Measurement Techniques, VoL 43, No.

6, 2000

GENERAL PROBLEMS OF METROLOGY

AND MEASUREMENT TECHNIQUE

ACCURACY PROBLEMS AND SCALE THEORY.

INTERPRETATION OF MEASUREMENT ERRORS

L. Z. Dich UDC 62-533:389

The possibility is considered o f applying scale theory (representative theory) to the description o f
measurement accuracy. It is proposed that the measurement error should be considered as an independent
object which is described by constructing its own scale which may not cofl2cide with that constructed in order
to describe the object o f the measurement itself. It is noted that the thesaurus used in metrology needs to be
extended.

A comparison was made in [1] between the approaches used in scale theory (the representative theory of measure-
ments) and in classical metrology for describing measurement procedures. An analysis is now made of the possibility of apply-
ing the methods of scale theory to a description of measurement error.
M e a s u r e m e n t E r r o r and the Formalism of Scale Theory, TlJe correctness of a representation constructed in scale
theory when formalizing a measurement procedure implies, in particular, that it is unique and that each represented state corre-
sponds to one and only one image number. Actually, however, any measurement algorithm which is practically implemented
cannot guarantee such uniqueness. Moreover, even if such a condition were observed then, for example, the nonlinearity of rep-
resentation which unavoidable exists in practice would prevent one from speaking of the construction of an isomorphic or homo-
morphic image. The total effect of these factors leads to a situation in which a measurement procedure creates a nonunique
image of reality, i.e., according to [2] a state a of a measured characteristic x is represented in sets of values of an abstraction
region: a ~ B a = {b}, B a = {b}, where {b} is some interval (subset) of numerical values. Since in [2] the property of isomor-
phism of representation is initially postulated, its author is unavoidably forced to ask whether the postulate corresponds to things
as they really are. Here it is mentioned that indeterminacy of representation can in principle be made compatible with the for-
mal requirement of isomorphism, but only for an extremely specific case in which one is essentially concerned with the repre-
sentation of a countable set of states. The question of the representation of a continuum, which precisely relates to the over-
whelming majority of measurements, remains open.
In this case the fact of the matter is that one cannot ignore the existence of measurement error as part of the proposed
formalism and that in order to utilize this concept it is necessary to find logical foundations which are no less weighty than those
for the measurement concept itself.
If one poses this question in the way it was done in [2], i.e., understanding the measurement process to be a process of
nonunique representation, then it will be necessary to admit that a quantity Q is represented into a set of value-error pairs {x, 6x}
or into a set of intervals {x', x"}. For example, instead of a representation r : Q --->R+ it is necessary to construct a representa-
tion r : Q ~ R~. Here there are two objections. Firstly, the inclusion of an error in a numerical model describing the proper-
ties of an object is not justified since it in no way characterizes the properties of the object but characterizes only the measure-
ment process itself. Secondly, nl-dimensional and n2-dimensional sets can be isomorphic only if the condition n 1 = n 2 is satis-
fied. Consequently R+ and R 2 cannot be isomorphous and neither can an isomorphism relationship be established between
Q and R~. Thus, neither the set of pairs which was mentioned nor the set of intervals can serve a satisfactory model of reality.

Translated from Izmeritel'naya Tekhnika, No. 6, pp. 3-9, June, 2000.

