Ejecución de experimentos
Tomado del libro Introduction to scientific research, E. Bright Wilson, Dover 1990
Versión 20 agosto 2002, RMM


Regresar a Página Principal
Ir a menú de Prácticas de Óptica

ÍNDICE

  • SOME GENERAL SUGGESTIONS
  •   Haciendo cambios
  •   Ensayos
  •   Lista de tareas
  •   Calibraciones
  •   Analisis de datos
  • LIBROS DE NOTAS Y RECORDS
  •   Numeros de identificación
  •   Químicos
  •   Etiquetas
  • CUESTIONES SICOLOGICAS
  •   Trabajo de grupo
  •   Terminando una investigación
  • TRAYENDO UN APARATO BAJO CONTROL
  •   Pincipios de búsqueda
      Métodos de variación artificial, de estabilización, de correlación y factorial
      14 recomendaciones
  • NOTAS GENERALES


    Para comentarios y sugerencias contactar a Roberto Machorro

    Regresar al inicio

    Some General Suggestions

    It goes without saying that if two investigators are otherwise equal, the one who understands best the theory on which his apparatus was designed has an advantage. This is one reason why a person who has built an apparatus is usually best fitted to operate it. With the present trend toward commercial instruments, tightly sealed in closed boxes, there is a dangerous tendency to develop mere "knob wirlers" with only a vague understanding of the insides of instruments and the theories on which they are based.

    Closely related to this requirement is the desirability of taking the time to think out and explain all the odd things an apparatus does. As equipment gets more complicated, it may develop unexpected quirks of behavior which are often not immediately understood by the operator. The urge to shrug these off as irrelevant side phenomena should he resisted, since they are at the least signs of a lack of complete understanding of the apparatus and at worst symptoms of serious troubles.

    These efrects can sometimes be very baflling and arise from most unexpected causes. In one apparatus a trouble arose which consisted of a regular galvanometer jump every time a certain automatic switch closed. It required some investigation to discover this correlation because the switch in question was not supposed to be connected to anything. Further study tracked down the eflect to static electricity on a moving belt, which discharged through the switch because of the proximity of an unused cable attached to the switch. Needless to say, this was not a very tidy piece of apparatus.

    Often one's faith in cause and effect is put to a severe test by apparatus troubles, and the weak are likely to fall hack on their ancestral inheritance of faith in gremlins and' psychic phenomena. The persistent and materially minded scientist will, however, generally find the trouble by logical methods. One of the features it is most important to understand is just what the apparatus really measures. This is usually not exactly what one wishes to measure but an approximation thereto under definite conditions, Are these conditions really being adhered to? Often they are not, with resultant error.


    Regresar al inicio

    MAKING CHANGES

    An important maxim is: Never try more than one new feature at a time. If it is desired to put in a new unit or change to a new operating procedure, be sure the old system was operating as it should before making the change, and limit the changes to one at a time if at all possible. When necessary, borrow tested components from another apparatus in order to tryout new units one at a time. A step-by-step attack, in which each stage is a sure and definite advance, usually saves time in the end. This advice can be modified if a factorial-type experiment is feasible.

    Related to this is the rule: Leave well enough alone, Although this should not be taken too literally, it is a great mistake to make untested changes, . however trivial they may seem, just before an important experiment. This is particularly true of critical tests such as eclipses, artificial earthquakes, atomic-bomb explosions, and demonstrations before the Board of Directors. Trivial changes have wrecked many such events.

    There is a story of an industrialist who was unable to duplicate in a branch factory the manufacture of a chemical in a desirable crystal form. On further investigation of the process in the main plant, he found it in charge of an elderly employee who had always operated it, apparently with a perfectly simple routine, Closer examination, however, revealed that at one stage he always injected tobacco juice into the vat. This was the missing step, presumably because of surface active ingredients which influenced the growth habits of the crystals.


    Regresar al inicio

    REHEARSALS

    When a crucial experiment is to be performed, where a fililure of the apparatus would be a serious matter, "dress rehearsals" are essential. A rehearsal is not sufficient unless it is as nearly as possible under the same conditions as the final experiment. If the grand test is to be attended by generals, admirals, and congressmen, be.sure to have stand-ins present at the rehearsal; otherwise it may be found that the guests obscure the operator's view at a crucial moment or otherwise introduce unexpected new features. Human foresight is a very weak reed on which to rely.
    One apparatus which had been very thoroughly tested in the laboratory gave trouble in the Held because it had not been mouse-proofed! Left overnight in the Held, it attracted a colony of field mice, who nibbled the wax in the transformer boxes.

    Naturally, every effort should be made to foresee troubles. A mental rehearsal of the experiment is valuable for this purpose. The observer stands in front of the apparatus and pretends to carry out all the operations in the proper sequence, thinking at each stage about what may go wrong. What will happen if various things break, or stick, or overheat, or plug up, or flood, or faiI to make contact, or leak, or burn out, or deflect beyond expectation, or below expectation, or spill, or get wet, or get contaminated, or are dropped, or freeze, or are vibrated, or arc-over? What accidents and fililures have occurred in previous experiments of a similar nature, and how can they be prevented or coped with if they occur? Is the setup safe with respect to explosion, electric shock, fire, poisonous vapors, radioactivity, mechanical collapse? Even extremely. able and experienced people sometimes make remarkable blunders which would surely not have happened if the above questions had been asked beforehand. There was a notable case in which an automatic pressure regulator was built using the electrolysis of water as the source of pressure. When the pressure had built up sufficiently to force the water below the electrodes, the dissociation was supposed to stop. It stopped all right, but the resulting spark initiated a rapid and disastrous reverse of the dissociation reaction.


