Types of Applications oF Measurement Instrumentation WHY STUGY MEASUREMENT SYSTEMS? J ‘The study of any subject matter in engineering should be motivated by an appreciation of the uses to which the material might be put in the everyday practice of the profession ‘Measurement systems are used for many detailed purposes ina wide variety of application arc approach will be to start with some specific epplications ina specific industry and then ever: picture by developing classification schemes that apply to all possible situations. ‘Whilemeasurementisused in many contexts, [want o introduce somebasic ideasusing the automotive industry as an example. This industry employs measurement in many ways and is thus a good choice {or exploring the various uses of measurement tools. Inthe text title, the term “measurement system” is meant to include all componen's in a chain of hardware and software that leads from the measured variable to processed data. Let us start examining the use of measurement in the automotive industry “at the beginning,” that is, with the conceptual design procese, where a new automobile or sick Is frst conceived and the basic configuration developed, Because a modern automobile uses as many as 40 or '50 sensors (measuring devices) in implementing various functions necessary tothe operation of the ear ‘an automobile designer must be aware of the instruments available for the various measurements and hhow they operate and interface with other parts of the system. As new sensors are invented, designers ‘must keep up with such developments since they may allow improvements in car design and operation. Lack of such sensor knowledge can severely resirict the range of designs that one can conce!ve. thus limiting improvements in overall car performance. While sensor specialists will at later stages of design ‘consider the measuring devices in great detail, the conceptual designer must have a basic appreciation of their capabilites, so that te intial design does not neglect any useft! possibilities. Once the conceptual stage of design is well underway, measurement system considemitions arise in new contexts. Many engineered products are nowadays designed using the methods of concurrent cemgincering where design and manufacturing are integrated, rather than being considered suena fas was ofien the case in carlicr times. Before concurrent engineering became commoa, dcsign was _generlly completed frst, manufacturing considerations addressed only Inter, and costly revisions and delays (or poor designs) were often the result. With concurrent engincering, product design conccntsHEBD Docbelin’s Measurement Systems are not “frozen” uni both funetion and manfacturabilty have becn reconciled. That is. the design and ‘manulacturing engineers work in coordinated teams, bending their expertise right from the beginning of the design process. Both functionality and manufacturability considerations often require the design process (0 include laboratory testing of one kind or another. For example, iP'a new material is being considered, we may need to ran strength tests to develop date needed by the design enincors. Or. a new or revised manufacturing process may require statistical response surface experiments! to find the effects of process variables on performance ani/or cos. Finally, availability from suppliers of new components, such as improved shock absorbers. may require performance testing to decide whether thei use is warranted in the new design. We see that laboratory testing and the associate! measurement systems are hus vital part ofthe design process. ‘As design and development proceed, prototype subsystems and finally entre vehicles will be produced, These are used as “test beds" to evaluate performance and then food back information to the design/manufacturing teams. Tht is, inital designs usually have unsuspected flaws, which are revealed by building and testing the prototypes. Als. “pencil and paper” or computer-aided designs afar are ‘based on theories that are never exactly correct, so experimental tostng is nooded to verify, oF improve, theoretical calculations. We begin to appreciate that design relies heavily on experimental testing at every stage ofthe process ‘We have seen that experimentation soften nceded during the design phase to help inthe development of the manufacturing processes forthe product. Once the design has been finalized, then manufacture of the product in quantity rather than the “one of a kind” mode use! during development. can commen. ‘When we examine actual production machinery and processes, ve often find that these manufacturing tools are controlled by a so-called feedback mechanism. In such a scheme, some quahity parameter of the par produced is mexsured with appropriate sensors. This measured value is compared with a desired value of the parameter, and ifthe desired and measured values do not agree within some allowable tolerance, aconiroller adjusts the machine or process until the prouct is “on specification.” Perhaps the most obvious example of this general situation isthe machining of parts to spectie dimensions. Here the measuring devices are precision gages that measure shaft diameters hole sizes, lengths, etc. Robots used to weld, spray paint, or assemble pars ae also usually feedback devices that use motion and force sensors to control the robots" operation. Again itis clear that measurement plays significant role in almost every manufacturing enterprise. “Turing now tothe final product, a modem automobile, as mentioned earlier, relies on a multitude of | sensors for its optimum operation. Somcor these play essentiatly monitoring” role. that i. they measure and display to the drive, information useful for safe and efficient operation ofthe car. Speedometers tell us the vehicle's speed, while tachometers display engine RPM. Fuel gages keep track of the gas supply, and temperature sensors war of overheating, Recent developments inelude use ofthe Global Positioning System (based on satellites) to locate the car on an electronic map and guide the driver to «desired destination. Many other sensors are part of feedback controls that optimize engine operation ‘by measuring such variables as tmospheri pressure, airflow rae, fuel/air ratio, engine temperatures, cite. Acccleration sensors (accelerometers) measure vehicle motion during a erash and signal air bags to L.Then (2.6) ‘The significance of Eq (2.6) isthat the effec of varationsin,y Ky and K, (asa resultof modifying inputs iy. iqy ad i,q), on the relation between input , and output s,, has been made negligible. Me snow require only that K, stay constant (unaffected by i) in order 16 mainiain constant input-ousput calitration as shown by Eq, (2.0) ‘You may question whether much really has been gained by this somewhat claborate scheme, since we merely transferred the requirements for stability from K,, and K,y10 K,, In practice, however, this ‘method often leads 1 great improvements in accuracy. One reason is that, since the amplifier supplies most of the power needed, the feedback device can be designed with low power-hancling capacity. In general, this leads to greater accuracy and linearity in the feedback- device characteristics. Also, the input signal e, need cary only negligible power; thus the feedback system extracts less energy from the ‘measured medium than the corresponding open-loop system. This, of course, results in ess distortion of the measured quantity because of the presence of the measuring instrument. Finally, i the open-loop chain consists of several (perhaps many) devices, each susceptible to its own spurious inputs, then -G. Chetwynd, “Selection of Structural Material for Precision Devices,” Precision Eng, vol 9, 0.1, January 1987, pp. 3-6.Genercllzed Configurations and Functional Descriptions of Measuring Instruments EB all these bad effects can be negated by the use of high amplification and a stable, accurate feedback device, ¢ us Fuca £ ¢ via { Fa] | teow |] fossa aor Se (erence system ja fa ye usa Fue Fun ; : Fa es ke | | west Moor ‘re Feeeck ace ¥ to (Ce oop toot yt (FIBER) Use of feedback to reduce effect of spurious inputs. Before we pass on to other methods, we should mention that application ofthe feedback prin ‘ot without its own peculiar problems. The main one is dynamic instability, wherein excessively high amplification leads to destructive oscillations. The study of the design of feedback systems is a whole field in itself, and many texts treating this subject are available; however, a brief discussion is presented in3.3.20. ‘The method of calculated ouput corrections requires one 10 measure or estimate the magnitudes of the interfering and/or modifying inputs and to know quantitatively how they affect the output. With this information, it is possible to caleulate corrections, which may be added to or subtracted from the indicated output, so as to Leave (ideally) only that component associated with the desired input. Thus, in the manometer of Fig. 2.10, the effects of temperature on both the calibrated seale's length and the density of mercury may be quite accurately computed ifthe temperatures known. The local gravitational acceleration is also known for @ given clevation and latitude, so that this effect may be corrected by calculation, Since many measurement systems today can afford to include a microcomputer to carry cut ‘Various functions, if we also provide sensors forthe spurious inputs, the microcomputer can implement the method of calculated output corrections on an automatic basis, giving a so-called smart sensor,EER Decbetin's Measurement Sten ‘The method of signal fiterin is based on the possibilty of introducing certain clements (“filters”) into the instrument which in some fashion block the spurious signals, so that ther effects onthe output are removed or reduced. The filter may be applicd to any suitable signal in the instrument, be ‘output, or intermediate signal. The concept of signal filtering is shown schematically in Fig. the eases of input and output fitering. The application to intermediate signals should be obvious. In Fig, 2.13). the inputs andi, are caused to passthrough filters, whose input-output relation is ideally) zero, Thus, fand i are zero, even if j, and jy are not zero, The concept of output filtering is illustrated in Fig. 2.13(b). Here the output, though really one signal, is thought of es a superposition of o, (output duc to interfering input), 0, (output due to desired input), and 0, (ourput due 10 modifying input). If it is possible to construct filters that selectively block o, and o, but allow o, to pass through, this may be symbolized as in Fig, 2.13(b) and resuls in o consisting entirely of 0, oe oom Ch oa = anime Fig!2N3") General principle of filtering. ‘The filters necessary in the application ofthis method may take several forms; they are bes illustrated bby examples. If put directly in the path of a spurious input, 2 filter can be designed (ideally) to block completely the passage of the signal. If, however, its inserted ata point where the signal contains bosh desired and spurious components, the filter must be designed to be selective. That is, it must pass the desired components essentially unaltered, while effectively suppressing all others. ‘Offen it is necessary to attach delicate instruments to structures that vibrate. Electromechanical devices for navigation and control of aircraft or missiles are outstanding, examples. Figure 2.14 (a) shows how the interfering vibration input may be filtered out by use of suitable spring mounts. The 'mass-spring system is actually a mechanical filter, which passes on to the instrument only a negligible fraction of the motion of the vibrating structure ‘The interfering tilt-angle input to the manometer in Fig. 2.10(c) may be effectively filtered out by ‘means of the gimbal-mounting scheme of Fig. 2.14(b). Ifthe gimbal bearings are essentially frictionless, the rotations 0, and 9, cannot be communicated to the manometer; thus it always hangs vertical. ‘In Fig.2.14(c), the thermocouple reference junction is shielded from ambient-temperaturefluctustions bby means of thermal insulation, Such an arrangement acts as a filer for temperature or heat-low gage circuit of Fig, 2.14(d) is shielded from the interfering 60-Hz field by enclosing it in a metal box of some sort. This solution corresponds to filtering the interfering input. Another possible solution, wihich corresponds to selective filtering of the output, is shown in Fig. 2.14(e). For this approach to be effective, itis essential thatthe frequencies in the desired signal occupy a rangeGeneralzed Configureions end Funtinel Descriptions of Meeswring Instruments ce rosin |_O a . < teste aeema mene noue o 7 sranony FiglZA4) (2)-e) Examples of filtering. considerably separated from those in the undesired component of the signal. In the present example, suppose the strains to be measured are mainly steady and never vary more rapidly than 2 Hz. Them it is, possible to insert a simple RC filter, as shown, thet will pass the desired signals, but almost completely block the 60-H interference. Figure 2.14(0) shows the pressure gage of Fig. 2.2, modified by the insertion of a flow restriction ‘between the source of pressure and the pision chamber. Such an arrangement is useful, for example, ifEBB Docbeli's Measrement Systems ree psa Fam | ety Li. ‘Chooper Fpl Taare jin if jmoore, ao \ o Fig 2.44') (1nd (@ you wish to measure only the average pressure in large ar tank that is being supplied by a reciprocating ‘compressor. The pulsations inthe air pressure may be smoothed by the pncumatc filtering effect ofthe flow restriction and associated volume. The variation of the output-input amplitude ratio lx/p) with frequency issimilar to that forthe electrical RC filter of Fig 2.|4(e). Thus steady or slowly varying input pressures are accurately measured, while rapid variations are strongly attenuated. The flow restriction ‘may be in the form ofa ncedie valve, which allows easy adjustment of the filtering effet. ‘AT chopped” radiometeris shown in simplified formin Fig. 2.14(g) This device sensesthe temperature 7, of some body in terms of the infrared radiant energy emitted The emited energy is focused on a deiector of some sort and causes the temperature 7, ofthe detector and thus its output voltage cto vary. ‘The difficulty with such devices is thatthe ambient temperature, as well as 7, affects 7, This effect is serious since the radiant energy to be measured causes very small changes in 7, thus small ambient sits can completely mask the desired input. An ingenious solution to this problem intcrposes a rotating sihuter between the radiant source and the detector, so that the desired input is “chopped,” or modulated, av aknown frequency. Ths frequency is chosen tobe much higher than the frequencies ai which ambient dfs may occur. The output signal e, ofthe detector thus i a superposition of slow ambient flactuations and a high-frequency wave whose amplitude varies in proportion to variations in 7 Since the desired and interfering components are thus widely separated in frequency, they may be selectively filtered. In this case, we desire a filter that rejects constant and slowly varying signals, but faithfully reproduces rapid variations. Such a characteristic is typical of an ordinary ac amplifier, and since amplification isGeneralized Configuretions and Funetionol Descriptions of Meeuring increment BEB ‘necessary in such instruments in any en as shown solves two problems at once. In summing up the method of signal filtering, it may be said in general, itis usually possible to design filters of mechanical, ‘electrical, thermal, pncumatic, etc. ature, which separate signals according to their frequency content in some specific manner. Figure 2.15 summarizes the most common usefil forms of such deviees. the use ofan ac amplifier Method of Opposing Inputs ‘The method of oppasing inputs con- sists of intentionally introducing into the instrument interfering and/or modifying inputs that tend to cancel the bad effects of the unavoidable spurious inputs, Figure 2.16 shows schematically the concept for interfering inputs. The extension to modifying inputs should be obvious. The intentionally introduced input is designed so that the signal 9, ard o,, are essertally equal but actin the op posite sense; thus the net contribu paw to the ouput i nearly zero. This method ac- mo tually might be consider as a varia ‘ the method of calculated output corrections, tray However, the “calculation” and application of cane the corection ae achieved automaticaly ov om = ing to the stare of the system, rather than @ by numerical calcultin by human operator. ‘Thus the two methods are similar; however. cms the distinction between them is a worthwhile oer Zz one since it helps to organize your thinking in a inventing new applications of these general- ized corection concepts Fig. 2.16 | Methods of opposing inputs. ‘Some examples of the method of opposing — ‘npatsare shown in Fig 2.17. Amilivoltmeter, shown in Fig. 2170, is basically 2 curen-sensive device. However, a long asthe total circuit resistance is constant, its scale ean be calibrated in voltage, since voltage and curent ae proportional ‘A modifying input here isthe ambient temperature, since i causes the coll resistance R_, to change, thereby ltering the proportionality factor between current and voltage. To correct for tts eon the compensating resistance Ri inroduced int the circuit, and its materia is caeflly chosen to have ‘temperature coefficient of Teitance ¢pporit to that of R,. Thus when the temperature changes, he total resistance ofthe circuit is unaected andthe calibration ofthe meter remains accurate Figure 2.17(b) shows a statc-pressue-probe design de to L, Pra. As the Maid flows over the surface of the probe, the velocity of the uid must ierease since these sreamtines are longer than those in the undisturbed flow. This velocity increase causes a drop in static pressure, so that a tap in the surface ofthe probe gives an incorrect reading. This underpessureeror varies withthe distanced, of the tap from the probe tip. Prandil recognized that the probe support wil have a stagnation pont (ine) along its front edge and that this overpressure will be felt upstream—the effect decreases as the distanceHEEB Dovbetin's Measurement Stem: a epg = De ns {j= we | SEE Sal tae Fig. 247 ) (0) Examples of method of opposing inputs. ocerrsire eto ow oversamac ee (nef = Psa iPestes ran FeBa7) wGencrlized Configurations and Furtina! Descrptions of Measuring Instruments 4, increases. By properly choosing distances d, and d, (by experimental test), these two effects can be ‘miade exactly to cancel, giving a true static-pressure value atthe tap, ‘Adevice forthe measurement ofthe mass flow rte of gases is shown in Fig 2.17(c). The mass flow rate of gas through an orifice may be found by measuring the pressure drop across the orice, perhaps by ‘means ofa U-tube manometer. Unfortunately, the mass flow rate also depends onthe density ofthe gas, which varies with pressure and temperature. Tous the pressure-drop measuring device usually cannot be calibrated to give the mass flow rate, since variations in gas temperature and pressure yicldéifferent mass flow rate for the same orifice pressure drop. The instrument shown in Fig. 2.17(c) overcomes his problem in an ingenious fashion. The flow rate through the orifice also depends on its How area, Thus 3f the flow area could be varied in just the right way this variation could compensate for pressure and temperature changes So that a given orifice pressure drop would aways corespond t0 the same mass flow rate. This is accomplished by attaching the specially shaped metering pin toa gas-filled bellows as shown. When the temperature drops (causing an increase in density and therefore in mass low rate), the gasinthe bellows contacis, which moves the metering pin imo the orifice and thereby reduces the flow area. This retums the mass flow