sis of Expevimental Data 1.1 Introdueti ncasured with perfect certainty; there are always errors in any © quantity and, then. repeat the No physical quantity can be measurement. This means that if we mi ure a different value the second time, How, measurement, we will almost certainly me: then, can we know the true value ofa physical quantity? The short answer is that we can’t. However, as we take greater care in our measurements and apply even more refined experimental methods, we cat reduce the errors and, thereby, gain greater confidence that our measurements approximate even more closely the true value. Some form of analysis musi be performed on all experimental data. The analysis may be a it may take the form of a complex theoretical simple verbal appraisal of the test results, « analysis of the errors involved in the experiment and matching of the data with fundamental physical principles. Even new principles may be developed in order to explain some unusual phenomenon. Analysis of data involve determination of errors, precision, general validity of experimental measurements and graphical presentation of data, Errors will creep into ail experiments regardless of the care exerted. Some of these errors are of a random nature, and some will be due to gross blunders on the part of the experimenter, 1.2 Accuracy and Precision Experimental error is the difference between a measurement and the true value or between two measured valucs. Experimental crror, itself, is measured by its accuracy and precision. ey measures how close a measured value is to the true value or accepted value. lue for a physical quantity may be unknown, it is sometimes not Accura Since a true or accepted val possible to determine the accuracy of a measurement. Precision measures how closely two or more measurements agree with other. Precision is sometimes referred to as repeatability or reproducibility. A measurement which is highly ach other. reproducible tends to give values which are very close to ¢Errors in experimental data are those factors that are always vague to some extent and carry some amount of uncertainty 1.4 Accuracy and Err Analysis Vhe exact value ofa variable is called the true value, The value of the variables as indicated m is called the measured value. The accuracy of 4 measurement But by a measurement s refers to the closeness of agreement between the measured value and the true value. the true value is rarely known exactly, and various influences, called errors, have an effect on both of these values. An appropriate approach to stating this closeness of agreement is to identify the measurement errors and to quantify them by the value of their associated uncertainties, where an uncertainty is the estimated range of value of an error. We define an error, €, as the difference between the measured value and the true value i.e. e=m-t, a Where € is the error, m, is the measured value f, true value of the variable been measured. While the true value is rarely known exactly. Errors exist and they have a magnitude as given by equation (1). 1,5 Error classification: Error can be classified in different ways Systematic / Fixed error: Fixed errors causes repeated readings to be in error by roughly the same amount but for some unknown reason. These fixed errors are sometimes called systematic errors, or bias errors. Generally the value may be constant / similar form / recur consistently every time measurement is measured. May result from improper condition or procedures employed i.e. calibration errors, errors of technique, uncorrected loading errors. Rendom/ Accidental errors: Random errors are caused by personal fluctuations, random electronic fluctuations in the apparatus or instruments, various influences of friction, and -so forth, These random errors usually follow a certain statistical distribution, but not always, Random error causes a random variation in measured values found during repeated measurements of a variable, Random errors are distinguishable by their lack of consistency, errors of this type are normally of limited duration and are inherent to specific environment. It could be as result of errors stemming from environmental variations, due to insufficient sensitivity of measuring system. 1.6 Uncertainty Analysis and Propagation of Uncertainty Uncertainty is a numerical estimate of the possible range of the error in a measurement. In any measurement, the error is not known exactly since the true value is rarely known exactly, But based on available information, the operator might feel confident that the error is within certain bounds, a plus or minus range of the indicated reading. This is the assigned uncertainty. Uncertainty is brought about by all of the errors that are present in the measurement system—its calibration, the data set~statistics, and the measurement technique. Individual errors are properties of the instruments, the test method, the analysis, and the measurement system. The uncertainty values assigned to an instrument or measurement system specification are usually the result of several interacting random and systematic errors inherent to the measurement system, the calibration procedure, and the standard used to provide the known value. the uncertainty