Download as pdf
Download as pdf
You are on page 1of 122
IN AMERICAN NATIONAL STANDARD 8 ASME PTC 19.1-1998 Test Uncertainty Instruments and Apparatus PERFORMANCE BS) Orgel TEST CODES Date of Isuance: December31, 1998 This document will be revised when the Society approves the issuance of the next edition. There wil be no Addenda issued to ASME PTC 19,1-1998. Please Note: ASME issues writen replies to inquiries concerning interpretation of technical aspects >this document. The interpretations are not fart of the document. ASME PTC 19.1-1998 is being issued with an automatic subsciiption service t the interpretations thet will be issued to it up to the publication of the next edition. [ASME is the egistorad vada of The American Society of Mechatical Engineers This code or standard was developed under procedures accredited as meeingthe citera for American National Sundards, The Consensus Committe that approved the code a standard wasbsalanced to assure ‘hat indivicuals fom competent and concemed interes have had an epportinity to parleipate. The Proposed code er siatdard was made availble for public review and comment which provides an pparunity for adstinal public input from industry, academia, regulatory ajenctes, and the public- at-large. ‘ASME dors not “apfrove” “ate,” or “enderse” any item, constuction, proptatary device, or activity. ASME does nattake ny position withroepact to tha validity af any pater sighs atserted in connection ‘with any items mentiored inthis document, and does not undetake' Insire anyone wilzing a standard gaina ability fr infrngement ol any applicable Letters Patent, nor assume ary such Hiabity, Users of 2 code or sundatd areexprssly avisoe that dotermination ofthe validity of any suck patent right and the risk of iningemen: of such rights, i enthely their own rexponsbility. Partcipation by fedes! agency representatives) cr person(s) filiated with indsty isnot tobe interpre- ted as government or dusty endorsement of this code or sandare. ASME accepts :esponsibilty for only those interpretations iived In accordance with governing ASME procedures and policis which preclude the Isuance of interpretations by indhvidual voluneers, Ho par of his document may be reproduced in any fon, in an electonic retrieval sytem cr otherwise, without the prior writen permission ofthe publisher. “The American Sodlety of Mechanical Engineers Three Pare Avenue, New Vad NY 10016-5900 Copyright © 1998 by THE AMERICAN SOCIETY CF MECHANICAL ENGINEERS All Rights Reserved Printed in US.A, FOREWORD (this Forewend isnt par of ASME PTC 181-1988) The work on the current edition was begun in 1991. The twc-fold objective was to improve the usefulness to the reader as regards clarity, conciseness and technical treatment of an evolving subject matter, as well as harmonization with the ISO "Guide to the Expression for the Uncertainty in Measurement.” This PTC Supplement was approved by the Board on Performance Test Codes on 30 May 1996. It was approved as an American National Standard by the ANS! Board of Standards Review on 20 March 1998, NOTICE All Performance Test Codes MUST adhere to the requirernants of PTC 1, GENERAL INSTRUCTIONS. The following information is based on that cocument and is included here for emphasis and for the convenience of the user of this Supplement. It is expected that the Code use is fully cognizant of Parts | and II] of PTC 1 and has read them prior ‘0 applying this Supplement. ‘ASME Performance Test Codes provide test procedures which yield results of he highest level of accuracy consistent with the best engineering knowlecge and practice currently available. They were developed by balanced committees representing all concerned inter- ests. They specify procedures, Instrumentation, equipment operaing sequiremenis, calcula tion methods, anc uncertainty analysis. ‘When tests are run in accordance with a Code, the test results themselves, without adjustment for uncertainty, yield the best aveilable indication of the actual performance of the tested equigment. ASME Performance Test Codes do not specify means to compare those results to contractual guarantees. Therefore, itis recommended that the partes to a commeréial test agree before starting the test and preferably before signing the contract on the method to be used for comparing the test results to the contractual guarantees. It is beyond the scoe of any Code to determine or interpret how such comparisons shall be made. PERSONNEL OF PERFORMANCE TEST CODE COMMITTEE NO. 19.1 ON TEST UNCERTAINTY (The following isthe roster ofthe Commitee at the time of approval ofthis Supplement) OFFICERS Ronald H. Dieck, Chair W. Glenn Steele, Vice Chair George Osolsobe, Secretary COMMITTEE PERSONNEL D Allan Coutts, Westinghouse Savannah River Co. Ronald H. Dieck, Pratt & Whitney Pilip A. jones, US. Generating Co. Paul K. Macisjewai, Univenity of Conectiut Pail E, McCarty, Sverdup Technelogy, Ine, AEDC Group Mark L. Price, Rols Reyee, Ine Jefrey A. Rabensttine, Power Generation Technologies Ralf B. Schumacher, Coast Quality Merology Systems, Inc. W. Glean Steele, Mississippi State University Jorn’. Wyler, Smiths Industies ‘The Committee wishes to icknowledge the contibuton of Past Member and Past Chair Robet B. Abernethy. RP. Allen 1. Banaiser R. Fiedran JeGerbe ~ M. Gerhart Ss. Hecklnger BOARD ON PERFORMANCE TEST CODES OFFICERS D.R. Keyser, Chale FM, Gethery Vice Chait WO. Hays, Secretary ‘COMMITTEE PERSONNEL RW. Henry C.B.Scharp DR. Keyser 1A. Silvaggo, i S.J. Kovels J. W. Siegmund J. maon W.G, stele G.H. Mitended jr Rf. Sormerlad ~ STP Nuspt J.C Weseor RP Perkins 1. Yost ALL Plunley Foreword .... 4. see Commitiee Roster... 66. see. Board Roster..e0.es45 CONTENTS Section 1 Introduction 2 Object and Scape,» 3 ions and Description of Terms 4 Fendameral Concepts... .. 5 Defining the Measurement Process . 6 Uncertainty of a Measurement 7, Uncenainty of a Result 8 Additional Uncerainty Considerations 9 Step-by-Step Calculation Procedure... 10 Examples 11 References... beteteeeeeneees 12 Bibliography |. . ‘Appendices A Statistical Considerations... 2. vee B Uncertainty Analysis Models. : : C Propagation of Uncertainty Through Tayior Series. D Uncertainty of Derived Results on K-Y Plots ...... Figures 4.1 Illustration of Measurement Errors . 4.2 Measurement Error Componenis. ... te 43 Distribution of Measured Values (Normal Distribution) «2... 4.4 — Test Dats (ample 4.5-1) 5.1. Generic Measurement Calibration Hierarchy. 6.1 Pressure Date Acquisition Systern 6.2 Uncertainty Interval 8.1 Schematic Relation Between Parameters Characterizing Nonsymmetric Uncertainty ..... seteeeees 8.2 Relation etween Parameters Characterizing Noneymmotic Uncertainty. beveteees 83 Three Posttes: Cases... oscilla 10.1 Traverse Points (Example 10.1). 10.2 Schematic ofa 6 x 4 in. Venturi. beste 103 Typical Fressire and Temperature Locations for Compressor Eiciency Determination. 83, 95 101 107 10 " R “4 16 20 2 30 32 34 48 55 10.4 Simplified Flow and Instrumentation Diagram for a Nonextraction Condansing Turbine . oe see 8B 105 Generator Electrical Losses [12] (Example 10.5)... 60 106 Throttle Pressure Correction [12] (Example 10.5). cece 61 107 Throttle Tempereture Correction 112] (Example 10.5) v1.44 e 108 — Exhaust Pressure Correction (12] (Example 10.5) 8 109 Installed Arrangement (Example 10.6) . 3B 10.10 Pump-Design Curve With Factory and Field Test Daa Shown (Example 10.6} ..... . 73 10.11 Comparison of Test Results With Independent | Cont Condition (Example 10.6) ..... . 7 10:12 Comparison of Test Results ‘sing ‘he Initial (Example 10.6) ciceeeeeeseteseees a 8 Tables 41 Discharge Presse Test Data Wh N = 37 Gxample-45-T--s0--6 13 8.1 Burst Pressures (Example 8.1) vice eeeeteteeeeeee 28 9.1 Table of Dara (Example 10.2). . : 38 9.2 Summary of Data (Example 10.2) 38 9.3 Report of Results (Example 10.2) wecteceeeeeeteeeererees 40) 9.4 Report of Results (Example 10.2). cee 40 95 Summay of Results (Example 10.2) wae) 10.1 Average Values (Example 10.1). 44 102. Summay of Average Velocity Calculation (Example 10.1) 45 10.3 Standard Deviations {Example 10.1) bieeeee 45 104 Standard Deviation of Average Velocity (Example 1€.1) 46 103 Uncertainty Due to Spatial Variation (Example 10.1). te 45 10.6 Uncertainty of Measurement Example 10.1) ceeteeeeeeee AZ 10.7 Uncalitrated Case (Example 10.2) «+... rere 48 40.8. Roltive Sensitivity Coefficients in Example 10.2 (Cacuiated Numerically) vectevetseutes a 109 Absolut: Sensitivity Coefficients in Example 10.2 (Calculated Analstically) . secre 5 10.10 Absolut: Contributions of Systematic and Random Unceriainties of Independent Parameters (Example 10.2)... - rn] 10.11 Summay: Nominal Value; Systematic, Random, an¢ Total Uncertainties in Absolute Terms for in... . ceeeeeees 51 10412 Relative Uncertainty of Measurement (Example 10.2. 31 10.13 Relative Contributions of Systematic and Random Uncersinties of Independent Parameters (Example 10.2)... .-.- « : 52 10.14 Summary: Nominal Value; Systematic, Random, anc Total Uncertainies in Relative Terms forthe Uncaltratd Case Example 10.2)... 32 10.15 Relative Uncertainties of Independent Parameters (Example 102)... 52 10.16 Relative Contributions of Systematic and Random Uncertainties of Indefendent Parameters (Example 10.2)... « . 3 10.17 Summay: Nominal Value; Systematic, Random, anc Total Uncertainties in Relative Terms for he Calibrated Case Example 10.2) ce veeeeeeees 8 10.32A Numerical Development of Sensitivity Coefficients (Example 10.5) 10.328 Numerical Development of Sensitivity Coefficients (Examole 10.5) 10.33 10.34 10.35 10.36 10.37 10.38 10.39 10.40 Results of Uncertainty Analysis (Example 103) ......... 4 Input Systermtic and Random Uncertainties (Example 16.4) | Systematic and Random Uncertainties (Example 10.4). Result Sytonatic and Random Uncertaines (ample 124 Specified Conditions Example 10.5)....... Summary of Test Data (Example 10.5). . Throttle Pressure Elemental Uncertainties (Example 10.5) ‘Throttle Tererature Elemental Uncertainties (Example 10.5)... Exhaust Pressure Elemental Uncertainties (Example 10.5).......... Condensate flow Elemental Uncertainties (Example 10.5.1... 0... Generator Output Elemental Uncertainties (Example 10.5) Generator Hydrogen Pressure Elemental Uncertainties (Example 10.5) Cycle Leakage/Storage Elemental Uncerainties (&xample 10.5), Measuremen: Uncertainty Summary (Example 10.5) : Propagation of Measurement Uncertainties (Example 10.5) . Uncertainty of Result (Example 10.5) Pump Desigr Curve Data With T, = 20°C ample TOO) ess Test Results Cxample 10.6). : : Test Results (xample 10.6). : Uncertainty Fropagation for Comparison With Independent Control (Example 10.6). Uncertainty Fropagation for Comparative Uncertainty (Example 10.6) Sensitivity Ceefficent Estimates for Comparative Analysis (Example 10.6)... : . ‘TEST UNCERTAINTY ASME PTC 19.1-1998 SECTION 1 — INTRODUCTION 1.1 HARMONIZATION WITH INTERNATIONAL STANDARDS ‘This Supplement has been substantially rewritton so that itis easier to use. In addiion, this Supplement is now in harmony with the International Organiza- tion for Standardization (ISO) “Guide to the Expres- sion of Uncertainty in Measurement” (hereafter re- ferred to as the “ISO Guide") (1. For the practicing engineer, this harmonization means the elimination cof such commonly misundersood terms as bias, precision, bias limit, and precision index. in addition, careful attention js paid to discriminating between errors and the effects of errors This Supplement treats recognizable terms such as bias and precision as follows The term bias limit, meaning the estimated limits of the error caused by bias error sources, Is now described by the term systematic uncertainty. The term precision (or preci sion index), meaning the estimate of the magnitude of the error caused by precision error sources, is now described by the term random uncertainty. The term for precision index has keen changed to the standard deviation of the men. While many will prefer to use the terms bias and precision, using them in strictly unambiguous contexs takes considerable care. This Supplement recommends the terms sy5- tematic uncertainty and random uncertainty as they are less ambiguots. In addition, this Supplement while in harmony with the ISO Guide in technology, differs in applica- tion. This Supplement emphasizes the effects of ncerainties rather than the source of information about them. The ISO Guide chssifies uncertainties by the source of information about them, i.e, are there data to calculate a standerd deviation or rot? If thete are, the ISO Guide calf those uncertainties “Type A," if not, they are called “Type 6.” This Supplement permits that kird of classification through the vehicle of an optional subscript on each uncertainty ‘erm, “A’ or “B.” However, the ommended and mest commonly used classifica tion system for uncertainties in this Supplement is by their effect. That Is, if an uncertainty source causes scatter in test results, itis a random uncer- tainty source and has been caused by random errors. If net, it is a sysematic uncertainty and has been caused by systeratic errors. The actual classification of each uncertainty source ic left to the analyst. Thie Supplement merely racom- mends grouping them as random or systematic subscripts to idenify the origin of information about their magnitudes, if desirable. ‘The harmonizaion of this Supplement with the ISO Guide has been achieved by recommending @ simplified Usss uncertainty model which combines the technologies in the ISO Guide and in.the earlier version of this Supplement [2]. With this model, detailed in Sections 6 and 7, several simplifying assumptions have been made, namely: (a)-Allsystomate uncertainty sources are assumed normally distributed and are estimated as 20 for 95% coverage If this assumption is not adequate for analy- sis, more complex methods are handled in Appendix 8. Although this assumption is not statistically rigor- us, it has been feund to work well in many applica- tions. (6) All random uncertainty sources are estimated as 25g, which is 2 95% confidence estimate of the effect on the avenge of a particular random uncer- tainty source. (©) These uncerainty estimates are grouped as sys- tematic or randon and roct-sum-squared to obtain the systematicandrandom uncertainties ofa measure- ment, B and 25, for large samples. (d) These are then rootsum-squared to obtain a 95% confidence uncertainty as follows: Us = (BP (2591"* This simplified uncertainty model is in harmony with the ISO Guide, which can be shown by rear- ranging it as Uss = 2108/2) + (591 Here the multiplier 2 is really Student’s ts which is taken as 2 for more than 30 degrees of freedom ASME PTC 19.1-1998 If fewer degrees of freedom are available, the meth: ods for hancling that situation are in Appendix B. This simple approach to uncertainty analysis will meet the needs of most users; details needed for more: complex analyses are presented in the Appendices. Harmonizetion with the ISO Guide has thus been achieved by using an uncertainty model already developed and usod in (2) while preserving historical groupings of elemental uncertainties, both systematic. (lormerly bias) and random (formerly precision). 1.2 APPLICATIONS This Supplement is intended to serve as reference to the various other ASME Instruments and Apparatus. Supplements (PTC 19 Series) and to the ASIAE Per- formance Test Codes in ‘general. TEST UNCERTAINTY fi TEST UNCERTAINTY ASME PTC 19.1-1990 SECTION 2 — OBJECT AND SCOPE 2.4 OBJECT The object of this Supplement is to define, de- scribe, and illustrate the various terms and methods used to provide meaningful esimates of measure- ment uncertainty and effects of individual measure- ment uncertainties on test results. Measurement uncertainty analysis is useful be- cause it {a) identifies dominant sources of error, theit effects on a test result, and estimates of their limits, uncer- tainties; (b) facilitates, communication regarding measure- ment results; (©) facilitates the choice of eppropriate and cost, effective measurement devices and procedures; (d) reduces the risk of makng erroneous deci- sions; and (e) demonstrates compliance with agreements. 2.2 SCOPE The scope of this Supplement is to specify proce- dures for evaluaton of uncertainties in individual test measurements arising from both random errors and systematic errors, and for the propagation of random and systematic uncertainties into the uncer- tainty of a test result. The various statistical terms involved are defined in the Nomenclature (para 3.1) and Glossary (para. 3.2) The end result o' a measurement uncertainty analy- sis is to provide aumerical estimates of systematic uncertainties, random uncertainties, ard the combi nation of these into a total uncertainty with an approximate confidence level. TEST UNCERTAINTY ASME PIC 191-1998 SECTION 3 — DEFINITIONS AND DESCRIPTION OF TERMS 3.1 NOMENCLATURE = the systematic uncertairty component of a parameter = (3) ‘bs the systematic uncertainty associated with an elemental error source Bg= the systematic uncertainty component of a result 5 y= the Covariance ofthe sysiematic uncer- tainties in x and y B*, B= the upper and lower values of a non- __ Symmetrical systematic uncertainty ‘C= the number o° coefficients in a re- gression ‘Mex the number of redundant instruments; also the number of samples N= the number of measurements or sam- ple points or observations available (the sample size) P= an independent parameter Re the result r= the sample conelation coefficient Se= the standard deviation of a result ‘S.