Fields Industrial Optimization Seminar
The inaugural meeting of the Fields Industrial Optimization Seminar
took place on November 2, 2004. The seminar meets in the early evening
of the first Tuesday of each month. Each meeting is comprised of
two related lectures on a topic in optimization; typically, one
speaker is a university-based researcher and the other is from the
private or government sector. The series welcomes the participation
of everyone in the academic or industrial community with an interest
in optimization theory or practice, expert or student .
Please subscribe to the Fields mail list
to be informed of upcoming seminars or
The Seminar Series has finished for the 2010-2011 year. Please
see this link
for the 2011-2012 Seminar Series.
| PAST SEMINARS
|Jun 7, 2011
Kim B. McAuley (Queen's University)
Optimization for Development of Reliable Fundamental Models
Fundamental models are used to design, debottleneck and optimize
chemical processes to ensure safe and economical manufacture
of high-quality chemicals, petroleum, plastics and pharmaceuticals.
Reliable optimization results require reliable models. This
talk will focus on some optimization tools and statistical
concepts that can be used during model building. One common
problem that arises when modeling chemical processes is the
large number of parameters that can appear in equations that
describe the rates of chemical reactions. Often, there is
insufficient information in the available data to reliably
estimate all of the model parameters. Parameter estimability
analysis techniques and model-selection criteria will be presented
that can help modellers to estimate parameters, simplify model
equations and design additional experiments to ensure good
model predictions. These methods will be illustrated using
models and data from several different laboratory-scale and
Keith Marchildon (Keith Marchildon Chemical Process
Modeling Successes in a Polymer Production Process
Mathematical modeling of chemical processes is an expensive
business, requiring skilled personnel to create the models
and to secure the supporting data, and requiring long-term
commitment by management. If the models are to be fundamental
rather than statistical then they additionally require persons
with the ability to interpret a process according to principles
of chemistry, physics and engineering. These conditions came
together in support of a particular polymer production process.
The work has been more than justified over the years by a
variety of profitable benefits, many of them only vaguely
foreseen: the very existence of the model was the catalyst
for new methods and entirely new ideas. We present an account
of the architecture of the model, an early major project to
gather and interpret supporting data, some applications in
process optimization and expansion, solution of a complex
problem in process control, and the use of the model in building
an on-line process monitor. Along the way we comment on some
issues in model building such as language, input and output,
maintenance, and use of commercial simulators.
|May 3, 2011
|Andrew Jardine and Dragan Banjevic
(University of Toronto)
On the Optimization of Condition-Based Maintenance Decisions
Condition-based maintenance is a popular maintenance tactic
for expensive, complex, and multi-component physical assets.
The proliferation of condition-monitoring techniques (such as
oil sampling and vibration monitoring) and the ubiquity of maintenance
databases (such as SAP-PM) make it possible to employ a wide
variety of data-driven, evidence-based maintenance policies.
In this presentation some recent developments in condition-based
maintenance for equipment are described, including estimating
the remaining useful life of an asset that is subject to condition
monitoring. Several real-world industrial examples will be introduced
that motivated the work.
Norm Hann (Hydro One)
Closing the Crevice: Achieving Valuable Maintenance Analyses
by Linking Corporate Data with Maintenance Analysis Software
There is currently a gap or crevice between corporate databases
and powerful maintenance analysis software, such as Exakt.
This crevice has impeded the development of usable maintenance
decision models. The CMMS has not ventured into this area,
and general-purpose data warehouses are ill-equipped to handle
the analysis and the complex requirements of maintenance and
reliability. This presentation describes a flexible technique
developed with the needs of reliability analysts in mind.
It enables the automated filtering of large volumes of work
and monitoring data in order to produce the "Events"
and "Inspections" tables of the quality and form
required for analysis, modeling and processing by an Exakt
decision agent. The process will be described using examples
from Hydro Ones experience in the challenging area of
data management and decision making.
|Apr 5, 2011
Boris Mordukhovich (Wayne State University)
GENERALIZED NEWTON'S METHOD BASED ON GRAPHICAL DERIVATIVES
This paper concerns developing a numerical method of the
Newton type to solve systems of nonlinear equations described
by nonsmooth continuous functions. We propose and justify
a new generalized Newton algorithm based on graphical derivatives,
which have never been used to derive a Newton-type method
for solving nonsmooth equations. Based on advanced techniques
of variational analysis and generalized differentiation, we
establish the well-posedness of the algorithm, its local superlinear
convergence, and its global convergence of the Kantorovich
type. Our convergence results hold with no semismoothness
assumption, which is illustrated by examples. The algorithm
and main results obtained in the paper are compared with well-recognized
semismooth and B-differentiable versions of Newton's method
for nonsmooth Lipschitzian equations.
