Part II

Meanwhile, back on the farm

The scientists and engineers at Salisbury had found there was no lack of work for both ARTVS and AGWAC. In April of 1955 Lonergan wrote to Peter Benyon, the young engineer who had assisted Gilbert in tests of the ARTVS amplifier. Benyon was then undergoing training in Charles Wass' Dynamic Analysis Division at RAE. AGWAC had been in operation for about three months and Lonergan was generally happy with its progress. They had investigated roll stabilisation of RTVl with Demon boosts and "at the moment we are determining the lateral response on the standard trajectory when the missile has been linearised in the manner of Barnsley". Not all faults had been cleared but the machine could be considered as 95% satisfactory. There would, however, be ample scope for Benyon to implement design improvements on his return so his developmental activities at RAE would not be wasted. Lonergan had a number of specific queries concerning TRIDAC performance and mentioned that, before WRE committed itself to building 30 Elliott amplifiers to expand the machine, they would like to know more about Benyon's new operational amplifier about which, he said, "good reports have been coming to hand". He ended by sending best wishes to Wass, Jack Gait, Gordon Herring, Don Verrian, Colin Shapley and Reg Bradbury.

Lonergan had a permanent maintenance staff to maintain and correct faults on the machine. This was most necessary as the analogue machines depended on accurate feed-back components. AGWAC had a large airconditioning plant to carry away the heat from the more than 1000 valves it contained so that the far from stable carbon resistors of the day would vary in value as little as possible. Bernie O'Driscoll assisted Tom Buckland until the latter became primarily an operator. George McCaulay then assisted and later Gerry Currie took over the maintenance section. There were those who claimed that the procedure whereby a specified number of amplifiers were removed from the machine weekly to be checked and recalibrated put in as many faults as it found. Be that as it may, those technicians who have grown up in an era of transistors and microcircuits would find it hard to imagine just how much work was involved in keeping a large valve machine running.

The work was not wasted, however. At the time it was built AGWAC was exceptionally flexible, easy to set up and easy to calibrate. Dr Rupert Trail-Nash of the Structures Department, Aeronautical Research Laboratories carried out pioneering work on Jindivik flutter problems on the machine. A recalculation of these equations, essential in aircraft design, years later showed that AGWAC's results were accurate to a fraction of one percent.

TRIDAC lagged behind AGWAC. Engineering problems and the sheer size of the machine caused most of the delay but by the end of January 1956 Jack Gait was able to write to Lonergan saying that they had managed to get most of the bugs out of the monitoring systems, had recently completed two very simple test problems and were shortly to test the hydraulic servos and axis transformation equipment. He included details of a "very worthwhile modification to the relay corrector units". These had previously sometimes caused the amplifiers to oscillate. Such oscillations were troublesome because they too easily spread to other amplifiers and the original fault could be hard to find. Relay corrector units though, were an indispensable part of any major analogue computer installation. Usually called choppers and coupled with an AC amplifier, they sampled the input of the operational amplifiers and compensated for any internal drifts within them. RAE had designed and specified for both TRIDAC and AGWAC a carefully engineered but extremely expensive chopper. One was needed for each operational amplifier and, at a cost of some 30 to 40 sterling each, as built by Elliotts, they added substantially to the cost of the machines.

WRE's ability to produce choppers cheaply, in-house, would prove to be the most important single factor in the expansion of its analogue computing facilities in the years to come. These were based on a simple modification of commercial 'vibrators', produced in the millions then for valve car radios. An inventive John Dunne, who had initially begun training as a chemist, destined for the Salisbury Munitions Factory, had tried to change to an electronics degree course but had been refused permission because of war-time manpower controls. He merely added enough subjects to allow him to be qualified in electronics as well. He used his knowledge of electronics to work out a simple electrical modification but needed to get the tungsten contacts on the stainless steel springs plated with gold to reduce resistance.