= the standard deviation of a data sam- ple; an estimate of the standard devia- tion of the population, oy where Xe is the kth valve of the vari= able and the degrees of freedom is N-1 2 unbiased estimate of the popula- tion variance Sy= the estimate of the standard deviation of the mean of N measurements Se= Se SEN SEE= the standard error of estimate of a feist squares regression or curve fit. Yee af@ values of the regression cor. responding, to the independent vari- ale, Yims are values Comesponding tothe N data points, C represents the number of Coefficients of the re- grssion tthe Students t value at 2 specified confidence level with » degrees of freedom, ie, 5, {5e= product of Se and the appropriate t statistic forms the random component of total uncertainty U= the total uncertainty Us, U-= the upper and lower values of the nensymmetrical total uncertainty X= an individual observation in a data sample the sample mean; the average of a sel of N individual observ yeit Rei dA A= the (unknown) true systematic error; the fixed or constant component of & = the (unknown) tctal error; the differ- cenze between a measurement ard the true value = the (unknown) true random error; the rardom component of 8 ASME PTC 19.1-1998 the absolute sexsitivity; the change the calculated result’ ¥ due to an incremental change in a parameter P ay oP 6’= the relative senstivity of a result “-RA)-10) y= the (unknown) tue average of the population v= the number of degrees of freedom = the unknown) trie standard deviation of the populatior v= the (unknown) tue variance of the population 3.2 GLOSSARY accuracy:the closeness of agreenentbetweena mea- sured value and the true value. calibration: the process of comparing the response of an instrument to 2 standard instrument over some ‘measurement range. calibration hiesarcty: the chain ef calibrations which ks or traces a measuring instument to a primary confidence level: the prob: falls within the specified lints, degrees of freedorn (v): a sample of N values is said to have N degrees of freedom, ind the first statistic calculated from it also is said to have N degrees of freedom. The dagress of freodomare reduced by one for each previously calculated statistic used to calcu- late a new statistic. ty that the true value ‘elemental error: the systematic or tandom error associ ated with a single error source. elemental uncertainty: the systerratic or random un- certainty associated with a single uncertainty source. estimate: a value calculated from a sample which is used to estimate the true value of ax entire population. For example, the sample standarddeviation, Sy is an ‘estimate of the population standard deviation, a. influence coefficient: see sensitivty. TEST UNCERTAINTY ‘mean (the arithmetic average of N readings. ‘measurement errer (6) the true, unknown difference between the measured value and the true value. parameter: a measured quantity, such as temperature, Dressure, or sitess, used in deriving a result. The defi nition dilfers from that used in statistics. population: the se of al possible measurements of a parameter, random error (@): sametimes called precision; thetrue random error which characterizes a member of a set of measurements, ¢ varies ina random, Gaussian- Normal manner, fom measurement to measurement, resul (Ra value calculated ftom a number of param eters. sample size (N): the number of individual measure- ments in a sample or used to calculate a mean. sensitivity the rato ofthe change in a result to a unit change in a parameter population standard deviation (a): & value which ‘uaniifies the dispersion of a population. standard error of esimate (SEQ: the measure of disper- sion of the dependent variable about a least squares regression or curve satistic:any numercal quantity derived from the sam- ple deta. X and Swe satis, Student's ta value used herein to estimate the uncer- tainty for @ given confidence level, obtzined from ta- bles entered with cogroes of freedom at the chosen confidence level. systematic error (B} sometimes called bias; the tue systematic or fixed error which characterizes every ‘member of any set cf measurements from the popula. tion. The constant component of the total measure rent error, & systematic uncertainty (B): the 95% confidence level estimate of the limis of a true systematic error, 8, often determined by judgment. traceability: the ability to trace the calibration of a measuring device through a chain of calibrations 10 a primary standard. Traceability does not guarantee accuracy, it only irdicates documented calibration hierarchy. true value: the actual value of the parsmeter being measured, TEST UNCERTAINTY ASME PTC 19.1-1998, urcertainty(U): + Uis the inteval about the mezsure- ‘ment or result that Contains the true value for a given confidence level. TEST UNCERTAINTY ASME PTC 19.1-1998 SECTION 4 — FUNDAMENTAL CONCEPTS 4.1 ASSUMPTIONS The assumptions inherent in measurement uncer- tainty analysis include the following: (2) the test process is defined; (b) the measurement syitem and test process are convolled; {c) all appropriate calisration corrections have beeen applied; (d) the test objectives are specified; and (e) the instrument packaze and data reduction pro- cedures are defined. For total uncertainty, 95% confidence levels have been used throughout this document in accordance with accepted practice. Other confidence levels may be used, if required. 4.2. MEASUREMENT ERROR Every measurement has error which results in a difference between the measured value X and the true value. The difference between the measured value and the true value is the total measurement ‘error, 8: Since the true value is unknown, the tetal error in a measurement cannot be known and there- fore only its limits can be estimated. The total measurement erro" consists of two components: 5/5- tematic error and random error (see Fig. 4.1). ACCU tate measurement requires minimizing both system- atic and random errors (see Fig. 4.2) 4.2.1 Random Error. Random arror, «isthe portion of the total measurement enor that varies in repeated ‘measurements of the true value. The total random error in a measurement ié usually the sum of the contributions of several 2lementai random error sources. Elemental random error sources include those which are known and are controlled, those which are negligible and are ignored, and those which are unknown and whose limits must be estimated, Elemental randem errors may arise from nonrepeatability in the measurement system, envi ronmental conditions, daa reduction techniques, and measurement method. 4.2.2 °Systematic Error, Systematic error, 8, is the portion of the total mezsurement error that remains constant in repeated measurements of the true value. The total systenatic error in a measurement is usually the sum of the contributions of several elemental systematic errors. Elemental systematic errors include those which are known and can be calibrated out, those which are negligible and are ignored, and those which are unknown and whose limits must be estimated. Elemental systematic ertors may arise from imperfect calibration corrections, data acqu tion systems, data reduction techniques, etc. The nonnegligible systematic errors that remain after cali- bration contrisute to the systematic uncertainty of the measurerrent. 4.3 UNCERTAINTY Since every observation or measurement i¢ the combination cf the true value of the parameter plus the total measurement error, there Is an inherent Uncertainty in the use of measurements to represent the tue value. The total uncertainty in a measure- mentis the conbination of uncertainty cue to systemn= atic error and uncertainty due to random error. 4.3.1 Uncertainty Due to Random Error. Since random error introduces variation or scatter in re- peated measurements of a parameter, the uncertainty. due to randorr errox may be estimated by inspection af the measurement scatter. The population standard deviation, o; § a measure of the scatter about the true population mean, 4, caused by random ertor. For anormal distribution (see Fig. 4.3), the interval # + 20 will include approximately 95% of the population. Ax estimate of the population standard eviation is the standard deviation of a data sample, Sy, which is determined by (any where N is the number of measurements made and ASME PTC 19.1-1998 Trot value Systematic etor(B) Total orror (54) Fraqueney of Occurrence TEST UNCERTAINTY Measurement population mosn(u) ‘population istibution ror le Neasured Valves FIG. 4.1 ILLUSTRATION OF MEASUREMENT ERRORS X is the mean of the individual measurements % given by 2% 42 zI- Averaging multiple measurements and using the mean in lieu of any individuel measurement reduces the random uncertainty. This indicates that the mean of several samples is a beter measure of the true population mean than one simple. The estimate of the standard deviation of the sample mean, Sz, is given by = = a3) 43.2 Uncertainty Due to Systematic Error. Since systematic error is constant for repeated measure- ments of a parameter (via the same technique and constant true value) and sincesystematic ertor cannot bbe quantified absolutely, the uncerainty due to systematic error must be estimated. The systematic uncertainty B is an estimate of the systematic error evaluated at 95% confidence, The systematc uncertainty estimate is based cn ‘engineering judgment and analysis of elemental sys- tematic errors. Ivery effort must be made to identify and account for all elemental systematic errors, Special tes's or data can -previde information on the systematic uncertainty in a measurement. The following examples are given in order of preference: (@) interlaboritory or interfacility tests; ©) compariscn of standards with instruments in the actual test environment; (©) compariscn of independent measurements that depend on different principles or thathave been made by independent calibrated instruments; for example, in a ges turbine test, airflow can be measured with an orifice or a bell mouth nozzle, or computed from ‘compressor specd-flow rig data, turbine flow parame- ters, or jet rozze calibrations; (Q) special calibrations which perturb a known cause of systematic error through its complete range to determine the range of systematic uncertainty; TEST UNCERTAINTY {a1 NegighalesytonatisEror ‘Small andor Error ‘rue value and Frequency ~ ‘obser Valucs, (2 Negligible systematic tor (lrge Meador Error ASME PTC 19.1-1998 Population Frequency Observed Values lbp Lege’ Syetomate Feror ‘Smell Random Error Population Frequercy ‘Observed Vales (Large Systemate Error Large Random Eror FIG. 4.2 MEASUREMENT ERROR COMPONENTS: (©) reports from instrument manufacturers and other references; (9 model ofthe process whichis known to generate the eror; () engineering judgment or experience. In most cases, systematic uncertainties are equally likely to be plus or minus about the measurement. That is, typically itis not known if systematic errors are postive or negative ard the systematic uncer- tainty reflects this as +B. Sometimes, however, the physics of the measurement system provides knowl- edge of the sign. For a discussion of nonsymmetic systematic uncertainties, see para. 8. In some cases, systemati: uncertainties may arise from the same source and are therefore correlated. See para. 8.1 for a detailed discussion, 4.3.3. Total Uncertainty. As rrentioned previously, the tial uncertainty in a measurement is the combi ration of uncertainty due to systematic error and uncertainty due to random error. For simplicity of presentation, 1 single value is often preferred to express a reasonable limit for the total measurement error. Since the true value is unknown, it is conve nient to define the limit referenced to the measured value. Uncerttinty, U, is defined as an interval, about the meisured value, that has a preassigned probability of containing the true value. Therefore, the interval (ay represents a band about the measurement mean, X, within which the true value is expected to lie at‘a given level of confidence (soe Fig. 6.2). ‘The uncertainty interval for 95% confidence, Uss, 1s calculated »y Mae ASME PIC 191-1998 Fraquetey of Measurement TEST UNCERTAINTY, Messurement population Sitrution 85% of Meseured Veluoe FIG. 43 DISTRIBUTION OF MEASURED VALUES (NORMAL DISTRIBUTION) 4.5) where Bis the 95% confidenceestimate of systematic uncertainty and Sy is the standard deviation of the meen. Several assumptions are required for this simple equation. These include: (a) @ 95% confidence interval is appropriate; () the degrees of freedom issufficiently large such that t = 2 can be used for the multiplier; and (cl the systematic uncertainty component is trom 2 distribution of possible values that is normally dis- tributed. Treatment of uncertainty Intervals with alternative confidence levels and alternate uncertainty equations Is addressed in Appendix B. 4.4 PRETEST ANALYSIS AND CORRECTIVE ACTION It is recommended than an uncertainty analysis be done before a test or experiment. This procedure allows corrective action to be taken prior to the test 10 reduce uncertainties, The pretest uncertainty analysis is based on data and information that exist before the test, such as calibration histories, previous tests with similar instrumentation, prior measurement uncertainty anal- ysis, expert opinions, and, if necessary, special tests. With complex tests, there offen are alteinatives to evaluate prior to te test such as different test designs, instrumentation atrangements, alternative calculation procedures, and so on. Pretest analysis can identify the most accurate test method Corrective action resulting from this pretest analy- sis may include: (2) improvements toall calibrations of instrumenta- tion if the eystomctic uncertainties are unacceptable, (6) selection ofa different measurement method to obtain the parameter of interest; and (c) repeated testing, or increased sample sizes, or both if the random uncertainties are unacceptable. Cost and time may dictate the choice of the corrective aciion. If the corrective action identified cannot be taken, there may be a high risk that test objectives will not be met because of the large uncertainty internal, end cancellation of the test should be a consderation, 4.5 POSTTEST ANALYSIS — PHILOSOPHY 4.5.1. Reported Results. The result from a test can normally be presented in terms of an average, X, ‘with an uncertain estimate, This can be written aé TEST UNCERTAINTY ASME PTC 19.1-1998, TABLE 4.1 DISCHARGE PRESSURE TEST DATA WITH N = 37 (EXAMPLE 45-1) Equation ia pai Systematic uncerainy, B 410 060 Kander uncerany, 25 comm oso Mean, X 7585 119.00 Sample stardard deviaton, S, 16.77 243 Standard deviation ofthe mean, 5 2.76 040 6.88 00 Total uncerainy, U X & U (95% confidence) which implies thet, under the con curred during a single test or 2 series of tests, the average is X, and an estimate of the possible error associated with the average X at 95% confidence is Uc If the test was to deternine the supply pressure and efficiency for a particular pump, the test report might read: Pump number 6 was tested on 23. July 1993 using company 2rocedure XYZ. For the specified flow of 0. 57 m/s (2500 gpm) at 20°C, the discharge pressure was 758 27 kPa or 110 * 1 pti (95% confidence), The estimated efficiency of the pump during the testing was 0.812 = 0.009 (95% confi- dence. This test was conducted at a single flow rate and had two results: (1) discharge pressure and (2) efiiciency. It musi be assumed thet the specified procedure presents an accestable range of flow and temperature for the testing, The statement above implies that the average discrarge pressure during the test was 756 kPa, and the measurement uncertainty associated with this pressire measurement was 7 KPa, The pressure would be considered a measure- ment, and the estimate of tre uncertainty would be generated as presented in Section 6. The statement does not provide information on whether the dis- charge pressure varied apcreciably during the test or if the discharge pressure was acceptable. 4.5.2 Posttest Reviews, A strict mathematical re view of test data is normally not adequate, Not only is it necessary ta algebraically demonstrate conformance to a requirenent (e.g, the measure- ment uncertainty was below that required by the test program), the basic aisumptions of para. 4.1 3 must be checked. Plotting of the data such as shown in Fig. 44 is one method to evaluate compliance with the assumptions in para. 4.1. Example 4.5-1 discusses some of the basic checks which could be made with any data set Example 4.5-1; Table 4.1 presents the test data for the pump that was described earlicr. For this, test, the pump was operated ‘or 3 hr and data were collected every 5 min. The systematic uncertainty, B, in Table 41 has been estimated based on the calibration data and previous test experience. The remaining quantiles have been computed from the raw test data The test data presented in Fig. 4.4 are typical of a set of test daa; while the measured reading varies with time, there is no observable pattern of behavior. If the data were clumped such that most of the early data lay above the mean while the later data lay below the mean, this would suggest that the test arrangement was not at steady-state. This would indicate that the process under test was not con- trolled {assumption (b), para. 4.1]. Also, if the data appeared cyclc with time, the process under test may not be controlled. A final check for any data set is performed by ‘comparison to a known condition. Do the test data meet expectations? For performance testing, the be- havior of different classes of equipment is well established. ‘Performance which deviates from an established nam indicates a need for additional evaluation Examples of why deviations might exist are: (2) the testeé device does not perform as expected; (b) the test facility has changed; and (6) an uncortrolled test variable exists. If the pretest and posttest uncertainties do not compare withix expectations, it is possible that the assumptions of para. 4.1 have been violated. For a further discussion of these issues, a text on experi- mental metho¢s should be consulted, ASME PTC 19.1~1998 TEST UNCERTAINTY 00: ‘Semple mean: 3 -Corfidence interval fr sainples Pressure, kPa 740 720 ° 80 wo 00 FIG. 4.4 TEST DATA (EXAMPLE 4.5-1) “4 TEST UNCERTAINTY ASME PTC 19.1-1998, SECTION 5 — DEFINING THE MEASUREMENT PROCESS The first step in an uncertainty analysis is to clearly define the measurement process. This simple step, often overlooked, is essential to successfully develop and apply the uncertainty information. Con- sideration must be given to the selection of the appropriate “rue value” cf the measurement and the time interval for classifying errors as systematic or random, 5.1 SELECTION OF THE APPROPRIATE “TRUE VALUE” Depending on the user's 2erspective, several “true values” may exist simultaneously in @ measurement process. For example, when anelyzing a thermocou- ple measurement in a gas stream, several starting point, oF “true valuoe,” car be solected. The starting point for the analysis coud begin with the “tue value” defined as the metal temperature of the thermocouple junction, the gas stagnation tempera- ture at the thermocouple point location if the probe ‘was not installed, or the mass flow weighted average of the gas temperature at the plane of the instrumen- tation. Any of these possitle “rue values” may be appropriate. The selection of the ‘wue value” for the uncertainty analysis must be consistent with the goal of the measurement 1]. 3.2. TIME INTERVAL Since measurement error sources are classified by their effect, the time interval for the measurement process must also be cleary stated. Errors that may be fixed over a short time period may be variable ‘over a longer time period. ‘or examole, the calibra- tion corrections, which may be assumed to be fixed over the life of the calibration interval, can be considered variable if the process consists of a time interval encompassing several recalibrations. The fimo interval must be clearly specified to classify an error as either systematic or rardom. ‘The time incerval used to classify an error as either systematic or random is ustally, but not always, the same as the test duration. When comparing results between various laboratories, it might be appropriate to classify a quantifiable evor source that has a time period much larger than the duration of the test as random even though it was constant for the duration of the tect 5.3 COMPARATIVE VERSUS ABSOLUTE TESTING ‘The objective of a comparative test (also known a a back-to-ack test is to determine, with the smallest measurement uncertainty possible, the net effect of @ design change. The first test is run with the standard cr baseline configuration. The second test is then run in the same ‘acility with the design change ard hopefully with instruments, setups, and calibrations identical to those used in the first ast. The difference between the results of these tests is an indication of the effect of the design change. Because of the effects of correlated uncertainties (see Section 8) the total uncertainty of the difference will usually te less than the uncertainty of each, separate test, An example of back-to-back uncer- tainty analysis is shown in para. 8.1. 