rate to its proper value. Similarly, should the pressure ofthe flowing gas ‘increase, eausing an increase in density and mass flow rate, the gas-filled bellows would be compressed again, reducing the low area and correcting the mass flow rate. The proper shape forthe metering pin is revealed by a detailed analysis ofthe system. ja —etoctet a [Sostenperare ence ‘Meatyroinnaty Lf esto (Desiedinea Figen) A final exanple of the method of opposing inputs is the rate gyroscope of Fig. 2.17). Such devices are widely used in acrospace vehicles forthe generation of stabilization signals in the control system. The action of the device is that a Vehicle rotation at angular velocity 8, causes a proportional displacement 6, ofthe gimbal relative tothe case. This rotation 6, is measured by some motion pickup (not shown in Fig, 2.17(@). Thus a signal proportional to vehicle angular velocity is available, and this is useful in stabilizing the vehicle. When the vehicle undergoes rapid motion changes, however, the angle 8, tends tooscillate, giving an incorrect angular-velocity signal. To contro these oscillations, the gimbal rotation is damped by the shearing action ofa viscous silicone fluid ina narrow damping gap. The damping effect varies withthe viscosity ofthe fluid and the thickness of the damping gap. Although the viscosity of the silicone fluid is fairly constant, it does vary with ambient emperature, causing an undesirable changeDocbetin’ MewsurementSytens in damping characteristics. To compensate for this, a nylon cylinder is used in the gyro of Fig. 2.17(4). ‘When the temperature inereases, viscosity drops, causing a loss of damping. Simultancously, however, the nylon cylinder expands, narrowing the damping gap and thus restoring the damping to its proper value. By proper choice of materials and geameiry, the two effects ean be made to very nearly cancel cover the operating temperature range of the equipment. a = nest Fanm}e— Erect otwseosty ‘ort soe Woden “| GL | erect eoticress | Soe era vec = a (Bociedinpat CONCLUSION CD einer eset erertton with cand te tion clement and the input-output configurations of measuring instruments and systems, Inthe analysis of @ given instrament or inthe design of a new one the staring point isthe separation ofthe overall operation into its functional elements. Here you must take a bread view of whar must be done, bu not be concerned with how itis aetually accomplished. Once the general functional concepts have been. clarified the details of operation may be considered fritfilly The ideas of active andpassive transducers, analog and digital modes of operation, and null venus deflection methods give a systematic approach for either analysis or desig. Finally, compensation of spurious inputs and detailed evaluation of performance are facilitated by application of input-output block diagrams. These configuration diagrams show cleaty, which physical analyses must be made to evaluate performance, with respect to accurate measurement ofthe desired inputs and rejection of spurious inputs. The evaluation ofthe relative quality of different instruments (oFGeneralized Configuration: and Functions Descriptions of Meoeurng inetruments EB the same instrument with different numerical parameter values) requires the definition of performance criteria against which competitive designs may be compared. This isthe subject of Chapter 3. a 2A Make block diagrams such as Fig, 2.1, showing the functional elements of the instruments depicted in the following: @ Fig.27. © Fig. 28. (©) Fig.2.10(0 (Fig. 2.11, Take Fes input and ¢, as ‘output. (© Fig. 2.14(@). Toke 7,as input and e, as output. Fig. 2.17(b). Take V as input and manometer Vas output. (@) Fig.2.17(@). Toke @ as input and 6, 4s output. 2.2 Identify the active transducers, if any, in the instruments of (a) Fig. 2.8, (b) Fig 2.10(9), () Fig. 2.11, (d) Fig, 2.1710), (€) Fig. 2.176). 23° Consider a man, driving a car along a road, who sees the opportunity to pass and decides to accelerate. (2) If the light waves entering his eyes are considered input) and accelerator-pedal travel is taken as ‘output, isthe man functioning as an active ora passive ransducer? (8) If the acceierator-pedal travel is considered input and ear velocity 45 output, isthe eutomobile engine ‘mactive ora passive transducer? 24 28 26 2 2s 2 210 Give an example of a null method of force measurement. Give an example of @ null method of voltage measurement, Sketch and explain two possible modifications of the systern of Fig. 2.4 that will allow measurement to 1/10 revolution. Identify desired, interfering, and modify- ing inputs for the systems of (a) Fig. 2.2, (Fig. 23. (© Fig. 24, () Fig. 2.5. ‘Why is tlt angle in Fig. 2.]0camodifying input? Suppose in Eq. (2.2) that Ky, = Ky. 6, = 1.0, Now let K,, change By 10 percent to 1.1. What is the change in s,7 In Eq. 25), let Ky, = Ky = Kyy =0,= 1.0, and K,, = 100" Now Tet Ky. change by 10 percent to 1.1. What isthe change in x, Investigate the effect of similar changes in K gp Kye and Kipp ‘The natural frequency of oscillation of the balance wheel in a watch depends ‘on the moment of inertia of the wheel nd the spring constant of the (torsional) Iaispring. A temperature tise resus in ‘8 reduced spring constant, which lowers the oscillation frequency. Propose a ‘compensating means for this effect. Non tomperature-tensitive hairepring material {snot an acceptable solution,Generalized Performance Characteristics oF Instruments INTRODUCTION ERD izes ere tn comers sae inners. te onemos uate for a proposed measurement, ox, altematively, if you are engaged in the design of instruments for specific measuring tasks, then the subject of performance criteria assumes major proportions. That is to make intelligent decisions, Unere must be some quantitative basis for comparing one instrument (or proposed desigh) with the possible alteratives. Chapter 2 has served as a useful preliminary to these considerations since there, we developed systematic methods for breaking down the overall problem into is component pars. Now, we propose 1 study in considerable desi, the performance of measuring insnuments and systems, with regard to how well they measure the desired inputs and how thoroughly they reject the spurious inputs. “The treatment of instrument performance characteristics generally has been broken down into the subaceas of static characterises and dynamic characteristics, and this plan is followed here. The reasons for such a clasification are several. Fist, some applications involve the measurement of ‘quantities that are constant or vary only quite slowly. Under these conditions, itis possible to dane a fet of performance criteria that give considerable description of the quality ef measurement, without becoming concerned with dynamic descriptions involving differential equations. These csitevia are called the stati characteristics, Many other measurement problems involve rapidly varying quantities Here, the dynamic relations between the instrument input and output must be examined, generally by the use of differential equations, Performance criteria hase! on these dynamic relations constte the dynamic characteristics. ‘Actually, static characteristics also influence the quality of measurement under dynamic conditions, tut the static characteristics generally show up as nonlinear o statistical effects inthe otherwise linear differential equations giving the dynamic characteristics, These effets would make the differentil exqations analytically unmznageable, and so the conventional approach isto treat the vo aspects of the problem separately. Ths the differential equations of dynamic performance generally neglect the cffcets of dry friction, backlash, hysteresis, statistical scatter, ctc., even though these effects influence the dynamic behavior. These phenomena are more convenienly studied as static characterises, andthe overall performance ofan instrument is then judged by a semiquantitative superposition of the staticGeneralized Performance Characteristic of Instruments ‘and dynamic characteristics. This approach is, of course, approximate but 2 necessary expedient for ‘convenient mathematical sudy. Once tentative designs and numerical values are available, we can of ‘course use simulation to investigate the nonlinear and statistical effet. STATIC CHARACTERISTICS AND STATIC CALIBRATION ea 3.2.1 Meaning of Static Calibration ‘We begin our study by considering the process of static calibration since all the static performance characteristics are oblained by one form or another of this process. In general, static calibration refers to a situation in which all inputs (desired, interfering, modifying) except one are kept at some constant values. Thea, the one input under study is varied over some range of eonstant values, Which causes the outpul(s) to vary over some range of constant values. The input-ouiput relations developed in this way comprise static calibration valid under the stated! constant conditions of all the other inputs, This procedure may be repeated, by varying in tum each input considered to be of interest, and thus developing. family of static input-output relations. Then we might hope to deseribe the overall instrument static behavior by some suitable form of superposition of these individual effects, In some ‘cases, if overall rather than individual effects were desired, the calibration procedure would specify the variation of several inputs simultaneously. Also if you examine any practical instrument criteally, you will find many modifying and/or interfering inputs, each of which might have quite small effects and Which would be impractical to control. Thus the statement “al ober inputs are held constant refers to an ideal situation which can be only approached, but never reached, in proctice. Measurement method describes the ideal situation, while measurement process deseribes the (hmperteet) physical realization of the measurement method, The statement that one input is varied and all others are held constant implies that all these inputs are determined (measured) independently ofthe instrument being calibrated. For interfering or modifying inputs (whose effects on the output should be relatively smal in a good instrument), the measurement of these inputs usually need not be at an extremely high aecuracy level. For example, suppose a pressure gaye has temperature as an interfering input to the extent that a temperature change of 100°C causes pressure error of 0.100 percent. Now. if we had measured the 100°C interfering input witha thermometer Which itself had an error of 2.0 percent, the pressure error setually would have been should be elear that the difference between an error of 0.100 and 0.102 percent is et in most engincering situations. However, when calibrating the response of the instrument tits desired inputs, you must exercise considerable care in choosing the means of determining the numerical values of these inputs. That is, if pressure gage is inherently capable of an accuracy of 0.| percent, you must certainly be able to determine its input pressure during calibration, with an accuracy somewhat greater than this. In other words, itis impossible to calibrate an insirument ion accuracy greater than that of the standard with which itis compared. A rule often followed is that the calibration system (ihe standard and ‘any auxiliary apparatus used with it) have a total uncertainty four times bettor than the unit under test! Details about standards for specific physical variables will be given atthe beginning of the chapter devoted to that variable. Here, we want to give some general information. A standard for a certain physical variable is often “just another” measuring device for that variable, However, 1o be called a standard, its accuracy must be at a higher level than the instrument to be calibrated, the 4-0-1 ratio ISO Guide 25, ANSUNCSL 2540-1Doebein's Measurement Systems ‘mentioned above being a common requirement. There is actually a hierarchy of standards which arranges them in order of decreasing accuracy, with primary standards being the most accurate (atthe top of the rarchy). Primary standards are considered the “state-of-the-art,” that is, the most accurate way known tomeasure the quantity of interest. Such standards are developed, maintained, and improved by national laboratories such as the National Institute for Standards and Technology (NISTY in the United States. ‘These labs also may provide calibration services to industry or other customers. Large, “high-tech” companies sometimes maintain calibration laboratories, which also are eapable of calibration at this highest level. ‘AS you might guess, primary standards tend to be complex and expensive and are needed only for the most critical situations. Thus we need /ower feve! (secondary, tertiary, etc.) standards, which ace simpler and cheaper to use for most engineering calibration work. Such standards are available for calibration service at the national laboratories, commercial calibration laboratories, and in-house calibration laborstories, associated with industrial companies, universities, ete. When planning a specific experimental project, we need to decide how accurate our measurements need to be, and then arrange ‘o calibrate eoch instrument against a standord that is about four times more accurate, if possible. Thus, if we noed a | percent accuracy ina pressure gage, we need to calibrate it against a standard accurate to about 0.25 percent orbetier. OFcourse, the gage must be capable of 1 pereent accuracy. [fit has random. crtors of, say 3 percent, calibrating it against a 0.25 percent standard will noi make it aI percent gage.” This, of course, would be discovered during calibration, but we do not want to waste our time, s0 our inital selection of instruments must be carefully made. In performing a calibration, the following steps are necessary: 1, Examine the construction ofthe instrument, and identify and lst all the possible inputs, 2. Decide, as best you can, which of the inputs willbe significant in the application for which the instrument is to be calibrated. 3. Procure apparatus that will allow youto vary all the significant inputs over the ranges considered necessary, Procure standards to measure each input 4. By holding some inputs constant, varying others, and recording the ouiput(s), develop the desired static input-output relations. [Now we re ready fora more detailed discussion of specific static characteristics. These characteristics may be classified as either general or special. General static characteristics are of interest in every instrament, Special static characteristics are of interest in only a particular instrument. We concentrate ‘mainly on general characteristics, leaving the treatment of special characteristics to later sections of the text in which specific instruments are discussed. 3.2.2. Measured Value versus True Value ‘When we measure some physical quantity with en instrument and obtain @ numerical value, usually we are concemed with how close this value may be to the “true” value. It is fist necessary 10 understand ‘that this so-called true value is, in general, unknown and unknowable, since perfectly exact definitions of the physical quantities to be measured are impossible. This can be illustrated by specific example, for instance, the length of a cylindrical rod. When we ask ourselves what we really mean by the length of this rod, we must consider such questions as these: 1. “Are the two ends of the rod planes? 2. Ifthey are planes, are they parallel? ‘Formerly (before 1989) called the National Bureau of Standards (NBS).Generalized Performance Characteristics of Instruments WEE 3. lfthey are not planes, what sort of surfaces are they? 4. What about surface roughness? ‘We see that complex problems are introduced when we deal with a real object rather than an abstract, ‘geometric solid. The term “true value,” then, really refers to a value called the reference value that ‘Nould be obtained if the quantity under consideration were measured by an exemplar method. th ‘method agreed on by experts as being sufliciently accurate for the purposes to which the data ultimately ‘will be put. ‘The measurement process consists of acially carrying out, as well as possible, the instructions for performing the measurement, which are the measurement method, (Since calibration is essentially a refined form of measurement, these remarks apply equally to the process of calibration.) If this process. is repeated over and over, under assumed identical conditions. we get a large number of readings from, the instrument, Usually these rsadings will not all be the sume, and so we note immediately that we may ‘ny to ensure identical conditions for each tral, but it is never exactly possible, The data generated in this fashion may be used to describe the measurement process so that, iit is used in the furure, we may bbe able to attach some numerical estimates of error to its outputs. Ifthe output data are to give significant description of the measurement process, the data must form. what is called 2 random sequence. Another way of saying this is that the process must be in a state of statistical contol: The concept of the state of statistical control is not a particularly simple one, but we try to explain its essence briefly. First, we note that itis meaningless to speak of the accuracy (ofan instrument as an isolated device. We must always consider the instrument plus its environment and method of use, tha is, the instrument plus its inputs. This aggregate constitutes the measurement process. Every instrument has an infinite number of inputs; tht is, the causes that can conceivably affect the output, iFonly very slightly, are limitless. Such effects as atmospheric pressure, temperature. and humidity are among the more obvious. But if we are willing to “spit hairs.” we can uncover a multitude of ether physical causes that could affect the instrument with varying degrees of severity. In defining a calibration procedure for a specific instrument, we specify that certain inputs must be held “constant” within certain limits. These inputs, it is hoped, are the ones that contribute the largest components 10 the overall error ofthe instrument. The remaining infinite number of inputs i Ie uncontrolled, and it is hoped that each of these individually contributes only a very smal effect and that inthe aggregate their effect on the instrament output will be ofa random nature. If this is indeed the case, the process is seid tobe in statistical control. Experimental proof that a process is in statistical control isnot easy to come bby; in fact, achieving sorcr statistical control is unlikely. Thus we can only approximate this situation, Lack of control is sometimes obvious, however, if we repeat a measurement and plot the result (output) versus te trial number. Figure 3.1(a) shows such a graph for the calibration of a particular instrument. In this instance, it was ascertained after some study that the instrument actually was much _more sensitive to temperature than had been thought, The original calibration was carried out in a room ‘without temperature control. Thus the room temperature varied from a low in the moming to 2 peak in the early aftemoon and then dropped again in the late afternoon, Since the 10 trials covered a period of about one day, the trend of the eurve is understandable. By performing the calibration in a temperature- controlled room, the graph of Fig. 3.1(b) was obiained. For the detection of more subtle deviations from statistical control, the methods of statistical quality-control charts arc useful.’ * Churchill Eiseahart, “Realistic Evaluation ofthe Precision and Accuracy of Insirament Calibration Systm.” J Res Natl Bur. SiC, vol. 67C, no. 2, April-June 963 1b SEB, Wilson, J, "An Introduction to Sciemific Research,” Chap. 9, McGraw-Hill, New York, 1982WIM doebetin's Measurement Systems Insert ouput Tot co ® Fig. 3.1) Effect of uncontrolled input on calibration. ‘Ifthe measurement process is in reasonably good statistical control and if we repeat a given measure- ‘ment (or calibration point over and over, we will eneraea set of data exhibiting random scatter. AS an example, consider the pressure gage of Fig. 3.2. Suppose we wish to determine the relationship between the desired input (pressure) and the output (scale reading). Other inputs which could be significant and Poster uno Eade Pape Pianos Som Pree, t Fora —£ sce tate, weson_f (al fe ase re pe eee ed == Figh 312) Pressure goge.Generalized Performance CharcterisiceofInstroments ‘which might have tobe controlled during the pressure calibration include temperature, zeceleration, and vibration. Temperature can eause expansion and contraction of instrument pars in such a way that the scale reading will change even though the pressure has remained constant, An instrument acceleration slong the axis of the piston rod will cause a seale reading, even though pressure again has remained unchanged. This input is significant ifthe pressure gage is to be used aboard a vehicle of some kind. A small amount of vibration actually may be helpful to the operation ofan instrument, since vibration may reduce the eflect of static friction. Thus, ifthe pressure gage i to be attached toa reciprocating air ‘compressor (which always has some vibration), it may be more accurate under these conditions tha ‘would be under calibration conditions, where no vibration was provided. These examples illustrate the _zeneral importance of carefully considering the relationship between the calibration conditions and the acwal application conditions 3.2.3 Some Basic Statistics Suppose, now, that we have procured a sufficiently accurate pressure standard and have stranged to ‘maintain the other inputs reasonably close to the actual application conditions. Repeated calibration at. a given pressure (say, 10 kPa) might give the data of Fig. 3.3. Suppose we now order the readings from. the lowest (9.81) to the highest (10.42) and see how many readings fallin each interval of, soy. Table3.1) Pressure gage citation data “True pressure (reference value) = 10.000 = 0.001 kPa Acceleration = 0 Vibration level = 0 ‘Ambient temperature = 20 + 1°C a8 wooo Hoa Fig. 3.3) Pressure-goge calibraton cate.SR Cocbetin's Mean starting at 9.80, The result can be represented graphically as in quantity Z by" Suppose we now define the (number of readings in an interval) total number of readings) aD e ‘width of interval and we plot a “bar graph” with height Z for each interval. Such a “histogram” is shown in Fig, 3.4(b). ‘Any spreadsheet software can plot histograms for us. It should be clear from Eq (3.1) that the area ofa particular “bar” is numerically equal to the probuifi that a specific reading will fallin the associated interval. The arca of the entie histogram must then be 1.0 (100 percent ~ 1.0), since there is 100 percent probability that the reading will fall somewhere between the lowest and highest values at least based on the data available. ft were now possible o take an infinite number of readings, each with an infinite number of significant digits, we could make the chosen intervals as small as we pleased nd still have each interval contain a finite number of readings. Thus the steps in the graph of Fig. 3.4(b) would become smaller and smaller, with the graph approaching a smooth curve in the limit. If Namba ot | Af] ‘Ssierading Pa : NN Z VX UN) N GaP 985 O55 995 1909,90st0 19Ta S10 20107510 ISTO TOASTOSD 7 Seen reasng tPa © Fig. 3.4} Distribution of data The symbol) means “equal by definition.”Generaized Performance Choracteristic of heiruments ER ‘we take this limiting abstract case as a mathematical woe ‘model for the real physical sitation, the function fos) is called the probability density function forthe ‘mathematical model ofthe real physical process (see 3.5(a)). From the basic definition of Z, clearly ° > Probability of reading lying between a end b fe « aMacxesyp foode 2) ‘The probability information sometimes i given interms ofthe cumulative distribuion finetion Fs), whieh is defined by F(x) A probability that reading is less than any chosen value of x Flx)= [onde and is shown in Fig 3.5(b). From the infinite number of forms possible for probability density functions. a relatively small ‘number are useful mathematical models for practical applications; in fat, one particular form is quite dominant. The most useful density function is the normal or Gauasian function, which is gven by Probability distribution function. 63) a)= moexctes F SO)= Fe 4) ‘A new variable 7 can be defined as tre ts 1h Ba) ‘The cumulative probability density function Fix) can be proved to be equal to P< y= Fx)= fl er(p] 4b) where erfis known as the ertor function that ean bbe expressed in terms of standard integral given B40) The error function can be evaluated and plotted as in Fig. 3.6. The error function for positive values of 7 is shown in Table 3.1. For vegative values of 7, since the error function i¢ odd, erf(-n) = ~erfin). Therefore. this table is Useful in determining the cumulative probability of Gaussian distri reeset a Fig: 3.6 | Error function of equation (3.4e)SER oebetin's Mewsurerent Systems In addition, using the ervor function, we can also determine the probability of obtaining 2 certain range of values as Posy = 6 =) readings will ogcur. This is one of the reasons why a tue Gaussian distribution can never occur in the real oes world; physical variables are always limited to EBA) austen eistrbutton, fini values, There is zero probability, for example, thatthe pointer on a pressure gage will ead 100 ” iPa when the ange of the gage is only 20 kPa, Real distributions must ths, in general, have their “ails” cut off asin Fig. 3.8. ‘Although aswal data will not conform exactly : to the Gaussian distribution, very often they are sufficiently close to allow use of the Gaussian Fig. 3.8} Non-Gaussian distribution. ‘model in engineering work. It would be desirable to have available tests that would indicate wheter the data were “reasonably” close to Gaussian. We ‘must admit however, tht in much practical work the ime and effort nesessay for such ets cannot be justified and the Gaussian model is simply assumed until troubles arise which justify a closer study of the patculr situation ‘One method of testing fr an approximate Gaussian distribution involes the use of probability graph paper. If we take the eumulatve distribution funetion fora Gaussian distribution and suitably distort the tertcal scale ofthe graph, the curve can he made to plot 28 a straight line, as shown in Fig. 39. (This of cours, can be done with any curvilinear relation, not ust probability curves.) Such graph paperGeneralized Performance Characteristic of Instruments ‘commercially available and may be used to give a rough, but useful qualitative test for conformity tothe Gaussian distribution. (If you have aecess to statistical sofiware, special graph paper is not needed: the software plots your points to the proper seales.) For example, consider the data of Fig. 3.3, These data ‘may be plotted on Gaussian probability graph paper as follows: First lay out on the uniformly graduated horizontal axis a numerical scale that includes all the pressure readings. Now the probability graph paper represents the cumulative distribution, so that the ordinate of any point represents the probability ‘that a reading will be less than the abscissa of that particular point. This probability, in terms of the sample data available, is simply the percentage (in decima! form) of points that fell at or below that particular value, Figure 3.10(a) shows the resulting plot. Note thatthe highest point (10.42) cannot be plotted since 100 percent cannot appear on the ordinate seale. Also shown in Fig 3.10(a) is the “perfect ‘Gaussian fine” the sirsight Fine thar would be perfectly followed by data from an infinitely large sample ‘oF Gaussian data whieh had the same sand 6 valucyas our setual data sample. To plot this line, we must ‘estimate j from the sample mean value X, using 5) where Q individual reading N A total number of readings and @ from the sample standand devtation s. using 66H ‘The data of Fig 3.3 give = 10.11 kPaands= 0.14 kPa, Two points that may be used to plot any perfect Gaussian line are (X, 50%) and (Y + 5, '84.1%), which yield the lin of Fig 3.9(a) for our data, Superimposing this line on our actual data, ‘we may judge visually and qualitatively whether ‘our data are “close to” or “far from” Gaussian. Note that there is no hope of ever proving real-world data tobe Gaussian. There are always physical constraints, which require real dats (o be at Feast somewhat ‘non-Gaussian, Since our graphical testis clearly subjective, it may be useful to provide some additional data to hnolp each the required decision. Figures 3.10(b) and (e) show “tdta” generated by a digital-computer Gaussian random-number generator. Such random-number algerithms are excellent simulations of perfect Gaussian distributions. Figure 3.10(b) graphs data generated when the program was asked 10 ‘Produce a sample of 20 readings with = 10.11 kPa and = 0.14 KPa. Note that even though the sample was drawn from a perfect Gaussian distribution, the points do not fall on the perfect Gaussian line. This does not mean thatthe computer algorithm is fal. It simply shows that a sample of size 20/is too small to give a statistically reliable prediction. In Fig. 3.10(c), the sample size was increased ta 100, which gives a clear improvement but still not perieetion. The results of Fig. 3.10(a) would be Fig.'3.9 ) Rectification of Gaussion curve.EER bocbetn’ Measurement Systems Ee gsdssas Perenage ct ecg ator blow abi aot ‘Vl chan ec rain, KPa ei Fig. 3.10 ) (0) Graphical check of Gaussian astribution. considered by most people to indicate a reasonable approximation to a Gaussian distribution. However, this testis clearly only qualitative, and its main usefulness perhaps is in eevealing gross departures from the theoretical distribution. Such deviations would lead us to examine the instrument and measurement process mare closely before attempting to make any statistical statements about accuracy. For a perfect Gaussian distribution, it can be shown that 68% of the readings lie within £10 of w 98% of the readings lie within £26 of en 99.79% of the reading lie within #30 of Thus, if we assume that our real distribution is nearly Gaussian [Refer to Fig. 3.3(6)], we might predict, for instance, that if more readings were taken, 99.7 percent would fall within £ 0.42 kPa of 10.11 KPa, Since the purpose of calibration is to convert instrument readings into estimates ofthe true value, let us suppose we now use the calibrated gage to measure an unknown pressure and the gage hhappens to read 10.11 kPa. Based on the calibration, we would say our best estimate of the true pressure {is 10.00 kPa, but this value is uncertain. We cannot remove this uncertainty, but we can quantify it. We are 68 percent sure thatthe true value is somewhere within + 0.14 of 10.00, 95 percent sure i is within £0.28, and 99.7 percent sure itis within #0.42 kPa of 10.11 KPa. FEO, Dosbel jineering Experimentation,” McGraw-Hill, New York, 1995,Generalized Performance Characteristic of Instruments SEB uessezes 3 roms e2| ‘canen iow oa Fora 10%) (rte 20h pt oot ‘ae fan ical eae. Pa Fig3.10) (0) AENITAD sate Fore ‘cass oe (Gere 0 acl sont f= 1028 = 13) Vale a i raing, a © Fig 3.105)EDR coebetin's Meaurement Systems instrument bias 4 reading — reference ("inve”) value 8) “The instrument bias is (10.11 ~ 10,00) = 40.11. The basic precisfon index iss (0.14), but it has been. ‘common practice to quote the imprecision as + 3s. Figure 3.3(b) displays these definitions. ‘When cumulative distribution graphs such as those in Fig. 3.10 are plotted, it can be shown’ that it is more correct to plot the fh point of an n-point sample, not as (Jin)100 percent, but rather as i-03 tas 100 percent. This also allows the plotting of the “last” point, which would otherwise always (& 23) pe plotting oi v be at 100 percent and thus not pletable. While this procedure gives a useful improvement for samples (of 15 oF les, it has almost no effect when, asin Fig. 3.10, samples of 20 or more are used. A somewhat more quamitative method for deciding whether data are close 10 Gat ehissquare (7) zoodiness-olfit test The reference explains this test but concludes “Gaussian graph paper” method; thus we pursue it no further. sian involves the is inferior w our 3. .4 Least-Squares Calibration Curves Up to now, we have been examining the situation (Fig. 3.2(b)) in which a single true value repeatedly and the resulting measured values are recorded and analyzed. In an actual calibration, the true value is varied, in increments, over some range, causing the measured value also to vary over a range. Very often there is no multiple repetition ofa given true value. The procedure is merely to cover the desired range in both the inereasing and the decreasing directions. Thus a given true value is applied, at most, vice if we choose to use the same set of true values for both increasing and Aecreasing readings. ‘Asan example, suppose we wish to calibrate the pressure gage of Fig, 3.2 forthe relation between. the desired input (pressure) and the output (scale reading). Figure 3.11(@) gives the data for such a calibration over the range Oto [0 kPa. [nthis instrument (asin most but not all), the input-output relation is ideally a straight line. The average calibration curve for such an instrument generally is taken as a straight line which fits the seattered dats points best as defined by some chosen criterion. The most common isthe least-squares criterion, which minimizes the sum of the squares ofthe vertical devi of the data points from the fitted line. (The least-squares procedure also can be used to fit curves other ‘than straight lines to scattered data.) The equation forthe straight line is taken 2s mg, +b @s) where 4, 4 outpu quantity (dependent variable) 4g, 4 input quantity (independent variable) im slope of ine ‘A intercept of line on vertical axis ‘The equations for calculating m and b may be found in several references.” TC Lipson and N. J. Sheth, “Statistical Design and Analysis of Exgincering Experiments,” MeGraw ill, New York, 1973, p18, B, O. Docbelin, “Engineering Experimentation,” McGraw-Hill, New York, 1995, p. SS. “HD. Young, “Statistical Treaument of Experimental Data,” McGraw-Hill, New York, 1962, p. 121Generalized Performance Characteristics of Instruments CHM G0) px Ea Ea?) ~ (Ea) (Ea) NEq? Ga)" ‘where N 8 otal number of datapoints By calculating this example, we get m = 1.0823 and 6 =-0.8471 kPa. Since these values are derived from scattered data, it would be useful to have some idea of their possible variation. The standard ncstosyp 12” 1063 4, 0.312 (i) Standard deviation ofthe output, 5, = 068 (Eq.3.14) Standard deviation ofthe slope, S700 (Eq.3.12) Standard deviation ofthe intercept, S, = 038 (Eq. 3.13) Standard deviation ofthe input. 5, = 064 (Ea. 3.16) (i) Ife insinament ead, 25.35 MPa, ve vals of pressure a or ths inca equation i given by 133 Dyn 25 35 MPa and is uncertainty willbe $508, = 192 MPa,TELM octets Measurement Sites 2-0 Limits in Defining imprecision ‘While the method just described was used for many years and does not cause any major problems, a more refined technique is available and is now recommended by national and international standards sroups.! The improvements have two major features Firs, our simple method considers a standard eviation calculated from a small numberof points to be as accurate as one gotten from a large number of points. Intuition alone tells us that this isnot true. Statistical theory (confidence intervals)" allows. Us o adjust the uncertainty fo suit the numberof points. The second improvement substitutes for our 35 limits (99.7 percent level limit analogous to 2s (95 percent level). This change can be justified without complex statistical arguments a follows. When we use + 3s limits, we imply that our uncertainty band, is 99.7 percent sure to include the correct value. For Gaussien end near-Geussian distributions, 99.7, percent puts us well into the “tails” ofthe distribution. In the tail regions it takes ver’ large samples to get reliable results for probabilities. In most engineering experiments (including calibration), the sample sizes are much too small to make any reliable predictions about the til regions. For example, to actually compute a 99.7 percent probability from a sample of real data would require about 1000 points (997 Within the band and 3 outside). If we had only, say, 100 points, we might get a 99 percent probability 9;n, 1 out) but we could nor compute a 99.1 percent mit. You just cannot compute valid probabilities Table 3.3} t-isiibucon values for uncertainty caution, ‘Drcetes oF rRszoow w-2 ‘Diansss oF Fréeow *B.W.Taylorand C. E, Kuyatt "Gi Results” NIST Tech. Note 1297, 1993, sec. 65; “Gaide tothe Expression of Uncertainty in Measurement,” Ist ¢d, 1SO, 1993, see. G12; "Measurement Uncertainty,” ANSVASME PTC 19.1-1986 Patt 1, 1986; R.B. Abernethy, “*Messurement Uncertainty Handbook.” Instrument Society of America, 1980: H.W. Coleinen “Experimentation ‘and Uncenainty Analysis for Engineers.” 2nd ed, Wiley, New Yor, 1999; “ASTM Standards on Precision and Bias for Various Applications.” Sth ed. ASTM, 1997. 