| Summary: | 碩士 === 淡江大學 === 土木工程學系 === 85 === Using nondestructive deflection measurements to evaluate
pavementstructural capacity has become popular for highway
angencies recently.Surface deflection basins obtained directly
from nondestructive testingdevices can be transferred to
deflection indices or layer moduli throughbackcalculations.
However, deflection indices can only give very roughdescription
of the pavement structural condition. Backcalculations oflayer
moduli, however, usually involve great computation times
andsometimes with the possibility that their solutions would not
converge. Artificial neural networks provide a fundamental
new approach tobackcalculation of layer moduli from
nondestructive deflectionmeasurements. An artificial neural
network is a highly interconnectedcollection of simple
processing elements that can be trained toapproximate a complex,
nonlinear function through repeated exposure toexamples of the
function. One backpropagation neural etwork was utilizedto
backcalculate layer moduli for four-layer pavement profiles.
Two types of nondestructive testing device, Road Rater and
FallingWeight Deflectometer, were simulated in this study.
Synthetic deflectionbasins generated by the ELSYM5 program and a
wide variety of layer moduliwere used as training sets.
Subsequent testing showed that the networkcould backcalculate
pavement layer moduli accurately. Four seperatenetworks were
then trained to backcalculate layer moduli, E1, E2, E3, E4,
respectively, using the same training sets. The results showed
thatfour-network approach performed better than one-network
approach especiallyin E4 backcalculation.
|
|---|
