, 367 pages; Vol. 2, 447
tions. pages) by Douglas at Syracuse in 1972 develops the most exten-
Process Control (374 pages) by Harriott at Cornell, published sive modeling book to-date, including state space and distributed
in 1964 (Harriott, 1964), begins to look like many current process parameter models (Douglas, 1972). Volume 1 is focused on model-
control textbooks. In addition to block diagram, Routh stability and ing, while volume 2 is focused on process control, including mul-
frequency response analysis, he covers feedforward, cascade and tivariable control, optimal control and periodic processing. While
ratio control, and introduces the challenges associated with multi- he notes the sections that are appropriate for an undergraduate
ple control loops. He covers different types of pneumatic and elec- course, this is largely a textbook for graduate level courses in mod-
tronic analog controllers and provide an analysis of the difference eling and control.
between series (interacting) and parallel forms of a PID controller. An Introduction to Process Dynamics and Control (434 pages)
He covers both the Ziegler–Nichols reaction curve (open-loop) and by Weber at Buffalo in 1973 (Weber, 1973), provides a chapter on
continuous oscillation methods, and also proposes a damped oscil- control actions early in the text, before applying them in the con-
lation method that is more appropriate for process plants. There text of feedback control later. While the text covers frequency re-
are chapters devoted to heat exchangers, level, flow, distillation, sponse methods and feedforward control, it does not cover cascade
reactor (including multiple steady-states), pH and blending pro- or ratio control.
cesses. Process Modeling, Simulation and Control for Chemical Engi-
Process Systems Analysis and Control (485 pages) by neers (558 pages) by Luyben at Lehigh in 1973 (Luyben, 1973),
Coughanowr and Koppel at Purdue in 1965, while widely represents the first process control textbook to provide extensive
considered the precursor to current process control textbooks, coverage of digital computation methods. In the preface he refers
surprisingly does not cover feedforward, cascade or ratio control to the pioneering textbook by Ceaglske (1956a,b), as well as other
(Buckley, 1965). It does cover control loop interaction through process control textbooks, but points out the limited mathemati-
a distillation control example. Like Harriott, they cover different cal modeling and no digital simulation content. The second edition
types of pneumatic controllers and provide an analysis of the (Luyben, 1990, 721 pages) adds dynamic matrix control (DMC), sin-
difference between series (interacting) and parallel forms of a PID gular value analysis and the relative gain array, and Tyreus–Luyben
controller. Nonlinear phase-plane analysis and multiple steady- modifications to Ziegler–Nichols tuning. Luyben and Luyben (1997,
state behavior of a CSTR is studied. Finally, analog simulation 584 pages) note the growth in textbook length and thus reduce
methods are covered. Coughanowr was the sole author of the the amount of content for their textbook on the “Essentials of Pro-
second edition (Coughanowr and LeBlanc, 2009, 566 pages), which cess Control.” Some of the mathematical details are reduced, while
added cascade, feedforward, ratio, Internal Model Control and dig- Fortran and MATLAB code is supplied.
ital control to the previous topics. In addition, state space models, The Eisen et al. (1975) survey found that Coughanowr
additional multi-loop content, and some BASIC programming is and Koppel (51.9%), Luyben (12.5%), Weber (7.7%) and Harriott
added. Finally, the third edition (2009, 602 pages), with LaBlanc (7.7%) were the more commonly used textbooks. The survey by
and Coughanowr as authors, incorporates MATLAB, Simulink and Seborg (1980) found that the leading textbooks were Coughanowr
Excel examples throughout the text. The digital control content and Koppel (50.7%), Luyben (15.4%), Weber (6.6%), Harriott (4.4%),
is reduced, and there is no coverage of the relative gain array or Douglas (4.4%), Perlmutter (2.2%), Smith (1.5%), with 19 other texts
model predictive control. reported.
Introduction to Chemical Process Control (204 pages) by Perl- Chemical Process Control: An Introduction to Theory and Prac-
mutter at Penn in 1965 (Perlmutter, 1965), while concise, manages tice (696 pages) by Stephanopoulos at MIT in 1984 provides a com-
to cover frequency response, cascade, feedforward and control-loop prehensive treatment of many topics, including more mathematical
interaction. He also covers phase plane analysis for nonlinear sys- modeling, and the design of control systems for entire plants – us-
tems. While frequency response and process reaction curve meth- ing the HDA example from Douglas (1982). Also, somewhat sur-
ods are discussed, the Cohen-Coon and Ziegler–Nichols PI parame- prisingly, given the power for multi-loop synthesis, this is the first
ters appear only as student exercise problems. textbook to cover Bristol’s relative gain array (RGA) analysis for
Automatic Control of Processes (467 pages) by Murrill at LSU MV-SISO controller synthesis. While the book has a strong model-
in 1967 (Murrill, 1967) covers feedforward, cascade and ratio con- based focus, it does not include internal model control or MPC;
trol, frequency response, and the Cohen-Coon and Ziegler–Nichols it does include digital control, including model-based techniques
open-loop methods. There are chapters on analog computer based such as Dahlin’s controller. Clearly this book had an immediate
simulations and controller hardware (pneumatic and electronic). impact, as the Eisen (1985) survey found that 48% of the courses
Automatic Process Control (272 pages) by Johnson at Princeton used Stephanopoulos, 20% Coughanower and Koppel, 11% Luyben
in 1967 (Johnson, 1967), like Perlmutter, is concise while covering and 20% other.
frequency response, cascade, feedforward, and ratio control. While Principles and Practice of Automatic Process Control (614 pages)
he shows an example multi-loop P&ID, he provides no numerical by Smith and Corripio at South Florida and LSU in 1985 (Smith and
analysis. He does show how a pneumatic PID control results in an Corripio, 1985) presents frequency response analysis, feedforward,
“interacting” (series) controller transfer function, such that the pro- cascade and ratio control, the use of the relative gain array and de-
portional gain is not truly independent of the integral and deriva- coupling for multi-loop systems, and includes a chapter on simula-
tive tuning parameters. Digital control receives a brief treatment. tion techniques. The second edition (Smith and Corripio, 1997, 768
548 B.W. Bequette / Computers and Chemical Engineering 128 (2019) 538–556