    Regresar al inicio

    CHECK LISTS

    Before an apparatus which requires any appreciable numnber of operations is used, a check list of these operations should be made up. This should be faithfully used for each experiment. An enormous amount of time is wasted in research because some switch is not closed during an experiment. It is fairly obvious that the use of a carefully prepared and tested check list is absolutely mandatory when the experiment can be done only once, as at an eclipse, or is very expensive or time-consuming. It is nevertheless true that there are many experimenters who have fdt themselves above such aids. For example, at a certain very large test explosion, someone forgot to insert the booster charge, with the result that there was no detonation, only a fire. Since the fire could easily have gone over into detonation, the large working crew had to remain inside the bomb proof for the considerable period required for the material to burn out. Check lists pay off not only in such large experiments but also in much everyday work, where an error is not fatal but wastes valuable time.

    Regresar al inicio

    CALIBRATIONS

    A motto some believe in is: Don't trust anyone. This is certainly applies to the question of calibrations. An experimenter of experience would as soon use calibrations carried out by others as he would use a stranger's toothbrush. This holds as well for calibrations provided by the manufacturer of commercial equipment, particularly after the apparatus has been used by others.

    When a new apparatus has been adjusted and put in good operating order, its performance characteristics should be measured and recorded. In this way, any future changes in adjustment can be readily detected. When the apparatus is taken down, altered, or repaired, a standard is available which the experimenter knows he should be able to attain. Closely associated with this is the advisability of having a quick method of checking over-all performance which should be used before investing time in an .. actual run. This is naturally more important if the final experiment is one involving considerable advance preparation.


    Regresar al inicio

    ANALYSIS OF DATA

    The really interesting experiments are those whi<.:h yield unexpected results. The sooner such results are recognized, the sooner the observer can follow up the new leads. For this reason, it is of great importance to analyze observations as quickly as possible after they have been made. If at all practical, this step of analysis should go hand in hand with that of observation. Often the conditions of an experiment are not correct, and much time is wasted by recording the results in a mechanical manner without thinking about them.

    It should be remembered that success£ul research requires the use of the mind as well as the hand and the eye. The modern pressure to "get results" tends to lessen the time spent in contemplation, but more thinking at every stage of an investigation is generally in order. Even at the cost of reducing the efficiency of the actual data gathering process, it is usually better to analyze and interpret observations while they are being obtained. '


    Regresar al inicio

    Notebooks and Records

    It is hard to conceive of a perfect laboratory notebook, and it is regrettably rare to find one that is even moderately satisfactory; yet the keeping of good records of work done is a major key to efficiency. There are bound to be dissenters to any set of fixed rules, but these will probably be rarer for the ritual of notebook keeping than elsewhere. Consequently, a set of rules which is generally regarded as satisfactory, or even as essential, will be somewhat dogmatically stated.

    Some great discoveries have been delayed because of careless record keeping. Thus it is related that the astronomer Le Monnier observed the planet Uranus on several occasions before its identification as a planet had been announced by Herschel, but decided that it was a fixed star. 1his was probably due in pait to the fact that he . wrote his measurements on scraps of paper, including a paper bag originally containing hair powder!

    Laboratory notebooks should be permanently and strongly bound and of sufficient size, say roughly 8 by 10 inches, with numbered pages. Loose-leaf pages or separate sheets are too easily lost to be satisfactory, especially since a laboratory notebook gets rather rough handling, and perhaps an Qccasional dousing with acid. An exception is the case of routine, repeated measurements, where a printed or mimeographed special blank is often useful if a good system is established for collecting and binding the separate sheets. Ruled pages are generally used, but this is a matter of personal taste, and some prefer ul1l'uled or cross-sectional pages. A rubber stamp may be used to provide headings for routine entries.

    Data should be entered directly into the notebook at the time of observation. It is intolerable to usememory or scraps of paper for primary recording, because of the inevitability of error and loss. Therefore, there should be a good place for the notebook at the operating position, and the experimenter should never he without his book when in action.
    Data should he recorded in ink, preferably a permanent brand, and a blotter should be handy. Otherwise the record is too ephemeral. Notebooks get hard usage, and pencil smudges rapidly. When the notebook may be used as evidence in a patent case, ink is much preferred.

    Rough, qualitative graphs can be drawn in directly, but more careful ones are usually best prepared on graph paper of the most appropriate type. These are then caref1.llly pasted in the notebook, a blank page being cut out in order to compensate for the bulk of the one added.

    Notebooks should carry the name of the user and the dates covered. It is convenient in a research group to agree on a standard size, but then some sort of external identification is a great timesaver. The first eight or ten pages should be reserved for a table of contents. This consists of a line added chronologically for each series of similar experiments, together with the page reference. 'The table of contents is enormously helpful in finding items later and is very simple to keep up. An index in the back of the book is advantageous but not indispensable.

    Each entry should be dated and, if several individuals use one book (not generally recommended), initialcd. The material should not be crowdcd on the pages; paper is cheap compared with other research expenses.

    The principal difficulty is in deciding what to put in. Obviously, one enters numerical results and those values of the independent variables such as temperature, composition, or pressure which are directly concerned. It is also necessary to have a syst m of entries or references so that years later it will be possible to tell what apparatus was used and under what circumstances. Somewhere there should be available a rather complete description of the apparatus. Then, when modifications are made, they should he described immediately in the notebook. It should also be possib]e to trace back the source of calibration curves, corrections, etc., which were appropriate to the data of a given day. It is helpful if the requirements for writing a paper, a thesis, or a book are kept in mind. Such a task, once carried out, usually leads to solemn resolves to keep a more careful notebook in the future. Also extremely salutary is the effect of trying to figure out something from another's book. All references to apparatus, places, times, books, papers, graphs, and people should be sufficiently explicit to be understandable years later. It should be possible to take each scientific paper and show just where every figure, description, or statement in it is backed up by original observations in the laboratory notebook, and exactly why the final and original numbers differ, if they do.

    Some statement of the purpose of each experiment and a summary of the conclusions reached make the notebook vastly more useful. Sketches, drawings, and diagrams are essential. Since so much observation is visual, it is important to record what is actually seen, including things not fully understood at the time.
    Bad or unpromising experiments, even those deemed failures, should be fully recorded. They represent an investment of effort which should not be thrown away, because often something can be salvaged, even if it is only a knowledge of what not to do.