He put the problem to B.H.S. (Basil) Day in the WRE Plating Shop. Day recalls it as a tricky plating problem because tungsten, in spite of its hardness, is chemically like aluminium and is affected by alkalis while the stainless steel is affected by acids.(10) By a logical materials analysis they worked out a innovative plating method. WRE could buy the commercial units and modify them at a total cost of about 30 shillings each but the effect went far beyond the saving in money. Because the sums involved were small it meant that manufacture of the computing equipment could be authorised at Superintendent or Assistant Controller level. Approval to spend large capital sums did not have to be sought and progress could depend on demonstrated, rather than postulated success.(11)

This simple device and the 'Benyon' amplifier became the standard building blocks of the growing SAD analogue computing complex. SAD had taken over AMAGUS, a simulator being designed for the Project E anti-tank weapon when the project was cancelled. ARL in Melbourne laid claim to it but, in dispute with the tough minded Tom Lawrence, Superintendent of SAD, had to settle for the visual simulation part of the machine. The analogue computing section, renamed ABACUS, remained with SAD and was expanded.(12)

Success and expansion

The analogue machines were almost instant successes, unlike WREDAC, which was got working only with difficulty and not for some time after installation. They were simple in principle to program and operate. Unlike digital machines they did not require such time consuming writing of programs or development of high-level languages before anything useful could be done with them. Most of the former Test Vehicles Group staff had been involved with longhand calculations of the aerodynamic and control equations which governed the performance of the missiles and found the ease of connection and the speed and flexibility of their new facilities well to their liking. Although the machines were initially fairly basic there was no shortage of problems for which computer solutions were sought. Overseas scientific and engineering journals as well as classified reports from Australia's allies contained many examples of computer applications. Hayman and Biggs as well as Lonergan had already gained computer experience during their attachments to RAE.

Just as data processing users found that work for WREDAC and later the IBM 7090 rapidly outgrew their capacity so did SAD need to expand the analogue computers. To a degree it was not so difficult at first. Because of the different way the machines worked it was simple enough in principle to add amplifiers and other accessories as they were needed. Jack Lonergan proposed a substantial increase in equipment in September 1956 to meet demands foreshadowed in the discussions which led to Chacksfield's directive. Simple models of the first missiles produced for service use, Blue Sky and Blue Jay had been successful and a great deal more work was in sight for Red Duster and Red Shoes.

First on the list was a WREDAC to AGWAC converter to enable digital trials data to be converted into analogue voltages so that SAD could check its model against actual trial performances. If the model could be made to behave as had the missile when fired then it gave the modellers confidence in their algebraic formulations.

An automatic programmer came next as trials had to be repeatable so that accurate measurements could be assured. They needed more electronic multipliers. Benyon had developed a variable mark-space type, based on an American design, while at RAE. A prototype looked promising and Lonergan thought it worth while producing the 20 and their associated power supplies, which would be needed for the Bloodhound model.

Three dimensional kinematics were essential. In the Bloodhound model five axes had to be resolved, yaw, pitch and roll of the missile relative to the earth and azimuth and elevation of the sight line relative to the missile. It was necessary to use servo operated sin-cosine resolvers to allow conversions between the axis systems. Lonergan estimated the cost of components for each of the six sin-cosine resolver units needed to be about 3 500. Sixty of Benyon's latest amplifiers would have to be built.

Photoelectric curve followers and other function generators for non-linear and other functions would be needed as well. Outside users would need to be catered for. ARL in Melbourne and Aerodynamics Division at WRE had displayed an early interest in using the analogue facilities. The addition of a Flutter Panel, essentially a large number of accurate potentiometers with some amplifiers which could be connected to the analogue computers would make some of their otherwise very time-consuming calculations very easy to solve.

There was also a need for a more complex trajectory computer, he argued. Lonergan thought most of the equipment could be manufactured within the group although outside help would be needed for some tasks and should be sought to ensure that all the equipment was as up-to-date as possible. The following February R.G. (Reyn) Keats proposed regular monthly meetings to control and monitor the expansion program. Keats was regarded by some as a slave driver. His management technique though was simple and effective. He would seek realistic estimates of work that had to be done and how long it would take and then invariably followed up to see if it had been completed. Since those involved had usually set their own targets the inexorability of the follow up visit induced the maximum effort in the responsible individual and proved effective in curbing some of the wilder enthusiasms of those who found new ideas much more interesting than more mundane tasks they had agreed to complete. Keats outlined the work program.

John Dunne was to be responsible for WREDAC to AGWAC Converters using both paper and magnetic tape, an Automatic Programmer and Curve Followers and Function Generators. Peter Benyon was to oversee ABACUS, Electronic Multipliers, a G.W. Programming Unit and Diode Function Generators while Jack Lonergan was given the Three Dimensional Kinematics and the Flutter Panel and Lawrie Henschke the Electromechanical Computing Annexe and Trajectory Computer. Much of the equipment would be designed by those responsible and such workshop engineering work as was needed would be under the control of Jim Harwood of the SAD Engineering Group.

