Transcript Document
Summary of Experimental Uncertainty Assessment Methodology F. Stern, M. Muste, M-L. Beninati, W.E. Eichinger 7/18/2015 1 Table of Contents Introduction Test Design Philosophy Definitions Measurement Systems, Data-Reduction Equations, and Error Sources Uncertainty Propagation Equation Uncertainty Equations for Single and Multiple Tests Implementation & Recommendations Introduction Experiments are an essential and integral tool for engineering and science Experimentation: procedure for testing or determination of a truth, principle, or effect True values are seldom known and experiments have errors due to instruments, data acquisition, data reduction, and environmental effects Therefore, determination of truth requires estimates for experimental errors, i.e., uncertainties Uncertainty estimates are imperative for risk assessments in design both when using data directly or in calibrating and/or validating simulation methods Introduction Uncertainty analysis (UA): rigorous methodology for uncertainty assessment using statistical and engineering concepts ASME (1998) and AIAA (1999) standards are the most recent updates of UA methodologies, which are internationally recognized Presentation purpose: to provide summary of EFD UA methodology accessible and suitable for student and faculty use both in classroom and research laboratories Test design philosophy Purposes for experiments: Type of tests: Science & technology Research & development Design, test, and product liability and acceptance Instruction Small- scale laboratory Large-scale TT, WT In-situ experiments Examples of fluids engineering tests: Theoretical model formulation Benchmark data for standardized testing and evaluation of facility biases Simulation validation Instrumentation calibration Design optimization and analysis Product liability and acceptance Test design philosophy Decisions on conducting experiments: governed by the ability of the expected test outcome to achieve the test objectives within allowable uncertainties Integration of UA into all test phases should be a key part of entire experimental program Test description Determination of error sources Estimation of uncertainty Documentation of the results Test design philosophy DEFINE PURPOSE OF TEST AND RESULTS UNCERTAINTY REQUIREMENTS SELECT UNCERTAINTY METHOD - DESIGN THE TEST DESIRED PARAMETERS (C D, C R,....) MODEL CONFIGURATIONS (S) TEST TECHNIQUE (S) MEASUREMENTS REQUIRED SPECIFIC INSTRUMENTATION CORRECTIONS TO BE APPLIED DETERMINE ERROR SOURCES AFFECTING RESULTS YES ESTIMATE EFFECT OF THE ERRORS ON RESULTS IMPROVEMENT POSSIBLE? NO UNCERTAINTY ACCEPTABLE? NO YES NO NO TEST IMPLEMENT TEST START TEST RESULTS ACCEPTABLE? NO YES YES NO CONTINUE TEST PURPOSE ACHIEVED? MEASUREMENT SYSTEM PROBLEM? SOLVE PROBLEM YES ESTIMATE ACTUAL DATA UNCERTAINTY - DOCUMENT RESULTS REFERENCE CONDITION PRECISION LIMIT BIAS LIMIT TOTAL UNCERTAINTY Definitions Accuracy: closeness of agreement between measured and true value Error: difference between measured and true value Uncertainties (U): estimate of errors in measurements of individual variables Xi (Uxi) or results (Ur) obtained by combining Uxi Estimates of U made at 95% confidence level Definitions Bias error : fixed, systematic k+ 1 Bias limit B: estimate Precision limit P: estimate of Total error: = + k+ 1 X X true Xk X k+ 1 (a) two readings FREQUENCY OF OCCURRENCE random k k of Precision error : = total error = bias error = precision error X X true MAGNITUDE OF X (b) infinite number of readings Measurement systems, data reduction equations, & error sources Measurement systems for individual variables Xi: instrumentation, data acquisition and reduction procedures, and operational environment (laboratory, large-scale facility, in situ) often including scale models Results expressed through data-reduction equations r = r(X1, X2, X3,…, Xj) Estimates of errors are meaningful only when considered in the context of the process leading to the value of the quantity under consideration Identification and quantification of error sources require considerations of: Steps used in the process to obtain the measurement of the quantity The environment in which the steps were accomplished Measurement systems and data reduction equations Block diagram showing elemental error sources, individual measurement systems, measurement of individual