©1997-2001 by M. Kostic Ch.5: Uncertainty/Error Analysis • • • • • • • • • • • • Introduction Bias and Precision Errors/Uncertainties Different Instrument Errors RSS-Errors Summing Error Sources Sensitivity or Dependency Rate Error Propagation (Combined Uncertainty) Design-stage Uncertainty (Instrument errors only) Advanced-stage Uncertainty (Instrument and measurement errors) Multiple- measurement Uncertainty Error Summation/Propagation (Expanded Combined Uncertainty) Problem 5-30 Ch.5: Uncertainty/Error Analysis ©1997-2001 by M. Kostic Errors (e = x-x’ ≅ x-xavg= d, also B, P, S≅σ; do NOT be confused, see NIST Guide): Bias (B), Precision (P), also Standard (S or σ) Uncertainty (u) is the range of errors (e, B, P, S) at corresponding Probability (%P) Remember: u = d%P = t ν,%PS (@ %P); z=t=d/S 1 Bias and Precision Errors/Uncertainties ©1997-2001 by M. Kostic Precision error Average value Bias error True Value Different Instrument Errors 1 3 2 4 u y = u y1 + u y 2 + ... + u yL = 2 2 ©1997-2001 by M. Kostic 2 5 L ∑u i =1 2 yi ; i.e. RSS ( Root − Sum − Square) procedure 2 ©1997-2001 by M. Kostic RSS-Errors Summing e1 e2 e3 e4 eSUM=|e 1 |+|e2 |+|e 3 |+|e4 | eR e4=Max < e R< eSUM e1 e3 e2 e4=Max e1 e R12 eR 2 2 L ∑u i =1 e4 …more probable …or, in general, when eyi is uyi : u y = u y1 + u y 2 + ... + u yL = e3 e R123 e R=e RSS = (e1 )2 +(e 2 )2 +(e3 )2 +(e 4 )2 2 …too conservative some errors are opposite signs and cancel out e2 2 yi ; i.e. RSS ( Root − Sum − Square) procedure ©1997-2001 by M. Kostic Error Sources 1. Calibration Error Source Group Table 5.1 - (B or P)1j 2. Data Acquisition Error Source Group Table 5.2 - (B or P)2j 3. Data Reduction Error Source Group Table 5.3 - (B or P)3j It is not important which group an error is assigned to, as long as it is accounted for. The groups and their items are for convenience only. 3 ©1997-2001 by M. Kostic Sensitivity or Dependency Rate ity ∂ y is tiv = ∂x sen θ y / x :e uncertaintyof y = p /x y ∂ Slo c δy ≈ y ux = u y = /x ∂x Ky δx ≈ ux , uncertainty of x If one variable " y"depends on another " x, " then a small change of x, i.e δx ≈ ux (error, uncertainty) will propagate as error of y, i.e. δy ≈ u y . Using the partial derivative, i.e. the rate of dependancy, or sensitivit y : δy ≈ ∂y ∂y ∂y δx ≈ ux = (θ y / x )u x = u y ; where sensitivit y K y / x = cy / x = θ y / x = ∂x ∂x ∂x ©1997-2001 by M. Kostic Error Propagation (Combined Uncertainty) If one variable y depends on another x, then a small change of x , i.e δx ≈ u x (error, uncertaint y ) will propagate as error of y , i.e. δ y ≈ u y : δy ≈ ∂y ∂y ∂y δx ≈ u x = (θ y / x )u x = u y ; where sensitivit y θ y / x = ∂x ∂x ∂x For a multyfunct ion variable y = y ( x1 , x2 ,... xi ,... xL ) : uy = ∑ (θ L i =1 2 y / xi u xi ) = L ∑u i =1 2 yi ; i.e. RSS (Root − Sum − Square) procedure u yi a known elemental error ∂y where u yi = or u yi = ∂x u xi = (θ yi / x ) u xi i 4 ©1997-2001 by M. Kostic Design-stage Uncertainty (Instrument errors only) Design - stage error/uncertainty u d = u 02 + u c2 Interpolation error Instrument error u0 = ± 12 resolution uc (by calibration) Advanced-stage Uncertainty ©1997-2001 by M. Kostic (Instrument and measurement errors) N th order uncertainty N u N = u + ∑ ui2 2 d i =1 Zero - and design order uncertaint ies u0 = ± 1 2 resolution; u d = u02 + u c2 First - order uncertaint y u1 ≥ u0 , u d 5 Multiple-measurement Uncertainty ©1997-2001 by M. Kostic ©1997-2001 by M. Kostic Error Summation/Propagation (Expanded Combined Uncertainty) uR = BR = ∑B i B R + ( tν R ,% P PR ) 2 , where : 2 2 i and PR = ∑P i 2 ; also : i 2 νR 2 ∑ Pi ; ν = N − 1, # of degree of freedom = i i i 4 Pi ∑i ν i Note, it could be Pi = Pyi = θ yi / xi Pxi or, it could be B i = B yi = θ yi / xi B xi 6 ©1997-2001 by M. Kostic Problem 5-30 ©1997-2001 by M. Kostic Problem 5-30 (Cont.) 7 ©1997-2001 by M. Kostic Problem 5-30 (Cont.) ©1997-2001 by M. Kostic Problem 5-30 (Cont.) 8 Problem 5-30 (Cont.) ©1997-2001 by M. Kostic ©1997-2001 by M. Kostic Problem 5-30 (Cont.) 9