5.4 MEASUREMENT UNCERTAINTY SOURCES Sources introducing uricertainty in a measurement process can te divided arbitrarily into categories, eB. (a) calibration uncertainty (b) data accuisition uncertainty (©) data reduction uncertainty (A) uncenainty due to methods (2) others While often helpful, dividing uncertainty sources by categories is not necessary for a correct uncer- tainty analysis To decide ita given elemental uncerainty source contributes to the systematic or random component of the total uncertainty, the effect of the error on the measurement must be determined. If the nature cf an elemertal error is fited over the duration of the definec measurement process, then a 95% ASME FTC 19.1~1990 TEST UNCERTAINTY <— — oS working senders | ws) | ws 4 iT 1 cn | [oan Meseurenent im ee Confiderice estimate of the error is classified as 2 systematic uncertainty. IF the error source tends to ‘ause scatter in repeated observations of the defined measuterfient process, then the source contributes to the random uncertainty. The methods used to estimate uncertainty, as presented in this document, may be used to estimate exch of the five above uncertainties. Secondarily, it may also be beneficial to character- Ize the elemental uncertainty sources by the manner in which they are estimated. “hose which are evalu. ated by statistical methods mzy be classified as Type ‘A while those which are evauated by other means may be classified 2s Type 3. Depending on the selection of the defined meas.rement process, there ‘may be no simple correspondence between random or systematic and Type A or Type B. 5.4.1 Calibration Uncertainty. Each measurement instrument may introduce random and systematic Uncertainties. The main purpose of the calibration proces isto eliminate large, Known systematic ears and thus reduce the meastiement uncertainty to some “acceptable” level. Heving decided on the “acceptable” level, the calibration process achieves that goal by exchanging the large systematic uncer- tainty of an uncalibrated or poorly calibrated instru- ment for the smaller combination of systematic un- 16 EIG..5.1, GENERIC MEASUREMENT CALIBRATION HIERARCHY certainties of the standard instrument and the random: uncertainties of the comparison. Calibrations re also used to provide traceability to known refereace standards or physical constants, ‘or both. Requirements of military and commercial ‘contracts have lad to the establishment of extensive hierarchies of stindards laboratories. In some coun- tries, a national standards labcratory is at the apex of these hierarcties, providing the ukimate reference for every standads laboratory. Each additional level in the calibration hierarchy adds uncertainty in the measurement process (see Fig, 5.1) 5.4.2 Data Accuisition Unceriainty. Uncertainty in data acquisition systems can arise from errors. in the signal condtioning, the sensors, the recording dovices, atc. The bast method te minimize the effects of many of these uncertainty sources is to perform overall system calibrations. By comparing. known input values with their measured resuits, estimates of the data acquisition system uncertainty can be oblained. However, it is not always possible to do this. In these cates, it is necessary to evaluate each of the elemental uncertainties and to combine them to predict the overall uncertainty. 5.4.3 Data Retuction Uncertainty. Computations ‘on raw data are done to produce output (data) engineering units. Typical uncertainty sources inthis TEST UNCERTAINTY category stem from curve its and computational resolution. With the recent advances in computer systems, the computational resolution uncertainty sources are often negligible; xowever, curve fit error uncertainty can be significant. 5.4.4 Uncertainty Due to Methods. Uncertainties due to methods are defined as those additional uncertainty sources that orginaie from the tech- riiques or methods inherent in the measurement process. These uncertainty sources, beyond those centained in calibration, dat acquisition, and data reduction, may significantly affect the uncertainty ofthe final resutts. Some corrmon examples include: Ww ASME PTC 191-1998 {@) uncertainy in the assumptionsor constants con- tained in the calculation routines; (b) uncetainy due to intrusive disturbance effects caused by instaled instrumentation; (©) spatial or profile uncertainty in the conversion from discrete pint measurements to station averages; (@) environmental effects on probes such 25 con- duction, convedtion, and radiation; ‘@) uncerainy due to instability, nonrepeatability, and hysteresis o the test process. These additional uncerainties may be of either a systematic or random nature depending on their effect on the measurement. ae TEST UNCERTAINTY ASME PTC 19:1~1998 SECTION 6 — UNCERTAINTY OF A MEASUREMENT 6.1 DETERMINATION OF SAMPLE STANDARD DEVIATION AND DEGREES OF FREEDOM 6:4 General Case. For T that is determined as the average of N' readings, the appropriate sample standard deviation of the mean (53) is given by Eq, (43). In a sample, the degrees cf freedom is the sample size (N. When a statistic is calculated from the sample, the degrees of freedom associated with the statistic is reduced by one for every estimated parameter used in calculating the statistic. For exam- ple, fiom 2 sample of size N, X is calculated by Eq, (42) and has N degrees >f freedom. The sample standard deviation (S,) and the sample standard doviation of the mean (Sz) ere calculated from Eqs. (4.1) and (4.3), respectively, and each has WV —"1 degrees of freesiom (2) veN-t en because X (based on the sime sample of data) is used in the calculation of both standard deviations. 6.1.2 Using Previous Values With Small Samples. In some test situations, the measurement of a variable may be only 2 single reading or a single average ‘of readings taken over a short time frame, as with a computer-based data acquisition system. In this fatter case, the time frame ever which the readings are taken may be onthe der of milliseconds or less while the random variatons in the process may be on the order of seconds, cr minutes, or even days. This “averaged” reading should then be handled in the same manner as a singe roading. Information about the possible variations in a single reading measurement must be obtained from previous readings of the variable taken over the time frame and conditions which cover the variations in the variable, For example, taking multiple samples of data as a function of “ime while holding. all ‘other conditions constant would identify the random variation associated with tre measurement system and the unsteadiness of the test condition. If the sample standard deviation of the variable being » measured is alio expected to be representative of other possible random variations in the measure ment, eg, repeatability of test conditions, variation in test configuretion, etc, then these additional error sources will hve to be varied while the multiple deta samples ere taken to determine the standard deviation ‘Another situation where previous readings of a variabie would be useful is when a small sample size (N) is used to calculate the mean value (X) of 2 measuremen. if a much larger set of previous readings fer the same test conditions is available, thon it could be used to calculate a more appropriata standard deviaion for the measurement [4]. Typi- cally, these larger data sets are taken in the early phases of an experimental program. Once the ran- dom variation of the test variables is understood, then this information can be used to sueamline the test procedure oy reducing the number of readings taken in the later measurements. When pievicus readings (X,) are known for the quantity being measured, the sample standard devia- tion for the variable can be calculated from these Np previous rexdings as Np= 1 at (2) where The appropriate standard deviation of the mean for the current measurement (X) is then i AN a (6) where N is the number of current readings averaged to determine X The number of degrees of freedom for this standard deviation of the mean Sy is ASME PTC 19.1198 Pressure anedvcer TEST UNCERTAINTY Exctotion valage Signai conditioning FIG. 6.1 PRESSURE DATA ACQUISITION SYSTEM Y= NeW1 (65) The case where a messurement is only a single reading is handled above wih N = 1 6.1.3 Using Elemental Random Error Sources. An- other mathod of dete-mining the standard devistion of the mean for a measurement is from information ‘about the elemental random error sources in the entire measurement process. As an example, con- the pressure data acquistion system shown in 6.1, Elemental random errors for this system can arise from several sources such as the excitation voltage, the signal conditioning process, the rec- ording device, the pressure transducer, probe errors, and environmental effects, The standard deviation for tre measurement mean is the rootsum-square of the elemental standard deviations of the mean from all sources divided by the square root of the number of current readings (N) averaged to determine X 5 = Jy [3 co] 6.6) where K is the total number of random error (or uncertainty) sources. Each of the elemental standard deviations of the mean (S5); is calculated using the methods described in para, 6.1.1 of 6.1.2 deperding upon which is appropriate. If in each of the N readings of the 20 measurement X the output of an elemental compo- nent ie averagoc AY, times to abtain X,, then the method in para. 6.1.1 would be used. If instead previous informetion is used to obtain: (S,);, then the method in para. 6.1.2 would apply. ‘The degrees of freedom for the measurement stan- dard deviation cf the mean () is dependent on the information used to determine each of the ele- ‘mental standard deviations of the mean and is caleu- lated as a where », is the aopropriate degrees of freedom for (Sp); and is obtiined from Eq. (6.1) or (6.5) as appropriate, Whe: all error sources have large sam- ple sizes, the calculation of is unnecessary. How- ever, for small sanples, when combining elemental standard deviations of the mean by the root-sum- Square method [see Eq. (6.6)], the degrees of freedom |v) associated with the combined standard deviation is calculated using the Welch-Sstterthwaite formula {5} above Eq. (67)]. TEST UNCERTAINTY Frequency of currence ' x Semple tleane » ASME PTC 1921-1998 ean of emasuremant opulstion CGstribution of sample means Uncertainty teers FIG. 6.2 UNCERTAINTY INTERVAL 6.2 COMBINING ELEMENTAL SYSTEMATIC UNCERTAINTIES, The systematic uncertainty (6) of a measurement wes defined in para. 4.3.2 2 the 95% confidence level estimate of the limits of he true systematic error (B. The systematic uncertainty of the measurement is the rootsum-square of the elemental systematic uncertainties (b) for all sources (de]" Where K is the total number of systematic error sources. For each error source in the measurement, the elemental systematic uncertainty must be esti- mated from the best available information, if any of the elemental systematic urcertainties are nonsym- metrical, then the method given in Section 6.2 should be used to determine the systematic uncertainty of the measurement, (6.8) a ‘There can be many sources of system: jin a measurement, such as the calibration process, instrument systematic errors, transducer errors, and fixed errors of method. Also environmental effects, such as radiation effects in a temperature measure- ment, can cause systematic errors of method. There will usually be some elemental systematic uncertain- ties that will be dominant. Because of the resulting cflect of combining the elemental uncertainties in a rootsum-squere manner, the larger or dominant cones will contro the uncertainty in the measurement however, one should be very careful to identity all sources of fixed error in the measurement. 6.3 TOTAL UNCERTAINTY OF A MEASUREMENT For simplicity of presentation, a single value is often preferred io express the estimate of the error beween the mean value (X) and the true value, ASME PTC 19.1-1998 with a defined confidence. Since the true value is, unknown, itis convenient to define 2 limit referenced ‘0 the measured value. Unceriainty (U) is defined as an interval, about the measured value, that has 4 preassigned probability of containing the tue value, The interval Fu 6.9) represents a band about X within which the tue value is expected to lie with a given level of confi dence (see Fig. 6.2). The uncertainty Interval composed of both the systematic and random uncer- tainty components ‘The general form of the expression for determining the uncertainty of a measurement is the root-sum- square of the systematic and random standard devia- tions of the measurement tmes the appropriate Stu- dent's t value for a specified confidence level Gee Appendix B). Depending on the application, various confidence levels may be appropriate. A Student's is chosen on the basis of the desired level of confidence and degrees 0 fresdom. A ¢ value of 1.96 (usually taken as 2) conesponds to large degrees of freedom and defines an interval with a level of confidence of approximatey 95%. For other confi dence levels see Appendix 8. ‘The systematic uncertainy (5) for a specific mea surement is the estimate of the limits of the fixed error in that measurement. The true systematic error G@ isunknown, but 6 is estimated so that it represents the limits of a 95% con‘idence interval for the possible # values. For the simple uncertainty expres- 2 TEST UNCERTAINT. sion given below, it is assumed that these possib| B values are from a Gaussian distribution and the enough information is known ahout the possibl systematic erors so that the degrees of freedor of the distitution is large ‘Then the systemati. uncertainty (6) can be considered to be two time the standard deviation of the systematic error distri bution. It should be noted that while 8 is a 95% estimate of the distribution of possible bias error in a measurement, the bias error that is present it 2 specific messurement is a fixed single value of 6 ‘The uncertzinty interval for 95% confidence (¢ 2) is calculated by Un = 2 (2) +59 (10 Where Bis the95% confidence estimate of systematic uncertainty 163. (68), Sy is the standard deviation ‘of the mean (Fg. (4.3), (64), of (6.6) as appropriate] ‘and Ups 's the uncertainty interval for 95% confi dence Several assumptions are required for thi equation. ‘These include: (a) a 95% confidence interval is appropriate; (©) the degrees of freedom is sufficiently large suct that (= 2 car be used for the multiplier, (c) the systematic uncertainty component is fron a distribution of possible values that is symmetric normally distibuted, and has large degrees o freedom. Treatment cf uncertainty intervals with alternate confidence levels and alternate uncertainty equation is addressed in Appendix B. imp TEST UNCERTAINTY ASME PTC 19:1~1958 SECTION 7 — UNCERTAINTY OF A RESULT PROPAGATION OF MEASUREMENT UNCERTAINTIES INTO A RESULT Results are not meatured directly; more basic parameters such as tempercture and pressure are measured and the required result is calculated as a function of measurement: of these parameters. Uncertainties in measurements of the parameters are propagated to the result through the functional relationship between the result and the parameters. ‘The effect of the propagation can be approximated by the Taylor series method (see Appendix ©. To obtain the uncertainty cf a derived result, such as turbine efficiency, based cn a series of measure parameters, such as temperaure and pressure, i necessary to express the result (R) in terms of the average values of the independent parareters (P}) that enter into the result. That is, 7A R= APs, Pay. BD ay where the subscript J signifies the total number of parameters involved In &, aid the average values of the independent parameters are obtained as 1 ele 72) where N; is the number of rpeated measurements of P.. N; will be 1 for 2 single reading, Ih some experimental situaions, a set of parame- ters (P)) is measured and 2 single result (R) is calculated. This case is called a single test result In this case, some of the parameters may be sin ‘measurements and others may be means of N, mea- surements. Nj can be different for each P. When, in an experiment, multiple test results can be run at the same condition and each parameter 1s measured during each tes, then these multiple sets of data can be used © determine multiple results, The reported result will usually be the mean of the results (R). This case is called a multiple test result in the discussions of uncertainty given in the following paragraphs. 23 7.2 SENSITIVITY It is convenient to introduce the concept of the sensitivity of @ result to a unit change in each of its parameters, sensitivity Is the ratio of the change in a result to ¢ unit change ina parameter. The ‘coefficient of each parameter is obtained in one of two ways. 7.2.1 Anaiytically. When there is a known mathe- matical relationship between the result (R) and its parametors (P, Pz, ..., F), then the absolute (di- mensional) sensttivity coefficient (@) of the parameter (P) may be obained by partial difere Thus if R= fF), Py, ..., BY, then aR # wa ‘Analogously, therelative (nondimensional sensitivity coefficient (6/) & ay 7.2.2 Numerically, Finite increments in a parame- ter also may be used to evaluate sensitivity using the data reduction calculation procedure: In this case, @ is given by a onde os and 6/ by Fi (ar ale) v6 ASME PTC 19.1-1998 The result is calculated using P, P; to obtain R and then recalculated using (F + AP) to ebtain (R + AR) (6). To best approximate the sensitivity that would be cbtained anaiyticaly, the value of AP, used should be as smell as practical (.., large enough to keep rounc-off errors from influencing the calculations). 7.3 RANDOM STANDARD DEVIATION OF A RESULT 7.3.1 Single Test. The absolute standard deviation of a single test result may 2€ determined from the Propagation equation (see Appendix C) as a-[Ze%] ‘The relative standard deviation of a result is on 7.8) ‘The symbols 4, and are ‘he absolute and relative sensitivity coefficients, respectively, of Eqs. (7.3) or (7.5) and (7.4) or (7.6), and Sp, is the standard deviation of the mean of the measured paremeter average (P), determined according to the methods presented in para. 6.1. 