Ian Hinckfuss of Information Studies Group and Ross Treharne and Roy Hynes of Transistor Group (TQD) were also scheduled to help. Much of the construction was undertaken in-house and, because of the urgency of the task, was assembled in a temporary fashion. The builders had many problems to solve. The most serious concerned the lack of suitable components. The cheap high precision resistors and capacitors of today did not exist. Better quality components which were necessary for analogue computing were hard to obtain and were very expensive. Superintendent, T.F.C. (Tom) Lawrence's name appears often as he lent his weight to demands or pleas that the stores system provide parts when they were ordered. Valve equipment, too, was bulky and racks and chassis had to be built and this took time and effort. Castings had to be made to support the servo equipment and the various gears, motors and tachometers carefully assembled and tested.

When the expansion was completed in 1958, AGWAC's original 95 drift corrected amplifiers, of which 21 could be used for integration, had in addition, 20 single product high speed electronic multipliers, five general purpose servo multipliers with 106 associated amplifiers, five electro-mechanical sin-cosine resolvers with 96 amplifiers and additional power supplies, automatic sequencing equipment and the WAC Mk. 2 Digital to Analogue Converter. ARTVS too, had grown but by no means as much. It had a total of 120 amplifiers, three servo multipliers and associated coefficient potentiometers but still had rudimentary control, recording and calibrating facilities.

The 1957 Computer Conference

When SAD scientists spoke at the Conference on Data Processing and Automatic Computing Machines in June 1957 their confidence in their new machines and their consequent ability to understand the behaviour of the missile systems under study was obvious. An exultant Jack Lonergan summed it up at a session on, "The Reliability of Large Scale Electronic Systems", chaired by Mr Emlyn Jones. Most of the session was taken up with details of the apparently inevitable failures of digital machines before they had finished a longish calculation and various ways of at least salvaging some of the work.

In a spirit of friendly rivalry Lonergan said, "I would like to answer that question from the analogue point of view. If the missile does not do what the analogue computer says it will do then we immediately get the technicians to work on the missile!". Others from SAD who presented papers and described a variety of studies of guided missile, aerodynamic, control and guidance problems, as well as proposed new equipment, included Peter Twiss, Jack Cartmel and Les Witchard and Peter Benyon.

Reyn Keats presented a paper which foreshadowed what was to be the major use of analogue computers at WRE for most of the next decade - the mathematical modelling of guided missiles being tested at the Woomera range. The early results from simulations of Blue Sky, Blue Jay, Red Duster and Red Shoes had been good enough to persuade those who planned affairs and allocated funds to agree to a considerable expansion of analogue computing facilities and effort.

The missile projects - Blue Jay, Red Shoes and Red Duster Three teams existed in SAD. Team Leaders were: - Blue Jay - Les Witchard under Bill Watson Red Duster - Alex Biggs under Reyn Keats and assisted by Dick Cawthorne of RAE Red Shoes - Ted Hayman under Reyn Keats and assisted by Wally Jeans of RAE

With time, differences emerged between the teams. This was in part due to differences in research philosophies between RRE and RAE. RAE was a very large organisation with every sort of specialist department. They had the resources to investigate all aspects of a problem. RRE, with fewer departments and less in the way of resources tended to argue for a simpler approach to problem solving. Not surprisingly, the Blue Jay team argued for a no more than adequate model'. To an extent the approach was dictated by the equipment available and the relative paucity of information on the missiles.

It was easier, in the 1950s, to make a missile which worked than it was to acquire a mathematical understanding of all the factors which made it work. Empirical success had gone ahead of mathematical design and incomplete formulas only were available for Blue Sky and, to a lesser degree, Blue Jay. On the side of those who proposed a simple model it should be remembered that ARTVS, even though a very simple machine, had proved itself in assisting an understanding of the dynamic behaviour of RTV1.

