Research

Overview

My research focuses on intelligent monitoring and control of complex industrial processes, with the goal of improving the safety, reliability, and intelligence of industrial systems. I develop theories and algorithms spanning three interconnected pillars:

1. Data-Physics Hybrid Modeling for Complex Industrial Processes

I develop methods that fuse data-driven machine learning with first-principles knowledge to build high-accuracy, physically consistent models. Key contributions include incorporating mechanistic constraints into neural network architectures, multi-source data fusion with iterative optimization, knowledge transfer using causal reasoning, and constructing reliable transfer learning models grounded in domain knowledge. These methods have been applied to refinery digital twins, biofeedstock co-processing, and renewable CO2 tracking.

Soft Sensors Hybrid Modeling Transfer Learning Physics-informed ML Digital Twins

2. Trustworthy Intelligent Monitoring and Fault Diagnosis

I address the "black-box" bottleneck in industrial AI by developing interpretable and reliable monitoring systems. My work includes causal discovery algorithms (including a novel polynomial chaos framework published at ICLR, CVPR, AAAI 2026), virtual sample generation with causal learning, alarm sequence alignment and threshold optimization, and attack detection for cyber-physical systems using causal representations.

Causal Discovery Process Monitoring Alarm Management Polynomial Chaos Interpretable AI

3. Multi-Channel Fault-Tolerant Control

I design control strategies that maintain system stability and performance under multi-type, multi-channel faults common in industrial settings. This includes mathematical modeling of multi-channel faults in 2D systems, observer design for fault estimation, and nonlinear time-varying fault-tolerant control with provable stability guarantees.

Fault-Tolerant Control 2D Systems Observer Design Robust Control

Application Domains

My research has been successfully deployed in real-world industrial scenarios:

Biopharmaceutical Manufacturing: Core participant in MIT/US FDA continuous mRNA manufacturing platform project
Refinery Digital Twins: Developed soft measurement and big data analytics for UBC-Parkland Corporation, deployed in production
Smart Energy Systems: Microgrid digital twin modeling and optimal control (NRC Canada)
Healthcare: Real-time intelligent monitoring for emergency departments (Vancouver Coastal Health)
Lithium-ion Battery Management: State of health estimation and capacity prediction
AI-Accelerated Materials Discovery: Inverse design using machine learning (Mitacs)