Research Interests
Our
research interests are in the general areas of
control, dynamics and optimization including both theory and
applications, computational process modeling and simulation, and applied mathematics. The central objective of our research is
the development of novel methods for the systematic and rigorous solution of complex
process control and systems engineering problems. Our group has
pioneered the development of nonlinear model-based control methods for
broad classes of nonlinear, hybrid and distributed parameter processes and has
demonstrated their successful application to complex industrial
processes. Our research team
is part of the UCLA Center for Systems,
Dynamics and Control and of the
Process Systems Engineering Group.
Our research work is carried out in collaboration with experimental
groups and major industrial companies. The
following research directions are being pursued:
Statistical machine learning in process modeling
and control
The goal of this research is to employ and further develop the
methodological framework of generalized error bounds from machine
learning theory for the development and verification of machine
learning (ML) models with specific theoretical accuracy guarantees and
integrate these models into model predictive control (MPC) system design
for nonlinear chemical processes. This research focuses on the following
broad objectives: a) the development of generalized probabilistic error
bounds for ML models accounting for the impact of the number of neurons
and layers on accuracy and guiding network structure selection and
training, b) the design of model predictive control schemes that
incorporate ML models that are computationally-efficient and ensure
desired closed-loop stability, performance, robustness and operational
safety properties, and c) the development of a methodology for on-line
adaptation of ML models in MPC to capture changing process dynamics and
model uncertainty using real-time noisy data. The machine learning
modeling and control methods are tested using large-scale process
simulators incorporating industrial-data informed parameters and noise,
and an experimental electrochemical reactor system at UCLA ((in
collaboration with Professor C. Morales-Guio)) employed to
convert carbon dioxide to liquid fuels.
Process operational safety and cybersecurity
This work focuses on the development of rigorous, yet
practical, methods for the design of optimization-based control systems
that improve process operational safety and cybersecurity for broad
classes of nonlinear processes. The research addresses two key
directions: a) the design of novel model predictive control systems that
explicitly account for operational safety considerations and ensure
simultaneous closed-loop stability and safety, and b) the design of
integrated machine-learning-based detection and control systems that can
improve process cybersecurity by efficiently detecting and mitigating
the impact of intelligent cyber-attacks. The methods are being tested
using large-scale chemical process network simulators and industrial
data. Networked, distributed and economic
model predictive control of nonlinear processes
Process control systems traditionally utilize dedicated,
point-to-point wired communication links to measurement sensors and
control actuators, to regulate process variables at desired values.
While this paradigm to process control has been successful, we are
currently witnessing an augmentation of the existing, dedicated control
networks, with additional networked (wired and/or wireless)
actuator/sensor devices which have become cheap and easy-to-install the
last few years. Such an augmentation in sensor information and networked-based availability of data has the potential to be transformative in the
sense of dramatically improving the ability of the control systems to
further optimize process performance and fault-tolerance. However,
augmenting dedicated, local control networks with real-time
wired/wireless sensor and actuator networks challenges many of the
assumptions in traditional process control and monitoring methods
dealing with dynamical systems linked
through ideal channels with flawless, synchronous communication. In
the context of designing process control systems which utilize
sensor and actuator networks, key fundamental issues that need to be
addressed include the use of asynchronous and delayed measurements in
the control system as well as the occurrence of network malfunctions due
to field
interference and device power losses. To address these fundamental
problems, our research currently focuses on the development of the
theory and algorithms needed for the design of networked and distributed
predictive control
systems accounting explicitly for asynchronous and delayed measurements
and random network malfunctions. In this direction, the integration of economic optimization and feedback control in the context
of economic model predictive control is currently investigated in
our group.
Multiscale computational fluid dynamics
modeling, control and optimization of multiscale process systems
Interest in the control and optimization of multiscale
(deterministic/stochastic) process systems has been triggered by the
need to achieve tight feedback control of complex processes, such as
deposition and sputtering of thin films in semiconductor manufacturing
and solar-cell systems,
which are characterized by highly coupled phenomena occurring at
disparate time and length scales. We develop general methods for
multiscale computational fluid dynamics modeling,
reduced-order modeling and feedback controller synthesis for multiscale
systems that efficiently address coupled macroscopic and microscopic
(e.g., thin film roughness, height-height correlation and porosity) objectives, and illustrate
their application to thin-film growth and sputtering processes of
industrial interest. Specifically, we develop: a) multiscale
computational fluid dynamics modeling archtectures that allow to build
accurate models of industrial processes, b) detailed microscopic modeling
approacheswith emphasis on the theory and
implementation of efficient kinetic Monte Carlo simulation, c) methods for
stochastic model construction and parameter estimation, and d) methods
for optimization, predictive and covariance controller design using stochastic partial
differential equation models.