San Yip (Suncor Energy)
Challenges of Integrating Planning and Scheduling in Oil
Decision making hierarchy in process industry is typically
broken down into several layers: Planning, Scheduling, Real-time
Operations Optimization (RTO) and Process Control. Each layer
executes the optimization problem at a different frequency.
Planning and scheduling layers consider a horizon of weeks
to months and their calculations are performed monthly and
weekly, respectively. The RTO calculation is executed every
few hours or minutes, depending on the dynamics of the processes,
to reject disturbances with frequencies higher than those
considered in the planning and scheduling layers. The lowest
process control layer executes its calculations every minute
or second to maintain the controlled variables at the setpoints.
Each layer optimizes its objective based on the decision
from the layer above. Planning layer determines the monthly
average production targets by maximizing long-term economics.
Scheduling layer calculates daily operating targets by minimizing
deviations from the monthly optimum targets determined from
the planning layer. The daily operating targets are then passed
to the RTO layer which optimizes the operating conditions
of process units. Finally, process control layer keeps the
units running at the optimum operating conditions.
This presentation will focus on planning and scheduling layers.
By reviewing current industrial practice, challenges of integrating
planning and scheduling layers will be discussed, and a strategy
to interface commercial planning and scheduling tools for
gasoline blending will be presented.
|Mar 1, 2011
|Nick Sahinidis, Carnegie Mellon University
Global optimization of nonconvex NLPs and MINLPs with BARON
We describe the theoretical and algorithmic foundations of the
branch-and-reduce approach to the global optimization of continuous,
integer, and mixed-integer nonlinear programs. These include
recent convexification strategies for constructing sharp polyhedral
relaxations of the convex hulls of nonlinear problems, domain
reduction techniques, and branching strategies that guarantee
finiteness in certain cases. Applications in a variety of areas
will be addressed and computational results with BARON will
Jose M. Pinto, Praxair R&D
Risk Management in the Industrial Gas Supply Chain
(co-authored by Atul Rangarajan)
The capital intensive industrial gases business involves the
production, distribution and sale of atmospheric gases (argon,
nitrogen and oxygen), carbon dioxide, hydrogen, helium and
specialty gases. Atmospheric gases are produced though cryogenic
processes in air separation plants. There are three basic
distribution methods for industrial gases: (i) via pipelines
(on-site); (ii) as cryogenic liquids via trucks
(merchant liquid); and (iii) as gas in cylinders
(packaged gases). These distribution methods are
often integrated, with products from multiple supply modes
coming from the same plant. The method of supply is generally
determined by the lowest cost means of meeting the customers
needs, depending upon factors such as volume requirements,
purity, pattern of usage, and the form in which the product
is used (as a gas or as a cryogenic liquid). A typical business
model is the so called sale-of-gas in which the
industrial gas company owns and operates the plants. These
capital investments are characterized by long payback periods
and long term contracts with customers. Due to the complexity
and expanding geographic reach of the companys operations,
a wide range of factors, many of which are outside of the
companys control, could materially affect its future
operations and financial performance. For example, the following
risks may significantly impact the company:
Project execution including construction, supplier
risk, new technology
Operations including reliability, maintainability,
Commercial and Market including onsite demand, liquid
demand, energy and raw material availability and costs, contracting
Financial like currency, inflation rates, country/regulatory
Others like general economic conditions, global financial
markets conditions, competitor actions; catastrophic events,
international events and circumstances
The objective of the talk is to discuss the several risk factors
and their impact on the industrial gas supply chain. In addition
we will present work that considers the simultaneous capacity
allocation and distribution planning under demand risk for
an industrial gas supply chain. A stochastic inventory approach
is proposed and it is incorporated into a multi-period two-stage
stochastic mixed-integer nonlinear programming (MINLP) model
to handle uncertainty of demand and loss or addition of customers.
This nonconvex MINLP formulation takes into account customer
synergies and simultaneously predicts the optimal sizes of
customers storage tanks, the safety stock levels and
the estimated delivery cost for replenishments. Three case
studies including instances with up to 200 customers are presented
to demonstrate the effectiveness of the proposed stochastic
models and solution algorithms.
|Feb 1, 2011
||Arian Novruzi, University
Optimal shape design and hydrogen fuel cells
After a general overview of hydrogen fuel cell (HFC) modeling,
we will present the problem of shape optimization of the cathode
channel in HFC.