pages) expands the first-principles modeling content, adds digital no discussion of plantwide control or the optimization and control
control (including discrete IMC) and presents DMC examples. The hierarchy.
third edition (Smith and Corripio, 2005) (2005, 580 pages) expands My textbook on Process Control: Modeling, Design and Simu-
the modeling and simulation content and includes two case stud- lation (800 pages) published in 2003 (Bequette, 2003), provides a
ies for plantwide control. more general treatment of state space models, as well as a compre-
Process Dynamics and Control (736 pages) by Seborg, Edgar hensive treatment of model-based control and has a set of modules
Mellichamp, in 1989, at UCSB and UT-Austin includes direct syn- that can be used as tutorial examples throughout a dynamics and
thesis, internal model control (IMC), digital control and is the first control course. The textbook fully integrates MATLAB-based exam-
process control textbook to treat model predictive control. The ples. As I write the second edition I am realizing that I have over-
Griffith (1993) survey found that 46% of the courses used Seborg, emphasized Internal Model Control (IMC), so this content will be
Edgar and Mellichamp, 20% Stephanopoulos, 11% Smith and Corri- reduced and placed a supplemental material.
pio, and 11% Luyben. So, like in 1985 (when Stephanopoulos was Introduction to Process Control (528 pages) by Romagnoli and
most popular), the newest book was the most widely used. The Palazoglu at Louisiana State University and UC-Davis in 2006
second edition (2004, 713 pages) reduces some digital control con- (Romagnoli and Palazoglu, 2006), while a concise introduction, also
tent, places safety into a separate chapter, and adds process mon- includes elements of process monitoring and data reconciliation.
itoring, batch and plantwide control and a brief appendix on the The second edition (Romagnoli and Palazoglu, 2012, 643 pages)
use of MATLAB. The third edition (528 pages, 2010) with Doyle as provides more exercises.
a co-author, adds two chapters on biological systems, and reduces Silverstein et al. (2016), in a 2015 survey with a 48.7% insti-
content by using supplemental material in digital form. The fourth tutional response rate, cited the following textbook usage: Seborg
edition (512 pages, Seborg et al., 2016) downsizes slightly by fur- et al. (48%), Riggs and Karim (15%), Ogunnaike and Ray (12%),
ther reducing lengthy derivations. Bequette (6%), Smith and Corripio (6%), with all others receiving
Process Dynamics, Modeling and Control (1260 pages) by Ogun- 13%. Thus, the dominant textbooks for the following periods were:
naike and Ray at DuPont and Wisconsin in 1994 (Ogunnaike and 1965–1984: Coughanowr and Koppel, 1985–1990: Stephanopoulos,
Ray, 1994) is a comprehensive tome covering virtually every pro- 1991-current: Seborg et al.
cess control topic, including statistical process control and several Designing Controls for the Process Industries (394 pages) by
case study examples. The books length is due to the particulary Seames at North Dakota in 2017 (Seames, 2017a,b) has reduced
large number of examples; since many different types of dynamic transfer function-based content (which appears in appendices),
behaviors are studied in-depth, the feedback control system design providing a focus on the process and specific unit operations con-
chapter does not begin until page 461! Many of the more com- trol problems.
plex examples are from Professor Ray’s lab at Wisconsin, and from There is certainly no shortage of available process control text-
studies by Ogunnaike and colleagues at DuPont, so the book could books. Current process control textbooks cover much of the same
easily be used as part of an introductory graduate course in pro- analysis content as Coughanowr and Koppel (1965), which was the
cess control. textbook used in the course that I took in 1979. Most books have
Process Control: Designing Processes and Control Systems for added topics and examples and make use of computer-aided soft-
Dynamic Performance (954 pages) by Marlin at McMaster in 1995 ware, such as MATLAB. Generally, the texts cover far too many top-
(Marlin, 1995) stresses the importance and integration of process ics for a one-semester course, so it is important that an instructor
design, operability and safety. The majority of the P&ID are more be selective about the material covered. As summarized in Fig. 1,
realistic than the over-simplified ones found in most textbooks. the textbooks grew in length during the 1980-20 0 0 period, but
The textbook is also integrated with an interactive learning web- have generally been reduced in length with supplemental material
site and a set of workshops or learning moduels. The second edi- placed on websites.
tion (Marlin, 20 0 0, 1017 pages) adds additional case study content. While the most important fundamental topics covered in pro-
It should be noted that the book is no longer distributed by Mc- cess control textbooks has changed little over the decades, per-
Graw Hill and is available for no cost by download as a .pdf. haps we should be satisfied with the knowledge they same is true
Chemical Process Control (361 pages) by Riggs at Texas Tech of control engineering textbooks used in electrical and mechanical
in 1999 (Riggs, 1999) presents an applied focus, with less transfer engineering as well. Dorato (2007) notes that most undergraduate
function and no frequency response analysis. The second edition, control engineering textbooks (for examples: Franklin et al., 2006;
expanded to 545 pages, was published in 2002 (Riggs, 2002). The Phillips and Harbor, 20 0 0) do not cover theory developed after
third edition with Karim (2007, 578 pages) incorporates biological state-space methods of the 1960s. Typical topics include modeling
systems, adds examples, includes a process simulator, and expands (transfer functions and state space models), dynamic response, sta-
on model predictive control. The fourth edition (570 pages, Riggs bility (Routh–Hurwitz), root-locus, frequency response, and digital
and Karim, 2016) adds process safety, MATLAB and Simulink con- control; these topics are typically covered during the first semester.
tent. Topics covered during the second semester often include state-
A Real-Time Approach to Process Control (307 pages) by space design (state and state-estimate feedback, LQR design), digi-
Svrcek et al. (20 0 0) presents a simulation-based approach with less tal control (state space design), and nonlinear systems; appendices
focus on transfer functions and does not cover frequency response. typically include matrices and complex variables.
Distillation control is used as an example for a number of control
loops, and a series of eight workshops provide “hands on” rein- 6. The flipped classroom approach at RPI
forcement of the methods covered. The second edition (2006, 325
pages) adds some material on decoupling. The third edition (2014, 6.1. Evolution of teaching by the author
342 pages) adds brief coverage of model predictive control.
Process Control: A First Course with MATLAB (330 pages) by My initial experience teaching process control was at UC-Davis
Chau at UC San Diego in 2002 (Chau, 2002) is the first process in 1987–88. I used the Stephanopoulos (1984) textbook and ana-
control textbook to incorporate MATLAB, and includes a compre- lytical solutions. I arrived at RPI in 1988 and was impressed with
hensive set of seven MATLAB Tutorial Sessions. The book con- the computing facilities and software, so I scheduled a weekly
tent is certainly reduced compared to more recent books, with computer laboratory for the process control class. Initially I used
less multi-loop coverage, no digital or model predictive control, the IBM simulation package, CSMP, but switched to MATLAB after
B.W. Bequette / Computers and Chemical Engineering 128 (2019) 538–556 549

Fig. 1. Length (pages) of process control textbook by publication year.

a year or so. The regular classroom sessions remained largely 6.3. Screencasts
lecture-based with handwritten derivations, but with as much
engagement and active discussions with students as possible. To A common characteristic of the flipped classroom is that stu-
further motivate students we developed a set of case studies that dents read and view important lecture material outside the class
the students, in groups of two, would work on during the final 4 so that more time can be spent using engaging, interactive learning
weeks of the course (Bequette et al., 1998). They would select from techniques in the classroom. I feel that it is better for the students
a set 4 topics, perform a literature review, develop models based to view a couple of short “screencasts” of 5–7 min each, rather
on step-testing, design SISO controllers, then use the relative gain than viewing a videotaped lecture of 50 min in length. The idea is
array (RGA) to decide on controller pairings for the multivariable to focus on a specific topic or technique and to have their full at-
case. Requirements included short reports each week, a final writ- tention during the short screencast. The University of Colorado has
ten report and a final presentation. I would serve as the advisor developed a large set of screencasts for most of the core chemical
for one case study, while the TA and two of my other graduate engineering courses, available at learncheme.com, with 58 videos
students would be the advisors for the other three case studies. related to process control. Sometimes the notation is slightly dif-
This was very time-intensive but the students were clearly moti- ferent than that used in my course, so I warn students of this both
vated by the opportunity to choose a case study in their area of pre- and post-viewing.
interest.
6.4. Brief on-line quizzes
6.2. Studio classrooms
The flipped classroom works best if all of the students have
Interactive learning is the hallmark of the Montessori method read or viewed the lecture material in advance. I require the stu-
developed over 100 years ago and used primarily in classrooms dents to take a brief on-line quiz, automatically graded, before
with young children. At the college level interactive learning often class. This typically consists of 5 multiple choice or true-false ques-
included students solving problems at the chalkboard (see Fig. 2.), tions that are relatively easy if they have covered the material.
and at RPI there are still many classrooms where chalkboards line Questions are sometimes related to the previous lecture, often in-
every wall of the room. volving impromptu discussions. For example, when Hurricane Har-
The studio classroom approach was pioneered at RPI, with ini- vey hit the Gulf Coast in August 2017, we discussed refining capac-
tial courses in physics and calculus in 1993 (Wilson, 2002). The ity and the effect of plant shutdowns on gasoline prices. I noted
mantra was to change the instructor’s role from “the sage on the that I had worked at a refinery in Port Arthur, Texas and asked
stage” to the “guide on the side.” The basic idea was to have re- the students to name a famous former blues/rock singer from that
duced lecture content, and instead, promote learning through dis- area; since no one could name Janis Joplin the next quiz included
cussion and problem-solving. The first studio classrooms had net- questions based on a Wikipedia page on Janis Joplin.
worked computer workstations, with 2–4 students per computer.
I started using a studio classroom for process dynamics and con- 6.5. In-class simulation exercises
trol in 1999 (Bequette et al., 1999). That classroom could handle
up to 40 students that could face forward during lectures and dis- In-class simulation exercises change in difficulty throughout the
cussions, then swivel 180° to work at computer workstations in semester. During early exercises I have the students construct sim-
groups of two. This was not an efficient use of space, and in 1999 ple Simulink diagrams. As we get into more difficult problems, e.g.
RPI began requiring that entering freshmen (class of 2003) pur- cascade control, I supply them with .mdl and script files that they
chase laptop computers supplied with relevant computational soft- can modify. I feel that over the years some Simulink blocks have
ware (Bequette, 2005). gotten too flexible and difficult for students to follow; I need to
550 B.W. Bequette / Computers and Chemical Engineering 128 (2019) 538–556

