You are on page 1of 30
ASHRAE Guideline 2-2005 ASHRAE GUIDELINE Engineering Analysis of Experimental Data ‘Approved the ASHRAE Standards Comites on October 3, 2004, and bythe ASHRAE Board of Decors on February 10, 2005. /ASHRAE Guidelines are updated on a five-year cycle; the date folowing the guideline number i the year of approval, The latest edition ofan ASHRAE Guideline maybe purchased from ASHRAE Customer Senice, 1791 Tulle Cite, NE, Atlante, GA 30329-2805. E-mait orders ashe org, Fax< 404-321-5478, Telephone: 404-695- £400 (worldde) or tol free 1-800-527-4723 for orders in US. and Canada). ‘©Copytigh 2005 ASHRAE, nc ISSN 1049-694x American Society of Heating, Refrigerating and Air-Conditioning Engineers, Inc. 1791 Tallie Circle NE, Atlanta, GA 30329 ‘wwwashrae.org [ASHRAE Guldeline Project Committee 2 Cognizant TC: TC 1.2, instruments and Measurements ‘SPLS Liaison: Donald B. Bivens Chad B. Dorgan, Char John W. Mroszczyk Douglas T. Reind, Vice-Chair LW. Potiet Peter R. Armstrong? ‘Ager Reddy" Sally A. Hooks Claudia Wood “Denotes members of voting status when the document was approved for publication "ASHRAE STANDARDS COMMITTEE 2004-2005 Dean S. Borges, Chair David E. Knebel Richard D. Hermans, Vice-Chair Mori F. MoBride Donald B. Bivens Mark P. Modera Paul W. Cabot Cyrus H. Nasseri Hugh F Crowther Davor Novosel Brian P. Dougherty George Reeves Hakim Elmandy John Sabet Matt R. Hargan ‘Stephen V. Santoro Roger L. Hedrick Gideon Shavit John F Hogan David R. Tree Frank E. Jakob James E, Woods ‘Stephen D. Kennedy Michael F. Beda, BOD ExO William A. Harrison, CO Claire B. Ramspeck, Manager of Standards ‘SPECIAL NOTE ‘This Guidaline was developed under the auspces ofthe American Society of Heating, Reigeating and Air-Conditioning Engineers (ASHRAE). ASHRAE Guidelines are developed unde: a review proces, identifying a guideline forthe design, testing pl ato, o evauation of a specie product, concept, or pracice. As a guideline isnot dofinive but encompasses areas where there ‘may be varety of approaches, none of which must be precisely correct, ASHRAE Guidelines are wit o assist professionals inthe ‘10a of concer and expertise of ASHRAE'sTechrical Committees and Task Groups, ‘ASHRAE Guidelines are prepared by project committees appoined specially forthe purpase of wring Gules. The ‘oject committee char and vicechair must be members of ASHRAE; whl other members ofthe projet commits may or may not De ASHRAE member, al must be technically qualified inthe subject area of the Guideline. Development of ASHRAE Guidelines flows procedures similar o those for ASHRAE Standards excep! that (a) committee balance is desired but not required, () an efor Is made to aesiove conseneus But consensus ie not required, (c) guidelines are not ‘ppeasbie, and (d) quceines are not submited te ANSI or approval “The Manager of Standards of ASHRAE shoud be contacted fr 1 inlerpretaon of th contants of tis Gulden, ©: partipation in the next review of the Guielne, ©. eng constructive elem for improving ne Guideline, 4 permission to reprint potions ofthe Guideline, DISCLAIMER ASHRAE publishes Guidelines in order to provide assistance to irterested patos on lsues tat coat to the design, testing, application, andor evaluation of predicts, concep, and practices where there may be more tan one acceptable epproach Guidlines are pot mandatory and only provide one soures of information that may be helpful in any given station ‘ASHRAE INDUSTRIAL ADVERTISING POLICY ON STANDARDS ASHRAE Standards and Guidlines are established to assist industry and the pubic by offering @ uniform mettod of testing for rating purposes, by suagesting safe practices in designing and insaling equipment. by providing proper deinions ‘ofthis equipment and by providing othe Information that may serveto uide the industry. The craton of ASHRAE Standards ‘and Guidelines is determined by the need for them, and conformance to them completly voluntary. In refering otis Standard or Guideline and in marking of equipment and in advertising, no clan shall be ma stated or imple, thatthe produit has been approved by ASHRAE. other CONTENTS ASHRAE Guideline 2-2005 Engineering Anlysis of Experimental Data ‘SECTION PAGE Foreword. 1 Purpose 2 Scope ... Definitions ‘Types of Measurements. 3 4 5 The Experimental Process 6 Uncertainty Analysis Methods and Techniques... 7 8 Data Validation Regression Analysis. 9 References sev 14 ‘Annex A: Uncertainty Analysis/Propagaion of Error 15 ‘Annex B: HVACAR System Examples, 20 ‘Annex C: Reporting of Results. ‘Annex D: Bibliography (This foreword is not part of this guideline. It is merely informative and does not contain requirements necessary for conformance to the guideline.) FOREWORD Guideline 2 was originally created from the material con- {tained in ASHRAE Standard 41.5-75, Standard Measurement Guide: Engineering Analysis of Experimental Data, when it was published in 1986 and reaffirmed with minor editorial ‘changes in 1990 and 1996, Because technology and use of computer programs has increased significantly since 1986, Guideline 2 has been revised to provide a greater focus not only on the analysis of the data but also on how to collect the data to ensure proper analysis. The scope of this guideline is therefore expanded to include engineers and technicians who collect, analyze, and report information on systems covered in ASHRAE standards and other publications. Although statistics are important in this context, this ‘guideline should not be considered a comprehensive statistical text. An attempt has been made to provide sufficient back- ‘ground and references for th statistics used inthis guideline, 1. PURPOSE ‘The purpose ofthis document isto provide guidelines for planning, analyzing data, and reporting the uncertainty of experiments, 2. SCOPE Appropriate terms are defined and statistically based procedures and formulae are recommended for evaluating experimental data related to heating, ventilating, air-condi- tioning, and reffigeration (HVAC&R). Numerous examples are provided to illustrate analysis of experimental dat, 3. DEFINITIONS: accuracy: the difference between a measured value and the tue value. The degree of inaccuracy is known as total ‘measurement etror and is the sum of bias error and precision bias, faxed, or systematic error: an error that persists and is usually due to the particular instrument or technique of measurement. confidence level: the probability that a stated interval will include the true value. In analyzing experimental data a level 0f 95% is usually used. confidence limits (used for multi-sample data) and uncer- ‘tainty interval (used for single-sample data): that range of ‘values that can be expected, given a stated probability, to include the true value. For example, a statement thatthe 95% ‘confidence limit is 5 to 8 means that there is a 95% probability (19 chances out of 20) thatthe interval between 5 and 8 will contain the true value data: information obtained by experimental means, assumed to be in numerical form; recorded values of the variables; readings. deviation: the difference between a single resultand the mean of many results. error: the difference between the truc value of the quantity ‘measured and an observed value, Since the true value is often ‘not known, it is estimated by the mean. The difference between the mean and an observed value is often called its deviation. experiment: a systematic approach for collecting information cnaphysical apparatus orsystem to determine the nature ofits operation ‘mean: the sum of measurement values divided by the number of measurements. Itis considered the est approximation of the true value. ‘parent population: synonymous with population but empha- Sizing the relation toa sample population: any finite o infinite agaregaton of items orndi- Viduals inanimate or animate). precision: the ability o consistently reproduce measurements ‘of outputs for a given system state (inputs). ‘Propagation of uncersainty: the degree to which the uncer- tainties in the values ofthe parameters affect the uncertainty in the result random error (or precision error): an error that is due to the unpredictable and unknown variations in the experiment and ‘that causes readings to take random values on either side of | some mean value. Measurements may be precise or imprecise epending on how well an instrument can reproduce the subsequent readings of an unchanged input. There are two types of random errors: * additive errors that are independent ofthe magnitude of the observations + multiplicative errors thet are dependent on the magni- tude of the observations repeatability: the ability to obtain the same observed value in repeated experiments. Lack of repeatability is commonly associated with precision error, the random component ofthe total error. replication: repetitions of measurements atthe same con tions that are taken to estimate the uncertainty in the results. ‘sample: a portion, subset, or limited number of items of a population; a set of values, experimentally obtained, that is representative of the parent population. standard deviation: the square root of the mean ofthe squates of the deviations. uncertainty: a measure of the error in a measurement or ‘experimental result that reflects the lack of confidence in the result toa specified level uncertainty range (or confidence interval); the band for the error in an estimate ata certain confidence level. The greater the acceptable confidence level (e.g., 95% rather than 50%), the wider the uncertainty range. variance: the square of the standard deviation, ASHRAE Guideline 2.2005, 4. TYPES OF MEASUREMENTS. 4.1 ‘Types of Measurements. Measurements are catego- rized as either primary measurements or derived measure ments 4.11 Primary. A primary measurement is one that is obtained directly from the measurement sensor. This can be temperature, pressure, speed, etc. The key is that a primary ‘measurement is ofa single item from a specific measurement device. 