Professional Documents
Culture Documents
Application of Statistical Methods in Chemical Engineering
Application of Statistical Methods in Chemical Engineering
ciples were clearly understood twenty ments that were already well along when
The normal law of error the book was being written do not ap-
years ago), and that new statistical tools
should be forged, more closely fitted to pear. S e w work on error rates and
chemical engineering situations. These ations. T h e work of Box and his associ- allowances-mainly by Tukey and his
contentions may well be true. Much ates (3, 6, 8 ) is finding increasing appli- associates (27)-on fractional replication
can be learned, however, by using the cation, especially in bench and pilot of factorial designs (7, 14) and, on the
available and tested methods. Their plant work. fitting of response surfaces (3-6) are in
lack of modernity does not mean that O n the other hand plant scale tests the writer’s opinion not fully treated
chemical engineers have used them with their great expense, limited range in this volume. But it is quite safe to
widely; indeed their age has been at- of variation of factors, and high vari- say that engineers who have mastered
tained in occasional and cautious ap- ability, are usually best run as “fac- the contents of this single book will gen-
plication. torial designs.” An unexpected ad- erally know what to read next, and they
Some parts of statistics apply with vantage of these designs is that they can will actually know more about the
about the same cogency to laboratory usually be arranged so that no out-of- field than most engineers who have
scale engineering research, to pilot plant specification material is produced. been brought up on less important texts.
work, to semiworks process development, Statistical Literature. There are now Probably the very next book to be
and to full scale plant tests. This degree available several works written expressly read by the engineer who has studied
of generality is claimed for regression for chemists and chemical engineers by Bennett and Franklin will be “Design
analysis and for the related area- statisticians of rank. Two of these de- and Analysis of Industrial Experiments,”
statistical design of experiments. Oper- serve immediate reference since thev edited by Davies (9). Many new ex-
An experimental plan which uses 1/2 (figure a t left), 1/4 (center), 1/8 (right) of the treatment combinations from a com-
plete factorial design with seven factors, each a t two levels. (Letters denote the factors, subscripts the levels of factors,
shaded squares the combinations to b e used)
amplrs of factorial designs are given In many universities chemical engineer- to add to these the departments at Vir-
and a n excellent summary of the main ing students can take courses in statistics ginia Polytechnic Institute and a t the
points in Box’s surface-fitting program given in other departments. The range University of Illinois.)
takes up a full chapter. The research or of excellence of these courses is so wide Statistical and Scientific Methods.
development engineer who has used or that it lvould seem safer to plan self- There is a feeling among some engineers
read both these works will be able to teaching until some more serious system and scientists that statistical methods
decide responsibly on the extent to which of accrediting is in wide use. are put forward as a substitute for other
statistical design can aid him in his work. Something can be judged by the methods usually referred to as “scien-
Some will wish to extend their insight quality of texrs used in statistics courses tific.” This feeling is admittedly re-
into the designs proposed by reading for engineers. If either of the two longer enforced by the scientific limitations
some of the historically important papers works praised above or Hald‘s book (73) of some of the proponents of statistical
and books. The leading papers are by is used. and if the instructor has himself methods. The defective integration of
Fisher and by his colleagues, especially some experience beyond the field of qual- modern staristical ideas with the general
Yates. Forty-three of Fisher‘s papers ity control in applying statistical meth- scientific program has resulted in the
are available in ( 7 7 ) Yates’s pamphlet ods to chemical engineering problems, more-or-less ingroxvn development of a
“Design and Analysis of Factorial then the course is in all likelihood a sizable bolus of material that now must
Experiments” (24), readable by chemi- recommendable one. But it appears- be accepted or rejected as a whole.
cal engineers with average mathematical a t least in the East and Midwest-that There is little content to the argument
background. A good summary of the these conditions are not met by many of often heard from scientists bvho are not
underlying distribution theory is to be the courses offered. statisticians to the effect that they prefer
found in Kempthorne’s book ( 7 4 ) . An interesting interim effort to bridgr exact methods and hence will have
Some engineers and chemists, finding the gap between demand and supply of nothing to do with statistics.