We consider a two-dimensional isothermal model of gases in
cathode channel and gas diffusive layer (GDL) of HFC, given
by a system of PDEs. This system involves the velocity, pressure
and concentration of oxygen
and water vapor. The objective is to minimize, with respect
to the channel shape, a certain energy functional which ``measures``
the oxygen at the contact of GDL with the catalyst layer,
the water vapor on channel outlet and the pressure drop.
The shape of the cathode channel which minimizes this energy
functional enhances the performance of HFC.
Under some assumptions we prove that this PDE system has
a solution, and that there exists a channel shape, in the
class of Lipschitz channel shapes, minimizing the energy functional.
Using classical shape optimization techniques we prove the
shape differentiability of state variables and of the energy
functional, and we give an explicit expression of the energy
functional shape gradient.
By using an appropriate adjoint problem we transform the
shape derivative of the energy functional in a form appropriate
for numerical computations. Furthermore, we prove that the
adjoint problem is well-posed.
We will conclude with the presentation of several numerical
solutions of optimal channel shape
Jeff Renfro, Honeywell Process Solutions
Overview of a Nonlinear Model Predictive Optimal Control
Technology used in Industrial Process Control Applications
Linear Model Predictive Control (MPC) has been successfully
used to address difficult process control problems in the
chemical and refining industries since the 1970s. It use in
new application areas continues to expand today. The linear
and/or quadratic programming components of this MPC technology
are highly reliable and can provide computed solutions to
large control problems in the required cycle time. However,
there are limitations to the class of control problems this
linear MPC approach can address due to the linear dynamic
models used in the technology. In particular, control of processes
that require changes in operating conditions to regions with
widely different process sensitivities and dynamics are difficult
to manage by linear MPC approaches, due to the dramatic model/process
mismatches that develop.
In the mid 1990s some commercial approaches to using nonlinear
dynamic models in MPC algorithms were developed in the chemical
industry to address the process control challenges that could
not be addressed by linear MPC. Some of these approaches for
nonlinear model predictive control algorithms required the
solution of nonlinear programming problems at each control
cycle. This presented challenges for obtaining predictable
solution times, reliable convergence and insuring physically
meaningful solutions not experienced with linear MPC. In addition,
the combination of prioritized control and optimization (optimal
control) objectives presents a challenging nonlinearly constrained
optimization problem formulation. These issues were addressed
to yield a practical nonlinear MPOC (Model Predictive Optimal
Control) technology that was productized and has successfully
solved a class of difficult process control problems in industry.
This seminar will present an overview of the theoretical
formulation, solution strategies, implementation experience
and benefits of a nonlinear model predictive optimal control
system that has been used in industry since 1994.
|Dec 7, 2010
|Yurii Nesterov, Center for
Operations Research and Econometrics, Université catholique
Recent advances in Structural Optimization
In this talk we present the main directions of research in
Structural Convex Optimization. In this field, we use additional
information on the structure of specific problem instances
for accelerating standard Black-Box methods. We show that
the proper use of problem structure can provably accelerate
these methods by the order of magnitudes. As examples, we
consider polynomial-time interior-point methods, smoothing
technique, minimization of composite functions and some other
Ivan Miletic, ProSensus, Inc.
Data-Driven Models in Industrial Applications of Optimization
Industrial uses of optimization methods cover a wide variety
of engineering applications ranging from process control,
feedstock and product blending, plant operations optimization,
real-time optimization, scheduling, and others.
One key aspect in all of these diverse cases is the need
for data and suitable modelling methods in order to develop
and drive robust optimization solutions. This important aspect
of using data and effective modelling methods to support optimization
is examined in this talk by looking at the application of
multivariate analysis methods that lead to working empirical
models, design of experiments, and improved knowledge and
insight into processes.
The examples examined in this talk include commercial applications
of optimization-based closed-loop batch control in the food
industry, and optimal product design and development. In both
cases, the successful use of optimization methods is tied
to the effective use of the information in process data through
empirical model building.
|Nov 2, 2010
||Tim Davidson, Dept. Electrical
and Computer Engineering, McMaster University
Semidefinite relaxation in action: Efficient "soft"
demodulation for wireless communication systems with multiple
Wireless communication systems with multiple transmit and multiple
receive antennas have the potential to provide reliable communication
at data rates that are substantially higher than those of the
single antenna systems. The core challenge in designing practical
multiple-input multiple-output (MIMO) systems is to achieve
these rates with reasonable computational complexity. A standard
transceiver architecture for moving towards that goal is MIMO
bit-interleaved coded modulation with iterative "soft"
demodulation and decoding (MIMO BICM-IDD). However, the computational
cost of the MIMO soft demodulator increases exponentially with
the number of bits transmitted per channel use, and hence there
is considerable interest in the design of approximate soft demodulation
schemes with lower complexity.