0543-1972/00/4306-0461 $25.00 9 Kluwer Academic/Plenum Publishers 461

 Taking into account that "a measurement is not an aim in itself but only a means of achieving the aim" [3] the exis-
tence on nonuniqueness of representation forces one to ask the questions:
1. Is one justified in attempting to construct a formal model in which the properties of an object are represented not in
number but in numerical intervals?
2. If one is not justified in doing this, is there any sense at all in invoking scale theory in order to describe measure-
ment errors even taking account of the advantages it gives when describing such measurements?
3. If one is justified, how can one still correctly introduce the concept of error?
If it turns out, for a positive answer to the first question, that it proves to be possible to construct the model being
sought, then the rationality of the succeeding possible steps is doubtful. The fact of obtaining some number as a result of a mea-
surement procedure very frequently proves to be only the first step in a chain o f further actions which, in particular, can result
in calculations or in the generation of control instructions. It is very difficult to imagine a situation in which, for given purpos-
es, one would employ not a single result number but a whole numerical interval. The latter, at the very least, would result in a
sharp increase in the volume of the calculations and would additionally cause considerable difficulties when solving poorly con-
ditioned problems and problems in which the result of the calculations does not depend in a continuous way on the value of the
measured quantity.
It makes sense to consider a representation model, such as a scale, only in interrelationship with other abstract models
which utilize numbers. Any anaiyticai dependence into which the measured value of a quantity is inserted will adequately
describe reality only to the extent that it itself depends on another model, on a numerical image of the object. It is for this rea-
son that, however rough the measurements, in the calculations one uses only their result itself as though it had been obtained as
part of an ideal representation. The final accuracy of the calculations or control actions is estimated separately. Thus, it is not
very productive to seek a correct description of errors by constructing nonunique representations.
In order to answer questions 2 and 3 let us turn our attention to the difficulties which arise when determining the error
of a number of types of measurement by classical methods.
In the case of an automatic calculation of the result z of indirect measurements, z = z ( x 1, x 2. . . . . xj.) it is impossible to
restore the original values of the input quantities x i w i t h which the measurement occurred or the calculation was made [4].
Meanwhile, one and the same value z can we obtained for different sets of vaiues (x 1, x 2..... xj) which correspond to different
error values ~z. In this case it is impossible uniquely to compare any accuracy with the result z despite the fact that the errors
~x i in each case can be indicated uniquely. On this basis it was even proposed that one should not at all class corresponding
measuring instruments as means of measurement. The diametrically opposite opinion was expressed in [5, 6] where it was pro-
posed to consider an instrument (for example, a wattmeter) as a means of direct measurements if all the calculating operations
are performed within the instrument while the operator taking down the readings is not involved in the calculations but records
only the final result. Nevertheless, the problem of determining the error ~z of the readings from such an instruments does not
disappear when this is done.
Another example is associated with a cited problem of monitoring the shape of a surface. It is well known [7] that
when describing the inaccuracy of manufacturing a surface having some given nominal shape, one indicates the distance from
points on the real surface to some reference surface which is constructed in accordance with some or other fixed rules and whose
shape coincides with the nominal shape. When using this method to estimate the shape, it becomes impossible to distinguish
between a distortion of the actual shape and a displacement of this surface in space without a distortion of its shape.
In optical manufacturing, when monitoring the shape of optical surfaces, test glasses are utilized as a standard surface,
but it is effective to use them only when the radii of the tested and standard surfaces are practically equal. "In addition, in this
case also they resort to concepts of general and local errors of shape, giving a pair of interference fringe numbers (N, AN). This
description can be informative only in combination with the observation of the interference pattern whose shape contains addi-
tionai albeit qualitative information. It is very difficult to judge the quality of manufacture of a surface by relying solely on this
pair of numbers and such monitoring o f the shape of a surface requires a detailed description of the rules governing its behav-
ior. In particular, the form of the interference pattern (rings, fringes) and the number of fringes observed can be stipulated. This
however is an additional veiled normalization.
A variation in monitoring the shape of a surface is that of making multiple measurements of the diameter of a cylin-
dricai or sphericai object in different directions. Thus, the assumptions are introduced that circles or spheres are unique geo-
metric objects whose sizes are constant for any direction of measurement. However, one can contrast so-called c o n s t a n t thick-
n e s s curves and bodies with these objects. The simplest example is that of the Reuleanx triangle and the corresponding body of

