Keywords: Power and Energy Systems, Optimization under Uncertainties
Abstract: Electrochemical technologies offer efficient and dynamic storage of electrical energy in batteries or via electrolysis in hydrogen. In addition, power-to-chemicals technologies, such as electrochemical CO2 conversion or production of expensive fine chemicals, promise a deep energy transition of the chemical industry. Almost all new electrochemical technologies, but also established ones, suffer from performance losses due to a lack of quantitative insight into processes in the cells. Modelling tools at higher levels are already well established and predictive, when it comes to heat and mass transfer, as these phenomena are widespread in (chemical) engineering. They struggle, however, to reproduce or predict the electrochemical behaviour of electrodes. This talk covers established and upcoming electrochemical energy and power-to-chemical technologies, their application and state of development. The complexity and challenges of the technologies at electrode level are complemented by strategies how to formulate and parameterise suitable models. The models are then used to give insight into the many unmeasurable processes or states, and for virtual design of better catalysts, electrodes, cells or operating conditions.

Keywords: Derivative-Free Optimization, Optimization under Uncertainties, Estimation and Robust Estimation
Abstract: Convex relaxations are a crucial tool in methods for global optimization, but are challenging to construct for dynamic processes. In this article, we investigate combining two recent approaches in convex relaxation in new nontrivial ways, to aid global optimization of dynamic chemical process models. Specifically, we combine recent approaches for automatically generating convex relaxations for solutions parametric ordinary differential equations with a recent sampling-based approach for tractably generating lower bounds of a convex relaxation.

Keywords: Estimation and Robust Estimation, Identification Methods, Optimization under Uncertainties
Abstract: An approach for the real-time state estimation for particulate systems using an early-lumping moving horizon estimator is presented. The synthesis of particles is characterized by a complex system of partial differential equations and ordinary differential equations and optimization-based parameter identification techniques are employed to precisely determine process parameters. Subsequently, model order reduction through the dynamic mode decomposition with control is performed to obtain a linear approximation of the inherently nonlinear system while preserving critical dynamic features. The resulting system serves as the foundation for the moving horizon estimator design, which operates by solving a constrained optimization problem. This optimization problem is designed to minimize the discrepancy between measured and estimated outputs, thus providing accurate real-time state estimation. The applicability of the proposed approach is illustrated by investigating the synthesis of aluminum-doped zinc-oxide particles as a case study with the goal to reconstruct the particle size distribution based on the estimated concentration of the aluminum-doped zinc-oxide nanocrystals.

Keywords: Estimation and Robust Estimation, Optimization under Uncertainties, Identification Methods
Abstract: Model-based monitoring and control of chemical and biochemical processes rely on estimators like Extended Kalman Filters (EKFs) to ensure accurate predictions in real time. The selection of suitable model parameters and tuning factors is crucial for precise predictions. An extended Simultaneous Parameter Estimation and Tuning (SPET) method is proposed to handle complex systems with nonstationary disturbances, time-varying parameters, multi-rate data, and measurement delay. Through a case study on a Continuous Stirred Tank Reactor (CSTR), we demonstrate that SPET outperforms traditional approaches, achieving improved online predictions and state estimator performance. Keywords: State estimator tuning, Extended Kalman filter, Parameter estimation, Stochastic process disturbances, Multi-rate sampling, Time-varying parameters, Measurement delay

Keywords: Identification Methods, Optimization under Uncertainties, Experiment Design
Abstract: The present work deals with the control and observer design for a one–stage spray drying tower with focus on the optimization based trajectory planning and tracking control for an operation setpoint change during continuous operation. Putting emphasis on the numerical performance by employing different discretization schemes in combination with CasADi, key achievements are (i) an optimization-based system inversion for target trajectory generation, and (ii) a real-time capable model predictive tracking controller with computation times below one second in combination with either an extended Kalman Filter or a moving horizon estimator. The performance is shown in numerical simulations for a previously validated process model.

Keywords: Model Predictive Control, Learning-Based Control, Optimization under Uncertainties
Abstract: Reinforcement learning (Rl) has attracted considerable attention from both industry and academia for its success in solving complex problems. However, the performance of Rl algorithms often decreases in environments characterized by uncertainties, unmodeled dynamics, and nonlinearities. This paper presents a novel robust Rl algorithm designed to ensure closedloop stability for industrial processes. The algorithm considers a wide range of potential scenarios across various operating conditions and different ranges of parameter uncertainties. Using the multiple-model adaptive control methodology, the algorithm evaluates all scenarios and ranks them based on their likelihood of accurately characterizing the actual process. The validity of the results is demonstrated using a benchmark continuous stirred tank reactor (CSTR) problem.