'E.O. Doebelin, “Engincering Experimentation.” p $20.Generalized Performance Characteristic of Intraments ESB resolved to tenths of | percent unless you have sbout 1000 or more points. With samples of size about 20, if we get | of the 20 outside the band, that would be 5 percent outside and 95 percent inside. Since ‘most engineering samples are relatively small, it was decided that quoting uncertainty bands at the 95 percent level of confidence was more realistic than using the older 99.7 percent criterion. Since our present interest isin specifying uncertainty for readings taken from calibration eurves, let us now define a recommended procedure which follows the spirit of the “new” methods, We will still find the best fit calibration line by the standard least-squares method given earlier. When we use this calibration line to get a true value from a measured value, we will still use Eq. (3.15). However, we need ‘anew method for calculating the uncertainty band that surrounds our nominal value. Mandel" provides 4 method which meets our needs. To make our uncertainty band ("confidence interval") sensitive tw sample size, we will need to use another of the common probability distributions, the so-called ‘distribution. Since we have decided that we want a 95 percent level of confidence, we will need only an abbreviated table (Table 3.3) of this distribution, Mandel’s Eq. (12.35) gives a+ 95 percent confidence interval, defined by two hyperbolss on either side of the leastsquares lines. When we read the instrument (7,), we draw a horizontal line through thot value. Ths line intersects the two hyperbolas, ‘and the , values ai these two intersections define the ends of a 95 percent confidence interval forthe true value, That is, we are 95 percent sure that the true value fies between these tivo valucs of g, ‘The “vertical” location of the two hyperbolas asa function ofg, is computed from a7) 24,)" is, the upper hyperbola is plotted by adding the positive Ag, tothe best-fit line, and the lower one is plotted by adding the negative one. The numerical valuc of,” is taken from thet table, using Was the total number of points on the calibration graph (22 points in Fig. 3.11(a) makes WV~ 2~ 20 and the | value = 2.086). The standard deviation s, is computed from Eq. 3.14), and q, has the same meaning asin Fig. 3.11(@). In some applications, we arc able to repeat the measurement 1 times (n= 2if two 4, readings are taken) and use forthe q, value the average ofthe readings. If this is done, the confidence imerval for g, will be smaller (bee). asthe formula shows. We now want to use the new method tO compute the uncertainty in , for the same g, value as used earlier and then compare this with an “old method” 2s, value of 2) (0.192) = 0.384 Using Fj. G.17) (with » = 1), we ean compute the values needed to plot the graph of Fig. 3.12 Visually, the two “hyperbolas” seem tobe straight lines, but inspection of the tabular results shows that 2a, does. vary with g, the largest value being 0.376 (atthe left and right ends of the curves) and the smallest boing 0.350 atthe center. Fora, reading of 4.32, the 95 percent confidence band fq, i found to be 4.78 40.392, Comparing the “old” and “new” methods. we note that the old method gives a fied! size othe confidence interval while the new will give slightly diferent value for diferent readings ‘Also, the numerical values are somewhat different 0.384 versus 0.392 in our example ‘While we have given various computing formulas for these curve-fitting and uncertainty calculations, many engineers nowadays have on their personal computers statistical software which makes the computing and graphing quick and easy. Such a calibration curve ¥, shown in Fig. 3.13, Because a ‘major assumption of the analysis is thatthe statistical variability ofthe measurements isthe same over 15), Mandel, “The Statistical Analysis of Experimental Data,” Wiley, NY, 1964, p. 286. A papesback rep (Dover, 1984) was sil available in 1999SFR Doebetin’ easement Syters the entre calibration range, itis good practice to also make a plot ofthe residuals versus, Such a graph can show whether ths assumption is reasonable. For our example, Fig. 3.14 shows no obvious trend in the size ofthe residuals: the variability seems to be about the samz over the whole range, When there is clear indication that the variability is not constant, other methods are available." However, these are rarely needed, and we thus leave them tothe reference Draw a horizontal lin of the measured value of, It intersections withthe prediction bends define 1.95% confidenc interval for true pressure g, Fig. 3.12 Colibrotion curve with 95% confidence interval. Be sure you are clear onthe meaning of our 95 percent confidence intervals. Ifyou use these methods in your day-to-day work, you will be corect 95 percent ofthe time in predicting where the “tue value” lies. That is, if you use the method for 100 ilferent projects, you can expect to be wong bout five mess the trv value will nor ie within the computed uncertainty band In actual engineering practice, the sccurncy ‘ofan instrument sometimes is given by a single swinerieal value: very ofien itis wot made clear just what the precise meaning of this num is meant to be. Even though a calibration, as Fig. 3. hasbeen carious, isnot calculated Instead. the eros tken asthe largest horizontal deviation of any data point from the ited ine. In Fig. 3.11 th and amounts 100.25, Po. The inaccuracy inthis case thus might be quoted as #2.5 percent of fll sale. Tis practice is no doubt duc to the practical viewpoint that Fig.3:43') couation curve. when a measurement i taken all we really want TU, Mandel, "Evaluation and Control of Measurement farce! Dekker, New Yotk, 1991, sect $4Generalized Performance Characteristics of Instruments EES is to say that it cannot be incorrect by more than some specific value; thus the “easy way out” is imply to givea single number. (FRESHA) check jor s2e of variapitty over calibration range. Irrespective of the precise meaning to be attached o accuracy figures provided, say, by instrument manuficturer, the form of such specifications i fairly uniform. More often than not, accuracy i quoted 4s percentage figure based on the full-scale reading of the instrument. Thus if'a pressure gage has range from 0 to 10 kPa and a quoted inaccuracy of +1.0 percent offal scale, this isto be interpreted as meaning that no error greater than #0.1 KPa can be expected for any reading that might be taken on this ‘ge, provided itis “properly” used. The manufacturer may or may not be explicit about the conditions required for "proper use.” Note that for an actual reading of | KPa, a0.1-kPa error is 10 percent ofthe reading. ‘Another method sometimes utilized gives the eror as a percentage of the particular reading with a qualifying statement to apply tothe low endof the scale, For example, a pring scale might be described as having an inaccuracy of #0.5 percent of reading or 20.1 N, whichever is greater. Thus for readings Jess than 20 N, the e1ror is constant at 40.1 N, while for larger readings the error is proportional tothe reading. 3.2.5. Chi-square Test CChissquare testis 2 quantitative method to determine whether data are close to Gaussian. ‘The method involves first arranging a given data in the increasing order and then dividing them into groups; itis not necessary to have the same number of elements in these groups and there is no specific guideline how these groups will have to be done. The general guideline, however, is that there should be at least 20 samples. Figure 3.15 shows a guideline for choosing the number of groups for various sample sizes. For sample sizes between 20 and 40, there should be at least five data points in each group. For sample sizes greater than 40, equal size groups as given in the figure should be used. For example, fora sample size (of 80, en groups of equal size (8 elements) should be used. Once the number of groups and elements per soup is decided, the following equation can be used for obtaining the chi-square valuc.HEEB Doebelin's Meaurement Systems KenauSuae guneg ge = tar ntie™ “smancer erie ae aise Saogeste GPRRISABY) Co Grupin suid for ct suare test * T | Si Lp 3 Probably of goting a x? ‘See ee oa Degree endo FIIBAS)) 1b) Probability of closeness of data with Gaussion,Generalized Performance Characteristics of Instruments ER G8) ‘xhere n, isthe numberof elements in the group, 1, is the number of readings that would be observed in the sime range ifthe distribution were Gaussian based on the mean and standard deviation ofthe ‘ntre sample data and m is the number of groups to which the sample date was divided. Once the Chi- square values obiined, the probability ofthe given sample das being closer toa Gaussian distribution can be obtained from Fig. 3.15(b). An additional information required to obtain this information isthe numberof degrees of freedom, which is equal tothe total numberof groups minus 3. ‘Consider the Table 3.1 (pressure gage caltnration data) The average value p= 10.11 KPa and the stondard deviation @= 0.14 KPa. Apply tho Ca-equare fast and dotor- ‘mine the probebilty ofthis date being closer tothe Gaussian dstibution Giovnses 7 2 z Mis [1008 21s 008 i022 10s 1025 10.09 i036 101 i042 101d | [1021 Max T tons | 10215 |_20 Number of element 3 [6 4 = Mince) 20 sa _| 4a | sis | ass Solution The data of Table 3.1 is first arranged in the ascending order and then divided into four groups of 5 5, 6 and 4 respectively. The third group has one more element than the fourth group since there is 4 repetitive value of 10,2 atthe end, Therefore, one more value is added to the third group. The maxi- ‘mum and minimum values are determined fore each group. The lowest valve of the dats ofthe entire set is taken os O and the highest valuc is taken as 20 KPa, although itis very unlikely any data will ever reach such extreme values, The intermediate value dividing the group is taken as the average value at the margins, that isthe average of the maximum value of one group and the minimum value of the next ‘group, Using the error function and the given values of mean and standard deviation, the probability of the number of elements of each group within the maximum and minimum ranges is determined this is the value, foreach group. The chi-square value from Eq. 3.18 can be obtained as ee (S-5.68)' , (S—4.61)' , (6-518) | 4-453) 3.68 461 S18 433 =03 Since there are four groups, the numberof degrees of freedom is 4~3 = 1. Figure 3.15(b) gives only the range of probabilitis for I degree of freedom. For the above Chi-square value of 03. the probability of the pressure data being Gaussian is between SO 10 75%BEIM oebetin’s Measurement Systems 3.2.6 Calibration Accuracy versus Installed Accuracy Because calibration is a vital function in every measurement application, we have devoted considerable effort to explaining the procedure, We must, however, remember tha it is not an end in itself, our real purpose is to measure something. When we earlcr stated that calibration removes the bias portion ofthe error, this is trac only forthe conditions under which the calibration was performed. When we use the instrument, i is rarely possible to maintain that carefully controlled environment in our experimental apparatus. This means thatthe measurement error (bias and imprecision) must be re-evaluated, taking into account, as best possible, the deviation of the measurement conditions fron the calibration conditions. ‘This re-evaluation is usually not as straightforward asthe ealibration was, beeause the measurement environment is rarely as controlled as a standards laboratory calibration. In fact, judgment and past experience (rather than “scientific” calcul often necessary." One aspect of fon is that the bias portion of the error is now not ze1o. (Revall that we eatlier said that calibration removes the bias.) Abernethy" classifies biases into five different types: 1. Large known biases. 2. Small known biases. 3. Large unknown biases. 4. Small unknown biases with unknown algebraic sign, ‘5. Small unknown biases with known algebra sign, Large known biases are eliminated by calibration. Small known biases may or may not be corrected, depending on the difficulty of correction and the magnitude of the bias. Large unknown biases are not ‘correctable; they may exist, but the magnitude and perhaps even the sign is not known. They usually come from human errors in data processing, incorrect installation and/or handling of instrumentation, and unexpected environmental disturbances. In a well-controlled measurement process, the assumption is that there are no large unknown biases. The reference gives extensive detail on methods for ensuring this, Small, unknown biases then remain as a contribution to the measurement error. Quoting from Abernethy: “In is both difficult and frustrating to estimate the limit of an unknown bias. To determine the exact Dias in a measurement, it would be necessary to compare the true value and the measurements. This ts almost always impossible... If there is no source of data for he bias, the Judgment ofthe most knowledgeable instrumentation expert on the measurement must be used.” The reference does give some hints on how to deal with this situation, but we need to admit that past experience and judgment are generally necessary, ‘A result of ths is that the bas in the measurement situation (as contrasted with calibration) is treated asa random effect rather than as systematic. The number that we estimate for this bias is called the Dias limit. Its defined as the range of values within which we feel that the actual bias will be found 95 percent of the time. The 95 percent value was chosen to make it consistent with our 95 percent confidence limit forthe imprecision, Using this scheme, the “total error” in the measurement is defined as the sum ofthe bias limit and the imprecision, and is given the name wacertainy Uncertainty AUAHB+ I, « 5) 19) "RB. Abernethy, “Measurement Uncertainty Handbook,” Instrument Society of Ameria, Research Triangle Park, NC. 1980,Generalized Performance Characteristics of nctremente. OEY where B A bias limit Jy. & Value from 1 table (Fig. 3.11) and Vis the number of readings used to compute s «5 sample standard deviation, computed from the usual formula “The meaning attached (othe uncertiniy Vi that we are 95 percent sre tha the “inc” value of the measured quantity les within +U ofthe measured value, While the term “ive value sin common we, itis prefered to cll this the reference valne since, a8 we argued earier, the tue value of a measured quantity can never be found. Rather, we can only talk about the “best available” measurement or calibration technique forthe quanti. The value given by ths method is elled a reference valve. ‘The basie measure of imprecision isthe standard deviations, and i's calle the precision index. To attach a certain confidence level to uncertainties, s is multiplied by some constant. called the coverage {factor tis most common, and our suggestion, thatthe coverage factor bef, giving 95 percent ‘confidence. I we could enforce at measurement, the same conditions 35 3telibation, then we could ‘use our calibration uncertainty number as the measurement uncertainty, Otherwise, we must add 1 the calibration uncertainty some adjustment to account for, a5 best we can, the increased uncertainty ofthe measurement conditions. Let us briefly corsier same examples to elf this concept. In Fig. 3.16) we showa simple spring-ype Force measuring scale, We could easily calibrate his with standard Iasies and finda best-fit ine and ncertenty, and remove ery scale bias present. If we then move this scale to our experimental appara to meeste an unknown fore, the unceiinty will increase for number of possible reasons. For example, if the temperatures at calibration and at measurement arc rot the same, the scale will exhibit an uncorrected bias. This bas as wo sources; thermal expansion (hich shifts the zero point) and temperature sensitivity ofthe spring's elastic modulus (which changes the spring stiffness). Unless we measure temperature and correct for these effects, a bias of unknown sign and magnitude will be precent and increase the measurement uncertainty in an unknown (and thus “random” way. Other posible effects include angular misalignment of the unknown force with the scale’s sensitive axis, giving a further increase in uncertainty. To compute how much we need 10 inreate the uncertainty, we nee to estimate there: temperature, misaligment, and ary other effects feito be significant. Note that we do not measure the temperature and misalignment and then correct for these effects, rather we estimate some limits om how lage we think these effets might be and then add this to the unceriiny Figure 3.16) shows »dispicerentsmeasuring dil indicator, fr both calibrtionand measurement situations Teinstruments ist calibrated fgeinst a more aceurate micrometer and then used to measure deflection ‘ofa beam, a part ofan experiment find the beans spring constant FI Such dial indicators have an intra spring to maintain contact between the indicator stylus and the surface being diplaced. When we measure the beam displacement, bias err will be introduced because the indicator spring force acs against the epplied fore F, causing the measured deflection tobe less than it should be. I the foe Fis always downward, this bies error would be tated es having an unknown tagritude, bata known sign, That the deflection is alvays measired v too low. If we esimate an upper imi fois magnitude, this bias would give an unsymmetrical uncertainty; for example, 75 jum to + 25 yum, Fig.'3.36:} (a) Differences ‘final example, Fig, 5.16(c), shows a thermocouple (electrical ~ between caltrtion and tempzrature sensor for both calibration and measurement conitcns. measurement stuetions.KEIR oebetin's Measurement Systems o K © SAE) oor ey Such devices are calibrated in an accurately controlled and measured temperature environment. Specifically, the wires arc immersed in a liquid-flled well whose temperature is uniform at 7,,, over a Jong distance. This is done to prevent conduction heat transfer along the wires, which would cause the sensing tip to read too low. When the calibrated thermocouple is used to measure the temperature of aGeneralized Performance Characteristics f Instnements hot gas, the wires ar in contact with a coo! duct ll; conduction is now not negligible, and the sensing tip will read low. We ean again estimate an upper limit fr this bias error and ada it to the calibration uncertainty. If such an error causes unacceptable uncertainty, we may measure the wall temperature, estimate the needed heat transfer parameter, and compute a correction. This correction will improve the uncertainty, but not elimina i, since the comecticn will iselFbe uncertain, which uncertainty we ‘ill have to estimate and include, For example. if our reading is 357°C and the eorection is “8°C, the nominal value is 365°C. Ifthe uncertainty in the correction is 42°C, and the uncertainty due to other sources was * 5°C, the temperature would be quoted as 365 +7°C ‘While the calibration and measurement environments are by choice often different, we should not overlook the possibility of performing the calibration with the instrunient already installed in our experimental apparatus, the so-called in situ calibration, This, if possible, would in many cases be prefered since, then the calibration numerical results would include all the effects contributing to uncertainty, and thus not require separate judgments and estimates based on experince rather than actual measured data. In a similar spirit, we should also consider the end-so-nd calibration. Here, rater than calibrating separately each link (sensor. ampli, filter. recorder, ec.) in our measurement chain and then combining the individual uncertainties mathematically, wee apply & standard to only the sensor input and record only the final ouput. The advantage here is that all interactions among the Finks are automaticaly taken into account and the procedure may be considerably quicker. A disadvantage is that \We do not see which components are contributing the most tothe total uncertainty. Even when we do perform the individual calibrations, a final end-to-end study may be desirable. ‘One way of gathering data on uncertainty is to repeat calibrations and measurements, though this may often not be technically or economically feasible. One situation where itis regularly use is inthe ‘development and use of industry standard tei, such asthe many documented in the ASTM standards volumes. These standard test wil be used over and over in many laboratories throughout the particular industry, s0 itis possible to devote considerable effort to their development. One type of uncertainty study leads to information on the repearbity of the test. Hee the testis repeated several or many times, using the same laboratory, apparatus, and operator. On the other hand, a reproccibility study repeats the tet using diferent laboratories, apparatus, and operators. (For example, astel rod may be cut into 100 “identical” test specimens, randomly selected into groups of 10, and one group sent to each of 10 laboratories for tensile strength testing.) As you might guess, the reproducibility studies lead to larger uncertainties, but they usually give more realistic results. When these results are carefully analyzed, they may give a keen insight ito the sources of uncertainty. ‘final bit of nomenclature concems the terms “type A uncertcinty” and “type B uncertainty.” which you may encounter. If you actually perform experimentation, such as we have described above, to set numerical values for uncertainties, these ae called “type A": all other sources are called “type 8." Examples of type B include expert opinion, past experience, and specification shoots provided by ‘equipment suppliers. Such sources of uncertainty numbers may have to be used, but we always want (0 ‘make clear that we did not actualy run a calibration experiment. When cornbining uncertainties to get the overall value, the calculation methods ae the same whether they are type Aor type B, however itis important to clearly state which ae type A and which are type B, since the use of type B values means that you are assuming these values ae correct, rather than verifying them by your own experimentation ‘Some additional references on the general subject of uncertainties may be belpful S.1. Kline and FA. MeClintock, “Describing Uncertainties in Single Sample Experiments,” Mech. Eng. vol 75, January 1953, p. 3: L. W. Thrasher and R. C. Binder, "A Practical Application of Uncertainty Calculations to Measured Data” Trans, ASME, February 1957, . 