of a result: in terms of clike th n ways to state : ie There are two commo) standard f a percent or fractional uncertainty, for which is s of a pel uncertainty, for which we deviation of the mean dj, OF 17 term: reservethe symbol, ¢ When stating o result and its uncertainty ina report, one typic ally uses th a ses the form, x + a,, units placed last. For example example, if the mass of an object is found an object is found to be 9.2 g and the uneertainty in the mass is 0.3 p, one would write m= 92 40:39 Sometimes one will present uncertainty in terms of €, but in this case, € is usuall : se, usually multiplied by 100, so that one would say, ‘The mass of the object is 9.2 grams with an uncertainty of 3 percent," 1,7 Propagation of Uncertainty ‘The method of computing the uncertainty in a result which depends on several variables, each with its own uncertainty, is called propagation of uncertainty, or casually error propagation, Suppose we have measured the length and width of a tabie and have computed the standard deviation of the mean value for both the length and the width. Our aim is to determine the area of the table and an associated standard deviation of the area. It can be shown that the best estimate of the area is simply the mean length times the mean width. What uncertainty should we associate with this same area? The answer is not obvious and in fact we can distinguish two distinct extreme cases: * The uncertainties in the length and width are completely independent: If the uncertainties are completely independent, the possibility of compensation occurs. ‘That is, if the uncertainty in the length causes the area to be too large, then the uncertainty in width may be such as to cause the area to be too small. On the average, the total uncertainty in the area will be algebraically less than the sum of the separate contributions to the uncertainty in the area. © The uncertainties in the length and width are completely dependent: For completely that the uncertainties are dependent uncertainties, we must take into account the fact 5 iii:ways correlated, Uhis leads te complications involving a quantity cal wovartance of bvo correlated quantities. Vable 1 Present common formulae for propagation ef uncertainties. [ se he Combined in the eases of more complicated! formulas Table 1 Vnetionat Poon Vormmla Uneert Vv fon Mocitet or Quotient ry ot f= r/y ayer Stn or Ditlerenee rtyorfar-y Produc Constant nultipliors f f Col fretors ratsed to powers fy" f Logatithinie finetions: f f f Exponential finetions Consider a result R, whieh is dete: nined through some functional relatio independent variablesx,3 xy x, defined by R Alar =x where L is the number of independent variables involved. Each variable contai measure of uncertainty that aflects the result. The best estimate of the true mean value R would be stated as R=Rt ug Where the sample mean of ® is found from R Wks and the uncertainty ty is found from Me = fille ttyy sc sertty, } cach wey, f= 1, 2s. represents the uneertainty associated with the best estimate of £ ‘The value of tty tetlee and so forth through %, S the contributions of the individual and e uncertainties as they are propagated through to the result, 6ener Bs vity index, 6), that relates how changes in each x, affect R, can be defins - for each variable as on Figay oie i= 1,2, The contributior inty i the re y the term 8, ibution of the uncertainty in x to sult R is estimated by the t The most probable estimate of up i Sr ed a: ¢ given by the second pow r pI sstimate of uz is generally accepted as that value given by the s ela ; RSS), T f fal f quares (RSS). TI ‘ uncertainty in the variables to the result is by Up = [Seemay} ‘Therefore aR 2 ar. \? ar yzy'2 a= (GE) + (ame) te .) | Q 1.8 Uncertainties for Product Functions In many cases the result function of Eg. (2) takes the form of a product of the respective primary variables raised to exponents and expressed as Ra xx? xt Equation (2) take the form wa yl" 6) Uncertainties for Additive Functions san additive form, R will be expre: When the result funetion has R= ayxy + O2%2 + ntn = Yan The uncertainty in the result may then be expressed as up = [E(a)') (4)Equations (3) and (4) may be used in combination when the result function involves both product and additive terms, Examples 1 The resistance ofa certain size of copper wire is given as R=R,[1+a(T-20)] where Ro = 62 + 0.3 percent is the resistance at 20°C, a = 0.004°C+1 percent is the Calculate the temperature coefficient of resistance, and the temperature of the wire is T= 3041 nce of the wire and its uncertainty. The two resistors R and Rs are connected in series as shown in the accompanying ‘The voltage drops across each resistor are measured as E=10V+01V 1.2V + 0.005V along with a value of R, = 0.0062 +3% From these measurements determine the power dissipated in resistor R and its ‘uncertainty, 3. For a displacement transducer having the calibration curve, y = KE, estimate the uncertainty in displacement y for B = 5.0 V, if K = 10.10 mm/V with ug = £0.10 mnVV and ug = £0.01 V at 95% confidence. 