Keywords: Optimization under Uncertainties, Optimal Control, Estimation and Robust Estimation
Abstract: Population balance model-based approaches have become ubiquitous in crystallization process design and control to drive crystallization systems to meet the required industry-specific critical quality attributes (CQAs). However, the reliability of model-based approaches is often subject to uncertain model parameters, which are usually determined through parameter estimation routines that process noisy experimental data. Disregarding these uncertainties during design often results in unexpected operational failures, suboptimal performance, or failure to attain desired CQAs. In this study, a reliability-based design optimization (RBDO) framework was applied for the open-loop control design of crystallization processes under parametric uncertainty. First, the concept of reliability-based design optimization was introduced to design crystallization systems under uncertainty to meet the target CQAs with a desired probability. A nested two-level simulation-optimization approach using surrogate modeling was used to solve RBDO problems. Finally, the above method was applied to demonstrate its effectiveness using a case study for batch crystallization process design. The results show that the RBDO-based approach provides reliable open-loop setpoint trajectories with higher probabilities of satisfying the desired CQAs when compared with open-loop optimization using nominal model parameters.

Keywords: Supervisory Control, Model Predictive Control
Abstract: This research paper introduces a novel approach to enhance closed-loop performance in manufacturing processes, catering specifically to industrial practice. The primary objective is to design a supervisory controller capable of shaping the response of an existing yet rigid, closed-loop system while adhering to crucial criteria like respecting process constraints or allowing for simple synthesis of the tuning options. The manuscript shows how to extend the universal tuning strategy with a lead-lag compensator, which decreases the rise time and removes possible oscillations. Two experiments are conducted: one employs a traditional approach, and the second verifies the lead-lag extension. Experimental results demonstrate the significant impact of the universal tuner on closed-loop performance, particularly in the overall effect on quality criteria of control performance.

Keywords: Fault Detection, Manufacturing Plant Control, Identification Methods
Abstract: The performance of a process control loop can be limited when nonlinear problems like deadband, hysteresis, backlash, stiction, etc. exist in control valve. Stiction occurs more frequently than the other valve problems and has potential to cause adverse oscillations in the control loop, resulting in poor quality products, excessive use of raw materials and energy, and an environmental footprint. Timely detection of sticky control valves can help control engineers to take appropriate actions (retuning the controller or using stiction compensation methods) to prevent further degradation of the performance of the control loop. In this regard, in the present work, a new stiction detection method is developed based on learning vector quantization neural network (LVQNN). Simulated database is generated and used to train the LVQNN with the training algorithm: LVQ2.1. To further improve the stiction detection capability of the proposed method, transfer learning is adopted to retrain the pre-trained LVQNN model by using industrial data. The retrained LVQNN is applied to control loops taken from chemical, paper and mining industries. Results highlight that the proposed method is able to provide correct verdict for majority of the control loops studied in the work.

Keywords: Identification Methods, Manufacturing Modeling for Management and Control
Abstract: The dynamic flux balance model (DFBA) is a constrained-based optimization modeling approach that has gained popularity for describing microbial cultures but has not been thoroughly investigated for mammalian cell cultures due to their relative complexity. This research aims to identify a DFBA model with minimal constraints and associated parameters to predict data for the fed-batch operation of a mammalian CHO cell culture. The Bayesian Information Criterion (BIC) is used to find a minimal set of kinetic constraints. The resulting DFBA model is used to predict 24 metabolites, biomass, and titer with 85 parameters that has a lower BIC and higher R2 as compared to previously reported kinetic models.

Keywords: Manufacturing Modeling for Management and Control
Abstract: We present an index-1 differential-algebraic equation (DAE) model for dynamic simulation of a calciner in the pyro-section of a cement plant. The model is based on first engineering principles and integrates reactor geometry, thermo-physical properties, transport phenomena, stoichiometry and kinetics, mass and energy balances, and algebraic volume and internal energy equations in a systematic manner. The model can be used for dynamic simulation of the calciner. We also provide simulation results that are qualitatively correct. The calciner model is part of an overall model for dynamical simulation of the pyro-section in a cement plant. This model can be used in design of control and optimization systems to improve the energy efficiency and CO2 emission from cement plants.

Keywords: Manufacturing Plant Control, Adaptive and Learning Systems
Abstract: In this paper, a new adaptive extended state observer based data-driven anti-disturbance control (AESO-DDADC) design is proposed for industrial nonlinear systems with unknown dynamics subject to external disturbances. By reformulating such system description into a compact-form dynamic linearization model with a residual term, a new AESO is firstly constructed to estimate the residual term using the partial derivative (PD) estimation from the previous time step, such that the residual term could be proactively counteracted by the feedback control law, in contrast to the existing data-driven ESO where the residual term in the PD estimation is absolutely neglected to facilitate the convergence analysis. Then, the bounded convergence of PD estimation and AESO is jointly analyzed by the Gerschgorin disk theorem, followed by robust convergence analysis of the established closed-loop system. Moreover, another AESO-DDADC scheme is developed using a partial-form dynamic linearization model of the system, along with rigorous robust convergence analysis. Finally, an illustrative example is shown to confirm the efficacy and advantages of the proposed designs.

