129
listed in Table 4.9. Only 14 out of 320 cases had errors above 30
percent but they did not exceed 40 percent. These cases occurred when
tx value was 3.0 in. (10 cases) and 4.5 in. (4 cases). Table 4.9 also
shows that only six cases with 20 to 30 percent predictive error
occurred for tx value of 6.0 and 8.0 in. More importantly, the percent
error decreased substantially for tx equal to 8.0 in. The majority of
the pavements with 8.0 in. asphalt concrete thickness had less than 10
percent prediction error.
The general tendency for Equations 4.18 and 4.19 was to predict
with a higher degree of accuracy for pavements with thick asphalt
concrete layers. Because of the high prediction errors from 3.0 in.
asphalt concrete pavements, all the 3.0 in. pavements were also deleted
from the data set. Subsequent regression analysis of the remaining data
set yielded Equation 4.20, with a significantly improved correlation (R2
= 0.993, and N = 240). Error analysis revealed that only 2 out of 240
cases had errors of 20 percent or more. Pavements with E predictions
having errors of 15 percent or more are listed in Table 4.10. This
table indicates that only the 4.5- and 6.0-in. pavements fall in this
category. The 4.5-in. pavements all underpredicted Ex when the actual
value was 1200 ksi. However, for the 6.0-in. pavements, Ex values of
150 and 300 ksi were generally overpredicted. Again, predictions were
very good for the 8.0-in. asphalt concrete pavements. This supports the
findings with the Dynaflect that for thicker pavements, the effect of tx
on E is small.
4.4.2.2 Base Course Modulus, E2 Two equations were derived to
predict the base course modulus using theoretical 9-kip FWD load deflec
tions. Equation 4.21, which was developed from a subset of the data