73.2 Multiple Test. When more than one test is conducted with the same instrument package (i.e, repeated tests), the uncertainty of the average test result will be less than that for one ‘est because of the reduction in the random uncertainty of the average. However, systemaiic uncertainty will re- main the same as for a single test. Careful consideration should be given to designing the test series to average as many causes of variation as possible within cost consraints. The test design should be tailored to the specific situation (71. For ‘example, if experience indicates that time-to-time and ig-to-rig variations are significant, a test dosign that averages multiple test results on one rig or for only one day ‘may procuce optimistic random uncertainty estimates compared to testing several rigs, each monitored several times over a period of several days. The list of test variation causes are many and may include the atove plus environmental and test crew variations. Historic data are invaluable 24 TEST UNCERTAINTY for studying hese effects. A statistical technique called analysis of variance (ANOVA) is useful for partitioning the total variance by cause [5]. If the pretest uncertainty analysis (see para. 4.4) identifi unacceptably largo orrer sources, special tests to measure the affects of these sources should be considered. The average result from more than one test is given by 79) where -M signifies the number of tests available. Following Eq, (4.1), the estimate of the standard deviation of the distribution of the results s [z am ‘ & Se |= (7.10) where Se inclides random errors within tests and variation between tests. The degrees of freedom associated with Sp 's determined by yp = M = 1, The samole standard deviation of the mean result from multiple ‘ests is Se so ay This standard ceviation of the mean also has 9m = M — 1 degrees of freedom. 7.4 SYSTEMATIC UNCERTAINTY OF A RESULT In para. 6.3, the assumption was made thet if the possible systematic errors for the measured parame- ters were irom Gaussian distributions, and that if enough information were known about the possible systematic orro's so that the degrees of freedom of the distributiors were large, then the systematic uncertainty for 2ach parameter would be essentially an estimate of two times the standard deviation of ‘the systematic eror distribution. Using these assump- tions, the propagation equation in Appendix C can be used to determine the absolute systematic uncer- tainty of a result as TEST UNCERTAINTY Be (7.12) (3, or]! The relative systematic uncertainty of a result is 7.13) The symbol By, is the systematic uncertainty of the measured parameter (see para. 6.2) 7.5 UNCERTAINTY OF A RESULT The general form of the expression for determining the uncertainty of a result is the root-sum-square of the systematic and random standard deviations of the result times the appropriste Students ¢ value for a specified con‘idence level (see Appendix B). The following simple expression for the uncertainty of a result applies in many cases: 25 fe re 195-100 Ung = (SF ssa)" where Bp is obtained from Eq. (7.12) and Sp is obtained! from either Eq, (7.7) for a single test result or from Eq, (7.11) for a multiple test result. The assumptions required to use Eq. (7.14) are that: (a) 2 95% confidence interval is appropriate; (b) the degrees of freedom is sufficiently large so that t = 2 can be used for the multiplier (see Appendix B for the method to calculate the degrees of freadom of the resul); {c) the systematic uncenainty of the result is from a cistribution ofpossible values that is symmetric and normally distributed with degrees of freedom large enough so that the standard deviation is approximated by Bg/2. The interval within which the true result should lie with 95% confidence is ay) RE Unas as) The special cases of correlated systematic uncer- talnties and noasymmetric systematic uncertainties are covered in paras. 8.1 and 8.2. respectively. ‘Treatment of uncertainty intervals with alternate con- fidence levels end alternate uncerainty equations Is addressed in Appendix B. ‘TEST UNCERTAINTY ASME PIC 19:1-1998 SECTION 8 — ADDITIONAL UNCERTAINTY CONSIDERATIONS 8.1 CORRELATED SYSTEMATIC UNCERTAINTIES, The expressions for the systematic uncertainty of the result in para. 7.4 (Eqs. (712) and (7.13)] assume that the systematic uncertainties of the measured parameters are all independert of each other. There are many situations where systematic errors for some of the parameters in a result are not independent Examples would inchide using the same apparatus to measura different parametors or calibrating different parameters against the same sandard. In these cases, some of the systematic errors zre said to be correlated and these nonindependent errors must be considered in the determination of the systematic uncertainty of the result (6) Consider an example where the result (R) is deter- ied from three parameters (X;, Xp, X3) which have correlated systematic errors. The resuk is calcu: lated as R= FR, Ky Xs) (8.1) and the absolute systematic uncerainty of the result is given as Be = [(0, By? + (0 Be, + (8s BR, #20; B34 #2010, Bra, +2640 Bx,74] " (8.2) ‘The first three terms under the square root in Eq. (6.2) are the same 2s those obtained by using Eq (7.12), and the fast three terms are those that account forthe correlation among the systematic uncertainties in X,, Xo, and Yq. The terms By.z, are the estimates of the covariances of the srstematic errors in X, and X; (see Appendix B). These terms must be included when systematic uncertainties for separate parameters, X; and X,, are ‘rom the same source ‘making them correlated. The units of the correlation 27 terms, Byjry we the product of the units of Ryand Xe “The covatiance terms in Eq, (8.2) must be properly interpreted. Eact By;z, term represents the sum of the products of the’ portions of Bz, and By, that arise from the sane source and are therefore perfectly Correlated (9]. For instance, if elemental systematic Uncertainties 1 and 2 for parameters 2 and 3 were from a common source, then Br,7, would be deter mined as Bex, = (bx) (beh + (brh lbmde (8.3) ‘ihe example in Eq. (8.2) can be extended to any number of parameters by including a term 26 cBe;z, for each pair of parameters with corre- lated systematic uncertainties Example 8.1: The use of back-to-back tests is an excellent method to reduce the systematic uncer- tainty when comparing two or more designs. This method is a special case of correlated systematic uncertainties. Consider a burst test for an improved container desige. Tha imorovement in the dosign can be expressed as the fraction >, where P, is the burst pressure of the new design and Pp is the burst pressure of the original or base aoe Table 8.1 prevides burst tests for two different programs. in the first test program, different pressure transducers were used in the tests on the two designs, There were no correlated systematic. uncertainties common betwe:n these two transducers. In. the second program, the same pressure transducer was used for both teits; therefore, the systematic uncer. tainty was the seme and was correlated for the two test measuremerts. ‘ASME PIC 19.1-1998 TABLE 6.1 BURST PRESSURES ‘EXAMPLE 8.1) ts ’ Base Design Improved Design By 10 Pa 108 Ps 108 Program 7 ‘Meter =1 400 02 Meter #2 520 02 Program? ter #3 420 547 os Program 1 (no correlated systematic uncertainties): 520 1.30 40.0 a= SR — 0.0325 1108 Pay" Py (0.0250 (10° Pa)- Bg? = ((-0.0325)(108 Pa) (0.2108 Pa)? + {(0.025)(10° Pay? (0.2108 Pa)? B= 0.0082 Program 2 (correlated systematic uncertainties): pu dhl 30 42.0 = ~0.0310 (108 Paj~ @,= 0.0238 {10° Pay Bq? = ((-0.0310)(10 Pa)- (0.5108 Pay]? + {(0.0238)(108 Pay? (0.5)(108 Pall? + 2-0.0310)(108 Par" (0.0236)(108 Pa)! (0.5)(108 Payi0.5410® Pa) Be= 0.0036 This example demonstates the strength of the back-to-back testing technique using the same instru- mentation. Even though the frescure taneducer in program 2 had a systematic uncertainty of more than twice those of the transduce’s in program 1, the systematic uncertainty of the result for program 2. was less than half of that for program 1. Example 0.2 (adapted fron [8): Consider the piping arrangement shown telow with the four flowmeter: 28 TEST UNCERTAINTY From consenation of mass, a balance check would yield ze mm - ms = 0 IF the errors in the flow rate measurements are predominantly systematic, then for the balance check to be satisfied lls 8, Consider the case where the dominant systematic errors are from the calibration standard and from the calibration curve fit. The calibration standard systematic uncerainty for each flowmeter is a per- cent of reading and is 3 llbm/hr for the three smell ‘meters and =9 Itrv/hr for the large meter. The curve fit systematic uncertainty for each meter is =¢1 Ibrv hr. Note that for this example the derivatives for Eq. (8.2) are Om = bom = and Ong = 1 Case 1 — Each flowmeter calibrated against a different, standard. In this case, all of the systematic uncertainties are uncorrelated with the systematic uncertainty for the three small flowneters determined as {oy +98)" 3.16 lomvar By, (1 = 1.2.3 and the systematic uncertainty for the large flowmeter calculated as Bg = (OP + 01718 9.06 tome Using Eq, (7.12, the systematic uncertainty for zis "st UNCERTAINTY Bz = (6,8 n,)? + (%Bmm)? + (08mg) (OBing)?) or 8 = [(Bm)? + (Byp)* + (Bay)? + (Boy)? T 0.6 Ieev/hr Case 2 — Flowmeters 1, 2, and 3 calibrated against the same standard and flowmeter 4 cali brated against a different standard. In this case, the systematic uncertainties are Bry = Bg = Bmy = 3:16 lb/hr Bigg = 9.06 lbmmihr and Brum, = Bmymy = Bmamg = (9 Ibi) 9 lore) Using Eq. (6.2) for four measurements with three of them having correlated systematic uncertainties, the systematic uneartainty far 7 becomes Bz = [(8, Bens)? + (8: Bm)? + (88m)? + (84 8g)? +.20,08m ny + 201%Bmym + 202938 mgm) or B, = {(Bm)? + (Bm)? * (Bny)* + (Bag) + Bey + 2B + 2Bmng] B, = 12.9 Ibevhr Note that in this case ihe ‘signs for all of the correlated terms are positive because all of the derivatives of z, with respect 10 m, ma, and ms, are negative. If flawmeters 1, 2, and 3 are calibrated against the same standard, and flowmeter 4 is cali brated against a different standard, the systematic uncertainty for z is larger than if all the meters had been calibrated against different standards (Case 1). Case 3 — Flowmeters 1, 2, 3, and 4 calibrated against the same standard, ‘In this case, the systematic uncertainties are 20 ASME PIC 191-1998 By © Brg = Bry = 3.16 lbmn/ne and Bing = 9.06 Ieee with Barymy * Bmyry = Brey = C3 Nbra“hs)( tbmn/he) and Bing = Bnang = Bmymy = @ Ibm) @ toe) Using £4. (82) for four measurements, all with correlated systematic uncertainties, the systematic tuncertainty for z is 8, = [lh By)? + (eB)? +B)? +O, Brg)? + 28) Barsony +2008 + 2&8 Boums + 20, 03Banyeny * 22% Emme + 20,048 mm)? [Gnq)? + (Bm)? + (Bm) + (Em)? + 2mm; + 2Bimmy ~ 28mg + Bry = 25mm = Bread B, = 2.0lbm/br Note the signs for each of the correlated tems. For this example, caliorating all the flowmetes against the same standard will yield the minimum systematic uncertainty for z. In general, correlated ystematic uncertainties can either decrease, in- ‘cease, or have no effect on the systematic uncer- inty of the result, depending on the form of the data reduction equation and on which parameters have cortelated systematic errors. ASME PTC 19.1-1998 TEST UNCERTAINTY Rev FIG. 8.1 SCHEMATIC RELATION BETWEEN PARAMETERS CHARACTERIZING NONSYMMETRIC UNCERTAINTY 8.2 NONSYMMETRIC SYSTEMATIC UNCERTAINTY In some experiments, physizal models can be used to essentially replace the arymmetric uncertain. ties with symmetric uncertainties in additional exper- imental variables. If this can be done, then it should be, but if not, then the'method of para. 8.2.1 should be used. This paragraph presents a method for determining nonsymmetric uncertainty intervals in these cases [10]. 8.2.1 Nonsymmetric Systematic Uncertainty Inter- vals in a Measured Variable. Ifthe distribution of the systematic uncertainty associated with a measured variable is not centered on tre measured value, then the overall uncertainty interval for the measured variable will not be certered on the measured value of the variable. In this case, the ‘ollowing procedure should be employed for construcing a nonsymmetiic, uncerainty interval for the messured variable (see Fig. 81) (a) Specify an interval (R - B, X + B" relative to the measured value within whichone may expect the systematic uncertainty to fall wih 95% confidence. The interval that is specified should contain the mea- sured value of the variable, (©) Define ¢ as the difference between the center of the interval specified in a) andthe measured value. (6) Estimate the component of systematic uncer tainty B by one-half the length cf the interval speci- fied in (a). (d) Estimate the total uncertainty for the measure~ ment using the standard formula: Ups = 2108/21? + (Sy) (8.4) (6) Fstimatea 95% confidence interval forthe mea- sured variable using (+ ql = ss es) (. Express the final result as an asymmetric 95% confidence interval for the measured variable as fol- lows, where the lower limit on this interval is given by: Kowerina = K = (Uys ge K-66) and the upper limt on this interval is given by Reppoctoss = X + (Uys + q) =X Ur B7) Example: Suppose a thermocouple is being used to measure the temperature of a ges stream, but the user of the thermocouple believes there may be a tendency for the thermocotiple to provide a tempera- ture reading thet is bwerthan the actual gas tempera~ TEST UNCERTAINTY ture due to a radiative heat tansfer mechanism. The user does not have enough information to properly correct the thermozouple reading for these effects, but wishes to account for them in an uncer- tainty analysis. From a sanple of more than 30 readings using the thermocouple, the user finds that X = 5347°C and Sy = 24°C. If the user believes that the tue gas temperature may be as much as 10°C higher than X due to raciaticn effects, then how should the user specify a ronsymmetric cont dence interval accounting for this nonsymmetric systernatic uncertainty? Step 1: Specify an interval (corresponding to 95% Confidence) for tte systematic uncertainty in question. In this case, the user of the thermocouple bdieves that the true gas temperature falls between the average measured with tie thermocouple, X = and a value that is 10°C higher than X, ie., 5447°C, Determine @, the difference between the center of the interval specified in Step 1 and the value measured with the thermo- couple, In this case, q = G44.7C + 534,7°CV2 = 534.7°C = 5°C. Estimate B by one half of the length of the interval specified in Step 1. In this case, B= (10°02 = S°C Estimate the (otal uncertainty for the mea- suromont, Uys. [n this case, Ugg 2idseCyay + (24°C) = 6.9°C. Estimate @ 95% confidence interval for the average using [X + g] = Us. In this case, this 95% confidence interval is given by [534.7°C + 5°C] + 6.9°C. Express the final result as an asymmetric 95% confidence interval for the average using X (Uss + @. In this case, this 95% confidence interval is given by 534.7°C + (6.9°C + 5°O. Note that in this example the asymmetric 95% confidence inierval for the average extends from Riowatinit = 534.7°C ~ (69°C ~ 3°C) = 532.8°C, iz, 19°C below the avenge X, t0 Kupperinit = 534,7°C + (6.9°C + S°C) = 546.6°C, ie, 11.9°C above the average X. This esult corresponds to the User's intuition that the true value ofthe gas tempera- ture may be higher than the value measured with the thermocouple. ‘step 2: Step 3: Step 4: Step 5: Step 6: 8.2.2 Nonsymmetric Systematic Uncertainty Inter- val for a Derived Result. A ronsymmetric systematic uncertainty in a measured variable may also result a ASME PIC 191-1998 in a nonsymmetric uncertainty interval in a derived result, The follewing procedure should be employed for propagating the nonsymmetric uncertainties in a set of measured variables to a derived result (see Fig. 82): (@) Detemine ,, Uss,xi, and qj for each average TF; that contributes to the deiermination of the derived result, ry, Xa...» Xl _ (b) Define gas the diference between rt + or, Kat Gar ee /Xnt Go) and Oy, Kay ves Fede (o) Detemine the Sersitivity coefficient, @, for each average X; that contributes to the derived result by the standard means. 'fa sensitivity coefficient de~ pends on the values of any averages, i.e, 6 = 6(%;, Xp, +, Xo then it should be evaluated at the point (+ 91,2 + Gay «++, Xa + Gn) (¢) Estimate the total uncertainty for the derived result, Uys,, using the standard formula: Uosr = (A Ussexi)? + (6,Usse)? +. + (GyUss.x9) (8.8) (@) Cotimate 4 95% confidence interval for the de- rived result using: HR + te Rat dap Rat Ge) * Ussy (8.9) (8. Express the final result as an asymmetric 95% confidence interval for the derived result as follows: HR Kaye) # (Uys ® 9) (B10) where the lower limit on this interval is given by howe tit tH, Kay oe oe Xe) = Wes.- 9) mre Ur (6.11) and the upper limit on this interval is given by Fappe tins = AE, Kay oy Xd + Was,.+ 90 sreue (8.12) Example: Suppose the user of the thermocouple in the example in para. 8.2.1 wishes to use this gis temperature to estimate the speed of sound for the gas using the ‘ollowing relation: ASME PTC 19:1-1998 AR + up Zsa aR — ‘TEST UNCERTAINTY FIG. 8.2 RELATION BETWEEN PARAMETERS CHARACTERIZING NONSYMMETRIC = teary? where k, the ratio of specific heats, and , the gas constant for the gas, are ‘aken to be constant with negligible uncortainty and 7 i: the measured value of the absolute temperature. in this case, T = (® + 273.2) K, where X is the average value of the temperature of the gas using the thermocouple, What is the uncertainty interval for ct Step 1: Determine 7, Uss, r. and qr for the mea- sured variable T. in this case, T = 807.7 K, Uss.7 = 6.8 K, and gy = 5K. Determine ge, the difference between e(7 gh BO Mag His ease ge "2 (812.7 K)" - (807.7 K)'8) = tea (0058 soo Determine the senstivity covfficient, 8, for the measured variable T. In this case, 8; = (1/2) (kRMT"S, Since this sensitivity coefficient depends on 7, it should be evaluated at T+ gy = 8127 K, so that here Oy “= (1/2)kRN812.7 Ki)" {kR1"* (0.0175 K, Estimate the total uncertainty for the de- Fived result: Us, this case, Uns « = HCkRI" (0.0175. 4) (6.9 KH = UE)" (0.1208 K") Step 2: Step 3: Step 4: 32 é UNCERTAINTY Step 5: Estimae a 95% confidence interval for the deived resul using (T+ q) = Ups, c In this case, this 95% confidence imerval i ven by "(612.7 40" = eR) (0.4203 Step 6: Express the final result as an asymmetric 95% confidence interval using c(7) + Wys.e = de) In this case, this 95% confi. dence interval is given by [kR''* [807.7 KY? = {{KR}"* (0.1208 RK) = [kR]"* (0.0878 K'%). In this example, the uncertainty interval for the speed of sound of the gas extends from 0.12% below to 0.73% above the value for the speed of sound assessed using the measured value of the temperature. 