Control theory, distributed parameter systems, and process systems engineering
Data analytics, machine learning (transfer learning, reinforcement learning, etc.), and artificial intelligence
Intersection between machine learning and control theory
Applications in smart sensing and automation of energy, manufacturing, and agriculture systems
Robust and constrained interface tracking using convolutional neural networks and optimistic moving horizon estimation
We propose a robust video-based interface estimation framework for primary separation vessels, combining CNN-based sensing with optimistic moving horizon estimation (OMHE). Designed to handle visual outliers such as occlusion and blurriness while respecting constraints, the method enables real-time monitoring with stability guarantees. Its effectiveness is demonstrated in both lab-scale and industrial case studies.
The detailed research can be found in J Xie, H. Liang, M. B. Tatlici, B Huang, JPC, 2025.
Moving horizon estimation for pipeline leak detection, localization, and constrained size estimation
Developing effective leak detection and localization methods is challenging due to the spatiotemporal dynamics of process variables, the presence of process/measurement disturbances and constraints, and the limited measurement data. To address this issue, this manuscript proposes a novel moving horizon estimation design for pipeline leak detection, constrained estimation of leak size and location by using an infinite-dimensional pipeline hydraulic model.
The detailed research can be found in J Xie, B Huang, S Dubljevic, CACE, 2024.
Back-and-forth nudging moving horizon estimation for discrete-time linear systems
To address the finite-sample issue in state estimation, we propose a novel back-and-forth nudging moving horizon estimation algorithm for discrete-time linear systems. The unbiasedness properties and comparison results with the conventional (forward-in-time) moving horizon estimation are discussed. The developed results are promising for safety-critical systems.
The detailed research can be found in J Xie, B Huang, Automatica, 2024.
Reinforcement learning for soft sensor design through autonomous cross-domain data selection
To address data availability issue in soft sensing of industrial processes, we propose novel asynchronous advantage selector-actor-critic RL method for cross-domain feature/sample selection and soft sensor design, with application in industrial SAGD processes.
The detailed research can be found in J Xie, O Dogru, B Huang, C Godwaldt, B Willms, CACE, 2023.
Approximate Moving Horizon Estimation for Switching Conservative Linear Infinite-Dimensional Systems
For health monitoring of distributed parameter systems with switching parameters, we develop optimization-based moving horizon estimation theory and algorithm, with application in a switching Schrödinger equation.
The detailed research can be found in J Xie, JP Humaloja, CR Koch, S Dubljevic, Automatica, 2023.
Constrained Receding Horizon Output Estimation of Regular Linear Distributed Parameter Systems
To address constrained output estimation and of complex systems with spatiotemporal dynamics, we propose new receding horizon estimation theory and algorithm, with applications in a 1D wave equation and a 1-D diffusion equation.
The detailed research can be found in J Xie, JP Humaloja, CR Koch, S Dubljevic, IEEE TAC, 2022.
Transfer Learning for Dynamic Feature Extraction Using Variational Bayesian Inference
To deal with the distribution drift and the unavailability of monitoring data, we propose a novel online transfer learning technique (transfer slow feature analysis) to dynamically learn cross-domain features based on the variational Bayesian inference, with application in soft sensing of industrial SAGD processes.
The detailed research can be found in J Xie, B Huang, S Dubljevic, IEEE TKDE, 2021.
Health monitoring and advanced process control of gas/oil/water pipeline networks
To ensure the operational safety, efficiency, demand, profitability, and sustainability of the gas/oil/water pipeline transportation networks, we propose a series of advanced hybrid observers, model predictive control (MPC), internal model control (IMC), and machine learning techniques for pipeline monitoring, leak detection, localization, size estimation, flow regulation, disturbance rejection, and strain demand prediction of buried steel pipeline under geo-hazards (e.g. at strike-slip fault crossings).
The detailed works are in J Xie, X Xu, S Dubljevic, AIChE J, 2019; L Zhang, J Xie, S Dubljevic, CACE, 2023; J Xie, S Dubljevic, JPC, 2021; J Xie, L Zhang, Q Zheng, X Liu, S Dubljevic, H Zhang, EAS, 2021.
