CHAPTER 2 oe, el ~ —_ / Static Characteristics of Instruments By static characteristics we mean attributes associated with static measurements or nicasurement of quantities which arc constant or vary very slowly with time. For example, the nieastirement of emf (electromotive force) of a cell or the resistance of a resist constant temperature are both static measurements. Static characteristics of instruments can broadly be divided into two categories—desirable and undesirable—each consisting of a few characteristics as shown in Fig. 2.1. Static characteristics Desirable Undesirable \ Accuracy Sensitivity Drift Dead zone Static error Repeatability 1 Reproduciblity Threshold Hysteresis Creep Resolution Fig. 2.1 Static characteristics tree. 2.1 Desirable Characteristics The nature of these characteristics is discussed below. Jhkecuracy Accuracy deterinines the closeness of an instrument reading to the true value of the measurand. Suppose, a known voltage of 200 V is being measured by a voltmeter and the successive readings are 204, 205, 203, 203 and 205 volts. So, the ac uracy is about 2.5%. Here, though the repeatability of readings is not too bad, the accuracy is low because the instrument may be having a Large calibration error. Hence, th i J + Hence, the accuracy can be improved upon by better calibration of the instrument. : pemveve> Static Characteristics of Instruments : precision iS another terat which is often used in the sane connotation as the necuracy. But in vali precision is different from accuraey, tn the above example, the reading. can be expressed San + 1 V; which means that the precision is a little less than 0.5% in this ease precision is, therefore, related to the repeatabi c ty! of the instrument reading and is a qharacteristic of the instrument itself. To improve the precision of an instrument, its design ruction have to be improved upon. \ y, therefore, if a denotes accuracy, p the precision and ¢ the calibration error thena = Pte Precision of a measurement also depends on what is called the number of significant figures. ‘An example will perhaps make the point clear. Suppose the resistance of a conductor is being measured by an analogue ohimmeter. The ohmmeter indicates the true value, but the observer js able to read the exact value because of lack of graduation beyond a certain number of decimals. Thus, though the instrument is showing the correct value, the precision of the measurement depends upon the number of significant figures to which the observer can read the value. And in an involved measurement where many measurands have to be combined, the number of significant figures plays a crucial role to determine the precision of the ultimate measurement. We discuss below this aspect in somewhat greater detail. Significant figures Significant figures convey information regarding the magnitude of precision of a quantity. For example, ifa measurement reports that the line voltage is 220 V, it means that the line voltage is closer to 220 V than it is to 219 V or 221 V. Alternatively, if the reported value is 220.0 V, it means that the value is closer to 220.0 V than it is to 219.9 V or 220.1 V. Talking in terms of significant figures, it is 3 in the former case and 4 in the latter case, Significant figures play an important role in figuring out the final value in an involved measurement. Suppose, four resistors of values 28.4, 4.25, 56.605 and 0.76 ohms are connected in series. 28.4 4.25 56.605 0.76 90.015 What should be the value for the total resistance? The general tendency is to report the result obtained by a straightforward addition, ic. 90.015 ohms. But a close look reveals that this result conveys a wrong information regarding the precision of the measurement. If we signify doubtful figures by italics, it will be evident from the calculation shown above that the value when reported as 90.0 ohms would convey the right information regarding its Precision. The simple rules arriving at such figures in mathematical manipulation of data are now enumerated Addition and subtraction. After performing the operation write the result rounded to the same number of decimal places as the least accurate figure. "See Section 2.1 at page 9.BASIC CONCEPTS OF MEASUREMENT § 3 Error onship exists for every transducer. If the I< with ideal workmanship, the output f the measurand It would follow ‘An ideal or theoretical output versus the measurand rela transducer was ideally designed and made from appropriate materia} of this ideal transducer would continuously indicate the true value 0 curve which specifies the relationship of the output to the applied m surand over the transducer’s range. Such a relationship