Fig. 2. Rensselaer Polytechnic Institute classroom, early 1900 s. Used with permission.

spend time explaining and discussing the different ways to im- can be specified for a particular failure mode, so I present cross-
plement a continuous PID controller, for example. Indeed, when I sectional diagrams of typical valves and actuators for discussion.
cover digital control I supply my own block that corresponds to my When discussing liquid surge drums we think about realistic val-
preferred derivation and implementation of digital PID. ues for the high and low level alarms. We also take groups of stu-
dents on a boilerhouse tour with the operator pointing out the var-
6.6. Weekly homework assignments ious vessels, sensors, actuators and the control room and control
system. This is of particular benefit to students that have not had
Homework problems are assigned weekly and cover analyti- summer industrial experience.
cal and MATLAB/Simulink-based solutions. At the time of the as-
signment the particular topics have not been covered yet, so we 6.10. Course topic overview
quickly go over the assignment and the particular challenges to
motivate them for the techniques being covered that week. The Chemical Process Dynamics and Control is a 4-credit course
daily in-class exercises are used as a springboard for the solution scheduled for 3 days/week (Tu/We/Fr) for one hour and 50 min
of the weekly homework assignments. each class period. As a practical matter, to give adequate time for
homework problem solving and to accommodate the instructors
6.7. Case studies and detailed problems travel schedule, during many weeks the third class period serves
as a recitation, with the TA reinforcing material, and assisting with
Most years I have used case study projects for the final month MATLAB, as examples. Since it is a 4-credit course, we tend to
of the semester, as discussed in Section 6.1. The past two years cover about 33% more topics than is covered in most 3-credit pro-
I decided to use a distillation control problem as a multivari- cess control courses. A typical 15-week semester will include 32
able study example. Another detailed example used and discussed lecture/regular classes, 7 recitations, 2 in-class exams, 1 campus
throughout the course is an automated insulin delivery system, as boilerhouse tour, 1 final course review, and 1 final exam. A sum-
presented in Section 6.11. mary of topics is provided in Table 3, with specific examples in
Table 4.
6.8. Mid-term and final exams
6.11. Motivating example: automated insulin delivery
While the in-class exercises and weekly homework assignments
involve the use of MATLAB and Simulink, the two mid-term exams In this learning module we first review the natural physiologic
and the final exam are based on analytical solution techniques and control loops involved when a healthy pancreas regulates blood
fundamental modeling and control understanding. glucose. We then describe the lifestyle of an individual with type
1 diabetes, who must monitor blood glucose levels and administer
6.9. Additional topics: safety, process design, boilerhouse tour insulin, either through multiple daily injections or using a contin-
uous insulin infusion pump. We describe typical ranges for basal
Process safety is discussed, in one way or another, in almost ev- (steady-state) insulin infusion, carb-to-insulin ratios for meals, and
ery lecture. Most control strategies involve a discussion of whether correction factors (amount of insulin required to reduce a glu-
a fail-open or fail-closed value should be used, for example. Early cose level by a certain amount). Someone with type 1 diabetes
on most students do not have an appreciation of how a valve serves as a controller (feedforward and feedback) through constant
B.W. Bequette / Computers and Chemical Engineering 128 (2019) 538–556 551

Table 3 bohydrates in a meal makes it safer to under-bolus to avoid hy-


Course topics.
poglycemia.
Incentives/Motivation for Process Control • Appreciate the importance of maintaining a consistent healthy
Mathematical modeling value of blood glucose. Blood glucose that is too low (hypo-
Linearization, state space models
glycemia) can cause short-term dangers, such as drowsiness
Laplace transforms, transfer function analysis
Connection between poles and eigenvalues or a diabetic coma. Blood glucose that is too high (hyper-
Dynamic behavior, step and impulse inputs glycemia) causes long-term risks, such as micro- and macro-
Introduction to feedback (FB) control, PID vascular problems resulting in retina and other problems.
Closed-loop stability • Understand the metrics used in reporting clinical perfor-
Real PID, filtering, derivative on output
mance of glucose control strategies, such as mean glucose,
Internal Model Control and IMC-based PID
Digital PID and digital filtering time-in-range (70–180 mg/dL), time in hypoglycemia (less than
Cascade (CC) and feedforward (FF) 70 mg/dL), time in hyperglycemia (greater than 180 mg/dL), and
Anti-reset-windup total daily insulin dose (insulin used each day).
Nonlinear PID – surge vessel example • Approximate the dynamic behavior between insulin and meal
Closed-loop control interaction, RGA
Singular Value Decomposition (SVD) inputs and blood glucose outputs using low-order transfer func-
Plantwide control, startup/shutdown (brief) tion models.
Model predictive control (brief) • Understand the importance of sensors that report discrete
Statistical process control (brief) quantized values (for example, a continuous glucose sensor re-
Optimization: Linear Programming (brief)
ports to the nearest 1 mg/dL).
• Understand the effects of sensor noise and bias on control sys-
Table 4 tem performance. A mis-calibrated glucose sensor can be off by
Example processes. 10–20 mg/dL or more.
Gas and liquid surge vessels
• Understand that providing an insulin bolus at mealtime (feed-
Insulin pharmacokinetics forward control) leads to improved glucose control performance
Biochemical reactor compared to feedback-only control.
Jacketed reactors (batch & continuous)
Steam drum, 3-mode (FF/FB/CC) Students are provided with a Simulink diagram (Fig. 4)
Gasoline blending – ratio control that includes the insulin-meal-blood glucose model of
Automated insulin delivery for type 1 diabetes:
(i) response to meals w and w/o feedforward,
Hovorka et al. (2004) and Wilinska et al. (2005). We added
(ii) effect of sensor bias, quantization & noise additional lags for the dynamics between the blood and interstitial
Drug infusion in anesthesia – multiple loops fluid and the sensor lag, with possible bias and measurement
Distillation control – multiple loops noise. The simulation model also includes a digital PID controller,
Gasoline blending – LP problem
and it is suggested that the students begin with the tuning param-
eters proposed by Palerm (2011). An outcome of the simulation
exercise is to suggest controller tuning and an appropriate blood
glucose setpoint if the sensor can be biased by up to 20 mg/dL,
with the objective of reducing the risk of hypoglycaemia to
less than 5% (that is, less than 72 min per day spent less than
70 mg/dL).

6.12. Discussion

In a process dynamics and control course the specific examples


used may not be as important as the instructors enthusiasm in
presenting them. I have been fortunate to be involved in a wide-
range of automation and control related problems, ranging from
classic chemical processes (petroleum refining, pharma batch reac-
tors) to biomedical systems (automated insulin delivery, drug in-
fusion in anesthesia). It certainly helps that I can bring specific
examples that I have worked on into the classroom – such as re-
sponding to level sensor failures while serving as an operator in
the refining industry, and conducting clinical trials of an automated
Fig. 3. Example of an automated insulin delivery system currently under study. The
insulin delivery system. The final course survey results, specifically
components, using Bluetooth communication, include a continuous glucose monitor
(CGM), a control algorithm on the smart phone and insulin infused by the insulin
for the instructor-added questions, shown in Table 5 indicates,
pump. that students were generally satisfied with the course; indeed it

Table 5
diligence; this motivates the development of an automated insulin Course survey responses to instructor-added questions (median, out of 5).
delivery system shown in Fig. 3.
The screencasts have been useful 4.28
After reading this module and performing the studies a student
On-line quizzes have been useful 3.60
is able to: Studio approach has been useful 4.40
I have benefited from reading the textbook 4.14
• Appreciate the challenges someone with type 1 diabetes faces I have adequate time for homework 4.13
in managing blood glucose. Meals increase and insulin de- I appreciate the importance of process control in industry 4.62
MATLAB has been easy to use 2.75
creases blood glucose. One challenge is providing the correct
The boilerhouse tour was useful 4.62
insulin bolus to compensate for a meal. Uncertainty in the car-
552 B.W. Bequette / Computers and Chemical Engineering 128 (2019) 538–556

Fig. 4. Simulink diagram for automated insulin delivery. Includes open-loop behavior by setting the controller gain to 0.

appears that the on-line quizzes actually provided encouragement 8. Comments/Perspective on process control practice and
for them to read the textbook. education