373,HEI oebetin's Meaurement Systems 3.2.7 Combination of Component Errors in Overall System-Accuracy Calculations ‘A measurement system is often made up ofa chain of components, each of which i subject to individual inaccuracy. Ifthe individual inaccuracies are known, how is the overall inaccuracy computed? A similar problem occurs in experiments that use the results (measurements) from several different instruments to compute some quantity. Ifthe inaccuracy of each instrument is known, how is the inaccuracy of the computed result estimated? Or, inversely if there must be a certain accuracy in a computed result, what errors are allowable in the individual instruments? ‘To answer the above questions, consider the problem of computing a quantity y; which is a known. function of n independent variables, x,y +X, that is, YER Shy eK) 20) For small changes in the independent variables from given “operating points.” a Taylor series expansion gives a good approximation forthe corresponding change iny Lge Pans Lids penr Lh tan Aart So teh sn tot @2n Think of the partial derivative as the sensitivity of y to changes in the particular x. When a partial derivative has a large numerical value, y is very sensitive to that particular x. Since the partial derivatives are numerically evaluated at the operating point, they are constants (not functions) in Eq, (3.21), so this equation defines y as a linear function of the 2's, even though the original function (/) may be nonlinear. If the Aes ate now considered ta be the uncertainties u,, in each measured value x, then the corresponding uncertainty U, in yis given by “This relation is called the rot. sum-square (rss) formula, We have not here proven is validity but this rests on the fact thatthe standard deviation of any linear function of Gaussian independent variables is sivenby the square oot ofthe sum of squares ofthe individual standard deviations. Its an approximate result because y is not really a linear function of the xs; itis close to linear only for small changes. Since U, is computed from individual uncertainties that are 9 percent confidence intervals, it has the same meaning, thats, it isa 95 percent confidence interval forthe dependent variable y. Equation (3.22) is the basis for computing the error of complete measurement systems from the errors in individual components ‘When we originally plan an experiment, we must decide how accurate the final results must be ‘o meet the goals of the study. Once this decision has been made, then we can explore the accuracy needed in each individual instrument. It should be clear that this kind of problem does not have a unique solution; there will be many combinations of individual errors that give the same overall error. To get started on such problems, we can use the method of equal effects. If we have nothing else to g0 on, it seems reasonable to, atleast initially, force all the instruments to contribute equally to the overall error. Using Eq. @.22), this leads 10Generalized Performance Characteristic of Instruments SER 623) ‘This equation defines initial values for the allowable error in cach measurement. We must then compare these requirements with the capabilites of the actual instrumentsavailable to us. Ifwecan find instruments ‘which meet all these needs, we have atleast one solution to our problem. Ifone or more requirements ‘carmot be met, it does not mean we are defeated. We then need to sec if some of our instruments are better than Eq, (3.23) requires. If 0, we can use them to relax the requirements that we were not able to ‘meet. A careful evaluation of all such possibilities may lead 1o a combination of instruments that meets the overall specification. It of course can also happen that, if the overall accuracy is set too high, we can find no set of real instruments that meets our needs. If so, the systematic approach used here allows a convincing argument when we need to discuss the dilemma with superiors or colleagues. [Anan axa ft abo proce conte an epee messy ineans of ynemontar be ore ower Casritod ya ohn sha The ‘Geni br povorcm be wrton ss aaa. ! a ster R rotten ofsbn cir ra Ftc a onto em ® Lg Lengen of torque arm, (m) {gt oon Solution A sketch of the experimental setup is shown in Fig. 3.17(a). The revolution counter is of the type shown in Fig. 2.4 and can be tumed ‘on’ and “off” with an electric switch. The instants of tuming “on’ and ‘off’ are recorded by a stopwaich. If is assumed the counter does not miss any counts, the ‘maximum error in R is 1, becsuse ofthe digital nature of the deviee (see Fig. 3.17(b). ‘There is a related error, however, in determining the time ¢ since perfect synchronization of the starting and stopping of watch and counter is not possible. The stopwatch might be known to bea quite accurate time-measuring instrument, but this does not guarantee that it will always measure the time interval intended. In assigning an error to 1, then, we are not helped much by the watch manufacturer's uarantee of 0.10 percent inaccuracy if our synchronization error is much larger than this. This synchronization error is certainly not precisely Known since it involves human factors. An experiment to determine its statistical characteristics would be a more expensive and involved undertaking than the power measurement of which it is a part, So we are in the rather common position of having to rely ‘on experience and judgment in arriving at an estimate of the proper numerical value. and we begin to appreciate that some of the statistical niceties and fine points of theory considered earlier somewhat academic in such a situation. They are always useful in terms ofthe understanding of basic concepts that they develop: however, they cannot be relied on to give clear-cut answers in situations where the basic data are ill-defined. In the present case, suppose itis decided that a total starting and stopping error is taken 28 +0.50.s whether this is tobe considered as an absolute limit oF as, say, a 9S percent uncertainty, becomes a matter of judgment and preference. Our preference, as stated catlict, isto teat this number {as an uncertainty atthe 95 percent level of confidence, The measurement ofthe torque arm length L isHERA oebetin's Measurement ystems also subject to similar vagaries, depending on the care taken in its measurement. Suppose we use a fairly rough procedure and decide on a 95 percent uncertainty of 0.00127 m. [He'SHP!) (0) dynamometer test setup. (0) Revolution-counting error. ‘The scales used to measure the force F can be calibrated with deadweight, yielding a set of data ‘analogous to that of Fig. 3.11(a), which can be processed to caleulate a 95 percent uncertainty as done there. Suppose this tums out to be #0.18 N. This calibravion uncertainty must however be translated into a corresponding measurement uncertainty, as we discussed above. When actually used, the scales ‘will be subject to vibration (not present at calibration), which may reduce frictional effects and decrease the error (uncertainty). At the same time, the pointer on the scale will not stand perfecily still when the dynamometer is running; thus, in reading the scale, we must perform a mental averaging process. which ‘may introduce a new error not present at calibration, Such effects are clearly difficult o quantify, and again we must make a decision based partly on experience and judgment. Suppose we assume the two ‘mentioned effects cancel and thus take the force measurement uncertainty as 0.18 N. Fora specific run, if the data, with 95 percent uncertainties fr eac R=12024 Irev F=45=0.18N L = 0397 £0.00127 m ® 1 602055Generalized Performance Charectrstics of Instruments C3 then the calculation proceeds a follows: In terms of inch units, we have w Bet o ‘Then, computing the various partial derivative to three significant figures gives OW _2nRL _ 2x x1,202x0397 = 50 Wan © OW, _ 2RFL _ 2 ASX0397 1 59 Watvrev o OY THRE ERNIE X AS © 5667 Wat a ae 60 BW _2RRFL _28X1,202%48%0.397 376 Woruy © ae oF While we have chosen to use the statistical uncertainty concept as our preferred measure of error, other viewpoints are possible. One such treats the individual errors as absolute limits; the measured quantity is assumed to never fall outside these bounds. When we combine such errors, we want the combined error to have this same meaning. Since the individual errors are as likely to be negative as positive (we usually have no information onthe sign, to get the worst case we would use Eq. (3.21) but ‘with absolute values for each term, so that they add up inthe worst possible way. If we call this absolute ‘error, our example calculation goes 2s follows: aw + a ret a o = $0 % 0.18 + 0.187 x 1 + $667 x 0.00127 + 37.5 x0.5 = 36.8 Watt ag ‘We now compute horsepower as = ERIN AS20371 «aso wa ay 2,250 4 36.8 Watt ‘We prefer to treat the individual errors as 95 percent uncertainties, giving the total uncertainty U as iF) -Sa) Sa) Fa) = (60x 0.18) +1.87" + (5667 x0.00127) + G75 x03) = 422.08 Watt aay The uncertainty J will of course always be less than the corresponding absolute error E,, a8 the two formulas show. We choose to use uncertainties because, due to random nature of the error, it is not likely that the worst possible combination will actually occur, making the absolute error method 100 conservative. As engineers, we do need to be appropriately conservative, but not when this leads to unrealistic and uneconomic decisions. Finally, we must again admit tha, since the individual errors are often estimates rather than “scientific” calculations with hard data the dstinetion between absolute crrors and uncertainties may be subject to some challenge from “practical” critics. Nevertheless, the uncertainty viewpoint is a reasonable one, is widely accepted, and we continue to recommend itEER Deebetn's Measurement Systems ‘When computing values forthe partial derivatives, using Eqs. (5) to (8), an approximate numerical ‘method (ther than the exact analytical oncused there) may be useful when programming dataacquisiton software, We simply compute the change in power AW caused by a smal (typically 1 percent) change in each ofthe independent variables, ay AF. The ratio AW/AF, for example, will be very close to the analytically computed value 4W/0F. Finally, suppose we wish to measure power to 0.5 percent accuracy in the previous example. What accuracies are needed in the individual @W/AF mesurements? Using Eq (3.23), we get ap aW Xaceuraey _ 2250 0.005 V4 x50 1125N W x0.005 _ 2250 0.005 2x87 =3rv 2250 0.005 5 arse 2 9925% 10% m 3) 2250 x 0.005 Apa X0.005 ow 2x37 wy OSs If itis found that the best instrument and technique available for measuring, say, F are good to only 0.18 N rather than the 0.1125 N called for by Eq. (13), this does not necessarily mean that power cannot be measured to 0.5 percent. However, it does mean that one or more of the other quantities, R, and 4 must be measured more accurately than required by Eq, (13). Making one or more of these ‘measurements more accurately may offket the excessive error in the F measurement. The given formu lac allow calculation of whether this wll be tu. xa EAB) 10 &c. (2.6), solve forthe strain « in terns ofthe other paremetors; « = fIGF, Ry Ey RR, 0). Then, take the natural log ofboth sts: In ¢ = In f. Now taka the differs! ofboth sides s0 that rms such 28 dc, de,/¢, ORY/R,, etc, are formed. This wil give the percontage error e/e ine 80 function of tho percentage omorsin tho other parameters. IGF. R., ER, and 0, are all mea sured to #1 percent error, what isthe possible errr inthe computed value of ¢ 7° Solution GARE, R, TcaF PND raboFs. 2. Strain, ¢, output voltage, GF: gage factor Gage resistance, E,: Bridge voltage 2B: Resistance of the adjacent armGeneralized Performance Charecterites of Instruments EI Equation (2.1) can be written as: ey +R)? GFRE,R o Taking natural log on both sides Ine= ine, +2 In(R, +R) InGF)~InR,~ In E,~In R, ® Differentiating Eq. (2) de dey , 248, +R) GF _ ally a, a, € & R+R, GFR, BR, ® using Eq. (3.20) and leing R= R, +R, ee = ee oS © lawn 16 bay | Rdeas | OF bn | Re Ione | las” Re lee ° Ifeach term of Eq. (4) contributed the same error of 1% ‘Applying the rot mean square formula of Eq. 3.21) to Eq. (3) A PTF on 3.2.8 Theory Validation by Experimental Testing (One of the most common and important application areas for experimental work is the testing of new theoretical relations. When a new theory is developed, we always question its validity until we get ‘experimental confirmation of its predictions. Such confirmation (or refutation) is usually ofa statistical nature since both the theoretical prediction and the experimental measurement have some uncertainty attached to them. A useful way of visualizing this situation is an “overlap graph.” Here we plot the ‘theoretical result and the measured result on the same line, each with its own uncertainty band. Ifthe ‘wo uncertainty bands do not overlap, then it is unlikely that the theory is correct. If there is some overlap, then the theory gets more and more likely to be correct as the overlap region increases. A simple example will ilastrate this concept. ‘The cantilever beam of Fig. 3.18 isa common structural element that has been theoretically analyzed in several ways. Ifthe beam is sufficiently “long and thin” and if several other assumptions arc also reasonable, then the end deflection & due tothe force F is given by sev Ene To check this theory. we might build a beam with certain dimensions (L, 6, ), from a material with a certain modulus of elasticity £, We then apply a foree F and measure the resulting end deflection 5. Suppose our interest isin the “spring constant” F/6. We have a measurement of the applied foree F and 324)HER Docbetin’s Measurement Sates an estimate ofits 95 percent confidencs interval ad similar data forthe deflection. The uncertainty in ‘the ato F/5.can be computed inthe usual way. We now have number forthe best estimate ofthe spring constant and_95 percent confidence interval fortis value ‘Turing now to theory, note that every “theoretical” result relies on one or more parameter values that can only be found by “experiment.” That is, to compute the theoretical value ofthe spring constant fora specific bear, we need numbers forthe dimensions and the material property E. These can only be found by experiment! Thus we must measure 6, f, and L and “look up’ value for Ein some reference, ‘When we measure the dimensions, we ean as usual attach an uncertainty to each. Ifyou look up an E value for, say, some aluminum alloy, you often do not find any data on its uncertainty, though the quoted value surely has uncertainty. Let us assume that ‘a phone call to the material supplier gets us at least an estimate of E°s uncertainty. We now have ‘cnough data to calculate, in the usual way, the uncertainty in the theoretical value of £18. ‘We now can plot the “overlap graph” in Fig. 3.19 which showsa situation where the theory is not validated. Thetis, at the 95 percent level of confidence, there is no overlap between the ‘two confidence bands, making it highly unlikely that the theory is accurate for the particular beam tested. (This might be the case if the beam is ‘prediction with experimental measurement.Generalized Performance Characteristics of Inctraments WEB ‘ery short, For short beams, the theory behind Eq. (3.24), which is based on pure bending, would be inaccurate since now shearing effects become important) Note that inthis example, asin the general ‘case, the question “Is the theory correct?” cannot be answered by a simple yes or no, As the confidence bands exhibit more and more overap, the theory becomes more and more accepble. Once more wwe see that stisical viewpoints give us a more realise picture, than would the more ‘comparison of only the “base valucs” of measured and theoretical spring constants, without considering, ‘the uncertainties involved in both, While the overlap graphs are often sufficient, a more numerical: comparison is available ™ 3.2.9 Static Sensitivity ‘When an input-output calibration such as that of Fig. 3.11 has been performed, the static sensitivity of the instrument can be defined as the slope of the calibration curve. Ifthe curve is not nominally astraight line, the sensitivity will vary with the input value, as shown in Fig. 3.20(b). To get a better definition of sensitivity, the output quantity must be taken as the actual physical output, not the meaning attached to the scale numbers. That is in Fig. 3.11 the output quantity is plotted as kilopascals; however, the actual physical output is an angular rotation of the pointer. Thus to define sensitivity properly, we must know the angular spacing of the kilopascal marks on the scale of the pressure gage. Suppose this is 5 angular R,,. Thus our earlier statement about the desirability ‘of high input impedance ean now be made more specific. The input impedance must be high relative to the output impedance of the sysiem to which the load is connected. Assuming that itis possible to define ‘generalized input and output impedances Z,, and Z,, in nonelectrie as well as electric systems, we may generalize Eq. (3.28) to Z, 742, 29) Gan Ge 29») where g,, Amessured value of effort variable qq. A undisturbed value of effort variable OF course, if we knew both Z, and Z,., we could correct ¢,,, by means of Eq. (3.29). However, this would be inconvenient; also Z,, is not always known, especially in ronelectric systems, in which Acfinition of both Z,, and Z,, is not always straightforward. Thus a high value of Z,, is desirable since then corrections are unnecessary and the actual values of Z, or Z_ need not be known. PEERED | Repent example 3.7, except now the vottage across R, isto be measured. Solution Objective: Error due to loading resistance Rif voltage across R, has to be measured. ‘Short circuit all the power sources and determine the equivalent resistance across R, at AN | 1X a 4 meme t daem L.. {| 7 Re . Y » “ 1 l Ray +R, Ry RFR, RyCR+ RoR + Re) (R + R)Ry + RCRy + Re) + (Ry + RAR, + Re) RR, =R,=R=R,= 1002 a Ra @ R,=502 (from Eg. (2))ZB ocbetin’s Measurement Systems Error due to loading resistance R, ® for for Toachieves high value ofinput impedance for any instrument, not just voltmeters a numberof paths are open to the designer. Now Wwe describe three, using the voltmeter as a specific example, The most ‘obvious approach is o leave the configuration ofthe instrument unchanged, but o alter the numerical values of physical parameters so thatthe input impedance is inereased, In the voltmeter of Fig. 3.23, this is accomplished simply by winding the colin such a way (higher resistance material and/or more turns) that is increased, While this accomplishes the desired result, cetain undesirable effees also appear. s type of voltmeter is basically a curren-sensitve ether than a volage-sensitve device, an increase in R, will reduce the magnetic torque available from a given impressed voltage. Thus if the spring constat ofthe restraining springs isnot ehanged, the angular deflection fora given voltage (the sensitivity) is reduced, To bring the sensitivity back to its former value, we must reduce the spring consant. Also, because of loner torque levels, pivot bearings with less friction must be employed. ‘These design changes generally result na less rugged and less reliable instrument, so that this method of increasing input impedance is limite in the degree of improvement possible, before other performance features are compromised, This situation occur in most instruments, not just in this specific example. inputimpedance isto be increased without compromising other characteristics, different approsches are needed. One of general uility employs a change in configuration of the instrument so as to inelude an auxiliary power source. The concept is that @ rugged instrument requires a fair amount of power to actuate its output