Keats, Biggs and Hayman had all spent time at RAE and they set the pattern for the Red Duster and Red Shoes teams. Keats' background in mathematics and assessment was also important. He was, what one would call, a true mathematician, one of those people who are not really convinced of the truth of anything unless they can completely understand it in mathematical terms. Another who played a critical role was Alex Biggs who, rather than fit the model to the machine, drew up a model requirement for Bloodhound 1 that required a machine the size of TRIDAC. It was his argument that launched the expansion of AGWAC.(13) Biggs' team believed that simplification could follow only after a complete model was assembled and understood.

Blue Jay

Blue Jay was a proportional navigation infra-red homing air-to-air missile which was intended for tail-on target interceptions from sea level to high altitudes. Because the infra-red eye had only a limited field of view there were restrictions on the launching angle between the attacking aircraft and target. The missile was of cruciform construction and used a single set of fins for both lateral control and roll rate stabilisation. The tracking system consisted of a telescope and infra-red detector system mounted on a gyro which was free to move about its gimbals. The gyro was made to precess by applying torques to the gimbals and so the whole detection system could be made to lock on automatically to an infra-red source. The signals driving the gimbal torque motors were derived from the infra-red sensing device at the principal focus of the telescope. An alternating current carrier system was used with the infra-red image at the detector being chopped by means of a rotating disk. Due to the physical construction of the rotating disk the detection system had a blind spot in the centre of the field of view. When modified by the torque motor drive electronics this appeared as a blind band in each of the two lateral planes.

A lateral control signal, proportional to that applied to the torque motors, was fed into the missile control system. This signal, summed with a roll control signal, resulted in a movement of the fins, the missile developing a lateral acceleration proportional to this demand and a roll acceleration appropriate to the roll demand. During the early boost phase the control and guidance systems were isolated by a 'K factor potentiometer' which reduced the proportional navigation to zero; this potentiometer was wound up during the later stages of boost. In addition the fins were locked in the central position and released at the beginning of potentiometer wind up. These precautions were necessary to ensure that the missile was not subjected to excessive accelerations due to incidence build-up during boost phase. The overall 'stiffness' of the acceleration control loop in the missile was maintained nominally constant with altitude by an automatic control unit which was barometrically operated.(14)

Initially only ARTVS was available for Blue Jay. A special laboratory was set up to supplement contractor's measurements. The first Blue Jay Acceptance Firing occurred on September 1957 and special firings were held to measure other parameters. The team worked to build their 'no more than adequate' model and by March 1958, Witchard reported a model had been set up which was able to account for the spread of miss distance which had been observed on trials. The behaviour of Blue Jay in flight could not be explained by simple proportional navigation theory and the miss distance was greater than that theory predicted. In qualitative terms the early model was a success in that the much larger than expected oscillations on accelerometer records could be explained as being due to Coulomb friction (15) in the homing eye gimbal bearings and other non-linearities explained by the way the blind spot at the centre of the homing eye was interpreted by the control electronics to produce two blind bands. Bruce Moon published a supporting Tech Memo in July (TN SAD 30) which contained the equations supporting these arguments.

In quantitative terms Blue Jay was harder to assess. During 1958 and 1959 the model was expanded to include representations of induced drag, roll, head noise and loss of lock. Development and understanding of the model was hindered at times as effort had to be diverted from the modelling work to assist in solving some of the difficulties being experienced with the Acceptance Trials firings. Defects in the model had become evident and effort had to be put into estimating their effect on the accuracy of model predictions.

Noise generated in the homing head had earlier been ignored, seemingly because there had been a lack of confidence in the telemetry. All concerned, not only the modellers but those concerned with other aspects of the trials were involved in a learning phase where they had a hard time determining the accuracy of observations, whether from the Range or from the machine. The cumulative effect of errors in an analogue machine was also more noticeable as the model grew larger and more complex.

John Highfield, who had by then joined the team, remembers one brainstorming session with Bill Watson and other team members which resulted in the discovery of a basic error, which had apparently existed in the model since Bruce Moon had first set it up. ARTVS, like AGWAC, used sin/cosine potentiometers to perform Eulerian mathematical transformations, i.e. to change from Cartesian to Polar co-ordinates. One of these potentiometers had been connected wrongly so that all previous results were, at best, of dubious value.

