Biomedical Engineering Reference
In-Depth Information
of the residuals about the individual regressions (
LPM
/
SPM
vs.
MMAD
), and by
definition reflects the precision of the fitted regression.
RMSE
is transformed by
dividing by the slope of the regression to express this statistic in terms of precision
of the estimated
MMAD
.
Both of these plots do the following:
1. They verify that setting the
LPM
/
SPM
boundary at the target
MMAD
provides
optimum precision.
2. They confirm that the precise selection of the boundary location with respect to
the
MMAD
value is not critical and can vary by an order of magnitude in the ratio
(from 0.3 to 3.0, as indicated by the placement of the solid vertical lines for ratio
in both plots), without serious degradation of precision.
7.7
Experimental Evidence for Ratio
LPM
/
SPM
as Measure
of Mean of the APSD
The most important aspects of applying abbreviated data acquisition and analysis
strategies for OIPs are the initial determination of the full-resolution APSD profile
for each product in a robust manner and subsequent confirmation that an AIM sys-
tem with a particular chosen boundary between
LPM
and
SPM
provides acceptable
predictive capability for
MMAD
.
Overall, Tougas et al. [
11
] showed in their investigations of OIP APSDs that
the
LPM
/
SPM
ratio appears to be capable of detecting small changes in
MMAD
of
the order of tenth(s) of microns. This finding is reflected in the magnitude of the
goodness-of-it statistic (
R
2
) and [
RMSE
/b], obtained for regressions of the
LPM
/
SPM
ratio versus
MMAD
reported in Table
7.1
.
Tougas et al. observed that the relationship between
MMAD
and the
LPM
/
SPM
ratio was approximately linear for every OIP type studied, illustrated by the magni-
tudes of the coefficient of determination and
RMSE/b
goodness-of-fit statistics [
11
].
They also noted that a small degree of systematic deviation from linearity was
observed in some cases and observed that such behavior is consistent with the
expectations for the ratio metric
LPM
/
SPM
. Thus, as values of
MMAD
approach the
lower bound (finest particles) of the size range of the CI, the
LPM
/
SPM
ratio trends
towards zero. Similarly, as values of
MMAD
approach the upper bound (coarsest
particles) of the size range, this ratio trends towards infinity.
The results in Table
7.1
reflect outcomes for the
LPM
/
SPM
boundary placement
that provided the best correlation between
LPM/SPM
ratio and
MMAD
(denoted as
the optimum boundary in this table). Figure
7.9
illustrates the nature and quality of
these regressions for two cases selected as examples (OIPs
w9k001
and
w9k901
).
The 95% prediction bounds at the mean
LPM
/
SPM
ratio were projected onto the
x
-axis (aerodynamic diameter in μm). Tougas et al
.
noted that the difference between
these projections of the upper and lower prediction intervals reflects the ability of
the
LPM
/
SPM
ratio to detect differences in
MMAD
and indicates that changes of a
Search WWH ::
Custom Search