1.9 Sources of errors (1) Nolse: It is defined as any signal that does not convey usefi information, nae(2) Design limitations: These are certain inevitable factors such as friction and resolving power which lead to uncertainty in measurements. (3) Response time: It is the time lag between the application of input signal and output measurement. 4) Deterioration of measuring system: Physical and/or chemical deterioration or other alterations in characteristics of measuring system constitute a source of error in measurement he change in atmospheric temperature may alter the clastic (5) Environmental effect: constant of a spring, the dimensions of a linkage, electrical resistance ete. similarly other factors such as humidity, pressure etc. also affect measurements (6) Errors in observation and Interpretation: It is the mistake of operators in observ interpreting and recording the data 1,10 Statistical Analysis of Experimental Data When ui set of readings of an cnt ts taken, the individual readings will vary somewhat from cach other, and the expermenter may be concemed with the mean of all etic mean the readings. If cach reading is denoted by x, and there are readings, the arit is given by O) The deviation di for cach reading is defined by d x Sum The standard deviation of root-mean-sq a= [dna ~ tn)? (6) and the square of the standard deviation is called th e. This esc the population or biased standard deviation because it strictly 5 only wher a number of samples is taken to describe the populatioIn many circumstances the engineer will not be able to collect as many data points as ne sary to describe the underlying population. Generally speaking, it is desired to have at least 20 meas rements in order to obtain reliable estimates of standard deviation and general validity of the data. For small sets of data an unbiased or sample standard deviation is defined by ” ‘The standard deviation of the mean value of a set of measurements oj, when all measurements have equal statistical we ieht, is given by Om (8) When we speak of the uncertai urements made under identical ly 0 of a set of mea conditions, we mean that number a), and not 0, Example The following readings are take of a certain physical length, Compute the mean reading, ed deviation, variance, using the “biased” and unbiased basis: 1.11 The Gaussian or Suppose an experimental observation is made and some particular result is recorded, We know (or would strongly suspect) that the observation has been subjected to many random errors. These random errors may make the final reading either too large or too small, depending on many circumstances which are unknown to us. Assuming that there are many 10small errors that oo NAL Contribute to the Hal error and that cach small error is ot equal and equally likely to be positive or negative. the may be derived. If the measure ent iy designated by x, the Gaussian distribution gives the probability ¢ the measurement will lie between Vand vv and is written P(x) = () In this expression x), the standard deviation. the mean reading and o to call P(x) the prodability density. function curve, P(x), detined by The area under the portion of the probability densi interval ¥ — z,0 number of observ ions ¥ mean of'n observations: AV mean of normal population whieh the samples are taken from Student then developed a distribution function f(t) such that f(t) =- = ( . 2) 2 reo =a (1 E\() Where Ky is a constant which depends on mand y is (2 1) degrees of the distribution function approaches the normal distribution. Table below gt sn a 99 percent dence. tn Student's (for different degrees of freedom and levels of con! confidence level.Table 5 Sutsenpr des Degrees & (6 a Ise for bas 1 307s 2 186 - 1.638 7 AS33 5 16 o 24. 3 ow bee a a 2 u a0 \ 2 i aos V3 = 6 tans 1 2 feet fags el is cal ae tat 1s ose zie 2 a ere Bhi = Bi 2 Lesa a0 ‘ pr Vet 6 (688 Tals 307140.1 V. Determine the 5 Ten observations of a voltage are made with @ 15 Vand o and 1 percent significance levels. Example of a metal plate which give 3.61, 3.62, 3.60, Ten measurements are made of the thicknes: 3.63, 3,61, 3.62, 3.60, 3.62, 3.64, and 3.62 mm. Determine the mean value and the tolerance limits for a 90 percent confidence level. Example In a certain pressure measurement a known precision o large number of measurements. This significance level of the pressure determination is f +6 kPa can be obtained with a directly related to rejection of a certain production part. The cost per part rejected is P and the cost per pressure measurement is C. Determine the relative relationship between P and C for levels of significance of 5 and 10 percent such that the mean pressure measured will be +3 kPa. 1.13 Chauyenet’s Criterion It is a rare circumstance indeed when an experimenter does not find that some of the data points look bad and out of place in comparison with the bulk of the data. The experimenter is therefore faced with the task of deciding if these points are the result of some gross experimental blunder and hence may be neglected or if they represent some new type of physical phenomenon that is peculiar to a certain operating condition. The engineer cannot just throw out those points that do not fit with expectations—there must be some consistent basis for elimination.Suppose 7 measuremer Ppesc 7 Measurements of a quantity are taken and is Large enough that we may expect the results to follow the G: issian error distribution, ‘This distribution may be used to compute the probability that a given reading will deviate a certain amount from the mean According to Chauvenet’s criterion a reading may be rejected if the probability of obtaining the particular deviation from the mean is less than 1/2 Table below lists values of the ratio of deviation to standard deviation for various values of n according to this criterion with the figure below furnishing a graphical representation. Table 6 Chauvenet’s criterion for rejecting a reading Number of Readings, Ratio of Maximum Acceptable Dev a to Standard Deviation, dru! 2 138 4 154 5 165 6 in 1 Ls0 10 1.96 15 : 50 100 300 500) 1000 SO 2os ' 10 Ino 1000 ‘= number of readings In applying Chauvenet’s criterion to eliminate dubious data points, one first calculates the mean value and standard deviation using all data points. The deviations of the individual Points are then compared with the standard deviation in accordance with the information in Table 6 (or by a direct application of the criterion), and the dubious points are eliminated. For the final data presentation a new mean value and stand: lard deviation are computed with the dubious points eliminated from the calculation Example Using Chauvenet’s criterion, test the data points of below for possible inconsistency. Eliminate the questionable points and calculate a new standard deviation for the adjusted data. 221.14 Method of Least Squares es of their ‘The sum of the squa Suppose we have a set of observations x1, 3 deviations from some mean value is Now, suppose we wish to minimize $ with respect to the mean value Xj, We set =o y =2(%) — Xm) = —2 (> xi- rn) st where n is the number of observations. We find that or the mean value which minimizes the sum of the squares of the deviations is the arithmetic mean. This example might be called the simplest application of the method of least squares. Suppose that the two variables x and y are measured over a range of values. Suppose further jon for yas a function of x. T obtain a simple analytis rhe simplest that we wish to ical expr function is a linear one; henes asa linear function of x. ding, the best linear fun e, we might try (0 establish y scatter a type of f The problem is one of fin ction, for the data may s 23considerable amount, We could solve the problem rather quickly by plotting the data points aight line through them by eye. Indeed this is common. ‘on graph paper and drawing practice, but the method of least squares gives amore reliable way to obtain a better functional relationship than the guesswork of plotting. We seek an equation of the form y=axtb The constants are readily obtained as nLouy) -— CxO) nYxp — (2x)? CyEx?) -VEwdAxd) nea xD? Example From the following data obtain y as a linear function of x using the method of least squares: Lo 16 a4 40 1.15 The Correlation Coefficient Let us assume that a suitable correlation between y and x has been obtained, by either le: : : ast- Squares analysis or graphical curve fitting, We want to know how good this fit is and th. : and the arameter whi i ion ii 7 hich conveys this information is the correlation cvefficient r defined by — nie) - CxdCy) vndat = Ox)? Indy? — Oy? 24At this point we must stress the need for graphical displays of data. When the data are plotted the experimentalists will get an “eyeball” plot of the correlating straight line before actually performing a least-squares analysis. 1.16 Graphical Analysis and Curve Fitting Engineers are well known for their ability to plot many curves of experimental data and to extract all sorts of significant facts from these curves. The better one understands the physical phenomena involved in a certain experiment, the better one is able to extract a wide variety of information from graphical displays of experimental data. If the experimenter has a good idea of the type of function that will represent the data, then the type of plot is easily selected. It is frequently possible to estimate the functional form that the data will take on the basis of theoretical considerations and the results of previous experiments of a similar nature ‘Table 7 summarizes several different types of functions and plotting methods that may be used to produce straight lines on the graph. The graphical measurements which may be arked that the made to determine the various constants are also shown. It may be re method of least squares may be applied to all these relations to obtain the best straight line tal data to fit the experim:Table 7, Methods of plotting various functions to obtain straight linesFunctional Rear shp y= atte yeatheper Method of Mot com seniiluy paper dis 2 oa tinens paper foes SOM y5 com linea paper ra Gray of Py rod phical Determt gation lee) i bbe i a * a Ae a ay A a 0 i r a“Graphical Determination often Slay eae ys. 4 on tnear paper | Ste et |i yeethe 98. Fon hiner paper % 1.17 Choice of Graph Formats The engineer has many graph formats ailable for pres; 2 experimental data or calculation results. While bar charts, column charts, pie charts, and similar types of'd have some applications, by far the most frequently used di 28choices of coontinates to mateh the sitiation, This basic ¢: aph has several variations in