Control of nonlinear and hybrid process networks
Chemical process networks are inherently nonlinear and cannot be
effectively controlled and monitored with conventional control and
estimation schemes which are developed on the basis of linear or linearized process models. To enhance our ability to operate chemical
process networks, our research focuses on: a) the development of a rigorous and
practical framework for nonlinear model-based control of chemical
processes that explicitly accounts for the presence of uncertainty in
the process model, and constraints and time-delays in the control
actuators and measurement sensors, b) the development of nonlinear state
estimation algorithms for process monitoring, and c) the development of a comprehensive
framework for nonlinear model-based feedback control of hybrid nonlinear
processes (i.e., processes with combined continuous dynamics and
discrete events) and large-scale process networks with significant
interactions. The theoretical studies are coupled with applications
to complex process networks used in the chemical and petroleum industry.
Model reduction, optimization
and control of nonlinear distributed parameter systems
Distributed parameter systems (DPS) like integro-differential equations and
partial differential equations arise naturally in the mathematical modeling of particulate
and transport-reaction processes. The main feature of DPS is that they are
characterized by infinite dimensional dynamic behavior. Therefore, it is impossible
to perform dynamical analysis, optimization and design, and synthesize
practically-implementable controllers for particulate and transport-reaction processes
based on full distributed parameter models. The objectives of this research are:
a) the development of model reduction methods for the derivation of low-order
systems that accurately reproduce the solution and dynamics of a DPS, b) the development
of optimization algorithms and the synthesis of nonlinear controllers, which are
guaranteed to work for the DPS, based on the low-order approximations, and c) the
development of a general framework for integrated optimal design and control of DPS.
The theoretical results are applied to particulate processes like
continuous and batch crystallization and aerosol production for particle
size distribution control, as well as to transport-reaction processes
used in advanced materials and semiconductor processing including
crystal growth, rapid thermal processing, and plasma-assisted chemical
vapor deposition and etching.
Fault-tolerant control and monitoring
Increased process automation tends to increase the
vulnerability of the
process to faults (e.g., defects/malfunctions in process equipment,
sensors and actuators, failures in the controllers or in the control
loops) potentially causing a host of safety, environmental and economic
problems. Many recent incidents are chilling examples of faults that
turned into disasters. Management of
abnormal situations is a major challenge in the chemical and process
industries since abnormal situations account annually at least for
$10 billion in lost revenue in the US alone. We work on the development
of
model-based based and data-based process monitoring methods which
utilize information from sensor networks to achieve quick and accurate
fault detection and isolation, as well as we develop fault-tolerant
control re-configuration strategies that achieve continuous, safe and
economically-optimal process operation in the event of faults.
Water
and energy system modeling and control
In this direction, our
research (in collaboration with Professor Y. Cohen) focuses on the
modeling, analysis and control of water processing and distribution
systems with particular emphasis on real-time fault diagnosis and
control of water desalination plants for energy-optimal operation. This research covers both
development of user-friendly software tools for the practical
implementation of UCLA-developed algorithms and applications to
experimental systems and pilot plants as well as the efficient
integration of water systems with renewable-energy systems.
Modeling and control of particulate systems
Advances in on-line particle size and shape distribution measurement including
laser absorption scattering and probe sampling techniques provide the
means for achieving real-time model-based feedback control of fine
particle synthesis and processing. We develop a systematic multiscale
approach to real-time control of processes involved in the synthesis
(e.g., crystallizers, aerosol reactors) and processing (e.g., thermal spray
processing of nanostructured coatings using nanosized powders) of fine
particles. We address the development of low-order approximations of
multiscale models linking macroscopic scale (e.g., thermal spray
process) and microscopic scale (e.g., evolution of coating
microstructure) and the integration of models, measurements and control
theory to develop real-time feedback control systems. Recent work
in the broader area of particle technology also focuses on the
development of a computational framework for controlling crystal size
and shape in protein crystallization.
|