Keywords: Manufacturing Plant Control
Abstract: Control of extrusion-based printing is fundamental to improving the traditional open-loop operation of commercial machines. However, the literature on feedback control based on nozzle motion is still limited in comparison with the developments for extrusion dynamics. In this work, we propose a model-based control strategy where the effects of the nozzle speed on the filament width are described by a first-order model, and the model uncertainty is used for robust synthesis. The feedback employs the internal model controller (IMC) to get an approximate inversion of the dynamics and the IMC filter is tuned based on robust performance criteria. The controller was experimentally tested and provided satisfactory results, especially when the parameters were refined depending on the speed regime.

Keywords: Experiment Design, Identification Methods, Model Predictive Control
Abstract: While designing excitation signals for identification of industrial processes, it is important to maintain desired model accuracies, reduce the experimental time and limit the output amplitudes within the specified bounds to avoid serious disruptions of the nominal process operation. In this work, we design a multi-frequency multi-amplitude square wave (multi-square) input based on either a nominal model by minimizing the experiment length and placing constraints on the model accuracy (in the frequency domain) and the output amplitudes. A separate design is carried out for each input where the resulting optimization problem has the same structure as a semi-definite program but with the decision variables restricted to integers corresponding to the number of half-periods of each square-wave. For processes with multiple inputs, the corresponding designs are carried out sequentially. The violations in the output constraints either due to model-plant mismatch or unmeasured disturbances should be mitigated by appropriate closed loop control actions. The efficacy of the proposed design is shown by means of a simulation case study.

Keywords: Identification Methods, Estimation and Robust Estimation, Data-Driven Optimization
Abstract: This paper explores some hierarchical extended parameter estimation algorithms for finite impulse response moving average (FIR-MA) model from observation data, including the hierarchical extended stochastic gradient algorithm, the hierarchical multi-innovation extended stochastic gradient algorithm, the hierarchical extended gradient algorithm, the hierarchical multi-innovation extended gradient algorithm, the hierarchical extended least squares algorithm and the hierarchical multi-innovation extended least squares algorithm. The proposed hierarchical algorithms for the FIR-MA systems can be extended to other stochastic systems with colored noises.

Keywords: Identification Methods, Estimation and Robust Estimation, Signal Processing
Abstract: This paper presents a systematic data-driven methodology to infer macroscopic reaction schemes and their associated kinetic laws from the measurements of concentration trajectories. The procedure uses sparse identification incorporated with a generic kinetic structure combining activation and inhibition factors. Only measurements of the extracellular species, i.e., biomass, substrates, and products of interest, are required, and measurement noise can be tackled using specific regularization techniques. The methodology is illustrated with a case study of a synthetic dataset from the production of therapeutic proteins using mammalian cell cultures.

Keywords: Identification Methods, Estimation and Robust Estimation
Abstract: Ensuring high prediction accuracy is essential for maintaining high quality of products in batch processes, given their inherent multi-phase characteristics and dynamic variations in real-world applications. Identifying quality-relevant process variations is crucial to address these challenges and produce interpretable and accurate predictions. This work aims to uncover critical quality-relevant process variables from raw measurements collected from batch processes. The proposed method consists of two key components. First, a dynamic subspace is designed for batch processes to extract the slow-varying features that are relevant to the quality index. Second, the quality-relevant features have been employed to achieve the reliable prediction of the performance index. Through the simulated experiment on a concentration batch production process, the proposed method is illustrated.

Keywords: Identification Methods, Production Planning, Data-Driven Optimization
Abstract: Chemical process monitoring is essential for product quality, plant efficiency, and safety. Conventional methods often prove inaccurate, particularly when dealing with nonlinear process behaviour. This paper presents a new approach that combines data-driven nonlinear system identification using smoothed L_1 regularisation and a state prediction method using a sequential importance resampling (SIR) particle filter to provide a basis for process monitoring. The results obtained from the polycondensation reaction in an operator training simulator (OTS) with real process conditions validate the effectiveness of the method in detecting anomalies, addressing challenges in nonlinear process modeling, and reliable state prediction for chemical process monitoring.

Keywords: Identification Methods, Machine Learning Assisted Modeling, Model Predictive Control
Abstract: The present paper introduces an innovative approach integrating recurrent neural networks, static models, and signal decomposition into base and residual behavior components for system nonlinear dynamic modeling and identification. The proposed methodology divides a nonlinear single-input single-output gas-lift oil production system into base response and residual components, assessing the first with a first-order dynamic model with variable gain and the latter with an Encoder-Decoder (E-D) GRU model. The study evaluates the methodology under various conditions, including noiseless and noisy data and scenarios with unmeasured disturbance. The percentage of stationary residues and the normalized root-mean-squared error (NRMSE) are applied to assess the model's performance. Overall, the proposed methodology demonstrates significant effectiveness, with NRMSE lower than 5% and percentages of stationary residues ranging from 90 % to 100 % across the study scope . The results stand out when compared to the direct application of E-D GRU model without decomposition, where the percentage of stationary residues was equal to 83%, and the training mean squared error was 10 times higher than that of noiseless scenarios.

