Automatic control, particularly the application of feedback, has been fundamental to the

development of automation. Its origins lie in the level control, water clocks, and
pneumatics/hydraulics of the ancient world. From the 17th century onwards, systems were
designed for temperature control, the mechanical control of mills, and the regulation of steam
engines. During the 19th century it became increasingly clear that feedback systems were prone to
instability. A stability criterion was derived independently towards the end of the century by Routh
in England and Hurwitz in Switzerland. The 19th century, too, saw the development of
servomechanisms, first for ship steering and later for stabilization and autopilots. The invention of
aircraft added (literally) a new dimension to the problem. Minorsky’s theoretical analysis of ship
control in the 1920s clarified the nature of three-term control, also being used for process
applications by the 1930s. Based on servo and communications engineering developments of the
1930s, and driven by the need for high-performance gun control systems, the coherent body of
theory known as classical control emerged during and just after WWII in the US, UK and
elsewhere, as did cybernetics ideas. Meanwhile, an alternative approach to dynamic modelling
had been developed in the USSR based on the approaches of Poincaré and Lyapunov.

Information was gradually disseminated, and state-space or modern control techniques, fuelled by
Cold War demands for missile control systems, rapidly developed in both East and West. The
immediate post-war period was marked by great claims for automation, but also great fears, while
the digital computer opened new possibilities for automatic control.

4.1 Antiquity and the Early Modern Period

Feedback control can be said to have originated with the float valve regulators of the Hellenic and
Arab worlds [4.1]. They were used by the Greeks and Arabs to control such devices as water
clocks, oil lamps and wine dispensers, as well as the level of water in tanks. The precise
construction of such systems is still not

entirely clear, since the descriptions in the original Greek or Arabic are often vague, and lack
illustrations. The best known Greek names are Ktsebios and Philon (third century BC) and Heron
(first century AD) who were active in the eastern Mediterranean (Alexandria, Byzantium). The
water clock tradition was continued in the Arab world as described in books by writers such as Al-
Jazari (1203) and Ibn al-Sa-ati (1206), greatly influenced by the anonymous Arab author known as
Pseudo-Archimedes of the ninth–tenth century AD, who makes specific reference to the Greek
work of Heron and Philon. Float regulators in the tradition of Heron were also constructed by the
three brothers Banu Musa in Baghdad in the ninth century AD. The float valve level regulator does
not appear to have spread to medieval Europe, even though translations existed of some of the
classical texts by the above writers. It seems rather to have been reinvented during the industrial
revolution, appearing in England, for example, in the 18th century. The first independent
European feedback system was the temperature regulator of Cornelius Drebbel (1572–1633).
Drebbel spent most of his professional career at the courts of James I and Charles I of England and
Rudolf II in Prague. Drebbel himselfleftnowrittenrecords,butanumberofcontemporary descriptions
survive of his invention. Essentially an alcohol (or other) thermometer was used to operate
avalvecontrollingafurnaceflue,andhencethetemperature of an enclosure [4.2]. The device included
screws to alter what we would now call the set point. If level and temperature regulation were
two of the major precursors of modern control systems, then a number of devices designed for
use with windmills pointed the way towards more sophisticated devices. During the 18th century
the mill fantail was developed both to keep the mill sails directed into the wind and to
automaticallyvarytheangleofattack,soastoavoidexcessive speeds in high winds. Another important
device was the lift-tenter. Millstones have a tendency to separate as the speed of rotation
increases, thus impairing thequalityofflour.Anumberoftechniquesweredeveloped to sense the
speed and hence produce a restoring force to press the millstones closer together. Of these,
perhaps the most important were Thomas Mead’s devices [4.3], which used a centrifugal
pendulum to sense the speed and – in some applications – also to
providefeedback,hencepointingthewaytothecentrifugal governor. The first steam engines were
the reciprocating engines developed for driving water pumps; James Watt’s rotary engines were
sold only from the early 1780s. But it took until the end of the decade for the centrifugal governor
to be applied to the machine, following a visit by Watt’s collaborator, Matthew Boulton, to the
Albion Mill in London where he saw a lift-tenter in action under the control of a centrifugal
pendulum. Boulton and Watt did not attempt to patent the device (which, as noted above, had
essentially alread been patented by Mead) but they did try unsuccessfully to keep it secret. It was
first copied in 1793 and spread throughout England over the next ten years [4.4