462
 Fig. 1. Reuleaux triangles as an example of figures of constant width.

rotation (see Fig. 1) where the width (the distance between two parallel straight lines or planes in which a plane figure or its
body of rotation is inscribed) is the same in all directions and equal to R 1 + R 2. Generally speaking, the number of types of such
objects is unlimited [8]. W h e n their transverse dimensions are measured in different directions, the results of the measurements
will be the same as in the case of similar measurements of the diameters of a true cylinder (sphere). This is a fact well-known
to mathematicians but ignored by metrologists [5, 6] despite the existence of publications concerning it [9]. This example par-
ticularly vividly indicates that estimates of the accuracy of the shape of a surface based on measurements of the linear dimen-
sions of a body bounded by it may in the best case be considered to be palliative but are in no way an adequate solution of the
given problem. An error in shape simply cannot be described in terms of linear dimensions.
More complex is an example with coordinate measurements. Let us indicate the following difficulties:
-the accuracy characteristics of different coordinate measuring machines are very poorly amenable to comparison with
each other;
- the results of measurements obtained using coordinate instruments do not always agree with those of measurements
of the same objects using instruments of a different type;
- it is frequently impossible to make either an a priori or an a posteriori determination of the error in the final results
of measurements obtained using a coordinate instrument;
- the error in measurements of identical objects using a coordinate instrument is substantially dependent on the choice
(and in certain cases also on the tracing sequence) of the points at which the coordinates are recorded; their choice in the liter-
ature therefore involves the concept of a measurement strategy which is not defined and is not legitimized by any of the known
authors of normative documents.
These difficulties result for reasons of a fundamental nature which amount to the following. If one speaks of mea-
surements of geometric quantities using coordinate instruments and seeks the original source of these errors in an inaccurate
determination of the coordinates then the error in determining the coordinates can always be considered to result from some cen-
troaffine coordinate transformation. Such a transformation is decomposed into the sum of a symmetric and an orthogonal trans-
formation. The error in determining the coordinates is then also decomposed into two terms corresponding to the types of trans-
formation. However, it is well known that only the symmetric transformation influences the result of the calculations of the geo-
metric parameters, but it is difficult to separate out the part of the error corresponding to it from the result. Therefore the error
ascribed to the coordinates can characterize the measurements only to the accuracy of the term associated with the orthogonal
transformation and cannot give an objective representation of their quality [10].
A second important factor is that even if it were possible to take account of the action of the symmetric transformation
separately one would be required to specify six (!) independent components of the corresponding matrix in order to find the error
vector Ax from the coordinate vector x whereas only three independent parameters are indicated when utilizing the adopted con-
cept of coordinate error. Thus, the projection of the concept of measurement error onto that of coordinate error fails to yield
fuU information for describing the measurement error. However, if one speaks of utilizing the components of the symmetric
transformation for this purpose one encounters the need to describe not only and even not so much the error of the actual num-
bers obtained (sets of coordinates) as the error of the model (representations, scales), i.e., the coordinate system as a whole.
The significance of the model in the context of the problem described has been discussed by many research workers
[2, 3, 5]. In its importance for metrology it should probably be considered on a level with the concepts of measurement and
quantity. However, it can be seen from the above discussion that not one but two models must participate in the measurement

463
process. These models are: the factor set of the images (if you wish, the models) of the objects (we shall call this and the actu-
al model of the object model A) and we shall call the other abstract (numerical) set model B. As has already been remarked, a
correct measurement procedure implies the equivalence of isomorphism of these models. In the majority of cases the main atten-
tion of metrologists is directed only to investigating the first of these. However, when the second is raised the discussion is usu-
ally only about numbers, it being the role of the numerical descriptions which can even be specially accented [11]. Moreover,
the measurements are associated only with the set R+ [2], the use of the numbers, for example of a natural series (i.e., counting)
is assumed not to be related to the measurement procedures [5]. Nevertheless, high-rank normative documents [12] exist in
which the concept of an integer is extremely important for the measurement algorithm. If, however, one is concerned with mod-
em methods of processing measurement information one can utilize the so-called data representations in systems of residual
classes [13] where the role of integers becomes most important.
However, despite the fact that the concept of a real number remains universal and predominant when modeling a hugh
circle of phenomena, it is by no means a natural abstract representation utilized in order to represent reality. It is sufficient to
cite the description of oscillatory processes using complex numbers and observations of Lissajou figures using an oscilloscope.
If we encounter the fact that when determining measurement errors it is necessary still to consider the properties of
model B, the problem passes into another plane.
In legislative metrology [ 14] an error is taken to refer to the result and only to the result of measurements of a param-
eter of an object model and to describe it in terms of the measured quantity. In the language of scale theory this denotes the
construction of the representation r : Q ~ R~, but we are convinced that such representation may prove to be incorrect.
However, by operating with two models it is possible to propose an aitemative and to attempt to relate an error not to the result
(model A) but to model B in which this result was obtained. The question can now be formulated as: does the model in terms
of which the results are interpreted provide a description that is sufficiently full that it can be considered to be adequate?
In other words, in order to describe an error one can choose the following example of discussions. In terms of scale
theory a measurement is considered to be a process of comparing numbers of the properties of objects. A property of an object
is made to correspond to one and only one number. It should be remembered that here we are not at all being specific about the
category of the objects described (!) and are not utilizing the concept of quantity. We are therefore justified in considering the
scale model itself and each representation obtained in practice using some or other instrument to be an object in its own right
on a level with other objects and to describe their properties utilizing all the same scale-theory formalism. We can consider the
error itself to be such an object. Then, if we constructed a scale r for measuring an individual parameter, another scale qb which
is already secondary with respect to the scale r but which is an isomorphic representation of a set of objects of the error type
can be constructed in order to describe the properties (the error) of the scale r. The difficulties associated with the incorrectness
of the representation r : Q ~ R~ are thereby eliminated.
Since as a role as part of any model there are no means for describing its own imperfections, the secondary scale will
unavoidably be more complex than the initial one, since the initial scale will be only a particular case of it. We must then con-
sider the error to be not the discrepancy between the obtained result and the true value of the measured parameter but a con-
struction describing the discrepancy between the models. The correct description of the error must then be constructed not in
terms of the initial model (the measured quantity) but in the terms of the secondary model which follow from its specific char-
acteristics. Following [3] let us refer to a representation constructed in order to describe an initial representation as a metamodel
or a metascale.
The following scheme can be proposed. Let instruments of a definite type be produced (used) in order to measure a
quantity Q. Each instrument constructs its own representation r" which differs from the ideal abstract representation r. By
describing the properties of the instrument we must already study not the object with an attribute Q but the representations r and
r'. However, these representations belong to another set which cannot be described by the model A and we must again pass
through the chain of steps previously indicated in Fig. 1 of [1], having changed the initial set and having obtained a new abstract
model C.
An dement of model C which will be placed to correspond to the representations r and r" can, for example, take the
form of a function. The whole assembly of instruments of a single type is then described by a set of functions in functional
space and such a description possesses greater completeness. In order to provide for the possibility of correctly ascribing the
error to any measured result, it is necessary to introduce into this functional space a norm (metric) [2, 15], by a method which
most of all corresponds to the aim of the measurement, and to utilize this norm as a characteristic of the error of this type of
instrument, already comparing it then, for example, with each measured result using each of the instruments of this type.