4.1.2. Derived. A derived measurement is one that is cal- culated using one or more measurements, This calculation can occur at the sensor level (an energy meter uses flow and temperature difference to report an energy rate), itean be done by a data logger, or it can occur during data processing. Derived measurements can use both primary and other derived measurements. 42. Categories of Dat of data: ype and sample, 42.1 ‘Type. The type of data is determined based on the ‘dependence of the data on time and can be either stationary or time dependent. 42.1.1 Stationary. Stationary data do not change with time, Examples of stationary data include the volume of @ tank, the area ofa room, the length of ductwork, orthe size of a building. ‘Therefore, whenever the measurement is repli- cated, the result should be the same, independent of time, within the bounds of measurement uncertainty 42.12 Time Dependent. Time-dependent data varies with time. Examples of time-dependent data include the tem- perature of space, the chilled water flow to a building, and the electrical power use of a facility. A time-dependent read- ing taken now would probably be different than a reading taken in five minutes, a day, or a year. Time-dependent data cam be recorded either as time-series or cross-sectional data, 42.12.41 Time-Series Data. Time-series data consist ofa multiplicity of data taken ata single point or location over fixed intervals of time. 42.122 Cross-Sectional Data, Cross-sectional data are data taken at multiple points ata single instant in time. 42.2 Sample. The sample of data is determined based on the number of measurements taken and can be single-sample or multi-sample. 42.2.1 Single-sample. A single semple is one or more readings taken under identical conditions atthe same or dif- ferent times, Many experiments that appear to be multi-sam- ple are actually, in part single-sample experiments. If the ‘same instrument is used for a set of observations, the fixed (bias) error inherent inthe reading and caused by the instru- ‘ment or observer will persist, no matter how many times each reading is repeated, 42.22 Multi-sample. A multi-sample is a repeated ‘measurement of a fixed quantity using different observers, different instrumentation, or both. In multi-sample measure: ‘ments, the uncertainty ‘and reliability can be evaluated through the use of statistics. Merely taking repeated readings ‘There are two primary categories ASHRAE Guideline 2-2008 with the same procedure and equipment does not provide ‘mult-sample data. 5. THE EXPERIMENTAL PROCESS. 5.1 The intent of this section is to provide a general over view of the relationship between equipment selection, data analysis, and reporting of results obtained from engineering experiments. A well-planned experiment requires that the experimental equipment and procedures be selected to obtain the required results asa function of the measured variables. A Getailed experimental plan and measurement methodology ‘must be developed, The basic experimental process involves the following steps: 1. Identify experimental goals and acceptable accuracy. 2. Identify the important variables and appropriate relation- ships. 3. Establish the quantities that must be measured and their expected range of variation 4. Tentatively select sensorsinstrumentation appropriate for the task. 5. Document uncertainty of each measured var 6. Perform a preliminary uncertainty analysis, 7. Study uncertainty results and reassess the ability of the ‘measurement methods and instrumentation to meet accept: able accuracy. 8. Install selected instrumentation in accord with relevant standards or best practices, 9, Perform initial verification of data quality 10. Collect experimental data subject to ongoing quality contol criteria. 11, Accomplish data reduction and analysis. 12, Perform final uncertainty analysis. 1. Report experimental results. ‘52 Identify Experimental Goals and Acceptable Accu- racy 52.1 A well-planned experiment begins with identifica- tion of experimental goals that can be achieved within the time and budget available for the experiment. Depending "upon the goals, the quantity of data collected and the type of instrumentation required are the primary variables that will affect the cost forthe experiment. An essential part of estab- lishing realistic goals is to decide upon the required accuracy ofthe results. A higher accuracy resulls ina lower uncertainty, ‘See Section 5.6 for detailed guidance on setting up the exper- iment and documenting the goals. 53 Identify Variables and Relationships 53.1 To further refine the goals of the experiment, the entire list of relevant measurable variables should be exam- ined, Not all of the measurable variables are independent. ‘Therefore, as part ofthis step, identify the relationships that ‘wil imit the measurements or otherwise affect their validity. For example, ina steam-to-water heat exchanger, the capacity ‘can be determined based on the measurement of hot water flow rate and inlt-to-outlet temperature difference, with the be. assumption of a constant heat capacity at the mean tempera: ture between the inlet and outlet, 53.2 Alternatively, capacity ean be determined by mea- suring steam flow rate, inlet enthalpy, and outlet enthalpy. Since each enthalpy requires measurement of pressure and temperature, more instrumentation and measurements are required, with the potential of increased uncertainty in the results, The effect on uncertainty of more instrumentation and measurements for enthalpy could be larger than the effect of assuming a constant heat capacity. 54° Establish Measured Variables and Limits 'SA4.1_ For each measured variable, determine its theoreti cal limits and expected bounds to match the selected instru ‘ment limits SA.L.1 Theoretical Limits—These are the limits ofthe measured quantities based on definitions or laws of nature For example, the relative humidity must always range between 0 and 100%. 5.4.1.2 Expected Bounds—These are the upper and lower bounds ofthe measured quantities and must be doca- mented to aid in identifying potentially false or physically impossible measurements as part of ongoing quality contro For exemple, if the temperature of a conditioned room is, below 10°C (50°F) or above 32°C (90°F), itis likey thatthe observation is in error. 54.1.3 Instrument Limits—All sensor and measure- ‘ment instruments have physical limits that restrict their ability to accurately measure quantities of interest. Such limitations may be attributable tothe environmental conditions to which the sensor is exposed (sensor melts at high temperature) or due tothe technology used to obtain the measurement (8 lin- ear response is only obtained between, say, 50°C [122°F] and 100°C [212°F)), The expected bounds should be within the instrument limits. 55. Select Preliminary Instrumentation $5.1 With the variables known and their limits of mea- surements documented, the next step is to select the equip- _ment to be used in the process of data collection. Generally, as the accuracy, repeatability, and features of the instrument increase, s0 does the cost forthe instrument. Regardless ofthe instrument chosen, it should have been calibrated within the last twelve months or within an interval required by the man- ufacturer, whichever is less. The required accuracy of the instrument will depend upon the acceptable level of uncer- tainty for the experiment. 5.6 Document Uncertainty of Each Measured Variable 5.6.1 Utilizing information gathered from manufacturers or past experience with specific instrumentation, document the uncertainty for each measured variable. This information will then be used in estimating the overall uncertainty of results using propagation of error methods. The items to be documented for each measurement include: + Variable (eg. chilled water flow rate) + Instrument limits (e., 0 Us [0 gpm] to S00 Ws [7,925 gpm) + Expected bounds (e.g, 100 1s [1,585 gpm] to 300 Vs [4,755 epm)) + Instrument accuracy (¢.,, #5% of full scale = 425 V/s) 362 _ Therefore, for the preliminary uncertainty analysis where field effects are not accounted fr, the uncertainty for the chilled water flow rate inthis example would be #25 Is (40 gpm) 5.7. Perform Preliminary Uncertainty Analysis 5.7.1 An uncertainty analysis of proposed measurement procedures and experimental methodology should be com= pleted before the procedures and methodology are finalized in order to estimate the uncertainty in the final results. For example, ifthe efficiency ofa chiller is calculated using poor instrumentation, the measured efficiency of the chiller could have a large uncertainty range of 450%, with a confidence level of 95%. This means that we would be 95% certain that the true chiller efficiency will lie somewhere from 30% less 0 ‘50% greater than the observed efficiency (e.2., ifthe observed efficiency was 3.52 kW,/kW, [1.0 kW/ton}, then the true ‘value is between 5.27 KW /kW, [0.67 kW/ton] and 1.76kW-/ KW, [2 kWiton). If ths level of uncertainty is unacceptable, beter measurement devices and methods are required to obtain the acceptable level (eg, 10% uncertainty range resulting in the true efficiency value between 3.87 kW7/kW, [0.91 kWiton] and 3.17 kW,/kWe [1.1 kWiton). The higher the accuracy required of measurements, the higher the accu- racy of sensors needed to obtain the raw data. The uncertainty analysis is the basis for selection of a measurement system that provides acceptable uncertainty at least cost. Section 6 contains additional details for the completion ofthe prelimi nary uncertainty analysis, 5.8 Final Instrument Selection and Methods 5.8.1 Based on the results of the preliminary uncertainty analysis, the selection of instrumentation is changed to achieve the acceptable uncertsnty in the experiment results. In addition, the methods used are evaluated to determine if field effects can be reduced to decrease the uncertainty in the results $9 Install Instrumentation 59.1 The next step is to install the instrumentation in accordance with manufacturer’s recommendations. Any deviation in the installation from the manufacturer's recom ‘mendations should be documented and the effects ofthe devi- ation on instrument