    Data should always be entered in their most primary form, not after recalculation or transformation. If it is a ratio of two observations which is of interest but it is the two numbers which are actually observed, the two numbers should be recorded. If the precise weight of an object is important, the individual balance weights used and their identification should be included, i.e., the serial number of the box. Otherwise it is not possible to apply calibration corrections later or to change the corrections if new values appear. Naturally, this detail is not necessary if only a rough weight is invo]ved. A tabular form is best for numerical data. Units should be noted.

    Where patent questions are involved, it may be desirable to witness and even to notarize notebook pages at intervals. The witness should be someone who understands the material but is not a coinventor. Material added to a page at a later date should be in a different-colored ink, and any alterations should be initialed, witnessed, and dated if they are likely to be important. Industrial concerns usually enforce their own rules concerning patent matters.


    Regresar al inicio

    IDENTIFICATION NUMBERS

    It is foolish to spend time and money making records of various kinds such as pen-and-ink recorder sheets, photographic records, or spectra if these are then lost or mixed up. Every such record should carry indelibly on it complete identification. A simple system of doing this which has worked well in practice is to write in ink on each record a symbol identifying the notebook and then the page number on which the auxiliary data are recorded. If more than one record occurs on a page, letters or further numerals can be added. Thus EBW II 85c would identify the third record discussed on page 85 of the second EBW notebook. This is better than a serial number, which doesn't tell without extra keying where to look for the notebook entry. A good filing system is indispensable for all films, photographs, charts, graphs, circuit diagrams, drawings, blueprints, etc. It is hardest to devise satisfactory filing methods for either very small or very large material. The former are easily lost and the latter very bulky. Small envelopes are useful for tiny films and also protect them from scratching.

    It is thoroughly worth while to save drawings and blueprints from which apparatus has been built, however rough these may be. These should be dated, initialed, and labeled; in fact every piece of paper containing useful material should be so marked. When an electronic or other similar piece of equipment is built, a careful circuit diagram should be prepared, fully labeled with all constants. 111e apparatus should carry a serial number which also appears on the diagram. When changes are made, these should be indicated on the diagram and dated or a revised, dated diagram made. The old one should not be obscured or thrown away because it may be required to explain earlier data, later found to be peculiar. It is convenient to draw such diagrams on tracing paper with a good black pencil. Cheap ozalid or similarly processcu copies can then be made. One should be kept in the laboratory, where it will usually prove inuispensable for trouble shooting. Sooner or later, however, it will be useu as scratch paper by someone with a brilliant idea to demonstrate; so the official copy and spares should be filed elsewhere. A great deal of time is unnecessarily wasted poking around the insides of an apparatus trying to find out where some wire goes.


    Regresar al inicio

    CHEMICALS

    Whenever chemicals are used, it is imporant to record in the notebook their source and grade and just what purification or treatment they have received. Nonchemists often make the grave mistake of thinking that "sodium chloride" is an adequate description. At a slightly higher level of sophistication they recognize the difference between "crude" and CP grades, although usually they confuse CP and USP labels, Most never get beyond the idea that CP is the ultimate in purity. The old saying that a chemist is a man who makes bad measurements on good material and a physicist is one who makes good measurements on bad material is too painfully true in many cases. Physical chemists, however, resent the extension that they make bad. measurements on bad material. It is as well to remember that any two samples of the same substance probably have demonstrably different quantities of impurities, impurities which in some experiments may be the origin of the whole eflect being observed. This is particularly true of certain biological experiments. Plants, for example, will visibly react to one part of ethylene in a million parts of air, so that effects formerly attributed to other causes were ultimately found to be due to traces of ethylenc.

    Regresar al inicio

    LABELING

    Related to the question of notebooks and records is that of labeling. Naturally bottles of chemicals must carry adequate labels, which should include not only the chemical name but also the source, or a notebook 'page reference if there has been any special treatment, or initials and date. One research worker departed from a certain laboratory to take another job and left a good deal of material behind. One bottle of clear liquid carried no label. Those assigned to clean up examined it, smelled it, finally concluded that it was water, and poured it down the drain. It was water, all right-heavy water at $30 an ounce. Some supervisors relentlessly throw out unlabled bottles on sight. It only needs to be done once or twice.

    Labels are also essential on all kinds of specimens, pieces of apparatus, gadgets, etc. Controls on apparatus should be labeled and the apparatus itself numbered. Every laboratory has orphaned pieecs of e
    The whole purpose of all these recording systems is to preserve values. They should be carefully thought out to fit the conditions of each laboratory and should be adc

    Regresar al inicio

    Psychological Questions

    Success in scientific work is not assured by any mechanical system of rules but is to a large extent the result of intensely human factors.

    No attempt will be made to go deeply into these psychological factors here, although they provide a subject of great importance which would repay' study by qualified persons.

    Everyone with any research experience is impressed by the great differences in mental attitude and behavior with which different investigators approach their work. Some are energetic, confident, rapid; others are slow, diffident, ineffcient. There are probably two indispensable characteristics: enthusiasm and self-confidence. If a man does not believe a job worth doing, he is unlikely to push ii' hard enough over the difficult places. Enthusiasm is very largely a product of the tnvironment and atmosphere. A few very unusual individuals can show it without help from their superiors, but, generally speaking, the responsibility here rests squareely on the director of research or his equivalcnt. If he isn't interested in the individual research problems going on in his laboratory, why should his employees be?

    Self-confidence comes naturally to some but has to be nurtured in others. As a man achieves successful results, he grows in this regard. Encouragement, sympathy, and advice from his supervisor can help. It is also very important that he not be allowed to get mired in a problem beyond his depth. After a reasonable period he should be given a different problem. Students given up as hopeless have often blossomed forth remarkably when shifted to problems better suited to their talents.

