132
set, had an R2 of 0.962 for 240 number of cases. Error analysis indi
cated that 11 out of the 240 cases had errors exceeding 20 percent, and
these were cases with tx equal to 1.5 or 6.0 in. These were the
extremes of the range of tx used to develop Equation 4.21.
When the data set was expanded to include tx and E4 values of 8.0
in. and 5.0 ksi, respectively, Equation 4.22 was obtained. This equa
tion had an R2 of 0.953 for 400 number of observations. Error analysis
indicated that 30 out of 400 observations had predictive errors above 20
percent. These simulated pavements are listed in Table 4.11, which
shows that two pavements (numbers 24 and 28) had E2 predictions with 30
to 40 percent error. One pavement (number 27) had 54 percent prediction
error. When all three pavements with 30 percent or more predictive
errors were deleted from the data set, the R2 value increased from 0.953
to 0.956. This increase was considered not significant enough to war
rant changing the Equation 4.22 format.
The most interesting thing from Equation 4.22 was that most of the
errors occurred at the extreme limits of tx (1.5 and 8.0 in.). As shown
in Table 4.11, most of these high errors were pavements with E2 values
of 42.5 and 175.0 ksi. Predictions were generally good for intermediate
t1 values, especially 4.5 and 6.0 in. Moreover, Equation 4.22 contains
sensor 8 deflection measurement (Dg) which is currently not made with
the conventional sensor array utilized by the FD0T. Therefore, the
reliability of this equation is contingent upon measurement of 0o. In
o
the absence of D measurement, Equation 4.21 would be the best choice
8
for prediction of the base course modulus.
4.4.2.3 Stabilized Subgrade Modulus, E^. Two equations were also
developed to predict E3. Equation 4.23 which holds for t values from