8.3 FOSSILIZATION OF CALIBRATIONS Definition of the Measurement Process is a prereq- uisite for determiring measurement uncertainty esti- mates. For example, different Defined Measurement Processes for the same test will result in different estimates of measirement uncertainty. Elemental er- rors are classified as random if they add scatter to a result. If they éo net, they are systematic. Final ‘TEST UNCERTAINTY classification is dependent on the Defined Measure- ment Process. Calibration errors are frequently reclassified as a result of the Defined Measurement Process. As an ‘example, for test measurerrents involving multiple calibrations (Le, calibrationsperformed prior to each of the multiple tests or peiodically throughout a test program), the calibration process random error will introduce scatter in the test data set and therefore should remain classified as a random error source ‘or random uncertainty in the uncertainty analysis. However, for test measurements involving a singe calibration (Le, calibrations performed only once during a test program, and ali data samples processed Using the same calibration constants), the calibration process random error does not have an opportunity to introduce scatter in the test data set and therefore should be reclassified (fossilized) as systematic error during the final uncertainty analysis. In this case, the calibration random uncetainty should be treated as another systematic elenental uncerainty and combined with the other calibration systematic un- certainty to obtain the tota systematic uncertainty for the calibration process as: BF = GREET. CF (8.13) where B= total systematic uncertainty of the sin- gle calibratior process by, bp. . -= elemental systematic Uncertainty com- ponents of the calibration process Sc= standard devittion of the calibration process, an estimated standard devia- tion of the random error components of the calibraion process 5, Sh n= number of calibration points within a single calibration ‘A more general form of Eq, (6.13) for a single calibration is Bt = 21(8,/2)? + (5371+ (e148) libration process initial systematic uncer tainty before reclassifying (fossilizing) the calibration random uncertainty into system= atic uncerainty 3 ASME PTC 191-1998 To illustrate the fossilization process, consider a master flowmeter installed in line with a test flow- meter for use ir establishing a calibration correction, for the test meter. As a result of the calibration process, the sysematic uncertainty of the test meter is, replaced by tha of the master meter. The calibration process randon uncertainty is a function of the Inial random unceraintes in both the master and test meters. When a data set of interest from the test meter involves multiple calibrations, the calibration process random uncertainty wil cause seater in tho individual test meter aata samples and thus should remain classified as calibration random uncertainty in the uncertainty analysis. However, when a data. set from the tesi meter involves only a single calibra- tion, the calibration process random uncertainty is common to al data samples and thus manifests itself in the dita set as a systematic uncertainty (becomes fossil zed). In this case, the initial random and systemat cess should be combined and carried forward as a systematic uncertain. : Fossilization of calibration random uncertainty can occur at any oF all levels of calibration, from NIST (National institute of Standards and Technology) to the test application, depending on the defined Calibration, or measurement processes, or both. Sometimes, multiple calibrations are periormed but the results are averaged into a single set of Calibration vormants for use in processing all data samples (e.g, pre- and postcalibrations). In this case, a portion of the random uncertainty still becomes fostilzed into systematic uncortaingy. Tho magnitude has been reduced by having averaged multiple cali- brations. The tem Szin Eq. (8.14) should be reduced by dividing by JN. Thus a more general form of Eq, (8.14) woud be: Br = 28/2) + (SVR (8.15) where N= numbe of repeated independent {cingla) calibrations averaged and used in obtaining a single et of calibration constants common to all amples within the test data set. 8.4 SPATIAL VARIATION [11]. Measurement requirements for a performance test are often such that an average measurement of individual parameters is required. Most instrumenta- tion, however, yields a poirt measurement of a ASME FIC 19.1199 TEST UNCERTAINTY Complete overinn FIG. 83 THREE POSTTEST CASES parameier rather than an average measurement While this point characteriste may be ucoful for other purposes, it raises a problem in determining performance level. In many instances, the quantity ‘measured varies in space, making the point meacure- ment inadequate, Thus, itoften is necessary to install several measurement sensors a! different spatial loca~ tions to account for spatial variations of the parameter being measured. Spatial variaton effects are consid- ‘ered ertors of method (see para. 5.4.4). A simple illustration of the impact of spatial varia~ tion is the measurement of average velocity of an incompressible fluid in 2 pipe. At low Reynolds numbers, the flow is laminar and has a parabolic velecity profile. One measurenent will not give the true average value directly. Forexample, the velocity at the center of the pipe is twice the average velocity. This is in contrast to the situation at high Reynolds numbers, where the flow is turbulent. At higher Reynolds numbers the profile spprcaches uniformity and any measurement will yied a good estimate of the average velocity. Usually, test corditions vary between these two extremes ind it is not possible to correct the readings in a simple manner. Circum- ferential variations may ako be present. Therefore, other approaches are chosen, such as installing ‘multiple sensors and averaging the outputs. An exarn- 34 ple of this is an everaging pressure probe often used to measure average velocity ata cross section of pipe. ‘The uncertainty of this mean velocity would be calculated by considering it to be a determined result. An additional systematic uncertainty may need to be assigned b the mean result to account for the possible difference between the determined mean and the true mean. 8.5 POSTTEST ANALYSIS OF REDUNDANT ‘MEANS Posttest analysis is required to establish the actual test uncerlainty interval. It is also used to confirm the pretest uncertiinty estimates, or to identify prob- lems, or both. When redundant instrumentation or calculation methods are available, the individval results and their uncertainty intervals should be compared with each other and with the pretest uncertainty analy a When comparing redundant means (¥, and X,) and theit uncertainty intervals, the three cases illus- trated in Fig. 8.3 need to be considered. Case 1. A problem clearly exists when there is no overlap between uncertainty intervals. Either uncertainty intenals have heen grossly undorost TEST UNCERTAINTY mated, an error exists in the measurements, or the true value is not constant. Investigation to identify bad readings, overlooked or underestimated system- atic. uncertainty, etc., is necessary to resolve this, discrepancy. Case 2. When the uncertainty intervals completely overlap, as in Case 2, one can be confident that there has been a proper zccounting of all major uncertainty components. The smaller uncertainty in- terval X, + U, is wholly contained in the larger Interval X; = Uy. Since the individuat measurements are valid, the weighing method technique described in para. A2 can be used to obtain a better estimate Of the true value than either of the individual mea- surements 33 ASME PIC 191-1998 Case 3. Case 3, where a partial overlap of the uncertainty intervals exists, is the most difficult to analyze. For koth measurements and both uncer- tainty intervals to be correct, the true value must lie in the regon where the Uncertainty intervals overlap. Consequently, the lerger the overlap, the more confidence we have in the validity of the measurements and the estimate of the uncertainty intervals. As the difference between the two measure- ‘ments increases, the overlap region shrinks. Standard statistical kypohesis testing may be used to evaluate the significance of differences observed. TEST UNCERTAINTY ASME PIC 18,1-1998 SECTION 9 — STEP-BY-STEP CALCULATION PROCEDURE 9.1 GENERAL CONSIDERATIONS. It is rocommended that an uncerainty analysis, following the methods of Sections 4 through 8 of this Supplement, be conducted before anc after each test, according to the procedure given below. The pretest analysis (see para. 44) is used to determine if the test result can be measured with sufficient accuracy, i.e, the predicted uncertainty should be smaller than the required uncertainty. It ray also be used to compare alternative instrumenta- tion systems and test desgns and to determine corrective action if the predicted uncertainty is unac- cceptably large. Furthermore, it may he used to evalu ate the need for calibration. The posttest analysis {see para. 4.5) validates the pretest analysis, provides data for validity checks, and provides @ statistical basis for comparing test results. 9.2 CALCULATION PROCEDURE (a) Define Measurement Process (see Section 5) (6) Review test objectives and test duration. 2) Listall independent measurement parameters and their nominal levels. ) _Listall calibrations and instrumentsetups that will affect each paremeter. Be sure to check for uncer tainties in measurement systen components thataffect ‘two or more measurements sinultanecusly (correlated, uncerainties). (4) Define the functional relationship between the independent measurement parameters and the test, result, () List Elemental Error Sources (see para. 5.4) (1) Make a complete and exhaustive list of all possible test uncertainty sources for zl parameters (c) Calculate the Systemaiic and Random Standard Deviation for Each Parameter (see paras. 6.1 and 6.2) (@) Propagate the Systematic and Rendom Starndard Deviations (see paras. 7.1 through 7.4) (1) The systematic and random (sample) stan- dard deviations of the independent parameters are propagated separately all the way to the final result. 7 (2) Propagation of the standard deviations ix done, according to the functional relationship defined instep (a)4) above, by using the Taylor series method (see Appendix C). This requires a calculation of the sensitivity factos, either by differentiation or by com- puter perturbation. (@) Calculate Uncertainty (see pata. 7.5) (2) Combine the sysiematic and random uncer- tainties to obtain the total uncertainty. ( Report ‘The uncertainty analysis for each calculated result should be reported on two tables. The first is a detailed report which displays all the information used in the calculation of the nominal value and uncertainty of the result, The second is a table that summarizes the uncertainty information at the result level, For most uncertainty analyses, all measured pa- rameters will have symmetric systematic uncertain- ties and large degrees of freedem. For some analyses, ‘one or more of the systematic uncertainties may be onsymmetric (ee para. 8.2), and for other analyses, the degrees of freedom may be small for some of the uncertainties (see Appendix B). ‘The detailed report table should include, as a minimum, the following information for each param- eter used in the calculation of the result: '¢ Symbol usec in the calculations © Description © Units ‘© Nominal value (average of measurements), X; Systematic uncertainty, 8; © Random (sample) standard devi: © Sersitivity, € ‘© Sysiematic contribution to the total uncertainty of (8 trent (84) «© Random consibution wo the wotal uncertainty of the result (5; 8;? tion, Sx,1 The summary report table should display the fol- lowing information associated with the result: ASME PIC 19.1198 TEST UNCERTAINTY. TABLE 9.1 ‘TABLE OF DATA (EXAMPLE 10.2) Independent Parameters an ‘aie posee Se tie. weit twit vit Cet, Ppt te Symbol Description unis "Value e) ‘Ss, Oy “§ Bisageaient aes ee ee o inewmce” = Whe gam 2 ws 7 Siting Eat, SERB soa hoteles 8.0 8003s nts ieee eaten TABLE 9.2 SUMMARY OF DATA (EXAMPLE 10,2, Gicnetat Abie Sc Gaara Ab an ac Atte Ucn, canton (MERE) (E08) (apa syint__oscion un“ Vi fran aks 3 Hi © Symbol used in the calculations © Description © Units © Calculated value (calculatec using nominal values for measured parameters), F which all degress of freedom are large and all systematic uncerkinties are symmetric and uncorre- lated. These tables are taken from Example 10. The calculations can be made and the values dis- played on either a relative or an absolute bi nothing in the procedure changos — but the headings is — ( 7 in the report tatles should always specify which ‘© Systematic uncertainty, By, 2 /S (@ 4) basis ic being td. 6 speciy wh \ vier When one or more of the parameters has a system- atic uncertainty that Is nonsymmetic, the report tables will have to be expanded. The detailed report table for an uncerainty analysis that includes systern- atic uncertainties that are nonsymmetric will be set up with the following columns: ‘© Random uncertainty, 25p, ( Pay 7) (BY + © Symbol used in the calculations © Description © Total uncertainty, Up, ( Vv 2 Tables 9.1 and 9.2 are exanples of the detailed and summary tables, respectively, that should be used for the reporting of uncertainty analyses for 28 © Units '® Nominal value (average of measurements), X, © Systematic uncertainty, 8, ‘TEST UNCERTAINTY © Offset from the center of the systematic uncertainty interval, q;(which is O for symmetric uncertainties) # Random (semple) standard deviation, 55,1 © Sensitivity, 8 Systematic contiibution to the total uncertainty of the resut, (Ba) + Random contribution toe ttal uncertainty ofthe result, (Se 8? The summary report table for a nonsymmetic uncertainty analysis will contain the following columns: '* symbol used in the calculations * Description © Units © Calculated value (calculsted using nominal valves for measured parameters), R Se) © Offset rom the center of he systematicuncertainty interval, qe © Systematic uncertainty, Bp, f . x ® Random uncertainty, 2s zr G& *) ® Total upper uncertainty imit, Us, (: |G)-#-»] © Total lower uncertainty imit, UR, (J) #-5) For symmetric uncertainy analyses in which one or more of the unceriaintes has small degrees of freedom (i.e, less than 10), the detailad report table will contain the following information for each of the measured parameters: ‘© symbol used in the calculations © Description © Units . © Nominal value (average of measurements), X; © Systematic uncertainty, 3; © Random (sample) standerd deviation, S5,; (© Number of measurements used in calculation of nominal value, N; 39 ASME PTC 13.1-1998, © Degrees of feedom associated with systematic un- certainty, 2, © Degrees of freedom associated with the random uncertainty, vs, , © Sensitivi “ * Systematic contribution to the total uncerainty of the resut, (% a) © Random contribution tothe total uncertainty of the result, (S559)? «© Value of ter associated with the systematic uncer tainty in the denominator of the Welch-Satterth- waite equaion used to calculate composite de- (G3) 1 (F ) agrees of freedom for the result, © Value of term associated with, the-random uncer- tainty ia th: denominator of the Welch-Satterth- waite equation used to calculate composite de- 22s of freedom for the rest, (“LS”) *u ‘The summary report tables “for an_uncortainty analysis that ircludes parameters with small degrees of freedom would have the following columns: ‘© symbol used In the calculations © Description © Units + Calculated value (calculated using neminal values for measured parameters), R «systematic ncerainyy, 38 © Random uncertainty, Se © Compesite degrees of freedom, vg (see Appendix B for formua) «# Studen’s ¢ associated with the composite degrees of freedom, | ae + tat ezine (1) VG + Se Tables 9.3 and 9.4 show the detailed report for Example 10.2 Table 9.5 siows the resulis table that would be used for this case. As Example 10.2 invelves no parameters having nonsymmetric uncertainties or small degrees of freedom, N/A has been entered into the colunns not required for this analysis. ASME PIC 19:1-1998 TEST UNCERTAINTY TABLE 9.3 REPORT OF RESULTS (EXAMPLE 10.2) Independent Parameters Number of Measuremests Degrees of Degrees of Used to Freedon Absolute Absolute Freedom for Calculate Abgoute for Offset of | Standard Standard Nominal systematic Systemate Uncertainty Deviation of Deviation of Nominal Value Uncerinty Uncertaity intenal” the Mean’ the Mean Syrbot__Deseription Units Value. —_(N). 6) Dag) © Discharge eoefclent 058 SONA «0.0075.——S*NA ° ° wa throat dtameter Inches NA oc! NR ° ° VA D bletcianeer incles 6 NA 002 NA ° ° A p Water dersty tml 62.37 NA oocs WA ° oo NA 60" 3) ‘hDierential presse in. HO 100 NA 03 WA ° oa NA head aerss ven a 687) TABLE 9.4 REPORT OF RESULTS (EXAMPLE 10.2: Independent Parameters forWeleh’ Uncertainty Satiorthwaite for Welch: fosalte Tonal Saltertwale Fandom Formula Absoite eran — (8) Nominal Seniviy (Bg Contiuton 2 ((@suat) Symbol___Deserition Units “Value 10) 2 ns me om © Discharge coeficienst =e? ~« TE ° wa NA 4 Thyost diameter Irches "4 e621 6e3 ° NA NA D Inlet dameer Wehes 11312984 © NA NK Water density, i tb 62.37 MASE BE NA NA oF (251 hifi pressire inHjO 100.682.0666 NA NA head sere 40 ASME PTC 18.1-1998, jy 209 moody 269 () BION wos z VN 0 ° lee wag pM wu (h—tq) eh +9) sxmpms yao (,¢6 45) 3M) ob teuann fs) amen sun uepduss90 joowls samen setup enema AUDEN prPR ‘nan Tomaiog So Mumma ‘Saag 1g vuawaouh ment Siseaung “PHTEIPIV Io ayuoy_yeuRS OSIY srysty inst stout 2840 ren wo ie PRRPDRO 701 TW) SLINSRE 10 RIVAWNS $6 18s {TEST UNCERTAINTY me ‘TEST UNCERTAINTY ASME PTC 19.1-1998 SECTION 10 — EXAMPLES 10.1 FLOW MEASUREMENT USING PITOT TUBE 10.1.1 Define the Measurement Process. The flow rate of an incompressible fluld Ina pipe may be determined by multiplying the integrated-average yelocity of the fluid by the cross-sectional flow area of the pipe. One technique for measuring the integrated-average velocity of the fluid is to traverse the cross-sectional flow area with a pitot tube. Measurements at cach traverse point can be used to determine local fluid velocity, Traverse points are typically specified at the centroid of equal areas so that the integrated-average velocity may ke estimated as the average of the meastred values for all traverse points, For this example, the velocity is measured at 40 unique traverse points (10 traverse positions along four equally spaced radii) corresponding to the centroid of equal areas as shown in Fig. 101. A total of 60 measurement is taken in succession at ‘ach traverse point once th: pitot tube is positioned. ‘Several simplifying assumptions are made for this example: (a) the pipe diameter is lerge compared to the pitot tube diameter such that bockage effects and wall interforance affects can be noglectodt (b) the velocity pressure developed across the pitot tube is automatically measured by a differential pres- sure transmitter; (©) the output ofthe differential pressure transmitter is measured and recorded by @ computerized data ‘acquisition system (DAS); (d) the DAS computes velocity for each measure- ment by taking the square root of the output of the differential pressure transritter, making the appro- priate corrections for fluid density, and making the appropriate calibration conections and unit conver- sions: (€) the DAS automatically takes 6Oreadings at each traverse point and computes and records average val- ues and siandard deviations for the data collected at each traverse point; (9 the pltot tube, differential pressure wansmitter, and DAS are calibrated togsther as 2 system; and a @ the flow rate (and hence the velocity profile) remains constant for the duration of the test. 10.1.2 Data Summary. The computerized data ac- uisition systen Is used to compute average values from the 60 measurements at each traverse point using Eq. (4.2) The resulting average values at each traverse point, Xj, are summarized in Table 10.1. The traverse points are located at the centroid of ‘equal areas so that the integrated-average velocity in the pipe is approximated by the average of the velocities determined at the averse polis. Fist, the average veocity along each radius, Vis approxi= mated as Vite = Kuits. aoa) fs 10 Then, the everage velocity in the pipe, V, is approximated as Views) = BO (10.2) The subscripts / and / are used in the previous ‘equations to designate radius and traverse positions, respectively. ‘The results of these calculations are summarized in Table 10.2. 