can he expressed in the form of a metnematical equation, graph, or table of values. The ideal output would be obtained regardless of the ambient environmental conditions and the operating conditions to which the device is subjected to Th actual practice, however. the output of a transducer is affected by the non-ideal behaviour of the device which causes the indicated measurand value to deviate from the true value. The algebraic difference between the indicated value and the true value of the measurand is termed the error of the device. This error is usually expressed in per cent of the full scale output (F-S.). The ratio of this error to the full scale output is a measure of the accuracy of the device. The above mentioned error is contributed by the existence of a large number of individual errors. In actual measurement the effect of these errors on the transducer behaviour should be known clearly. The knowledge of these individual errors, which are described below, can be often used to correct the final data and thereby increase the overall accuracy of the measurement. exactly the prescribed or known theoret (a) Intrinsic, Absolute and Relative Errors The error observed when the instrument is under the reference condition is termed as the intrinsic error. The absolute error is the difference obtained by substracting the true value of a quantity from the observed value. The relative error is the ratio of the absolute error'to the true value. In certain cases, it may be necessary to express this as relative linearity error K which can be written as where K, = average slope measured over the central 80% of F.S., and K, = average slope measured at the lower extreme 10% of F.S. VO) Uncertainty and Random Error Uncertainty and random errors are indicated when repeated ‘measurements of the same quantity result in differing values. The magnitude and direction of these errors are not known and as such are considered indeterminate. They aro caused by such effects as friction, spring hysteresis, noise, and other phenomena. The contributing factors are any random changes in input signal, combined with noise and drift in the signal conditioner. Such errors occur more in dynamic data analysis. The uncenainty is expressed as the average deviation, probable error, or statistical deviation. re on value is estimated as the amount by which the observed or calculated value departs from the value. (© Systematic and Instrumental Error Other etrors resulting from the characteristics of mate- Fials used in fabricating the measuring device or system are systematic and instrumental errors, Systematic errors are relatively constant errors occurring due to such effects as sensitivity drift, zero effect, known non-linearities, etc. The phenomenon is frequently hidden and not evident in direct observations. The value is obtained by a statistical procedure based on repetitive measurements, cared Out under different conditions or with different equipment or by new methods. They are eliminated by ‘suitable corrections.12 Introduction te Measurements and Instrumentation 7 ss errors. Gross errors are basically human mistakes in reading or recording values Suppose an instrument shows a value of 47.0 while the observer reads it a8 42.0. Or, even if he reads the correct value, records it as 41.0, Such errors can be eliminated by automation oy minimised by taking multiple readings of the same value at different times and by different observers. Misuse. A casual approach on the part of the operator is the cause of this error. For example, in electrical measurements, if the leads are not connected firmly, or an ohmic contact? is not established or if the initial adjustment such as zero-checking is not done properly, or for a microvolt order measurement proper care is not taken to avoid thermo-emfs arising out of junctions of dissimilar metals, etc. errors will creep in. Alertness and perception on the part ‘of the operator are the only remedy for such errors. Observational errors. As distinct from gross errors or errors arising out of misuse, observational errors are caused by the observer's lack of knowledge in measurement methods, Parallax is one such error. There may be many more sources of observational errors from set-ups which depend on the so called eye estimation or human reflexes. One such example is the measurement of time period of a pendulum by a stopwatch. Here the precision of the measurement depends on the reflexes of the observer who clicks the stopwatch ON or OFF by noting the position of the bob of the pendulum visually. Systematic errors As shown in Fig. 2.7, this error may have two possible origins—instrumental and environmental. Instrumental