# Error Propagation, Sensitivity and Total Uncertainty

According to the general uncertainty analysis formalism (Ronen 1988; Bevington and Robinson 1992), the total uncertainty of pressure p is described by the error propagation equation

where ptj = cov(CiCj)/[var( Ct)var( Ci)]1/2 is the correlation coefficient between the variables Ct and Ci, var(Ct) = < ACt2 > and

cov(Ci C j) = < ACi AC j > are the variance and covariance, respectively, and the notation < > denotes the statistical ensemble average. Here, the variables (Cj, i = 1–M} denote a set of the parameters Dt(At), Dx(Ax), Dq0(At), V,

Vref, nc/ncref, Пf/nfref, h/href> c/cref, q0/q0ref, pref, T, A and B in Eq.

(7.2) . The sensitivity coefficients St are defined as St = ( C/p )( dp/dCt ). Eq.

(7.3) becomes particularly simple when the cross-correlation coefficients between the variables vanish (ptj = 0, i Ф j).

Table 7.1 lists the sensitivity coefficients, the elemental errors and their physical origins. Many sensitivity coefficients are proportional to a factor (p = 1 + [A(T)/B(T)] /(p/pref ). For Bath Ruth + silica-gel in GE RTV 118,

Figure 7.1 shows the factor 1 + [A(T)/B(T)]/(p/pref ) as a function of p/pref for different temperatures, which is only slightly changed by temperature. The temperature sensitivity coefficient is ST = – T[B’ (T) + A'(T) pref/p]/B(T ), where the prime denotes differentiation respect with temperature. Figure 7.2 shows the absolute value of ST as a function of p/pref at different temperatures.

After the elemental errors in Table 7.1 are evaluated, the total uncertainty in pressure can be readily calculated using Eq. (7.3). The major elemental error sources are discussed below.

 Table 7.1. Sensitivity coefficients, elemental errors, and total uncertainty of PSP Variable Sensi. Elemental Variance Physical Origin £ Coef. s var( C t) 1 Dt(At) f [(dV/ dt)( At)/V]2 Temporal variation in luminescence due to photodegradation and surface contamination 2 DJAx ) (p [ (dV/dx)2 a2 + (dV/д y )2 a 2y]V ~2 Image registration errors for correcting luminescence variation due to model motion 3 Dq0(At) (p [( дqo / д t)( At)/qoref]2 Temporal variation in illumination 4 Vef (p Vref ghv Bd Photodetector noise 5 V –

Note:

(1) <7x and <ry are the standard deviations of least-squares estimation in the image registration or camera calibration.

(2) The factors for the sensitivity coefficient are defined as <p = 1 + [ A(T)/B(T)](pref / p) and St = -[T/B(T)][B'(T) + A'(T)(p„f/p)] .

Fig. 7.2. The temperature sensitivity coefficient as a function of p/pref at different temperatures for Bath Ruth + silica-gel in GE RTV 118. From Liu et al. (2001a)