Keywords: Data-Driven Optimization, Experiment Design, Machine Learning Assisted Modeling
Abstract: Bayesian optimization (BO) has proven to be an effective paradigm for the global optimization of expensive-to-sample systems. One of the main advantages of BO is its use of Gaussian processes (GPs) to characterize model uncertainty which can be leveraged to guide the learning and search processes. However, BO typically treats systems as black-boxes and this limits the ability to exploit structural knowledge (e.g., physics and sparse interconnections). Composite functions of the form f(x, y(x)), wherein GP modeling is shifted from the performance function f to an intermediate function y, offer an avenue for exploiting structural knowledge. However, the use of composite functions in a BO framework is complicated by the need to generate a probability density for f from the Gaussian density of y calculated by the GP (e.g., when f is nonlinear it is not possible to obtain a closed-form expression). Previous work has handled this issue using sampling techniques; these are easy to implement and flexible but are computationally intensive. In this work, we introduce a new paradigm which allows for the efficient use of composite functions in BO; this uses adaptive linearizations of f to obtain closed-form expressions for the statistical moments of the composite function. We show that this simple approach (which we call BOIS) enables the exploitation of structural knowledge, such as that arising in interconnected systems as well as systems that embed multiple GP models and combinations of physics and GP models. Using a chemical process optimization case study, we benchmark the effectiveness of BOIS against standard BO and sampling approaches. Our results indicate that BOIS achieves performance gains and accurately captures the statistics of composite functions.

Keywords: Fault Detection, Machine Learning Assisted Modeling, Adaptive and Learning Systems
Abstract: Differentiating between various types of faults and classifying them based on their importance is essential for process fault detection and diagnosis. This classification helps operators to prioritize their actions based on the severity of the faults. This paper proposes a reservoir computing-based slow feature analysis (RCSFA) to model complex and nonlinear industrial processes and study its application in fault classification while integrated with a graph neural network (GNN) and majority voting ensemble causality detection. To make the algorithm robust to unseen faults, real-time operator feedback is included by utilizing operator eye tracking. The practical applicability of the proposed method and its application in fault classification is studied through an industrial application.

Keywords: Derivative-Free Optimization, Data-Driven Optimization, Model Predictive Control
Abstract: Model predictive control (MPC) is one of the most effective technologies for optimal control of constrained multivariable systems. The closed-loop performance of MPC, however, can be sensitive to the choice of several tuning parameters that can appear in the prediction model, constraints, and/or cost function. Due to inherent limitations of manual tuning and the performance function depends on these parameters in an unknown manner, there has been increasing interest in “auto-tuning” using derivative-free optimization (DFO) methods. Bayesian optimization (BO) is a particularly powerful framework for data-efficient DFO of noisy, black-box functions Several recent works have shown the effectiveness of BO for MPC tuning when the number of tuning parameters is relatively small; however, MPC problems often involve a much larger number of parameters for which BO tends to struggle. In this paper, we propose to exploit a new type of Gaussian process surrogate model defined on sparse axis-aligned subspaces to mitigate the curse of dimensionality in BO. The approach is effective when closed-loop performance is sensitive to a small subset of tuning parameters, which is often the case in task-specific tuning problems. We demonstrate an order-of-magnitude performance improvement can be obtained with the proposed method compared to standard BO on a challenging inverted pendulum on a cart problem controlled by MPC with twenty independent tuning parameters

Keywords: Machine Learning Assisted Modeling, Fault Detection
Abstract: A novel Metabolic Graph Neural Network (MGNN) model is proposed for simulating the dynamic behavior of metabolites involved in oxidative stress metabolic pathways in a bacterial cell culture. The developed MGNN model is trained and validated with in-silico data generated from the mechanistic model. By using the a priori known metabolic network, the proposed MGNN model effectively reduces the overfitting issue as compared to a fully connected network that does not uses the metabolic network knowledge. The MGNN exhibits a superior fit for both training and testing datasets. The proposed MGNN is highly interpretable since it efficiently computes the relevance of each metabolite on any other metabolite by applying gradient computation and back-propagation operations to the neural network. The proposed model is also shown to be useful for fault detection.