error. The instrumental error, in turn, may originate from four different causes— inherent shortcomings, interference, multiple earths and loading effect Inherent shortcomings. As the name implies, this error creeps in owing to malfunctioning of the components of instruments due to ageing, etc. For example, the spring of a galvanometer may become weak, thus changing its calibration. Therefore, to avoid this error the calibration of the instrument should be checked from time to time. Interference. ‘Thevenin’s theorem states that a network consisting of linear impedances and voltage sources can be replaced with an equivalent circuit having a voltage source and a series impedance, while Norton’s theorem states that such a network can be replaced with an equivalent circuit consisting of a current source in parallel with an impedance (Fig. 2.8). Linear * y, network | Z| Zz, Z| an AA _@) (b) () Fig. 2.8 (a) A linear circuit and its (b) Thevenin, and (c) Norton equivalents. 2If materials to be contacted are metals, it < it is easy to establish an ohmic contact between them through proper cleaning of their surfaces. But if the contact is between a metal and a semiconductor, it is necessary ta consider their Fermi levels or else, rectifying contact may result. For a discussion on this, see Solid State lectronic Devices, 4th ed., by B G Streetman, Prentice-Hall of India (1993), pp 187-189,24 Introduction to Measurements and Instrument change their lengths in humid conditions, while metal scales, althou, mmunity fron, humidity, are affected by changes in temperature. Vibrations cause s machinery o vehicles play havoc in measurements involving highly-sensitive int Ich as electron microstopes. Most electromagnetic instruments are affected by str ‘gnetic fields, ‘The remedy from such factors is, of course, controlling tempera nidity or using vibration-free mountings for instruments and shielding electrom: 8. ‘Theoretical corrections may be made for factors such as atmospheric pressure. which cannot be controlled. éiéndom error Even if all the sources of error, as narrated above, are taken care of, some fluctuation jn the measured value remains. This fluctuation is caused by many random happenings such as cosmic ray showers, changes in geomagnetism, thunder-cloud activities, minor earth tremors, etc. The effect of such disturbances which we are not aware of, are grouped together and are termed random (or residual) errors. These errors can be estimated from a statistical treatment of data. In the next chapter, we will consider the estimation of static errors in measurement. Review Questions 2.1 What are the different errors encountered in measurements? Explain with suitable examples. 2.2 Define and explain briefly the static performance parameters of instruments. 2.3 Explain, giving an example, as to why the input effective resistance of the measuring instrument should be very high like that of a potentiometer or a voltage follower if we want to measure temperature by a thermocouple accurately. 2.4 What are various sources of gross, systematic and random errors in a process of measurement? How are these errors minimised? 2.5 What are various sources systematic errors? How do these errors influence the accuracy of measurements? 2.6 The true value of a voltage is 100 V. Values indicated by a measuring instrument are 104, 103, 105, 103 and 105 volts. Find the accuracy of the measurement and the precision of - -the instrument. 2.7 A voltmeter has a uniform scale with 100 divisions. The full-scale reading is 5 V and 1/5th of a division can be read. What is the resolution of the instrument? 2.8 The dead zone in a pyrometer is 0.125% of the span. The instrument is calibrated from 500 to 1800°C. What temperature change must occur before it is detected? 2.9 Define the terms accuracy, precision, resolution and sensitivity. 2.10 List the desirable and undesitabie static characteristics of instruments. 2.11 Explain with an example the difference between accuracy and precision of measurement On what factors does precision depend? How can the accuracy be improved upon”Introduction cats and Instrumentation t Mode is defined as the most frequent ‘e frequency distribution of heights of students Fig. 3.1°. the mode value is 66 incl The mode may not exist, or eve! nay not be unique. Example 3.7 What are the mode values for the f se sets? a) 2. 3.5.6.9. 10. 10. 10, 11. 12 (b) 3.5.6.9.10.11,12.1418 ¢) 2.3.5. 5.5.6.9. 10, 11, 11. 