variables, data reduction equations, and experimental results ELEMENTAL ERROR SOURCES 1 2 J INDIVIDUAL MEASUREMENT SYSTEMS X 1 B ,P X 2 B ,P X J B,P MEASUREMENT OF INDIVIDUAL VARIABLES 1 1 2 2 J r = r (X , X ,......, X ) 1 2 J r B, P r r J DATA REDUCTION EQUATION EXPERIMENTAL RESULT Error sources Estimation assumptions: 95% confidence level, large-sample, statistical parent distribution MODEL FIDELITY AND TEST SETUP: - As built geometry - Hydrodynamic deformation - Surface finish - Model positioning TEST ENVIRONMENT: - Calibration versus test - Spatial/temporal variations of the flow - Sensor installation/location - Wall interference - Fluid and facility conditions CONTRIBUTIONS TO ESTIMATED UNCERTAINITY SIMULATION TECHNIQUES: - Instrumentation interference - Scale effects DATA ACQUISITION AND REDUCTION: - Sampling, filtering, and statistics - Curve fits - Calibrations Uncertainty propagation equation Bias and precision errors in the measurement of Xi propagate through the data reduction equation r = r(X1, X2, X3,…, Xj) resulting in bias and precision errors in the experimental result r A small error (Xi) in the measured variable leads to a small error in the result (r) that can be approximated using Taylor series expansion of r(Xi) about rtrue(Xi) as r = r ( X i ) rtrue ( X i ) = X i dr dXi The derivative is referred to as sensitivity coefficient. The larger the derivative/slope, the more sensitive the value of the result is to a small error in a measured variable Uncertainty propagation equation Overview given for derivation of equation describing the error propagation with attention to assumptions and approximations used to obtain final uncertainty equation applicable for single and multiple tests Two variables, kth set of measurements (xk, yk) r = r ( x, y) k xk = xtrue + xk + xk xtrue y x k x = r x; y = r y k y xk r xk xtrue + r yk ytrue + R2 x y r = rk rtrue = x x + x + y y + y k r k rtrue 1 sensitivity coefficients k k yk r k = r (xk , yk ) The total error in the kth determination of r k x k yk = ytrue + yk + yk rk rtrue = y x k rk rk ytrue Uncertainty propagation equation We would like to know the distribution of r (called the parent distribution) for a large number of determinations of the result r A measure of the parent distribution is its variance defined as 2 r 1 N 2 = lim rk N N k =1 2 Substituting (1) into (2), taking the limit for N approaching infinity, using definitions of variances similar to equation (2) for ’s and ’s and their correlation, and assuming no correlated bias/precision errors 2 = x2 2 + y2 2 + 2 x y + x2 2 + y2 2 + 2 x y r x y x y x x x y 3 ’s in equation (3) are not known; estimates for them must be made Uncertainty propagation equation Defining uc2 estimate for 2r bx2 , by2 , bxy2 estimates for the variances and covariances (correlated bias errors) of the bias error distributions S xw , S y2 , S xy estimates for the variances and covariances ( correlated precision errors) of the precision error distributions equation (3) can be written as uc2 = x2bx2 + y2by2 + 2 x ybxy + x2 Sx2 + y2 S y2 + 2 x y Sxy Valid for any type of error distribution To obtain uncertainty Ur at a specified confidence level (C%), a coverage factor (K) must be used for uc: U r = Kuc For normal distribution, K is the t value from the Student t distribution. For N 10, t = 2 for 95% confidence level Uncertainty propagation equation Generalization for J variables in a result r = r(X1, X2, X3,…, Xj) J 1 J J J 1 J J U = B + 2 i k Bik + P + 2 i k Pik 2 r i =1 2 i 2 i i = i =1 k =i +1 i =1 r X i 2 2 i i i =1 k =i +1 sensitivity coefficients Example: CD = D = CD D, ,U , A 2 1 2 U A J U 2 CD J = B + i2 Pi 2 2 i i =1 2 i i =1 2 C C C C = D BD2 + PD2 + D B2 + P2 + D BU2 + PU2 + D BA2 + PA2 D U A 2 2 2 Uncertainty equations for single and multiple tests Measurements can be made in several ways: Single test (for complex or expensive experiments): one set of measurements (X1, X2, …, Xj) for r According to the present methodology, a test is considered a single test if the entire test is performed only once, even if the measurements of one or more variables are made from many samples (e.g., LDV velocity measurements) Multiple tests (ideal situations): many sets of measurements (X1, X2, …, Xj) for r at a fixed test condition with the same measurement system Uncertainty equations for single and multiple