Machine learning to enhance the management of highway pavements and bridges

Abstract

The adoption of machine learning in transportation asset management is hindered by the perception of being a black box, the natural resistance to change, and the challenges of integration with existing management systems. This paper aims to enhance the understanding of machine learning and provide guidance for the development and implementation of machine learning to support decision-making in the management of highway pavements and bridges. The paper identifies successful research efforts using machine learning, identifies opportunities and challenges in adopting machine learning, and derives recommendations on when and how to apply different machine learning algorithms to support asset management decisions. Four main challenges were identified: the trade-off between accuracy and interpretability, the shortage of machine learning engineers, data quality, and the limitations of machine learning algorithms. Although the complexities associated with training machine learning algorithms challenge the short-term implementation, machine learning offer a wide range of opportunities when compared to traditional approaches. The development of hybrid systems combining machine learning algorithms with expert opinions and traditional approaches seems a reasonable step forward to support agencies asset management decisions.