10.1.3 List Elemental Uncertainty Sources. The sources of uncertainty which are considered random in this simplified example are those causing variation in the 60 repeited measurements of velocity at each traverse point, The sources of uncertainty which are considered systematic in this simplified example are the uncertainty of the calibration of the instruments used to measule and record velocity at each traverse point, and the uncertainty of the integrated-average velocity due tv spatial variation. 10.1.4 Calculate Random Standard Deviation. The DAS is prograrimed to output the standard deviation of the 60 measurements at each traverse point using ASME PTC 19:1~1998 TEST UNCERTAINTY fai 81 Racivan Redives2 FIG. 10.1. TRAVERSE POINTS (EXAMPLE 10.1) TABLE 10.1 AVERAGE VALUES (EXAMPLE 10.1) Raiius 1, <1 Radlis2,7 = 2 Radius 3,7 3 Radius 47 =a Traverse Point thy ye) % Xin 72a 331 527 sat 50 j=? 545 S33 335 sie 523 335 561 337 ant Sis 568 547 5a 5s 574 350 530 365 37 se 535 37) 370 Bes 53 524 57% 365 565 536 575 370 Se 372 50 30 he aso TEST UNCERTAINTY ASME PIC 19.1-1998, TABLE 10.2 SUMMARY OF AVERAGE VELOCITY CALCULATION (EXAMPLE 10.1) Parameter Value Gem 2 iw Ser Bie 532 Kw 55 Vis TABLE 10.3 STANDARD DEVIATIONS (EXAMPLE 10.1) Radlus 1, Radius 2, Seq (tl) 5a 12 130 1.05 161 1103 136 121 131 iD 1.96 109 126 oat 1.03 1.00 os 15 134 oi vas Eq, (4.1). The standard deviation at each point, Sq, ie summarized in Table 10.3 ince there are 60 measurements at each traverse point, Nj = 60, the cegrees of freedom at each traverse point is y= pot = 59 (10.3) The standard deviation fo: each average velocity along a radius, Sy, 1s calctlated from Eq, (7.7) 2. Sx va)" toa where The standard deviation of the average velocity in the pipe, Sp, is then calculited as 4s fos) 05) where The results ae summarized in Table 10-4. Note that the standard deviation of the average velocity in the pipe is influenced by the random ‘errors in the velocity measurements and actual va tions in the velocity with time for the 60 measure- ments at each raverse point. These values are not influenced by variation of velocity as a function of spatial position. 10.1.5 Caiculete Systematic Uncert 10.1.5.1 Caibration. Velocity is measured at each point with the same pitat tube, digital pressure transmitter, and DAS. The pitot tube, digital pressure ies ASME PTC 19.1-1996 TABLE 10.4 STANDARD DEVIATION OF AVERAGE VELOCITY (EXAMPLE 10.1) ‘TEST UNCERTAINTY TABLE 10.5 UNCERTAINTY DUE TO SPATIAL VARIATION (EXAMPLE 10.1) Parameter Vale Taranter Value ‘9, 9 0.0840 Site 010 5, (5) 0.0523 i 4 5, (vs 00618 Bets) 010 50, 9 0.0687 tess oe transmitter, and DAS are calibrated together as a system. For this example, the calibration uncertainty of the instruments is 3% of measured velocity. Since the same instrumentation is used for all measuie- ments, the systomatic uncoraintias due to calibration for all traverse points are correlated. Equation (8.2) can be used io show that the uncertainty in the average velocity measurement. V, is (106) 03 5.57 (tus) = 0.17 (fs) 10.1.5.2 Spatial Variaticn. The true value being ‘measured is the integrated.average of the velocity over the cross-sectional flow area. Averaging tho velocities at the centroid oi equal area points is a numedical approximation of this integrated-average. Even if exact values of velacity are known at each of the traverse points, the analyst must recognize that the numerical average may not equal the integrated- average value over the entire cross-sectional flow atea. This is due to incomplete sampling of a profile which varios as a function oF spatial position. There- fore, there Is an inherent uncertainty in the method used to approximate the integrated-average velocity. This uncerainty of method is sometimes referred 10 as uncertainty due to spatial variation. Since the velocity profile remains fixed for the duration of the test the spatial variation is a source of systematic Uncertainty for the test resuit. This uncertainty can be estimated in a variety of ways, including: (2) special tests which provide independent know!- edge of the velocity profile; (b) special tests which compare the measurement technique to other techniques which yield the desired integrated-average; () published reports which document the uncer- tainty of similar measurement techniques in similar measurement situations; and (2) evaluaticn of the variation in test measurements 2s a function of spatial position. For this example, the velocity profile is distoned due to the presence of flow disturbances upstream, of the measurenent location. Therefore, special tests were conducted to map the velocity profile along each of the feur equally speced radil. The results (of these tests indicate that the average of the 10 traverse locations along each radius sufficiently char- acterizes the integrated-averege value along each radius. However, since no additional pipe taps ere available for ‘esting, it is unknown whether the average of the traverses of the four radii suficiently characterizes the integrated average around the pipe ‘due to circumferential variations. Therefore, the un- certainty due ‘0 spatial variation is estimated by evaluation of the variation in the radial averages. ‘Assuming that the four measured radial averages come irom a population cf values which are normally distributed, an estimate of the systematic uncertainty in the average velocity measurement due to spatial variation Is etd son) were p= the sytemaic uncertainty die t spa! variation Ss= the standard deviation of the racial averages [computed using Eq. (4.1) and the radial averages presented in Table 10.2} Le the number of locations (radii) used in the compuiation of the average value The calculation is summarized in Table 10.5. 10.1.5.3 Combining Elemental Systematic Un- 2. The systematic uncertainties due to cali- 46 TEST UNCERTAINTY TABLE 10.6 UNCERTAINTY OF MEASUREMENT (EXAMPLE 10.1) 020 0103 330 0.21 avenge” Re ~=SST bration and spatial variatior are combined using Eq. (6.8) as follows: Bm (B24 Belt = 0.201Hs) (10.8) 10.1.6 Total Uncertainty ofthe Measurement. The total uncertainty of the measurement is determined using Eq. (6.10) as follows: Us = 2) (2) + (se? = 0.21 us) 10.9) The uncesainty ofthe measurement is summarized in Table 10.6. 10.2 FLOW RATE UNCERTAINTY [12] 102.1. General Description. In this example, the test objective is to determine the flow of water using a 6 x 4 in, venturi Gee Fig, 10.2) within an uncer- tainty of 0.5%. A pretest analysis is required to determine if an uncalibrated venturi could be used to satisfy the test objective, and, If not, whether calibration of the venturi would achieve the desired objective. The clearest way to present the results of the steps in the uncertainty analysis is to develop a table in which the names, definitions values, uncertainties, , etc, are displayed for each of the vari- ables required for the analysis. The table associated with the uncalibrated case is shown here firs, as Table 10.7. The table has bee1 developed in accord. ance with the step-by-step procedure of Section 9. The steps in the development of the table are as follows: (@) Define Measurement Process. Flow rate can be calculated by making the measurements required to define the independent variab'es found in Eq. (10.10) (see Fluid Meters Research, ASME, 6th Edition, 1971, p. 54): ASME PTC 19.1-1998 0.099702C a? Jah 3) ‘Ihe defintions and values for each of the measure- ments used in the calculation of the mass flow rate are displayed in the first four columns of Table 10.7, (10.10) Uncalibrated Venturi Case () (2. (@ List Elemental Systematic Uncertainty Sources; Estimate Elemental Uncertainties; Calculate the Systematic and Random Uncertainties. The sys- tematic and rardom uncertainties are determined from experience. The degrees of freedom associated with each of the uncertainty estimates is greater than 30. The systematic uncertainties (also known as bias limits) for each input parameter are shown in Table 107 displayed on an absolute basis. The standard deviation of the mean of the measurements taken is shown for each of the independent variables. These values are also expressed in absolute terms, so the units are the same as the units for the measured pa- rameter. (6) Propagate ‘he Systematic and Random Uncer- tainties. The sersitivity of the result to each of the individual parameter uncertainties is calculated, ci- ther numerically or analytically, in accordance with Section 7.2. ‘A quick and sasy way to numerically calculate the sensitivity coafficients of the independent param- iets Is to develop a table using a spreadsheet program on the personal compuier. Table 10.8 shows the results obtaired using such a spreadsheet, along with the formules used In the spreadsheet. ‘s can be seen by looking at the formulas shown in Table 10.8, the same basic equation for the calculation of fm was repeated seven times (once for each independent parameter, P) with P; being replaced with P,+ PQp where Qp is the quantity by which P| is to be perturbed. A perturbed fn is therefore calculaed for each independent parameter, and the baseline value of ih is subtracted from each perturbed rh. Dividing the difference by the amount by which P, was perturbed (P,Q) provides the abso- lute sersitivity efficient ‘As noted in pera. 7.2.1, the sensitivity coefficient can also be determined analytically, by finding the paral derivatives of the result with respect to each of the measured parameters. Table 10.9 shows the formulas for the partial derivatives for each of the measured parameters with respect to the calculated ASME PIC 1911-1998, TEST UNCERTAINTY TABLE 10.7 UNCALIBRATED CASE (EXAMPLE 10.2) Independent Paramaters ‘Absclute Absolute Systematic Standard Uncertainty Deviation of the symbol Description Units Nominal Value (a) Mean (S,), € Discharge coeiicert = 0.964 (0.0075 0 é “Throat dameter Inchoe * 0.001 ° D Inlet dianeter Inches 6 0.002 ° ° Water density 2 60°F (25) Ibm ear 0.008 0.002 h Difirontial pressur> esd in, HO 100 3 oa 221055 venti (68°F) GENERAL NOTES: (@)Thesystematicandrancom esimates for density are based on water temperature meaturemens having sysematicand randomunceralnies Of o2"F and 0.1", respectvely () The systematic unceranty forthe citferential pressure head is assumed to be ene-half the least count ofthe manometer scale 2-610. FIG. 10.2 SCHEMATIC OF A 6 « 4 in, VENTURI mass flow rate, and the sensitivity coefficients found. using these formulas. The absolute contributions cf the systematic and random uncertainties of the incependent parameters, are shown in the sixth and seventh columns of Table 10.10, The absolute contribution of the systematic, uncertainties of each of the independent parameters. to the uncertainty in the result (Le,, mn) are calculated by multiplying the sensitivity coefficient (8) for each parameter by half of the systematic uncertainty (treat- as a standard deviation) of the parameter and squaring the product. The absolute contribution of the random uncertainties is found in a similar fashion, with the standard doviation, of the mean of the measurements being mulipliel by the sensi and the product squared. The square root of the sum of the systematic contributions (times Students t which equals 2, as all degiees of freedom are greater than 20) yields the systematic uncertainty component of the result, and the random component 40 of the result uncertainty is found by following the same procedure vith the random contributions. By looking at the numbers in the last two columns of Table 10.10, it can be seen at once which Parameters contritute most to the uncertainty of the result. In this ease, it i clear that the largest contibu- tor is the systemitic uncertainty in the discharge coefficient, C. @ Calculate Uncertainty. The total uncertainty of the resultis then calculated by root-sum-squaring the systematic and random contributions. Table 10.11 shows the nominalvalue, and the systematic, random, and total uncertairtes for rh, calculated as described above. 10.2.1.1 Relative Sensitivities and Uncertainties. Sensitivities and uncertainties may also be stated in relative terms. To convert the uncertainty of any parameter from absolute to relative terms, itis neces- saty only to civide the absolute uncertainyy by the ‘TEST UNCERTAINTY ASME PIC 181-1996 TABLE 10.8 RELATIVE SENSITIVITY COEFFICIENTS IN EXAMPLE 10.2 (CALCULATED NUMERICALLY) ‘Quantity by Which to Perturh Independent Parameiers (Qp): 0.001% Formulae for Abeolute Sonaithity ‘Absolute Seritvily Noma om on Symbol Values wR F, © 0.98% 0.099702 (C~ CQ) Poh 130 €Qy 4 4 0.099702 (4 + Qe Vph 862 & 20) D. D 6 0.099702C dah “4 a J + DQ, DOr ° 62.97 0.999702 d Vier oO, wn h 10 0.099702€ a? igh + FQ) 0.692 Or Symbol Resut Formula for sn tn 138.336 0.099702 o gh nominal value of the parameter. Multiplying by 100 would provide an expression of the uncertainty in terms of a percentage of the nominal value. To convert the sensitivity coefficient to relative terms, the absolute sensitivity rust be divided by the nominal value of the result and multiplied by the nominal value of tre independent parameter. ‘Multiplying by 100 would yield the percentage change in the result for a 1% change in the measured parameter. Tables 10.12 through 10.14 display the same parameters as Tables 10.6, 10.10, and 10.11, respectively, bu‘ with the sensitivities and uncertain- ties expressed in relative terms rather than absolute terms Calibrated Venturi Case (a) Define the Measurement Process (0), (0, (@) Lis: Elemental Uncertainty Sources; Est- mate Elemental Uncertainties; Calculate the System atic and Randoms Uncertainties. In this case, the sys- tematic uncertiinty of the discharge coefficient changes. In addtion, the systematic uncertainties of 4“ ASME PIC 19.1-1998 TEST UNCERTAINTY TABLE 10.9 ABSOLUTE SENSITIVITY COEFFICIENTS IN EXAMPLE 10.2 (CALCULATED ANALYTICALLY) Formals for Absolute Sensitivity am ‘Abeolute Senctvty Nominal ‘2 Symbol__Valene " a, c oes see: 0B 140 locszoacye ahi # (ai (SZ) ‘ ‘ (2) 0.039702¢ ka ~ neo 0997026 # haat > 6 Ta 4 (-@ye aa oommrorc a Le a 6237 a . (5) 2 0,039702¢ d? o » 100 0.992 FT the throat and pipe diameters are eliminated provided that the same dimensions which are used in the cali- bration process are also used in the test. By experi- ence, for a calibrated venturi, the following estimate is used: U, = 0.122%, (©), (0 Propagate the Systemaiiand Random Uncer- tainties; Calculate Uncertainty. These new values can be inserted into the tables (spreadsheets) used for the uncalibrated case. As the formulas do not change, there is nothing more that needs to be done. The parameters and results are disphyed using a relative 50 basis for the uncetainties and sensitivities in Tables 10.15 through 10.17. le can be seen that the test objective of in + 0.5% is marginaly satisfied by using a calibrated venturi, and that the most promising path to take 20 obtaining additional reductions in the uncertainty of the mass flow rate determination would be through an increase in the number of differential head readings. (@) Report. Tables 10.12 through 10.17 would be included n the report, with text ncting that calibration ‘TEST UNCERTAINTY ASME PIC 13.1-1998 TABLE 10.10 ABSOLUTE CONTRIBUTIONS OF SYSTEMATIC AND RANDOM UNCERTAINTIES OF INDEPENDENT PARAMETERS (EXAMPLE 10.2) Independert Parameters Abwolate Random Absolute Uncertainty Nominal Sensivity Cottrbution Symbol Description Units Value Gut? © ‘Discharge conficent 0.96 ° d ‘Throat dameer Inches 4 0 D Inlet ameter Inches 6 ° 2 Water density a 60°F (251 Ibi e237 49266 4 Difereral pressurz head in HO 100 7.6662 across vent (2 68°F TABLE 10.11 SUMMARY: NOMINAL VALUE, SYSTEMATIC, RANDOM, AND. TOTAL UNCERTAINTIES IN ABSOLUTE TERMS FOR mm Cakelatid Rel Abie Absalate systematic Tota Absolute ny Radom Uncertainty, Ge Uncen, Se i" Calculated = Lies Symbol Description Units Value V224) NAF sudnter (8) 79 Wa tow nolo 838 198 ass z Fy TABLE 10.12 RELATIVE UNCERTAINTY OF MEASUREMENT (EXAMPLE 10.2) dependent Parametens walaive eltve Standard systematic Deviation of Nominal Uncertainty the Mean Symbot Description ‘unite Value’ @ Sa © Discharge coefficient osas 0.76% 0% d Throw diameter Inches 4 9.0250% 0% D Inlet iameter ches 6 9.03339 0% 2 Wate deity at 60°F 25] emi 6237 0.0064% © .0032% 5 Diteentil pressure head inv HO 100. 0.300% 9.400% cose venturl (at 68°F) 3 ASME FIC 191-1998" TEST UNCERTAINTY TABLE 10.13 RELATIVE CONTRIBUTIONS OF SYSTEMATIC AND RANDOM UNt PARAMETERS (EXAMPLE 10.2) ‘Independent Parameiers ERTAINTIES OF INDEPENDENT Telane sycmac Eelatve Uncertainty Random Cony Uncertainty Nominal hap Syrian symtot Desripton Unit Vive 2) ai © Bicone oan ~ a8 Tas E05 ¢ ¢ Tron dane inchs : 971 e008 ° 5 inletdamaer Inches : 673 809 8 & Wier denaty 60°F) Ibm 237 257 Evo asre10 h Dieta prere nad in thO “0 oso = S27 © hag ron vet 607) TABLE 10.14 SUMMARY: NOMINAL VALJE; SYSTEMATIC, RANDOM, AND TOTAL UNCERTAINTIES IN. RELATIVE TERMS FOR THE UNCALIBRATED CASE (EXAMPLE 10.2) = Caleulited Rerult Relative systematic Total Relative felathe Uncectany, andon Uncertainy, 1 Caleulted a Symbol Description Unite Value NESAT — sudents ¢ th MassFlowrate bm 1308 0.76% 0.40% 2 0.87% TABLE 10.15 RELATIVE UNCERTAINTIES OF INDEPENDENT PARAMETERS (EXAMPLE 10.2) ledependent Parameters Relative Reative Standard Systematic Deviation of Nominal Uncertainty the Mean Symbol Daseription Unite Value ® Cy © _Dischargecoeficient - 0968 On2% om Threat dsmeter Inches 4 0.0:50% 0% D Inlet diameter Inches 6 0.0:33% om Wiser density at 60°F (25) Ibm? 62.37 KH = 0.0082%. hy Diflerentl pressure Neal = in KO «100, 0330% 0.400% cross ventut (at 68°F) Sama 52 ‘TEST UNCERTAINTY TABLE 10.16 ASME PIC 19.1-1998 RELATIVE CONTRIBUTIONS OF SYSTEMATIC AND RANDOM UNCERTAINTIES OF INDEPENDENT PARAMETERS (EXAMPLE 10.2) Independent Parameters ~ Relive Systematic Relative Uncertainty Fandom Retatve Conti Uncerainy Nominal Sensitvity (Gay Contribution Symbol Description Units Yalue 7 2 Sue" © ‘Discharge coafciont ona T0030 Ther © a Throat dameter inches 4 20 ones ° > Inlet diameter Inches 6 02 onto ° » Water density at 60F (251 tom 037 0500 257610 257610 h Diferentol pressue heed in. HO 108 050 5.62607 4006-06 267085 Vent (at 68°F TABLE 10.17 SUMMARY: NOMINAL VALUE; SYSTEMATIC, RANDOM, AND TOTAL UNCERTAINTIES IN RELATIVE TERMS FOR THE CALIBRATED CASE (EXANPLE 10.2) Gace Ret anveSstomatio ; Taal anv = Reatve ro sce ine vei : coulated B neering, [Ose sts oon on SRS VEE) ES ar VE) Wier few mie Tonk 8838 oe ios 3 ou of the venturi isrequired to meet the objective of 0.5% uncertainty or less in themasi flow rate determination. 