A potential automated pavement evaluation system to address multisensor applications; integrate different types of sensors, techniques, and information; and offer more sophisticated and intelligent processing capabilities for improved pavement management is described. The separate components of this system either now exist in prototype form or are under development. Such a system could automate in real time much of the pavement data acquisition, interpretation, and evaluation process, and capture the experience and judgement of expert pavement engineers in performing condition assessments and identification of appropriate rehabilitation and maintenance strategies. This research is directed toward an innovative, noncontact, intelligent nondestructive evaluation (INDE) system, using a novel artificial intelligence (AI)-based approach that would integrate three AI technologies: computer vision, neural networks, and knowledge-based expert systems, in addition to conventional algorithmic and modelling techniques. The focus of the current, initial research is development of an advanced sensor processing capability using neural network technology to determine the type, severity, and extent of distresses from digitised video image representations of the pavement surface acquired in real time. The properties of neural networks provide potential solutions to the inherently difficult nature of sensor integration and output interpretation in automated pavement evaluation. The background and conceptual development of an INDE system for automated pavement evaluation, and initial research results that demonstrate the feasibility of a neural network approach in a case study application using a multilayer perception and a backpropagation learning rule, are described.
Abstract