In this talk we will discuss how the power of semidefinite
relaxation can be harnessed to yield a computationally-efficient
approximate MIMO soft demodulator. Semidefinite relaxation
techniques have previously been proposed for "hard"
demodulation problems for which the output is a vector of
binary symbols. Rather than making a "hard" decision
on each bit, soft demodulators seek to provide more information
to the decoder by providing an approximation of the posterior
log likelihood ratio of each encoded bit. A key step in the
development of the proposed soft demodulator is the use of
an approximation of the randomization step in the semidefinite
relaxation technique to generate a list of candidate bit vectors
over which the likelihoods can be approximated.
The talk will include comparisons with the key competing
approaches to MIMO soft demodulation, including the "minimum
mean square error soft interference canceller", and the
various "soft sphere decoders", which have their
roots in tree-search methods for finding the closest point
on a lattice. The computational properties of these algorithms
have some distinct features, and present some interesting
choices to system designers.
This talk is based on work with Mehran Nekuii, who is now
with Wavesat, Montreal; Mikalai Kisialiou, who is now with
Intel, Portland; and Zhi-Quan (Tom) Luo at the University
Ramy Gohary, RIM-Carleton Research Project, Manager
Jointly Optimal Design of The Transmit Covariance and The
Relay Precoder in Amplify-and-Forward Relay Channels
The use of relays in wireless communication networks enhances
the coverage area of these networks and enables them to operate
at higher data rates. The extraction of these gains require
the employment of effective signal processing techniques,
which in many cases are too complex for practical implementation.
Amplify-and-forward is a
computationally efficient scheme that, in some cases, was
shown to provide better performance than more sophisticated
decode-and-forward and compress-and-forward techniques.
In this work we consider designing a rate-optimal amplify-and-forward
relay-assisted communication system in which the relay is
assumed to be capable of sending and receiving at the same
time and the same frequency. In addition to the transmitter-relay
and relay-receiver links, we assume that there is a direct
transmitter-receiver link. In this case a rate-optimal design
of the system involves the joint optimization of the input
signal covariance matrix and the relay precoder; a non-convex
problem with potentially high dimensionality.
To solve this problem, we note that the design problem of
the input covariance is convex for any given relay precoder.
Using this observation, we obtain closed form solutions of
the corresponding Karush-Kuhn-Tucker (KKT) system of equations.
For each of these solutions, we study the corresponding optimization
of the relay precoder. We show that for some of the KKT solutions,
a closed form expression for the optimal precoder can be obtained.
However, for other solutions, we find necessary conditions
that the optimal precoder must satisfy. Finally, for the latter
case, we identify a class of precoders that meet the necessary
conditions. For each case, an efficient algorithm is developed
for obtaining the final pair of input covariance and relay
This is joint work with Professor Halim Yanikomeroglu, of
Carleton University, and is funded by Research In Motion (RIM).
|Oct 5, 2010
Pietro Belotti, Clemson University
Couenne, an Open-Source solver for non-convex Mixed Integer
Mixed integer nonlinear programming (MINLP) problems are
among the most general and difficult in Optimization, especially
if the nonlinear functions expressing the objective or the
constraints are also non-convex. Because of their non-convexity,
an optimal solution is in general sought using branch-and-bound
techniques. These methods recursively partition the feasible
set and obtain a lower bound on the optimal solution value
by generating convex relaxation of the original problem.
The talk focuses on Couenne (Convex Over- and Under-ENvelopes
for Nonlinear Estimation), an Open-Source software package
whose development started within a collaboration between Carnegie
Mellon University and IBM, and that is part of the Coin-OR
initiative. Couenne is a branch-and-bound method which implements
several techniques for obtaining tight lower bounds, heuristics
for feasible solutions, and procedures for reducing variable
bounds. We describe its main features and show how it can
be used, extended, and adapted to solve several classes of
Alkis Vazacopoulos, FICO
Using Mixed Integer Programming to Solve Sequencing, Scheduling
and Packing Problems
Recent advancements in Mixed Integer Programming solvers
give us the ability to solve larger and more complex sequencing
, scheduling and packing problems. We will demonstrate this
fact by showing examples from tournament scheduling, space
retail optimization, production scheduling and sequencing
in energy applications.