10.3. FLOW RATE UNCERTAINTY INCLUDING. NONSYMMETRIC. DETERMINISTIC UNCERTAINTY This exemple is nearly identical to that presented in para. 10.2 for the uncaliarated venturi case. The only differences are the following: (1) the discharge coefficient will have a nonsymmetric absolute sys- tematic uncertainty represented by Bx = 0.0095 and 8° = 0.0055, and (2) the differential pressure head across the venturi (at 63°F) will have a nonsym- metric absolute systematic uncertainty represented by BF = 0.4 in. HzO and B* = 0.5 in. H20. The results of this uncertainty analysis are presented in Table 10.18 in which each symbol has the same description as in Table 10.7, and in which each Corresponding number has the same units, as those given in Table 10.7. Following the method given in para.’ 8.2, the absolute systematic uncertainty, 8, for each variable is estimated by B = (B' + 6/2 and the apparent offset, 4, if etimated by g = (8 — B)2. The absolute sensitivity, @, for each variable is estimated with all veriabes set to their offset values, X + g. The apparent offset in the mass flow rate, gy is determined by the difference between the mass flow rate evaluated when all variables are set to their cifset values, Rz..g, and the mass flow rate evaluated when alll variables are set to their measured values, Ry. The lower and upper limits on the uncertainty of the mass flow rate are given by U- = Usse~ Go and U" = Uys, + Gp respectively. 3 ASME PTC 19:1-1998 TEST UNCERTAINTY TABLE 10.18 RESULTS OF UNCERTAINTY ANALYSIS (EXAMPLE 10.3) ‘Symbol ¥ = 7 Ree a S Us 6 © daca oases aoc) -o000 0982 osama é 2 Gor Gor ty? 8 aoe ae é ¢ too awe kB a fear Boot tome caty Samson ater § Soot aoa ams ‘Symbol Re Beeg % saa as uw eat 20a TABLE 10.19 INPUT SYSTEMATIC AND RANDOM UNCERTAINTIES (EXAMPLE 10.4) in wares Ui slematic Uncertainty Dogrece a OMY nando ME Nutro Maumee Nomi tnareiny frelon Neca Symbol ___Dwerinion “thie Ven ey @ mney GS a Tl essce sa 1470 ea 9a 350 aes Be base tr ote et Ce a rr Tikacah Etwmenue koe et Noms Sh alin gate uncaring an asad ely canoe ad prada 95% cove. 13 Seaport ccc: ttrotursgant Foe snc a fh Townecowiea os x0 sao Soniye : to a Dac ten ° oo ts) ot ron tse ne om 66 50 paves to tS) creit a es “These uncertainties ae covelated wit hove of he inlet temperatures, COMPRESSOR PERFORMANCE UNCERTAINTY 10.4.1 General Description. In this example, the test objective is to measure air compressor perform- ance by computing adiabatic efficiency. The mea~ surements required are total pressure ard total tem- perature at inlet and exit. A schematic of the locations of the required pressure and temperature measure- meni is shown in Fig. 10.3. 10.4.2 Determination of Uncertainty. According to the step-by-step procedure cf Section 9, we have (a) Define Measurement Process. The functional re- lationship involved is (PyPr4 — YT) — 9.11) where = adiabatic compressor efficiency at 7 0.871 P= inlet pressure, nominally 14.70 psia P,= exit pressure, nominally 95.52 psia inlet temperature, nominally 530°R xit temgerature, nominally 960°R tio of soecitic heats, nominally 1.40 TEST UNCERTAINTY PT tates ASME PTC 19.1-1988 exit FIG. 10.3 TYPICAL PRESSURE AND TEMPERATURE LOCATIONS TOR COMPRESSOR EFFICIENCY DETERMINATION (b), (@) List Elemental frror Sources; Calculate the Systematic and Random Uncetainties for Each Param- eter. Table 10.19 lists typical systematic and random Uncertainties for pressure an temperature. (d) Propagate Systematic and Random Uncer- tainties. The individual parameter errors ere propa- gated into efficiency unitsby aTaylor series expansion as given in Appendix C. The random uncertainty for efficiency Sey = WOn? (Str? + (Ong? Se)? + (On) Ser? + (Or) (Ser (10.12) where the absolute sensitivites are = ly = WoAliPysP Yr” [PARP] = WANE PY PAP Or, tanh iF 0.04090 = r= Yn 0) WP 9 + m yh) = o0829 _ UL UPymyersvm — 9) oo, = Halles = 1 — -ocomer 6, = ot {ucr/Po’ 9 = 1h 6 oo203 1, = TH The systematic uncertainty for efficiency is (Gr,? (Br) + (61,9? (Br)? +6 p,? (Bn)? + (On.? (Br? +2 Gr, Or (Bryn) (0.13) where Bin = (0.5) (05) + (0.5) (0.5) + (0.07) (0.97) + (0.4 (04) = 0.655 (‘RP Bryr_ is the cevariance of the error sources com: mon to T, and Tz Note that in fq. (70.13) an additional term has been added to account for the correlation between the inlet temperature systematic uncertainty 8, and that portion of th: exit temperature systematic unce tainty B;, with which it correlates. In this case, four of the elemental systematic uncertainties in Br, and Bry arise from identical cerror sources. The absolute values of the sensitivities along with their corresponding systematic and random uncer. tainties ate summarized in Table 10.20. Utilizing Eqs. (10.12) and (10.13), the following expressions for the random standard deviation and the systematic uncertainty of the result are ob- tained by: {(-0.040:0)2 (0.033 + (0.006297 (0.17 + (0.00367) (9.607 + (-0.00203)? (0.707) = 11.506 « 10-* + 1.143 x 10-* + 4.849 y 10°F + 2.019 x 10°5)"% Sin (10.14) as ASME PIC 191-1998 ‘TEST UNCERTAINTY TABLE 10.20 SYSTEMATIC AND RANDOM UNCERTAINTIES (EXAMPLE 10.4) 1 Result Unite 7 Dograee Number of Mesure: Random of Measure- nent Nominal Semitiviy — (B- [e(e\] urge adem me Symbol___Descripton Units Value) ) So @ (9) ese) a Inlet presse sla. —«14.70 004080... ,.. O1GEO6 1506605 530 1 a Exit pressure fla 95.52 0.00673 02m £06 1143 £08 330 1 Ty (ote |1)) inlet empeature R330 a.00367 2.26 6.05 4.89605 330 i 7, ote )) Exietempannre —@__—_-940.0_—000202, 274 £06 2019505 +30 1 NOTE: (1) These systematic uncensintes have some components that are contelated (same uncertatty sources) TABLE 10.21 RESULT SYSTEMATIC AND RANDOM UNCERTAINTIES (EXAMPLE 10.4) Te Result Unis Syitematc Degree Random} Number of est cate, _UASEHtANY_pcertaimy Freecom Measuremenss Uncertainty “ive Symbol Description _Units_Result_ (872) @/D (Sz) & Ww Uys Ut 7 Compresor Hfcimey O6714 00015 0901S 0.00323 t 3.0067 0.0067 fene : Sky = 0.003 efficleney units (10.15) _freedom greater tan 30 for the uncerainty, thus Stu: dent's tis 2.00. The Lbs uncertainty is calculated as > 2 24/2 = (0.040301 (2922) « co.00629» (217) Usg = + (2). 1B Y/2P + (Sy,)21"% (2) (0.00157 + (0.003) 120.0067 at 95% coverage (10.18) + .ooaene (222) 5 -ooo20s? (ey + (2) (0.00367) (~0.00203) (0.665/4))"* = (0.167 10 + 0.286% 10-8 + 2.26 x 10° 4274 104~2.50% 105" (10.16) shown in (f Report. Thefinal uncertainty report is Table 10.21. The calculation of uncertainty for the efficiency measurement is now complete. If it is desired to (©) Calculate Uncertainty. .the uncertainty is ob- know the percert uncertainty, now is the time to tained by combining the systematic uncertainty ofFq. calculate that. The efficiency level is 0.8714 = (10.17) and the random standard deviation of Eq. 0.0067, or 0.88714 + 0.77% for Uss. (10.15). Because all the rancom uncertainties are When multiple measurements of the result are based on experience, the degres of freedom, », are made (N times), the above Sg., value would need assumed to be greater than 30 throughout. Therefore, _to be divided by -(N. This accounts for the averaging using Welch-Satterhwaite wculd yield degrees of of several measurements of the result B,/2 = 0.0015 efficiency units (10.17) 36 SST UNCERTAINTY [ASME PTC 19.1~1998 TABLE 10.22 SPECIFIED CONDITIONS (EXAMPLE 105) Paramiier Specied Cenditions Throtle pressure, PF Thre trmperatue. Te Shave prasure, Pe Goneratr output power hector, pF= Generar hydrogen pressure, Fa Generar size 800 pig 850°F 35 In. Hg 35%, 30 psig 109,000 kw 0.5. STEAM TURBINE PERtORMANCE UNCERTAINTY 105.1 General Description. One of the more com- vlcated tests described by Performance Test Codes 5 that of a steam turbine. as an example of the uncertainty procedures for a -urbine test, the steam cate test described in [12] is analyzed. Note that the number of readings, the quality of est Instrumentation, and the test data proner were assumed for the purpose of ‘his example. The test described in this example is not in strict accordance with PTC 6. The objective of the test is t0 determine the corrected steam rate of a nonextracting, condensing turbine used in a gas turbinsteam turbine come bined cycle power plant. The corrected steam rate is determined by measuring the steam rate at test conditions, $a and correcting the sieam rate for specified conditions, SR°. The specified conditions for the cycle are shown in Table 10.22. A simplified diagram of the nonextraction condensing turbine is shown in Fig. 10.4. 105.2 Define Measurement Process 10.5.2.1 Description of Test Instrumentation. The location o test instrumentation is shown in Fig. 10.4. Throtle steam pressure is measured at two locations with calibrated, digital absolute-pressure indicators. Throttle steam temperature is measured attwo locations with calibrated, digital TD tempera- tute indicators. The exhaust pressure is measured at eight diferent locations in the exhaust casing with calibrated, absolute, digtal pressure indicators connected to basket=ype probes. The generator out put power 2rd power factor are determined with a calibrated, 0.000573) 0.0097 The total comparative uncertainty is then 0097" ua=2 (em Y #10.00697 = 0.0169 The uncenainyy for each test can be calculated as shown in Tables 10.39 and 10.40. The uncertai ties for each test are presented in Table 10.37. The results ate plottec in Fig. 10.12, The test conclusion that can be drawn from this figure is the pump is ‘operating consistently when compared since ‘the uncertainty band: overlap. TABLE 10.39 UNCERTAINTY PROPAGATION FOR COMPARATIVE UNCERTAINTY (EXAMPLE 10.6) a 240 4p ‘Ox, a a 5 By Sy OS rial tet 292) o 13866 Taeaee 280 aoe 001s e001 O02 B Tene 2 tote 700 e097 $000 80002 & 13986 -«7510 oees 8002 0.000; 00 ~0.9000 Fa 1ore 700 aorsso aes 0003 80 2000 a Tse 786 “ooi3e aos 0.0003 20 0.0000 * Vere were sraes 33 3.000 an 0.0000, > 1306 aos ses 12 2.0001 39, -nnnco é 1aste coor uate 10000 2.0001 08 8.000 Conve test 6/880) @ —3eS 1363S wa 0006 “301550008 ~o.06 be abr 7 interval for the population standard devation, ‘A confidence interval for a population pe:- centile. Finally, a new time-sha‘ing computer pro- ‘gram calculates a wide variety of statistical intervals, including corfidence, tolerance, and prediction intervals, Reference 6. THE EXAMPLE FROBLEM The ealculalon cf the thee intewals ae iesated here by the following numerical example. Assume that ‘readings cbtaired ona normally disvibutedperformance parameter based on a random sample of fve urits are: 514,493, 487, 493, and 51-4, From thisinfoeation, thesampie mean p and the sample standard deviation + are calculated by well-known expressions: fie] [ete sah tio” 6-0 sia. where i, Irarethe valuesof n given cbeerations. REFERENCES (1) Hahn, G1, “Additional Factrs for Calculating Prediction Intervals for Samples from a Normal Distribution,” journal ofthe American Statistical ‘Associaton, 65, Decerter, 1970. "Factor. for Calelating Two-Sided Prediction Intervals for Samples from a Normal Distribution." Journal of he American Staetical ‘Assoclavon, 64, Septerter, 1969. , Siatetiealinercalefor a Normal Pop: Ulation,*Jourral of Qualty Technolegy, Volume 2, Number’, pages 115-125, July 1970; Volume 2, Number 4, pages 195-206, October 1970, Navella, Mary Gibbons, Expoument Stacie, National Buresu of Stancards Handbook $1, US ‘Government Printing Offee Nelson, W, 8, TwosSamle Prediction,” Ganeral Electric Company TIS Repor 68-C-404, Novern- ber 1968. (Auilable from Distibution Unit, PO Box 43, Building 5, Room237, Schenectady, New York 12305) ‘Summary Siastics Paclage — ONE-SAMS***" ocumert 003401, Genial Electric nfarmation Service Deparment, 7735 Old Georgetown Rosd, Bethesd, Marylaed. a ° @ 6 © A.2 WEIGHTING METHOD Whenever the value of a parameter is approxi- mated by several different measuring methods, then ‘weighted means of the systematic erors, random 87 certors, and parmeter uncertainties may be calcu- lated. Let X; represent best estimates of a parameter by N measurement methods. Then X, the weighted mean of the measurements, can be given by X= 3 WE, where W; are the required weighting factors. ‘A weighting principle which is statisteally valid and is based on weighting by variances is applicable in this case (5) This is true since the systematic uncertainty component of a parameter B is assumed to equal two tines the standard deviation (square root of the variance) of the possible distribution of systematic uncertainty. The variance of this distribu- tion is (B/2P. This is combined with S33, which: i tho variance of the average measurement. Therefore, Uss is a combitation of variances: une me)” Theroforo, aca wn 2(a) where U; represents the uncertainties of X, by the Uss model __for two measurement methods with the means %, and Xp, Eq. (A.1) yields Ue “= Ore and ue treat Using these sane weighting factors, the systematic and random uncertainties of the weighted mean are given by the roct-sum-square relations = (witat + weee)” and 3 = (wits? + 252)” where 5 and 5: are Se, and Sey Similarly, the degrees of teedomn of Uss is given by the Welch-Satterthwaite formula: where S, and S) ere Sz, and Se,. These values are combined as Uss to obtain the weighted uncertainty of X cecording to fq. (4.5) a5 follows: Tos = tos (Bra? + 5341 > 3 5, B, and Uss should be reported. where AS usual ee A3- OUTLIER TREATMENT 31 Genial, All measurement systems may pro- dice spuridus data points. These poin's may be caused by temporary or intermittent malfunctions of the. measurement system. Ewvors of this type should not be included as part of the uncertainty of the ‘measurement. Such points are considered to be meaningless as stesdy-siate ‘est data, and should be discarded. Figure A.1 shows a spurious data point called an outler. All datz should be inspected for spurious data points as a continuing check on the measurement identification criteria should be based on engineering analysis of instumentation, thermody- namics, flow profiles, and past history with similar data. To ease the burden 0” scanning lage masses of data, computerized routires are available to scan steady state data and flag suspected outliers. The suspected outliers should then be subjected to an engineering analysis. The effect of outliers is ta increase the standard aviation of the system. Test are available to deter- mine if @ particular point from a sample is an outlier. In most of these tests, the arobability for rejecting good point is set at 5%, This means that the odds against rejecting 2 good point are 20 to 1 (or less). The odds could be increased by setting the probability of rejecting a good data point lower. 29 However, this practice decreases the probability of rejecting bad data points. For small samples, bad data points are hard to identify. Two tests are in common usage for determining whether or not spurious data are outliers. These are the Thompson + Technique [18] and the Grubbs Method (19]. The Thompson rTechnique is excellent for rejecting outliers, but also may reject, some good values. The Grubbs Method does not reject as many ‘outliers but the number of good points rejected is smaller. is Supplement, the Modified Thompson 7 ue! described below is recommended for ing posible outliers for further examination. 3.2 Thompion rTechnique (Modified). Consider a sample (X) 0' N measurements. The sample de tuon (Sq) and the mean (X) of the sample are calcu- lated. Suppose X; the jh observation, is the suspected outlier. Then, he absolute difference of ¥; from the mean (X) is cileulated as = KX] Using Table A.2, a value of 7 is obtained for the sample size (M at the 5% significance level. This limits the protabiliy of rejecting a good point to 5%. (The probability of not rejecting 2 bad data point is not fxed. It will vary as 2 function of sample size.) The test for te outlier is to compare the difference (0 with the product 75,. Wf 8 is larger than or equal to *Sy, we say X; is an outlier. If 8 is smaller than 75x, we say % is not an outlier. A3.3. Outlier Examples. Example 1: There were 440 temperature probes installed In one stage of the turbine of a jet engine. The 40 probe readings were averaged and 5 was calculated from the average for each probe. 2 79 581-103-121 -220 137 0 1 m9 38s 60 ug 32 26 12-3698 “07-00 3-486 126-22 179417 m5 3 H55 555 ard 334 are sopected oul “resent et rmestomsnnts ieee ie TABLE Az MODIFIED THOMPSON 7 (AT THE 5% SIGNIFICANCE LEVEL) (EXAMPLE B.3) N r N 7 3 Tist a 1932 4 135 33 1933 5 1371 54 1.934 ‘ 1356 5s soaa 7 iwi 36 1.935, a 19 ° 1.935 8 7 58 1.935 10 798 59 1336 i" sis 60 1.936 2 128 61 1337 3 1.840 a 1937, “4 11850 3 1337 15 1858 6 1.938 16 ves 65 1.938 ” vat 66 1.938 6 1876 o 1939 19 1.681 68 1939 20 1.885 68 1.939 2 1.889 70 1940 2 193 7 1.940 23 1.296, n vo40 24 1693 3 25 1s01 7 26 1.04 75 2 1.506 76 28 11508 7 2 110 7 30 ag 79 3 113 60 32 1.315 ar 33 1316 a2 38 1917 3 35 1319 84 36 1.020 35 37 iar 86 38 192 a7 39 1.923 88 40 1974 as 4“ 15 90 a 1926 1 a 1.937 2 a tox 23 5 13%8 94 46 v9 95 a 1.979 96 48 1.930 7 49 1931 98 20 1331 99 31 1932 100 89 To illustrate the calculations for determining whether ~555 is an outlier, the following steps are taken, Mean @ Sample standard deviation (5,) = 140.6 Sample size (N) = 40 By the above equation for- 6, $ > [-555 — 14125] — 556.125 From Table A2, 15g = 1.914 x 1408 = 270.9 Since 8 > 1Sy, we conclude that ~555 is an outlier according to the modified Thompson r Technique. Repeating the above procedure for 334, 8 = [934 - 1.125] 332.875 Since 8 exceeds 75; for the suspected point, 334, ‘we conclude that 334 also is ¢ probable outlier and also would be deleted. This procedure would be coxtinued until no more outliers are found. Whenever outliers are found, the validity of the data should be questioned. Recalculate X and 5, wher oulliers are deleted and use these new values in reporting results. In the report on the test, a desciption of the use of the statistical data rejection stould be given. Example 2: In the following sample of 12 values from the distibution of a qualiy characterisic (data from {20}}, the suspected outler is 17. 345677899 101117 Mean (X) = 6 Sample standard deviation (5) = 3.717 Sample size (N = 12 By the above equation for 8, S=7-a)=9 From Table A.2, 75, = 1.829 x 3.717 = 6.80 Since 8 > +S,, we conclude that 17's an outlier according to the modified Thompson + Technique. 90 ‘A4 PARETO DIAGRAMS It is often useful to display the relative sizes of the components of a whole with a bar chart, One particular type ofbar charts called a Pareto diagram after Vilredo Pareto, an Italian economist, who used this type of diagram in his studies of the unequal distribution of wealth. Most of the uses today, with extensive activity in the area of quality control, are attributed to Joseph Juran who defined the gencral iple known as the “Pareto Principle” — the “Vital Few, Triviel Many." Mathematics were devel- ‘oped which described the distibution, but for the purposes illustrated here, the diagram can be defined as a bar chart with the bars arranged in descending. order of size. To apply this to a measurement uncertainty exam- ple, the first step is to define the individual systematic and random uncertainties of the mean in terms of their relative individual percentages of the total uncertainty, Uss. To derive the percentages for Uss, the calculations ‘must be made in terms of Uss?. The reason for this is that bs =2f(8) +52 Since the squares of the random and systematic uncertainties are additive under the square root, the total uncertainty must be squared before making the calculation. First, for the systematic uncertainties, Bj, define where B= D166, And that term is the following fraction of total uncertainty squared, Uss? «() AAZE = the systematic component of Uss? Use ” where Lo Spurious data ring x " pararater Level x x x x x pao x XX x error OR KKK KY KK Ot 2? x EO FIG. A.2 OUTLIER OUTSIDE THE RANGE OF ACCEPTABLE DATA 4 (503? 