W.G. (Bill) Watson and D.G. (Dave) Strahle checked progress on the Blue Jay model during 1959.(16) The model equations had been fixed and, while they thought some parameters might be changed, there was no intention of adding further complication to it. They saw as the major problem to ensure the equations they had were calculated correctly. As a result, M.J. Hood, assisted by Highfield, carried out an investigation of the equations and checked the computing ability of ARTVS by comparing its results against those which Bruce Moon had obtained with the RRE digital computer in 1956. Hood's report could not have been much comfort for the proponents of the 'no more than adequate model'. He admitted the checking procedures had not been ideal "but in the time scale, and with the computational facilities at present available, little improvement is believed to be possible." Every element of the computer was checked but the procedure did not guarantee against an omitted or incorrectly connected patch cord. "How much variation between sets of '"boot strap"' checks is due to non-linear behaviour, and how much to small undetected errors is not known. Since the missile behaviour, particularly at high altitude, is dominated by non-linearities which can produce multiple-model behaviour, a reasonable variation is therefore accepted."

The actual Acceptance Trials ended in December 1959, and yielded only a very sketchy picture of Blue Jay performance. While the originally planned statistical trials were by no means seen as a success those, who had expressed doubt about the adequacy of the modelling approach were also partly vindicated.

However, as the inadequate model was all they had that could possibly extract more information from the trials, WRE was encouraged to press on in an attempt to define pass path conditions of the weapon under a wide variety of initial conditions so that the G.W. Department at RAE could determine lethality. This they did in an exercise involving some 1400 computer runs. RAE also proposed an 800 run exercise to use the model to give some indication of the missile performance at altitudes above 42,000 feet. Very few trials had been carried out this high because of the unavailability of launching and target aircraft. As well, launching and target aircraft were all sub-sonic. RAE set out a statistical plan for this exercise, carried out over one Easter long weekend because of the urgency. SAD issued reports on both exercises and David Strahle (who replaced Les Witchard) as team leader reported in May 1961, on the adequacy of model. (17)

Strahle's words put as optimistic a gloss on matters as could be justified as he concluded:

"It would seem that, within limits due to the small angle approximations, the model is capable of reproducing acceptable times of flight and look angles, and that ARTVS, as a computer, generates results of an acceptable degree of repeatability.

"However, with miss distances and accelerations at pass, there would appear to be inadequate flight-trials information for a thorough check of the overall model, even though most of the comparisons, as far as they extend, suggest that the results are of the correct order.

"Significant changes in miss distance distributions occur when a head noise characteristic is included which is probably not severe in relation to effects observed on telemetry records, but the number of flight-trials observations is usually insufficient to identify which is the more correct miss distance distribution. It is unfortunate that sightline spin information has been inadequate to check the head representation and make reliable measurements of head noise directly (see Part 1). This, and the doubt on the characteristic of normal force versus incidence mentioned in paragraph 8, appear to be the most serious limitations of the model.

"In conclusion, it is emphasized that these difficulties are not general to the model approach to weapon assessment; rather, they have been peculiar to Blue Jay and would be avoided where Research and Development and Acceptance Trials had as a central object the construction and checking of a mathematical model. The truth of this statement has been demonstrated on subsequent exercises in Systems Assessment Division."

Agnes Jordan (now Jackson), the junior author of the report doubts that RAE got any useful information from their work. The model was inadequate, as was the machine and, since the missile and trials had not been designed for modelling purposes, the trials information was often suspect. However, all the difficulties found in attempting to build the 'no more than adequate model' pointed SAD firmly in the direction they must take if modelling was to be a success.

A more successful simulation -- Bloodhound I and Thunderbird I

These were the first of the new generation of surface-to-air homing missiles. Thunderbird was a symmetrical cruciform missile (i.e. its four rear control fins and wings formed a cross when viewed from the rear). It was boosted to supersonic speed by four wrap-around boosters and powered by a solid state sustainer motor. It was intended for the British Army.(18)

Bloodhound was quite different, being of monoplane configuration employing twist-and-steer control rather like a normal aircraft, except it had no rudder and the elevator and aileron motions were achieved by rotating the main wings, both together and differentially. Bloodhound was also boosted to supersonic speed by four wrap-round boosters but powered in flight by two kerosene (paraffin) fuelled ramjet motors. The RAF had ordered Bloodhound. Both missiles had radome nose-cones inside which their radar antennas could be pointed in the direction of the target. The target was tracked and "illuminated" by a radar on the ground and the missile in turn locked on to, and tracked the target using the energy scattered from the illuminated target.