clement, but that this power need or necessarily be taken from the measured medium. Rather, the low power signal from the primary sensing element may control the output ofthe uiliary power Source so 2s to realize a power-amplifying effect. “To continue our voltmeter example, this approach is exemplified by a transistor version ofthe classic vacuum-tube volmeter (VTVM) shown in rudimentary form in Fig. 327(a). When the input voltage is 2zo (shor circuit across the E, terminals), adjustment of R allows sein ofthe meter volige E0220, fan input E, is applied, the gate bias is no longer the sane, the curentsj, andi, are no longer equal, anda meter voliage exits, While the meter current Z, may tl be as large asin conventional meter, the curent that determines the inpat impedance isi, Which will be very small. Thus a very high input impedance canbe ealized whike a rugged meter clement is used. A block diagram forthe FET voltmeter is seen in Fig. 3.27(t), which may be compared with tht for an ordinary meter in Fig. 327(c). ‘tll another approach to the problem of increasing input impedance uses the principle of feedback cr null balance. For the specific area of voltage measurements tis technique is exemplified by the potentiometet. The simplest form of this instrument is shown in Fig. 3.28, Clearly in Fig. 3.28(a) cach position ofthe sliding contact corresponds to adefinite voltage between terminals ¢ and 4 Thus the scaleGeneralized Perjermance Characteristics of struments EM ‘ean be calibrated, and any voltage between zero and the battery voltage can be obtained by properly positioning the slider. If we now connect an urknown voltage and a gelvanometer(curent detector), as in Fig. 328(b), then if E, i ess than the battery voltage, thefe will be some point on the slider seale at which the voltage picked off the side wire just equals the unknown Ths point of null balance can bbe detected bya zero deflection of the galvanometer, since the net loop Yotage a,c, db, will be 2210. ‘Then, the unknown voltage can be read from the calibrated scale. ‘We should note that under conditions of perfect balance the curent drawn from the unknown voltage source is exactly zero, which yields an infinite input impedance. In actual practice, there must always remain some unknown unbalance current, since a galvanometer always has a threshold below which currents cannot be detected. The interpretation of poteniometic voltage measurement as a feedback scheme is shown in Fig. 3.28(c). While the manual balancing described above is adequate when the unknown voltage is relatively constant, the procedure may be made automatic by use ofthe instrument servomechanism (in this case called a selfbalancing potentiometer), as shown in Fig. 27. Thea, by provding a pen and recording chart, varying voltages may be accurately measured. hh pene weLoce ts | Reg Etat women Te ea » ° ERB) reta-espece transistor (FET) voltmeter. In generalizing the concepts of input impedance, a reasonable santing point might be a listing of 9, and corresponding q, variables for some common measurement situations In general, the quantity q, \Whichis of primary concer, may be either alow variable oran efor vartabe> Briefly, energy transfer across the boundaries ofa system may be defined in terms of two variables the product of which gives EO: Dosbelin, “System Modeling and Response,” Chap. 7, Wiley, New York, 1980.MEER Doebett's Measurement Sterns the instantaneous power. One of these variables, the flow variable, is an extensive variable, in the sense that its magnitude depends on the extent ofthe system taking part in the energy exchange. The other variable, the effort variable, is an intensive variable, whose magnitude is independent of the amount ‘of material being considered, In the literature, flow variables are also called “through” variables, and effort variables arc called “across” variables, When, isan effort variable, Bq, (3.28) and subsequent developments apply. (Arter Oo » ‘Votagecouce (FSB) roteniometer tage mesuremen. However, ifg, isa flow variable, the situation is somewhat different. Then, iis appropriate to define a generalized input admitiance ¥,, 38 flow variable _ 4 Ys 9 stor variable ction variable fa 630) rather than a generalized input impedance Z,- effort variable 63n flow variable [Now we can write the power drain ofthe instrument from the measured medium, in terms of the measured variable g, as Peaga= 632) Ye and we note that now a large value of input admittance ¥, is required to minimize the power drain. ‘A familiar electical example of this situation is the ammiete. In Fig. 329{a), we are interested in ‘measuring the current by means of an ammeter inserted into the circuit. Applying Thévenin’s theorem, ‘we can reduce Fig. 3.29(a) to Fig, 3.29(b). We see thatthe measured value of the current i Ey Ey BER, WigGeneralized Performance Charectrktcs of Instruments WIA Bm i 7 Fe] by Lee Btn & ee ° a Anmeter lao eet In Fig. 9.20, what porcentage erormay be expected in measuring the current through RR, Ry" R,=R,=R,=100end R, = 1007 1R,=1 27 Solution As a first step, short circuit all the power sources of Fig. 3.29(a). x fa ana tan ns . Re a fe ‘ (ee, Te @ ° a AOE ay Bo] % Eu FE G34) Gvithout mete: o Gvithout meter) o ft with 2) Rel (Wvithmers @ Error due to meter loading @ o 157. (from Fig, 3.300)es 10_ 6% 267+10 1 = 037% ‘ 267 +1 ‘whereas the truc (undisturbed) value of the current would be Ba Ea aay Reis Soiisleartatf, ie oapposeh , went sean anmeter ith Y_> thats, he metereinance ‘must be suiiciently low, just the opposite of that desired in a voltmeter. Ths result can be generalized to.pply to all effort variables (such as voltage) and al flow variables (such as current). Equation (3.29) applies to those cases in which the measured variable q is an effort variable, and Eq, (3.35) gives the corresponding relationship when q, i alow variable: —t_ ia, 35) Gin = where Y,, generalized output admittance of preceding element generalized input admittance of instrument 4, 4 measured value of ow variable dna S undisturbed value of flow variable Repeat example 2.9, except now the vollage across R, is o be measured. (B+ RR + Red RAR +Rak Rae +R o Rise Fe cy @ If R=R,=R,=R,=R,- 1002 = 2000 {from Eq,(1)}Generalized Performance Characteristics of Instraments EI Bu From Bxample3.9 = 9 @ Error due to loading 10 y for R,= 100 ee for R,=10 0.5% For some instruments, in the case of a static input, the power drain from the preceding element is zero inthe steady state, although some total energy is removed in going from one steady state to another. Jn such an instance, the concepts of impedance and admittance are not as directly useful as one would like, and it is appropriate to consider the concepts of static stiffness and static compliance. These make it possible to characterize the energy drain (in the same way that impedance and admittance define ‘the power drain) in those situations in which impedance or admittance becomes infinite and thus not directly relevant. The terms “stiffness” and “compliance” come from the terminology of mechanical systems, which afford some of the best examples of the application of these concepts. However, we ‘generalize their definitions to include all types of physical systems, just as we did with impedance and admittance. ‘Consider the system of Fig. 3.32(a), as an idealized model of some elastic structure under applied load f,,¢ Ths load wil cause forces in the various structural members. Suppose we wish to measure the force th the member represented by the spring k,. A common method of foree measurement employs a calibrated elastic link whose deflection is proportional to force; thus a deflection measurement allows a force measurement, Such a device, with spring constant 4,, is shown in Fig. 332(b). To measure ‘the force in link k,, we insert the force-measuring device “in Series” withthe link A, as shown in Fig. 3.32(¢). The usual difficulty encountered here is that the insertion of the measuring instrument alters the condition of the measured system and thus change the measured variable from its undisturbed valuc. ‘We wish to assess the nature and amount ofthis error, and we use this example to introduce the concept of static stiffness. ‘The measured variable, force, is an effort variable. Thus if we try to use the impedance concept, ‘we must find an associated flow variable whose product with force will give power. Since mechanical power has dimensions newton-meter per second, we have power ms ‘effort variable N ‘Mechanical impedance is thus given by effort variable __ force Mechanical impedance = eee” velehy 37 If we calculate the static mechanical impedance ofan elastic system by applying a constant force and noting the resulting velocity, is zero we get Flow variable = 2 velocity 639 Sic hrc impedane = «= 039| yi —| “, » a a Anite * TTT o ‘ L AWA [rh lgin ! 1 i aE = | | wit —f asa tw ie { jw Pies So hn am « wen » ee st ts { SPS (ew " ae ren » (CIR) rece gone tong eet ‘This difficulty may be overcome by using energy rather than power in the definition of the variable associated with the measured variable. Ithis is done, a new term forthe ratio ofthe two variables must be introduced, since the use of mechanical impedance as the ratio of force to velocity is well established. We thus define force force Mechanical static stiffness 4 2 ____foree_ sates Gieplacement ~ [velocity since G39) Energy = (force) x (displacement) 6.40) ‘Thus, in general, whenever the measured variable is an effort variable and the static impedance is ‘impedance, we use a generalized static stifiness 5, defined byGeneralized Performance Chracteristies ef Inctrument: EER effort variable (flow variable) dr ean If this is done, it can be shown that the same formulas can be used for calculating the error due to inserting the measuring instrument as were utilized for impedance, except S is employed instead of Z. Thus, Eq. (3.29) becomes 42, where qu, measured value of effort variable Gnu undisturbed value of effort variable S,, 4 generalized static input sifiness of measuring instrument S._ & generalized static output stiffness of measured system Let us apply these general concepts tothe specific ease at hand. The output stiffness ofthe system of Fig. 3.32(a), atthe point of insertion ofthe measuring device, i simply the ratio of force p to deflection yet terminals a and b in Fig. 332(¢), This siffness can be found theoretically by applying «fictitious load p and calculating the resulting.y; or, ifthe structure (ora scale model) has been constructed, we can obtain the stiffness experimentally by applying known loads and measuring the resulting deflections. A theoretical analysis might proceed as below: E forces = 0 park thQ.-y,) <0 3.43) Son 0-9) Ky, = 0 G44) CA AD, Hy =P G45) By, + KRY, = B40) Using determlonets yields +h)+ fa (bs) Pils hd + Say b) aa (byte) ea) ean “The output siffness is now obisne from Eq (3.47) by lening be 2e0: Ss, =2=-—2--____P____ G48) YMA Pe PFE) OF ED GAB 5,-—__1_____ TFRs EDDA] oe”)AER octets Measurement Systems ‘The input stiffhess ofthe measuring instrument is given by _fowe cement 7 50) We may now apply Eq. (3.42) to get Measured value of force _ a os True value of force” 1/{1/R, + (+R) + KE, + KEIR, From Eq. (042), it i apparent that in general, we should like to have S, >> S__ inorder to have the ‘measured value close tothe true vale. In this example, this requirement coresponds to 1 ‘ Ti+ + ky) hk + hk + ky) a ‘Thus the measuring device must have a sufficiently stiff spring. Ini 332 woe pcan ary asc ase bk Wek sigeh= feo Won an = 100 Rvon? f= 1000 en Sotution Objective: Percentage error in measuring force in k, Given kk ky, ~ 100 Nem Aa feed Ne, 10008 a Be FE 1 Try 100 (100+ 100)(100-+ 100) ~ 1007 Sgo =60 Neem 1000 Nicm = k, = 1000 Nem (Input stiffness) al So S,)= 60460 + 1000) = 5.66% 000 Nem 596%Ceneraized Performance Characteristics of Instruments ER ‘We saw earlier that when the measured variable is not an effort variable, admittance (rather than impedance) is a more convenient tool. Again, however, under static conditions it may happen that admittance is infinite; thus a concept analogous to stiffness is needed to facilitate the treatment of such situations, For such cases, the generalized compliance C, is defined by flow variable , few variable ’ 9 Teton variable) dt os ‘Asa mechanical example, suppose we wish to measute the displacement xin Fig. 3.27(a) by means ‘fa dial indicator. Such indicators generally have a spring load (sifiness k,) to ensure positive contact with the body whose motion is being measured. This spring load add a force tothe measured system, thereby causing error in the motion measurement. Iti clear thatthe indicator spring load should be a5 light as possible, but we wish to make more quantitative statements about measurement accuracy. Analysis shows that: Measured value of deflection _ 1 “True valucof deflection ~ hy /{k +k, + Kk +h) +1 and k, must be sufficiently small to get accurete displacement measurement G4) EEEREMD Corie Ee, (3.54) Solution From Example 3.11, the following equation can be written rt thy hae Bk +h) © Mth thy thy) +E +h thy) ‘Equation (1) simplifies as * 2 billy thy he) 4 hylh +45) ® BG +h th +k) hh hy +k) Equation (2) simplifies as ra Bulk + + ha) + hal +) ® Bath +h) +k +h) + Gh) Equation (3) simplifies as 1 FiO +h) “ bt hy) + halk +h) 57EO,Docbelin, “Measurement Systems,” sth ed p85,HEIR ocbetn's Measurement Sisems Equation (4) simplifies as 0 why ky th, © kD kk te (s ° me) Equation (5) ste same Eq, @.54) a pe 4 by state 4 tea] Mi | Is Beton ” Figs 3133) Olsplacement-goge leading effect. FSEEEICEREN | 200! Exampo 3.1, except now th tere hk, 8 be measur. Solution (Note the degree of freedom x tthe junction of end k,) ‘ » Fig. 3.34 ‘The stifiness matrix [X] for the above system is given by thth 0 [x]; why yt hy aly ayGeneralized Performance Cheracteritice of etruents EER It is easier to write down the above stiffness matrix since it follows a certain pattem—the diagonal elements are formed by adding stiffness of all the elements connected to a degree of freedom: off- diagonal clements are formed by subtracting stiffness of clements connecting to the other degrees of ‘freedom. ‘Stiffness matrix is related to applied forces by in) [0 [Kle=/Q)Where r= 425} f=} 0 @ es) htkth hy 0 V[x) [0 “ko hth, Hk, [J tey 0 © 0 ake ketal) Ln (+h + hye, kx, 7 0 “ wee tA By 0 © het Et RD Soe oO The messed force is Le=ke-5) © In order that we do not have to solve explicitly for x,,x,,x, to determine the error due to the measuring system, Eqs. (4), (5), (6) are rearranged so they can be expressed in terms of f,. From Eq. (4) tk, +h), = ly) ® From Eq. (5) KG, 5) “hGy—a“f, ® From Eq, (6) KA) * KE = (10) fen Lt kaa From Eq. (8) +s, f, an which gives aay From Eqs. (9) and (12) 5- |= al a) tn 13)MEER oeteti’s Measurement Systems Which gives nsf 1 } a bth From Eqs. (9) and (14) wfa-2[z as Which gives neath as From Eqs. (10) and (16) forte} on Sa 1 Fam 144s. “ ky hk Equation (18) relates the force measured by the instrument in response to the applied force, which accounts forthe spring stiffness of the instrament. If one were to be interested in direetly obtaining the applied force on the measuring instrument, the calibration constant defined by Eq.(18)has tobeaccounted for. However, the objective of this problem is to determine the percentage error in measurement duc to If k, were to be extremely large, it represents the situation of least error in measurement. Let (undisturbed value of force due to&,) be given by i 7 (19) mia (64,18) for 0 gives Ea (19) From Eqs. (18) and (19) (20) Qty From Eq. (20), for, = 1000 NiemGeneralized Performance Characteristics of Ietruments WR 1 ut BOT 1000 2 Error=1-0.9615=385% (from Eq. 21) From £4. (20), fr A, = 10000 Nin \ oy 100 1 Fae thts ror 1-09960=04% (rm Ha. 21) ‘Therefore, increasing the stiffness of the foree transducer will decrease error, but will have other disadvantages related to sensitivity, resolution, etc. Although we have used some specific electrical and mechanical measurement examples to develop the basic concepts of impedance, admittance, sifiness, and compliance, as they apply to loading effects im measurement systems, these concepts can be easily generalized to other measured variables. This ‘generalization and proofs of formulas, such as Eq. (3.42) are let forthe reference. In concluding this section on loading effects, the following comments are appropriate. In every ‘measurement, the instrument input causes 2 load on the measured-medium output Ifthe instrument system consists of several interconnected stages (the “elements” of Fig, 2.1), there may be significant loading effects Between stages. Tis is often the case when the “elements” are general-purpose devices such as amplifiers and recorders, which are connected in various ways at different times, to create a ‘measurement system suited to a particular problem. In using such a “building block" approach, loading problems must be carefully considered, and methods such as have been outlined above must be used if satisfactory and predictable results are to be obtained. If, however, the elements are merely functional components (permanent!y connected) of a specific instrument, then loading between elements may be a necessary consideration for the instrument designer but not forthe user. The user nced be concemed only ‘with the loading situation at the interface between measured medium and primary sensing element. nf 2.33 nt pmcertag ory ect a marig a cto x ¥ k,=k,= ky =h,= 1 Nom and k,= 0.1 Nom? if, = 0.01 Nem? Solution ¥E 0. Doebelin, “Engineering Systems,” 3rd ed, McGraw-Hil [New York, 1983, pp. 87-97.HED Dosvetin's Meazurernent Syston (Kees hy thy thy thy ky Yfxt]_[ 0 A byt lL SV ‘Where x is measured by the dial gage. from Eq. (2) Oth th thor aks, Which gives =a th th tk)" From Eqs. @)and (6) a (6a thy Ha) A thy he) ‘The undisturbed value of x* can be obiained by putting &, = 0 in Eq. (5) as follows. (PS; Eq. (5) is also part of Prob, 3.17) x - K the that hi th) From Eqs. (5) and (6) Oth tht ith +) FT thy thy thy) + (tke thn) +e bth thy) Emor=t—# Given for A= 0.1 Niem From Eq. (7) ” Le1414(040 xp Ge1+140.) 