I particularly enjoyed reading the detailed discussions at the


end of early process control papers. This discussion summary was
6.13. Approaches and material from other sources
also a very valuable record of the international conferences on
chemical process control through 2006. Now that selected papers
Huang (2017) discusses the results of integrating three
appear in special issues of Computers and Chemical Engineering
MATLAB-based learning modules into a flipped classroom by us-
their has only been a brief overall summary in the preface of the
ing teaching videos to be viewed a week before class and quizzes
special issue – I would like to see future CPCs include panel pre-
taken at the beginning of class. Marlin (2017) discusses “blended
sentations and discussions as part of the published special issue.
learning” and develops a flipped classroom as one example. E-
The scope of the process control course (and to a larger
lessons consist of pre-prepared slides with audio that can be stud-
extent the process control textbooks) has grown tremendously
ied via the internet at any time, followed by an un-graded quiz.
over the years, with an early focus on single loop PID de-
Class time then involves a mini-lecture reviewing the e-lesson,
sign/analysis/tuning. The current course at many schools could
then workshops (hands-on, interactive learning) to reinforce learn-
probably be better titled process operations, with some safety, sta-
ing material, followed by a mini-lecture on the next e-lesson).
tistical process/quality control, multivariable and even model pre-
Seames (2017a,b) takes a simulation-based approach, including a
dictive control at least briefly covered.
new textbook, and requires outside material be studied, with on-
There are several challenges in educating students in process
line quizzes; re-quizzes can be taken up to 3 weeks later. It
control. One, it would be nice if students were introduced to more
should be noted that the text by Svrcek et al. (2014) also takes
general dynamic balance equations in their introductory material
a simulation-based approach.
and energy balances course. Also, process control students are of-
There are many learning resources available on the internet. For
ten using MATLAB for the first time, and any software package re-
example, cache.org contains links to material from many sources.
quires a certain amount of time to learn. At RPI we are now teach-
The learncheme.com site is particularly strong in the use of screen-
ing our own numerical computing class, which includes basic pro-
casts. The Hedengren process dynamics and control course at BYU
gramming skills. We have chosen MATLAB as the programming en-
has lecture material and a low-cost temperature control experi-
vironment, which should be a tremendous help in the student use
ment (Hedengren, 2018; Learncheme, 2019).
of MATLAB in the process control course. I would also like to see
better integration of control topics in other courses. For example,
when I teach chemical reaction engineering I give examples and
7. Conclusions of flipped classroom assignments that require a temperature controller to keep a reac-
tor operating “isothermally.”
A flipped classroom makes better use of a students and instruc- There are many important topics that are relevant to process
tors time. Students come to class prepared, after taking an on-line control and design courses, including “big data” and data analytics,
quiz, and are more engaged in discussions and problem solutions. and at RPI there is now a “data dexterity” requirement, where stu-
Because of the interactive nature of the class-time, it is important dents take two data intensive course, including one in their major.
to have a teaching assistant willing to engage with the students. This has not been fully implemented, and likely we will include
B.W. Bequette / Computers and Chemical Engineering 128 (2019) 538–556 553

data intensive modules in more than one of our core chemical en- After graduation, my first day as a process engineer occurred
gineering courses. immediately after the refinery (American Pertrofina, Port Arthur,
So what is the practical reality of process control education dur- Texas) went on strike, so I began working 12 h shifts as a process
ing the next 20 years? Content wise, it is not likely that much will operator in the boilerhouse unit. Since all process units either con-
change. To maintain a 128–135 h eight-semester curriculum with sume or generate steam, return condensate to the boilerhouse, and
various breadth and depth requirements, the total chemical en- take or contribute to the fuel gas system, it was a good way for
gineering content cannot change much. Newer teaching methods, me to understand the entire refinery operation. My first stint dealt
such as the flipped classroom approach, make the learning process with the de-ionization unit, which was largely a sequential batch
more efficient – to me, by properly combining screencasts, on-line operation, with beds being switched on and off-line, being regen-
quizzes, homeworks and in-class discussions, we can make certain erated, etc. An important part of my responsibility was to intervene
key concepts clearer. Can we truly get them to “learn” the entire and force relays to switch to the proper position when they failed
content presented in a 10 0 0 page textbook? No. Then again, text- to do so automatically. The unit supplied de-ionized water to the
books have covered too much material for a long time to enable boilerhouse, with a storage tank that had a least a 12-hour resi-
instructors to “pick and choose” the content they wish to cover. dence time. Thus, I could play with the few continuous controllers
This works well for instructors with expertise in process control in the de-ionization unit, setting up continuous oscillations, for ex-
applications, but can be very difficult for instructors with other ample, without worrying about causing problems in the rest of the
backgrounds. My experience is that those instructors focus more plant.
on the development of mathematical models and their analytical I then moved to the main boilerhouse, where three boilers pro-
solution, often barely getting through basic PID loops before the duced much of the plant steam needs. Although this was 1980,
end of the semester. Now, this is not such a bad thing as long as all process units had analog pneumatic controllers, except for one
the students have developed an appreciation for process dynamics that had been retrofitted with electric analog controllers in the
and the importance of mathematical models. 1960s. Some of the controllers were the circular chart recording
controller variety (many textbook P&IDs still show TRC or PRC
symbols for temperature or pressure-recording-controller), where I
Acknowledgments
quickly learned that knocking the recording pen changed the con-
troller output because of the direct hardware linkage. Many impor-
I wish to acknowledge the enthusiasm and clear love of teach-
tant variables were recorded in log books at different times during
ing that Jim Turpin exhibited in my undergraduate process con-
the shift – often, if you did not like a gauge reading, you could
trol course at Arkansas. The various engineering projects, and time
tap the gauge and the needle would jump to a different value. A
spent serving as a process operator during strike duties at Amer-
control room in those days did not suffer from “alarm flooding” –
ican Petrofina motivated me to focus on process control in my
there were only a few alarms and when they were activated you
graduate students. Fortunately, Tom Edgar accepted me in to his
knew to take immediate action. I got plenty of exercise sprinting
research group at Texas – his ability to multi-task is unparalleled
up a level of stairs whenever the deaerator level alarm went off.
and he has been a great influence on me throughout my academic
I got experience switching to manual control under certain con-
career. Also at Texas I had the chance to “co-teach” (OK, provide
ditions – for example, when a drum level sensor failed. In that
a few lectures) in a course on Advanced Process Control that Jim
case the redundant “sensor” consisted of a light shining through
Rawlings was teaching; his unique ability to question the “standard
the sight glass into a mirror that was angled towards the ground;
assumptions” involved in what, I thought, were solved problems,
thus, an operator could monitor the level visually, typically main-
was quite impressive and I have enjoyed our many discussions over
taining it by setting the boiler firing rate to a constant value and
the years. As this paper was coming together I had the opportunity
making minor feedwater flowrate adjustments.
to serve as a “second opponent” for a thesis in Sigurd Skogestad’s
My primary role as a process engineer did not directly involve
group at NTNU in Trondheim. I was lucky that the “first opponent”
process control, but rather on “energy conservation,” Thus, my
was Karl Astrom; during our discussions as we walked between
main activities were justifying retrofit projects that could reduce
the hotel and campus each day I learned a lot about his experi-
plant energy consumption, and there were many projects with a
ences with computer control at IBM in the late 1950’s and with
payout period of less than two years. I then was given the respon-
different forms of PID controllers. I also wish to acknowledge the
sibility as serving as the lead process engineer for project involv-
outstanding efforts of Edmund Tang, Sambit Ghosh, Shu Yang and
ing the installation of an Aromatics Extraction Unit. Although the
Pranesh Navarathna, who served as the TAs the first three times I
bulk of the design work was done by a contract engineering firm, I
used a flipped classroom approach.
was responsible for the raffinate splitter design, including develop-
ing the P&ID. I certainly made use of the books by Greg Shinskey
Appendix. Background and biases of the author in this endeavour. In any event, when I decided to go to gradu-
ate school (my plan all along), there was no doubt that I wanted
I was an undergraduate at the University of Arkansas, a school to study process control, and who better to study with than Tom
that was applications oriented and took a “unit operations” rather Edgar at the University of Texas (see acknowledgements section).
than “transport phenomena” approach in most courses, and I was While at RPI I spent a one semester sabbatical at Merck work-
fortunate to have co-operative education experience in specialty ing on reaction calorimetry and performing simulations for a new
chemicals and summer process engineering work experience in multi-scale pilot plant that was proposed to use different heat
refining. I was also fortunate to take an elective course in pro- transfer fluids than previous plants. My focus, then, was to un-
cess control from Jim Turpin, who used Coughanowr and Kop- derstand the effect of proposed changes on the ability to control
pel (1965) as the textbook. Professor Turpin used the chalkboard a wide range of reactor sizes and characteristics. This experience
approach, with a pace that encouraged much thought and discus- provided examples that I use in the control class.
sion; while fewer topics were taught than in a typical current pro- Early in my career at RPI I became involved with Rob Roy (an
cess control course, I would argue that the ones covered were pre- anesthesiologist and head of Biomedical Engineering) and Howard
sented in more depth. I’ll admit, however, that I did have this vi- Kaufamn (adaptive control) on a drug infusion project, with goals
sion of operators cycling plant inputs to get frequency response re- of regulating blood pressure and cardiac output during and post-
sults – but, more on that later. surgery. This effort provided motivating case study examples of
554 B.W. Bequette / Computers and Chemical Engineering 128 (2019) 538–556

