Propagation of Uncertainty Jake Blanchard Spring 2010 Uncertainty Analysis for Engineers 1 Introduction We’ve discussed single-variable probability distributions This lets us represent uncertain inputs But what of variables that depend on these inputs? How do we represent their uncertainty? Some problems can be done analytically; others can only be done numerically These slides discuss analytical approaches Uncertainty Analysis for Engineers 2 Functions of 1 Random Variable Suppose we have Y=g(X) where X is a random input variable Assume the pdf of X is represented by fx. If this pdf is discrete, then we can just map pdf of X onto Y In other words X=g-1(Y) So fy(Y)=fx[g-1(y)] Uncertainty Analysis for Engineers 3 Example Consider Y=X2. Also, assume discrete pdf of X is as shown below When X=1, Y=1; X=2, Y=4; X=3, Y=9 0.4 0.4 0.35 0.35 0.3 0.3 0.25 0.25 0.2 0.2 0.15 0.15 0.1 0.1 0.05 0.05 0 0 0 1 2 3 4 5 6 0 5 10 15 20 Uncertainty Analysis for Engineers 25 30 4 Discrete Variables Example: ◦ Manufacturer incurs warranty charges for system breakdowns ◦ Charge is C for the first breakdown, C2 for the second failure, and Cx for the xth breakdown (C>1) ◦ Time between failures is exponentially distributed (parameter ), so number of failures in period T is Poisson variate with parameter T ◦ What is distribution for warranty cost for T=1 year Uncertainty Analysis for Engineers 5 Formulation e x f ( x) x 0,1, 2, ... x! x0 0 w h( x ) x x 1, 2, ... C 0 x ln(w) ln(C ) w0 w C , C 2 , ... e w0 p ( w) ln(w) ln(C ) e 2 w C , C , ... ln(w) ! ln(C ) Uncertainty Analysis for Engineers 6 Plots 0.4 0.35 0.3 0.25 0.2 0.15 0.1 0.05 0 0.4 1 1.5 2 2.5 3 3.5 x 4 4.5 5 0.35 0.3 0.25 C=2 =1 0.2 0.15 0.1 0.05 0 0 5 10 15 20 25 Uncertainty Analysis for Engineers w 30 35 7 CDF For Discrete Distributions If g(x) monotonically increases, then P(Y<y)=P[X<g-1(y)] If g(x) monotonically decreases, then P(Y<y)=P[X>g-1(y)] …and, formally, F ( y) F g 1 ( y) Y p (x ) X 1 x i xi g ( y ) y y x x Uncertainty Analysis for Engineers 8 Another Example Suppose Y=X2 and X is Poisson with parameter Y X g( X ) 2 1 g (Y ) X Y px t py t x x! e t y y ! e t x 0,1,2,3,... y 0,1,4,9,... Uncertainty Analysis for Engineers 9 Continuous Distributions If fx is continuous, it takes a bit more work f FY ( y ) g 1 ( y ) x ( x)dx x g 1 ( y ) f x ( x)dx x g 1 y dg 1 ( y ) dx dy dy 1 dg ( y) 1 FY ( y ) f x g y dy dy y or 1 dF dg f y ( y) f x g 1 dy dy Uncertainty Analysis for Engineers 10 Example Y X X g 1 ( y ) Y dg 1 dy im agine Normal distribution Mean=0, =1 2 1 1 X fx exp 2 2 y2 1 fy exp 2 2 fy y2 1 exp 2 2 Uncertainty Analysis for Engineers 11 Example Y ln(X ) X is lognormal Normal distribution X g 1 ( y ) exp(Y ) dg 1 exp(Y ) dy im agine 2 1 1 ln(x) fx exp 2 x 2 2 1 1 y exp(y ) fy exp 2 exp(y ) 2 2 1 1 y fy exp 2 2 Uncertainty Analysis for Engineers 12 If g-1(y) is multi-valued… k fY ( y ) f x g i1 i 1 dg 1 dy Exam ple U cS 2 S u c dS 1 du 2 cu u u 1 fs f u f s c 2 cu c S l ognorm al( , ) Uncertainty Analysis for Engineers 13 Example (continued) 2 1 1 ln(s ) fs exp 2 s 2 lognormal 2 u ln 1 1 c 1 fu exp 2 u 2 cu 2 c 2 1 1 lnu lnc 2 fu exp 2 2 2 u 2 u lnc 2 u 2 Uncertainty Analysis for Engineers 14 Example FV 2 Z aV 2 1400d im agine v 1 v 0 f v exp v0 v0 v Z a dV 1 dz 2 az z z 1 z 1 fv f z f v f v a 2 az a 2 az a z 1 1 a exp v0 2 az v0 Uncertainty Analysis for Engineers 15 A second example Suppose we are making strips of sheet metal If there is a flaw in the sheet, we must discard some material We want an assessment of how much waste we expect Assume flaws lie in line segments (of constant length L) making an angle with the sides of the sheet is uniformly distributed from 0 to Uncertainty Analysis for Engineers 16 Schematic L w Uncertainty Analysis for Engineers 17 Example (continued) Whenever a flaw is found, we must cut out a segment of width w w h L sin f U 0, w w sin L 1 2 1/ 2 d 1 w 1 dw L L 1 L2 w2 Uncertainty Analysis for Engineers 18 Example (continued) g-1 is multi-valued </2 f1 w f 2 w 1 L w 2 2 >/2 1 L w 2 0w L 2 f w f1 w f 2 w 0w L 2 L w 2 2 Uncertainty Analysis for Engineers 19 Results 5 4.5 4 3.5 pdf 3 2.5 2 1.5 1 L=1 0.5 0 0.1 0.2 0.3 0.4 0.5 w 0.6 0.7 0.8 0.9 1 1 0.9 0.8 0.7 0.6 cdf 0.5 0.4 0.3 0.2 0.1 0 0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 w Analysis for Engineers Uncertainty 1 20 Functions of Multiple Random Variables Z=g(X,Y) For discrete variables fz f x, y g ( xi , y j ) z ( xi , y j ) If we have the sum of random variables Z=X+Y fz f xi y j z x, y ( xi , y j ) f x , z x x, y i i all xi Uncertainty Analysis for Engineers 21 Example Z=X+Y 0.7 0.6 0.4 0.3 0.2 0.45 0.1 0.4 0 0 0.5 1 1.5 2 2.5 3 3.5 0.35 0.3 x fy fx 0.5 0.25 0.2 0.15 0.1 0.05 0 0 5 10 15 20 25 30 35 y Uncertainty Analysis for Engineers 22 Analysis X Y Z P Z-rank 1 10 11 .08 1 1 20 21 .04 4 1 30 31 .08 7 2 10 12 .24 2 2 20 22 .12 5 2 30 32 .24 8 3 10 13 .08 3 3 20 23 .04 6 3 30 33 .08 9 Uncertainty Analysis for Engineers 23 Result 0.3 0.25 fz 0.2 0.15 0.1 0.05 0 0 5 10 15 20 25 30 35 Z Uncertainty Analysis for Engineers 24 Example Z=X+Y 0.7 0.6 fx 0.5 0.4 0.3 0.2 fy 0.1 0 0 0.5 1 1.5 2 x 2.5 3 0.45 0.4 3.5 0.35 0.3 0.25 0.2 0.15 0.1 0.05 0 0 0.5 1 1.5 2 2.5 3 3.5 4 4.5 y Uncertainty Analysis for Engineers 25 Analysis X Y Z P Z-rank 1 2 3 .08 1 1 3 4 .04 2 1 4 5 .08 3 2 2 4 .24 2 2 3 5 .12 3 2 4 6 .24 4 3 2 5 .08 3 3 3 6 .04 4 3 4 7 .08 5 Uncertainty Analysis for Engineers 26 Compiled Data z fz 3 .08 4 .28 5 .28 6 .28 7 .08 fz 0.3 0.25 0.2 0.15 0.1 0.05 0 0 1 2 3 4 z 5 6 7 8 Uncertainty Analysis for Engineers 27 Example Z X Y vt x fx exp(vt) x and y are integers x! y t fy exp( t ) y! f z f x ( x) f y ( z x) all x vt t x!