11, 12, 12, 14, 18 Solution The mode of set (a) is 10. Set (b) has no mode. Set (c) has two modes, 5 and 11. Median The median is the value below which half the values in the sample fall. So, if the number of data N. arranged according to magnitude, is odd, it is the value corresponding to the N +2+0.5)th data. If N is even, the median is represented by the average of the (N + 2)th and (N’ Example 3.8 Find the medians of the given sets of data: (a) 2.3.4, 4.6.7.7, 7,9 (b) 3.3, 7, 8. 12, 13, 16, 19 2+ 1)th points. Solution (a) The number of data points is 9 which is odd. So, the (9-+ 2) +0.5 = Sth point, Le. 6, is the median. (b) Here the number of data points, 8, is even. So, the median is the average of the 8+2 = 4th and 5th points, ie. (8+ 12)+2= 10. “Arithmetic mean Of all these types of averages, the arithmetic mean 1 is considered the most probable value of the ineasurand and is defined as 1 i< w= Stat + tn) at i= where 1;’s are individual data and n is the number of measurements. If the data 1, 12, -.-ytm occur fi, fay -.-sfe ti ctively, (i i + D2, sey , fay ..-sfx times respectively, (i.e. occur with frequencies fs, fo, ---+ fi), the arithmetic mean is a = fit + fata +--+ Site _ 1 Witte a_Bstimation of Static Errors and Reliability 37 Example 3.9 1f 10, 16, 12 and 4 oceur with frequencies 5, 3, 4 and 2 respectively then what is the arithmetic mean? Solution The arithmetic mean is — (5)(10) + (8)(16) + (4)(12) + (2)(4) _ 5434442 Geometric mean The geometric mean gp of a set of n numbers 21, 22, 73, ...,2n is the nth root of the product of number. Written mathematically Example 3.10 Gm = YEiRaEy.. Ey The mass of a substance is being measured in a faulty common balance having unequal arm lengths. Show that the true mass of the substance is the geometric mean of the masses determined by placing the substance once on the left pan and next time on the right pan of the balance. Solution Let the true mass of the substance be m and the lengths of the left and right arms of the balance be xr; and a» respectively. Initially, the substance is placed on the left pan and a mass my on the right pan balances it. Then ma, = myx2 22 or m=m— a Next, the mass is placed on the right pan and a mass m2 on the left pan balances it. Then mag = myx, = ™,, or mam Substituting this value of xr in the first equation, we get mm or or. Harmonic mean ‘The harmonic mean fm of a set of m numbers 21, 2, 23, ...,2n is the reciprocal of the arithmetic mean of the reciprocals of number. Written mathematically,a8 Introduction to Mensurenienty ind Hitrinentation Examplo 3.11 A person travels from X to Yoat un average qpeed of 60 ban/he and retinnis by the sine My, AC an average speed of 60 kan/h Bind dhe avenge peed for the round trip, Solution Let the distance between the two pliees bea kin Phen, Hf, and ty be the tiie Gn by pny, for the onward and retinn trips, we have x Ww ” 0 0 Therefore, the average speed vay for the round trip is r 2n 7 hey Beet On oo total distance ty = total time We observe that ugy is nothing but the harmon nian of the two Kpeeds. Root mean square The root mean square (rms) or quadratic mean of a set of n numbers 21, x2, 74, - sn in defined as This averaging is quite common in engineering and physical applications. Measures of Spread The average height, in our carlier example, may be the most important. characteristic (statistic) of the students. However, it is also equally important to know how the observations spread out or disperse around the average value. Like the measures of central tendency, there are several measures of spread or dispersion: 1. Deviation 2. Mean absolute deviation 3. Variance, and 4. Standard deviation \Beviation Usually written as d;, deviation is the scatter of an individual datum from the mean. Symbolically, Obviously,Estimation of Static Errors and Reliability 39 Man absolute deviation ‘The mean absolute deviation’ D of a set of data is defined as the average of absolute values D= aM = ad be—al Ta1,t2,-..,tm occur with frequencies fi, fo,..., fy respectively, then et D mile a] Example 3.12 Find the mean absolute deviation of heights of 100 male students of a class as given in table below. ——————— Height (in) 60-62 63-65 66-68 69-71 72-74 No. of students 5 18 42 27 8 Solution Here, the arithmetic mean __ (61. x5) + (64 x 18) + (67 x 42) + (70 x 27) + (73 x 8) - 5+ 1844242748 = 67.45 in The rest of the calculation is presented in the following table: Height (in) las = wl = |x - 67.45) fi files - wl 60-62 6.45, 5 32.25 63-65 3.45, 18 62.10 66-68 0.45 42 18.90 69-71 2.55 a7 68.85 72-74 5.55 8 44.40 = 100 226.50 226.50 7 Therefore, the mean absolute deviation D = =) >~ = 2.26 in Variance Intuitively, the mean absolute deviation is a good measure of spread; but it is mathematically intractable. One difficulty is the problem of differentiating an absolute value function. To40 Introduction to Measurements