Keywords: Machine Learning Assisted Modeling, Model Predictive Control
Abstract: We present a data-driven modelling and control approach for batch processes utilizing information from thermal images for feedback control. This work is driven by the requirement of utilizing the thermal image data that is the sole output of the system for feedback control. The overall goal here, like in many batch processes, is to obtain products with quality variables which match the user's specifications. The quality variables of the product cannot be measured online and is only measurable after the batch has terminated. The control problem is therefore not a setpoint tracking problem. We propose a multi-layered modelling approach. We first have a dimensionality reduction technique to reduce the high dimensional image to a set of few representative outputs. Then, we apply subspace identification (SSID) to identify a Linear Time Invariant (LTI) State space (SS) model between the inputs and the reduced outputs, and finally we construct a Partial Least Squares (PLS) model between the terminal states of a batch (identified using SSID) and the product qualities obtained for that particular batch. This model is utilized in a Model Predictive Control (MPC) formulation. We demonstrate the working of the MPC by showing its ability to achieve products with good quality.

Keywords: Model Predictive Control, Machine Learning Assisted Modeling, Optimization under Uncertainties
Abstract: In scenarios where high-fidelity physical models are either unavailable or are impractical due to their high complexity, data-based models offer a viable solution to obtain the system model necessary for predictive control. However, the accuracy of the predictions obtained by data-based models is limited. We propose to use neural networks with Bayesian last layer to obtain information about the uncertainty of the predictions. The weights of the Bayesian last layer are assumed to be Gaussian distributed, resulting in Gaussian distributed predictions. This paper demonstrates the use of Bayesian last layer surrogate models in a robust nonlinear model predictive control setting. The nonlinear model predictive control problem is adapted by considering the predicted uncertainty of the surrogate model, which can be efficiently computed using the Bayesian last layer method, in the cost function. The controller thus takes model uncertainty explicitly into account and by its formulation also avoids areas of extrapolation. The proposed method is applied to a mixed-suspension, mixed-product-removal crystallizer and simulation studies show that it outperforms a standard data-based model.

Keywords: Manufacturing Modeling for Management and Control, Power and Energy Systems, Optimal Control
Abstract: Chemical-looping Reforming (CLR) offers a promising process option for Blue Hydrogen production. Fixed-bed CLR reactors are ideal for process intensification, but require an effective control strategy to maximize hydrogen production in a dynamic process that needs to satisfy constraints imposed by process safety concerns and concomitant target products. To realize this goal, we use dynamic modeling and optimization for the design and control of an optimal fixed-bed CLR reactor. The optimal control strategy in the CLR reactor not only enables autothermal operation but also efficiently manages the heat recovery from the exhaust gas for feed gas preheating. As a result, the reactor produces syngas with an H2/CO ratio of 3 and generates stream of a high N2 concentration (> 98%) in each CLR cycle. Without hydrogen shift units, the reactor achieved a hydrogen yield efficiency of 62%.

Keywords: Model Predictive Control, Estimation and Robust Estimation, Power and Energy Systems
Abstract: Power-split HEVs can improve their overall fuel economy by applying an appropriate energy management strategy. One of the most popular energy management strategies is model predictive control (MPC) which has attracted a great deal of attention in the HEV research community. In this paper, we focus on the energy management of HEVs and propose a novel energy management system using a new robust predictive control based on almost strictly positive real properties. We also verify that the proposed method can improve fuel efficiency compared with the existing rule-based control through numerical simulations.

Keywords: Optimization under Uncertainties, Plant-Wide Optimization, Power and Energy Systems
Abstract: Optimizing subsea oil production systems utilizing recirculated gas lift and limited produced gas treatment capacity presents challenges. Real-time optimization (RTO) is a used method for optimizing such systems, but it is restricted by the lack of reliable sensors and the high cost of developing and updating models. As a result, the RTO is typically executed infrequently, and the optimal set points are not updated in real time, leading to suboptimal plant performance over extended periods. This study implements self optimizing control (SOC) techniques as an alternative solution that can handle frequent disturbances and drive the plant towards near-optimal performance without requiring frequent model updates or solver use. It compares different SOC structures in recirculated gas-lifted oil production optimization, their advantages, and disadvantages. The study concludes that SOC structures are an effective and suitable alternative to RTO, particularly in large and complex systems with limited measurement capabilities, given sufficient process system knowledge is considered for SOC design. This conclusion reinforces previous research, but with a more realistic case study.

Keywords: Identification Methods, Power and Energy Systems
Abstract: Lithium-ion batteries play a pivotal role in modern energy storage, offering high power, specific energy, and volumetric density, thereby establishing themselves as an eco-friendly alternative for diverse applications. As the demand for these batteries continues to grow, a comprehensive understanding of their intricate mechanisms becomes imperative. This study delves into the fundamental modeling of lithium-ion batteries, elucidating electrochemical processes and addressing aging complexities. Empirical models for Lithium-ion batteries, such as equivalent circuit models and open circuit voltage models, stand out for their real-time applications. Though these data-driven models achieve high accuracy and require low effort, their lack of explainability poses a limitation. Consequently, to comprehend the intricacies of batteries, it is essential to analyze the plausible causality inherent in internal electrochemical processes and aging effects, aspects challenging to capture in a simplified approach. This study, with a particular emphasis on aging, systematically scrutinizes model parameters, fitting them to experimental data, thereby unveiling subtle impacts on performance. The insights gained enhance predictive capabilities and contribute to the formulation of strategies for mitigating aging effects, ultimately extending the lifespan of lithium-ion batteries.

