CIM PROGRAMS AND ACTIVITIES

Fields Industrial Optimization Seminar
2009-10

When applied to molecular conformation, graph localization refers to computations that are used to determine atomic coordinates when given all or some subset of the inter-atomic distances. The problem becomes somewhat difficult when the inter-atomic distances are all below some low threshold, say 6 Angstroms, and even more of a challenge when the data are corrupted by noise. This last possibility is typical of the data that are provided by nuclear magnetic resonance experiments, a technique that has gained considerable recent importance for the structural determination of macromolecules that are difficult to analyze using conventional x-ray methods. This talk will review some of the computational approaches to graph localization, highlighting the extra assumptions that are put in place when the data are both sparse and approximate. This discussion concludes with a presentation of our approach which involves the
following steps: A biologically motivated heuristic strategy is used to define a set of overlapping cliques in the graph representing the problem. We then apply an iterative procedure that alternates between truncation of a spectral decomposition and the averaging of distances contained in the overlap regions. After these smoothing operations are completed, the coordinates of atoms can be computed for each clique. Since each clique will have its own frame of reference, the remainder of the algorithm uses a progressive coalescence approach to bring all the cliques into a common frame of reference. We conclude the talk with some future directions dealing with geometric analysis of ligand binding to drug receptor sites.

============

Zsolt Zsoldos, SimBioSys, Toronto
Algorithmic and mathematical challenges in protein-ligand docking and scoring

Scientific and technological advancements during the past two decades have altered the way pharmaceutical research produces new bioactive molecules. Traditional "trial and error" drug discovery efforts are gradually replaced by structure based rational drug design, virtual screening(VS) and 3D modelling. There are two approaches for VS, ligand-based similarity search and docking. The flexible ligand docking problem is divided into two subproblems: pose/conformation search and scoring function. The search algorithm must be fast, provide a manageable number of candidates, and be able to find the optimal pose/conformation of the complex. In eHiTS, both the receptor cavity and the candidate ligands are described by geometric shape and chemical feature graph based on distorted polyhedra. The graphs are mapped to each other with a specialized clique detection algorithm. A linear time complexity matching algorithm will also be presented that guarantees the solution with optimal score. The scoring sub-problem can be solved with various approaches including physical, empirical and statistics based models of the interactions. A new statistical model will be described that is based on interacting surface points and their normal vectors.

Larry Megan, Advanced Control and Operations Research R&D, Praxair Inc.
Optimizing the Industrial Gas Supply Chain Across Organizational Boundaries

Global competition and increasing energy costs make it more important than ever to optimize the energy efficiency of US manufacturing industries. While several industries and organizations have focused on the efficiency of individual units or plants, organizational boundaries have limited most from doing a higher level optimization of an entire supply chain. One specific area of potential impact is the Industrial Gas supply chain, where the typical supply chain includes the energy utility, the industrial gas producer, the manufacturing plant using the industrial gases, and the downstream manufacturing or production operations. The operational cost of the entire supply chain is likely much higher than it needs to be as a result of uncoordinated supply and demand. Such a lack of coordination inherently limits the efficiency of the overall process, increasing energy use, emissions, and labor. This presentation discusses the impact of optimizing the overall supply chain by developing tools to facilitate such an analysis, identifying the data management infrastructure necessary to enable such collaboration, and developing methods to show how different companies, each with their own financial objectives, can best share the benefit of such collaboration. Such work is needed to transform our approach in energy intensive industries and lead to a transformative step change in manufacturing competitiveness.

Dominique Pelletier, Canada Research Chair on Analysis, Characterization and optimization of Complex Flow Systems, École Polytechnique de Montréal
Simulation-Based Engineering Analysis and Design: Issues and Challenges

The so-called high fidelity models are very popular in design optimization. While there is a long track record of this approach in structural optimization, design optimization of flow systems is a much younger discipline fraught with success stories and failures. We take a look at Computational Fluid Dynamics that have an impact on the performance of the optimization process. How does one control numerical errors so that they don't lead to sub-optimal designs or even worse, to a breakdown of the optimization process? How does one identify key parameters controlling the flow responses? How does one determine the uncertainty of the flow response induced by uncertainties in the input data? (How to cascade uncertainty though a CFD code?) Usually users of design optimization are not CFD experts and most often rely on commercial or third-party code. How can they make sure that these CFD tools work correctly? Finally, the question of how appropriate the flow model is for the flow physics at hand must be settled. We will illustrate these various points with examples taken from our research.