464
 As regards simple forms of measurement, such an approach is entirely in agreement with the existing methods of nor-
malization and does not require them to be fundamentally reconsidered. However, it can prove to be extremely effective when
overcoming difficulties with the description of errors in the cases described above.
Let us consider the measurement of power using a wattmeter. W h e n we attempt to describe such a measurement we
construct a scale (representation) r : W --~ R In terms of this representation we are not able to distinguish whether the method
of measurement is direct or indirect, they both implement the same scale. However, when describing the error of measurements
we cannot abstract ourselves from the method implemented in the instrument and must take it into account. In this case we must
first consider the representation r : (U, I) ~ R~, although it is not constructed in explicit form, and then the representation
R +2 ~ R + ~-- W. In the general case, a sequence of representations is constructed: r : (Q1, Q2 ..... Qn) -'~ Rn --~ R+ +-- Q. Since
the representation R n --~ R+ can be isomorphic only for n = 1, information is unavoidably lost at this stage. In this case we are
fundamentally unable to compare uniquely the result number with the error number if intermediate information concerning the
values of the quantities QI, Q2 ..... Qn is inaccessible. One correct method for describing the error in the case of a wattmeter
could be to compare the representations r : (U, I) ~ R~l w= const and r ' : (U, 1) ~ R+I w= const in terms of the metamodel, the
functional space, and to determine the distance between the corresponding functions along the curve W = const.
W h e n measuring the shape of surfaces, we are mainly interested in the property of the object to have a definite shape
as such, but not at all its property to have definite linear dimensions. Model B, which must be constructed, must be invariant
with respect to dimensions.
In the latter case it is required to compare with the shape of an element not the numerical but the functional sets F spec-
ifying in them some particular shape, i.e., one must construct the representation r : S --) F '~. In particular, the function describ-
ing the surface must be considered to be a vector L 2 in Euclidean or Gilbertian space of functions with an integrable square,
constructed on an orthonormalized basis. The result of a "measurement" r : S --->L 2 will be a set of coefficients giving a repre-
sentation of the function by a corresponding Fourier series. Thus, when describing deviations of the shape of a surface from
that of a circular cylinder it can prove to be convenient to expand in trigonometric functions or Legendre polynomials P,(cos0).
For estimates of sphericity one can expand in general spherical functions Ynm = Pn~(COS0)e ira9 or also in Legendre polynomials.
Polynomials Pn(cos 0) describe, for example, a function specified on the surface of a sphere, the zeroth polynomial giving a
sphere of unit radius. Normalizing the expansion to the radius of a sphere constructed in accordance with a given criterion (in
particular, one can take a sphere having a nominal radius or constmct a root-mean-square sphere), all the subsequent terms of
the expansion can be considered to be the actual error of the shape of the surface, irrespectively of the linear dimensions of the
object bounded by it. In a number of cases the components of the expansion can be interpreted as independent measurements
in the classical sense of a quantity. For example, in optics the aberrations of an optical system (measured in units of length) can
be found as coefficients of an expansion of a wave function in orthogonal Zernike polynomials Z m = Rn"~(p)ejrn0 when process-
ing the interference pattern obtained in a similar way. At the same time, an estimate of the aberration can also be obtained by
direct measurements.
A n estimate of a the error of a shape in vector space of orthogonal functions can immediately also carry information
concerning the spectral composition of the perturbations. This is important for completeness of representation concerning the
quality of its manufacture. For example, mention was made in [7] of measurements of the shape of a real surface at a necessary
number of points. How to find the "necessary" number was not clear. This is another manifestation of veiled normalization,
providing evidence of a lack of completeness of the methods adopted for estimating the error. Having an expansion on the spec-
ified basis one can easily find the maximum distance between the points of the real and nominal surfaces, i.e., one can obtain
the traditional estimate in linear measure. However, the inverse transition is impossible. This approach can also be carried over
to describe dynamic errors.
W h e n utilizing coordinate methods of measurement one constructs a formal representation which places each point of
physical space in correspondence with an element of space R 3. This representation does not describe any definite property but
is of a universal nature. Even when the units of measurement are precisely indicated and a metric is introduced which trans-
forms this space into Euclidean E 3, it is not possible uniquely to specify a representation; an arbitrariness remains associated
with the choice of the axes and of the coordinate origin. It then becomes necessary to separate that which is associated with the
inherent geometry of the studied object and the internal geometry of the coordinate system implemented in order to perform the
measurements from that which is introduced by the random choice of the coordinate system. In such a formulation, the prob-
lem coincides with the so-called fundamental problem of tensor calculus. If one has in mind the classical measurement of geo-
metric quantities and speaks of the comparison between a representation which was conceived as an ideal and one which proved

