What Is a Hybrid Computer and Why
What is a Hybrid Computer and Why
The use of the word hybrid in this context is to link the two
most widely known forms of electronic computer, namely the digital
computer of which most people are most aware, and which essentially
Works in numbers in a particular form, and the analogue computer.
The analogue computer derives its name from the word analogy; in this
case an exact electrical analogy, in voltages, is constructed to
represent a working system, be this a mechanical plant, aeroplane in
flight, chemical production plant, or whatever.
Numbers in the digital computer are represented in what is called
binary form. The familiar form of calculation used is, of course, the
decimal form, i.e. numbers expressed in multiples of ten. We, in Britain,
are currently in the throes of converting from a duodecimal to a decimal
system of counting money. When moving into the workings of a digital
computer it is necessary to convert to a system of counting based on 2
rather than 10.
The representation of numbers in binary form only requires the
statement of two quantities, O and 1
since decimal 2 = binary ‘10
" 3 = binary 11
" 4 = binary 100 etc.
The state of O or 1 can therefore mean electrically whether a
voltage is present or not, i.e. whether a switch is made. The transistors
of which one is otherwise familiar, perform the switching functions,in
the very early machines actual switches, electromagnetically driven-relays
were used.
The form of calculation that can be performed by a digital computer
is essentially arithmetical, i.e. addition, subtraction, etc., but what
are termed logical operations based on the algebra of the binary form are
inc luded.
The analogue computer, however, is quite different in that it solves
differential equations, rather than arithmetic and algebraic equations,
although these are not excluded and do, in fact, form part of the total
analogy. The velocity or speed of a vehicle is the first differential
of its position along a given path or road, while its acceleration is the
second differential. The equations presented to the analogue computer
are in this form, i.e. displacement, velocity and acceleration or their
equivalents.
The complete set of equations which represent, say, a motor car
travelling along a road, to include the effect of the drivers action,
the bends in the road, the road roughness, the movement of the car
suspension etc, are solved simultaneously, so that it can be said that
the true history in time of the movement of the car can be represented
by the voltage variations in time calculated by the analogue computer.
When considering the hybrid computer, the first requirement is to
communicate the two dissimilar forms of data, i.e. numbers in binary
form and voltages, this is done through analogue to digital converter -
which successively divide the voltage present by factors of 2, and
digital to analogue converters which reconstruct voltages from the binary
value by multiplying a given basic voltage by factors of 2.
The control of the two computers working together form the other
major task.
With the achievement of a hybrid computer the value of the total
system is many times greater than the sum of the two component parts, as
can be easily imagined from the scope of problems that are now possible.
Facilities - Vital Statistics
The group's facilities are essentially grouped into
(a) that for basic experiment and research.
(b) that associated with computing and the application of computers.
(c) that for instrumentation, which in the main bridges the first
two in that instrumentation of experiments and processes is
aimed at producing computer intelligible data.
The facilities for experiment are centred in the main Laboratory which
provides a comprehensive selection of electronic instruments and test
gear and workshop facilities.
The computing facilities are:
(i) The access to the large digital computer, Titan, sited at
the Mathematical Laboratory. Two remote teleprinter consoles
are provided and are situated along with the other comparable
machines for paper tape preparation.
(ii) The hybrid computer. This comprises an I.C.L. 4130 digital
computer linked to PACE 231R V analogue computers.
There are in fact four such analogue computers, all of which
can, however, be linked together in the solution of a large
problem. They are physically divided into two groups of
two consoles. Those remote from the digital cannot be
controlled except indirectly from the 4130 digital. However,
the group of analogues directly adjacent to the digital
computer are fully interfaced to the digital in terms of
data, logic and control and addressing.
The group also possess two major data logging facilities for the
collection of data from industrial plant. One set is designed to operate
on slow processes such as encountered in chemical engineering and in
consequence, the system has been designed to collect the data on to punched
paper tape for subsequent analysis on the 4130 digital, and in such a
Way that the logger may be left to collect the data automatically over
extended periods of time amounting to months, when the only attention
required is to replace the used reels of paper with new ones.
A second set is in the process of manufacture and is intended for
the collection of data on fast processes ~ usually mechanical, such
as the production of steel sheet in a rolling mill. There it is essential
to use the much faster recording rate possible with the magnetic tape
machine, and to record in a form suitable for replay on the 4130 digital
magnetic tape decks. This fast data logger uses a small digital computer
as the controlling device.
Vital Statistics of the Hybrid Computing system
I.C.L. 4130 digital
Store size:: 32,768 words
24 bit words
2usec store access time.
