Neural Networks For Highway Maintenance Investigations Using Ground Penetrating Radar*

The early detection of potential subsurface erosion in highway systems allows for the application of appropriate<br>maintenance rather than more expensive rehabilitation or reconstruction. Currently, there are few practical test<br>procedures to assist a highway maintenance engineer in detecting these problems or in evaluating how effective<br>remedial measures were in repairing the problem. Existing procedures rely on complex algorithms that are<br>computationally expensive and are not amenable to real time applications. Our research evaluates the synergy<br>between Neural Networks (NNs) and Ground Penetrating Radar (GPR) as tools for early detection of subsurface<br>highway problems. We trained NNs to identify the following pavement characteristics in GPR SC~IIS:<br>(i) pavement thickness (ii) the degree of moisture in the surface layer, (iii) the degree of moisture in the base<br>layer, (iv) voids or loss of support beneath slabs, and (v) overlay delamination. Undesirable conditions can then<br>be treated by maintenance, and major structural problems requiring extensive repair can be avoided. Our research<br>shows that the NN/GPR approach is a viable solution. The NNs extrapolate over noisy or incomplete data, and<br>they provide faster responses than existing methods.