465
 TABLE 1

Described category Scale Metascale

Quantity R+ Functional space F

Object shape, Space C of summed functions, Metric functional space, set of coordinates in functional
functional dependence metric functional space space and of Hermitian matrices Hij

Space point Vector space, metric space Metric space. Riemannian manifold

to be capable of realization, the single objective basis for such a comparison is provided by a metric tensor describing the rules
for calculating the lengths along the coordinates. The structure of the metric tensor determines the characteristic features of the
coordinate system which exist in order to find dimensional parameters and it makes it possible to compare the "true" and "dis-
torted" representations with each other [ 10].
Such a comparison, however, is not capable of amounting to either a comparison of scalars (coordinates considered inde-
pendently) or of two vectors. Here the measure of the error assumes the form of a metric tensor, a more complex object. If, for
example, it is proposed to construct a representation by introducing an orthogonal Cartesian coordinate system (model B) in
which the metric tensor gij is a unit tensor (g(/= k/j), i.e., is described by a Kronecker delta function and is specified by only one
component, then it can turn out in practice that instead of this an oblique angled coordinate system is realized in which six (!)
parameters are required for fully describing the metric tensor. The oblique angled coordinates will then be a metamodel (model
C) for which rectangular coordinates are merely a particular case. Attempts to normalize the error of coordinate instruments by
specifying three parameters, namely the errors in measuring lengths along the coordinate axes (as is done at present) funda-
mentally lacks the necessary completeness and leads to the above-mentioned contradictions. It is impossible to agree with the
statement in [16] that it is always possible to partition multidimensional quantities into several separately measured
single-dimensional quantities. When calibrating the coordinates of an instrument we cannot "tune" each coordinate displace-
ment converter independently. Correct calibration is possible only by performing a set of measurements and finding all the com-
ponents of the tensor gij" In other words, in this case we are in no position to describe an error o f a measuring instrument flz
the same terms as the m e a s u r e d parameters themselves, namely the coordinates.
If it turns out that an arbitrary curvilinear coordinate system is constructed then one can speak from the start only of
arithmetization (namely of ascribing numbers to points in space) while in no way linking this with the choice of the units of
measurement. Further steps are possible only after introducing a metric tensor and using it to transform to the measured quan-
tities. Model C then takes the form of a still more complex system and is a Riemannian manifold.
It should be noted that the expansion of a function on an orthonormalized basis, given above as an example, is also inter-
preted in terms of coordinate representations, and so the problem of describing the properties of an object independently of the
coordinates also arises here (the role of the metric tensor can then be performed by a Hermitian matrix H~). In the general case
one must be concerned with seeking a definite class M of objects Q(x 1, 2" ..... ; f ) which are functions of a point and are deter-
~lmn: 1 9
mined by the scalar components ~dijk t x , x" ..... x ~) such that the properties of these objects can be described in terms of opera-
tions which are independent of the frame of reference. In the broad sense of the word, these objects can also be called tensors.
Continuing the example with coordinate measurements of geometric parameters it can be mentioned that for sec-
ond-order surfaces, which in the overwhelming majority of cases are shaping surfaces, invariants exist which make it possible
to calculate the geometric parameters of the surface without reference to the specific way in which the equations specifying the
surface in some or other coordinate system are written. In an ideal case it is desirable to have the possibility of comparing such
an invariant both of the model to the object, using the chosen scale, and of the "scale" itself considered as the object when study-
ing the error. When comparing the object of measurement and the instrument used for it will be necessary to compare the invari-
ants and to use them for normalization. In this case the results of the measurements and of the verifications can obtain a unique
interpretation [17], i.e., the possibility appears of distinguishing whether the error is a result of discrepancy between the object
and the model A or a result of imperfection of the measuring instrument. Table 1 gives the relationship between the models
(scale and metascale) for the examples considered above.