    It is claimed that observers often can make finer distinctions and more delicate sense perceptions than they believe themselves capable of: Furthermore, their ability in this regard is said to increase as their self-confidence is increased by encouragement or othelwise. Thus one subject who was engaged in some tests of fine taste distinctions was encouraged to believe that he was endowed with extrasensory perception. This so boosted his self confidence that he quite exceeded his previous per!()rmances. However, it is reported that people rapidly become fatigued if driven to the limit of their sensory capacities, especially when they believe that they are just guessing amI are not really able to make true distinctions.

    Because of the importance of the personal, factor, the investigator is quite justified in demanding a reasonable degree of comfort and suitability in his laboratory, when these are practical. He should be sure that his apparatus is such that he can operate it in safely and convenience. Improvisations which interfere with these requirements may sometimes have a temporary justification but they should never be allowed to lapse into permanence.

    Linked with this topic is the advisability of planning schedules so that adequate time is available for each experiment. Hurried observations are likely to need redoing. Small bits of time need not be wasted; there are usually short tests or apparatus-performance measurements which can be done quickly, or preparations for later experiments. If it is possible, a schedule which sets aside whole days for research is much more useful than one allowing the same total amount of time in short intervals.


    Regresar al inicio

    TEAMWORK

    Teamwork in science is becoming more and more frequent. Sometimes this is quite necessary, but in these cases it is usually a question of organizing a large project into parts, so that an individual may still work as such on his own part. Generally speaking, true teamwork is rather difficult in theoretical problems. In experimental work it is often not only possible but desirable. The advantages are filirly obvious. People are usually happier working together, they can build up each other's enthusiasm and confidence, they can complement each other's special abilities and knowledge, the work is less likely to drag out over long periods of time, and many jobs can be done more efficiently by two together than by two separately. Also there is less danger of a false conclusion based on personal bias.

    There are disadvantages in teamwork, too. Usually it is rather inefficient from a man-hour viewpoint because so much of the time one of the partners is merely standing around waiting and watching the other do something, or perhaps interfering. It is very desirable to develop the habit of being useful during these 'periods. Ordinarily there are plenty of jobs to do; it is seldom, for example, that a research laboratory would suffer if someone went around picking up things and putting them where they belonged. It should be remembered that if one man requires no more than 99 per cent longer to do a job alone than to do it with an equal partner, man-hours will be saved by doing it alone. Partners are not always congenial, in which case they would often do better to separate. It takes an amiable man indeed to put up with a partner who insists on turning one set of knobs while his companion is trying to adjust another. Good communication between partners is essential. Each should know at all times what the other has done, what he plans to do, and just what his own responsibilities are. Ambiguous and indefinite arrangements are a curse. If you want Jones to clean the muck out of the vacuum pump, don't just politely hint that it might be a good idea if someone did it.

    The question of who is boss or, if the partnership is an equal one, who referees, disagreements should be answered. Also it is very important that cooperative work should not be entered into unless all parties understand clearly the share they will have in any results. Who is going to have his name on the publications or patents, whose name comes first, and who can expect a mere acknowledgment of help? It must not be forgotten that scientists are human beings and as prone as anyone to magnify the importance of their own contributions. No enemy is as deadly as one who thinks that someone has stolen his intellectual property. Sometimes arrangements just drift into existence without clear understandings; these are frequent sources of later grief. One practical difficulty with large numbers of collaborators is the length of the list of authors of publications. This looks more like the name of a New York law firm than something which will bring pleasure to the' eye of a librarian. As a consequence, the prestige which can result from a good paper is much diluted.

    Another drawback of partnerships, especially among students, is that they tend to blunt the development of self-reliance and to f()ster a one-sided growth. If Smith solders better than Jones, Smith will probably do all the soldering, so that Smith gets better and Jones poorer at this art. A research worker should serve a period of apprenticeship during which he gradually develops his own powers. At the right moment, neither too soon nor too late, he should try his own wings.

    Senior research supervisors sometimes take a few years to learn that their students or assistants are not mere automatons and that they require encouragement, praise, and recognition. One sometimes sees cases where the supervisor takes visitors, distinguished or othelwise, through the laboratories, completely ignoring the men working there, as if they were of no importance in his "empire." Such supervisors usually end by having few students, if any.


    Regresar al inicio

    COMPLETING A RESEARCH

    Bringing an Apparatus under Control One of the commonest psychological Haws among research workers is the tendency to drop a project that is half completed in favor of some new research which seems more interesting at the time. Frequently the new project then suffers the same fate. The almost inevitable result is a complete dissipation of energy. This desire to try something new usually arises when a difficulty has made the old project less attractive than it seemed at first. Then a newer idea appeals because its difficulties have not yet become manifest. Further more, it is human nature grossly to underestimate the time and effort which will be required to carry a new research to completion. The argument is. therefore advanced that the new problem can be worked out in a very short time and then the old one taken up again. In practice this seldom happens. It turns out that there are difficulties associated with the new idea, too, and in its turn it is abandoned in favor of a still newer one. It is hard to suggest any cure for this dis ase, which to a greater or lesser degree affiicts nearly all scientists, but it is quite obvious that giving in to this very strong pressure is completely fatal to any accomplishment.

    Regresar al inicio

    Bringing an Apparatus under Control

    When an apparatus has been first set up, it seldom performs as planned. 11le problem is then to put it right. Of fundamental importance at this stage are a finn belief in cause and effect and confidence in the possibility of ferreting out the trouble. The difficulties often appear extremely mysterious at first but patience and a systematic attack can locate their origins.

    Sometimes new eHects arise either after a period of successful operation or during the process of eliminating faults. It is then certain that something pertinent to the apparatus has changed. It is usually only necessary to go over all the things about the apparatus which have changed in order to get a list which includes the cause of the new phenomenon. Internal changes caused by aging must be included, e.g., corrosion, warping, running down of batteries, etc. This simple principle is a very powerfull tool for finding causes of new phenomena in apparatus, but it is important to list all the changes. As a general tool for scientific research this rule has flaws, but usually the number of things which can affect a piece of equipment is very limited.