Keywords: Power and Energy Systems, Model Predictive Control
Abstract: Controlling gas turbines (GTs) efficiently is vital as GTs are used to balance power in onshore/offshore hybrid power systems with variable renewable energy and energy storage. However, predictive control of GTs is non-trivial when formulated as a dynamic optimisation problem due to the semi-continuous operating regions of GTs, which must be included to ensure complete combustion and high fuel efficiency. This paper studies two approaches for handling the semi-continuous operating regions of GTs in hybrid power systems through predictive control, dynamic optimisation, and complementarity constraints. The proposed solutions are qualitatively investigated and compared with baseline controllers in a case study involving GTs, offshore wind, and batteries. While one of the baseline controllers considers fuel efficiency, it employs a continuous formulation, which results in lower efficiency than the two proposed approaches as it does not account for the semi-continuous operating regions of each GT.

Keywords: Production Planning, Optimal Control, Plant-Wide Optimization
Abstract: We recently investigated the optimal flexible operation of electrified biodiesel production, employing different process configurations with buffer tanks but without heat integration (https://arxiv.org/abs/2308.09537). Herein, we study the implications of incorporating heat integration on process flexibility. We present two process configurations that include heat integration across all three process columns. Within these configurations, one incorporates additional heating units for reboilers, while the other operates without them. Expectedly, introducing additional heating units increases process flexibility, yielding higher energy cost savings. We also propose an alternative configuration wherein we deploy two decentralized optimizers, reducing computational expenses while achieving comparable energy cost savings.

Keywords: Optimal Control, Environment and Agriculture, Model Predictive Control
Abstract: Image-based control and sensing has been applied in a wide variety of next generation manufacturing fields. Utilizing methods of simulating closed-loop image-based control may be advantageous for improving control performance and design without the need for an experimental setup. One software capable of these simulations is the open-source 3D modeling software Blender, which has many capabilities aided by a Python API. This work explores the use of Blender as an image-based control test bed, where both the process and the controller are simulated, in the context of a greenhouse supplemental lighting control system.

Keywords: Model Predictive Control, Optimal Control
Abstract: The paper extends recent advancements in self-stabilizing eNMPC formulation without pre-calculated setpoints, which leverages norm-based steady-state optimality conditions to enhance system robustness. To facilitate practical implementation, a generalized time-domain formulation is proposed, accommodating the discrete-time nature of control instrumentation and the continuous-time nature of first-principle models. The online computational time of the self- stabilizing eNMPC is improved via the simplification of the Lyapunov function. A case study involving a modular membrane reactor illustrates the applicability of self-stabilizing eNMPC in real-world industrial scenarios.

Keywords: Model Predictive Control, Optimal Control, Manufacturing Plant Control
Abstract: A predictive modeling framework for silicon production in fluidized bed reactors is proposed to characterize the particle size distribution of the product and the powder loss. Two different flow regime modeling approaches are considered to describe the silane pyrolysis reaction and characterize the deposition rate that contributes to particle growth. A discrete population balance equation is used to estimate the particle size distribution as a function of the deposition rate. A nonlinear model predictive control is then utilized to regulate the system at the desired operating conditions. Detailed open-loop and closed-loop simulation studies demonstrate the successful integration of nonlinear MPC and the proposed predictive modeling approach.

Keywords: Model Predictive Control, Optimal Control
Abstract: We propose a feasibility condition for reference governors that are used to tune the behavior of a reference-tracking explicit model predictive controller. Governors allow for a simple tuning of existing controllers without modifying them. However, since properties like feasibility are often not accounted for by the governors, their applicability may be limited. We prove that feasibility can be guaranteed using information about the explicit solution, and demonstrate the resulting algorithm by controlling the flow rate in a laboratory-scale hydraulic plant.

Keywords: Model Predictive Control, Optimal Control, Smart Cities
Abstract: Energy efficient building climate control involves maintaining thermal comfort across a wide range of environmental conditions while minimizing energy usage. However, the design of energy efficient control poses a significant challenge owing to the strong coupling between temperature and humidity. In this work, we present a control oriented model for the heating, ventilation and air conditioning (HVAC) system and provide a polytopic approximation of thermal comfort in terms of temperature and humidity ratio. A novel energy optimal control formulation based on generalized disjunctive programming is proposed to systematically account for the strong coupling effects and latent heat consideration. An extensive simulation study is performed to validate the efficcacy of the proposed control strategy across a wide range of operational and weather conditions.