404140 Error = | - 0.9615 = 3.85% (from Eq. (8) for 4, =001 Nem x L+d4) xy (1+1+1+0.01) + (1 +1+0.00 = 0.9615 =0,9960 i 0.9960 = 0.4% a @ ® O) © © ®ERR Cocbetin's Measurement Sten G7 hag Fe os 44 complementary-function part of solution 4g partcular-integral part of solution ‘The solution ,_,has 1 arbitrary constants; g,,, has none. These nt arbitrary constants may be evaluated ‘numerically by imposing intial conditions oh Ea, (3.57) The solution gis obtained by caloulating them roots ofthe algebraic characteristic equation arta, D+ +aD+a,=0 58) where the operator D is treated as if it wore an algobraic unknown. Once these roots sy.» 4, have been found, the complementary.function solution is immediately writen by following the rules sated below: 1. Real roots, wnrepeated. For each real unrepeated roots one term of the solution is written as Got, where Cis an arbitrary constant. Thus, for example roots -,7,+3.2, and 0 give solution Geir Ge#4C, 2. Real roots repeated. For each roots which appears times, the solution is written as (C, + Ct + Cf +--+C, ¢P Yet, Thus, if there are roots -1, —1 + 2, +2, +2, 0, 0, the solution is written 38 (G+ Ce (G+ Cr+ GPE C+ Cy. 3. Complex roots unrepeated.. Acomplex root has the general form a +b. It can be shown that if the a's of Eq. (3.64) are themselves real numbers (which they generally will be, since they are physical quantities such as mass, spring rte, ec), then when any complex roots cecut, they will always occur in pairs of the form a+b. For each such root pair, the corresponding solution where Cet sin(be+ 6) where C and @ are the two arbitrary constants, Thus roots ~3 +/4, 245, and 0 + /7 give a solution Ce sin (41+ 94) + Cye" sin (51 + 4.) + C, sin (Te + 6., 4. Complex roots, repeated. For each pair of complex roots a ib which appears p times, the solution is Cye*sin (br + 6.) + Cyc" sin (r+ 9) +--+ C, Pet sin (br+ g,). Roots 3 ‘J2,3 £2, and +3 £2 thus give a solution C,e™ sin (2r+ @) + Cyer™ sin (2F+ 4) + Ce" sin @r~ 9) ‘The complete complementary-function solution is simply the algebraic sum of the individual parts found from the four rules. Whereas the above method for finding ¢., always works, no universel method for finding the particular solution g,. exists. This is because, q,, depends on the form of g, and we can always define a suficiently “pathological” form of g, to prevent solution for g,,,. However, if 9, is restricted to functions of prime engincering interest, a reltively simple method for finding available. Ths the method of undetermined eoefiiets, whichis bel reviewed Since the meiod does not work for alg, the first question to be answered is whether it will work for the 4, of interest. For a given g, the right side of Eq. 3.55) is some known function of time fc). To test whether this ‘method can be applied, we repeatedly differentiate) and then examine the functions created by these differentiations. There are three possibilities: 1. Afiera certain-onder derivative, all higher derivatives are ero, 2. After a certain-order derivative, all higher derivatives have the same functional form as some lower-order derivative.Generalized Performence Choractelities of ietruments ER 3. Upon repeated differentition, new functional forms continue to arse. Iease 1 oF 2 occurs, the method will work. Ifcase 3 occurs, this method will not work, and others must be tried. If the method is applicable, the solution gis immediately wr Gg AND + BPO + C1" + ~ 059) where the right side includes one term foreach functionally different form found by examining /() and all its derivatives. The constants, A. B,C, ee., can be found immediately hey do not depend on the initial conditions) by substituting a, [as given in Eq. (3.59) into Eq (3.55) and requiring (3.55) to be an denriy. This procedure always generates as many simultaneous algebraic equations inthe unknowns A,B, C, ee, a8 there are unknowns; thus the equations ean be solved for 4, B,C, et. 3.3.2. Digital Simulation Methods for Dynamic Response Analysis, Im studying the dynamic performance of measurement systems, the classical operator and Laplace ‘transform methods outlined or referenced earlier are capable of providing analytical solutions even for high-order differential equations. However, digital simulation becomes an increasingly attractive altemative as system complexity increascs, and becomes a necessity if noalincar and/or time-variant effects are to be examined. Since a detailed treatment of digital simulation application is not within the scope of this text, we use it only as a working tool whenever appropriate. Fortunately, the basic ‘methods are rather easy to comprehend, so the reader with litle o no previous experience should have no difficulty following the examples, We use the MATLABYSIMULINK language, which is well documented” forthe reader who desires greater detail. 3.3.3. Operational Transfer Function In the analysis, design, and application of measurement systems, the concept of the operational transfer function is very useful, The operational transfer funtion relating output qo input gis defined by treating Eq (3.36) sit were an algebraic elation and forming the rato of ouput opt Operational transfer function b,D" thy Dh AD+ by 2GD'4a,.D aD ta 8.60) In writing transfer functions, we always write (g./¢ MD), not just q/a to emphasize thatthe transfer function is a general relation between q, and q, and Very definitely nof the instantaneous ratio of the time-varying quantities and g, [boron OTOL oe op Dens Ooo +m General operational transfer function, EO. Doebelin, “System Dynamics,” 1998octets Measurement Systems ‘One of the several useful features of transfer functions is their utility for graphic symbolic depiction of system dynamic characteristics by means of block diagrams. That i, if we wish to depict graphically ‘a device with transfer function (3.60), we can draw a block diagram as in Fig. 3.36. Furthermore, the ‘ansfer fanction is helpful in determining the overall characteristics ofa system made up of components ‘whose individual transfer Functions are known. This combination is most simply achieved when thers is, negligible loading” (the input impedance of the second device is much higher than the output impedance of the first, etc.) between the connected devices. For this case, the overall transfer function is simply the product of the individual ones, since the output of the preceding device becomes the input of the following one, Figure 3.37 illustrates this procedure, When significant loading is present, we may apply the impedance concepts of See. 3.2 (extended to the dynamic case) or simply analyze the complete system “from scratch’ without using the individual transfer fuctions. — x SS we | wee — ot % Bea jeter ee nie ewe peti ose eee ree 22a) (FIRB combination of individual transfer functions. Io the technical Itrature, the Laplace-ransform method isin common use forthe study of linear systems, When such methods are employed, the Laplace transfer function is defined a the ratio ofthe Laplace transform ofthe ouput quantity othe Laplace transform ofthe input quantity, when all inital conditions are zero. Thus, analogous to Eq. (360), the Laplace transfer function would be written as gals) lt bs +b a) sn wheres 4.0% i@is the complex variable of the Laplace transform. We note that, as far as the form ofthe ‘ansfer function is concerned, we can shif from the Laplace form to the D-operator form (or vice versa) imply by interchanging s and D. Thus, if we encounter a block i the Laplace notation, can always convert to the D notation by a simple substitution. Then all the methods we subsequently '0, Ea, (3.70) (D+ 19,= 9, G80) 41 cam be shown generally (by mathematical reasoning) or in any specific physical problem, such as the thermometer (by physical reasoning) thatthe initial condition for this situation i g, = 0 for 1= 0° ((=0° means an infinitesimal ime after = 0). The complementary-functon solution i, 9g Ce" 681 ‘hile the particular solution is n= Ka, G82) giving the complete solution as, 4,7 Cet Ka, 6.83) Applying the intial condition, we get 0-C+Ke, Ka, which gives finally 4,240") G34) Examination of Eq, (3.84) shows that the speed of response depends on only the value of = ~“""} Ti andi faster iis smaller. Thus, in fistoorder instruments, we strive to minimize +forfsithfal Step-functin response of fstorder ddynaic measurements. Instrument.GAR boebetin's Measurement Systems ‘These results may be nondimensionalized by writing . Ke 85) and then plotting q,/(Kg,) versus i/t, asin Fig, 3.42(a). This curve is then universal for any value of K,q,, 08 that might be encountered. We could also define the measurement error &, ‘ fe (3.86) ra AG K 3.86) e249 €°9 and nondimensionalize fo poting in Fig. 3426) a5 a es ‘A dynamic characteristic useful in characterizing the speed of response of any instrument is the setting rime. This isthe time (afer application ofa step input) forthe instrument to reach andstay within 1 stated plus-and-minus tolerance band around its final value. A small sting time thus indicates fast response. It is obvious thatthe numerical value of a settling time depends on the percentage tolerance band used; you must always state tis. Thus you speak of, say, S percent setling time. For fstorder instrument a 5 percent setting time is equal to three time constants (see Fig. 343). Other percentages ‘may be and are used when appropriat ‘Knowing now that fast response requires a small value off, we can examine any specific first-order instrument to see what physical changes would be needed to reduce © If we use our thermometer example, Eq. 3.92) shows that ¢ may be reduced by 1. Reducing p, C,and V, 2. Increasing Uand 4, ‘Since p and Care propertis of the fluid filling the thermometer, they cannot be varied independently of each other, and so for small x, we search for fluids with a small pC product. The bulb volume V, ‘may be reduced, but this will also reduce 4, unless some extended-surface heat-iransfer augmentation (uch as fins on the bulb) is introduced. Even more significant is the effect of reduced V, on the static sensitivity K, as given by Eq. (3.78). We sce that attempts to reduce t by decreasing V, will result in reductions in K. Thus increased speed of response is traded-off for lower sensitivity. This trade-offs not unusual and will be observed in many other instruments. (HARRY Honcimensional step function response of frt-order instrument.Generalized Performance Characteristics of struments EIEN ‘The fact that ¢ depends.on Umeans that we cannot 2-4 state that a certain shermomeler has a ceriin time" constant but only thata specific thermometer used in 4 certain fluid under certain hea-transfer conditions (G03, free or forced convection) has a certain time ‘constant, This is because, U depends partly on the value of the film coefficient of heat transfer at the outside ofthe bulb, which varies greatly with changes in uid liquid or ga), flow velocity, ete. or example, thermometer in stirred oil might have a time constant a8 F of 5 s, while the same thermometer in stagnant air would have a £ of perhaps 100 s. Thus, you must (EAUSIBD) Setting-time definition always be careful in giving (or using) performance (setting tine) data, to be sure thatthe conditions of usc correspond to those in force during calibration or that proper corrections are applied. ‘A temperature prote is transferred from air at 25°C to alr at 35°C, then to water at 70°C, and back to air at 35°C, Assume that in each case the transfor is “nstonta- neous”. The effective tine consiants and the Liming sequence are 8s folows: In ait, probe dry, = 20; In water, s: In at, probe wot, += 20s; for 10, T= 25°C (inital temperature) w O 7, our differential equation is (2+ 0g," Ke,=0 6.10 which gives g,= Ce? .102) The constant Cis found by imposing intial condition (3.100), Haare cen 6.103) ai @.109 giving finally g, =A40—2 2 @.105) Ter Figure 3.484 shows a typical response, and Fig. 3.48¢ shows the effect of cuting Tin balf. As Tis, ‘made shorter and shorter, the fist part (t< 7) of the response becomes of negligible consequence, so that ‘we can get an expression for q, by taking the limit of Eq, (3.105) as T—> 0. 6.106) <—=5 an indeterminate form Applying L’ Hospital's rule, yieldsGeneralized Performence Choracteistes of struments EX. ee 1 li = (3.107) Mera a Thus we have finaly for he impulse reponse of fit-onder instrument KA ain (108) + ‘which is plotted in Fig. 3.484. ‘We note thatthe output g, is ls0 “peculiar” in that it hes an infinite (vertical) slope at = 0 and thus _g0es from zero to a finite value in infinitesimal time. Such behavior is clearly impossible fora physical system, since it requires energy transfer at an infinite rate. [n our thermometer example, for instance, to cause the temperature ofthe fui inthe bulb suddenly to rise a finite amount requires an infinite rate of| ‘heat transfer. Mathematically, this infinite rate of heat transfer is provided by having the input 7) be infinite, i., an impulse function. In actuality, of course, 7, eannot go to infinity; however, if itis large ‘enough and of sufficiently short duration (relative tothe response speed ofthe system), the system may respond very nearly as it would fora perfect impulse. a o7 08 Os ERI) Cxct and coprosimate impulse response. ‘To illustrate this, suppose in Fig, 3.48a we take A = | and T= 0.01 +. The response to this approximaic it opule WOK Gem) Osis 6.109) = panngit 100K Teise gar 16EID doeveitn’s Measurement Siem Figure 3.49 givesatabularand graphical comparison ofthe exact and approximate response showing ‘excellent agreement. The agreement is quite acceptable in most cases if 7/t is even as large as 0.1. It an also be shown tat he shape of the puss tmaterial as lng its duration suffice short, only is awa mates. The plausbilty ofthis statement may be shown By integrating the terms inthe differential equation as follows: a, at t= Ka, @.) Wp ta, + fy" goat = fy" Kyat @.112) tJ. 4,)+ 0= K (area under g, curve from 1=0t01= 0°) @.113) GiJe. = K'¥(area of impulse) @.14) ‘This analysis holds stricily for an exact impulse and is 2 good approximation for a pulse of arbitrary shape, if ts duration is sufficiently short. It should be noted that, since the right side of the differential ‘equation is zero for 1 > 0°, an impulse (or a short pulse) is equivalent to a zero forcing function and a Kir at is exactly the same as the unit impulse response. ‘Another interesting aspect of the impulse function is its relation to the step function. Since a perfect step Func- tion is also physically unrealizable because it changes ‘from one level to another in infinitesimal time, consider an approximation such asin Fig. 3.50. If this approximate step function is fed into a differentiating device, the out- ” ry te Do il bea polyp faction. Aste approve ep wi) function is made to approach the mathematical ideal more ani more closely, te oupt ofthe difeentiating device vl approach a perfect impulse function, Za ths sense, Ihe inpuse fonction maybe tought ofes the deivetve othe step nction, eventhough the dsconttisin he Step funtion pci the rigorous aplication ofthe ba Si detntion othe denvate. In Fig 3 50, the uh of these fe 3.80) Approximate step npuie Semcon Ying te Opt he feet device ough an ineraing device (1D) 5) sy m0 3.3.11 Second-Order Instrument A second-order instrument is one that follows the equation G.16)Generalized Performence Cheracterlities of struments REED Again, a second-order equation couldhave more terms onthe right-hand side, butin common engineering usage, Eq. (3.116) is generally accepted as defining 2 second-order instrument, ‘The essential parameters in Eq. 3.116) can be reduced to three: 42 4 static sensitivity oan 0 99 wtp a egy ine Guy fa nar 8 damping ratio, dimensionless G19) which gives ( 22.1) (3.120) ‘The operat ser funni ths £O=sg7aeaT (aan A good example of @ second-order instrument is the force-measuring spring scale of Fig. 3.51. We assume the applied force f,has frequency components only well below the natural frequency of the spring itself. Then the main dynamic effect of the spring may be taken into account by adding one-third of the spring's mass to the main moving. ‘mass. This total mass we call Mf, The spring is assumed linear with spring constant K, newtons per meter. Although in a real scale there might be considerable dry friction, we assume perfect film lubrication and therefore a viscous, damping effect with constant 2 (in newtons per meter per second). ‘The scalecan beadjustedsothatx,=0 when = O(gravity force will then drop out ofthe equation), which yields E forces = (mass) x (acceleration) Pix: 3.31.) Second-order instrument, és #s, 12) Si BFP- Kay = ME (MDt+ BD + Kx,=/, G12 ‘Noting tis to fit the second-order model, we immediately define 1 Kam @.124) dls 3.128)(GEAR oebetin's Measurement Systems B casey as Fart spring sal of ig 3.1, levee he tae btwoon sense and speed frosoane mating fom change KS. Solution From Eq. 3.124, increase in stiffness reduces static sensitivity From Eq. 3.125, increase in stiffness increases speed of response GEEIIGERERD ind tnotranster function ota spring scale (Fig. 3.51) whose mass is nogigiblo. Show that the steady-state me lag fr @ ramp input i the seme whether mass is zero or not Solution Second onder system of Fig 3.51 From Fig, 3.53, time-lag t for the second order system for ramp input os a ‘B: Damping coefficient kz Stiffness, Nim First order systems of Fig, 3.51 itmis mate DB ©, isvery lage J 0 ee 8B Since ok from Eq. (1), the transfer function of the first order system becomes fee. od Bor ari where i From Fig, 3.54 steady state time lag for ramp input toa fist order system ~ +~ Blk, ‘Therefore, iespective of whether mass neglected or not, time lag remains the same because of constant velocityGeneratzed Performence Chorctertstls of Instruments SEED 3.3.12. Step Response of Second-Order Instruments Fora step input of size g, we get (& Balan, en ofa, ihn condions qe0 ar=0" feng aime osm ‘The particular solution of Eq (3.127) sclearlyq,,~ Ka,- The complementary function soliton takes enone of tee possible forms, depending on whut the ato the charter eustion eee and trrepeted (overdamped cs) rea and fepentd(eriealy damped ease, or complex (underdamped case). The complete solutions of Eq. (3.127) with intial conditions (3.128) are, in nondimensional fern, SANE pce iine SET et gon ae ae +1 overtamped (3.129) Br atone +1 critically damped @.130) out £ = = sin (iG a¢+9)+1 underdemped 3) Ka, i-G 4 ogsin' fF Since 1 and @, always appear as the product «, the curves of 9,((Kq,) may be plotted against a Which makes themn universal for any «as in Fig. 3.52. This fact also shows that a, is adirect indication ofspeed of response. Fora given &, doubling @, will halve the response time since @,f[and thus ¢,(Ka,)] achieves the same value at one-haif the time. The effect of Cis not clearly perceived from the equations but is evident from the graphs. An increase in ¢ reduces oscillation, but also slows the response, in the sense that the frst crossing of the final value is retarded, A settling time actually mey be a better indication of response speed; however, then the optimum value of ¢ will vary with the chosen tolerance band. For example. if we choose a 10 percent setling time. the curve for £~ 0.6 gives asetiling time of bout 2.4/o,, and this is optimum, since either larger or smaller ¢ gives a longer setting time. However, it;we had chosen a 5 percent settling time, a £ between 0.7 and 0.8 would give the shortest value, In choosing a proper {value fora practical application the situation is further complicated by the fact that the real inputs will not be step functions and their actual form influences what willbe the best £ value. If the actual inputs are quite variable in form, some compromise must be struck. Many commercial instruments use = 0.6 oF 0.7, We show shortly that this range of gives good frequency response over the widest frequency range.Dovbelin's Measurement Systems a oe Fig, 2:52.) Nonatmensional step-function responce of second-order instrument. 11 Ramp Response of Second-Order Instruments “The defferential equation here is ant! se Bt 2p ® SO ia, = Ka. (3.133 (& o ie i @.132) x ' eo ona =H 20 120 Figure3.53 shows the general characterof the response, There —+ is a steedy-state error 264,/o, Since the value of dj, is set by the measured quantity. the steady-state error can be reduced only Fig. 3.53.) Ramp response of by reducing ¢ and increasing 0, For a given a), reduction in £ —second-erder instrument. resulsin lager oscillations. There i also a steady-state time ag 2c, Figure 3.54 gives ase of nondimentionalized curves that summarize system behavior.Generalized Performance Cherctertetis of struments KEES 3.3.14 Frequency Response of Second-Order instruments ‘The sinusoidal transfer function is be ao) 3.139) 4” Gore.) B.39) ‘which canbe pt in the form LK nym G3 % ater G25) 1@,-0,/0 Figure 3.55 gives the nondimensionalized frequency-response curves. The flat response region for Coley, <0.1 isthe usefull portion of the curve, because the transducer, modeled as a seond order system a ale Fig. 3.58°) Frequency response of second-order instrument.Doebein's Measurement Systems can be used to measure dynamic signals inthis region, without having to calibrate at every frequency. Clearly, increasing o, will increase the range of frequencies for which the amplitude-ratio curve is relatively lat; thus high oy is needed to measure accurately high-frequency qs. An optimum range of values for is indicated by both amplitude-ratio and phase-angle curves. The widest flat amplitude ratio ¢xists for of about 0.6 to 0.7, While zero phase angle would be ideal, itis rarely possible to realize this even approximately. Actually, ithe main interest isin 9, reproducing the correct shape of 9, and if ‘time delay is acceptable, we show shortly that @ need not be zero; rather, it should vary nearly with frequency a2 Examining the phase cures of Fig. 3.55, we note that the curves for ¢= 0.6 0 0.7 are neatly straight forthe widest frequency range. These considerations lead tothe widely accepted choice of $= 0.6 100.7 as the optimum value of damping for second-order instruments 3.3.15 Impulse Response of Second-Order Instruments In the section on firstorder instruments we showed tha the impulse response is equivalent tothe fee (unforced) response, ifthe intial (¢ + 0°) conditions produeed by the impulse are taken into account. To find the inital conditions produced by applying an impute of area to second-order instrument, redraw the block diagram of Fig. 3.56(a) as in Fig. 3.56(t). (The equivalence of the two diagrams is casly demonstrated by tracing through the signals in Fig. 3.56(o) to get the differential equation relating 49,104.) In Fig. 3$6(c) the impulse is applied at q, and the “propagation” ofthis input signal is traced through the rest of the diagram. This analysis shows that at «= 0° we have q, = O and 4, = KAal. The differtial equation tobe solved is then & G36) Block-diagram analysts of impulse response.Generalized Performance Choracterlithes of struments KK The solutions are found to be Ge LU (em. p Kaw, 2-1 e Ge ye! Fig, HE cntialy damped fog Noten sin (fie Fin pipe le a Figure 3.57 displays these resulis graphically. 