    Causes can be very elusive. A photographic-film manufacturer once had trouble with film from one unit (so the story goes). The symptoms indicated contamination by mercury, but a thorough search failed to disclose how it was getting in. Hejection of product continued, with growing desperation on the part of the management. Finally, a good batch was produced! Nothing had been changed except the operator. This led to close questioning of the former workman, whose brother was found to be under treatment with mercurials. , Usually a new apparatlls does not work properly because some important variable is not under control. One of the commonest of such variables is the temperature, and it should always be high on the list for consideration. Another common troublemaker with electrical equipment is variation in supply voltage. . .

    METHOD OF ARTIFICIAL VARIATION

    When a variable is suspected, a standard method of testing the hypothesis that it is the cause of some trouble is to vary it artificially by an amount somewhat greater than its normal fluctuations and see if the troublesome effect correlates closely with these variations. For example, the windows may he opened and later closed to change the room temperature fairly markedly, or a variable transformer may be inserted in the electrical-supply line and the supply voltage changed by 10 or 15 volts each way.

    However, this method may fail. In the first place it is often impossible or very inconvenient to produce the desired artificial variation. For example, the suspected cause might be the incidence of cosmic rays. Second, the response to variations may be very nonlinear so that the efIect, say, of 10 degrees change in temperature is several times greater than that of a 5-degree shift. The 1O-degree test might then be positive, whereas in fact natural fluctuations of 5 degrees actually produced an unimportant efIect. Finally, there may be a number of causes nearly equal in importance. When the voltage is pushed up, .some other variable may by chance go down, counteracting and thereby hiding the efIect of voltage changes.

    METHOD OF STABILIZATION

    An alternative to accentuating the changes in a variable is to reduce them. A voltage-stabilizing transformer can be introduced in the supply line and the elluipment tested again, for example. However, the trouble often goes away when such a move is made merely because of a coincidence, the true cause of trouble becoming quiescent for a period. Sometimes it is not a mere coincidence; the process of making some irrelevant change jars a bad connection or otherwise alters the situation, usually only temporarily. Conversely, the trouble may remain, even though the variable under consideration is the true culprit, simply because the control measures were insufficient. They should always be tested for effectiveness.

    When a trouble due to unknown causes disappears without known reason, there is a strong urge to let sleeping dogs lie. This is not unreasonable but is unsound if some unique or very expensive experiment is in preparation, because the trouble may reappear at an embarrassing moment. It should also be disturbing as evidence that there is a lack of complete understanding of the equipment.

    METHOD OF CORRELATION

    I t is sometimes impossible or inconvenient to reduce the fluctuations of a variable or to give it controlled variations larger than its normal fluctuations. In this case it may be possible to record the changes naturally occurring and compare them with the effects under study. If the effect follows the observed changes in the variable, these changes may be the soughtfor causc. Correlation technilIues may be required in difficult cases. The correlation may be obscured by time lags, especially if flows ofhcat or matter are involved.

    THE FACTORIAL METHOD

    If the most important variable has effects several times larger than those of the second variable (or distinguishable in kind) and the second variable is several times more important than the third, etc., they' can often be located one at a time, the largest one being found first and then brought undcr control, the second largest next. But if there are a number of variables of roughly equal effect, it may be difficult to identify them one by one, because all of the above, methods can filii when applied to one variable if the system is very unsteady because of fluctuations in other variables.

    In this situation, the method of factorial experimentation seems indicated. For example, suppose that temperature, water pressure, and line voltage are not controlled and are suspected as causes of some measurable effect, such as shifts in a meter reading. Then arrangements are made for giving each of these variables a high and a low value, somewhat outside their normal ranges of variation. There are eight combinations possible, one being high temperature, low pressure, low voltage. One observation is made under each of the eight conditions. By averaging the four high temperature runs and comparing with the average of the four low temperature runs, the effect of temperature may show itself in spite of fluctuations due to other variables. . . . Exactly the same treatment can be applied to the same data in order to determine whether pressure or voltage is producing any significant effect. If the error calculated from this analysis is too great, it may be necessary to enlarge the experiment by adding other variables. . .

    Another technique of great usefulness at this stage of an experiment is the quality control chart. Although this can be used with no statistical knowledge, it is better understood with a small background of statistical principles. . . .


    Regresar al inicio

    Search Principles

    Scientific research is always concerned with searches of one kind or another. These may be as trivial as searching for the proper setting of three screws to bring a galvanometer into a level position or as largescale as a search for a cure for cancer. The problem may be to locate the cause of trouble in a complex apparatus or to locate the maximum in a function. However diverse these search problems may appear, they have certain aspects in common, and it is possible to set down a few principles which apply to most of them.

    1. Know as much as possible about the subobject of the search. It is particularly important to be familiar with those properties which can be used to distinguish the object from the surroundings.

    This principle is obvious when the objective is a physical thing, a collar button under the dresser or a submarine at sea. Thus a knowledge that the latter reflects sound, is made of magnetic material, produces a noise, leaves a wake, is a certain length, can submerge to a certain depth and run at a certain speed is of the utmost importance in designing methods of location. It is equally important in seeking for the cause of a disease to learn as much as possible about the disease. If the search objective is a trouble in an apparatus, the characteristics of the difficulty should be determined before looking for its origin. If the trouble is a leak in a vacuum system, how does the leak behave? Does the pres, sure continue to rise until atmospheric pressure is reached, or does it level off at a lower value, as if a vapor were responsible? Does the leak vary with the conditions of operation, as, for example, with room temperature? Generally speaking, a little tim spent in learning the characteristics of the object may shorten the search materially.

    2. Prove, if possible, that the object exists in the area to be searched. (This rule, unfortunately, is not always applicable.) It is often possible, in mathematical work, for example, to prove that a solution to a given set of equations exists before undertaking to find the solution itself. The alertness and efficiency of an observer are much greater if he believes that he is looking for something which certainly exists. This is a strong argument for ensuring that the observer knows the whole background which led to his carrying out a certain set of operations. The greatest secret connected with the atomic bomb was probably disclosed when the explosion at Hiroshima showed to the rest of the world that such a bomb was possible.

