The main goal of this online seminar series is to provide a platform for researchers to share their work with the process control community in an attempt to foster collaboration and contact between people working on process control, optimization, and data analytics theory and applications. Although geared to early-career researchers, this seminar is open to researchers at all stages that are interested in sharing technical results that may be of broad interest to the community. The monthly webinars will be fully open, meaning anyone is welcome to join and ask questions or provide comments to the speakers, and will last for around 1 hour (with an approximately 35 minute presentation followed by a 20 minute Q&A). Please contact Joel Paulson ([email protected]) if you have any questions or would like to nominate a speaker. Spring 2023 Schedule: Optimization Over Trained Neural Networks in Scheduling and Control Applications - Calvin Tsay Date: February 28, 2023 Abstract: In combination with the fast-improving performance of optimization software, optimization over trained neural networks (NNs) appears tractable (at least for moderate-size NNs). In this talk, we discuss recent advancements in optimization formulations and software for NN surrogate models, including OMLT, the Optimization and Machine Learning Toolkit. We will then outline applications to scheduling and control problems. For supply chain scheduling, we show how optimization over trained NN state-action value functions (i.e., a critic function) can explicitly incorporate constraints, and we describe two corresponding reinforcement algorithms. For process control, we show how optimization can be used to evaluate ‘correctness’ of NN-based controllers before deployment. Bio: Dr. Calvin Tsay is Assistant Professor (UK Lecturer) in the Computational Optimisation Group at the Department of Computing, Imperial College London. His research focuses on computational methods for optimisation and control, with applications in machine learning and process systems engineering. Calvin received his PhD degree in Chemical Engineering from the University of Texas at Austin, receiving the 2022 W. David Smith, Jr. Graduate Publication Award from the CAST Division of the American Institute of Chemical Engineers (AIChE). He previously received his BS/BA from Rice University (Houston, TX). Spring 2022 Schedule: Process Modeling and Control of Digital Biopharmaceutical Manufacturing - Moon Sun Hong Date: January 26, 2022 Abstract: Recent trends in biopharmaceutical manufacturing provide opportunities for process systems engineering to make major advances in biomanufacturing. These trends ultimately lead biomanufacturing to digital manufacturing, which is an integrated approach to manufacturing centered around a computer system. Digital manufacturing is enhanced by using modern system engineering tools. This presentation describes how modern system engineering tools are applied for digitalization of biomanufacturing systems through process modeling and control. The presentation describes mathematical models and optimal control methods for multiple bioreactor configurations including microbioreactor systems and stirred-tank bioreactors. The presentation describes laboratory unit operation systems constructed for downstream process including protein crystallization and continuous viral inactivation. These systems are optimally designed and controlled based on first-principles models. Speaker Bio: Moo Sun Hong is a postdoctoral associate at the Massachusetts Institute of Technology (MIT) in the group of Prof. Richard D. Braatz. Moo Sun received a B.S. in Chemical and Biological Engineering from Seoul National University and an M.S. in Chemical Engineering Practice and Ph.D. in Chemical Engineering from MIT. His research focuses on the mechanistic modelling and model-based design and control of biopharmaceutical processes. Awards for his research include the AIChE FP&BE Division Poster Presentation Award in 2019 and the AIChE Separations Division Graduate Student Research Award and AIChE PD2M Award for Excellence in Integrated QbD Practice in 2021. Reinforcement Learning with Guarantees - Mario Zanon Date: February 23, 2022 Abstract: RL struggles to provide strong guarantees on the behavior of the resulting control scheme. In contrast, MPC is a standard tool for the closed-loop optimal control of complex systems with constraints and limitations, and benefits from a rich theory to assess closed-loop behavior. Because of model inaccuracy, however, MPC can fail at delivering satisfactory closed-loop performance. Mario and Sébastien will show how to leverage the advantages of the two techniques, offering a path towards safe and explainable RL. Speaker Bio: Mario Zanon received the Master's degree in Mechatronics from the University of Trento, and the Diplôme d'Ingénieur from the Ecole Centrale Paris, in 2010. After research stays at the KU Leuven, University of Bayreuth, Chalmers University, and the University of Freiburg he received the Ph.D. degree in Electrical Engineering from the KU Leuven in November 2015. He held a Post-Doc researcher position at Chalmers University until the end of 2017, after which he became Assistant Professor and from 2021 Associate Professor at the IMT School for Advanced Studies Lucca. His research interests include reinforcement learning, numerical methods for optimization, economic MPC, optimal control and estimation of nonlinear dynamical systems in particular for aerospace and automotive applications. Distributionally Robust Model Predictive Control for Linear and Nonlinear Systems - Antonio Del Rio Chanona Date: March 30, 2022 Abstract: Classical stochastic model predictive control (SMPC) methods assume that the true probability distribution of uncertainties in controlled systems is provided in advance. However, in real-world systems, only partial distribution information can be acquired for SMPC. The discrepancy between the true distribution and the distribution assumed can result in sub-optimality or even infeasibility of the system. To address this, we present a new distributionally robust data-driven MPC scheme to control stochastic linear and nonlinear systems. We first propose a data-driven MPC scheme to control constrained stochastic linear systems using distributionally robust optimization. The resulting distributionally robust MPC framework is computationally tractable, efficient, and recursively feasible. Additionally, the information about the uncertainty can be determined empirically from the data. Subsequently we present a few alternatives to adapt this strategy to nonlinear systems in a tractable way. We show examples both on linear and nonlinear systems while examining the benefits and shortcomings of the proposed approaches. Speaker Bio: Antonio del Rio Chanona is an Assistant Professor at the Sargent Centre for Process Systems Engineering at Imperial College London. His main area of research is on the automation