multivariable control for students. After Rob retired I began to fo- Cohen, G.H., Coon, G.A., 1953. Theoretical considerations of retarded control. Trans.
cus my biomedical control activities in the field of automated in- ASME 827–834.
Cooper, D.J., 1996. Picles: a simulator for “Virtual World” training in process dynam-
sulin dosing for people with type 1 diabetes; Bruce Buckingham at ics and control. Comput. Appl. Eng. Educ. 4 (3), 207–215.
Stanford became a long-time collaborator of mine in this area. Coughanowr, D.R., Koppel, L.B., 1965. Process Systems Analysis and Control. Mc-
Graw-Hill, New York.
Coughanowr, D.R., LeBlanc, S.E., 2009. Process Systems Analysis and Control, third
References ed. McGraw-Hill, New York.
Cox, R.K., Smith, J.F., Dimitratos, Y., 2006. Can simulation technology enable a
Ahrendt, W.R., Taplin, J.F., 1951. Automatic Feedback Control. McGraw-Hill. paradigm shift in process control? Modeling for the rest of us. Comput. Chem.
Aikman, A.R., Rutherford, C.I., 1951. The characteristics of air-operated controllers. Eng. 30, 1542–1552.
Automatic Manual Control: Proceedings of the Cranfield Conference. Butter- Cutler, C.R., Ramaker, B.L., 1980. Dynamic matrix control - a computer control al-
worths, Tustin A (ed.) pp. 176–187. gorithm. In: Proc. Joint Automatic Control Conference. San Francisco, CA Paper
Alford, J.S., 2006. Bioprocess control: advances and challenges. Comput. Chem. Eng. WP5-B.
30, 1464–1475. Cutler, C., Yocum, F.H., 1991. Experience with the DMC inverse for identification. In:
Allen Jr., L.H., 1943. Industrial control instrument settings. Ind. Eng. Chem. 35 (12), Proceedings of the Fourth International Conference on Chemical Process Control
1223–1229. (CPC IV). Padre Island, TX, February 17-21 (1991). CACHE, pp. 297–317.
Allen Jr, L.H., 1944. Automatic-control instrument adjustment. Ind. Eng. Chem. 36 Daoutidis, P., Bartusiak, D., 2013. Perspective on CPCVIII. Comput. Chem. Eng. 51,
(8), 687–693. 1–3.
Arant, J.B., 1948. Process instrumentation in chemical engineering curriculum. J. Davis, J., Edgar, T., Porter, J., Bernaden, J., Sarli, M., 2012. Smart manufacturing, man-
Chem. Educ. (May) 283–285. ufacturing intelligence and demand-dynamic performance. Comput. Chem. Eng.
Arkun, Y., Garcia, C.E., 1996. Process control. In: Carnahan, B. (Ed.), Computers in 47, 145–156.
Chemical Engineering Education. CACHE, pp. 193–210. deFlorez, L., 1937. Automatic Control and the Chemical Industries. Ind. Eng. Chem.
Astrom, K.J., 1985. Process control – past, present and future. IEEE Control Syst. Mag. 1210–1213.
(August) 3–10. Denn, M.M., 1982. Industrial theme problems in process control. In: Seborg, D.E.,
Astrom, K.J., Hagglund, T., 2005. Advanced PID Control. ISA. Edgar, T.F. (Eds.), Proceedings of CPC2, pp. 609–612.
Athans, M., 1976. Trends in modern systems theory. In: Foss, A.S., Denn, M.M. (Eds.). Dillon, C., 2012. Models: what do engineers see in them? In: Bissell, C, Dillon, C.
Chemical Process Control, 72 AIChE Symposium Series, Number 159. pp. 4–12. (Eds.) Ways of Thinking, Ways of Seeing: Mathematical and other Modelling in
Axelby, G, 1964. The gap – form and function. IEEE TAC 125–126. Engineering. Springer, pp. 47–69.
Ayers, E., 1952. An automatic chemical plant. Sci. Am. 187 (3), 82–96. Dobson, J.G., 1951. Quality control in the process industries. Ind. Eng. Chem. 43 (12),
Bennett, S., 1979. A History of Control Engineering. Peter Peregrinus, pp. 1800–1930. 2695–2703.
Bennett, S., 1984. Nicolas Minorsky and the automatic steering of ships. IEEE Control Dorato, P., 2007. Review of introduction to feedback control. IEEE Control Syst. Mag.
Syst. Mag. (November) 10–15. (February) 75–76.
Bennett, S., 2001. The past of PID controllers. Ann. Rev. Control 25, 43–53. Douglas, J.M., 1972. Process Dynamics and Control. Prentice Hall Volumes 1 and 2.
Bennett, S., 2002. Production control instruments in the chemical and process in- Douglas, J.M., 1982. Process operability and control of preliminary designs. In:
dustries. In: Morris, P.J.T. (Ed.), From Classical to Modern Chemistry: The Instru- Edgar, T.F., Seborg, D.E. (Eds.), Chemical Process Control 2. Engineering Foun-
mental Revolution. Royal Society of Chemistry, pp. 149–167. dation, pp. 497–524.
Bequette, B.W., Schott, K.D., Prasad, V., Natarajan, V., Rao, R.R., 1998. Case study Douglas, J.M., 1988. Conceptual Design of Chemical Processes. McGraw-Hall.
projects in an undergraduate process control course. Chem. Eng. Educ. 32 (3), Dow, H.H., 1930. Economic trend in the chemical industry. Ind. Eng. Chem. 22 (2),
214–219. 113–116.
Bequette, B.W., Chow, J.H., Li, C.J., Maby, E., Newell, J., Buckbee, G., 1999. An inter- Downs, J.J. 2002. Linking control strategy design and model predictive control. In
disciplinary control education studio. In: Proc. Conf. Decision & Cont. Phoenix, Proceedings of the Sixth International Conference on Chemical Process Control.
pp. 370–374. Rawlings, J.B., Ogunnaike, B.A. and Eaton, J.W. (eds.), CACHE/AIChE, pp. 328–341.
Bequette, B.W., 1998. Computer applications in process dynamics and control Downs, J.J., Vogel, E.F., 1993. A plant-wide industrial process control problem. Com-
courses. Comp. Applic. Eng. Educ. 6 (3), 193–200. put. Chem. Eng. 17 (3), 245–255.