Keywords: Power and Energy Systems, Model Predictive Control
Abstract: We use model predictive control (MPC) for the optimal energy distribution in nonresidential buildings. Our approach is special in that it treats thermal and electrical energy flows simultaneously. Our sample application is a real office building, where components such as heat pumps and heating rods introduce discrete variables. This implies the optimal control problem that must be solved for MPC is a mixed-integer quadratic programming (MIQP) problem. Because both continuous and integer variables are involved, the computation times may become prohibitive for use in real-time. We explore a computationally efficient approximation that replaces integer variables by continuous variables for later time steps along the horizon. The performance of MPC using this method is investigated in simulations and the results are compared to those for the original MIQP problem and solution. Our method significantly reduces computational time while achieving a nearly optimal solution.

Keywords: Model Predictive Control, Manufacturing Plant Control, Supervisory Control
Abstract: As biomanufacturing transitions from batchwise to continuous operations there is an eminent need for increased attention to dynamic process control. In this contribution we describe a nonlinear model-predictive controller that we designed and implemented for control of a perfusion (continuous) Chinese hamster ovary (CHO) bioreactor producing a monoclonal antibody therapeutic. Through experimental results, we have demonstrated the superior regulatory performance of this advanced control strategy compared to an existing, PI control scheme.

Keywords: Fault Detection, Signal Processing, Data-Driven Optimization
Abstract: The Hall-Héroult process is limited in terms of on-line signals to enable successful application of advanced analytics to improve process control and monitoring. Although some successful small scale applications have been presented, the industry is still struggling to deploy those techniques at industrial scale. This presentation reviews some advanced analytics initiatives, pertaining to the Hall-Héroult process and discusses factors preventing success.

Keywords: Manufacturing Modeling for Management and Control
Abstract: This paper illustrates the use of MPC in fiber- and display-glass-making processes at Corning Inc. Methods used include linear- and nonlinear multivariate MPC, sometimes in combination with surrogate models. The authors will also discuss some future directions in the use of real-time optimization.

Keywords: Plant-Wide Optimization
Abstract: This paper conducts a retrospective analysis to understand why, despite initial triumphs in real-time optimization, the momentum of this technology waned. We review technology and some key dimensions of human factors, elucidating their pivotal roles in the outcome of automation projects. By relating historical insights with contemporary possibilities, we aim to inform new strategies for future sustainable industrial real-time optimization projects.

Keywords: Data-Driven Optimization, Plant-Wide Optimization
Abstract: Surrogate models are important in real-time optimization (RTO) when the first-principles model is unavailable or computationally challenging for online optimization. Among different surrogate models, the continuous piecewise linear (CPWL) model enjoys the universal approximation ability and potential computational benefits. However, the CPWL surrogate model poses three challenges to current RTO algorithms. First, the solution of a CPWL model is always located on the boundary of a polytopic subregion, while the plant optimum may be in the interior of a subregion. Second, the CPWL model is nonsmooth, which cannot be handled by RTO methods that rely on gradient matching. Third, the resulting nonsmooth optimization subproblems are hard to solve. This paper addresses the difficulties by adding a quadratic function to the CPWL surrogate model, extending a classical RTO method to accommodate nonsmoothness, and exploiting the difference-of-convex structure of the surrogate model for efficient solution. The advantages of the proposed method are demonstrated through a benchmark problem in the RTO literature.

Keywords: Learning-Based Control, Manufacturing Plant Control
Abstract: Iterative learning control (ILC) has been considered a powerful strategy for repetitive process control. However, a fundamental assumption of conventional ILC is that each cycle must start from a predetermined fixed initial state. This assumption can be strict and challenging to achieve in real-world industrial applications. To address the issues arising from varying initial states, we propose an ILC framework that learns from a virtual cycle generated using historical data. We establish three conditions for generating the virtual cycle, and theoretical results demonstrate guaranteed convergence. To ensure the practicality of our framework, we relax one of the conditions, enabling the virtual cycle to be generated by solving a convex optimization problem. The effectiveness of our framework in improving control performance is verified through an injection molding example.

Keywords: Fault Detection
Abstract: In this paper, a data-based methodology for performance monitoring of control loops under set point tracking and disturbance rejection is presented. A benchmark based on historical data is validated using well-known time domain performance indexes and then used for performance monitoring. Performance indexes are proposed based on the tasks performed by the controller and statistical tests provide evidence about changes in the performance. The methodology is illustrated through its application to a temperature control loop subject to set point changes and measured and unmeasured disturbances. The introduced faults were detected and discriminated based on the change in the proposed performance indexes.

