The Program

Arrival in Buenos Aires and Check-In

Registration

Student Meet & Greet Reception

Breakfast (On Your Own)

Registration (in the back of the room)

Lecture 1: Thinking about Data: Representations, Transformations, and Sustainability Applications

Victor M. Zavala Tejeda (University of Wisconsin-Madison)

Coffee Break

Tutorial 1: Modeling with Julia and JuMP

Joshua L. Pulsipher (Carnegie Mellon University)

Lunch with Roundtables on Diversity, Equity, and Inclusion

Break

Seminar 1: Storage of Energy from Renewable Intermittent Sources

Ana I. Torres Rippa (Carnegie Mellon University)

Seminar 2: Optimization Models for Sustainable Development of Energy Supply Chains

Diego C. Cafaro (National University of Litoral)

Coffee Break

Seminar 3: Supply Chain Design and Planning for the Forest Industry: An Opportunity for the Circular Bioeconomy Development in Argentina

M. Analia Rodriguez (IPQA, CONICET-UNC)

Dinner (On Your Own)

Breakfast (On Your Own)

Lecture 2: Nonlinear Programming: Concepts and Algorithms for Process Optimization

 Lorenz T. Biegler (Carnegie Mellon University)

Coffee Break

Lecture 3: Mixed-integer and Disjunctive Programming Techniques for the Optimization of Process and Energy Systems

Ignacio E. Grossmann (Carnegie Mellon University)

Lunch

Break and Optional Social Activities

Seminar 4: Dynamic Optimization of Bioprocesses: Use of Metabolic Models through Dynamic Flux Balance Analysis

Galo Antonio Carrillo Le Roux (University of Sao Paulo)

Seminar 5: Dynamic Real-Time Optimization: Advances and Challenges

Argimiro R. Secchi (Federal University of Rio de Janeiro)

 Coffee Break

Seminar 6: Ecohydrological Modeling and Dynamic Optimization for Water Management in Integrated Aquatic and Agricultural Livestock Systems

M. Soledad Diaz (National Southern University, PLAPIQUI)

Dinner (On Your Own)

Breakfast (On Your Own)

Tutorial 2: Modeling with Python and Pyomo

Carl D. Laird (Carnegie Mellon University)

Coffee Break

Lecture 4: Integration of Data-Driven and Mechanistic Models for Optimization: Applications in Carbon Capture and Resilient Power Grid Operation

 Fani Boukouvala (Georgia Institute of Technology)

Lunch

Break

Seminar 7: Towards Advanced Optimization Models for Power-Intensive Processes under Time-Sensitive Electricity Prices

Carlos A. Mendez  (National University of Litoral)

Seminar 8: Optimization Models for Sustainable Agrifood Supply Chains

C. Aldo Vecchietti (National Technical University)

Coffee Break

Seminar 9: The Role of Biorefineries in the Transition to a Low Carbon Economy: How PSE Tools can Contribute

Felipe Fernando Furlan (Federal University of Sao Carlos)

Seminar 10: Design, Modeling, and Optimization of an Induction Heated Steam-Methane Reformer

Cornelius Mduduzi Masuku (Purdue University)

Banquet Dinner with Tango Performance

Breakfast (On Your Own)

Tutorial 3: Modeling and Dynamic Simulation with EMSO

 Argimiro R. Secchi and Simone C. Miyoshi (Federal University of Rio de Janeiro)

Coffee Break

Seminar 11: Enhancing the Efficiency, Sustainability, and Reproducibility of Batch Column Distillation in Spirit Production

J. Ricardo Perez Correa (Pontifical Catholic University of Chile)

Special Topics: Overviews of PSE across Participating Countries

Selected Speakers

Closing Reception and Lunch Presentation: Challenges and Opportunities for PSE: Connections with the 11th World Congress of Chemical Engineering

Rafiqul Gani (PSE for SPEED Company Limited)

Modeling with Julia and JuMP

We will address how to use JuMP.jl to model/solve a variety of optimization problems that arise in PSE. JuMP.jl is Julia-based open source software package mathematical programming that is compatible with 50+ solvers and supports a wide envelope of optimization problem classes such as linear, conic, nonlinear, mixed-integer, positive semi-definite, constraint, and multi-objective programming. This hands-on course will feature practical overviews/tutorials on programming in Julia and using JuMP.jl to tackle a variety of optimization problems.

Modeling with Python and Pyomo

This course will provide a tutorial on how to model optimization problems in Python using Pyomo. Pyomo is a Python package that provides a rich suite of methods for tackling optimization formulations of interest to PSE. 