3. @437) 38) underdamped (3.139) 16 Logarithmic Plotting of Frequency-Response Curves ‘We will find the frequency response of measurement systems extremely useful, so that rapid methods. for getting the amplitude-ratio and phase-angle curves would be helpful. Certain logarithmic methods are in wide use and are expla ‘The sim .d now. 3! transfer function of a measurement system generally can be put in the form Koy" (Jot, +1)... JOT, + 1) [GO,) + 26, joo, +1) Ge (Gola, joie, +(e" tdt)...(e! Ua@)= a (3.140) % Gor, + 1). Gory + GO! +26, jolay, +1) G.140) AGW yp)? + 2p fO/Dyy + 1) ‘This follows fromthe factthatthe polynomials —_g, in the numerator and denominator of Eq. (3.60) Far { can ingeneml befactorediniotemmsofthe forme Dr tD+ 1, and Die + 25DIa, + 1. Replacing D by j@ then gives Eq. (3.140) when dead-time ‘elements are also included. * Since Eq, 3.140)is inthe formofproductof complex numbers, the use of logarithms suggests a itself as a means of replacing mulipicaion by addition. That is, 0 aa = GYOG.{jo)...G,jo) G.141) ° where the G{Ja) functions represent the various _,y terms of Eq. (3.140). The amplitude ratio would be given by os fe oo (G40 16,00) 16, ge0y “oe O12) oa C2) GPiESIST) Nondimerstonal pulse espns of second-order instrument.ITD oebetn’s Measurement Sater ‘A widely used logarithmic method uses the decibel novavion to express amplitude ratios. An amplitude ratio dis given in decibel (4B) by Decibel value A dB A 20 log 4 143) Then 20 log in] - 20 log 16,(a| |G,ia)) ...1G,an) G4, 20 og |e cel20 log IG,(ia)] + 20 log 1G, /a)| + -- +20 log (6, @.145) “Thus, if we get the amplitude-ratio curves forthe individual terms in Eq. (3.140) in the decibel form, we can obtain the overall decibel curve by simple graphical addition, The phese-angle curves also are obtained by simple addition, since we add phase angles when multiplying complex numbers. ‘We now show how to obian, with a minimum of effort, the amplitude-ato (dB) and phase-angle curves for each of the types of terms in Eq. (3.140). To put the curves in their simplest form, we plot against the logarithm of frequency rather than against frequency itself The simplest term isthe sensitivity K, which is areal number K 0°; thus its decibel curve i a straight horizontal line through the decibel value ofK, while its phase-angle cure is straight horizontal line through zero degrees (see Fig. 3.58). “The next type of term is (j where n= +1,42,.... The phase angle of such terms is constant with frequency and is given by 90x". The amplitude ratio is ar, so thatthe decibel value is 20 log c= 200 log @ Since we plot against log «the decibel curves become straight lines of slope 20n dB/decade 3.58). A decade is defined as any 10: | frequency range: an octave is any 2: 1 range. eo abo Fig@lB’SH) Imcegrotor and aifferentiacor frequency response.Generalized Performance Characteristics of Istruments EKED ‘Terms of the form javt + 1 and 1/0" + 1) give, respectively, a8 = 2010g lar) +1 B.146) and dB=~20l0g Ylory +1 G47) ‘When o>, these become 4B =20 log wt= 20 log t+ 20 log @ @.148) and dB ~-20 log t= -20 log 1-200 @ G49) ‘We see that both represent straight lines of slope 620 dBidecade, and these straight lines will be the high-frequency asymptotes of the actusl ampliude-ratio curves. Similarly, for wt 1, dB ~20 log 1=0 3.150) and dB=-20 og G.151) so thatthe low-frequency asymptote is simply the 0-dB line. The two straight-line asymptotes will meet at @r= I because, this is where (3.148) and (3.149) are zero. The point «w= 1/7 is called the breakpoint, (or corner frequency. In plotting curves for such terms, we first locate the breakpoint and then draw the ‘two asymptotes. The true curve is obtained by correcting the straight-line asymptotes at several points, ‘using the data of Fig. 3.59. The phase-angle curves may be quickly plotted by using the data of Fig. 3.60. ‘A numerical example illustrating these methods is given in Fig. 3.61 2008 eae 2328 Ce 2067 ae 2097 ov a OF 93 8805 GF 1 IS First-order-system amplitude-ratio corrections. T ‘OT OF OS OOS OF ITE Tem Fig. 3.805) First-order-system phase angle.Doebeln’s Measurement Systers ‘Terms of the form [jaro,)? + 2¢)a/a, + 1}* have low-frequency asymptotes of O dB and high- frequency asympiotes of slope +40 dB/decade. They intersect at «= @, The exact curves for a given are obiained by applying the corrections of Fig. 3.62. The phase-angle curves are obiained from Fig. 3.63. Figure 3.64 give numerical examples of second-order terms. 08y+) i | i os) i | | | ere) Fig{B61) Gxomple of first-order terms. 510 Seconchorder-system amplitude-ratio corrections. The final type of term considered is the dead-time term v’*, Since the amplitude Frequencies, the decibel curve is simply the O-dB line. The phase-angle curve ¢=~wr, forany given dead time, ‘Tillstratethe procedure for combining the individual terms to obtzin the overall frequoncy-response curves, we consider the following example: is 1.0 for all easily plotted from @.152) Figure 3.65 shows the procedure and results.Generalized Performance Characteristics of Instruments EK ‘OF 02 a3 aaes OF 10 2 3 4 5 6 T BoD (FRBBIB)) second-order system phase angle. 0 90 smsaee sop0 __so.ec0 tae ll & Prasad deg Arete ao, & bbovs PEPE (BRISA) crop te of secondorder tems. ‘The plots in Fig. 3.65 are known as bode plots in control engineering in which logarithm plots of the ‘open-loop transfer function are drawn to investigate stability [aera o 8h Phase oe deg & Fig! 3.65)) Example of freavency-response plot.EEPA ocbetin's Measurement Systems ‘The“manual” graphing technique just explained is today mainly used asa quick and rough she‘ tool; we use computer software when we need accurate graphs. Manual sketching will aay remain 2 usefl skill a the system design stage since it gives clear insight into the effect of individual system components on the overall sysiem behavior. When we do finally use computer aids to plot accurate traphs, the manual sketches allow us to also catch computer data-entry errors since we know roughly What the curves showld look like. Most mathematical software provides for easy frequency response plotting. For the MATLAB used inthis book, we enter, den=[a,a,.,---4, 4) num =[5, by. +6) By) bode (num, den) using the a's and d's from Eq, (3.60). Figures 3.66 and 3.67; and 3.64, respectively. the results, corresponding to Figs. 3.63, Freie ae ‘e200 1] dane800"2 ate rum en)Generalized Performance Characteristics of instruments EES Fara te ce nessun pln Fi. 292 bic own, esi eet teins iat, ais ootheceaael bao tion analogous to Eq. (3.64). Solution ‘ be 6 be pan “ “s yes im “ee Ee ig 67, hk “eek o Looky, @ Equation of motion [i alse eld ® ‘Taking Laplace transform of Eq. (3) and assuming zero initial conditions (+e, + EM (5) A XY5) = 0 @ KK) + y+ ht Si Fgg() © From Eq. (4), 6 { Bet PK thet PMO» ye nayte) Flt o X= = [+A Ms + (bs +k IM] Sli lik gh + gky = AEPED oebelin's Measurement Systems Taking Laplace transform of Fg. 2) F,(3)=4,X) begs) F,(s)= 7] Fata Fa [U6 My Ob +h ® A hiky yk + hy + gk KE FA ik ? hs © Fegi() Milas +[(h, +h, )Mz + (b+ KOM, Stik thik kyl hyky KE pak =k=INém k,=100Niem M.= M001 ke ‘Abode diagram of the function of Ea. (9) is shown on net page as Fig 3.68 It is clear that, upto 9 radi, the measured and applied osds are almost same. Beyend 9 rads they are not acceptable a f bo mal thé «rsGeneralized Performence Cheracterstles of Instruments EEECELAED | Peenatyzo the vottage-measuring problem of Fig, 3.23 fr dynamic operations, replace the batteries with sources of te-varying vllage. Aso, lt the vollage-mes- ‘suring device bo an oscifoscope with R, shunted by a capactor C,, Solution For an oscilloscope with resistance R, shunted with a capacitor C, R, Bebe 0 Equation (3.41) can be modified as En Biz ® Svbsining £91) in@) ® O} Reanaiyze the currentsneasuring problem of Fg. 329 for dynamic operation: i. ‘replace the batteries with sources of me-varying voltage. Also, lt the current.moa- suring davice have an Inductance Lin sores with Ry. Ratz, a =R,+ LD ® From Eq. (1) and (2) Ray 1 "RFR 37D ® By factoring out R, +R, in Eq, (3) )EIN coebetin's Measurement ystems 3.3.17 Frequency Spectra of Amplitude-Modulated Signals Interest in amplitude-modulated signals stems mainly from two considerations: 1. Physical data that are to be measured and interpreted sometimes are amplitude-modulated, 2. Certain types of measurement systems intentionally introduce amplitude modulation for one or more benefits. While, in general, the signal that modulates the amplitude of a carrier wave may be of any form (single sine wave, general periodic function, random wave, transient, et.) and the carrer may be given \ soe rection tra erptute a TA ois rejects és Application of amplitude ‘modulation. We now extend the amplitude-modulation concept to signals other than just a single sine wave. Ifthe modulating, signal isa periodic function (0), itmay be expanded ina Fourier series to get the output ofthe modulator MEL Sy ain MRL wo Eade som wey ‘hich can be writen as Output = 4,4, sin @, 1+ (4, 4,608 ©, rsin @, 1+ 4, A, cos @, sin @, 1+ -~) +(B. Asin, rsin@, 1+ BA, sino sine r++) BSB) Now, sin asin t= Ecos o~ )— 5 00s a=) , 1 i and sinzeos B=4sin(a+B)—4 sin sin cos B= sin (a+ B)— > sin(a+ B) and so Output = 4,4, sin @, 1+ C, sin (sin[@,+@)—a)]+sinlo-a)erad + — G.159) where cat [a+ BP Game & G.160)(ERI Doebetin’s Measurement Systems \We see thatthe spectrum ofthe output signal isa diserete spectrum containing the frequencies 9, a, a), @.£ @, @, + ©, etc. Thats, each frequency component ofthe modulating signal produces one pair of side frequencies (sce Fig. 3.75) Ifthe output ofthe moduletor is epplied to the input ofa system with known frequency response, the methods of Fig. 3.72 ean be used again to find the steady-state output. If the modulating signal isa transient, the spectrum of the modulator outpat may be determined with the help of the modulation theorem” for Fourier (or Laplace) transforms. Ifthe modulating signal is a tronsient f(0, it will have a Fourier transform F{/a), which can be obiained in the usual ways. The ‘modulation theorem leads tothe following result i) is multiplied by the carrier 4, sin «, to produce the modulated output: Fourier transform of modulated output 81Q.Uw1 20,70) G61) vthere (Q(e9)= 4 magninde (Fo a} 162) £0, (10) = angle (F(w>— «9)}} - 90° Gta) and 0S @ < e, Note that the argument (i is now je @,), so that if we want to find Q, (4,000) and c= 3,000, we must evaluste (1,000). Also note tha, to get Q{je) for 0 < w, < %, we must know Fi) for negative a's, since any @, @, gives F{(@— @)) a negative argument. While generally we have worked with positive as, the transform for negative as always exists and is easily found from the previously given symmetry rules. The spectrum given by Eq, (3.161) will ea continuous one, and ifthe modulated output is applied as an input to some systom with known frequency response, then the corresponding output canbe obtained by using magnitude and phase relations. “ware alll. = i (HERBY) Frequency spectrum when modulating signal is period. For measurement systems in which amplitude modulation is intentionally introduced to allow the use of carrier-amplifier techniques, the carticr frequency must be considerably greater (usually 5 to 10 times) than any significant frequencies present in the modulating signal. For such a situation, the 'G. A. Korn and T. M. Korn, “Mathematical Handbook for Scicntsts and Engineers” McGraw 1961, p. 219; Papoulis, “Fourier integral and its Applications.” p. 15. il, New York,Generalized Performance Characteristics of netruments KEEL pertinent frequency spectra are as shown in Fig. 3.76, We see again that the amplitude-modulation ‘process shifts the frequency spectrum by the amount a, ny = FABRE’) Frequency spectrum when modulating signal is transient. ‘When amplitude modulation is intentionally introduced to facilitate data handling in one way or another, it generally plays the role ofan intermediate step, and the amplitude-modulated signal usually is not considered a suitable final readout. Rather, the original form of the modulating signal (the basic ‘measured data from, say. transducer) should be recovered. The process for accomplishing this involves demodulation (ot detection, a it is sometimes called) and filtering. Demodulation may be fll-wave, half-wave, phase-sensitive, or non-phase-sensitive (Fig. 3.77). Here we treat the form giving the best reproduction ofthe original data full-wave phase-senstive demodulation, and consider only the process, ‘not the hardware for accomplishing it. Again, itis necessary to consider whether the form of te original signal was a single sine wave, periodic function, or transient. Fora single sine wave A, sin «which is modulating a carrier 4, sin ©) , the expression forthe fall-wave phase-sensitive demodulated signal is Demodulator output ~ (4, sin @, | 4, sin «4 G16 asseen fiom Fig. 3.784. Now [4, sin «is periodic function and may be expanded in 2 Fou The results are A, sinc 4A, sin £603 209, 1 - “4 sin w, t00s 40 1+ ise ‘@.167) ‘Now, by a trigonometric identity, tems of the form (sin o£) (1,2, can be writen as (sin (20, + sin ney — a) 442. Thus we can write Eq. (3.167) as(BEAR oebetin’s Measurement Systems i (CBB) rer 1 srnttin Demoulor ouput = 244. sino, ¢- 244. fin Qo, +0,)¢ ~sin(20,-@y) ~ 244 (sin 4, + oy-sin oy] + ~~ (3.168) From this we see thatthe frequency spectrum ofthe demodulator output signal i a discrete spectrum swith frequency content at @, 20, + 0, 40, , etc, as shown in Fig. 3.78(b). If this signal were an imput to system of known frequency response (such asthe filts of Fig. 3.77)sthe output ofthis system ‘would be found by the methods of Fi. 3.72. Ifthe outputof the filter is to look like the original data, the filter must be designed to reject frequencies 20, £ a, 4, * @, etc, while passing with & minimum of distortion the signal frequency «, The design of such a low-pass filter is made simpler ifthe passband And the rejection band are more widely separated. This is the basis of our earlier statement that carrier frequencies usually are chosen to be 5 to 10 times the highest expected signal frequency. ‘When the modulating signal is a periodic wave rather than a single sine wave, a procedure similar to that just used is employed, except now the modulating signal is also expressed as a Fourier series of the formGeneralized Performance Charactetties of ietruments EEL azine (CBR reqvency spectrum of put-wavephase-senstvedemodultion, Modulating signal =4,,+4, sin (@, 1+ a) +4, sin (2ay 6+ 0,) + ~~ G.169) ‘When this is multiplied by |A, sin o,f as given by Eq, (3.165), we find exactly the same situation as for ‘single sine wave, butt must be applied for each signal frequency (0, @, 20, 3a), ete). The frequency spectrum of the demodulated signal thus will be 2 discrete spectrum with frequency content st «= 0, (0, 20, 20, £ 0, 40), 40,4 0, 60, 60), + ©, ete), (20,20, + 2(09, 40, + 20, 60 + 2a, te), te Figure 3.79 illustrates these concepts. Again, if such a signal is applied to a system of known frequency response, the methods of Fig. 3.72 allow calculation ofthe output. Magne Denoa.ta 8 a Se Face | tare | oot oe tat, 24 tM, noo porta (BBR cereasaton spectrum for pertoce putEEE Coebetin's Measurement Systems ‘esha te ie em sal 4 sine wih anteens broatiot 024 (2-Zendq 1-2 crt.) arm oar ming fulated signal = 24 yr) 44 gcry00s Demodulated signal = "4: 4 -"2. fpcos 20,1 44, Ad 5 (9 cos 4, ea re F(t) 008 4,1 6.17) Application of the modulation theorem to each of the modulated terms of Eq, (3.171) leads to the result Fourier transform of demodulated signal 4 |Q(/o)| O(a) 24, 24, 24, wh 32k 2 et jw - 24 eo . 5 ere |e) EGON + Sl fo — 202.) + FLO — 40) + Gan) and 0S wo». The expression for ZQ(/) is not easily given since in Fourier-transforming Eq, (3.171) (cach term may be treated separately since superposition is allowed). we find Q, (jc) as astm of complex ‘numbers rather than a product. To find the overall angle of sum of complex numbers, we must express ach number 2s a+ jb, and then add the a's and B's to get the overall a +b. The angle is then tan" (bia). The angle due to transforming /(0 is ZF(je), the angle due to transforming -f{0) cos 20, «is ~180° + ZF, [a ~ 20}, the angle due to transforming ~f() 0s 40), ts - 180° + ZFI(o— 4e)], ete. These results allow calculation of ZQ{/e). The frequency spectrum of such a signal is shown in £80. Since it isa continuous spectrum, the respanse ofa system to such an input may be found by igen ou of ee ee Fig! 3{80)) Demoduiation spectrum for transient input.Generalized Performance Choracteisties of ietruments EY using magnitude and phase relations, In Fig. 3.80 itis assumed thet the F(a) is practically zero for @ > a,,and that ©, > @,. For such a situation, the phase-angle calculation is greatly simplified, since the individual terms of Bq, @.172) do not coexist over any frequency range. That is, for 0< @ < 0 oy only the first term is nonzzro; for (2a, - @, ..) 1.0), no oscillations exist, and the determination of {and a, becomes more difficult. Methods for dealing with this case are left to a reference.™ Frequency-response methods also may be used to find Cand @, or #, and 4, Figure 3.84 shows the application of these techniques. The methods shown use the amplitude-ratio curve only. Ifphase-angle curves are available, they constitute a valuable check on conformance to the postulated model, For measurement systems of arbitrary form (as contrasted to first- and second-order types), description of the dynamic behavior in terms of frequency response usually is desired, This information. ‘may be obtained by sinusoidal, pulse, or random signal testing, following the general methods" used 10 ‘experimentally determine mathematical models for physical systems. When the physical system being, studied is a measurement system, the output signal q, is itself generally useful and no separate output sensor is required. However, we do usually need to measure the input signal g, with separate sensor, ‘which serves as the calibration standard and whose accuracy is known, and, if possible, is about lO times better than that ofthe system being calibrated. If we can obtain {q./q) (a) thus for the measurement. system, this defines the frequency range over which corrections are negligible and provides the data ‘needed to make dynamic corrections (using the transform methods of the previous section) if we wish 10 use the instrument in its nonfat range of frequency response. 3EO. Deosbelin, “Measurement Systerts, 4th ed. p. 192. »B.0. Doebelin, “System Modeling ané Response,” chap. 6.1980,