performance evaluated. In contrast to ‘measurements made under controlled laboratory conditions, ‘measurements made in the field are subject to higher uncer- tainty levels due to a number of factor, including: + Less accurate instrumentation + Little or no control over conditions + Instrumentation installation constraints (eg., insufT- cient straight length before and after a flow sensor) + Impracticality of making independent checks + Less sophisticated dats sampling and archiving methods ASHRAE Guideline 22005 59.2 A change in instrumentation or location may be required if in-situ uncertainty exceeds acceptable limits deter- ‘mined by the preliminary uncertainty analysis 5.10 Perform Initial Data Quality Ve 5.10.1 To ensure thatthe measurements taken are not 100 uncertain and represent reality, instrument calibration and independent checks of the data are recommended, Indepen- dent checks can include sensor validation, energy balances, and material balances. Detailed information on independent checks is contained in Section 7 ‘The calibration procedure establishes the measurement system's random (precision) and fixed (bas) errors for the experiment. The variation about the mean yields the random error, whereas the comparison between the mean response of the measurement and reference toa standard yields the fixed error. For example, ife thermocouple verified in en ice bath witha series of readings yelds an average of 1°C (33.8°F) and a standard deviation (Sd_) about the mean of 2025°C (40.457), the resulting random error estimate would be ofthe onder of 2.4. or20.5°C (40.9°F) and the fixed error estimate would be 1°C (.8°P 5.10.2 Verification of calibration is accomplished using instruments traceable to standards. deal, the entire measur- ing system is calibrated repeatedly to guard against dri. Asa ‘minimum, the stability of the measuring system should be ication checked by comparing post-est calibration with pre-test cal ibration 5.10.3 If the entie measuring system (e.g, sensors, amplifiers, analog to digital converters, and curvecfts of cal- ‘bration results that translate measured voltages to desired physical quantities suchas temperature, pressure ete.) cannot be subjected to calibration, as many subsystem calibrations a3 possible should be conducted to clearly document evidence relating to random and fixed errors. Estimates ofthe uncer- tainty inthe esult due tothe entire measuring system are then calculated using techniques for estimating the propagation of error, usually based on use of root-mean-square-error (RMSE) estimation techniques 5.104 ‘Traditionally, instrument calibration procedures determine and compensate for fixed (bias) errors. Usually the gain and offset of the instrument itself are adjusted to bring the output into agreement with the known input. Altema- tively the result can be compensated for fixed error during data analysis. Ifa particular instrument drifts out of calibra- tion during the course ofthe experiment, the latter approach allows compensation by using the simple average ofthe pre- test and post-test fixed error or a more complicated weighted average versus time, if warranted 5.1 Collect Data S.ll.1 The challenge for data acquisition in any exper iment is to collect the required amount of information while avoiding collection of superfluous information. Superfluous information can overwhelm simple measures taken to follow the progress of an experiment and complicate data analysis and report generation. The relationship between the desired result—whether the result is a static result, a periodic station ASHRAE Guideline 2.2005 ary result, or a transient result—and time is the determining factor for how much information is required. SALA A static, unchanging result requires only the steady-state result and proof tha ll transients have dissipated (ee, supply air temperature once setpoint has been reached), S12 A periodic stationary result, the simplest dynamic result, requires information for one period and proof thatthe one selected is one of three consecutive periods with identical results within acceptable uncertainty (eg. the power required by a unitary air-conditioning unit cycling to meet a Toad. S113 Transient or nonrepetitive results, whether a single pulse or a continuing, random result, require the most information. Regardless of the result, the dynamic character- istics of the measuring system and the full transient nature of ‘the data must be documented for some relatively short inter- val of time. For a single pulse, a controlled stat time mini- the time interval of interest, whereas a continuing, random result must be extracted from records over a long interval of time (e.g., average outdoor air temperature). 5.11.2 The data collected should be subjected to the meth- ‘ods outlined in Section 7 to ensure validity and quality con- tol 5.12 Accomplish Data Reduction and A\ 5.12.1 Data reduction involves the distillation of raw data {nto a form that is usable for further analysis. Data reduction ‘may involve averaging multiple measurements, quantifying necessary conditions (e.g, steedy state), comparing with physical limits or expected ranges, and rejecting outlying ‘measurements 5.12.2 Dataanalysis involves the manipulation of reduced datato establish composite measures of interest, For exemple, series of temperature and flow rate data for some input and ‘output streams are combined using a physical law (conserva tion of energy) to estimate a rate of energy transfer in a pro- 13. Perform Final Uncertainty Analysis 5.13.1 A final uncertainty analysis is completed prior to reporting of data to account for unknown field effects and variances in instrument accuracy during the experiment. Sec- tion 6 contains additional details for the completion of the final uncertainty analysis. 5.14 Report Results 5.14.1. Reporting is the primary means of communicating the results from an experiment. Te report should be struc- tured to clearly explain the goals of the experiment and the evidence gathered to achieve the goals. Ibis assumed that data reduction, data analysis, and uncertainty analysis have pr ‘cessed all data to render them understandable by the intended audiences, 5.14.2. Different audiences require different reports with various levels of detail and background information. The dif- ferent reports that are often developed during the course of an experiment are described in the following sections. 5142.1 Experiment Notes intended for someone who is thoroughly familier with the project and can focus specifi cally on the evidence from the experiment. 5142.2 Progress Report details the current status, which may not properly present complete evidence on how the goals were attained, if indeed they were attained at the time of the progress report. 5.1423 Final Report pulls together information from all the progress reports. It therefore provides an interested reader notin direct touch withthe project a complete picture ‘of what was done and wiry. $142.4 Technical Papers are brief summaries of @ project, often distilled from the final report, to provide infor- tation about the project within the time allowed for its pre- sentation to an audience of interested partes (e., owners, operations and maintenance staf, researchers, et.) or its pub- lication ina relevant journal 5.143 All reports should include the results ofthe uncer- tainty analysis to an identified confidence level (typically 95% for HVAC&R applications). Uncertainty limits can be siven by 4nn (a one-significantfigure number in the same Units asthe results) or by 4nn% (a percentage of the results). S144 Graphical ond mathematical representations are often used, On graphs, error bars placed vertically and hori- zontally on representative points are a very clear way 10 present expected uncertainty. With mathematical expressions, Uncertainty limits can be given by nn or =nn% following each numerical coefficient. These limits arc especially vaku- able with regressions that are intended to allow the audience to predict results that were not measured directly 5.145 Annex Cpresents an example ofa short reportona. laboratory exercise that established the fixed and random errors in thermal resistances (R-values) of large (about 6m? in area) roof test sections when their R-values were measured in a particular climate simulator. SAS Other Considers 5.15.1 _Allengineering projects including design and exe- cution of experiments to acquire engineering data, require consideration of cost. Decisions must be made about the resources expended in planning, conducting, analyzing, and reporting an experiment 6. UNCERTAINTY ANALYSIS METHODS AND TECHNIQUES 6.1. The intent of this section is to provide an overview of the steps needed to perform an uncertainty analysis. Det ‘on completing an uncertainty analysis are contained in Annex A, with examples contained in Annex B, 62 Preliminary Uncertainty Analysis 62.1 A preliminary uncertainty analysis allows for better ‘experimental designs and should be done while planning an experimental program. If done carefully, it determines ‘whether itis feasible to pursue a particular experiment, what ‘measurements to make, which variables to vary, and what instrumentation is neoded 622 A preliminary uncertainty analysis involves the fol lowing steps: 1. Document the accepable level of uncertainty in the results (ee, accept 5% uncertainty in the results, providing 95% confidence). 2. Selecta certain typeof sensor or instrument type (eg. if flow is to be measured, the decision may be fo use a pitot ‘ube, paddle wheel, or an ultrasonic device. Atthis point do not differentiate between different manufacturers; only use generic types of instruments 3. Fined (bias) ertor can be neplected because a specific instruments not involved Document the random error due tothe instrumentation (the accuracy is #1% of ill scale between therange of 040-250 Us (0 0 4,000 gpm. 5. Develop the data reduction equations forthe quantity of interest (¢g, coeffcient of performance, capacity, or room enthalpy) 6. Calculate the uncertainty of the quantity of interest using propagation of eror equations, 7. Reselect instrumentation or refine data colletion/calcula- tion methodologies to obtain acceptable uncertainty (steps 26). 