 |
Ali ElkamelProfessor, PhD, PEng |
Process Systems Engineering; Process Analysis and Improvement; Planning and Scheduling of Process Operations; Supply Chain Management; Product and Process Design; Pollution Prevention and Waste Minimization; Waste Utilization and Management; Planning for Energy Production (Oil, Gas, and Biomass Energy); Robust Optimization; Soft Computing; Combinatorial Optimization
Due to increased complexity, quality, safety, and environmental requirements in process manufacturing operations, many leading companies have identified Process Systems Engineering (PSE) as a strategic technology. Process Systems Engineering deals with the discovery, design, manufacturing, and distribution of products under many conflicting goals. The use of PSE enables companies to operate inherently safe processes while at the same time reduce production costs, improve quality, increase efficiency, reduce pollution, and bring products to market faster.
With this outlook in mind, the goal of my research program is to develop theory and applications for PSE. The applications will focus on planning and scheduling of process operations, planning for energy production, pollution prevention, waste minimization, and product and process design. Effective solution strategies for these problems (exact and heuristic) will be developed as well as effective models.
During the last few years, batch and semi-continuous chemical processing has received increasing attention as a mode of production for high value-added specialty chemicals. A typical facility processes a number of related products on a given set of equipment. The sharing of equipment necessitates the scheduling of production tasks.
Our goal in the area of process scheduling is to develop effective mathematical programming formulations and solution strategies that exploit the structure of the scheduling problem and processing facilities. In addition, we are concerned with scheduling of start-ups, shut downs, and maintenance of facilities.
In a real processing plant, there is always an uncertainty factor associated with various parameters such as market demand, processing time, and resource availability. We are taking two complementary approaches for handling uncertainties in scheduling process operations. First, the scheduling models must be formulated to be robust (stable) under variations. Stabilized formulations that are able to absorb disturbances are being considered and performance robustness criteria are being defined. The second approach we are taking for handling uncertainties is the development of a reactive scheduling strategy that is able to recommend new schedules while keeping the original scheduling decisions intact as much as possible.
Pollution prevention is one of the most serious challenges that are currently facing the industry. With increasingly stringent environmental regulations, there is a growing need for cost and energy efficient pollution prevention techniques. Our research in the area of pollution prevention has been focusing on the integration of the systems methodology to pollution problems.
In general, pollution prevention can be divided into two categories: short-term prevention and long-term prevention. Our aim in the area of short-term pollution prevention is to develop solutions at the operational level by incorporating our background and expertise in process scheduling. It is a well-known fact that the implementation of operational modifications often requires the least capital when compared to other prevention strategies. To this end, we are now looking at reactive scheduling to absorb disturbances that cause violations of environmental standards. We are also preparing scheduling models with the objective of determining schedules that minimize waste and abide to environmental constraints.
Long-term pollution prevention involves the determination of strategies that must be implemented to meet environmental standards over a long period. There are usually an infinite number of possibilities that must be screened economically. We have successfully prepared and solved mathematical programming models for the selection and planning of air pollution prevention options. Both the primal selection problem and the retrofit problem were considered. These models offer the appropriate scheme for evaluating and ranking available waste minimization and pollution prevention technologies. Later, we plan to look at means of measuring wastes in a process. These means will be integrated with the pollution selection models and equipment cost equations to arrive at comprehensive models that can be used as an engineering tool for simultaneously evaluating, planning, and designing pollution prevention strategies on a continuous basis.
Process modification is another long-term strategy for pollution prevention. These modifications involve the way a production process is operated and maintained. Process conditions as well as reaction parameters can play an important role in waste minimization. We are currently developing mathematical programming models to suggest process modifications for a number of industrial processes.
A longer-term strategy for pollution prevention is to develop sustainable processes during the planning and design stage. The quest for pollution prevention and increased pressure and demand for environmentally sustainable processes and products have been creating new rules in the process industry. Sustainability is defined as “economic development that meets the needs of the present generation without compromising the ability of future generations to meet their own needs”. We have recently prepared a mixed integer linear programming model and solution strategy for the development of a petrochemical industry with sustainability as the objective. The model can account for the different interactions among units and provides at the same time a suitable mathematical representation of the decision variables of interest. We plan to extend the above models so that safety is also incorporated in the planning stage as well as product selection and design.
In this group of projects we seek new solutions to one of the grand challenges of this century: supplying energy to a growing population while reducing greenhouse gas emissions. Since no single technology is likely to be able to meet this ultimate energy challenge of the future on its own, it is essential to use a systems approach that can provide insight and data on how viable a technology can be. Our long term vision is to propose optimal solutions to effectively manage carbon dioxide reduction, capture, and sequestration while meeting growing energy demands. These carbon management solutions shall include physical and natural processes associated with decarbonization; carbon dioxide capture, transport, and sequestration; the use of new and/or improved fuel sources (nuclear, fossil fuels, hydrogen energy, renewables, etc.); improved efficiency of energy conversion and utilization; economic and market analysis; and alternative energy policy options.
