Home
|
About Me
|
Classes
|
Research Interests
|
Publications
|
Research Data Sets
|
Research Group
|
Gross Error Detection
and Data Reconciliation
The fault detection area is
concerned not only with detection of plant process and equipment faults, but
also with identification and correction. With advancements in computer
technology the last three decades or so, a number of approaches have appeared
from those that rely strongly on knowledge to those that are strongly data
driven. This approach is somewhere between these two extremes and requires a
model of the process in terms of material and energy balance closure. This
knowledge is exploited along with statistical knowledge of measurement noise
behavior to create effective methodologies to estimate the values of process
variables under faulty measurement devices. A program was launched with Rollins
and Davis (1992) and introduced a methodology we called ¡§the unbiased
estimation technique¡¨ or ¡§UBET.¡¨
Predictive Modeling and Control
This
research focuses specifically in block-oriented modeling (BOM) to describe nonlinear
static and nonlinear dynamic behavior. In BOM, nonlinear (N) static and linear
(L) dynamic functions are connected in parallel or series or both, in an arrangement
that can be described in a block diagram. The work to date has consisted mainly
in NL (Hammerstein) or LN (Wiener) arrangements which are the simplest and most
popular. My current BOM thrust is NLN and LNL sandwich systems. My most
significant contributions in BOM have been the development of multiple-input,
multiple-output (MIMO) techniques for NL and LN discrete-time (DT) and NL and
LN continuous-time (CT) modeling under N-Block non-invertiblilty.
I have focused on the development of CT prediction algorithms with
closed-formed exact solutions to BOM.
Human Thermoregulatory Modeling
HTR modeling is critical in the development of
clothing, environmental suits, and equipment for working in extreme
environmental conditions or humans that have thermoregulation challenges in
normal environments. There have been two extreme approaches in HTR modeling ¡V
theoretical, which has the drawback of lacking sufficient knowledge to
accurately model individuals; and empirical, which has the drawback of being
limited to the subjects and conditions of the data used for modeling. Since BOM
is semi-empirical, it able to overcome these limitations.
Bioinformatics
We have recently developed a powerful method for
developing assay-specific gene signatures from microarray
expression data. This methodology is based on the statistical multivariate
method called ¡§principal component analysis (PCA)¡¨ which is a data mining
method that is able to extract biological relationships from large data sets.
There are several unique aspects of our approach in signature development.
First, it exploits both eigengenes and eigenassay principal components. Secondly, it is the only
method to determine and use gene or assay contribution.
Thirdly, signatures are determined with all the genes as possible candidates
and ranked order signatures of the genes are provided using special criteria
for signature size developed in this work.
Chemical Vapor
Infiltration
Chemical vapor infiltration (CVI) which involves the
development of light weight carbon/carbon composites with high temperature,
high strength, and low wear properties. The challenge in producing these
composites is to reduce development time by maximizing deposition rates which
are challenged by deposition profiles in porous structures that are difficult
to control. My program focuses on building CT dynamic and spatial models over
critical response spaces such as temperature and infiltration density.
Current Thrusts
Research Program in Human Thermoregulation
The
basic goal of my research program in human thermoregulation (HTR) modeling is
to provide an ability to produce models for individuals using easily attainable
personal characteristics and property data. Now that we have demonstrated the
ability of our BOM approach to model the HTR system, we are working to prove
that we can produce models for individuals without subjecting them to
environmental chambers for data collection. We have developed a study that has
IRB approval that is designed to demonstrate this ability.
Research Program
in Type 2 Diabetes
My approach is to produce a modeling method capable
developing predictive models for individuals from noninvasive data under
free-living conditions. With this model, an individual will know how to
personally change their lifestyle to get optimal results. They will also get
immediate feedback, via model prediction, of the consequences for ¡§cheating¡¨ in
certain ways. Although no method has demonstrated an ability to model TTD from
noninvasive variables, we have seen promising results with our BOM approach
from studies using a limited number of inputs. Another modeling objective is
the determination of sub-classes of TTD. The goal here is to determine behavior
profiles that optimize glucose control for particular classes. To accomplish
this goal we will apply informatics to the large volumes of data that we will
collect and apply classification methods. Long term, my goal is to develop
software packages and training methods to assist medical workers with the
implementation of the methods that we will developed.
Research
Program in Predictive Modeling and Control
My
predictive modeling research will focus on the development of sandwich BOM. One
type will be NLN models which occur in practice when an input to a Wiener (LN)
type system passes through equipment that behaves as a nonlinear static
process, e.g., pressure drop through an orifice plate sensor used for flow
rate. Other types will be LNL and NLNL models which occur in practice when an
input to a Hammerstein (NL) type system has L or NL dynamic behavior. My
current research focus strongly on advancing model predictive control (MPC) by
modeling input behavior and using these models to provide more accurate values
for input changes. We will apply our sandwich modeling methods to determine the
dynamic behavior, the sinusoidal methods that we have developed to model
periodic behavior, and the CT stochastic process methods we are developing to
treat this type of behavior.
Research and Teaching Program in Material and
Bioinformatics
The
recent development of our PCA method to determine assay-specific signatures has
catalyzed our informatics program. Not only will I be extending this method
more broadly to similar areas such as proteomics and metabolomics,
but also to chemical and materials informatics. We used this method to analyze combi-experiments in catalysis and an analysis focusing on
frequency contribution for neutron spectroscopy data from the cyclic
deformation of a cobalt super-alloy. An outgrowth of our work will be the development
of multidisciplinary courses in informatics to support directions the college
is taking in the development of a materials informatics program.
Research Program in Carbon/Carbon Composites
Development
of dynamic/spatial CT models from experimental data as well as the techniques
is focuses work in this area. Methods to estimate rate constants for gas phase
reactions as a function of temperature from simulated and real data will be
developed. For a thermal gradient CVI process using real data from the
literature we developed a CT dynamic/spatial model for temperature. We are in
the process of developing CT dynamic/spatial models for temperature and pore
volume from simulated data for an isothermal/isobaric CVI process. This work
will be key in learning how to develop these types of
models using plant data. Application of the results will include optimization
and control.
Research Program in Dynamic/Spatial Modeling of Drug
Delivery Data
Another
application of our ability to develop dynamic/spatial models is in modeling
drug release data from pH and temperature-sensitive polymer systems. These
dynamic models are extremely useful to predict and control the modulated drug
release behavior from such systems.