63 Final Uncertainty Analysis 63.1 A detailed uncertainty analysis is performed after the entire experiment has been completed and when the results ofthe experiments are to be documented or reported Te end results will have the same degree of confidence as ‘was ascribed to the individual measurements 63.2 The detailed uncertainty analysis is more laborious than the preliminary uncertainty analysis in the following respects: ‘+ Both the random and fixed errors have to be considered and the error propagation equations formulated accord ingly + The values ofthe random and fixed errors specific to the instrumenvsensor/data collection system have to be quantified from in-situ performance instead of using estimated values. 63.3 As pointed out by Coleman and Steele (1999), the fixed errors needed for the detailed uncertainty analysis are usually more difficult to estimate. Minimizing fixed errors cean be accomplished by careful calibration with referenced standards. 63.4 A final uncertainty analysis involves the following steps: 1. Estimate fixed (bias) error based upon instrumentation cali- bration results 2. Document the random error due to the instrumentation based upon instrumentation calibration results. 3. Calculate the uncertainty ofthe quantity of interest using propagation of error equations ASHRAE Guidtine 2.2005, 7. DATA VALIDATION 7A Data validation involves processes undertaken to ensure ‘that both individual data and ensembles of data ae reasonable and accurate. An important component of analyzing engi- neering experimental data is the process of validating both raw and processed data. With properly validated data, the probability of drawing incorrect conclusions is minimized. ‘There area numberof steps that can be taken to strengthen the process of data validation, This section presents three activi- ties useful in validating data: limit checks, independent checks, and techniques for dismissing outlier data. 72 Limit Cheeks. Fortunately, many of the measurements ‘made in conjunction with HVAC&R systems have identifi- able limits. Limits are useful in a number of experimental phases such as establishing a basis for appropriate instrumen- tation and measurement techniques, rejecting individual experimental observations, and bounding/bracketing mea- surements. Measurements can often be compared with one or ‘more ofthe following limits: physical, expected, and theoret- ical 72.1 Physical Limits. Appropriate physical limits should be identified in the planning phases of an experiment so they can be used to check the reasonableness of both raw and post- processed data, Under no circumstances can experimental observations exceed physical limits. 72.1.1 The following are some examples of physical limits that are frequently applicable in HVAC-related experi ‘ments 7241 Psychometrics + Dry-bulb temperature can never be less than the dew: point temperature or wet-bulb temperature, + Dew-point temperature can never exceed the wet-bulb temperature. + Relative humidity is always in the range between 0 and 100%. TAAA2 Refrigeration Systems ‘+ Refrigerant temperature should equal the saturation tem- perature based on pressure when two phases of pure refrigerant are present. + Reftigerant saturated condensing temperature should always be greater than the outdoor air dry bulb (for ai ‘cooled condensers) or outdoor air wet bulb (for evapor tive or water-cooled condensers) while operating in a cooling mode. + Reftigerant saturated evaporating temperature should always be lower than the entering air temperature (for refrigerant-based cooling coils) or entering water tem- perature (for chillers) while operating in a cooling mode. 7213 Hydronic Systems + Valve positions must be between 0% and 100%. ‘+ Pump head cannot exceed the dead-head condition for a ‘given pump, ASHRAE Guideline 22005 T2144 Secondary Systems + Adiabatically mixed air temperature and humidity ratio can never be greater than the relevant highest entering, stream value or lower than the relevant lowest entering. stream value (the mixed ar state will always lie along a line that conneets the two entering states). 72.1.2 Experimental observations or processed data that exceed physical limits should be flagged and closely scrutinized to determine the cause and extent oftheir devia- tion from the limits. The reason for data persistently beyond ‘physical limits is usually instrumentation bias or errors in data analysis routines/methods. 72.1.3 Data that infrequently exceed physical limits _may be caused by noise or other related problems. Resolving problems associated with observations that sporadicelly ‘exceed physical limits is often difficult. However, if they ‘occur, the experimental equipment and data analysis routines should be inspected and repaired. In situations where data ‘occasionally exceed physical limits, it is justifiable to purge such observations from the data set prior to undertaking any statistical analysis or testing of hypotheses. 7.2.2. Expected Limits. In addition to identifying phy: cal limits, expected upper and lower bounds should be iden- tified for each measured variable. During the planning phase of an experiment, determining expected ranges for measured variables facilitates the selection of instrumentation and mea- surement techniques. Prior to taking data, itis important to ensure thatthe measurement instruments have been calibrated and ae functional over the range of expected operation. Di ing the execution phase of experiments, the identified bounds serve as the basis for flagging potentially suspicious data, If individual observations exceed the upper or lower range of expected values, those points should be flagged and closely scrutinized to establish their validity and reliability. Another suspicious behavior is constant values when varying values are expected. Typically, this is caused by an incorrect ower or upper bound in the data reporting system so that limit values are being reported instead of actual values 72.2.1 The following are some examples of expected ranges of operation for quantities regularly measured in the HVAC field, 72214 Psychrometries + Outdoor air dry-bulb temperatures in a range spanning the site-specific ASHRAE design weather conditions (from a lower bound corresponding to the 99% condi tion to an upper bound corresponding to the 0.1% condi tion) + Indoor air dry-bulb temperatures in a range dictated by the controls for the specified equipment + Outdoor air wet-bulb temperatures in a range spanning. the site-specific ASHRAE design weather conditions (from 99% outdoor air dry bulb to the mean coincident ‘wet bulb atthe 0.1% condition on the high side) + Relative humidity in conditioned spaces in the range between 20% and 65% T2212 Refrigeration Systems ‘+ Reffigerant pressures in the range of limits established bby safety cut-out switches forthe compressor + High-side and low-side reftigerant pressures in a range dictated by operating controls on the reftigeration sys- tem + Power measurements for electric motors in a range from Oto approximately 120% of nameplate motor rating 722.13 Hydronic Systems + Volume flow rate in the range from 0% to 100% of design flow 72.214 Secondary Systems + Volume flow rates in a variable-volume secondary sys- tem ina range from 0% (or the design minimum volume flow rate) to 100% of the design volume flow rate 72.3. Theoretical Limits. The third measurement limit type is associated with theoretical values. Theoretical limits ‘may be related to physical properties of substances (e-., fluid freezing point), thermodynamic limits ofa subsystem or sys- tem (e.g, Camoteffciency fora vapor compression cycle), or thermodynamic definitions (e.g., heat exchanger effective- ness between zero and one). During the execution phase of experiments, theoretical limits can be used to bound measure- ‘ments, If individual observations exceed theoretical values, those points should be flagged and closely scrutinized to establish their validity and reliability. 7.3. Independent Checks (Mass and Energy Balances). In 1a number of cases, independent checks can be used as a ‘method to establish viability of data. Examples of indepen dent checks include comparison of measured (or calculated) values with those of other investigators (reported inthe pub- lished literature) and intra-experimental comparisons (based on component conservation principles). Only the intra-exper- imental comparisons will be discussed here. 7.3.1 _Intra-experimental comparisons involve collecting data and applying appropriate conservation principle as part Of the validation procedures. The most commonly applied ‘conservation principles used for independent checks include ‘mass and energy balances on components, subsystems, and systems. All independent checks should agree within the range of expected uncertainty of the quantities being com- pared. See Annex B for examples. 7A. Cheeks on Outliers. The nature and complexity of ‘making experimental measurements can occasionally lead to observations quite different from expected values. A number of diferent terms have been used to describe such observe- tions or datapoints, including “ier.” “stagger,” “spurious data point,” and “outlier.” The “production” of spurious data points or outliers i usually infrequent, and resulting values tend to clearly fall outside of expected bounds Percent Full Load Power ° 0 oo io Paond Reto) Figure 7-1 Stationary dataset including an outlier at part- ya ratio of 40%, 1085 10 coy ‘Outdoor Air Dry Bulb Temperature [F] 8 5 ss 7 on we Tie of Day ve a 2 34 Figure 7-2. Time-dependent data set with single outer near hour 15 3:00 p.m). 74.1 Outliers can be found in both stationary and time- dependent observations. Figures 7-1 and 7-2 illustrate out! ers in stationary and time-dependent datasets, respectively. 