format th the simple table of ney data shown b fe shall illustrate by plotting Six formats for plotting the data are shown in Fig.a through 7°. The choice o depends on both the source and type of data as well as the eventual use to be made of the The computer graphies were display. The following paragraphs discuss the six alternatives generated in Microsoft Excel. may be employed for either raw experimental data points or for points calculated from an les analytical relationship. With computer graphics a wide selection of data marker s\ available. b. This display presents the points with the same data markers connected by a smooth curve stem; in this case, by computer. This drawn either by hand or by a computer graphies s display should be used with caution, If employed for presentation of experimental data, it smooth curve describes the physical phenomena represented by the data at the implies points. The engineer may want 0 avoid such an implication and choose not to use this format. Jisplay is the same as (4) but with the data markers removed. It would almost c. This be employed for presentation of experimental data because the actual data points are not displayed. It also has the same disadvantage as (8) in the implication that the physical phenomena are represented by the smooth connecting curve. In contrast, this type of display is obviously quite suitable for presenting the results of calculations, The calculated 29points could be designated with data markers as in (b) or left off'as in (c). The computer- generated curve offers the advantage of a smooth curve with a minimum number of calculated points. «This display presents the data points connected with straight-line segments instead of a smooth curve, and avoids the implication that the physical situation behaves in a certain “smooth” fashion, ‘The plot is typically employed for calibration curves where linear interpolation will be used between points, or when a numerical integration is to be performed based on the connecting straight-line segments. If used for presentation of experimental data, the implication is the same as in (6) and (c) that the physical system actually behaves as indicated, in this case with a somewhat jerky pattern. e. The format in (e) is the same as (d) without the data markers. It might be used for calculation results where the engineer wants to avoid computer smoothing between the calculated points. JF Finally, the format presented in (f ) is one that is frequently selected to present experimental results where uncertainties in the measurements are expected to result in scatter of the data points. A smooth curve is drawn through the data points as the experimentalist’s best estimate of the behavior of the phenomena under study. The smooth curve may be drawn by hand or generated through a least-squares process exeouted by the computer. A trend line equation may or may not be displayed along, with the eurve. When experimental uncertainties are expected to contribute significantly to the scatter of data, as they do in many cases, a full discussion of their nature should be offered in the accompanying narrative material. In some cases a display like that shown in (/) may be used to present calculated points along With a particular type of curve fit for the points which differs from the calculation equation.~ a ~ = 7 pce Graphical Presentations Most engineering reports include some form of graphical presentation: a simple plot of raw data, a statistical distribution of data, correlation of data, and perhaps a comparison of experimental data with analytical predictions. 2 The x-y chart, and others, are part of the overall graphics assembly which may bemrequired for good report. Some of the graphics are: 1. General curves or “trends” Bp . Detailed plots of experimental data 3. Accurate graphs which are used for calculation purposes 4. Schematic diagrams 5. Scale diagrams for experimental apparatus 6. Graphs in a format to be used for presentation 31Regandless of the purpose of the geaph, there are certain elements of good style which should be common to all, 1, Label the graph, Make sure all coordinates are labeled consistently; ie., don’t use all on one coordinate and mixed upper- and lowercase on the other. 2. 1fnumerieal v uppercase ‘alues of variables are graphed, make sure that each coordinate has scale markings and that ea ch label for a variable includes the units for that variable. ry to maintain the same unit system for all variables; i.e., do not mix English and SI units. 3. Recognize that some graphs may be reduced in size later because of space limitations. This is particularly true for those to be published in techt al papers or symposia volumes. If any reduction is anticipated, the size of the labels and unit specifications on the graph must be increased so that they will be legible in their eventual format. 4, The purpose of graphs is to convey information. Sometimes this is best accomplished by plotting several curves on one graph with markers and legends to distinguish between each curves as shown in the figure below. Condenser . 55 85 | ‘Temperature Bg Degrees F a 95 = 1 105 2 “a 3 11S 3 47 8 Ed 4s Boa a a . 39 rf # A fy Oo 35 78 55 60 - ue Evaporator Temperature, Degrees F