Research Interests
Estimation
Kalman filtering | Approximate nonlinear filtering | Moving horizon estimation | Simultaneous input and state estimation | Distributed estimation and fusion
Kalman filtering | Approximate nonlinear filtering | Moving horizon estimation | Simultaneous input and state estimation | Distributed estimation and fusion
System identification
Nonlinear identification | Closed-loop identification | Continuous-time system identification | Model reduction
Nonlinear identification | Closed-loop identification | Continuous-time system identification | Model reduction
Control
Adaptive control | Robust control | Optimal control | Model predictive control
Adaptive control | Robust control | Optimal control | Model predictive control
Optimization
Convex optimization
Convex optimization
Applications
Energy management | Oceanography | Mechatronic systems | Cyber-physical systems
Energy management | Oceanography | Mechatronic systems | Cyber-physical systems
High-Performance Battery Management
Batteries are ubiquitous, having an irreplaceable role in the age of electrification. Their popularity is not only limited to consumer electronics like smartphones and tablets and various power tools. More than that, they are being used in electric vehicles and even in smart grids as large-scale energy storage devices. Many battery-related applications require effective battery management, which monitors the battery status and regulates the charging and discharging processes for real-time battery protection and performance enhancement.
My research has been dedicated to addressing several key challenges arising in battery management: state-of-charge (SoC) estimation, state-of-health (SoH) estimation and optimal charging and discharging strategies. An indicator of the present battery capacity, the SoC needs to be estimated from the applied current and the terminal voltage. Currently, questions are being raised about how to do SoC estimation even in the presence of uncertain and time-varying battery dynamics. Our answer is to build adaptive SoC approaches to estimate the SoC and the model parameters simultaneously. We have developed high-fidelity and easty-to-implement methods for adaptive SoC estimation. Preliminary studies suggest that such methods yield the best potential for estimating the SoC adaptively and accurately. The success would lead to further improvement and provide strong incentives for development of more adaptive SoC estimation algorithms.
Monitoring the health status of a battery is another central problem. We are developing simple yet promising solutions based on the system identification theory to estimate the parameters indicating the health. We are also conducting research on optimal control for batteries to improve their performance and extend their life.
My research has been dedicated to addressing several key challenges arising in battery management: state-of-charge (SoC) estimation, state-of-health (SoH) estimation and optimal charging and discharging strategies. An indicator of the present battery capacity, the SoC needs to be estimated from the applied current and the terminal voltage. Currently, questions are being raised about how to do SoC estimation even in the presence of uncertain and time-varying battery dynamics. Our answer is to build adaptive SoC approaches to estimate the SoC and the model parameters simultaneously. We have developed high-fidelity and easty-to-implement methods for adaptive SoC estimation. Preliminary studies suggest that such methods yield the best potential for estimating the SoC adaptively and accurately. The success would lead to further improvement and provide strong incentives for development of more adaptive SoC estimation algorithms.
Monitoring the health status of a battery is another central problem. We are developing simple yet promising solutions based on the system identification theory to estimate the parameters indicating the health. We are also conducting research on optimal control for batteries to improve their performance and extend their life.
Ocean Observing Using Drifters
The constantly moving ocean water has a significant influence on the marine environment, climate change and human life. A fundamental issue in oceanography is to monitor the ocean and its water flows with sufficient accuracy. An ocean observing system is being developed based on small, inexpensive drifters to monitor the ocean flows and the composition, plants, animals, nutrients and pollution, etc. In this system, a group of drifters are launched. Each of them follows a submerging-surfacing (or going down-up) movement pattern to traverse the ocean. It also measures the marine biological and physical variables and records its motion information (mainly acceleration). When at surface, it determines its position by GPS and transmit the data to the remote central server for analysis.
A crucial part of the observing system is how to analyze the data. Currently, my interest and efforts have been focused on reconstruction of the flow velocity field of an ocean domain since flows play a key role in phenomena such as the transportation of nutrients, the motion of biological species in their early life, and the diffusion of contaminants and algal blooms. We tackle the problem using two approaches. The first one is nonlinear simultaneous input and state estimation, because the flow velocities can be regarded as the unknown input that drives the lateral motion of a drifter. We have developed a series of algorithms, including filtering and smoothing, to accomplish the goal. The other solution is based on application of nonlinear filtering methods to ocean models with measurements collected from drifters.