( z x)! allx fz t exp v t v exp v t x z zx x allx zx x!( z x)! Uncertainty Analysis for Engineers 28 Example (continued) v x zx z x!( z x)! z! allx v t z fz z! exp v t The sum of n independent Poisson processes is Poisson Uncertainty Analysis for Engineers 29 Continuous Variables Z g( X ,Y ) Fz ( z ) f x, y g ( x , y ) z ( x, y )dxdy 1 g Fz ( z ) f x, y ( x, y )dxdy x g 1 ( z , y ) g 1 z dg 1 Fz ( z ) f x , y ( g , y ) dzdy dz 1 z 1 dg Fz ( z ) f x , y ( g 1 , y ) dydz dz Uncertainty Analysis for Engineers 30 Continuous Variables 1 dg f z ( z ) f x , y ( g 1 , Y ) dY dz dg 1 f x, y ( X , g ) dX dz 1 if Z aX bY Z bY X a dg 1 1 dz a 1 Z bY f z f x, y ( , Y )dY a a Uncertainty Analysis for Engineers 31 Continuous Variables (cont.) 1 Z bY f z f x, y ( , Y )dY a a 1 Z aX f z f x, y ( X , )dX b b x, y independent 1 Z bY fz fx ( ) f y (Y )dy a a Uncertainty Analysis for Engineers 32 Example W U V ; U ,V 0 u fu exp 2 m u 2m 1 v fv exp 2 m v 2m 1 fw f u (u ) f v ( w u )du 1 w 1 fw exp 2m 2m u 1 u (w u) exp exp du 2 m u 2m 2 m( w u ) 2m 1 1 du wu du w fw exp 2 m 2m 0 u ( w u ) 1 fw 1 w exp 2m 2m w Uncertainty Analysis for Engineers 33 In General… If Z=X+Y and X and Y are normal dist. fz f x ( z y ) f y ( y )dy fz 2 2 y y 1 1 z y x 1 1 dy exp exp x 2 y 2 x 2 2 y 2 2 y z y x 1 1 1 y dy fz exp 2 x y 2 x 2 y fz 1 2 x2 y2 2 z 1 x y exp 2 2 2 x y Then Z is also normal with z x y 2 2 2 UncertaintyzAnalysis forx Engineersy 34 Products Z XY Z X Y X 1 Z Y 1 z f z ( Z ) f X ,Y ( , y )dy Y y im agineX i all l ognorm aland n Z Xi i 1 n ln(Z ) ln(X i ) i 1 n Z X i 1 i n X2 2 Z i 1 i Uncertainty Analysis for Engineers 35 Example W, F, E are lognormal WF C E 1 ln(C ) ln(W ) ln(F ) ln(E ) 2 1 C W F E 2 1 E 2 2 C 2 W 2 2 F Uncertainty Analysis for Engineers 36 Central Limit Theorem The sum of a large number of individual random components, none of which is dominant, tends to the Gaussian distribution (for large n) Uncertainty Analysis for Engineers 37 Generalization More than two variables… Z g ( x1 , x2 , x3 ,..., xn ) f Z ( z ) ... f x1 ,...,xn 1 g g 1 , x2 , x3 ,..., xn dx2 dx3 ...dxn z Uncertainty Analysis for Engineers 38 Moments Suppose Z=g(X1, X2, …,Xn) E ( Z ) ... z f X 1 , X 2 ,...,X n ( X 1 , X 2 ,..., X n )dX1dX 2 ...dX n E ( Z ) ... g ( X 1 , X 2 ,..., X n ) f X1 , X 2 ,...,X n ( X 1 , X 2 ,..., X n )dX1dX 2 ...dX n im agine Y aX b E (Y ) Y f y ( y )dy (ax b) f x ( x)dx E (Y ) a x f x ( x)dx b f x ( x)dx aE( X ) b Uncertainty Analysis for Engineers 39 Moments ax b Var (Y ) E Y Y 2 2 f x ( x)dx Y Var (Y ) 2 ax b aE ( x ) b f x ( x)dx Var (Y ) a 2 2 x E ( x ) f x ( x)dx Var (Y ) a 2Var ( X ) Uncertainty Analysis for Engineers 40 im agine Moments Y aX1 bX 2 E (Y ) aE( X 1 ) bE( X 2 ) Var (Y ) y 2 f x1 , x2 ( x1 , x2 )dx1dx2 y Var (Y ) ax bx 2 a x1 b x2 1 2 f x1 , x2 ( x1 , x2 )dx1dx2 Var (Y ) a 2 x 2 1 x1 f x1 , x2 ( x1 , x2 )dx1dx2 b2 x 2 x2 2 f x1 , x2 ( x1 , x2 )dx1dx2 2ab x 1 x1 x2 x2 f x1 , x2 ( x1 , x2 )dx1dx2 Var (Y ) a 2Var ( X 1 ) b 2Var ( X 2 ) 2abCov( X 1 , X 2 ) Cov( X 1 , X 2 ) E X 1 X 1 X 2 X 2 E (Uncertainty X 1 X 2 ) Analysis E( X )E( X ) for1Engineers 2 41 Approximation Y g( X ) E (Y ) g( X ) f x ( X )dX dg g ( X ) g ( x ) X x dx dg E (Y ) g ( x ) X x f x ( X )dX dx dg E (Y ) g ( x ) f x ( X )dX X x f x ( X )dX dx dg X x f x ( X )dX E (Y ) g ( x ) f x ( X )dX dx E (Y ) g ( x ) Uncertainty Analysis for Engineers 42 Approximation Var (Y ) g ( X ) 2 y f x ( X )dX dg g ( X ) g ( x ) X x dx Var (Y ) 2 dg g ( ) X x y f x ( X ) dX x dx Var (Y ) 2 dg X f x ( X )dX x dx dg Var (Y ) dx 2 2 X f x ( X )dX x dg Var (Y ) Var X dx x x 2 Uncertainty Analysis for Engineers 43 Second Order Approximation Y g( X ) E (Y ) g( X ) f x ( X )dX 2 dg 1 d g 2 g ( X ) g ( x ) X x X x dx 2 dx2 2 1 d g 2 E (Y ) g ( x ) X x f ( X )dX 2 x 2 dx 1 d 2g 2 E (Y ) g ( x ) X f x ( X )dX x 2 2 dx 1 d 2g E (Y ) g ( x ) 2 dx2 Var ( x) x x Uncertainty Analysis for Engineers 44 Approximation for Multiple Inputs Y g ( X 1 , X 2 , X 3 ,..., X n ) E (Y ) g X1 , X 2 , X 3 ,..., X n g Var (Y ) i 1 X i n 1 n 2 2 g xi 2 2 i 1 xi 2 2 Xi Uncertainty Analysis for Engineers 45 Example WF C E Example 4.13 Do exact and then use approximation and compare Waste Treatment Plant – C=cost, W=weight of waste, F=unit cost factor, E=efficiency coefficient median cov W 2000 ton/y .2 F $20/ton .15 E 1.6 .125 Uncertainty Analysis for Engineers 46 Solving… W lnWmedian 7.6009 F lnFmedian 2.9957 E lnEmedian 0.4700 W ln 1 covW2 0.19804 F ln 1 cov2F 0.149166 E ln 1 cov2E 0.124516 1 lnC lnW lnF lnE 2 1 C E lnC W F E 10.36 2 1 2 2 2 C W F E 0.25563 4 1 C exp C C2 32,620 2 C C exp C2 1 8477 Uncertainty Analysis for Engineers 47 Approximation 1 2 2g 1 2 2g 1 2 2g E (C ) g W , F , E W F E 2 W 2 2 F 2 2 E 2 g W , F , E W F E W 2039.6; F 20.223; E 1.6124 W 407.915; F 3.033; E 0.2016 W F 32620 32483 32,483; error 0.4% 32620 E 2g 2g 0 W 2 F 2 2 g 3W F 2 E 4 E5 / 2 W F 1 2 2 g E 32673 2 E 2 E error 32673 32620 0.16% 32620 Uncertainty Analysis for Engineers 48 Variance 2 g 2 g 2 g Var (C ) W F E W F E 2 2 F Var (C ) W E C 8370 2 2 F2 W E 2 2 E2 W F 2 3 / 2 E 2 8477 8370 error 1.3% 8477 Uncertainty Analysis for Engineers 49