PAVEMENT STRUCTURE EVALUATION WITH MACHINE LEARNING FOCUSING SHAPE OF CRACK AND RUTS ON SURFACE

Abstract

 The technical standard for pavement inspection in Japan attaches importance to structural soundness of subbase layer and below. The authors have built a new way to evaluate structural soundness without traffic restrictions. This evalutation way uses crack shape, rut shape, and machine learning method (convolutional neural network:CNN). Deflection D₀ was estimated using the crack diagram, cross-sectional profile and asphalt layer thickness. This evaluation way estimated the deflection D₀ with higher accuracy than the conventional method using cracking ratio and rut depth.