The novelty in this research is that it represents a multi-region, multi-technology decision framework that will provide provincial and national strategies for the effective reduction of carbon dioxide. It is based upon a bottom-up view of industrial activities and a top-down view of energy and other product demands. The framework will also account for the predictable trends and interactions that occur in a dynamic setting of a metropolitan region or a country as a whole. Included within this last category will be factors of regional growth, technological development, availability and limitations on resources, and interrelationships between different industrial sectors. One major contribution from this research will be the development of a decision support system that can aid management and policy makers in constructing equitable comparisons among different carbon dioxide abatement proposals. This will permit the selection of the least cost solutions from among a series of alternate carbon dioxide reduction schemes.
Integer and mixed integer programs (MIP) can be used to model a wide variety of problems encountered in many areas including plant layout and design, process scheduling, pollution prevention, and even molecular simulation and design. The general MIP is, however, known to belong to the class NP-complete or hard combinatorial optimization problems and no technically good algorithm is known to be available for its solution. Our aim here is to develop novel modeling techniques and effective solution methods for optimization problems.
The presence of various uncertainties (uncertainty in price, production rates and costs, labor, demand, raw material availability, etc.) complicates the optimization process. Traditionally, the treatment of uncertainty is realized by the use of a stochastic optimization approach. This approach recognizes the presence of multiple data instances that might be potentially realized in the future. The optimization will then attempt to generate a decision that maximizes (or minimizes) an expected performance measure, where the expectation is taken over the assumed probability distribution. In many cases, when multiple uncertain factors exist in the input data, assumptions of distributional independence among factors are made. After possible data instances (scenarios) or probability distributions are fed into a model, a stochastically optimal solution is generated.
There are multiple drawbacks of the stochastic approach in handling uncertainty. First of all, a decision has to be made on probabilities to the various data instances (future scenarios) or probability distributions for the different uncertain factors. Assigning such probabilities is far from a trivial exercise for many decision makers. Another more important drawback is that every decision has associated with it a whole distribution of outcomes, depending on what data scenario is actually realized. Decision makers are more interested in having information about the whole distribution of outcomes and the risk of poor system behavior. Clearly, a more robust approach is needed. We are looking at two different novel approaches for handling uncertainties: interval mathematics and fuzzy set theory.
Our research in the area of Artificial Neural Networks (ANNs) has focused on applications in new problem domains: fluid flow problems in reservoirs, air pollution forecasting, and modeling of complex refinery operations. In the latter case, the ANN models, in addition to their good prediction capability of product yields and properties, have many other uses. They can be used to optimize units, evaluate different feeds, train new operators, and can be integrated within general refinery mathematical programming software for planning and scheduling purposes. We are investigating other refinery operations with the goal of embedding the prepared ANN models within mathematical programming models for supply chain optimization of refinery operations. Existing commercial softwares for refinery production planning, such as RPMS (Refinery and Petrochemical Modeling System) and PIMS (Process Industry Modeling System) are based on models which are mainly composed of linear gain relations. Despite the use of mathematical programming for general production plans of the whole refinery, these models do not use accurate process models and are based on yield vectors, instead.
Another interesting application that we will pursue shortly is in product design and formulation. During my experience at Procter and Gamble (P&G), a key challenge was the identification of a formula that enables the product to best meet potentially conflicting functional, processing, and cost requirements. In addition, the available data for product formulation do not usually meet design of experiments criteria due to limitations of time and available facilities to run experiments. Also, both ingredients and process parameters need to be incorporated in the product formulation process.
In our research in planning for oil production, we have shown that the combination of ANNs and scaling is successful in solving Partial Differential Equations (PDEs) describing fluid flow problems in reservoirs. We plan to investigate the effectiveness of combining standard numerical methods (finite differences and finite elements) with neural networks in solving PDEs. We are also planning to use ANNs to generate models from observed data for a particular phenomena. This might be looked at as another mean of data mining. Since we were able to use numerical solutions of PDEs to provide data to train ANNs, and since these ANNs were able to predict the physical phenomena well, we can argue that they contain the same knowledge or have the same semantic content as the PDEs. Maybe then, the analysis of the connection weights of an ANN can lead to restoring the PDE describing the phenomena. This brings an important application to modeling complex processes that cannot be modeled directly from first principles. The use of ANNs to construct first principle models for these complex processes will open many doors for modeling and learning physical laws from observed data.
University of Waterloo
200 University Avenue West
Waterloo, Ontario, Canada N2L 3G1
519 888 4567
contact us | give us feedback | http://www.uwaterloo.ca