tests The total uncertainty of the result U r2 = B2r + P2r 4 Br : same estimation procedure for single and multiple tests Pr : determined differently for single and multiple tests Uncertainty equations for single and multiple tests: bias limits Br : J 1 J J B = B + 2 i k Bik Sensitivity coefficients 2 r i =1 2 i 2 i i =1 k =i +1 r X i i = B : estimate of calibration, data acquisition, data reduction, conceptual i bias errors for Xi.. Within each category, there may be several elemental sources of bias. If for variable Xi there are J significant elemental bias errors [estimated as (Bi)1, (Bi)2, … (Bi)J], the bias limit for Xi is calculated as J 2 k =1 k B = Bi 2 i B : estimate of correlated bias limits for X and X ik i L Bik = Bi Bk =1 k Uncertainty equations for single test: precision limits Precision limit of the result (end to end): Pr = tSr t: coverage factor (t = 2 for N > 10) Sr: the standard deviation for the N readings of the result. Sr must be determined from N readings over an appropriate/sufficient time interval Precision limit of the result (individual variables): J Pr = ( i Pi )2 i=1 Pi = ti Si the precision limits for Xi Often is the case that the time interval is inappropriate/insufficient and Pi’s or Pr’s must be estimated based on previous readings or best available information Uncertainty equations for multiple tests: precision limits The average result: 1 r= M M r k k =1 Precision limit of the result (end to end): M rk r 2 S r = M 1 k =1 tS r Pr = M t: coverage factor (t = 2 for N > 10) Sr : standard deviation for M readings of the result The total uncertainty for the average result: U = B + P = B + 2 Sr 2 r 2 r 2 r 2 r M 2 Alternatively Pr can be determined by RSS of the precision limits of the individual variables 1/ 2 Implementation Define purpose of the test Determine data reduction equation: r = r(X1, X2, …, Xj) Construct the block diagram Construct data-stream diagrams from sensor to result Identify, prioritize, and estimate bias limits at individual variable level Uncertainty sources smaller than 1/4 or 1/5 of the largest sources are neglected Estimate precision limits (end-to-end procedure recommended) Computed precision limits are only applicable for the random error sources that were “active” during the repeated measurements Ideally M 10, however, often this is no the case and for M < 10, a coverage factor t = 2 is still permissible if the bias and precision limits have similar magnitude. If unacceptably large P’s are involved, the elemental error sources contributions must be examined to see which need to be (or can be) improved Calculate total uncertainty using equation (4) For each r, report total uncertainty and bias and precision limits Recommendations Recognize that uncertainty depends on entire testing process and that any changes in the process can significantly affect the uncertainty of the test results Integrate uncertainty assessment methodology into all phases of the testing process (design, planning, calibration, execution and post-test analyses) Simplify analyses by using prior knowledge (e.g., data base), concentrate on dominant error sources and use end-to-end calibrations and/or bias and precision limit estimation Document: test design, measurement systems, and data streams in block diagrams equipment and procedures used error sources considered all estimates for bias and precision limits and the methods used in their estimation (e.g., manufacturers specifications, comparisons against standards, experience, etc.) detailed uncertainty assessment methodology and actual data uncertainty estimates References AIAA, 1999, “Assessment of Wind Tunnel Data Uncertainty,” AIAA S-071A-1999. ASME, 1998, “Test Uncertainty,” ASME PTC 19.1-1998. ANSI/ASME, 1985, “Measurement Uncertainty: Part 1, Instrument and Apparatus,” ANSI/ASME PTC 19.I-1985. Coleman, H.W. and Steele, W.G., 1999, Experimentation and Uncertainty Analysis for Engineers, 2nd Edition, John Wiley & Sons, Inc., New York, NY. Coleman, H.W. and Steele, W.G., 1995, “Engineering Application of Experimental Uncertainty Analysis,” AIAA Journal, Vol. 33, No.10, pp. 1888 – 1896. ISO, 1993, “Guide to the Expression of Uncertainty in Measurement,", 1st edition, ISBN 92-67-10188-9. ITTC, 1999, Proceedings 22nd International Towing Tank Conference, “Resistance Committee Report,” Seoul Korea and Shanghai China.