Bequette, B.W., 2003. Process Control: Modeling, Design and Simulation. Prentice Doyle III, F.J., Gatzke, E.P., Parker, R.S., 1998. Practical case studies for undergradu-
Hall, Upper Saddle River, NJ. ate process dynamics and control using process control modules. Comput. Appl.
Bequette, B.W., 2005. A laptop-based studio course for process control. IEEE Control Eng. Education 6 (3), 181–191.
Syst. 25 (1), 45–49. Doyle III, F.J., Gatzke, E.P., Parker, R.S., 20 0 0. Process Control Modules: A Software
Bequette, BW., 2018. Innovations in process control education: a flipped class- Laboratory for Control Design. Prentice Hall.
room/studio approach. In: Eden, M, Ierapetritou, M, Towler, GP (Eds.), Proceed- Eckman, D.P., 1945. Principles of Industrial Process Control. Wiley.
ings of the 13th International Symposium on Process Systems Engineering – PSE Eckman, D.P., 1958. Automatic Process Control. Wiley.
2018. San Diego, CA, pp. 63–70. Edgar, T.F., 1990. Process control education in the year 20 0 0: a round table discus-
Biegler, L.T., Rawlings, J.B., 1991. Optimization approaches in nonlinear model pre- sion. Chem. Eng. Educ. 24 (2), 72–77.
dictive control. In: Proceedings of the Fourth International Conference on Chem- Edgar, T.F., Ogunnaike, B.A., Downs, J.J., Muske, K.R., Bequette, B.W., 2006. Ren-
ical Process Control (CPC IV). Padre Island, TX. February 17-21 (1991) CACHE, ovating the undergraduate process control course. Comput. Chem. Eng. 30,
pp. 543–571. 1749–1762.
Bode, H.W., 1940. Relations between amplitude and phase in feedback amplifier de- Eisen, E.O., Hubbard, R.M., Perna, A.J., 1975. Summary report: teaching of under-
sign. Bell Syst. Tech. J. 19, 421–454. graduate process dynamics and control. Presented at the 1975 AIChE Annual
Bode, H.W., 1945. Network Analysis and Feedback Amplifier Design. Van Nostrand. Meeting, Los Angeles, November.
Bode, H.W., 1960. Feedback – the history of an idea. In: Bellman, R., Kalaba, R. Eisen, E.O., 1985. Summary Report: teaching of undergraduate process dynamics
(Eds.), Proceedings of the Symposium on Active Networks and Feedback Sys- and control. Presented at the 1985 AIChE Annual Meeting, Chicago, IL.
tems. Polytechnic Institute of Brooklyn. Reprinted in Selected Papers on Mathe- Ferrell, E.B., 1945. The servo problem as a transmission problem. In: Proceedings
matical Trends in Control Theory. Dover, pp. 126–143. 1964. IRE, November, pp. 763–767.
Bristol, E.H., 1966. On a new measure of interactions for multivariable process con- Foss, A.S., 1973. Critique of chemical process control theory. AIChE J. 19, 209–214.
trol. IEEE Trans. Autom. Control 133–134 AC-11. Foss, A.S., Denn, M.M. (Eds.), 1976. Chemical process control. AIChE Symposium Se-
Brown, G.S., Campbell, D.P., 1948. Principles of Servomechanisms. Wiley. ries, 72 Number 159.
Brown, G.S., Campbell, D.P., 1952. Control systems. Sci. Am. 187 (3), 57–64. Franklin, G.F., Powell, J.D. and Emami-Naeimi, A. 2006. Feedback control of dynamic
Brown, R.P., 1918. The automatic control and measurement of high temperatures. systems. fifth ed. Prentice Hall.
Ind. Eng. Chem. 10 (2), 133–135. Franks, R.G., Worley, C.W., 1956. Quantitative analysis of cascade control. Ind. Eng.
Buckley, P., 1965. Techniques of Process Control. Wiley. Chem. 48 (6), 1074–1079.
Ceaglske, N.H., Eckman, D.P., 1953. Automatic control of a pressure process. Ind. Eng. Friend, L., 1956. Transient chemical process behavior and control. Ind. Eng. Chem.
Chem. 45 (9), 1879–1885. 48 (6), 999.
Ceaglske, N.H., 1956a. Automatic Process Control for Chemical Engineers. Wiley. Froisy, J.B., Richalet, J., 1986. Industrial applications of IDCOM. In: Morari, M.,
Ceaglske, NH., 1956b. Fundamentals of automatic process control. Ind. Eng. Chem. McAvoy, J.T. (Eds.), Proceedings of the Third International Conference on Chem-
1002–1007. ical Process Control – CPCIII. CACHE and Elsevier, pp. 233–244.
Chau, P.C., 2002. Process Control: A First Course with MATLAB. Cambridge University Froisy, J.B., 2006. Model predictive control – building a bridge between theory and
Press. practice. Comput. Chem. Eng. 30, 1426–1435.
Chestnut, H., 1965. Bridging the gap in control – status 1965. IEEE Trans. Autom. Fuld, G.J., Dunn, C.G., 1957. New process control applications in fermentation. Ind.
Control 125–126. Eng. Chem. 49 (8), 1215–1220.
Clark, H.A., 1922. Some phases of automatic process control. Ind. Eng. Chem. 14 (11), Fuller, A.T., 1963. Directions of research in control. Automatica 1, 289–296.
1016–1017. Garcia, C.E., Morari, M., 1982. Internal model control. 1. A unifying review and some
Clark, R.N., 1965. 1965 Automatic Control Conference. IEEE Spectrum, pp. 81–82. new results. Ind. Eng. Chem. Process Des. Dev. 21, 308–323.
Clarke, D.W, 1991. Adaptive Generalized Predictive Control. In: Proceedings of the Garcia C.E. and Prett, D.M. 1986. Advances in industrial model predictive control.
Fourth International Conference on Chemical Process Control (CPC IV). February In: Proceedings of the Third International Conference on Chemical Process Con-
17-21 (1991). CACHE, Padre Island, TX, pp. 395–417. Padre Island, TX, February trol (CPCIII), Janurary 12-17, 1986. Morari, M and McAvoy, T.J. (eds.) CACHE and
17-21 (1991). CACHE. Elsevier pp. 245–293.
B.W. Bequette / Computers and Chemical Engineering 128 (2019) 538–556 555