Their warheads which were triggered by a proximity device contained sufficient explosive to destroy the target for the expected range of miss-distances. Such was the cost of development of these systems that the main contractors, English Electric of Luton for Thunderbird; and Bristol Aircraft, Filton & Ferranti, Wythenshawe for Bloodhound, could afford to fire barely enough missiles to demonstrate that they worked in some fashion and were safe to fire.

Answers to the many questions that remained were to be sought in the next phase of development, the so-called Acceptance trials. Unfortunately for the planners, financial constraints dictated that considerably less than 100 missiles could be fired in these latter trials. Yet so complex was the behaviour of these missiles that vastly more would be required to explore their operating envelopes to the required degree of accuracy. Clearly every ounce of information had to be wrung from the available trials. The missiles were to be prepared and fired at Woomera by Joint Services Trials Units (JSTU's) while a special division within the RAE at Farnborough was set up under Clifford Cornford to ensure that the scientific objectives of the trials were achieved.

In terms of scientific and engineering advances the Bloodhound project exceeded anything WRE, or anyone else, had attempted in the modelling of guided missiles. First of all, mathematical descriptions, consisting of linear and non-linear differential and algebraic equations or inequalities had to be distilled from the mass of sometimes contradictory or inapplicable information provided by the contractors. The model was multi-disciplinary, having to describe aerodynamics, control and servomechanisms, thermodynamics and propulsion, radar and guidance as well as radar noise, etc.

It had to describe the motion and orientation of both the target and missile from launch to the instant the warhead exploded near the target and to calculate the probability of destroying the target.(19) The model had to be subdivided and modularised in sections which could be validated and the trials and instrumentation necessary to validate the sections specified. Never before had such demands been made on range, target-borne and missile-borne instrumentation and the data reduction facilities at Salisbury.

To validate and run the model, AGWAC had to be more than doubled in capacity and many new computing functions added. New analogue multipliers were developed by Peter Benyon, and to link the output of the data reduction facilities digital computer to AGWAC, John Dunne developed WAC, a nine-channel transistorised digital-to-analogue converter.

The weak link in this ambitious assembly of hardware and software turned out to be the new digital computer, WREDAC.(20) Its reliability was far worse than even the most pessimistic forecasts. Instead of getting millions of points of trials data SAD got thousands. Urgent efforts were made to improve the situation and the requirements of modelling were given priority for the small proportion of time the machine operated correctly. The time table for model validation was put back nearly two years. Instead of doing model-checking experiments immediately after a firing and thereby providing the results in time to influence the next firing the modellers were forty or fifty firings behind the field trials. In fact it was not until WREDAC was replaced with IBM's new 7090 computer in 1961 that a satisfactory rate of model checking was achieved.

Convincing the sceptics

Not all of the problems were technical or scientific, however. There was a fair amount of scepticism, particularly in the UK, that the new techniques being developed at Salisbury could provide an accurate model of the weapon. The problem was recognised by management in both countries and regular visits were made both ways by the scientists and engineers. The scientists weren' t the only ones who had to be convinced.

Senior Service officers have a good deal of say in the use and specification of weapons and it was essential to convince them that the statistics produced with the help of the computers were valid. During the period when Bloodhound 1 was being modelled the tall austere figure of Tom Lawrence (21) was frequently seen in the AGWAC room as he led parties of senior officers through while explaining how the machine worked and showing them, over a light box, how the records obtained from Range instrumentation matched those produced by the chart recorders on the computer.

SAD also built a simple machine, appropriately titled, 'Pass the Buck Machine', to demonstrate that electronic circuits could indeed produce analogues of physical behaviour. It consisted of two heavy steel weights, each suspended from a frame by two springs. A metal bar formed the cross bar of an 'H' as it coupled the springs together. In operation, one of the weights was pulled down to an electromagnet and held while the other weight was steadied. When the first weight was released it moved up and down. The other weight, initially at rest, gradually began to move up and down as energy was transferred from the first until the first was at rest and all movement was occurring in the second.

The transfer then began in reverse and so on until all the energy of the system was slowly lost in friction. Alongside the mechanical system a display board showed the mathematical formulas which governed the motions of the cross-coupled spring/weight system.