466
 Let us turn our attention to the fact that in many ways the appearance of the problem with a unique description of errors
is associated with the possibilities of the "language" in which the description of the measurement procedure is given. The set
of concepts of traditional metrology makes it possible, in the main, to describe measurements of quantities correctly. The set
of concepts of scale theory possesses greater width and makes it possible, when describing measurements, to utilize an exten-
sive class of procedures leading to the construction of the representation itself and to a description of its "quality", namely the
measurement error. Thus, one can speak of the expediency of supplementing the thesaurus of metrology. Bearing this aspect
in mind, let us quote one more formulation defining a measurement as an experiment having the aim offorming pronouncements
on an ilwestigated object in the language of this thesaurus [3]. If the "language of the thesaurus" is taken to mean adequate
abstract models (groups, rings, fields, spaces, etc.) in which the described properties are represented, then we have a definition
of a measurement which coincides with the definition of this concept in scale theory.
The normative base of metrology allows one, unfortunately, to utilize essentially only one type of representation, a rep-
resentation in R+. Measurement practice, however, leads to different types of representations being used in metrology on the
spur of the moment and becoming fixed in the standards postfactum. For example, in the case of high-accuracy measurements
of end gauges [12] representations are used in both N and R. Moreover, here a heuristic procedure of their "selection" is used
when seeking integers which is not provided for in the normative definitions. This once more indicates that the circle of prob-
lems of metrology, even in its most established regions, is far wider than the indicated set of concepts used to describe them.
Expediency of Axiomatic Approaches when Describing Errors. There is yet another aspect of the problem being
discussed. As a result of the universality of metrology, it borrows methods of disciplines which deal with measurements and
becomes extremely heterogeneous in its structure. The examples given can partly serve to confirm this. Consequently the
approaches of metrology in many ways rely on intuitively adopted solutions and common sense considerations. Therefore
numerous attempts have been made to produce general and rigorous approaches for seeking the "original" concepts which are
capable of forming the basis of some "unified" measurement theory. Successes in the deductive construction of many concepts
in modem physics and mathematics lead some metrologists to wish to postulate the initial premises and to utilize axiomatic
approaches for developing a measurement theory. This primarily concerns the question associated with measurement accuracy.
In particular, attempts have been made to postulate the following [6]:
1) a true value of a measured quantity exists (postulate ~);
2) it is impossible to find the true value of a measured quantity (postulate 6);
3) the true value of a measured quantity is constant (postulate ~t).
Any axiomatic constructions are good when they are productive and make it possible to arrive to some significant
results using theoretical constructions. However, it is not possible to make any constructive logical conclusions from this sys-
tem of postulates, not to mention the fact that the validity of the last of them is doubtful.
Another system of postulates which was used as a basis for attempting to construct a universal applied theory of mea-
surements was proposed in [2]. This is as follows.
1. An isomorphic relationship exists between the states of a given characteristic and between the values of the corre-
sponding quantities.
2. The representation of the state of a given characteristic in the image of the state is nonunique.
3. The nonuniqueness of the representation of a state in the image of the state can be established on the basis of a math-
ematical model.
4. The image of reality formed corresponds to certain arbitrarily established standard states.
It was added to these postulates that the key problem of measurement theory is the model of the errors.
Meanwhile it was mentioned above that nonuniqueness of representation is incompatible with the relationship of iso-
morphism, i.e., proposals 1 and 2 contradict each other. This fact alone makes it impossible to speak of the correctness of fur-
ther steps which could be undertaken in constructing a theory. In addition, it is difficult to consider proposals 3 and 4 as pos-
tulates at all since they essentially fall to contain any statements. In fact, both groups of postulates are only declarative state-
ments which in no way relate to the requirements of completeness of a system of axioms (not to mention their noncontradicto-
ry nature) which is obligatory when constructing a rigorous theory, and they can scarcely by considered seriously. Indeed, every-
thing presented in [2, 6] following the postulates fails to rely logically in any way on the postulates.
Let us consider the other remarks made in [2]. It was noted that:
a) a special characteristic of measurement theory is its secondary nature relative to research methods;