Ocean observing and monitoring using underwater vehicles like drifters will stimulate further interest among both the oceanography and control communities. For control theorists and practitioners, it will be a place where filtering, distributed control, sensor networks and optimization will meet.
A crucial part of the observing system is how to analyze the data. Currently, my interest and efforts have been focused on reconstruction of the flow velocity field of an ocean domain since flows play a key role in phenomena such as the transportation of nutrients, the motion of biological species in their early life, and the diffusion of contaminants and algal blooms. We tackle the problem using two approaches. The first one is nonlinear simultaneous input and state estimation, because the flow velocities can be regarded as the unknown input that drives the lateral motion of a drifter. We have developed a series of algorithms, including filtering and smoothing, to accomplish the goal. The other solution is based on application of nonlinear filtering methods to ocean models with measurements collected from drifters.
Ocean observing and monitoring using underwater vehicles like drifters will stimulate further interest among both the oceanography and control communities. For control theorists and practitioners, it will be a place where filtering, distributed control, sensor networks and optimization will meet.
Active Vibration/Noise Reduction
Humans and machines alike love peace. Traditional absorber-based methods for reducing unwanted vibration or sound have poor performance at low frequencies. Hence, active vibration/noise control, which suppresses vibration/noise by applying force/sound with equal magnitude and opposite phase, has been given much attention and led to many significant industrial applications.
We are interested in an even more challenging problem - rejecting the external time-varying vibration/noise disturbance in the presence of system uncertainty, understanding that uncertainty is widely present in real-life plants. The solution is to design an adaptively tuned robust controller. Based on approximate knowledge on system dynamics, adaptive tuning of a feedback controller is done within the framework of a Youla parameterization. The Youla parametrization provides the benefit of tuning directly based on closed-loop time domain measurements, whereas an uncertainty description based on the Youla parametrization allows the formulation of explicit bounds on the tuning parameter to guarantee stability robustness.
The approach has been tested on a benchmark problem that involves the tuning of an active damper system for vibration suppression subject to unknown disturbances (see the left figure). It is an active suspension system using an inertial actuator and equipped with a shaker and a measure of the residual force.
We are interested in an even more challenging problem - rejecting the external time-varying vibration/noise disturbance in the presence of system uncertainty, understanding that uncertainty is widely present in real-life plants. The solution is to design an adaptively tuned robust controller. Based on approximate knowledge on system dynamics, adaptive tuning of a feedback controller is done within the framework of a Youla parameterization. The Youla parametrization provides the benefit of tuning directly based on closed-loop time domain measurements, whereas an uncertainty description based on the Youla parametrization allows the formulation of explicit bounds on the tuning parameter to guarantee stability robustness.
The approach has been tested on a benchmark problem that involves the tuning of an active damper system for vibration suppression subject to unknown disturbances (see the left figure). It is an active suspension system using an inertial actuator and equipped with a shaker and a measure of the residual force.
Networked Control Systems
The development of control theory during the past five decades has largely been built with the abstraction that information is transmitted along perfect communication channels, with numerous success stories witnessed and reported. However, the demand of future industrial applications for teleautomation, modularity, reduced complexity, quick maintenance and decentralization of control will stimulate the construction of control loop over real-time communication networks to connect sensing, actuating and computing components.
Though bringing many benefits, the introduction of networked communications presents some new challenges, one of which is missing data caused by packet dropouts. Our research was concerned with two key issues, system identification and adaptive control. To identify the parameters of a system in a network environment, we modified the classical Kalman filter to obtain an algorithm capable of handling missing output data caused by the network medium. We also established its convergence properties, proposing the conditions to ensure that the parameter estimates given by the algorithm converge to the truth. We further developed an adaptive control scheme for networked systems such that the output of the plant can track expected signals even though the measurements are missing.
Though bringing many benefits, the introduction of networked communications presents some new challenges, one of which is missing data caused by packet dropouts. Our research was concerned with two key issues, system identification and adaptive control. To identify the parameters of a system in a network environment, we modified the classical Kalman filter to obtain an algorithm capable of handling missing output data caused by the network medium. We also established its convergence properties, proposing the conditions to ensure that the parameter estimates given by the algorithm converge to the truth. We further developed an adaptive control scheme for networked systems such that the output of the plant can track expected signals even though the measurements are missing.