Modeling and Dynamic Simulation with EMSO

This tutorial provides an overview of the EMSO simulator and its interface with OpenLCA. The EMSO simulator is an equation-oriented process simulator with a graphical interface that can be used for modeling complex dynamic or steady-state processes. New models can be written in the EMSO object-oriented modeling language. With the new plugin EMSO_OLCA, life cycle assessment and process simulation or process optimization can be solved simultaneously. This hands-on course will feature practical tutorials on building simple examples of this integration.

Thinking about Data: Representations, Transformations, and Sustainability Applications

A dataset can be represented in different mathematical forms; for example, a micrograph can be represented as an image, as a matrix, as a graph (network), or as an intensity function. These representations are used to perform transformations of the data with the goal of extracting different types of features such as spatial patterns, geometrical patterns, correlations, principal components, gradients of light, and frequencies. These features contain key information that facilitate visualization and analysis, detection of abnormalities, and construction of predictive models.  In this talk, we show how to use representations and transformations in innovative ways to analyze complex datasets arising in flow cytometry, liquid crystals, chemical processes, and molecular dynamics. We show how these tools can be used to design chemical sensors for the detection of dangerous contaminants in air and liquid mixtures, for the selection of environmentally-friendly solvents, and to characterize plastic waste.

Nonlinear Programming: Concepts and Algorithms for Process Optimization

A survey is presented of advances in nonlinear programming (NLP) algorithms for challenging, large-scale applications of process optimization. These applications include topics in process modeling, design and control, for both steady-state and dynamic systems. The lecture outlines concepts that are embedded within state-of the-art NLP methods and provides a descriptive comparison of widely used NLP solvers, especially those that are incorporated into advanced modeling languages like those discussed in the hands-on tutorials. In addition, applications in process design, operations and control will be developed for a number of challenging case studies in the area of sustainability. Finally, a number of hands-on examples will illustrate these concepts.

Mixed-integer and Disjunctive Programming Techniques for the Optimization of Process and Energy Systems

Mixed-integer programming provides a powerful framework for modeling many optimization problems involving discrete/continuous decisions, including design, scheduling and control of process and energy systems. In this workshop we emphasize the importance of modeling in these problems. We first introduce the modeling of typical constraints arising in mixed-integer linear programing problems, stressing the importance of linear 0-1 inequalities. Next, we introduce the use of propositional logic as a higher level modeling framework from which linear 0-1inequalities can be systematically derived. We provide several examples to demonstrate the usefulness of this approach for the systematic modeling of discrete and continuous optimization problems. We also introduce disjunctions on continuous constraints, and its two major reformulations as mixed-integer linear constraints big-M and hull reformulations and analyze their corresponding LP relaxations. Next, we give an overview of the major methods for solving mixed-integer linear programs, branch and bound and cutting planes. This is followed by methods for solving mixed-integer nonlinear programming models, namely branch and bound, outer-approximation (OA), Generalized Benders Decomposition (GDB), and Extended Cutting Plane (ECP). We introduce the concept of Generalized Disjunctive Programming, as a high level representation of optimization problems in terms of continuous and Boolean variables with algebraic, logic and disjunctive constraints. Aside from showing that the GDP can be used a modeling framework to derive algebraic mixed-integer linear and nonlinear problems, we also briefly describe the concept of basic steps for tightening the relaxation of these problems, as well as the disjunctive branch and bound and logic-based outer approximation algorithms. Finally, we briefly discusses global optimization of nonconvex NLP and MINLP models, stressing the use of convex envelopes within spatial branch and bound methods.

Integration of Data-Driven and Mechanistic Models for Optimization: Applications in Carbon Capture and Resilient Power Grid Operation

Over the past years, we have experienced continuous improvement of high-performance computing capabilities for simulation, development of powerful high-level programming languages and Machine Learning (ML) methods, and increase in data availability. All the above are creating new opportunities for in tandem use of data-driven techniques with mechanistic equations (i.e., hybrid modeling) for decision-making. In this workshop, we will work through different scenarios of hybrid modeling paradigms, motivated by critical energy and sustainability challenges. The first motivating case study will be based on simultaneous material screening and process optimization for efficient and economical carbon capture processes. This will be used to teach students how to: (a) design computer experiments to sample an expensive computer simulation, (b) fit various surrogate or ML in Python to represent correlations between mixed-integer inputs and outputs, and (c) adaptively optimize-sample-refine surrogate approximations and re-optimize to get to optimal solutions within the existing optimization Python package, Pyomo. The second case study will use a chemical reactor design problem, to train students on how to embed mechanistic knowledge within ML models, following a Physics-Informed Neural Network approach. Through this case study, students will be shown the advantages and challenges of training these constrained ML models, with respect to training costs and accuracy for interpolation and extrapolation. Finally, the same case study will be used to present an alternative hybridization strategy, in which a fitted ML model is embedded within an equation-based optimization problem. This strategy is preferable when the different components can be decoupled, and often this approach can enable solutions for problems that can become intractable if a fully mechanistic approach is used. This final exercise will be further motivated by the discussion of a hybrid model for solving optimal power flow problems under contingency events.