74.2 The root cause of spurious data points is often diffi- cult to pinpoint. However, possible causes include intermit. tent transducer connection(s), interference (electromagnetic ‘or electronic), software/computer interface interrupts, or ing components (transducers, multiplexers, data acquisition hardware, computer hardware), 7.4.3 To prevent biasing datasets and conclusions derived ‘rom their analysis, spurious data points or outliers need to be ‘identified and potentially discarded prior to further analysis of the data. Outliers should not be included in the process of estimating random (precision) errors or as part of uncertainty analyses. In analyzing large data sets, the limit checks described in section 7.2 should be used to sereen and flag potential outliers. These flagged outliers can then be sub: -ASHRAE Guideline 22005, TABLE 7-1 Chauvenet's Criterion for Rejecting Outliers ‘umber of Readings 2 3 4 154 3 165 6 1B 7 179 10 196 15 23 25 233, 30 257 100 281 300 3.16 500 329 1000) 3.48 Jected to more rigorous tests prior to rejecting them and purg- ing them from the dataset. 75 Chauvenet’s Criterion for Rejecting Data Points. A sed method that can be used as a basis for reject- is Chauvenet’s criterion. Chauvenet’ criterion states that a suspect data point or reading can be rejected ifthe probability of obtaining its particular deviation from the mean exceeds a specified threshold. To apply the criterion, a trial ‘mean and tral standard deviation are computed using all data points (including suspected points). Then the deviations of individual points from the mean are divided by the tral stan- dard deviation. All values that exceed the criterion values «given in Table 7-1 (or, alternatively, from Equation 7-1) can ’be rejected and removed from the dataset. After the data set has been purged of outliers, a new mean and standard devia- tion should be calculated, 4, 0.819 + 0544In(n)~0.02346In(n") rejection can be found in Anscombe (1960)? 8, REGRESSION ANALYSIS 8.1 Overview 1 This section provides a brief overview of regression ‘analysis for use in energy systems analysis, with an emphasis ‘on the development of regression models and assessment of | the model performance, Regression analysis isa formal tech- nique that can be used to examine data for the purpose ofiden- tifying and establishing meaningful relationships. The principles and practices of regression analysis are based :mainly on statistics. With the advent of widespread computer use, the application of regression analysis is becoming more ASHRAE Guideline 22005 important. In its simplest sense, applied regression analysis is ‘too! that can be used to generate a simple compact equation that describes the response ofa system, subsystem, or compo- rent. More sophisticated applications of regression analysis can help practitioners and researchers better understand the variables in a system that influence a given response (e.g., ‘how outdoor sir wet-bulb temperature affects cooling system COP), 8.1.2 In applying regression analysis principles, it is important to distinguish between two types of variables Independent variables, also called predictor variables, are variables that can be freely adjusted, floated, or otherwise var- ied (c.g, pressures, temperatures, or flow rates). Dependent ‘variables, also called response variables (e.g, capacity or efficiency), cannot be adjusted or floated. Regression analy- sis often is used to develop and understand the relationship ‘between the response variables and the predictor variables. 8.13 Ofparticular concer in using regression analysis in the context of energy systems analysis is the reliance on ‘unplanned dats, Unplanned data are data that arise from the continuous normal operation of a system and not from a spe- cifically designed experiment. Do not accept regression anal- ysis results without analyzing for the following two Unplanned data sources. 8.13.1 The first issue involves the situation where ‘model errors are not random due to the effects of variables that may not be measured but have an influence on the response (50 called latent variables). For example, for &cool- ing tower, suppose thatthe efficiency was calculated using an assumed airflow rate based on the fan curve. Due tothe influ- ence of wind speed and direction, the actual airflow through the tower may have varied, This changes the actual tower e ciency but docs not affect the calculated efficiency. 8.1.3.2 The second issueis associated with possible lim- its on specific predictor variables that may have a significant influence on the response of interest. If those variables do not span a wide range during data collection, the regression anal- ysis may suggest that they do not influence the response. This ‘may lead to an incorrect conclusion about the relationship ‘between the predictor variables and response. For example, for a cooling tower, the efficiencies may only have been mea- sured at part-load under a constant wet bulb temperature, leading to an unclear understanding of full-load operation or of the effect of changes in wet-bulb temperature, 8.2 Regression Model Development 82.1 In every energy system, it is possible to identify a number of different variables (predictor variables) that when changed have an influence on system performance (response variables). Since the functionality of energy systems follows ‘well-understood physical principles, a possible conclusion is that the development of models by statistical methods is ‘unwarranted. Although this is strictly true, the complexity of the system and uncertainty in all of the variables that influ- ence its operation (e.g., properties, geometry) are often not well known or characterized. In addition, the development of detailed physical models may nat be cost-effective to achieve the objective of the designer or practitioner. A simple regres- sion-based model can be viewed as a convenient means to organize masses of data that are often collected by an energy ‘management or data acquisition system, 8.22 Itis especially necessary in regression analysis that the governing laws of physics, mechanics, and thermodynam- jes be understood and thatthe limitations, assumptions, and ‘required application procedures for analysis methods be doc- ‘umented. In collecting data that can serve 8s predictor vari- ables, itis essential to recognize thatthe use of any resulting regression model should be limited to the range tha the vari- ables spanned. In other words, regression models are meant to be used for interpolation, not extrapolation. There is no general way to limit the use ofa regression model to ranges thatthe predictor variables spanned. It is recommended that the analyst report the range of each predictor variable used in the development of a regression equation (see Equation 8-5 for an example). 8.2.3 Most of the underlying statistical methods used in the development of regression models assume that the predic- tor variables are not subject to random variation but thatthe response is. In nearly all situations (especially where predic tor variables are experimental observations), this is not strictly tue. This assumption can be relaxed ifthe random Variation in individual predictor variables is small relative to the size of the predictor variables that form the basis for a regression model. Other assumptions and limitations will be presented in the sections that follow. 83 Types of Models 83.1 The first step in developing a regression model is to identify and collect a number of individual observations. The nomenclature in this guideline presumes that the response variable, ¥, correlates with one or more predictor variables, Xp XpXj, %y. Anumber of predictor response data “pairs” ‘then form the basis to establish a relationship (i.e, an equa- tion) between the response and the predictor variables, The resulting regression equation will provide predicted values, Y, given the input of predictor variables within the range used to-develop the original equation. However, duc to regressing to a single variable, inserting actual sets of individual predic- tor variables into the repression equation will result ina cal- culated predicted value, ¥, being slightly different from the actual observed value ofthe response, Y. The goal of regres- sion analysis isto make this difference (called the residual), c, as small as possible forall pairs of predicted and observed val- 83.2 Linear Models. Nearly all ofthe regression model- ing or analysis used in the area of energy systems can be sccomplished with linear models. In this context, a model is linear if itis tinear in the parameters 8.3.2.1 For example, Equation 8-1 shows a firstorder single variable regression that is linea for each of the param- eters by and by. Fe aprox 1) 83.2.2 Equation 8.2 is an example of a model that is nonlinear in its parameters fe Fr = age @2) where ay and a te parameters, However, notice that trans- formation of variables, accomplished by tsking the natural logarithm ofboth sides of Equation 8-2, results in a model for the natural logariti of (yin = In ay» ay that i inca in cach ofthe parameters Indy and 83.23 Themost widely used method to make the resid- uals as lose to zero as possible in lnear models isthe method af least squares. The method of lest squares produces esti- mates of the model parameters that minimize the sum of squares of all residuals, 5, 38 shown in Equation 83. S-rd= Sai (83) 83.2.4 A secondary goal is to achieve an acceptable degree of model accuracy with the fewest number of model parameters as possible. In fact, over-specifying the model parameters can lead to unacceptable results. Determining an acceptable degree of model accuracy is difficult and often bbecomes the modeler’s choice. The degree of accuracy achievable with a model is also dependent on the discipline For example, engineering-related models are often able to achieve a much higher degree of model fit (or accuracy) than biological system models. 