Asset Management

This report explores the use of Artificial Intelligence (AI) and Machine-Learning (ML) in pavement asset management and details a methodology for developing new use-cases. While there are some existing uses of AI and ML technology in pavement asset management the development of new use-cases have often failed due to the inappropriate use of data and models, and difficulty articulating the constraints of different technical disciplines.

Two case studies demonstrate the application of the methodology using data from four member authorities.

The first case study sought to understand to what extent, historic condition, inventory and treatment data could be used to train machine learning models to reproduce expert pavement asset management decisions. Models were developed and evaluated on four Austroads member authority datasets, predicting six treatment classes over four future time-horizons. Treatment predictions were made for short road sections prior to clustering into candidate projects. The results were promising, and suggest that with further development machine learning models could provide decision support and efficiency gains to programme developers.

The second case study explored an extension to conventional Pavement Management System (PMS) optimisation to provide insight into the network-wide implications of various multi-criteria funding allocation scenarios. The resulting data captures the implications of all achievable funding splits (freight / non-freight routes, and metropolitan / regional routes) on network Level of Service. The project team developed a system to achieve specific multi-criteria splits using existing PMS optimisation tools, which suggests this methodology could be adopted without significant disruption or risk.

Code

The two case studies were developed using open-source Python data science tools, such as Python, NumPy and Pandas.

The code, with instructions, can be downloaded from github.com/Austroads/aam6201

Quick Reference

This Quick Reference contains a set of practical resources to help Austroads member authorities develop small teams to explore new AI/ML projects and successfully prosecute research and development through to integration into existing business processes. It includes resources which can help you define and scope projects, form teams, recruit specialists, manage projects, understand key terminology and concepts, and assess the viability and risks of specific project ideas.

Watch the first webinar in a two-part series with authors Tim Cross and David Rawlinson. The session describes tips, tricks and success strategies to ensure AI/ML projects proceed to delivery and successful integration with existing business processes. The presenters introduce the Quick Reference document with practical examples on how to use it.

Watch the second webinar on the Development of Machine Learning Decision Support Tools for Pavement Asset Management.