Also on the board, but quite unconnected with the mechanical system, were simple transistor amplifiers, connected to produce electrical analogues of the mechanical movements. These were displayed on a pair of cathode ray tubes placed alongside the weights so the the sceptical could convince themselves that the actual mechanical movements matched the analogue voltages precisely.

All this careful demonstration and explanation did not stop misunderstandings completely as is illustrated by the following story. One of the problems evident with Bloodhound 1 was that in the high altitudes of the stratosphere the ramjet motors which pushed the missile along at supersonic speeds became unreliable and flameouts often occurred. These reduced the probability of successful interception at high altitudes. The computer model could explain these flameouts only by postulating the existence of much larger collimation errors in the missile guidance than the manufacturing tolerances implied. Collimation error is the misalignment between the direction the radar antenna is pointing and the spin axis of the gyroscope to which it is slaved. Alex Biggs recalls the arguments. "We found that large collimation errors caused the missile to twist and turn so violently that the air-breathing ramjet engines lost their compression and flame-outs occurred. We could reproduce all of these events in our model.

"The missile contractors said we were wrong. There was no way such large collimation errors could exist but they sent two engineers to Australia to investigate. We went through all our computer experiments with the engineers and finally convinced them that the large collimation errors did exist and were the cause of the ramjet failures.

"However, when the two engineers returned to the UK their colleagues refused to believe them. So there the matter stood for many months until an alert RAF officer found the evidence to confirm our belief. At that time the RAF was starting to fire Bloodhound missiles at the UK Aberporth range. When running up a missile on the launcher this officer noticed some erratic behaviour of the guidance antenna. Investigations showed that this only happened when the missile ran on its own internal power supplies, rather than external launcher supplies and was due to the 'noisy' waveform of the missile alternators. That was indeed the cause of the large collimation errors postulated by the Australian team."

In 1962, in recognition of their success, RAE and WRE were invited to present a joint paper to the Royal Aeronautical Society in London on the Bloodhound 1 work.(22) The paper made the following points.

"The Uses of the Checked model
Once it had been established by the methods described (in the paper) that the mathematical model was a good representation of Bloodhound I, in all the varied conditions in which the missile was fired at Woomera, the model becomes a very powerful tool with a variety of uses. While the primary object of the WRE model was to provide engagement geometry under conditions which could not be provided at all in test firings of the weapon, the model also proved to be an extremely valuable research tool for the improvement of the existing weapon or the development of future weapons. Thus the uses of the model can be divided into two main categories.

"Missile Research and Development
Throughout the Bloodhound 1 evaluation trials the weapon manufacturers retained an active interest in the model activities. They were kept fully aware of progress and, (see sec. 10 of the paper) gave every assistance in tracking down the causes of model discrepancies. As a result some defects in the design and understanding of the weapon were discovered and corrected. This valuable result can be regarded as a by-product of the model checking phase.

"During the weapon's development stage the manufacturers had themselves devoted great effort to the development of models and simulation techniques, indeed some of the design features in their models had been incorporated in the AGWAC model. Nevertheless, at the end of the AGWAC model checking phase, it was recognised by the missile manufacturers that the AGWAC model was the most comprehensive Bloodhound 1 model in existence.

"At this stage the manufacturers were invited to suggest investigations which would help them in their weapon research and development programme. Nearly all of these investigations were accepted and done, using the AGWAC model. Two of the more important results are summarised below:

(a) The results of the AGWAC model checking programme showed that the wind tunnel information upon which Bloodhound I aerodynamic models were based was incorrect. More tunnel tests were done and the results compared favourably with earlier tunnel tests, but the information did not correctly describe Bloodhound I. Meanwhile, the AGWAC model was corrected to agree with flight evidence. At first sceptical of the AGWAC results, the manufacturers now fully agree with the AGWAC findings and, in collaboration with Weapons Research Establishment, they have suggested theoretical reasons why the tunnel information was wrong. The AGWAC findings have proved to be of vital importance in the development of a future weapon.

(b)As was mentioned (in the paper) a number of missile-to-missile variations were discovered or confirmed as a result of the AGWAC model checking programme. They were generally in the nature of varying electronic biases and gains and aerodynamic misalignments. The manufacturers wanted to know what the effects of these variations were on overall missile performance. More than two thousand firings with a variety of conditions were simulated on AGWAC, each representing a different missile with bias and gain variations drawn at random from their populations, and the results of this work showed that some of the variations discovered as a result of the AGWAC model checking programme had a very profound adverse effect on missile performance. This work also explained some of the peculiarities in missile performance noted in the Woomera firings and never before satisfactorily explained. The manufacturers have since corrected these faults in the missile design and the results of the AGWAC model work also show the considerable improvement in missile performance to be expected.