4.2 Stability Analysis in the 19th Century

With the spread of the centrifugal governor in the early 19th century a number of major problems
became apparent. First, because of the absence of integral action, the governor could not remove
offset: in the terminology of the time it could not regulate but only moderate. Second, its response
to a change in load was slow. And thirdly, (nonlinear) frictional forces in the mechanism could lead
to hunting (limit cycling). A number of attempts were made to overcome these problems: for
example, the Siemens chronometric governor effectively introduced integral action through
differential gearing, as well as mechanical amplification. Other approaches to the design of an
isochronous governor (one with no offset) were based on ingenious mechanical constructions, but
often encountered problems of stability. Nevertheless the 19th century saw steady progress in the
development of practical governors for steam engines and hydraulic turbines, including spring-
loaded designs (which could be made much smaller, and operate at higher speeds) and relay
(indirect-acting) governors [4.6]. By the end of the century governors of various sizes and designs
were available for effective regulation in a range of applications, and a number of graphical
techniques existed for steady-state design. Few engineers were concerned with the analysis of the
dynamics of a feedback system. In parallel with the developments in the engineering sector a
number of eminent British scientists became interested in governors in order to keep a telescope
directed at a particular star as the Earth rotated. A formal analysis of the dynamics of such a
system by George Bidell Airy, Astronomer Royal, in 1840 [4.7] clearly demonstrated the propensity
of such a feedback system to become unstable. In 1868 James Clerk Maxwell analyzed governor
dynamics, prompted by an electrical experiment in which the speed of rotation of a coil had to be
held constant. His resulting classic paper On governors [4.8] was received by the Royal Society on
20 February. Maxwell derived a third-order linear model and the correct conditions for stability in
terms of the coefficients of the characteristic equation. Unable to derive a solution for higher-
order models, he expressed the hope that the question would gain the attention of
mathematicians. In 1875 the subject for the Cambridge University Adams Prize in mathematics
was set as The criterion of dynamical stability. One of the examiners was Maxwell himself
(prizewinner in 1857) and the 1875 prize (awarded in 1877) was won by Edward James Routh.
Routh had been interested in dynamical stability for several years, and had already obtained a
solution for a fifth-order system. In the published paper [4.9] we find derived the Routh version of
the renowned Routh–Hurwitzstability criterion. Related, independent work was being carried out
in continental Europe at about the same time [4.5]. A summary of the work of I.A. Vyshnegradskii
in St. Petersburg appeared in the French Comptes Rendus de l’Academie des Sciences in 1876,
with the full version appearing in Russian and German in 1877, and in French in 1878/79.
Vyshnegradskii (generally transliterated at the time as Wischnegradski) transformed a third-order
differential equation model of a steam engine with governor into a standard form

ϕ3+xϕ2+yϕ+1=0

where x and y became known as the Vyshnegradskii parameters. He then showed that a point in
the x–y plane defined the natureof the system transient response. Figure 4.3 shows the diagram
drawn by Vyshnegradskii, to which typical pole constellations for various regions in the plane have
been added. In 1893 Aurel Boreslav Stodola at the Federal Polytechnic, Zurich, studied the
dynamicsofahigh-pressure hydraulic turbine, and used Vyshnegradskii’s method to assess the
stability of a third-order model. A more realistic model, however, was seventh-order, and Stodola
posedthegeneralproblemtoamathematiciancolleague Adolf Hurwitz, who very soon came up with
his version of the Routh–Hurwitz criterion [4.10]. The two versions were shown to be identical by
Enrico Bompiani in 1911 [4.11]. At the beginning of the 20th century the first general textbooks on
the regulation of prime movers appeared in a number of European languages [4.12,13]. One of the
most influential was Tolle’s Regelung der Kraftmaschine, which went through three editions
between 1905 and1922[4.14.The later editions in cluded the Hurwitz stability criterion.