Garcia, C.E., Ramaker, B.L., Pollard, J.F., 1991. Total process control – beyond the de- Mason, C.E., Philbrick, G.A., 1940. Automatic control in the presence of process lags.
sign of model predictive controllers. In: Proceedings of the Fourth International Trans. ASME 295–308.
Conference on Chemical Process Control (CPC IV). Padre Island, TX. February Mason, C.E., Philbrick, G.A., 1941. Mathematics of surge vessels and automatic aver-
17-21, 1991 CACHE, pp. 335–361. aging level control. Trans. ASME 63 (7), 589–601.
Ginsberg, 1929. Automatic control in the chemical industries. Ind. Eng. Chem. 21 McAvoy, T.J., 1983. Interaction Analysis. ISA.
(6), 410–414. Marchetti, J.L., Mellichamp, D.A., D.E. Seborg, D.E., 1983. Predictive control based
Grebe, J.J., 1937. Elements of automatic control. Ind. Eng. Chem. 29 (11), 1225– on discrete convolution models. Ind. Eng. Chem. Process Des. Dev. 22 (3), 488–
1228. 495.
Grebe, J.J., Boundy, R.H, Cermak, R.W., 1933. The control of chemical processes. Mehra, R.K., Rouhani, R., 1982. Model Algorithmic Control (MAC): Review and Re-
Trans. AIChE 29, 211–256. cent Developments. Chemical Process Control 2. In: Edgar, T.F., Seborg, D.E.
Griffith, D.J., 1993. The teaching of undergraduate process control. Survey prepared (Eds.), Engineering Foundation, pp. 287–309.
by the Chemical Engineering Education Projects Committee. AIChE, 4 November Mellichamp, D.A., 1993. Foreword: industrial challenge problems in process control.
1993. Comput. Chem. Eng. 17 (3), v–ix.
Grosididier, P., Morari, M., Holt, B.R., 1985. Closed-loop properties from steady-state Minorsky, N., 1922. Directional stability of automatically steered bodies. J. Am. Soc.
gain information. Ind. Eng. Chem. Fundam. 24, 221–235. Naval Eng. 342, 280–309.
Harriott, P., 1964. Process Control. McGraw Hill. Morari, M., Ray, W.H., 1980. Integration of real-time computing into teaching: part
Harris, H., 1946. The frequency response of automatic control systems. Transactions II, the undergraduate course. Chem. Eng. Educ. 14 (1), 32–36.
65, 539–545. Morari, M., 1982. Integrated plant control: a solution at hand or a research topic
Hedengren, J. 2018. Dynamics and control. (accessed 7 February 2018) http:// for the next decade? In: Seborg, D.E., Edgar, T.F. (Eds.) Proceedings of CPC2,
apmonitor.com/che436/index.php/Main/HomePage. pp. 467–495.
Henson, M.A., Ogunnaike, B.A., Schwaber, J.S., Doyle, F.J., 1994. The baroreceptor re- Morari, M., 1983. Flexibility and resiliency of process systems. Comput. Chem. Eng.
flex: a biological control system with applications in chemical process control. 7 (4), 423–437.
Ind. Eng. Chem. Res. 33 (10), 2453–2466. Morari, M., 1987. Three critiques of process control revisited a decade later. In:
Henson, M.A., Ogunnaike, B.A., Schwaber, J.S., 1995. Habituating control strategies Prett, D.M., Morari, M. (Eds.), Shell Process Control Workshop. Butterworths,
for process control. AIChE J. 41 (3), 604–618. pp. 309–321.
Henson, M.A., Badgwell, T.A., 2006. Preface: CPC-VII. Comput. Chem. Eng. 30, 1425. Morari, M., 1988. Process control theory: reflections on the past and goals for the
Holt, B.R., Jerome, N.F., Buck, U., et al., 1987. CONSYD—integrated software for com- next decade. In: Prett, D.M, Garcia, C.E., Ramaker, B.A. (Eds.), The Second Shell
puter aided control system design and analysis. Comput. Chem. Eng. 11 (2), Process Control Workshop. Solutions to the Shell Standard Control Problem.
187–203. Butterworths, pp. 469–488.
Hougen, O.A., 1967. Progress and future in chemical engineering education. Phillips Morari, M., Lee, J.H., 1991. Model predictive control: the good, the bad, and the
Petroleum Lectureship. Oklahoma State University 11 May 1967 https://che. ugly. In: Proceedings of the Fourth International Conference on Chemical Pro-
okstate.edu/sites/default/files/u14/pls1967hougen%5B1%5D.pdf. cess Control (CPC IV). Padre Island, TX, February 17-21 (1991). CACHE, pp. 419–
Hovorka, R, Canonico, V, Chassin, LJ, Haueter, U, Massi-Benedetti, M, Federici, MO, 4 4 4.
Pieber, TR, Schaller, HC, Schaupp, L, Vering, T, Wilinska, ME, 2004. Nonlinear Morshedi, M., 1986. Universal Dynamic Matrix Control (UDMC). In: Morari, M.,
model predictive control of glucose concentration in subjects with type 1 dia- McAvoy, J.T. (Eds.), Proceedings of the Third International Conference on Chem-
betes. Physiol. Meas. 25, 905–920. ical Process Control – CPCIII. CACHE and Elsevier, pp. 547–577.
Hurwitz, A., 1895. On the conditions under which an equation has only roots with More, J.L., Quail, F.J., Bain, J.W., 1945. Application of automatic controllers to heat
negative real parts. Mathematische Annalen. In: Bellman, R., Kalaba, R. (Eds.). exchangers. Ind. Eng. Chem. 37 (10), 912–916.
In: Reprinted in Selected Papers on Mathematical Trends in Control Theory, 46. Munch, R.H., 1951. Instrumentation. Ind. Eng. Chem. 43 (1), 63A–66A.
Dover, pp. 273–284. (1964), pp. 70-82. Murrill, P.W., 1967. Automatic Control of Processes. Int. Textbook Co.
Ivanoff, A., 1933. Theoretical foundations of the automatic regulation of tempera- Muske, K.R., Rawlings, J.B., 1993. Model predictive control with linear models. AIChE
ture. J. Inst. Fuel 7 (33), 117–130 Followed by a discussion on pages 130-138. J. 39 (2), 262–287.
James, H.M., Nichols, N.B., Phillips, R.S., 1947. Theory of servomechanisms. Vol. 25 Nyquist, H., 1932. Regeneration theory. Bell Syst. Tech. J. 11, 126–147.
of MIT Radiation Lab Series. McGraw-Hill. Ogunnaike, B.A., Ray, W.H., 1994. Process Dynamics, Modeling and Control. Oxford
Johnson, E.F., 1951. A pneumatic process analog for instruction and research. Ind. University Press.
Eng. Chem. 43 (12), 2708–2711. Palerm, C.C., 2011. Physiologic insulin delivery with insulin feedback: a control sys-
Johnson, E.F., 1958. Automatic process control. Adv. Chem. Eng. 2, 33–90. tems perspective. Comput. Methods Programs Biomed. 102 (2), 130–137.
Johnson, E.F, 1967. Automatic Process Control. McGraw Hill. Peters, J.C., 1941. Getting the most from automatic control. Ind. Eng. Chem. 33 (9),
Kestenbaum, A., Shinnar, R., Thau, F.E., 1976. Design concepts for process control. 1095–1103.
Ind. Eng. Chem. Process Des. Dev. 15 (1), 2–13. Peters, W.A., 1923. Automatic fractionating columns. Ind. Eng. Chem. 15 (7),
Koppel, L.R., Sullivan, G.R., 1986. Use of IBM’s advanced control system in under- 734–735.
graduate process control education. Chem. Eng. Educ. 70–106. Perkins, J., 2002. Education in process systems engineering: past, present and future.
Learncheme.com, 2019. Educational Resources for Chem. Eng. Univ of Colorado, Comput. Chem. Eng. 26, 283–293.
Boulder Process Control Screencasts http://www.learncheme.com/screencasts/ Perlmutter, D.D., 1965. Introduction to Chemical Process Control. Wiley.
process-controls. Phillips, C.L., Harbor, R.D., 20 0 0. Feedback Control Systems, fourth ed. Prentice Hall.
Li, X., Huang, Z., 2017. An inverted classroom approach to educate MATLAB in chem- Prett, D.M., Gillette, R.D., 1980. Optimization and constrained multivariable control
ical process control. Educ. Chem. Eng. 19, 1–12. of a catalytic cracking unit. In: Proceedings of the Joint Automatic Control Con-
Lee, W., Weekman, W.W., 1976. Advanced control practice in the chemical process ference. San Francisco paper WP5-C.
industry: a view from industry. AIChE J. 22, 27–38. Prett, D.M, Morari, M., 1987. The Shell Process Control Workshop. Process Control
Lees, S., Hougen, J.O., 1956. Pulse testing a model heat exchange process. Ind. Eng. Research: Industrial and Academic Perspectives. Butterworths.
Chem. 48 (6), 1064–1068. Prett, D.M, Garcia, C.E., Ramaker, B.A., 1989. The Second Shell Process Control Work-
Luo, Y., Westmoreland, P.R., 2015. Chemical engineering academia-industry align- shop. Solutions to the Shell Standard Control Problem. Butterworths.
ment: expectations about new graduates. 1 November 2015. AIChE accessed Qin, S.J., Badgwell, T.A., 1997. An overview of industrial model predictive control
13 January 2018. https://www.aiche.org/sites/default/files/docs/conferences/ technology. In: Proceedings CPC-V. Lake Tahoe, CA, pp. 232–256.
2015che_academicindustryalignmentstudy.compressed.pdf. Ray, W.H., 1983. Multivariable process control – a survey. Comp. Chem. Engng. 7 (4),
Luyben, W.L., 1973. Process Modeling. Simulation and Control for Chemical Engi- 367–394.
neers. McGraw Hill. Ricker, N.L., 1985. Use of quadratic programming for constrained internal model
Luyben, W.L., 1990. Process Modeling Simulation and Control for Chemical Engi- control. Ind. Eng. Chem. Process Des. Dev. 24, 925–936.
neers, second ed. McGraw-Hill, New York. Ricker, N.L., 1991. Model-predictive control: state of the art. In: Proceedings of the
Luyben, M.L., Luyben, W.L., 1997. Essentials of Process Control. McGraw-Hill, New Fourth International Conference on Chemical Process Control (CPC IV). Padre
York. Island, TX, February 17-21 (1991). CACHE, pp. 271–296.
Luyben, W.L., 2007. Chemical Reactor Design and Control. Wiley. Riddle, F.H., Royal, H.F., 1930. Process control in continuous production. Ind. Eng.
Luyben, W.L., 2013. Distillation Design and Control using ASPEN Simulation, second Chem. 22 (1), 14–20.
ed. Wiley. Ridenour, L.N., 1952. The role of the computer. Sci. Am. 187 (3), 116–131.
McKetta, J.J., Schechter, R.S., 1961. A flexible undergraduate chemical engineering Riggs, J.B., 1999. Chemical Process Control. Ferret Publishing, Lubbock, TX.
program. J. Chem. Educ. 38 (5), 260–262. Riggs, J.B., 2002. Chemical Process Control, second ed. Ferret Publishing, Lubbock,
McMahon, J.B., 1937. Control of liquid level in vessels under pressure. Ind Eng. TX.
Chem. 29, 1219–1224. Riggs, J.B., Karim, M.N., 2007. Chemical and Bio-Process Control, 3rd ed. Ferret Pub-
Marlin, T.E., 1995. Process Control: Designing Processes and Control Systems for Dy- lishing, Lubbock, Texas.
namic Performance. McGraw Hill, New York. Riggs, J.B., Karim, M.N., 2016. Chemical and Bio-Process Control, fourth ed. Ferret
Marlin, T.F., 20 0 0. Process Control: Designing Processes and Control Systems for Dy- Publishing, Lubbock, Texas.
namic Performance, second ed. McGraw-Hill, New York. Romagnoli, J.A., Palazoglu, A., 2006. Introduction to Process Control. Taylor and
Marlin, T.F., 2017. Flipping the chemical engineering process control class with e- Francis.
lessons. 2017 ASEE Ann. Mtg., Columbus, OH, June 2017 http://cache.org/files/ Romagnoli, J.A., Palazoglu, A., 2012. Introduction to Process Control, second ed. CRC
sum17- flipping- class- asee.pdf. Press.
556 B.W. Bequette / Computers and Chemical Engineering 128 (2019) 538–556