4 Magnetic tape handlers: 33/12 Kch/sec.
Line printer: 300 lines/minute. 120 ch frame
Digital plotter: 12" paper width
200 steps/sec.
0.005 inches per step.
2 tape readers: 1000ch/sec.
2 tape puncheg: 110ch/sec.
1 teletype controller: 10ch/sec.
Central processor fitted with autonomous transfer.
PACE 231R Vs together:
60 Integrator amplifiers
112. other amplifiers
18 multipliers
80 servo-set potentiometers
40 potentiometers
20 function generators
20 comparators
12 limiters
16 track store units
24 AND gates
20 relay drivers
16 D/A switches
PACE 231R and 221R, together:
48 Integrator amplifiers
62 other amplifiers
21 multipliers
6 squarers
160 potentiometers
12 function generators
10 comparators
Interface to 4130
48 analogue to digital converters
64 digital to analogue converters
16 logic trunks
13. interrupt trunks
Full control and addressing.
"So you think you've got problems!" or
"How to use a hybrid computer"
To explain the use of the facilities available in this Department
consider that the problem concerns a hypothetical chemical plant, currently
in use and earning its keep. The plant management will be aware that the
plant produces scrap when it starts up and is shut down as the vroduct
does not meet the minimum quality control requirements. The management
also is aware that during a typical day further product will also be
scrapped because of, for example, hopper levels changing or feed water
flows altering to less than optimum levels etc. These may be small in
themselves and probably the scrap product has not even been attributed
to these disturbances. (If the product takes fifteen minutes to make
then a disturbance at the start of manufacture may not be detected until
it reaches Quality Control some five minutes or more after it has left
the production line - 20 minutes after the causative disturbance.)
The disturbing factors may themselves be very small but two or more
values being less than optimum may be very significant in the performance
of the plant.
This is where the Control Engineer steps in. A preliminary study
of the plant will reveal where the major factors affecting performance
may be measured and he will initiate the gathering of data from the
plant and recording on to punched paper tape or magnetic tape.
He will synthesise a mathematical model of the plant based on
the chemistry of the process and on the data and physical information
he has gathered. He will make intelligent guesses at the unknown
coefficientsof his model and will estimate the most favourable values to
fit the data using an iterative process called hill climbing. The
resultant represents an accurate mathematical model of the plant. This
model is next converted into a set of first order differential equations
and by matrix manipulation and transformation of the coefficients a reduced
model is obtained which contains all the dominant disturbing factors
which grossly affect the performance of the real plant. This model may
be patched on the analogue part of the hybrid system for the study of how
the plant variables affect one another and change with time. The
recorded disturbances and data may be "played" to the model through the
digital computer to confirm that it performs like the real plant.
Optimising control strategies may now be tried out on the model to
evolve the ideal. The Control Engineer is aided in his task by being able
to study the plant on totally different time scales from that of the real
time operation - a year's performance may be simulated in a matter of
hours.
Production on the chemical plant will not have been affected whilst
the plant was studied and yet, having identified the sensitive areas,
the Control Engineer is able to advise where it will be necessary to
regulate and control the chemical plant to achieve greater efficiency
and a better quality product,
Postgraduate Research work in Control and Systems Engineering
leading to the M.Sc. and Ph.D. Degrees
A candidate for a higher degree is required to devote at least six
terms to full time research before proceeding to the M.Sc. degree and at
least nine terms before proceeding to the Ph.D. degree. For three of
these terms a candidate must be resident in Cambridge and, if not already
a Cambridge graduate, a further three terms residence in Cambridge is
required. Provided the residence qualification has been satisfied,
permission may be granted for the candidate to work away from Cambridge
for all but one year, provided he is fully employed during that period
on his approved subject of research. Each candidate is examined by means
of a dissertation, written at the end of his research period, followed
by an oral examination.
A research student will not in the first instance be registered for
a degree. After two terms of residence he will be required to write a
brief report on his chosen research covering a literature survey, such
work as he has carried out in his first two terms and a description of
the further work he proposes to do. The Head of the Research Group and
his research supervisor will discuss the report with him and it will later
be decided whether he shall be registered for the M.Sc. degree or Ph.D.
or whether he should be advised that it is not in his interest to proceed
further with University research work.
The programme of research in the field of control and systems
engineering being carried on in the Engineering Department is aimed at
the development of generally applicable methods of analysis and optimisation
of non-linear control systems and filters with random inputs, it includes
the following areas of work:
Theoretical and experimental investigation of methods of optimisation
of non-linear control systems and filters.