"Bloodhound Engagement Geometry
The primary purpose of the model was to provide engagement geometry for missiles fired against a variety of different targets flying with various speeds, heights, ranges, tracks and manoeuvres. To this end several thousand AGWAC runs were made, including in the model the various random variables that had been discovered or confirmed during the model checking phase. Thus every run represented a different missile drawn at random from the population; random variables resulted in a scatter in engagement geometry. From the engagement geometry results the lethality in each condition was determined. The work can conveniently be divided into three categories as follows:

(a) Replication
As described (in the paper) the evaluation missiles were fired at Woomera with a wide variety of target conditions. The number of fully representative results obtained in each condition was usually limited to less than three for two reasons. First, for economic reasons the total number of missiles allocated to the trails was limited, and second, despite the intensive research and development programme that preceded the evaluation trials some missile modifications became necessary as the trials proceeded. The result of this was that the number of missiles built to the weapon standard was less than had been expected. There was, however, no difficulty in adjusting and checking the model to the current weapon standard.

In these circumstances the model was used to generate a very much larger sample of engagement geometry than could be obtained in each condition from the Woomera firings. Excluding those missiles which developed faults in flight, the engagement geometry obtained from the Woomera firing trials results was, in every case, easily recognisable as a possible member of the distribution given by the model, and in some cases the small sample of firing results was shown to give a misleading impression of missile performance.

(b) Interpolation
There was quite a straightforward use of the model to provide information in conditions lying between the conditions investigated in the Woomera firings.

(c) Extrapolation
This was the most valuable use of the model and the one which imposed the greatest demands on its validity. Despite intensive efforts to provide realistic targets at missile testing ranges the targets used differ in performance and other characteristics from those likely to be encountered in the operational use of the weapon. In the Bloodhound trials the model results constituted the only source of information on the performance of the missile in such operational conditions.

Many operational conditions were investigated using the AGWAC model and from the results a clear picture was obtained of the performance of the current missile in such conditions."

Part III

References:
!0. WRE Electrofinishing Section Instruction J3014
11. To those unfamiliar with Public Service bureaucracy this may sound amusing but it is indicative of the environment within which creative people had to work.
12. By the mid-sixties ABACUS was no longer needed and following a request from Dr Peter Hetzel was refurbished and lent to the Royal Adelaide Hospital to further his investigations of pulmonary function.
13. Report SAD 20 by A. G. Biggs (Sept. 1959) presented not only a complete mathematical model of Bloodhound but also a series of experiments on each firing to validate the model. From this the requirements for computing hardware and range instrumentation were derived.
14. Description of missile operation taken from L.J. Dunne's Report SAD 26.
15. Best explained as something like what engineers call 'stiction' , the force or torque needed to begin moving a body on a plane surface which is greater than the force needed to maintain motion. Rather than behaving as if viscous friction applied when torque was applied to the gyro it jumped suddenly to its other state with the result that missile motion was more than expected.
16. TN SAD 53, Published January 1960
17. Reference Technical Note SAD 75 "The Adequacy of the WRE Mathematical Model of Blue Jay" by D.G. Strahle, J.A.F. Highfield and A.E. Jordan, May 1961.
18. As mentioned before in "Conflict between establishments", WRE did little work on Thunderbird 1
19. WRE was not required to provide a mathematical model of the warhead. Very few rounds were fired with warheads. Instead Tom Lawrence arranged with ARL structures department to examine and report on every target aircraft which was damaged or destroyed.

20. Fire across the desert p. 386-389.

21. Lawrence, T.F.C., Hayman, E.G. and Benyon, P.R. Use of a Mathematical Model in the Evaluation of Guided Missile Performance. Jour. I.E. Aust., September, 1961.

22.Biggs, A.G. and Cawthorne, A.R. Bloodhound Missile Evaluation. Journal of the Royal Aeronautical Society, Astronautics and Guided Flight Section, Vol. 66 September 1962.

AGB COMPUTER ART home page