the ideas of statistics unfamiliar and serious statistical teaching is being made The real decisions to be made are :
abstruse, may wish to start their reading by some schools in the form of short a. Between archaic and modern sta-
,,.
with Youden’s little book (25). Written “intensive” courses. These courses tend tistics
by a chemist who has himself made to become propaganda sessions if they 6. Between less and more efficient
major contributions to statistical meth- are shorter than a week, and are some- methods of collecting data
ods, this work has been criticized only times merely extended exhortations when c. Betlveen judging by fashion and
as being perhaps too persuasive. longer. The minimum course from current practice and judging by taking
An optimist might report cheerfully which something can be gained by a the time to make a more logical ap-
praisal
on the prospects, the trends and the op- willing chemical engineer is: in this
portunities in statistical education, but ivriter‘s opinion. a 6-week course a t one Each of these choices has been stated
the record of current activities is short. of the major statistical centers of the ro make clear which is the desirable alter-
A term of statistics is now required in the country. (hlost statisticians would judge native. Scientists can hardly maintain
course work of undergraduates in chemi- the statistical departments of the State the convention that whatever they are
cal engineering a t Cornell. As far as College of North Carolina, of Iowa Stare noiv doing is scientific and that all elsr
the writer knows, no other school has University, and of Princeton. Columbia, is not scientific.
such a requirement. A similar elective Stanford. and the University of California Many introductory- texts in physical
course a t Princeton is well attended. as in this category. hlan)- ivould ivant chemistry. elementary physics, and ai)-
These devices, most of them by no The claim made here is only that such where the constants 63, 61,and bs are to
means new, are discussed because so variation is necessary, not that it is be estimated from the data. I t is further
few chemical engineers are aware of their sufficient. assumed that there is only a certain
advantages. The improvements are Perhaps the simplest example of (known) range of X I and x2 in which ex-
aimed a t the deliberate elimination of the statistical design of a multifactor experi- perimentation is practicable. The ex-
two disadvantages of ‘.unplanned” data ment can be constructed to answer the perimental conditions (but not the results
just outlined. Data will be taken, then, (rhetorical) question, if we vary more of the experiments) can be plotted on
where they d o the most good for detecting than one factor a t a time, say XI and x ? , an XI, x2 grid (Figure la), inside or on the
influences or effects and so that there hoxv can we tell Lvhich produced the practicability boundaries.
are no correlations (or minimal ones) be- observed variation in J ? It is curious The “classicist,“ varying only one
tween the independent variables. that this question \vas not raised in factor a t a time, will d o the threr
Balanced Schedules of Experimental connection with the gleaning of in- runs indicated by the three dots in
Runs. Deliberate variation of factors formation from planr data since it is Figure I b . Comparing this “design”
(independent variables) is more costly only the rarest of accidents to find single with that indicated in Figure I C suggests
than is the unselective taking of data. and independent variation of two factors that the latter might be somewhat better
This commonplace observation should under normal plant operation. 41- since it seems to span the available ex-
he followed by the more important though the question asked above is perimental region a little better. Cal-
statement that it is only by the deliberate usually not meant to be answered, it is culation confirms this suspicion, thr
variation of factors that the engineer exactly in its careful answer that we find slope 6 1 being determined just as well b?
can become sure that observed correla- new gains in the power of variable data the symmetrical design, but 6 2 being de-
tions are in fact in causal relation to one to give information about effects. termined with somewhat better pre-
another. The more experienced and Suppose that the “response surface” cision-namely, with three fourths as
pessimistic operating engineer would is a plane to a close enough approxima- large a variance. This gain, though
want to add several more comments, one tion over the practicable range of varia- small, becomes proportionately greatei
being that even after such deliberate tion of 1 1 and xz. The equation is then as more factors, and more constants.
variation we are often not sure enough. 1- = 6n + 61.~1 + 62x2 (5) are included in the prediction equation.
It need not be overemphasized that
the three points in the second design are
kx2
/ y3
-----
J
/
/ kx27 ----- --_
4
crampedness that was only just notice-
able in the two-factor case is here rathei
painful. No one would claim that the
four points of Figure 2a cover the per-