Dr Yiming Zhang
- Marie Sklodowska-Curie Future Roads Fellow
- Young Chair Professor, School of Civil Engineering, Southeast University, Nanjing, China
Connect
Location
- Digital Roads of the Future - Data Science
About
Dr Yiming Zhang was a Maria Skłodowska-Curie Future Roads Fellow at the University of Cambridge (UOC). He received his PhD degrees in Civil Engineering at Monash University (Australia) and Southeast University (China) in 2021. Before joining the UOC, he worked as a Postdoctoral Fellow at the Hong Kong Polytechnic University (PolyU) from 2021 to 2022, funded by PolyU Postdoc Matching Scheme. Throughout his professional career, Dr Zhang has attained extensive industrial experience and written more than twenty well-recognized publications regarding developing probabilistic machine-learning methods to address engineering problems from a data-driven perspective. He also serves as an editor in Frontiers in Built Environment and a reviewer for over ten high-impact international journals. Dr Zhang was a recipient of Best Paper Awards (2021), Key Scientific Articles in Advances in Engineering (2021), First-Class Award of Science and Technology of China highway society (2021), and Outstanding PhD thesis of Jiangsu Province (2022).
Fellowship period: 13 Sep 2022 - 18 Dec 2023
Research
Research Interests
Dr Yiming Zhang’s research interests primarily lie in the crosscutting field of data science and engineering, including probabilistic machine learning algorithms for predictive modelling, analysis, and optimization; structural health monitoring of civil infrastructures; Bayesian inference for uncertainty quantification; and digital twin-enabled intelligent maintenance. He dedicates to the research and development of data science in Civil Engineering, with both fundamental investigations and real-world engineering applications.
Strategic Themes
Data Collection, Processing and Analysis
Anomaly detection and condition assessment based on Bayesian dynamic model.
Probabilistic Predictive Model
Probabilistic prediction of time-series data using Bayesian and deep learning methods.
Spectrum analysis
Adaptive multi-taper approach for spectrum and coherence estimation of stationary and nonstationary processes.
Research Project
This project aims to develop an automated and efficient decision-making tool through the digital twin, reinforcement learning, and Bayesian dynamic linear model involving computational efficiency, readily interpretability, and uncertainty quantification.
Project Title: Digital twin-driven Bayesian dynamic linear model and reinforcement learning for predictive maintenance of pavements
Abstract: Highway agencies are facing with challenges of ageing pavements, deteriorating networks, and limited budgets, highlighting the importance of effective maintenance strategies. Predictive maintenance is a prominent strategy to deal with the above issues, which enables pre-failure interventions based on the condition monitoring and prediction models. The digital twin (DT) integrates a physical-virtual connection that paves the way to predictive maintenance by monitoring, modelling, and analyzing through the lifecycle of pavements. However, the DT-based predictive maintenance suffers the following limitations: (1) anomalies that could affect the reliability of pavement performance prediction inevitably exist in the measurements; (2) the computationally expensive black-box models with inefficient interpretability and uncertainty consideration are usually adopted to predict pavement performance; and (3) it is challenging to design an efficient reinforcement learning (RL) architecture for maintenance planning optimization. This project aims to develop an autonomous and efficient decision-making tool through the DT, RL, and Bayesian dynamic linear model involving computational efficiency, readily interpretability, and uncertainty quantification. It will focus on establishing the system for automatic data anomaly detection and correction, developing an effective probabilistic prediction model for pavement performance, and presenting the RL framework with efficient modelling and training for maintenance decision-making. The technical results will be finally integrated into the DT platform. Three outcomes will be (i) work toward developing autonomous and maintenance and communicated through publications, conferences, and meetings; (ii) development of the applicant’s modelling, analysis, and communication skills; and (iii) strategic collaborations between partner organizations.