Page 14 - Instrumentation Reference Book 3E

Introduction

1 Techniques and applications tion about the mean, which is zero if there is no
systematic error.
We can look at instrumentation work in two This implies that we should quote errors based
ways, by techniques or by applications. When we on a certain probability of the whereabouts of
consider instrumentation by technique, we survey the true value. The probability grows steadily
one scientific field, such as radioactivity or ultra- wider as the range where it might be also grows
sonics, and look at all the ways in which it can be wider.
used to make useful measurements. When we When we consider a measurement chain with
study instrumentation by application, we cover the several links, the two approaches give increas-
different techniques to measure a particular quan- ingly different figures. For if we think of possibil-
tity. Under flowmetering, for instance, we look at ities/impossibilities then we must allow that the
many methods, including tracers, ultrasonics, or errors in each link can be extreme and in the same
pressure measurement. This book is mainly appli- direction, calling for a simple addition when cal-
cations oriented, but in a few cases, notably pneu- culating the possible total error. On the other
matics and the employment of nuclear technology, hand, this is improbable, so the “chain error” that
the technique has been the primary unifying theme. corresponds to a given probability, e,, is appreci-
ably smaller. In fact, statistically,

2 Accuracy
e, = de: + e: + . . .
The most important question in instrumentation
is the accuracy with which the measurement is where el, e2, etc. are the errors in the different
made. It is such a universal issue that we will talk links, each corresponding to the same probability
about it now, as well as in the individual chapters as e,.
to follow. Instrument engineers should be skepti- We can think of “influence quantities” as the
cal of accuracy claims, and they should hesitate to causes of random errors. Most devices that measure
accept their own reasoning about the systems a physical quantity are influenced by other quanti-
they have assembled. They should demand evi- ties. Even in the simple case of a tape measure, the
dence, and preferably proof. Above all, they tape itself is influenced by temperature. Thus, a tape
should be clear in their own minds about the level measure will give a false reading unless the influence
of accuracy needed to perform a job. Too much is allowed for. Instruments should be as insensitive
accuracy will unnecessarily increase costs, while as possible to influence quantities, and users should
too little may cause performance errors that make be aware of them. The effects of these influence
the project unworkable. quantities can often be reduced by calibrating under
Accuracy is important but complex. We must conditions as close as possible to the live measure-
first distinguish between “systematic” and “ran- ment application. Influence quantities can often be
dom” errors in an instrument. “Systematic” error quite complicated. It might not only be the tempera-
is the error inherent in the operation of the instru- ture than can affect the instrument, but the change
ment, and calibrating can eliminate it. We will in temperature. Even the rate of change of the
discuss calibration in several later chapters. Cali- temperature can be the critical component of this
bration is the comparison of the reading of the influence quantity. To make it even more complex,
instrument in question to a known “standard” we must also consider the differential between the
and the maintenance of the evidentiary chain temperatures of the various instruments that make
from that standard. We call this “traceability.” up the system.
The phrase random errors implies the action of One particular factor that could be thought of
probability. Some variations in readings, though as an influence quantity is the direction in which
clearly observed, are difficult to explain, but most the quantity to be measured is changing. Many
random errors can be treated statistically without instruments give slightly different readings
knowing their cause. In most cases it is assumed according to whether, as it changes, the particular
that the probability of error is such that errors in value of interest is approached from above or
individual measurements have a normal distribu- below. This phenomenon is called “hysteresis.”