83.2.5 There are several indicators that are commonly used for judging the performance of regression models. One dicator, Rs measure ofthe model's overall “goodness of fit.” R? will range between 0 and 1, with 0 implying that the predictor variables in the model had absolutely no success at Teflecting the actual response and 1 suggesting thatthe pre- dicted values from the model are indistinguishable from the actual observed response values. Other indicators, the t-value and F-ratio, are a measure ofthe usefulness of individual pre- dictor variables (and their corresponding parameters) and the overall regression equation, respectively. Relatively high val- ‘ues of individual parameter t-values or the overall F-ratio iply more usefulness in explaining the variation in the response. 83.2.6 Elaborating further on the mathematics behind the least squares method is beyond the scope ofthis guideline. ‘Those interested in more details on least squares estimation should consult a regression analysis text such as Draper and Smith (1981)? 833 Polynomial Models. Polynomial models in general assume thatthe response is only a function ofa single predic- tor variable but of possibly a higher order than a linear fune- tion. In determining statistical measures of the goodness of the regression, each power from 1 to n ofthe single predictor variable is considered an independent variable in the predic- tion. One of the best methods for developing a polynomial- ‘based model is to generate a plot of the response versus the predictor variable using all of the data available. The shape (Gtraight line, curve, ete.) of the response will provide some Indication of the polynomial degree required to achieve acceptable model accuracy. ASHIRAE Guide 2-2005 Py Soa 3040 50670 Bw 100 ‘opm Figure §-1 Example of heat transfer da 8.3.3.1. The simplest model to describe the behavior of a response variable, given observations of a single predictor Variable, isto assume that the response is & constant equal to the average of the observed responses. This “zero-order ‘model is almost always inadequate 83.3.2 Beyond the zero-order model, there isthe first- order single variable (or variate) model or linear model that is described by Equation 8-1 83.33 [faplotof the data appears to have some degree of curvature, a higher order model may be warranted. A gen- eral n! order polynomial is given by Equation 8-4, but yes tt" eH 8334 Rarely wll regression analysis in energy-related systems requite higher than second order (quadratic) polyno- tials. One caution in attempting to use higher order polyno- rials is that the higher order polynomials may lead to ‘unwanted behavior of the predicted values in the range ofits use (ie. wigely lines that have no resemblance to behavior expected between measurements) 8335 Consider the following example where the response is an overall heat transfer coefficient, U,and the sin- ale predictor variable is a flow rate, gpm. The fist step in considering the formulation of a single variable model is to plot the response, U, 8s function ofthe single-predictorvari- able, gom, as shown in Figure 8-1 83.3.6 It is clear that a zero-order model will not yield ‘ useful representation ofthe overall heat transfer coefficient over the entire range of flow rates. If the only interest (oF the only collected data) is fora flow rate range from 10 to 50 gpm, then it appears that a first-order model may be sufficient to predict the overall heat transfer coefficient. Using a least squares approach, Equation 8-5 details the results for predict- ing U given the one predictor variable, gpm, in this range. 2744+09598-gpm — (10 the dis cemnible pattems that may appear when plotting the residuals are what provide useful information. Figure 8-7a depicts an residuals Figure 8-7 Residuals exhibiting behavior that lustrates ‘the model is missing linear term, Figure 8-74 Residuals exhibiting behavior tha the model is missing a quadratic term. acceptable residual shape while Figures 8-7 b, ¢, and d show residual shapes with deficiencies, One of the more difficult situations to deal with isthe nonconstant variation shown in Figure 8-7c. It may be resolved using some sort of weighted least squares analysis 9. REFERENCES ‘Coleman, H.W., and W.G Steele, Experimentation and Uncertainty Analysis for Engineers, 2d ed. (New York John Wiley, 1999). Anscombe, FJ, “Rejection of Outliers’ Vol. 2, No.2 (1960). Spraper, N., and H. Smith, Applied Regression Analysis, 2d ed. (New York: John Wiley & Sons, 1981). Technometries, ASHRAE Guideline 2.2005 (This annex is not part of this guideline. It is merely informative and does not contain requirements necessary for conformance to the guideline.) ANNEX A UNCERTAINTY ANALYSIS/PROPAGATION OF ERROR ‘The Intemational Organization of Standardization (ISO) and six other organizations, including the American Society of Mechanical Engineers, have published guides (e.g, ANSU ASME Standard PTC 19.1-1988!) that have established the international experimental uncertainty standard. It is this international uncertainty analysis standard that is described below. AA Need for Uncertainty Analysis. Any measurement exhibits some difference between the measured value and the ‘true value and therefore has an associated uncertainty. A statement of measured value without an accompanying uncer- tainty statement has limited meaning. Uncertainty is the inter- ‘val around the measured value within which the true value is ‘expected to fall with some stated confidence. The term “good data” does not describe data that yield the desired answer. I¢ describes data that yield a result within an acceptable uncer- tainty interval or, in other words, provide the acceptable degree of confidence in the result. ‘Measurements made inthe field are especially subject to potential errors. In contrast to measurements made under the controlled conditions of a laboratory setting, field measure- ‘ments are typically made under less predictable circumstances and with less accurate and less expensive instrumentation. Furthermore, field measurements re vulnerable to errors aris- ing from the following factors: + Variable measurement conditions so that the method ‘employed may not be the best choice for all conditions ‘+ Limited instrument calibration, because field calibration is typically more complex and expensive than labora- tory calibration ‘+ Simplified data sampling and archiving methods ‘+ Limitations inthe ability to field-adjust instruments With appropriate care, many ofthese sources of error can ‘be minimized through the systematic development ofa proce- dure by which an uncertainty statement can be ascribed tothe result and through the optimization of the measurement system to provide maximum benefit fr the least cost. The results of practitioner who does not consider sources of eror are likely to be questioned by others, especially since the HVAC&R community is increasingly becoming sophisticated about the proper reporting of measured data A2 Basic Uncertainty Concepts: Random and Fixed Errors. The degree of inaccuracy of a measurement is quan- tified sing Equation A- (ay Bre, where: 8, {otal measurement error ASHRAE Guideline 2.2005 B fixed (bias) error & random (precision) error, and {= asubseript to denote a particular experiment or reading Fixed (bias) errors are the constant deviations that are typically the hardest to estimate or document. They include such items as miscalibration and improper sensor placement. Biases are essentially offsets from the true value, which are constant over time and do not change when the number of observations is increased. For example, a bias is present if temperature sensor always reads 1°C higher than the true value from a certified calibration procedure. Random (precision) errors are random differences from ‘one observation tothe next duc to both sensor noise and extra- neous conditions affecting the sensor. The random error ‘changes from one observation to the next, but its mean (aver- ‘age value) over a very large number of observations approaches zero. Random error generally has a well-defined probability distribution that can be used to bound its variabil- ity in statistical terms when a finite number of observations is ‘made of the same variable, Based on measurements of variable , it can be speci- fied thatthe true value of X lies in the interval Xpexe#U) where Xjqi is usually the mean value of the measurements taken and Uy is the uncertainty in X that corresponds to the estimate ofthe effects ofthe combination of fixed and random ‘The uncertainty being reported is specific toa confidence level. The confidence level defines the range of values thatcan bbe expected to include the true value with a stated probability. For example, a statement thatthe 95% confidence limits are 5.1 108.2 implies thatthe true value will be contained between the interval bounded by 5.1 and 8.2 in 19 out of20 predictions (95% of the time), or that we are 95% confident that the true value lies between 5.1 and 8.2 ‘An uncertainty statement with a low confidence level is usually meaningless. For the example in the previous para- ‘raph, if. confidence level of 40% is used instead of a 95% confidence level, the confidence interval becomes a tight 7.6 to 7.7. However, only 8 out of 20 predictions will likely lie between 7.6 and 7.7. Conversely, itis useless to seek a 100% confidence level. This always means that the true value of some quantity lies between plus and minus infinity ‘A3- Uncertainty of a Measured Variable ABA General Considerations. Muli-sample data (cepeated measurements of fxed quantity using altered tet conditions, such sdferent observers or diferent instrumen- tation or both) provides greater reliability and precision than single-sample data (measurements by one person, using asin- gle instrument). For the majority of engineering eases, itis impractical and too costly to perform a true mult-sample experiment. Although, strictly speaking, merely taking repeated readings with the same procedure and equipment oes not provide mult-sample results, such a procedure yp- ically provides a fair approximation of mult-sample exper- iment. A32_ Sample Size Considerations for Random Uncer- tainty. Depending upon the sample size of the data (greater than or less than about 30 samples), diferent statistical con- siderations and