    It is wise not to tear an apparatus down looking for a supposed fault until it is quite certain that the fault exists in that particular equipment. Very often, after a lengthy and fruitless search, the trouble is located elsewhere. For example, a public clock in a university town began striking thirteen every midnight. Workmen dismantled it several times looking fe)r the trouble, but without success. It finally developed that an undergraduate with a room opposite had provided the thirteenth stroke with the aid of an air rifle.

    3. Use the most efficient method of detection. It is obvious that searching for a lost milligram of radium in a dump may be practically hopeless if carried out by eye but perfectly feasible if a Geiger counter is available. A life raft at sea may be truly lost except for the fact that it possesses a radio transmitter on which bearings may be taken. Finding a needle in a haystack isn't too difficult if the hay can be passed along a conveyor belt under a powerful electromagnet.

    The first principle-know the properties of the object-provides the basis on which to choose the method of detection. There is one qualification, however, to rule 3. The cost in time, money, and effort of providing a more sensitive detector has to be considered. It may be quicker to use a less efficient system rather than to spend a long time building a better one. This is one of the innumerable places where good judgment enters into research. If the problem is one which is very expensive and is repeated many times, sometimes mi\thematical solutions can be worked out on which to base such decisions but usually this is not practical.

    4. Be sure you would see the object if it were encountered. This means that tests should be made at frequent intervals to ensure that everything is working correctly and that the sensitivity is adequate. In searching for a spectral line with a microwave spectrograph, for example, most of the time no maxilna will show on the cathode-ray screen. This may be because there is no line in the region being scanned or because the apparatus is not working properly. This is a general ditIiculty in many search prohlems. If possihle, some test arrangement should be available and frequently used to prove that the object sought could be detected if encountered. This test is best if it consists of an artificial object as similar as possible to that being sought. Thus if some other gas with known spectral lines is present in the microwave spectrograph, its lincs can provide the desired assurance.

    If an inspection technique is used in order to find certain dt}fects in a product, it may be a good idea to insert a defective part, occasionally and on purpose, to see whether or not it is detected, provided that arrangements are made to catch it if it escapes inspection. Similarly, if a mass spectrograph is being used to detect isotopic enrichment, it may be desirable to insert some artificially enriched specimens, unknown to the analyst. This is called a positive control.

    The converse of the fourth principle provides the fifth.

    5. Be sure you wouldn't see the object when it isn't there. In other words, the sensitivity should not be so high that false alarms are continually being given. How serious this is depends on the circumstances. All apparatus will show some form of "noise" ,if its sensitivity is pushed too high. In a general sense the function of controls is to guard against false conclusions caused by neglect of this precaution.

    6. Search systematically instead of haphazardly. This may seem trite, but it is certainly not a compelling mode of behavior for human beings. It also assumes that the object of search is not undertaking conscious evasive action. The path of the search should be planned in advance, although the plan should be sufficiently flexible to allow for modification in the light of later knowledge. Those with power to alter the plan should be well aware of the reason for the original choice.

    7. If possible, devise a way of determining the approximate direction and distance of the object at every point of search. Thus Newton's method of finding the roots of an ecluation guides the next step by just such an estimate. Obviously the search plan will be very different in cases where such guidance is possible from what it would be when no indication of the object is available until it is actually found. In the former case, the search path can be guided at every step by the estimates of distance and direction, even if these are very rough. In tuning up an apparatus involving several controls, a problem of searching for an optimum condition in several dimensions, such indications are often available since the increase or decrease of some performance characteristic tells whether one is approaching or receding from a maximum.

    8. In many-dimensional problems it is usually necessary to devise a one-dimensional path. Thus, in searching an area, a path needs to be laid out so that the whole region is covered. Clearly a two-dimensional region cannot be covered by a one-dimensional path of finite length. It is necessary that the path have a finite width. In many scientific problems, the search is a many-dimensional one; thus the tuning up of an apparatus which has four controls is a four-dimensional search. If the object sought could not be detected unless the four controls were to pass exactly and simultaneously through certain precise settings, success would be impossible. It is necessary that it be possible to "see" the object in a finite four-didJensional region along the search path through the four-dimensional space. If this is so, then a path. can be devised so that the whole region can be swept.

    It is obvious that searching becomes progressively and rapidly more difficult as the number of dimensions increases. This is untrue only when it is possible to search each dimension separately and independently, as with an apparatus whose controls each have an optimum setting which can be found without regard to the setting of others.

    There will always be various ways of choosing a path through a many-dimensional region, just as there are many ways of sweeping out an area. It is important that the whole region be covered and that overlapping be reduced to a minimum but not completely eliminated. Further considerations on the choice of path are given below,

    9. If possible, mark the starting point, and record the path actually followed, If this is not done, the search is very likely to lose any planned character and become more or less random, with increasing waste through overlap, For example, in determining the best position for focus with an ultra-violet microscope, photographs are taken at successive positions of the plateholder. It is obviously important to know where each was taken. It is very helpful if all adjustments carry scales; where this is not provided, temporary cardboard scales are useful. If the plan of the search is simple enough and the starting point known, it may be possible to f(>llow the plan without recording the path. However, it is often desirable to record observations made along the way, for example, the values of the ordinate if what is being sought is the maximum of some quantity. Then if progress along the' path produces a decrease in the ordinate, it is important to be able to get back to the best previous position.

    10. Use a convergent procedure. If possible, use a path and method which are bound to close down on the object. For example, in searching for trouble in an apparatus, it may be possible to divide it into two sets of components and test to find out in which half the trouble lies. This set can then be further divided into two parts and tested, etc. When the search needs to be discontinuous and an indication of direction is available, it is usually best to overshoot the target each time and approach it in a zigzag manner. This is analogous to the summing of an alternating series, where the crror is always less than the term after the last one used. Thus in carrying out weighings with a chemical balance, weights of a given size range are always added until the true weight is exceeded, before proceeding to weights of a smaller size.