Ship, Aircraft and Industrial Control Before WWII

The first ship steering engines incorporating feedback

appearedinthemiddleofthe19thcentury.In1873Jean Joseph Léon Farcot published a book on
servomotors in which he not only described the various designs developed in the family firm, but
also gave an account of the general principles of position control. Another important maritime
application of feedback control was in gun turret operation, and hydraulics were also extensively
developed for transmission systems. Torpedoes, too, used increasingly sophisticated feedback
systems for depth control – including, by the end of the century – gyroscopic action. During the
first decades of the 20th century gyroscopes were increasingly used for ship stabilization and
autopilots.ElmerSperrypioneeredthe activestabilizer, the gyrocompass, and the gyroscope
autopilot, filing various patents over the period 1907–1914. Sperry’s autopilot was a sophisticated
device: an inner loop controlled an electric motor which operated the steering engine, while an
outer loop used a gyrocompass to sense the heading. Sperry also designed an anticipator to
replicate the way in which an experienced helmsman would meet the helm (to prevent
oversteering); the anticipator was, in fact, a type of adaptive control [4.16]. Sperry and his son
Lawrence also designed aircraft autostabilizers over the same period, with the added complexity
of three-dimensional control. Bennett describes the system used in an acclaimed demonstration in
Paris in 1914 [4.17]

Significant technological advances in both ship and aircraft stabilization took place over the next
two decades, and by the mid 1930s a number of airlines were using Sperry autopilots for long-
distance flights. However, apart from the stability analyses discussed in Sect.4.2 above, which
were not widely known at this time, there was little theoretical investigation of such feedback
control systems. One of the earliest significant studies was carried out by Nicholas Minorsky,
published in 1922 [4.19]. Minorsky was born in Russia in 1885 (his knowledge of Russian proved to
be important to the West much later). During service with the Russian Navy he studied the ship
steering problem and, following his emigration to the USA in 1918, he made the first theoretical
analysis of automatic ship steering. This study clearly identified the way that control action should
be employed: although Minorsky did not use the terms in the modern sense, he recommended an
appropriate combination of proportional, derivative and integral action. Minorsky’s work was not
widely disseminated, however. Although he gave a good theoretical basis for closed loop control,
he was writing in an age of heroic invention, when intuition and practical experience were much
more important for engineering practice than the oretical analysis.
Importanttechnologicaldevelopmentswerealsobeing made in other sectors during the first few
decades of the 20th century, although again there was little theoretical underpinning. The electric
power industry brought demands for voltage and frequency regulation; many processes using
driven rollers required accurate speed control; and considerable work was carried out in a number
of countries on systems for the accurate pointing of guns for naval and anti-aircraft gunnery. In the
process industries, measuring instruments and pneumatic controllers of increasing sophistication
were developed. Mason’s Stabilog, patented in 1933, included integral as well as
proportionalaction, andbythe end of the decade three-term controllers were available
thatalsoincludedpreact orderivativecontrol. Theoretical progress was
slow,however,untiltheadvancesmade in electronics and telecommunications in the 1920s and 30s
were translated into the control field during WWII.