and Instramentation Mutt ee obviate this difficulty, the variance, which is nothing but the mean squared deviation, rm defined as 1< 2 Jarinnce = — S0(xi - Varinnce = = Le Hw ‘The statistical theories distinguish between a sample and a population of data. The Sample denotes that the number of data is less than or equal to 20 while the population indicates the size more than 20. According to the statistical theories, though the variance or MSD is good measure of dispersion for a population, the divisor should be (n — 1) rather than n to make it an unbiased estimator for a sample \Standard deviation Denoted by or s, standard deviation is defined in two ways—one for a sample (n < 20) ang the other for a population (n > 20) as follows z a= n> 20 n Variance which is just the squared standard deviation is thus written as 0? or s?. Example 3.13 A set of 10 independent measurements were made to determine the diameter of the bob of a simple pendulum. The measured values in cm were: 1.570, 1.597, 1.591, 1.562, 1.577, 1.580, 1.564, 1.586, 1.550 and 1.575. Determine (a) the arithmetic mean, (b) the average deviation, (c) the standard deviation, and (d) the variance. Solution ~ The calculation is presented in a tabular form below with the last row in bold face letters indicating sums of corresponding columns: Ea |a| a& 1.570 0.005 0.000025, 1.597 0.022 0.000484 1.591 0.016 0.000256 1.562 0.013 0.000169 1.577 0.002 0.000004 1.580 0.005 0.000025, 1.564 0.011 0.000121 1.586 0.011 0.000121 1.550 0.025 0.000625, 1.575 0.000 0.000000 15.752 0.110 0.001866 w= 1.575 D=001 5 = 0.0143 0.000204Estimation of Static Errors and Reliability (a) The maximum possible error (absolute error) in the determination limits. (b) Root-sum square error in the determination of D based on 30 limi Sobution (a) At 3.296 limit. 6L = 0.02 x 3.29=0.0658mm — 6W =0.5 x 3.29 = 1.642 From Eq. (3.2). we get ap [22 +|22 0 _(,_¥\, (Ww = (: = 2) bbs (3) aw =fi- a = [aang] 0918 + [ri] 5 =0.115mm (b) At 3o limit. re =002x3=006mm ry =05x3=1.5mm From Eq. (3.8). NET ae = yw (0.06)? + =007%mm os) (sp 49 13.290 Note: The instrument makers often mention a quantity called tolerance of the instrument. It is stated either as an absolute amount such as £0.5°C or as a percentage such as +1%. This tolerance may either be the limiting error or the standard deviation. 35 Chi-Square Test While calculating the probable error by Eq. (3.7), we tacitly assumed that our data closely followed 2 normal or Ganssian distribution. Although in most physical measurements data do follow such a distribution. it is someshat better to check quantitatively that the distribution is close to Ganssian. Chi-square (,7) goodness-of-fit test helps us to do that.ah, VF is a measure of discrepancy existing between the observed and the expected frequencie, Av a given range (denotedt hy o and ¢ respectively) and is defined as 30 Introduction to Measurements and Instrumentation (3.9) Wy? = Q the observed and the expected frequencies agree exactly, and therefore, the data exactly obey a Gaussian distribution, while if ? > 0, there is some discrepancy. ‘The larger the value of \?, the greater is the discrepancy. . te practicn, the computed value of \?. given by Eq. (3.9) is compared with some critical value, such as (As OF Vy. which correspond to 95% and 99% confidence limits. These values sre wvailsble in tabular form in which values are listed for different confidence limits against mamber of degrees of freedom », which is defined as v l-m, (3.10) where 1s the total number of groups and m is the number of population parameters. If in the above comparison. the calculated value were less than that listed in the table, wwe wouk! conclude that the observed frequencies do not differ significantly from the expected mes snd would sccept that the data are Gaussian at the corresponding level of significance. We should also look at the situation where x? is near zero, such 9 0.05 or 0.01. If the computed value of \? is less than the listed value for Xs or x%), We may conclude that the Sgreement S too good. The following example will make the procedure clear. Example 3.22 Test the goodness-of-fit of the data given below Height (m) 1521.56 157-161 162-166 167-171 _1.72-1.76 No. of students 5 18 42 27 8 ‘Solution ‘The work may be organised as in Table 3.1. po = (ESH)(S) + (1.59008) + (1.89192) + (1.69)(27) + (1.74)(8) _ sbdre as a (28 1.6175)? + 18(1.59 — 1.6475)? + 40.4 — 1.6475)? + 7 100 [209 — eats? + TAS 1.6475)? eee + 874 ~ 1.6475)? y = 0.04867 Neate: In Table 3.1: L 2 for class-boundaries have been calculated from 2 = (— he id s have been calculated by the standard procedure. is pee hae 2. Areas under normal curve