The use of functionals and the application of methods of statistical
mechanics to control system analysis and synthesis.
Methods of parameter estimation and the development of mathematical
models of complex systems.
General methods for the design of optimal and insensitive multivariable
control systems.
Theoretical and experimental investigation of methods of non-linear
control systems and filters.
Analysis and optimisation of control systems with randomly varying
loads.
Recording and analysis of random processes.
Application of analogue, digital and hybrid computers to the control
of complex industrial plant, aircraft and economic systems.
Use of analogue, digital and hybrid computers in control system
design and the preparation of universally applicable design procedures.
Analytic techniques associated with hybrid computation.
Important classes of control systems such as on/off and saturating
systems, multi-loop adaptive and predictive control systems.
Analysis and optimisation of extremum regulators.
Analytic and programming techniques associated with hybrid computation.
Self-adjusting non-linear filters and plant models.
Development of sub-optimal control systems, and the determination
of their performance in relation to optimal systems.
Some current Research Topics
The Application of Control Theory to Macro-Economic Models
undertaken by D.A. Livesey
A macro-economic model of the U.K. economy has now been built which
consists of sixteen nonlinear differential equations, with five controls and
two exogenous indices. The model, based upon a set of six account ing
equations, explains the behaviour of the main aggregated economic variables.
The model's parameters have been obtained by fitting, for quarterly data,
estimated values of the variables to actual variable values over the period
1957-66. The equations have been fitted both as single equation models using
ordinary least squares and as a complete set of simultaneous equations using
a hill-climbing technique.
The model has been programmed as a computer game on the university TITAN
computer. The player can set the taxation rates, government expenditure and
the bank rate for several years. The resultant behaviour of the economy is
determined by the model. The output from the game is in the form of the
standard accounting tables published by the government. At present the game
is under evaluation for use in undergraduate teaching in the Faculty of
Economics.
Dynamic response tests have been carried out to evaluate the structural
accuracy of the model. One or two weaknesses are at present being corrected
and the project will move into the final stage shortly. The model has been
constructed so that control algorithms can be applied. It is expected to
be able to demonstrate the improvement, measured according to some criterion,
which results in comparison with the actual events over the same period.
An Automatic Landing Control System for a V.T.O.L. Aircraft
undertaken by C.R. Guy
The object of this research project is to design, and mechanise using
the hybrid computing facility, an automatic landing control system for a
vertical take-off/landing (V.T.O.L.) aircraft.
For such a manoeuvre to be accomplished the aircraft is required to
follow a given flight path. This can be broken down into two separate
parts thus:
1. the approach region.
2. the touch-down region.
As both the dynamics of the aircraft and the form of the controller are
different for the two flight path regions, the project can be divided into
two halves.
A model of an aircraft has been set up on two analogue computers.
This model is based on the SHORT S.C.1. V.T.O.L. research aircraft, and
represents motion of this aircraft in six degrees of freedom for low speed
flight conditions (i.e. flight in the touch~down region).
A touch-down controller is at present being developed for use with
the model. As this problem is characterised by the necessity of satisfying
multiple performance requirements and constraints, the controller design
lends itself to the use of optimization theory. The system being developed
uses optimization theory to establish the form of the time varying feedback
gains. When the design of the controller is finalised, the complete system
will be set up on the hybrid computing facility.
Future work will include developing the model for the transition and
wing-borne regions of the flight envelope, and mechanising a complete landing
system.
Study of a Nitric Acid plant
undertaken by V. Sobotka.
A model of an absorption column has been constructed. Such a plant
1s a part of factories which produce nitric acid. The main demands on
these plants are to keep the concentration of nitric acid constant, to
produce as much acid as possible and to keep the concentration of waste
gas very small.
The whole model of the plant has two different parts. It is firstly
a model of the connecting pipes and valves among the towers of the column
and secondly a model of each tower. Account has to be taken of all those
factors which influence the desired output values.
It is not very difficult to construct the model for the physical
part of the plant because the time constants are relatively small and
the origin of the different disturbances is readily determined.
The second part of the model is more complicated because the
chemical reactions which occur inside the absorption towers are not very
well known. Many attempts have been made by different chemists to discover
the equations for such a process. The articles about their results are
written only about small columns in laboratories and there is no proof
that it is possible to generalise these results for big plants.
It will be necessary to compare and to verify these equations with
the experimental data about the plant which are being obtained from
Czechoslovakia during the summer.
Seng