    11. Searclt tlte most probable place first. This is a consideration to be utilized in selecting a definite search path. The ideal path would pass through the regions of highest prior probability first and then into regions of successively lower and lower probability.

    12. Distribute the available time, facilities, or effort in reasonable proportions in different regions. If the effort available is so small that it can all be put into one region without producing overlapping, it works out in most cases to be best to put all the effort in the most probable region. An exception would be a situation where the efficiency of search varied markedly from one region to another. When greater effort is available, the law of diminishing returns may set in, because of overlapping and duplication, if all the effort is applied to the most probable region. In looking for a collar button there is a limit to the eflort which can profitably be expended on one drawer. Then some should be applied to Jess probable regions. In simplified special cases the optimum distribution can be worked out mathematically, but usually this is not possible or worthwhile. Just as it is possible to put too much of the available effort in the most probable place and lose because the same ground is covered repeatedly, it is also possible to put too little into this region and go on to another too soon. This is particularly bad if changing from region to region is expensive of time, effort, or moncy, so that coming back later to a region previously imperfectly searched is wasteful. If a little more effort will make it practically certain that a region is empty, this certainty should be achieved.

    13. Take into accoullt tlte finite probability of missing the object on passing by it. Do not lay aside one region with complete finality if there is any reasonable chance that the object may still be there even though the space was thoroughly searched. This is a particularly important maxim in connection with the broad search for a solution to a scientific problem. It has often happened that a given approach was tried and abandoned in favor of others, it being found only much later that the first approach was after all the right one. This usually happens because in the meantime some new discovery or technique opens the way. When such new techniques do become available, abandoned avenues of approach should be reconsidered.

    14. Consider any effect the search procedure may have on the search object. This is obvious in military problems: the enemy submarine will take evasive action based on its expectation of the search procedure to be used. It is also true in many purely scientific problems that the search method modifies the situation. For example, in hunting trouble in apparatus by cutting a system in parts, the act of cutting may eliminate the trouble. It may be that there was a feedback loop circling through the whole apparatus. A biologist looking for rare species may easily frighten the animals out of the district. The introduction of check items in an inspection system is easily frustrated if the inspectors can learn the check plans.


    Regresar al inicio

    GENERAL REMARKS

    The reader may find it helpful to consider problems with which he has personal acquaintance in the light of the above principles. The analogies between various search problems of a somewhat abstract nature and the more concrete one of looking for a physical object are instructive, as are the relations with the method of successive approximations applied to the extraction of a root of a mathematical equation.

    In many investigations the results consist of a number of points plotted on a graph. In choosing the experimental conditions which determine the abscissa of the points, it is not always the case that the experimenter has kept clearly in mind just what he is looking for. The best placement of the points will be different for different aims. For example, one set is best to determine the slope of a line, another for the intercept, another for the position of a maximum. As in engineering, the cautious planner will not be content with the bare minimum required for a given purpose; he will introduce a liberal factor of safety. But even with a several-fold factor of safety, the engineer does not merely guess the sizes of beams to employ.

    An example of a search problem is given in the next section.

    Trouble Shooting A tremendous amount of time is spent in seeking the causes of trouble in apparatus. Much of this time is wasted because efficient and systematic methods are not used. The principles of the previous section are nearly all applicable to this problem and can pay handsome dividends.

    As an example which certainly causes the loss of thousands of man-hours annually, consider the search for a leak in a vacuum system. This problem is not only important in itself but is also completely typical of trouble shooting in general; so it should be useful even to those who never deal with vacuum systems. The first principle has already been illustrated with this example. Often the behavior of the trouble will tell whether it is due to a fixed hole (the pressure will rise to atmospheric), to a poorly greased stopcock (the leak may behave erratically as the stopcocks are used), to adsorbed gas (the rate of pressure rise may be less each time the system is pumped out, especially if it is h ated during pumping), or to a material with a high vapor pressure (the pressure will rise to this value and go no higher). TIle apparatus may be such that the chemical constitution of th unwanted gas can be determined or at least a decision made as to whether or not it is air. The rate of pressure rise is important because it is a measure of the size of hole, which determines the sensitivity required for the test procedure.

    The second principle obviously applies here. Is there really a leak, or is the trouble due to some other cause? Perhaps the pressure gauge is out of order. For example, in one elaborate apparatus several days were spent looking for a leak which finally turned out to be in the tip of the pressure gauge (McLeod) and not in the apparatus.

    Some methods of locating leaks are certainly enormously better than others. For glass parts the spark tester is usually best, although it can punch new holes in thin places. The real problem is with metal apparatus. The best (but very expensive) method is probably the special mass spectrograph, which detects helium. The helium is sprayed over the outside of parts of the apparatus and the detector watched for an indication that the helium has entered the system. This major instrument is generally regarded as worth the unfortunately large investment when cyclotrons, commercial vacuum systems, etc., are used. Recently other valuable leak detectors have become commercially available.

    A simple device consists of an acetylene bumer with an aspirator device for drawing air from the neighborhood of suspected leaks. Freon under slight excess pressure is introduced into the vacuum system. If it gets into the aspirator through a leak, a distinctive color change takes place in the flame, If a continuous-reading vacuum gauge is available (such as a Pirani, thermocouple, Knudsen, ionization, or Phillips), leaks can sometimes be found by pumping down and then watching the gauge while hydrogen is sprayed over different places (but watch out for fire and explosion hazard). This works best with thermal-conduction gauges. Also, if a volatile liquid such as acetone is wiped over the leak, enough may be sucked in to produce a sudden pressure rise. Or a low-vapor pressure putty may be pressed over various places and the effect on thc rate of risc of the pressure noted.

    An old stand-by is to watch the color of a spark discharge in the system as ether or acetone is wiped over suspected places. This requires experience, however, and is usually disappointing.

    The fourth requirement, that of sufficient sensitivity, would require that the experimenter know that his test method would find a leak of the given size. If he does not know this from past experience, he should introduce an artificial leak, say at a drawn off tip of glass tubing or a hole punched with a leak tester. This should be of a size to double the rate of leak and then can be used to check the test method. It is seldom in this application that oversensitivity is a problem, although it could be in continuously pumped systems where only large leaks are of importance. In these, time might be wasted on small leaks.