equations apply. A324 Random Uncertainty in Large Samples (more than about 30). The best estimate of a variable is usu: ally its mean value. The mean value for a variable is caleu- lated using the set of readings for that variable, as shown in Equation A-2. aa where x ‘mean value of ll readings » sample size ofthe readings taken X= an individual reading ‘The best estimate's worth as an estimate ofthe true value ofthe variable must be determined by constructing the limits ofthe confidence interval. Thisis accomplished by calculating te standard deviation (more eorrecty, the population sta dard deviation) ofthe variable using Equation A-3 (a3) where S, = standard deviation and he other terms areas defined in Equation 2. “The typical procedure is to then assume thatthe individ- ual data values are scattered about the mean following a certain probability distribution function, within 22 standard deviations of the mean. Usually a normal probability eure (Gaussian distribution) i assumed to represent the dispersion in experimental data. For anormal distribution, the standard deviation indicates the following degrees of dispersion ofthe values about the mean: Zo1: 68.3% of the data will be within +15, of the mean 2-2: 95.5% of the data will be within 42S, of the mean, and Z-3: 99.7% of the data will be within 43S, ofthe mean, Thus, the confidence interval for the mean of multi- sample random data, ie, data that do not have any fixed error is shown in Equation A-4. Xnin -(—=) and =) (a4) ru 2 ¥-(@) ) ay whee = entiation by Exton 43 sd 2 = ‘mmberofsandad deviations out emer, “which implies the confidence level. Forexample, if Z=2, there is 95.5% probability that the correct value lies in the interval between ty in Small Samples (less ‘than about 30). In many circumstances, the engineer will not be able to collect a large mumber of data points and may be limited to data set of less than 30 values (n<30). Under such conditions, the mean value is computed as before, while the standard deviation and the Z value have to be modified. A ‘more accurate relation forthe standard deviation (called the sample standard deviation) is shown in Equation A-S. As) ‘The Z value applicable for the normal distribution cannot, bbe used for small samples. The new values, called t-values, are tabulated for different degrees of freedom F(=n— 1) and for the acceptable degree of confidence. The confidence interval forthe mean value of X, when no fixed (bias) erors are present in the measurements, i given by Equation A-6, 18) y Statstis handbooks and textbooks give tabulations of values for various levels of confidence and degrees of fee- dom. The following truncated list of values for 95% coni- dence with a twossided dstibution as 8 funtion of the numberof degrees of feedom, nis taken from Coleman and Steele (1999)? 1S; Faw E(B) mt Fa CS) 30 208 20 209 1s 213, 10 223, 8 231 6 245, 4 278 2 430 A33 Fixed Uncertainty. Estimating the fixed (bias) crtorata specified confidence level (e.g, 95% confidence) is described below (Coleman and Steele 1999).? The fixed error By fora given value X is assumed to be a single value drawn from some larger distribution of possible fixed errors. The ‘treatment is similar to that of random errors with the major difference that only one value is assumed even though several observations may be taken, Lacking further knowledge, anor- ‘mal distribution is usually assumed forthe larger distribution Hence, if a manufacturer specifies thatthe fixed uncertainty ASHRAE Guideline 2.2005, By is 41.0°C with 95% confidence (compared to some stan- dard reference device), then one assumes that the fixed error belongs to a larger distribution (Gaussian) with a standard deviation $3 = 0.5°C (assuming the corresponding t-value to bbe 2.0) The value By is all that is needed. A34 Overall Uncertainty. The overall (combined) ‘uncertainty of a measured variable X has to combine the ran- tye (T2-T) (cold side) (8-20) Op = MaepTy-Te)—hotside 621) ‘The heat exchange to the cold stream, Q,, should equal that from the hot stream, Oy, within the range oF uncertainty of| ‘each quantity. The propagation of eror relations based on the energy balances forthe cold and hot sides are: + Seed Ge coy = Bion) (Sle) (Cur (C2 ‘The resulting expressions for uncertainty forthe cold and hot sides are given by: = GG) feel lal (B-24) QY-G+Es a) leral le ‘Table B-1 lists the values of uncertainty associated with «each ofthe measured values onthe heat exchanger. (Note: The specific heat is not measured but the tabular values are assumed to be accurate to within 5% of the rue value.) TT (B25) ASHRAE Guideline 22005 7,737,184 ls 6,750,000 2 2 5,140,000 Cold-Side 4,304,246 Figure B-4 “Measured” hot-sde and cold-side heat exchanger loads including respective uncertanties asa range. TABLE B-1 Parameters and Uncertainties Used in the Example Heat Exchanger Parameter _| Nominal Value | _Uncertainty CP 09 Buh 35% ty 450,000 10% % 55°F °F T aor °F, ‘The resulting energy balance and corresponding uncer- , tnt forthe old side ofthe heat exchanger are: 2, = myery(T,-T) ess 09 B20 = 51a ot (B-26) Heep 2 2 °. a (225) = con? (0057+ [EP fhe @ lara tal 216.2% - ‘The energy balance and corresponding uncertainty forthe hot side ofthe heat exchanger are: Oy = riseplTs- Ta) Ib Btw Btu 450, 00052 1 GET (55—40)F = 6,750, 0008 (8-28) a Since the hot-side and cold-side energy balances are not identical, itis natural to question whether or not the energy balances are in reasonable agreement. Uponcloser inspection, itisconeluded thatthe measured loads for the hot side and cold side of the heat exchanger overlap when their respective uncertainties are included. Figure B-4 illustrates that conclu (This annex is not part of this guideline. It is merely informative and does not contain requirements necessary for conformance to the guideline.) ANNEX REPORTING OF RESULTS [This annex is extracted from a research report on the calibration of a guarded hot box that is used t0 measure the thermal resistance (R-value) of full-scale roof assemblies. The urpose of the annex is to illustrate how results with uncer- tainty are reported.) ‘To establish the random error (precision) and fixed error (bias) ofthe thermal resistances of large roof assemblies tested in a device known as the Large Scale Climate Simulator at 2 USS. national laboratory, 37 steady-state tests were carried out vera petiod of nearly two years. The calibration test panel was 12 10.16-m-thick layer of expanded polystyrene (EPS) foam insulation. Itwas placed between a climate chamber above and ‘metering and guard chambers below. The metering chamber is in the form of an open box with its walls and floor exposed to the guard chamber. The tops of the walls were tightly sealed agains the botom of the test panel. The EPS thermal conduc- tivity was measured independently, with a probable error of| +£1.2%, and fit the expression in Equation C-1 gpg = 003361303 «10 *Typcay cy where eps = thermal conductivity ofthe expanded polystyrene, Wikmnky, Toon arithmetic average ofthe outer surface temperatures, °c. ‘The accepted thermal resistance or R-value of the test panel is its effective thickness divided by kp from Equation C-l. ‘The purpose of the tests was to compare accepted R- values to R-values determined by measurements at steady- state in the Large Scale Climate Simulator. The expression to determine these R-values from the measurements is Ay, ge MTD Qyanet © where R [R-value of the test panel, m?-K/W, A= fective ares ofthe panel exposed ote metering chamber, m?, average temperature atthe bottom of the panel, r average temperature atthe top ofthe panel, °C, and pane! * eat flow rate through the pane into the metering ‘chamber, W. The effective area ofthe panel must be estimated because the 11.1-cm-thick walls of the metering chamber are in ther ‘mal contact with the bottom of the test panel. The lower bound ‘on this area is 5.95 m:, the open area of the metering chamber. The upper bound, assuming symmetry due to balanced temperatures inside and outside the metering chamber walls, is 6.50 m?, the arca measured tothe centerlines of the metering chamber walls. The lower bound applies to very thin test panels, the upper bound to very thick panels. The effective area for panels of intermediate thickness must lie between the bounds. An estimate of the effective area using a wo-dimen- sional heat conduction model for the heat flow through a seetion of panel over a metering chamber wall, yielded 6.45 ‘Thermocouples for measurements of temperatures 7, and , were placed on the bottom and top surfaces of the panel, respectively, so that a total of 25 thermocouples were equally spaced on each surface over the 5.95 m? open area of the ‘metering chamber below the panel. The heat flow rate was determined by an energy balance on the metering chamber: an Quatt* ear * Qrpecimen ™ 0 3) where Qu = energy input from circulation fans, measured by voltage times current, Qyou = Sum ofthe heat gains forthe four walls andthe floor of the metering chamber. ‘This heat flows through 11.1. em of polyisocyanurate foam insulation from the guard chamber surrounding the metering chamber. The guard chamber is kept at approxi- ‘mately the same average temperature as the metering cham- ber. Heat flow for each of the four walls and floor is determined using Equation C-2 rearranged to find heat flow. For each component, the average temperature difference is ‘measured by six differential thermocouples, the surface areais known, and the thermal resistance of the polyisocyanurate foam was measured independently. rear" enerey input from resistance heaters, measured by voltage times curent et heat flow rate through the pane}, found with Equation C3 rpeciner Both surfaces of the test panel were painted to protect ‘them since the panel was intended for reuse to check the meter- ing chamber energy balance regularly. After painting, ther- 'mocouple beds for surface temperature measurements were pressed into the surfaces of the insulation so that the beads ‘were flush with the outer surfaces. In the absence of compli- cations, the