have been obtained using standard table.Estimation of Static Errors and Reliability 51, — Table 3.1. Example 3.22 “hed Class 2 Area under Area for Expected Observed undaries for.class the normal each _— frequency frequency x boundaries curve from class e o 0 toz fae 1515 -2.72 0.4967 ff 0.0413 4.13 or 4 5 1.565 -1.70 0.4554 1 0.2068 20.68 or 21.18 1615 -0.67 0.2486 1.62-1.66 0.3892 38.92 0r 3942 1.665 0.36 0.1406 1.67-1.71 0.2771 2771028 27 1.715 1.39 0.417 1.72-1.76 0.0743 7.43 or 7 8 1.765 2.41 0.4920 * Obtained by adding with the area corresponding to previous z value, because there is a change of sign. 3. The area for each class has been calculated by subtraction from the area under the preceding z, except in one case where it has been added because z has changed sign there. 4. Expected frequency has been obtained by multiplying the area for each class by the total frequency 100. Now, a_ (5-4)? | (18-21)? | (42-39)? | (27-28)? | (8-7)? _ ee ee ee Here, i=5 and m = 2 (namely, 1 and g); therefore, v = 5-1-2=2. From the standard table! for Percentile Values for the Chi-Square Distribution we get, for v = 2, x%s = 5.99. Since the value we obtained is much lower than this, we conclude that the fit of the data is very good. Now, for v = 2, x%5 = 0.103. Since our value is greater than this value, we may say that the fit is not too good. 1.08 A6& Curve Fitting Methods More often than not the data show a scatter and we draw an average curve by what we call the eye estimation. But there are a few methods of drawing the most probable curve provided, of course, we know the relation between the parameters. We discuss three methods, namely 1. Method of sequential differences 2. Method of extended differences, and 3. Method of least squares for a linear relation such as, y 9 + are. ‘See Table 2, Appendix D at page 866.Estimation of S1 The measurement can thus be reported as 1,575 + 0.014 em where the indicated error | standard devintion 4 Error Estimates from the Norm Distribution IC a rather large number of careful measurements are frequency of occurrence of a particular value ig plotted ayant the corresponding, values, the reeulting histogram usually assumes the form af a bell-shaped curve called the womeit or Gaussian’ curve as shown in Fig. 3.2. Here we have plotted the Gaussian distribution curve for a typical height measurement of students of a class as given in Example 3 12 at page 39 jal (or Gaussian) carried out of a measurand, and the 40 Gaussian curve £ 30 3 5 ‘S20 z 10 0-4 + + + 4 60 65 70 75 Height (in) Fig. 3.2 Height vs. number of students of a class showing a Gaussian distribution. Errors that are made in physical measurements do often have a normal distribution. This distribution is defined by the equation 1 1(@e—? YS Wim oo { 2 63) where is the mean, o is the standard deviation, and x © 3.14159. It is easily seen from the following calculation that the total area bounded by the normal distribution curve and the abscissa is 1. 2 ) } iis aes ete ees [. ee wal. o{-5 ca = 4 { exp(-u?)du where u= T Sco ov2 T(1/2) VE Named after (Johann) Karl Friedrich Gauss (1777-1855), German mathematician. His contribution to electrostatics gave birth to the electromagnetic field theory apart from his seminal contributions to probability theory and other branches of mathematics. 1a 42 Introduction to Measurements and Instrumentation 5 Hence the area under the curve between x = a and x = b, where, a < b, represent, th probability that 2 lies between a and b and, therefore, can be written as Pr{a < ¢ < . We note that in the very special case when = 0 and o = 1, the normal distributio , Eq, (3.3) reduces to 1 But more important, regardless of what 1 and ¢ may be, the Eq. (3.3) can be transla the form of Eq. (3.4) by defining z = (x — )/o, when it becomes ed ty This is called the standard form of the normal distribution. A plot of the standard norma) distribution is shown in Fig. 3.3. Fig. 3.3, Standard normal curve (Figures, e.g. 68.27%, show the probability within the indicated limits) The probability of occurrence of a particular value between limits 2) and zz given by the standard normal distribution is Pra Sesa}= [” Foe (-2) ae 35) Areas under this curve bounded by the ordinates at z = 0 and at any positive value of 2 are given in tabular form in Table 1, Appendix D at page 865. From such a table the are between any two z-values can be found by using the symmetry property of the curve, namely J(2) = f(—2). Thus, the probability of occurrence of a particular value within a certain limit can be found out from this table. From the graph of this standardised (mean 0 and variance 1) normal curve as shows ia Fig. 3.3 we observe that