============

Jean-Francois Hétu, Industrial Materials Institute, National Research Council of Canada
Putting CFD to work in industry: issues and challenges for optimization

We will outline our strategy for developing industrial strength CFD tools for analysis and optimization of manufacturing processes, including discretization by finite elements, modeling complex physics, parallel processing and its implications for simulations software. We will highlight cases that lend themselves to optimization and others that have resisted for many years with explanations as to why. In some instances, advances made at Polytechnique have had a strong impact on our ability to provide support to the Canadian Manufacturing industry. We close with a discussion of our most challenging problems.

Biomass in Canada: A renewed opportunity for Operations Research

C. Tattersall Smith, Dean and Professor of the Faculty of Forestry, University of Toronto
Biomass in Canada and the requirement for development and management of new supply chains and technology.

Mitigation of climate change, scarcity and high prices of fossil fuels and securing of the energy supply have brought forests into focus in global energy strategies. If produced sustainably, biofuels have the potential to reduce greenhouse gas (GHG) emissions in the transport sector, diversify Canada's energy supplies, contribute to domestic energy security, reduce the risk of disease and wildfire, and provide opportunities for investment, economic growth and job creation in the economy of Ontario. This talk will give an overview of the biomass industry in Canada, its potential and the many areas that OR and optimization can have an impact on, such as the new "Biomass Feedstocks for Energy Markets" program that is currently being launched by the International Energy Agency.


Eric J. McCarthy, Director of Fuels at Energy Markets, Ontario Power Generation
Biomass at OPG

Ontario Power Generation is actively studying the use of renewable biomass as a replacement fuel for coal in some of its electricity generating units. We are assessing the impacts of biomass on equipment, required modifications and the safe handling and storage of biomass fuel. In addition to this, sustainable fuel supplies must be developed. This talk will focus on the challenges and logistical issues that will be need to be resolved for a successful implementation of a biomass program, and the experiences that we have gained sofar.

B. Wayne Bequette, Isermann Department of Chemical and Biological Engineering Rensselaer Polytechnic Institute, Troy, NY
Systems and Control Applications in Diabetes

An individual with Type 1 diabetes no longer has the ability to produce insulin and must receive multiple daily insulin injections, or continuous insulin infusion using a pump. Diabetics must closely monitor their blood glucose levels by pricking their fingers several times each day to obtain blood glucose values from glucose test meters. The availability of a continuous sensor for subcutaneous glucose concentration has the potential to radically improve blood glucose regulation, and is also necessary for the development of a closed-loop artificial pancreas. In this talk we will discuss the application of optimal estimation theory to continuous glucose monitoring, including an alarm warning of possible future hypoglycemic (low blood glucose) threshold violations, which could lead to dangerous conditions. In addition, we discuss progress towards a closed-loop artificial pancreas using an optimization-based model redictive control (MPC) approach.

============

Jeffrey D. Kelly, Honeywell Process Solutions, Toronto
On the Modeling and Solving of Large-Scale Manufacturing Optimization Problems in Diverse Industries

Optimizing both the process and the production in complex manufacturing facilities in off, on and in-line environments is the focus of this presentation. With the astounding advancements in both computer hardware and computational software, going beyond simulation into optimization with respect to automated decision-making has tremendous industrial significance and has been the focus of decades of academic research and practical application. One of the major challenges in applying these technologies is "how do we describe the hierarchical, spatial and temporal complexities of the problem with enough accuracy to significantly improve the manufacturing?". Our approach is to successively develop the "allegoric", "arrayic" and "algorithmic" details of the problem into a new form of algebraic-like modeling-system which can be easily interfaced to any commercial and public-domain mathematical and meta-heuristical solving-system such as mixed-integer linear programming, successive linear/quadratic programming and evolutionary-type algorithms. The deployment of the solution in terms of whether it is executed in-line (scheduling), on-line (control) or off-line (analysis) following what we call the continuous-improvement plan-perform-perfect loop is also discussed.