effective thickness of the panel is the average distance between thermocouple beads from the bottom to the ASHRAE Guideline 2.2005 TABLE C-1 Precision and Bias for Tests on a 10.16-em-Thick Layer of Expanded Polystyrene Foam Thickness 10.16 em ‘Thickness~9.96 em UR 95% Regression Confidence Known Known Resistance, Interval Resistance, | Bias, | Resistance, | Bias Mean Temp,°c | _M@Kw % m@k/w % | wexw | % 346 2.90 +19 3.06 32, 3.00, 33 691 278 H19 294 a 2.88 33 2336 2.60 322 27 “62 271 43 Figure C-1 Thermal resistance vs. mean temperature for 10.16-cm-thick layer of expanded polystyrene foam. top of the panel. One possible complication is thatthe paint could penetrate into the foam beyond the thickness ofthe ther- mocouple beads. Figure C-1 shows a plot of 37 R-values forthe 10,16-cm- thick panel versus the mean temperatures at which they were ‘obtained, The mean temperatures are the arithmetic average of the temperatures measured on the top and bottom of the panel foreach test. The symbols represent the results ofthe 37 tests using Equation C-2 to determine the R-value from temperature and heat flow rate measurements, ‘Theol line plots R-values from the linear regression of| the measured R-values with mean temperature. Stictly speaking, the conductances (the reciprocal of the R-values) are directly proportional to the thermal conductivity of the EPS and are expected, ike Equation C-1, to be a linear func- tion of temperature. Over the temperature range shown in Figure C-1, there is no significant difference between the R and U/R fits. The /R fitis better for extrapolations beyond the measured temperatures. The dashed lines are the limits ofthe 95% confidence interval for predicting an individual R-value ‘rom the linear regression of the R-values. The legend inthe figure gives the equations of the regression and the confidence interval limits, The coefficient of determination (R2) of the regression is 0.936. The residuals showed no systematic vari- ation with mean temperature or with the order in which tests were performed. ASHRAE Guideline 22005 ‘Table C-1 shows precision (as 95% confidence intervals) and bias for the R-values of the 10.16-cm-thick layer of ‘expanded polystyrene foam at low, mid-range, and high mean temperatures. The precision ranges from 1.9 to 422%, which is essentially constant. Two sets of numbers are given forthe bias: the first uses the full 10.16 em thickness of the foam, and the second uses a thickness of 9.96 em. These results are summarized in Table C- ‘The latter thickness takes into account thatthe thermo- couple beads were about 0.10 cm in diameter and were pressed into the surfaces of the foam. Known R-values are obtained by dividing these respective thicknesses by the ther- mal conductivity from Equation C-1. With 10.16 em thick- ness, the measured resistances are 5.2% to 6.2% lower than the ‘known values; with 9.96 cm, the measured values are 3.3% to 4.3% lower. Using the bias for 9.96 em and the +1.9% to 22.2% range for precision, overall uncertainty by the RSSE formula is 3.8% to 24.8%, ‘Accounting fr the penetration ofthe thermocouple beads is astepin the right direction. Possible causes forthe remai ing bias include the following: bias in measuring temperature difference or heat flow rate, eror in the effective heat flow area, and error in the thermal resistance used as the known value. The biasis relatively independent of mean temperature, ‘whereas temperature differences and heat flow rate are larger at low mean temperatures and therefore more accurate at low ‘mean temperatures. The heat flow area needed to account for ‘the remaining bias exceeds the upper bound area for thick specimens discussed above, Therefore, the most likely cause of the remaining bias is thought to be the thermal resistance ‘used asthe known value, found from the thickness divided by the thermal conductivity, Regarding the thermal conductivity, a thin-heater appa- ratus was used fo measure the thermal conductivity of one specimen of the expanded polystyrene foam in 1988 and again {in 1991, with results that agreed to within 0.5%, With two ‘other specimens from the same lot of material, a round robin ‘was conducted that involved four laboratories and three differ cent types of apparatus (thin-heater, heat flow meter, and ‘guarded hot plate apparatus). Thermal conductivities from this round robin ranged from 7.1% lower to 0.6% higher than that given by Equation C-1. Use ofthese conductivities would generally make the bias larger. It is concluded thatthe bias isnot likely due entirely to ‘an uncertainty in the thermal conductivity of the foam. Error in assigning the thickness of the panel also contrib- ues. If, on both sides of the foam, the paint penetrated 2.85 mm, which is roughly three-fourths the diameter of a 2s bead of expanded polystyrene, the effective thickness ofthe foam would be 9.59 em, not the 9.96 em between thermo: couple beads. The average bias would be zero, Six ‘measurements on an unpainted panel, made from EPS from the same ot as the painted panel but with the thermocouple ‘beads taped to the unpainted surface, showed bias that was small and bidirectional, varying from -1.3% to +2.4%, Combining this range with the +1.9% to #2.2% range for precision, overall uncertainty decreases to 42.3% to 43.3%, In summary, a series of experiments was performed to establish the precision and bias of the Large Seale Climate ‘Simulator. The primary calibration test panel was inthe form of 10.16-cm-thick layer of expanded polystyrene foam insu- lation with thermocouples imbedded in its top and bottom surfaces and both surfaces painted. For mean panel tempera- tures from ~S°C to 23°C, R-values were obtained that had limits of precision that ranged from-=1,9% to 2.2% with 95% ‘confidence. The bias was consistently negative, ranging from ~4.3% to -3.3% if corrections were made only for penetration Of the thermocouple beads into the surfaces. Combining the precision and bias by the RSSE formula, overall uncertainty ‘was less than 35% with 95% confidence. Further measure- ‘ments using a panel with neither painted surfaces nor thermo- couple beads pressed into the surfaces lowered the uncertainty to about =3% with 95% confidence. (This annex is not part of this guideline. It is merely informative and does not contain requirements necessary for conformance to the guideline.) ANNEX D BIBLIOGRAPHY Anscombe, FJ, “Rejection of Outliers,” Technometrics, Vol 2,No.2 (1960). Beckwith, IG, N.L. Buck, and R.D. Marangoni, Mechani- cal Measurements, 34 ed. (Reading, Massachusetts: ‘Addison-Wesley, 1961), Coleman, H.W. and W.G Steele, Experimentation and Uncertainty Analysis for Engineers, 2d ed. (New York John Wiley, 1999), D'Albora, E.G, and K.L. Gillespie, 1999, “Evaluating the Uncertainty in Cool Storage Inventory Using an Energy Balance Method,” ASHRAE Transactions, vol. 105, pt. 2, paper SE-99-16-2. Draper, N., and H. Smith, Applied Regression Analysis, 24 ed. (New York: John Wiley & Sons, 1981). Fuller, W.A., Measurement Error Models (New York: John Wiley, 1994), Hald, A., Statistical Theory with Engineering Applications (New York: John Wiley, 1952), 150, Applied Guide to the Expression of Uncertainty in Mea ‘surements (Geneva: Intemational Organization for Stan- dardization, 1993). Moffat, R.J., “Describing the Uncertainties in Experimental Results," Experimental Thermal and Fluid Science, 1 (1988), pp. 317. Reddy, T.A., J.S. Haberl, and 4.S. Elleson, “Engineering Uncertainty Analysis in the Evaluation of Energy and Cost Savings of Cooling System Alternatives Based on Field Monitored Data,” ASHRAE Transactions, Vol. 105, pt. 2, paper SE-99-16-01, June 1999, Reddy, T.A.,J.K. Kissock, and D.K. Ruch, “Uncertainty in Baseline Regression Modeling and in Determination of Retrofit Savings,” ASME Journal of Solar Energy Engi- neering, Vol. 120, p.185, August 1998. Schenck, H. Theories of Engineering Experimentation, 24 ced. (New York: McGraw-Hill, 1968. ‘Theil, H., Principles of Econometrics (New York: John Wiley, 1971), ASHRAE Guideline 22005 POLICY STATEMENT DEFINING ASHRAE’S CONCERN FOR THE ENVIRONMENTAL IMPACT OF ITS ACTIVITIES ASHRAE is concerned withthe impact ofits members’ activities on both the indoor and outdoor environment. ASHRAE’s ‘members will strive to minimize any possible deleterious effect on the indoor and outdoor environment of the systems and ‘components in their responsibilty while maximizing the beneficial effects these systems provide, consistent with accepted standards and the practical state ofthe ar. ‘ASHRAE's short-range goal is to ensure thatthe systems and components within its scope do not impact the indoor and ‘outdoor environment to a greater extent than specified by the standards and guidelines as established by itself and other responsible bodies. ‘As an ongoing goal, ASHRAE will, through its Standards Committee and extensive technical committee structure, continue to generate up-to-date standards and guidelines where appropriate and adopt, recommend, and promote those new and revised standards developed by other responsible organizations ‘Through its Handbook, appropriate chapters will contain up-to-date standards and design considerations as the material is ‘systomaticaly revised ‘ASHRAE wil tke the lead with respect to dissemination of environmental information of is primary interest and will sek ‘out and aisseminate information from other responsible organizations that is pertinent, as guides to updating standards and ‘guidelines. ‘The effects of the design and selection of equipment and systems will be considered within the scope of the system's intended use and expected misuse. The disposal o hazardous materials, i any, wil alo be considered. ASHRAE's primary concem for environmental impact will be at the site where equipment within ASHRAE's scope operates. However, energy source selection and the possible environmental impact due to the energy source and energy transportation will be considered where possible. Recommendations concerring energy source selection should be made by Its members 86803 PC 3/05

You might also like