Enabling methods for predictive digital twin in pavement performance modelling

Chen, Kun (2025). Enabling methods for predictive digital twin in pavement performance modelling. University of Birmingham. Ph.D.

Preview

Chen2025PhD.pdf
Text
Available under License All rights reserved.
Download (4MB) | Preview

Abstract

Roads are vital assets and the backbone for any transportation system and support societal development by providing the foundation for constant mobility of goods and people. However, pavements are experiencing accelerated deterioration in most developed countries due to increased traffic volume and load, combined with rapidly changing climate. The existing reactive road asset management approach cannot keep up with the rate of pavement degradation, due to lack of condition data from infrequent inspection surveys and simple models that do not consider the factors influencing pavement performance holistically.

Digital twins have been popularly utilised in recent years enabled by the increasing capacity in data collection using intelligent sensors, digital innovations with technologies such as internet of things, cloud computing, big data analytics with machine learning, as well as artificial intelligence. Despite the growing interest in applications of digital twins in the built environment such as bridges and buildings, current digital twin research related to roads is still at an early stage.

To this end, this study investigates the development of digital twins for the road sector. Based on the literature, a digital twin-based decision-making support theoretical framework for road lifecycle is presented and discussed. In particular, two case studies, as applications of this framework, are conducted to demonstrate the impact of predictive digital twins on roads in the areas of pavement performance and data collection. As part of the road digital twin framework, it is found that integrating physics-based simulation with machine learning, decreased the root mean squared error by at least 25% compared to traditional machine learning in one year prediction, and reduced the 90th percentile range in multi-year predictions by as much as over 30%. In addition, this research also identifies that a substantial amount (approx. over 95%) of sensor data collected could be reduced while achieving acceptable prediction accuracy, thereby minimising the data related costs within the same framework. The findings are useful for the understanding and consideration of the on-going road digital twin development.

Type of Work:

Thesis (Doctorates > Ph.D.)

Award Type:

Doctorates > Ph.D.

Supervisor(s):