Electronics, Feedback and Mathematical Analysis

The rapid spread of telegraphy and then telephony from the mid 19th century onwards prompted
a great deal of theoretical investigation into the behaviour of electric circuits. Oliver Heaviside
published papers on his operational calculus over a number of years from 1888 onwards [4.20],
but although his techniques produced valid results for the transient response of electrical
networks, he was fiercely criticized by contemporary mathematicians for his lack of rigour, and
ultimately he was blackballed by the establishment. It was not until the second decade of the 20th
century that Bromwich, Carson and others made the link between Heaviside’s operational calculus
and Fourier methods, and thus proved the validity of Heaviside’s techniques [4.21]. The first three
decades of the 20th century saw important analyses of circuit and filter design, particularly in the
USA and Germany. Harry Nyquist and Karl Küpfmüller were two of the first to consider the
problem of the maximum transmission rate of telegraph signals, as well as the notion of
information in telecommunications, and both went on to analyze the general stability problem of a
feedback circuit [4.22]. In 1928 Küpfmüller analyzed the dynamics of an automatic gain control
electronic circuit using feedback. He appreciated the dynamics of the feedback system, but his
integral equation approach resulted only in a approximations and design
diagrams,ratherthanarigorous stability criterion. At about the same time in the USA, Harold Black
was designing feedback amplifiers for transcontinental telephony. In a famous epiphany on the
Hudson River ferry in August 1927 he realized that negative feedback could reduce distortion at
the cost of reducing overall gain. Black passed on the problem of the stability of such a feedback
loop to his Bell Labs colleagueHarry Nyquist, who published his celebrated frequency-domain
encirclement criterion in 1932 [4.23]. Nyquist demonstrated, using results derived by Cauchy, that
the key to stability is whether or not the open loop frequency response locus in the complex plane
encircles (in Nyquist’s original convention) the point 1+i0. One of the great advantages of this
approach is that no analytical form of the open loop frequency response is required: a set of
measured data points can be plotted without the need for a mathematical model. Another
advantage is that, unlike the Routh–Hurwitz criterion, an assessment of the transient response can
be made directly from the Nyquist plot in terms of gain and phase margins (how close the locus
approaches the critical point). Black’s 1934 paper reporting his contribution to the development of
the negative feedback amplifier included what was to become the standard closed-loop analysis in
the frequency domain [4.24]. The third key contributor to the analysis of feedback in electronic
systems at Bell Labs was Hendrik Bode who worked on equalizers from the mid 1930s, and who
demonstrated that attenuation and phase shift were related in any realizable circuit [4.25]. The
dream of telephone engineers to build circuits with fast cutoff and low phase shift was indeed only
a dream. It was Bode who introduced the notions of gain and phase
margins,andredrewtheNyquistplotinitsnowconventional form with the critical point at −1+i0. He
also introduced the famous straight-line approximations to frequency response curves of linear
systems plotted on log–log axes. Bode presented his methods in a classic text published
immediately after the war [4.26]. If the work of the communications engineers was one major
precursor of classical control, then the other was the development of high-performance servos in
the 1930s. The need for such servos was generated by the increasing use of analogue simulators,
such as network analysers for the electrical power industry and differential analysers for a wide
range of problems. By the early 1930s six-integrator differential analysers were in operation at
various locations in the USA and the UK. A major centre of innovation was MIT, where Vannevar
Bush, Norbert Wiener and Harold Hazen had all contributed to design. In 1934 Hazen summarized
the developmentsofthepreviousyearsinThetheoryofservomechanisms [4.27]. He adopted
normalized curves, and parameters such as time constant and damping factor, to characterize
servo-response, but he did not given any stability analysis: although he appears to have been
aware of Nyquists’s work, he (like almost all his contemporaries) does not appear to have
appreciated the close relationship between a feedback servomechanism and a feedback amplifier.
The1930sAmericanworkgraduallybecameknown elsewhere.There is ample evidence from prewar
USSR, Germany and France that, for example, Nyquist’s resultswere known– if
notwidelydisseminated.In1940, for example, Leonhard published a book on automatic control in
which he introduced the inverse Nyquist plot [4.28], and in the same year a conference was held
inMoscowduringwhichanumberofWesternresultsin automatic control were presented and
discussed [4.29]. Also in Russia, a great deal of work was being carried out on nonlinear dynamics,
using an approach developed from the methods of Poincaré and Lyapunov at the turn of the
century [4.30]. Such approaches, however, were not widely known outside Russia until after the
war.