467
 O~ ~ ~ I ( : ~ ] ~ e T t i ~ r " I M~ B]

Factor s e ~

RESULT 1

Fig. 2. Relationship between the concepts of scale, measurement, and error in a general
scheme of measurements.

b) the diversity of problems results in a difference in the models of the physical and metrological phenomena (in ter-
minology [2]) and so no unified theory of measurements exists or will exist.
The diversity of the problems solved primarily predetermines the variety of special-purpose measuring apparatus. By
interchanging this apparatus one can make the same property of the same object conform to the elements of the different abstract
models (both A and B). However, the choice of the model is for the present an operation which it is difficult to imagine sup-
porting a formalism, and this forces one to doubt whether a formal axiomatic method can be productively utilized in theoretical
metrology. However, when speaking of the rigor of the approaches used at present, it is clear that one should primarily pose and
solve the question of what should be considered to be the measurement result corresponding to some particular special-purpose
apparatus and what should be a necessary and sufficient set of parameters enabling one to give a full and accurate view of the
purpose of measuring the characteristic of this result.
On the other hand, axiomatic approaches are important since they make it possible to find definite solutions by deduc-
tive means while always dispensing with a more general case. Moreover, many problems which are encountered when describ-
ing measurements arise due to the fact that inductive deductions are used and particular cases are extrapolated to general cases.
It can be seen from this discussion that the tendency is nearly always directed at extrapolating the concepts and methods devel-
oped and justified for measuring quantities to measurements of the kind which do not apply to more general cases. Here there
is a need for a point of view to be generated which while not possessing the irreproachable rigor of an axiomatic theory never-
theless makes it possible to avoid making elementary logical errors.
Returning again to scale theory, it can be seen that it immediately suggests a sufficiently general approach which is sup-
ported by categories axiomatically introduced into mathematics and is applicable to practically any measurement procedure and
to estimating these procedures themselves if one begins to construct it while discarding the concept of representing one set in
another. The problems which we have touched upon in the present section are not so much problems of axiomatics as problems
of the choice of an adequate thesaurus. In summary, the generality sought can be reduced to the following set of proposals:
- an object o f measurement is taken to be a selected characteristic (property) of a subject or phenomenon consisting of
a fragment of reality [2];
- the aim o f a measurement is understood to mean the formulation of conclusions concerning the state of an observed
fragment of reality or the formation of an image of reality which is objective (from the chosen point of view) [2];
- an assembly of objects is considered for the description of whose properties it is necessary to set up a formal descrip-
tion (scale, model B) in accordance with the postulated aim of the measurements;
- in accordance with this same aim a representation of the object is formalized, a model of the object is constructed
(model A);
- a f a c t o r set is constructed from clear definitions of model A which are equivalent from the point of view of the aim
of the measurement;