4 SP and where where . 452 = 43 (Seal? 45 = 4 Seo" ‘And that term is the following fraction of total uncertainty squared, Uss? 4 Sq the random component: of Uss? The percentage of Ups? for exch individual system- 4 Se atic uncertainty is Usst The percentage of Uss? is 4 (55,6) 458 Proof that this sum is 1, or 100%, for a case of For the random uncertainty portion cf total uncer three individual srstematic and random uncertainties tainty, the fraction is defined as is shown below: a B,, By, Bs and Sx, Sry, Sry ard = = = 1 4 (Sa)? 100 shoes of Uys? = For the sum: Sum = B,% + 8: %+ 8% +5q% + Sa% + S,a% GFF) 4(B) sae __ Alsat + Sa? + 5x7] 100 +(2) vase 4 (+ as] 4S) ase 100% Using the compressor perormance example in para. 10.4 for illustration: Posolute Fandom Absolute Lrcenainyy Systematic Absolute Parameter ofthe Mean Uncartainyy —Sensivity Py O03 psa OM pia ~0.0209 Py 017 pea. pia +40.00629 n ose owe 0.00367 a or Let 0.00203, Stop 1: For By% of Uss Boley, % By? + (Oy % BaP + py * PM a aaa 59 where Br, = (0.5) (C05) + (0.5) (05) + (0.07) (0.07) + (0.4) (0.4) 0.665 (FY Br, 7, is the coveriance of error sources common to Ty'and Ty B= [(-0.0409 x 0.02? + (0.00629 x 0.17 + (0.00367 x 0.8) + (~0.00203 x 1.6)" + 2(0.00367) (-0.00203) (0.665)"* = + 0.003 Sx= U-0.0409 x 0.03)? + (0.00629 x 0.17 + (0.00367 x 0.6} + (-0.00203 x O74 = 0.003 Therefore 2 Use =4 (8) «4 59 ~ 0.00005 W Bp,%6 of Uns = ——90000855 = 135% Similarly, and ao Fe 387 af Bak W781 yeas 2 s216 Ba 24 3 231 aw Bo Uncerisinties nao -oE1_i tp tt oy 8h 8H 8 my BA FIG. A.3 PARETO CHART FOR RANDOM AND SYSTEMATIC UNCERTAINTIES (EXAMPLE 10.4) Br, 1% Of Use Sh, = 9.24%, G7, = 39.17%, Sr, - «(22 0367)(~0.00203)(1 = 16.31% 0.00005 = -19.0% Figure A. illustates the Pareto charts for the percentages csleulated for Urs. However, since the systematic uncertainties for 7; and Ty are correlated, the relative percentages for Br, Brz, and By,7, should be combined as Step 2: For $p,% of Uss shown’ in Fig. A.4. ‘As can be seen from this diagram, Sr, is the largest contribrtor to the total uncertainty, gs, and Bp, the strallest, The ideal erd result of an analysis such as this would be to take corrective action, if 4 5,04 possible, to recuce the contribution of major factors = FEE «100 in the final uncertainty through changes in methods, ‘s instrumentatior, or both, Although this application > (of Pareto diagrams has been used to determine the = £10.03 x -0.0409% 59 relative contributions to total uncertainty of system 0.000495, atic and randem uncertainties, the method can be applied just as easily to the individual estimates of = 12.16% the elemental errors that contribute to. systematic Similarly, and random uncertainties. 3 o 2ea7 3 Bo : 8 4s 5 2 3 1601 z Ew zw 5 oy Sy 88h Oy Bry Bey Bary FIG. A.4 PARETO CHART FOR RANDOM AND SOME COMBINED SYSTEMATIC UNCERTAINTIES (EXAMPLE 10.4) 94 APPENDIX B — UNCERTAINTY ANALYSIS MODELS This Appendix is rot a port of ASME PTC 19.1-1998 ard is included for information purgoses only.) Bt ISO UNCERTAINTY ANALYSIS MODEL Appendix B is adapted fron [21]. The uncenainty model given in the ISO Guide to the Expression of Uncertainty in Measurement (1993) {1] is presented in this paragraph. The uncertainty model requires estimates of the uncertainties or each of the elemen- tal error sources for each parameter in the data reduction equation. These «stimates are combined through use of the SO model to calculate the band about the experimental resut where the true result Is thought 10 lie with C% confidence. The estimates of the elemental errors fal into. two categories, systematic uncertainties, (bj, for the fixed errors and sample-standard deviations of the mean, 5), for the random errors, where i represents each pa- rameter. The systematic unceraintes are related to those fixed errors that remain afterall calibration correc tions are made. Systematic uncertzinties can be esimazed through manufacturer information, calibre= tions, and, in most cases, through sound engineering judgment. There will usually 2¢ a set, K;, of elemental systematic uncertainties for each parameter, 1. AS discussed in paras. 6.2 and 63, each elemental estimate, (bj, is taken to te the prediction of the 95% limits on a particular distrbution of possible cerors for that particular error source. Typically, these ‘enor distributions are assumed Gaussian (normally ‘ibuted) or reciengular (uniformly distributed). The ISO model uses the standard deviation of the assumed distribution, not the 95% estimate, to calcu- late the uncertainty in the result. Obtaining the standard deviation from the 95% estimate is simply a matter of dividing the estimate by the appropriate distribution factor, D [ue., 2.0 for Gaussian, 1.65 for rectangular (1.73)(0.95)]. The systematic. uncer- tainty for parameter /at the standard deviation level, (53), is determined from the estimates for the K; elemental error sources for that paremeter as os sar = len/DP (Ba) Note tha: Dy can be different for each elemental The estimate of the random error for a parameter is the sample standard deviation of the mean, or the estimate of the error associated with the repeat- ability of a particular measurement. The sample standard deviaton of the mean for each parameter determined fom N, readings as +o laa) 2, m-mF]" en where os For the case of single readings (Nj = 1), previous information must be used to calculate 5; [4). Consider an experimental result that is determined from j measured variables as 2s AM, RY oy ‘The ISO Guide defines the combined standard unce- tainty as s 2, | +3 5" ie ug = (Sel? 42 3D HA(Sadn (8.5) where a= 6.6 a. ‘The first two terms on the right side of Eq, (B.5) represent the systematic uncetainty of the result at the standard deviation level. The third term is the standard deviation of the result, or the random Uncertainty at the standard deviation level, The co- Variance of the random errors is assumed to be Zero. The covariance of the systematic errors, oF the correlated systematic uncerainty, (5p), 1s deter- mined by summing the products of the elemental systematic uncertainties at tre standard devietion level for parameters /and k that arise from the same source and are therefore perfectly correlated {9} (see pare. 8.1). In order to obtain the oven! uncertainty in the result, Up, at a specified confidence level, the ISO Guide recommends that the combined standard un- certainty be multiplied by a coverage factor. The coverage factor is the value fiom the ¢ distribution for the required confidence level corresponding to ‘the effective degrees of freedom in the result, 1. The values ere given in Table B.1 To find 2g, the Welch-Saterthwaite formula is used: [2 tla(Saul? + (6:57) = 8.7) s fest! +3 tbudDa) rl wy boy where »4, is either 2s, = N= (8.8) or the degrees of freedom of theprevious information if Sis estimated 14]. The degrees of freedom of the systematic uncertainties 2,3, may be known from Previous information or estimated. The ISO Guide recommends the approximation 6.9) i) Pon = 2 ib in parentheies is an estimate of ity of estimites of (b),. For in- stance, if one thought that the estimate of (5), was reliable to within +25%, then 96 TABLE B.1 STUDENT'S f DISTRIBUTION [8} © V\{ 0900 | 4950 | 0990 | 0995 | 0999 1 | 6314 | 706 | 62.657 | va7az1 | exer9 2 | 2920 | #303 | 9525 | Woes | “a1s90. 3 | 2353} 31e2) sasi | 7453 | 12924 4 [2932 | 2276 | 4608] 5590] sei0 5} 2015] 2571] s032] 4773] 6068 6] 1943 | 2447 | az0r | 4ar7} 5959 7 | 1995} 2368 | 3499] 4029} 5.408 8 | 1e60 | 2306} 3355 | 3033] Soa 9 | 1833 | 2362} 3230 | 3690] 4761 wo} e2} 2228] 310 | sar] 4587 | 1796 | 2201} 3106] 3497} aaa we | r7e2} x79} 3085] 3428 | aare Bz} r160} 3012 | 3272 | 422s us| r7er | 2a4s | 2977 | 3326 | 4140 ws} 753] vast | 2047 | 2386 | dors a6 | 1246 2s2 | 3252] aais a | 1740 2e08 | 3223 | 3965 ve | s7a4 zere | sisr | 39x 19 | 1729 2061 | 3174 | 3885 20 | urea 284 | 3153] 3.850 aif azar 2es1 | 3135] 3819 22 | air 209 | 3ais| 3792 23 | 1214 zeo7 | 310 | 3768 26 fan 2797 | 300] 37s 25 | 1708 2787 | 3076] 3725, 26 | 1700 | 2056 | 2779 | 3067 | 3.707 27 | 1703 | 3osa| 2771 | 3087 | 3.s50 2] 1701 | io4s | 2763 | 3047 | 3.574 20} ve99 | 3045 | 2756 | a.020 | 3.558 30] 1697 | 2002} 2750 | m0] 3.46 40 | v6s¢ | 2021 | 2704} 297 | 3551 60 | 1671} 2000 | 2560] 215 | 3.460 v0} 1656] 1980 | 2017 | 2560} 3.373, | 1645 | 1960 | 2876 | 2807 | 3291 GENERAL NOTE: Gives are the values of for confidence level Cand numbor of degees of reedom oy — $10.25) = 8 (10) With 4 known the proper t value is obtained from Table B.1 for C% confidence and multiplied by ug from Eq, (6.5) to obtain the overall uncertainty in the result, Up, at a C% confidence level gg % OSs) a2 (2, (OSes? +2 + & ay} Separating the systematic and random torme in Eq, (8.11) and taking the square roct yields = te{ 3, raison a 18.11) 18.12) and (Se) = «[Zusr]" 8.13) Where Bee represents the systomatic uncertainty of the result and t¢Se represents the random uncenainty Of the result each at C% confidence level. B.2_ LARGE SAMPLE UNCIRTAINTY ANALYSIS APPROXIMATION The method described in pare. B.1 above is the strict ISO method. In the previous version of this, stndad 2}, the recommendation was made that the t value for 95% confidence with 30 or more degrees of freedom could be taken as 2. Therefore, fot large degrees of freedom in the result (xq > 30), Eq. (B.11) becomes {assuming a Gaussian distribution for all systematic uncertaintes) 8, on (09) 25 308% + $s] 10 The large sample systematic uncertainty of the resut 28 95% confidence is then ne =[fe00r ok Fi eta " which is equivalentto Eq, (82) and the 95% random uncertainty estimate is or i Me 250 = [3 wasir] e19 which is Eq. (77) multiplied by 2. By squaring Eg. (8.15) and (6.16) and substituting into Eq. (8.14), the large sample uncertainty expression given in Eq. (7.14) is obiained (including the correlated systematic uncertainty terms) ve, = 2[(2ee)«cst] It has been shown [22-24] that for most engi- applications, when the degrees of freedom for the result from Eq, (8.7) is 9 or greater, ys can still be taken as2 toa good approximation (93% to 95% coverage) and the lerge sample approximation given by Eq, (6.17) can be used, (8.17) 3 EXAMPLE PROBLEM To demonstrate the ISO Uncertainty Analysis Model given in para. B.1, an example is presented of the experimental determination of the loss coeff. cient for water flow through a globe valve, Cyy, 25 shown in Fig. B.1. The loss coefficient is determined from the expression Bl «- th = hel 6.18) where g is the acceleration of gravity, t is tho timo required to fill a container with the water flow during the duration of the experiment, d is the pipe inner diameter, p is the density of the water, m is the mass of the water in the container at the end of the experiment, fy is the flow head entering the globe valve, ard hy is the flow head exiting the globe valve, Aso included in the above equation ‘are hy and hg, the entering and exiting head measure- ments across a section of pipe equivalent in length to the pipe tha: penetrates the globe valve fittings. This head loss was deducted from the drop across the valve since t was found in preliminary investige- tions to have a significant effect on the calculation of the loss coeficient. The flow heads were measured by extending tub- ing vetically fiom the taps against a surface that had been scaled for measuring purposes, with all four tubes being open to ambient pressure. The FIG, B.1 SCHEMATIC OF EXPERIMENT APPARATUS (EXAMPLE B.3) valve vas fully open, allowing water to flow through the apparatus and into the container. The duration Of the experiment was deperdent upon how long it to0k to fill the container and was measured with a hancheld stopwatch. Once. the container was filled, the mass of water colected was measured, {a pretest studies, i was found that the uncertainty in the pipe inner diemeter was a major contributor to the uncertainty in Cyy. Therefore, it was necessary to determine the mean diameter other than by direct measurement. The length and volume of a section of pipe. from the same lot as that used in the experiment were used to determine the mean diame- ter. This auxiliary test was repeated rine times, During the experiment, six multiple readings of the head measurements wera made in order to obtain the mean and standaid deviaton for each. ‘The random uncertainty estimate in the time mea- surement 25, attributed to the reaction of the person timing the event, was estimated using good judgment while the rondom uncerteintis in the diameter and dersity were taken to equal ze’o. The rendom uncer- tainty of the mass, 25,, was estimated based on the ability to read the scale. The systematic uncertainty estimates for the head measurements were taken to result from the accuracy of the scale used for the measurements and the variablity in the placement of the taps on the test apparatus. The systematic, 98 uncertainty in the time measurement was taken as the accuracy information provided by the manutac- turer of the stopwatch. For the diameter, the estimate of the standard deviation of the mean’ provided by the repeated dianeter measurements was multiplied by a factor of 2 and “fossilized” into a systema uncertainty with 8 degrees of freedom (nine read- ings). The systematic uncertainty in the density was estimated using good judgment while that of the mass measurement was provided from manulactur- er’s information concerning the accuracy of the weight scale. The mass systematic uncertainty was taken to be unitormly distributed while the others were taken to be Gaussian. The inputs for the variable measurements and their respective elemen- tal uncertainty esimates are listed in Table B.2 along with the degrees of freedom of each estimate, A degrees of feed>m of 50 was arbitrarily assigned fo each uncertainty that was estimated, With the random and systematic uncertainties for each variable determined, the uncertainty analysis ‘was performed using 69. (B.11) for a 95% confidence level. For this example, the systematic uncertainties attributed to the head measurements are correlated because the same measuring scale was used to graduate each vater column. The results for the uncertainty analysis are given in Fig. B.2. The plot TABLE B.2 DETAILED UNCERTAINTY ANALYSIS INPUT (EXAMPLE Symbol Values ae 72.72 (ec) 0.8 (sec) ae 0.547 (in) (0.00308 fn cry 223 thmst) 10.05 lot?) aim 30.0 tbr) (05 (be) Gh 125.3 (in) 0375 (in) hy 112 Gn) 0375 (hn) by 137.2 (in) 0375 cin) he 13055 (in) 0375 tin) 2 %y 3 (0.1 (ec) 50 8 ing NA Ey ator) NA 0 0.2 (im) 50 50 0.125(in) 5 2 ‘1 (in) 5 50 ot (nd 5 50 0.15 tn 5 NA = not applicble of the relative contributions shows the contributions of the terms shown below: OF SP; OP (Sa)2;26,8(Sa)i Each term is normalized by the largest of these terms. The plot of the normalized contributicns for this example is shown in “ig. B.2 along with its legené and the uncertainty analysis results. ‘Note that the degrees of ‘reedom of the result is 52 even though some of the variable degrees of freedom are small. In this case, fos is equal to 20. B44 COMPARISON OF UNCERTAINTY MODELS The recommended uncerainty interval equation (Eq. (8.17) differs from the uncertainty interval equa tions presented in the earlier version of this standard [2]. Previously, two equatiors were suggested, Uroo Corresponding to 99% coverage and Usss corres- ponding to 95% coverage. These equations were Yoo = 8 bs Sr 8.19) and, 9° Ups = VBP # Tas SP To simplify the uncertainty analysis process, this document has standardized to one equation corres- ponding to 95% confidence with large degrees of freedom. Detais are provided in this Appendix to construct intenals with alternate confidence levels and with small degrees of freedom. Although the recommended uncertainty equation appears diferent in form, it is Important to note that the Usss method and the new recommended Uys equation converge to the same answer for the case with large degrees of feedom. The interval size is identica ‘Monte Carlo simulation comparisons of the Uso and Uggs uncerainty models zgainst the ISO model were presented in (22. This simulation work indi- cated more cossistent coverage for the ISO model as compared to the Usp and Upss techniques for a wide range of assumptions for the degrees of freedom, different distribution forms for the system- atic ertos, and varying magnitudes of the systematic and rendom components. It was also shown that the Usy medel in Eq, (6.17) had a reasonably good ‘comparison wit the ISO model for 95% confidence. For this reason, U3; is preferred over both Uspo and Uxss (8.20) See Graph Output eM ee ee ‘contriouung Teer Graph Legend \denttving bel and Normalized Centibution fort Term 16 SyTemid = 00% 9 2+ SyeTermia = esx 10 3s Sys Termig) = 04k 11 4+ Bye Termim) = 102.0% 12 5 Sys. Termini = 2.1% 13 8+ Sa Termim= 2am 4 7 Sys. Tere (rahe 21% 15 8 Sys. Termin) = 21% 18 Ran, Termin» 5.2% 17 = Cor Term (6.8) = -4.2% Ren. Termidl = 0.0% 10 = Cau Term (9.7! = ~8.2% Ran. Termish = 00% 19 = Gor Tern iS) = 4.2% Ran. Term (om = 426% 20 = Cw. Term iO.) = 4256 Ran. Term{hyl = 09% 21 « Cx. Term (62) = —$2% Ran. Term {gl = 08% 22 «Cor Term (78) = ~62% Fon. Term (hy) = 0.0% Ren. Term (hy) = 1.2% Uncertainty Analysie Results Cy = 28 sy, = 0472 EH Meg ® 2 Begy > 0.417 2.378) tsStqy = 0220 01.75% FIG. B.2 DETAILED UNCERTAINTY ANALYSIS RESULTS (EXAMPLE B.3) 100 APPENDIX C — PROPAGATION OF UNCERTAINTY THROUGH TAYLOR SERIES (This Appendix is rot a part of ASME PTC 191-1986 and is included fo1 information purposes only.) CL INTRODUCTION Experimental results are not always directly mea- sured, it is quite common foran experimental result, 1 Kay ons Xqly 10 be defined as a function of certain variables, X,, X1, + Xq that are directly measured. The aim of this Appendix is to provide a method by which the variance of an experimental result that is not directly measured, r(X,, Xp, « y X;), an Be expressed in terns of the variances and covariances of its arguments, Xi, Xv ++ Xqe which are directly measured. The approach will be to relate the deviations in Hy, Yap =u, Xq) to deviations in Xy, Xap ee Xo) by moans of 2 first order approximation to the Taylor series expansion for xj, x2, . X») in the neighborhood of the point (m1, sar ---» #fn where poy isthe true value o! the measured variable X,. In order to facilitate this project, the function rl%, %, «1 %) will be assumed to be continuous with continuous partial derivatives in the neighbor. hood of the point (ye, 2x2: ---r Ha) C2 DEFINITIONS The primary goal of Appendix C is to present an expression for the variance in r in terms of the variances and covatlances of X;, Xp) «+ Xq. The definitions below fer “mean,” “variance,” and "co- variance” will be used throughout this Append The expected value ofa function £13) of a random variable X is given by 80001 = [fa pin ox where p(x) is the probability density function for X. 101 The mean (or expected value) 44y of the random variable X presents the special case where 10). = X, ie, cua = [_ poo de = ax The variance 7? of the random variable X presents the special case where £0) = (X ~ py? ie, BU — al = [x a pled de = one The expected valu of a function A(X, Xp,

You might also like