Storage of Energy from Renewable Intermittent Sources

Decarbonization of the economy has largely increased the exploitation of renewable energy sources (RES) such as wind and sunlight, that are intermittent in nature. Integration of these sources into already existing processes poses challenges to their operation. Energy storage systems can be used to make better use of RES power generation capacities and also stabilize operations.  This lecture will describe  the formulation of optimization problems to select the best energy storage solution(s) in different settings. We will  discuss the  formulation of a generic P2P superstructure, its division in different stages and the equations that link the stages and will act as the constraints of the proposed optimization problem. We will comment on key equations that should be taken into account for modeling each stage, focusing on models that are not usually covered in undergraduate programs. The lecture will be run in a tutorial mode with hands-on case studies published by our research group.

Optimization Models for Sustainable Development of Energy Supply Chains

In the coming years the world expects a substantial growth in demand for energy products, while the need to reduce GHG emissions is unavoidable. Since the operation of supply chains accounts for more than 80% of the environmental impact of goods and services, tracking the carbon intensity (CI) of energy products appears as a key measure to quantify carbon footprint along their value chains. This indicator will also constitute a key factor of survival, competitiveness and validity in the market for many companies in the near future. In this context, the optimal design, planning and operation of energy supply chains require rigorous techniques to make energy products accessible and affordable to all, under CI limitations. Unlike the tracking of other intensive properties, the CI associated to product flows depends on numerous decisions such as the network topology, the selection of means of transportation, their speed, the production processes, and the energy used at each facility, among others. The need to manage alternative energy sources globally, together with final markets imposing different CI limitations, makes detailed monitoring of CI a critical task. This presentation introduces a series of mixed-integer optimization models to help energy companies boost their transition to a more sustainable matrix. We address the optimal design and expansion planning of shale oil and gas gathering networks through pipelines; the supply of water, its processing, recycling and reutilization; the integrated design and operation of bioprocesses to produce renewable natural gas from residues; the optimal placement and interconnection of turbines on a windfarm; and the planning of green and blue hydrogen supplies across the globe. For the latter case in particular we assess the impacts of the CI on the optimal solutions and how companies could effectively manage their CI targets to remain competitive. 

Supply Chain Design and Planning for the Forest Industry: An Opportunity for the Circular Bioeconomy Development in Argentina

It is now recognized that the next progress step of the world economy has to do with the circular bioeconomy (CBE) development. Local initiatives are more and more observed as part of a change in the cultural interests and consciousness of the societies. However, in order to expand the impact and application of the CBE, cost-efficient supply chains should be designed aligned with this concept. Forest industry has a big potential in the development of CBE. On the one hand, several processes can be interconnected in order to diminish the amount of residues, recycle products and extend the life cycle of the whole production. On the other, a large amount of biomass which is poorly used could be transformed into high value products, replacing fossil-based energy, materials and chemicals. In this talk, we will present some of the problems arising in this area. First, a generic mathematical model, using a logic-based approach, is developed to address the optimal forest bio-refinery supply chain design and planning problem, that simultaneously considers a dynamic capacity allocation approach and conversion facilities co-location. Second, the model is extended to introduce more details in the processes and energy systems involved as well as the consideration of uncertainty in some critical exogen parameters. Finally, the long-term forest harvesting planning problem is considered integrating a simulation-based approach to estimate harvesting yields for alternative forest treatments and a GDP optimization model where several logical constraints are considered. Alternative formulations will be presented and discussed for the Argentinian forest sector case study. 