Seames, W.S., 2017a. Updating the process controls and dynamics course for the Truxal, J.G., 1955. Automatic Control System Synthesis. McGraw-Hill.
21st century. Presented at the 2017 AIChE Ann. Mtg., session 145d. Minneapo- Tustin, A., 1951. Automatic Control. Nature 168 (4271), 404–406.
lis, MN (Oct. 2017) Abtract (accessed 7 February 2018). https://www.aiche.org/ Tyner, M., May, F.P., 1968. Process Engineering Control. Ronald Press.
conferences/aiche- annual- meeting/2017/proceeding/paper/145d- updating- Van Antwerpen, F.J., 1942. The automatic control problem. Ind. Eng. Chem. 34 (4),
process-controls-and-dynamics-course-21st-century. 387–391.
Seames, W.S., 2017b. Designing Controls for the Process Industries. CRC Press. Van Antwerpen, F.J., 1980. The origins of chemical engineering. Chapter 1 in history
Seborg, D.E., 1980. A survey of process control education in the United States and of chemical engineering. Advances in Chemistry (Furter, ed.). American Chemi-
Canada. Chem. Eng. Educ. 14 (1), 42–44. cal Society (1980).
Seborg, D.E., Edgar, T.F., Mellichamp, D.A., Doyle, F.J., 2016. Process Dynamics and Vannah, W.E., Slater, L.E., 1956. Process control enters a new era. Ind. Eng. Chem. 48
Control, fourth ed. Wiley, New York. (6), 10 0 0–10 01.
Seider, W.L, Lewin, D.R., Seader, J.D., Widagdo, S., Gani, R., Ng, K.M., 2016. Prod- Velton, H.J., 1937. Automatic mixing and proportioning of gases and liquids. Ind.
uct and process design principles: synthesis. Analysis and Evaluation, fourth ed. Eng. Chem. 29 (11), 1214–2128.
Wiley. Vinson, D.R., 2006. Air separation control technology. Comput. Chem. Eng. 30,
Shinskey, F.G., 1967. Process Control Systems. McGraw-Hill. 1436–1446.
Shinskey, F.G., 2002. Process control: as taught vs. as practiced. Ind. Eng. Chem. Res. Vogel, E.F., Downs, J.J., 2002. Industrial experience with state space model predic-
41 (16), 3745–3750. tive control. In: Rawlings, J.B., Ogunnaike, B.A., Eaton, J.W. (Eds.), Proceedings of
Seinfeld, J.H., 1982. Control strategies for process control and estimation. In: Seborg, the Sixth International Conference on Chemical Process Control. CACHE/AIChE,
Edgar (Eds.), Proceedings of CPC 2, pp. 619–621. pp. 438–442.
Silverstein, D.L., Vigeant, M.A., Staehle, M., 2016. How we teach process control: Wankat, P.C. and Peppas, N.A. 2011. 100 years of chemical engineering at Purdue
2015 survey results. Presented at 2016 ASEE Ann. Conf. & Expo. New Orleans, University, 1911–2011. (accessed 1 January 2019). https://engineering.purdue.
LA (accessed 8 Feb 18) doi:10.18260/p.25495. edu/ChE/aboutus/publications/100_years.
Smith, C.A., Corripio, A.B., 1985. Principles and Practice of Automatic Process Con- Weber, T.W., 1973. An Introduction to Process Dynamics and Control. Wiley.
trol. Wiley, New York. Wilinska, M.W., Chassin, L.J., Schaller, H.C., Schaupp, L., Pieber, T.R., Hovorka, R.,
Smith, C.A., Corripio, A.B., 1997. Principles and Practice of Automatic Process Control, 2005. Insulin kinetics in type-1 diabetes: continuous and bolus delivery of rapid
second ed. Wiley, New York. acting insulin. IEEE Trans. Biomed. Eng. 52 (1), 3–12.
Smith, C.A., Corripio, A.B., 2005. Principles and Practice of Automatic Process Con- Wilson, E.D., 1951. Introduction – instrumentation. Ind. Eng. Chem. 43 (12), 2694.
trol, third ed. Wiley, New York. Wilson, J., 2002. The development of the studio classroom. In: Goodan, P.S. (Ed.),
Smith, E.S., 1936. Automatic regulators, their theory and applications. ASME Trans. Technology-Enhanced Learning: Opportunities for Change. Lawrence Erlbaum
58 (4), 291–303. Assoc, Mahwah, NJ Preprint: (accessed 7 Feb 18). http://www.jackmwilson.net/
Stephanopoulos, G., 1983. Synthesis of control systems for chemical plants: a chal- ArticlesTalks/Studio20 0 0.pdf.
lenge for creativity. Comput. Chem. Eng. 7 (4), 331–365. Williams, D.C., Tarrar, A.R., 1986. Undergraduate process control. Chem. Eng. Educ.
Stephanopoulos, G., 1984. Chemical Process Control. Prentice Hall, Englewood Cliffs, 20 (2), 74–77.
NJ. Williams, T.J., Harnett, R.T, Rose, A., 1956. Automatic control in continuous distilla-
Stice, J.E., Swanson, B.S., 1963. The use of analog computers in teaching process con- tion. Ind. Eng. Chem. 48 (6), 1008–1019.
trol. Chem. Eng. Educ. (March) 23–26. Williams, T.J., 1958. Process control and automation. Ind. Eng. Chem. 50 (3),
Stout, L.E., Ceaglske, N.H., 1945. Automatic control laboratory installed at Washing- 520–524.
ton University. Chem. Eng. 23 (14), 1245–1247. Young, A.J., 1955. An Introduction to Process Control System Design. Longman, New
Stout, T.M, Williams, T.J., 1995. Pioneering work in the field of computer process York.
control. IEEE Ann. Hist. Comput. 17 (1), 6–18. Young, R.E., Bartusiak, R.D., Fontaine, R.W., 2002. Evolution of an industrial nonlin-
Svrcek, W.Y., Mahoney, D.P., Young, B.R., 20 0 0. A Real-Time Approach to Process ear model predictive controller. In: Rawlings, J.B., Ogunnaike, B.A., Eaton, J.W.
Control. Wiley, Chichester. (Eds.), Proceedings of the Sixth International Conference on Chemical Process
Svrcek, W.Y., Mahoney, D.P., Young, B.R., 2014. A Real-Time Approach to Process Con- Control. CACHE/AIChE, pp. 342–351.
trol, third ed. Wiley, Chichester. Zafiriou, E., 1991. On the effect of tuning parameters and constraints on the robust-
Swan, C.J., 1928. Automatic control through temperature or pressure. Ind. Eng. ness of model predictive controllers. In: Proceedings of the Fourth International
Chem. 20 (11), 1152–1155. Conference on Chemical Process Control (CPC IV). Padre Island, TX. February
Thomas, C.A., 1937. Symposium on automatic control. Ind. Eng. Chem. 29 (11), 1209. 17-21 (1991) CACHE, pp. 362–393.
Thompson, G.W., 1917. The progress of american chemical engineering. Ind. Eng. Ziegler, J.G., Nichols, N.B., 1942. Optimum settings for automatic controllers. Trans.
Chem. (November) 1007 Page. ASME. 64, 759–765.

Potrebbero piacerti anche