    The sixth principle-search systematically-is the most important of all. The apparatus should he mentally divided into parts and the various possible places for a leak catalogued. Whatever procedure is used should be systematically applied to each place according to some definite order or search path. No place should be overlooked, and duplication should be postponed.

    It is not easy to tell the direction of a leak, but sometimes there are ways of closing off parts of the apparatus and testing the parts separately, thus localizing the trouble. Tests on the parts should be very carefully carried out; much time is wasted by jumping to conclusions at this stage.

    The eighth principle-devise a one dimensional path-is fairly trivial here. Likewise, unless the , pparatus is very large and complicated, the ninth rule-record the path-can usually be done mentally, but it is .important that it be done so that no place is overlooked.

    The use of a convergent procedure is important where successive subdivision into parts is possible. By dividing first into half, then half of the half, etc., the trouble will usually he localized more rapidly than if all the possihle cuts are mi1de at once and each part tested. In stuhhorn cases, it is often necessary to do some glass blowing am} cut off parts that way.

    In planning a path, the eleventh principle of choosing the most probable places first is important but re
    The twelfth and thirteenth principles go together. A leak may be missed the first time over and yet found on a second or third try at the same place. Therefore overlap may be important. Judgment is thus required in spreading the effort suitably over the most probable places, taking into account also the relative ease and certainty of searching different kinds of places. Thus the leak tester on glass is so easy and relatively unambiguous that it should always be used first on the glass portions, including hidden parts normally covered with clamps.

    If a trouble app ars in a previously satisfactory apparatus, some factor must have changed. Thus, if a vacuum system which was once tight develops a leak, stopcocks can often be responsible. However, this rule can appear to conflict with the principle of isolating a trouble by cutting the apparatus into parts. For example, a difficulty arose with a system consisting of a piezoelectric gauge, vacuum-tube amplifier, and cathode-ray oscillograph. This system had a long record of successful performance but showed troublesome drift on one occasion. The above argument pointed to the amplifier, which was a new one; but the method of cutting into parts indicated the gauge, since the trouble disappeared when the gauge was disconnected. Ultimately it was found that the gauge responded to changes of temperature but so slowly that the resulting signals could not pass through the original a-c amplifier. The new amplifier responded better to low-frequency signals and thus amplified the temperature drifts.

    The reader will readily translate all of the above remarks to fit his own problems. Considerable experience shows that a rational approach to trouble shooting is far more effective than the usual procedure of playing random hunches. In fact, the development of habits of orderly and rational procedure is one of the signs of scientific maturity in an investigator. The experienced worker lets his brain help his hands.

    One additional principle is useful with apparatus troubles. When a trouble has been located, record in an accessible place the detailed and distinctive symptoms by which this trouble can be recognized if it occurs again, the best known method of locating its position, the cure which worked before, and any remarks deemed useful, such as afterthoughts on better ways of doing the job next time. Enormous amounts of time are frittered away because experience of this kind is not effectively passed on from one person to the next. On an apparatus of sufficient complexity, especially if used by a succession of operators, a book can profitably be kept just for recording its "diseases." With modern electronic equipment, for example, quite large tomes dealing with their pathology are possible.

    Another maxim is: When a trouble is found and cured, be sure the cure is a permanent one. It is inexcusable to patch it up just well enough to hold out until the next shift takes over. Trouble shooting is made easier by certain design features, such as accessibility, demountability, adequate instrumentation, provision for test points, etc.

    Getting The Most out of Observations

    When Sherlock Holmes mystified his friend Dr. Watson by his amazing deductions, he was utilizing to the fullest degree the data available to him-the muddy boot, the ash of a cigar, the torn ticket. This is a practice which is also basic in the highest forms of scientific research. It might seem impossible to learn very much about a star, a mere point of light in the sky. Yet by assiduously utilizing all the information available, astrophysicists have deduced for many stars the distance, mass, velocity, size, period of oscillation, magnetic field, temperature, and some infonnation about chemical composition. Most of this information comes from stellar spectra. The presence or absence of spectral lines and their relative intensities yield information about the elements present in stellar atmospheres and the temperature. The doubling of lines indicates the existence of binary stars and gives information on their rotational motion. Even the shape of some stars forming pairs has been inferred from the variation of light intensity with time as one star eclipses the other.

    Whatever the nature of the phenomena being observed, it is very likely that it is capable of yielding still further useful information, even without changing the conditions of the observations. By changing the conditions, still greater opportunities can be realized. It is usually worthwhile to observe the effects produced by changes in variables which can be conveniently varied. If theory indicates that changes in some more difficult variables might yield useful information, it may be desirable to go to considerable trouble to produce these changes.

    Related is the importance of bringing the most powerful experimental aids to observation to bear on the problem. The recent progress in the study of filterable viruses illustrates the value of using new techniques, such as the electron microscope and the shadowing of specimens with gold.

    Not only should the experimenter be constantly watching for observational features which can be used in solving the problem in hand, but also he should be on the alert for new phenomena of all kinds. The discovery of penicillin has publicized the importance of the chance discovery by an alert, inquisitive, and trained scientist. The history of science is full of such instances.

    It must be said, however, that for every odd occurrence observed by scientists which leads to a fundamentally new discovery there are thousands which are ultimate ly explainable as due to apparatus defects, errors, or quite well-known laws. The overconservative scientist will pass by the new phenomena; the overenthusiastic one will endow commonplace events with world shaking interpretations. Herein lies one good reason for understanding an apparatus and its quirks thoroughly. The momentary signal on the cathode-ray screen may be worth a Nobel prize; but it may also be the thermostat turning on the room heat.

    The chance discoveries which have meant so much to science and to mankind have generally been made by men who were well trained in the special field involved. It is too much today to expect an outsider to know what is easily explained and what is novel, or to follow up and get to the bottom of a phenomenon that is new.