468
 - establishment o f a scale occurs, this consisting of a speculative searching or construction of an adequate representa-
tion of the factor set referred to the property of the object in an abstract set (for example, a numerical set). Model B is made to
correspond to the specific definitions of model A;
- a measurement occurs, this being a process of empirical construction of the representation (or search for an element
of the representation) of a factor set referred to a property of the object in the abstract set described by model B;
- the empirical construction of a representation is also considered to be a fragment of reality on the same level as other
objects (but having its own nature and not being identical to them[). In order to describe the lack of correspondence of the empir-
ical representation with model B, i.e., the measurement error, its scale (model C) is established (a "metascale" "metamodel");
- the error is described as part of and in terms of the metamodel; the structure and terms of this description may not
agree with the description of the result of the measurements itself;
- the complete characteristic of the result of a measurement (from the point of view of the aim) is considered to be the
assembly of parameters which represents the result together with an assembly of parameters describing its error.
Figure 2 is a schematic diagram of the proposed scheme.
It follows from the discussion undertaken in [1] and in the present work that the principal quality of scale theory which
could facilitate the solution of certain topical problems is not so much the use of specific methods for "ordering" the objects of
a description and constructing corresponding types of scales (nominal, ordinal, etc.), as the fact itself of a transition to the lan-
guage of representations and the possibility of their further study with more general postulates of the theory of models. Much
of what was considered above has already in some way or other been treated in the literature. However, the problem of a gen-
eral description of measurements is too broad and the scope of a single article is too restricted for one to be able to speak of any
complete solution of the problems discussed.
Nevertheless, it can be proposed with a large degree of confidence that if metrology continues to be restricted to its tra-
ditional limits and if only a single numerical model is normatively admitted for use in it, then it risks being deprived of its sig-
nificance as a branch of knowledge which provides a unified view of measurements. This may mean the appearance in each
subject field of "its own metrology" having its own group of concepts and it becomes extremely difficult to compare the results
obtained in different fields. Here, one should again indicate the urgency of strengthening the concepts of scale theory in the
technical normative documentation and the importance of issuing such documents [18].
If, however, one attempts to look at metrology as though it is an applied theory of models having as its aim a search
for and determination of conditions for correctly applying mathematical constructions in order to give a complete and noncon-
tradictory description of measurement procedures and of the accuracy of these procedures, it will be able to lay claim to a place
in a number of fundamental branches of knowledge and occupy one of the leading positions in natural science and modem indus-
trial high technology.

REFERENCES

1. L. Z. Dich, Izmer. Tekh., No. 5, 17 (2000).

2. Ya. Piotrovskii, Theory of Measurements for Engineers [in Russian], Mir, Moscow (1989).
3. V. Ya. Rozenberg, Introduction to the Theory of the Accuracy of Measurement Systems [in Russian], Sovetskoe Radio,
Moscow (1975).
4. P. V. Novitskii and I. A. Zograf, Estimating the Errors of Measurement Results [in Russian], t~nergoatomizdat,
Leningrad ( 1991).
5. M. A. Zemel'man, Metrological Fundamentals of Technical Measurements [in Russian], Izdatel'stvo Standartov,
Moscow (1991).
6. S. G. Rabinovich, Errors of Measurements [in Russian], l~nergiya, Leningrad (1978).
7. V. V. Leonov, Analysis of Methods of Measurements of Deviations from Linearity and Planeness of Surfaces [in
Russian], Izdatel'stvo Standartov, Moscow (1982).
8. D. Hilbert and S. Cohn-Fossen, Geometry and Imagination, Chelsea Publishing (1952).
9. F. G. Rol't, Gauges and Precision Measurements, Vol. 1 [in Russian], OGIZ, Leningrad (1933).
10. L. Z. Dich, Probl. Mashinostr. Nadezhn. Mashin, No. 3, 73 (1995).
11. L. I. Sedov, Thoughts About Scientists and Past and Present Science [in Russian], Nauka, Moscow (1973), p. 67.

469
12. State Standard GOST 8.367-79, Standards of Length of First and Second Order End Plane-ParallelMaster Gauges and
Working Classes of Accuracy O0 and 0 of Length up to 1000 ram. Methods and Means of Checking [in Russian], State
System for Providing Traceability of Measurements (GSI).
13. Yu. N. Solodkin, l~jner. Contr. Avtomat., No. 2, 27 (1991).
14. International Vocabulary of Basic and General Terms in Metrology, International Organization for Standardization,
Geneva, Switzerland (1984).
15. A. N. Kolmogorov and S. V. Fomin, Elements of the Theory of Functions and FunctionalAnalysis [in Russian], Nauka,
Moscow (1972).
16. V. G. Knorring, Theoretical Fundamentals of h~formation-Measurement Technology. Basic Concepts of Scale Theory
[in Russian], Polytechnic Institute, Leningrad (1983).
17. A. Wirtz, Annals of the CIRP, 37, No. 1,493 (1988).
18. Measurement Method/VII 2365-96, Measurement Scales. Basic Postulates. Terms and Definitions [in Russian], State
System for Providing Traceability of Measurements (GSI).

470
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.