Dynamic Optimization of Bioprocesses: Use of Metabolic Models through Dynamic Flux Balance Analysis

One of the motivations of Systems Biology is to predict the phenotype (behavior) of a living microorganism based solely on its genetic information. The aim of this is to go from the very basic genomic information, which can nowadays be obtained in a very short time (days) and cost (thousands of dollars), to the prediction of the behavior of the microorganism, which would allow the improvement of the strain for a given objective. A technique that is very popular for this is the Flux Balance Analysis (FBA). In FBA, a Metabolic Model, which is a collection of reaction stoichiometries and constraints is obtained, together with uptake models, based on Michaelis-Menten equations is obtained. This model is composed of a stoichiometric matrix with thousands of reactions, and of some constraints, and in order to obtain a simulation, an optimization problem is solved. If the model is for a dynamic system, the model is classified as a Dynamic Flux Balance Analysis (DFBA) where in order to obtain the derivatives of the system states an optimization problem needs to be solved. If, in addition, this model is to be used for dynamic optimization or for parameter estimation a bi-level optimization takes place. In this talk, we present some strategies that can be used to solve bi-level problems that include surrogate models and complementary constraints. Those strategies are compared and illustrated with examples.

Dynamic Real-Time Optimization: Advances and Challenges

Dynamic real-time optimization (DRTO) is a model-based technique to optimally drive process operation towards its optimal condition according to an objective function while respecting constraints. This approach is especially suitable for processes subject to frequent changes in the process operating conditions and production planning. Nevertheless, nonlinear dynamic optimization problems are computationally demanding making real-time applications a challenge. Besides, the lack of information for validating dynamic phenomenological models makes this challenge even more difficult. On the other hand, steady-state real-time optimization (SRTO) approaches depend on reliable stationary operating data to overcome uncertainties issues. This dependency may add long wait times to acquire reconciled data to update the model, losing performance. Several approaches were developed over the years to use transient data inside static optimization frameworks trying to remove these drawbacks of the SRTO. In this talk, the advances in DRTO, SRTO, and hybrid formulations are discussed, together with their alternative interactions with the supervisory process control layer. The benefits of combined data- and model-based approaches are also analyzed. The main challenges to enabling the use of DRTO in large-scale continuous processes are brought to light.

Ecohydrological Modeling and Dynamic Optimization for Water Management in Integrated Aquatic and Agricultural Livestock Systems

Water is an extremely scarse resource and much work has been devoted to modelling of water bodies with different objectives and approaches. The identification of meteorological, hydrological and ecological components and driving forces behind problems such as eutrophication, floods and salinisation in water bodies, as well as the integration with agricultural and livestock activities constitutes a challenging problem towards achieving sustainability goals. 

In this work, we formulate optimal control problems to plan management strategies in eutrophic lakes and reservoirs, based on ecohydrological and biogeochemical models represented by differential algebraic equation systems that include mass balances, evaporation, kinetic equations, etc. Dynamic optimization problems are formulated within a control vector parameterization framework in gPROMS (PSEnterprise, 2022).

On the one hand, we address the control of algal blooms in a reservoir that provides drinking water to two cities, by considering simultaneous application of different restoration strategies, including reduction of external nutrient charge to the water body through artificial wetlands (control variable: fraction of nutrient rich water stream that is diverted through wetlands), fish removal (control variable: fish removal rate) and artificial floating islands within the water bodies. 

On the other hand, we propose integrated models that include ecohydrological, agricultural, livestock and carbon capture submodels, to address the hydrological instability of a salt lake endorheic basin within a semiarid region, focusing on preserving valuable fish species, native and valuable crops and cattle. The objective is to propose water management policies to address mitigation of extreme events like droughts and floods through the formulation of an optimal control problem that allows a) Optimizing management of water resources in salt lakes and in constructed freshwater reservoirs; b) Keeping salinity in the lake within a desired value for valuable fish species growth, as well as salt lake volume, during dry and wet periods; c) Ensuring provision of drinking water, food and shade to cattle; d) Proposing restoration strategies for native tree species; e) Proposing combination of new plantations with drought resistant crops and pasture; f) During wet periods, preventing flooding of nearby towns by diverting part of lake tributary flows into constructed reservoirs; g) Addressing the economic valuation of ecosystem services in the basin. 

Numerical results provide useful insights on optimal management strategies and the effects of their implementation, as well as their impact on the valuated services in the watershed under study; which shows that the formulation of integrated models for ecosystems constitute valuable tools to help the sustainable socieconomic development focusing on water management in water bodies and their basins.