of chemical processes via optimization, machine learning and control. Antonio received his BSc from UNAM in Mexico, and his PhD from the University of Cambridge where he was awarded the Danckwerts-Pergamon Prize for the best doctoral thesis of his year. He received the EPSRC fellowship from the Engineering and Physics Research Council from the UK to adopt automation and intelligent technologies into bioprocess scaleup and industrialization. He has received several awards from different institutions such as the International Federation of Automatic Control (IFAC), The NASA, and the Institution of Chemical Engineers (IChemE). Antonio is very active in pursuing teaching and community building activities, particularly for young people from disadvantaged backgrounds, please feel free to get in touch for such activities. Computationally Efficient Distributed Decision-making for Interconnected Systems With Convergence Guarantees - Dinesh Krishnamoorthy Date: April, 27, 2022 Abstract: This talk considers the problem of distributed decision-making using decomposition-coordination framework, where the overall optimization problem is decomposed into smaller subproblems that are coordinated by a master problem. Repeatedly solving the subproblems from one iteration to the next can quickly add to the overall computation time. Although there have been several developments in the distributed optimization literature, these have predominantly been focused on the master problem formulation, such that the number of iterations required can be reduced. As such the computation burden of solving the subproblems itself remains. Noting that between each iteration, the subproblems solved are similar, the first part of the talk will show how the computational cost of the procedure can be significantly reduced by exploiting the parametric sensitivity of the subproblems. The second part of the talk will deal with this issue by transforming the distributed optimization problem into a distributed feedback control problem. Convergence guarantees for both the approaches will also be discussed. Speaker Bio: Dinesh Krishnamoorthy is currently a post-doctoral researcher at the Harvard John A. Paulson School of Engineering and Applied Sciences. Dinesh received his PhD in Chemical Engineering from the Norwegian University of Science and Technology (NTNU), MSc in Control Systems from Imperial College London, and B.Eng in Mechatronics from the University of Nottingham. He has more than four years of industrial research experience from Statoil Research Center and Novo Nordisk R&D. He is also the recipient of the Dimitris. N. Chorafas Foundation Award, EFCE Excellence in Computer-Aided Process Engineering (CAPE) PhD Award, IFAC-ABB Young author award, and NTNU Faculty of Natural Sciences Best PhD Thesis Award. His research interests include distributed optimization, numerical optimal control and model predictive control, extremum seeking control, real-time optimization, and Bayesian optimization, in particular for energy and biomedical applications. Control-Theoretic Cyberattack Detection for Optimization-Based Control of Nonlinear Systems - Helen Durand Date: May 25, 2022 Abstract: While most undergraduate process control courses focus on the dynamics and control of chemical processes that can be described by linear transfer function models, advances in computing over the last several decades have enabled the more complex nature (i.e., nonlinearities, interactions between variables, constraints) of the underlying process physico-chemical phenomena to be taken into account in the models used for controller design. A major trend in industry over the last 40 years has been employing constrained mathematical optimization techniques to compute control actions that optimize a quadratic objective function with its minimum at a process steady-state, subject to linear or nonlinear process models and practical constraints such as bounds on flow rates due to valve limitations. These optimization-based control designs (referred to as model predictive control or MPC) are typically implemented within a two-layer architecture. The upper layer (referred to as real-time optimization or RTO) solves an optimization problem that determines the economically-optimal operating steady-state for the process given recent process data, and then communicates its solution to an MPC system that computes values of the control actions that drive the process state to the economically-optimal steady-state. Recent trends in the process industries have motivated tighter integration of the economic optimization and feedback control layers to improve process economics with off steady-state operation, giving optimization-based control greater flexibility for realizing the efficiency, sustainability, and profitability goals of next-generation manufacturing. However, as the capabilities of these systems increase and their autonomy is relied upon more strongly to achieve day-to-day operating objectives, it becomes more critical to understand how malicious actors could undermine these systems and affect profits or even safety. These trends have motivated our research in the development and analysis of detection techniques for cyberattacks on sensors, actuators, or both sensors and actuators at the same time, where the control and detection policies are designed in tandem to take advantage of the guarantees of control theory for revealing when a process is not acting according to expectations. We will discuss stealthy attacks and cases where the control and detection policies can be simultaneously designed so that a lack of detection does not compromise safety. Throughout the talk, we will present applications of our methods to a chemical process to demonstrate their applicability and performance in meeting next-generation manufacturing goals related to improving process economics, safety, and security. Speaker Bio: Helen Durand is an Assistant Professor in the Department of Chemical Engineering and Materials Science at Wayne State University. She received her B.S. in Chemical Engineering from UCLA, and upon graduation joined the Materials & Processes Engineering Department as an engineer at Aerojet Rocketdyne for two and a half years. She earned her M.S. in Chemical Engineering from UCLA in 2014 and her Ph.D. in Chemical Engineering from UCLA in 2017, and subsequently started at Wayne State. She received the Air Force Office of Scientific Research Young Investigator award, and her work has also received support from the National Science Foundation including the NSF CAREER award.She received a Faculty Research Excellence Award and an Excellence in Teaching Award within the College of Engineering at Wayne State University and served as the chair of the Next-Gen Manufacturing Topical Conference for the 2021 Annual Meeting of the American Institute of Chemical Engineers. Her research interests are in the area of process systems engineering with a focus on process control.