Towards Advanced Optimization Models for Power-Intensive Processes under Time-Sensitive Electricity Prices

The competitiveness of power-intensive industries is highly tied to their ability to adjust production according to time-sensitive electricity prices. A classic example are Air Separation Units (ASUs), where large, electric-power air compressors are used to reach cryogenic temperatures. Due to the volatility nature of the energy markets, there is significant opportunity to reduce production costs by scheduling production during the cheapest hours of the day. Mixed-integer linear model (MILP) based on a discrete-time scheduling formulation can be used to represent and optimize operating decisions for any process under time-sensitive energy prices. The main goal is to find an optimal production schedule over a given time horizon that guarantees product demand satisfaction and that minimizes total energy cost. The formulation can be used to model transitions between operating modes results in a very efficient and robust model. The model is applied to a simplified industrial case. The results show optimal solutions for the proposed methodology with modest computational effort.

Optimization Models for Sustainable Agrifood Supply Chains

Agriculture 4.0 is a term to define the future of a sustainable agriculture production by integrating the ultimate digital technologies into food production, from harvest to production and distribution. Nowadays, agricultural machinery has entered the digital age, by enhancing its current features, incorporating incorporates electronic controls and autonomous devices. In addition, electronics, using sensors and drones, support the collection of data on several key aspects of agriculture (climate, soil, type of seeds). However, even with these advances, the use of adequate methods to improve the performance of agricultural supply chains remains a challenge. Therefore, the question remains about how to support a better decision-making process, or how to support agribusiness stakeholders for effective decision-making based on objective data. This proposal aims to present mathematical models for agro-industrial supply chains (CSA) focused on their operations, with the primary objectives of minimizing costs, increasing yields, reducing environmental impact and optimizing infrastructure and resources to reach a sustainable agricultural production.

The Role of Biorefineries in the Transition to a Low Carbon Economy: How PSE Tools can Contribute

It is a scientific consensus that the climate crisis derives from anthropogenic causes, with fossil fuel use as the main one. This perspective implies that in the near future there will be an abrupt shift toward a low carbon production matrix. In this sense, biorefineries appear as the main alternatives, producing biofuels, high-added value products, and energy (heat and electricity) from biomass. Nevertheless, there are still technological and economic gaps to be overcome for biorefineries to be generally feasible. Great research efforts and investments are being made in this direction. In this sense, Process Systems Engineering tools can contribute. One of these tools, Retro-Techno-Economic-Environmental Analysis, can be used to construct windows of feasible operation, based on economic and environmental metrics, which relate to the main process variables of a biorefinery, for example. Using this information, research teams can focus on the most promising process conditions, obtaining experimental data under industrially relevant conditions.

Design, Modeling, and Optimization of an Induction Heated Steam-Methane Reformer

Steam-methane reforming and other widely applied endothermic reactors can benefit from innovative electrically supplied heat generated from renewable resources instead of burning fossil fuels as is current practice. We propose the use of inductive heating to provide energy to the reaction system to abate the need to combust fossil fuels. This work focuses on the design of an electrically driven reactor, and models the reaction to generate the temperature and concentration profiles. The implementation of a heat induction coil surrounding the tubular reactor is also analyzed. ANSYS packages are used to design and effectively model this process. ANSYS Chemkin Pro is used to implement the reaction kinetics, while geometries are constructed in Maxwell 3D along with the assignment of desired electric properties, and this data is then transferred to ANSYS Transient Thermal Analysis ANSYS Fluent to observe transient thermal results as well as conductive and convective heating properties in the fluid flow. The results show that the reactor can be constructed to achieve the desired output fluid temperatures and controlled with precision through variable electric inputs.

Enhancing the Efficiency, Sustainability, and Reproducibility of Batch Column Distillation in Spirit Production

Fruit wine distillation is a growing economic activity due to the increasing demand for high-quality and distinctive spirits. These distillates are characterized by a particular flavor and aroma provided by several minor components from the fruit of origin. Distillate quality relies on the operation recipe and artisan expertise to compensate for disturbances during the distillation and fruit variability. Traditionally, distillers focused on ethanol productivity, preserving valuable minor components, and minimizing toxic or undesirable components, disregarding cooling water and energy consumption. This work aims to design sustainable and reproducible operating systems for packed-column wine stills, considering productivity and distillate quality and the consumption of cooling water and energy. We formulated a dynamic multi-objective optimization problem constrained by a nonlinear first principles model calibrated with experimental data to design an operating recipe that equilibrates ethanol productivity and cooling water expenditure, considering a fixed distillation time and energy consumption. Additionally, to ensure a high-quality product, constraints on the distillate's average and minimum ethanol content were included in the formulation of the optimization problem. Finally, a robust model-based control algorithm was implemented and tuned to minimize the impact of unmeasured disturbances and model uncertainties, achieving a reproducible operation from batch to batch. The methodology developed in this work can be easily extended to any packed-column batch distillation system if a reliable dynamic process model is available.