Keywords: Fault Detection
Abstract: Process faults often lead to process/equipment failure or an emergency abnormal situation and are of greater concern in process industries. Multivariate statistical process monitoring methods using Singular Spectrum Analysis (SSA) have proved to be an effective tool for chemical process monitoring among other multivariate multiscale methods and are extensively studied and widely used for fault detection. In this study, Singular Spectrum Decomposition, a data-adaptive nonparametric method originated from SSA is used for the decomposition of signals into multilevel components which take care of autocorrelation within the process variables. In SSD, unlike SSA, the determination of crucial parameters like the embedding window dimension for building the trajectory matrix and the number of principal components for grouping and reconstructing time series is automated through consideration of the data's frequency content. The proposed approach is applied to detect faults in simulated and industrial data. The evaluation of results showed that the proposed method effectively detects faults in lower scales/levels as compared with the conventional SSA-based method.

Keywords: Resilient, Safe, and Cyber-Secure Systems, Model Predictive Control, Learning-Based Control
Abstract: Cyberattacks on control systems can create unprofitable and unsafe operating conditions. To enhance safety and attack resiliency of control systems, cyberattack detection strategies can be developed. Prior work in our group has sought to develop cyberattack detection strategies that are integrated with an advanced control formulation known as Lyapunov-based economic model predictive control (LEMPC), in the sense that the controller properties can be used to analyze closed-loop stability in the presence or absence of undetected attacks. In this work, we consider neural network-approximated control laws, concepts for mitigating cyberattacks on such control laws, and how these ideas elucidate concepts in how to fight back against cyberattacks. We begin by providing sufficient conditions under which a neural network (NN) that approximates an LEMPC maintains safety for a sampling period after a cyberattack by inheriting safety properties from the LEMPC formulation. Then, we discuss a second concept inspired by neural network repair in the presence of adversarial attacks for attempting to ensure safety of controllers for a time period after undetected attacks, even those not based on a rigorous control law formulation like LEMPC. We examine the potential conservatism differences between the LEMPC-based safety strategy and one based on repairing problematic control actions, and discuss how this concept can inspire ideas for fighting back against attacks.

Keywords: Fault Detection, Large-Scale and Networked Systems
Abstract: Stimulated by the increasing demands for system safety and process reliability in complex industrial processes, this paper investigates a weighted average consensus algorithm based distributed fault detection scheme for large-scale interconnected systems, using a sensor network where each node is equipped with a Kalman filter (KF). To reduce the communication and computation efforts, the proposed distributed fault detection scheme is splitted into two phases: distributed offline training and online fault detection. To this end, the Expectation-Maximization (EM) algorithm is firstly addressed to identify the unknown measurement matrices and covariance matrices of noise vectors. It is followed by an average consensus algorithm so that the identical Kalman filters can be designed in parallel at all sensor nodes. On this basis, distributed residual generators and test statistics are constructed for fault detection purpose using the average consensus algorithm. Considering that there exist some special conditions, such as the occurrence of node failures, a variation of the distributed Kalman filter based fault detection scheme is proposed by dynamically adjusting the consensus weight. Finally, the feasibility and effectiveness of the proposed scheme are demonstrated through a case study on the waste water treatment plants (WWTPs).

Keywords: Resilient, Safe, and Cyber-Secure Systems, Model Predictive Control, Manufacturing Plant Control
Abstract: Prior research from our group developed a control-integrated active actuator cyberattack detection strategy. This strategy continuously probed for cyberattacks by updating target steady-states at every sampling time and then moving the process state toward these over the subsequent sampling period. Attacks were flagged if a Lyapunov function around the target steady-state did not decrease over a sampling period. This strategy had the benefit of ensuring safety of the process until an attack was detected. However, the continuous probing for attacks could decrease profit from the process compared to not probing for the attacks, which could limit the attractiveness of the method in practice. This work marks our first step toward attempting to develop a framework for modifying this detection strategy to make guarantees that the profit over a sampling period would be no worse than that of a stabilizing controller. This is achieved through utilizing two auxiliary controllers, in addition to the one which facilitates the attack-probing, with constraints on profits in the various controllers to enable the profit proofs over a sampling period (in the absence of disturbances) to be developed. A process reactor example is used to demonstrate the implementation of the detection strategy.

Keywords: Data-Driven Optimization, Derivative-Free Optimization, Supply Chain and Enterprise Integration
Abstract: Reinforcement learning (RL) has been widely applied in supply chain management due to its performance in dynamic, uncertain environments. However, most RL studies focus on a single objective, differentiable reward functions, and lack the ability to handle multiple conflicting non-differentiable objectives which is the case in many real-world problems such as in inventory control. The proposed multi-objective algorithm deploys a derivative-free approach to effectively optimize non-differentiable objective functions. The framework leverages the advantages of both reinforcement learning (RL) methods and multi-objective evolutionary algorithms (MOEAs) to obtain a Pareto set of policies. The effectiveness of our method is demonstrated through two case studies, each illustrating the adaptability of the policy of choice in varying scenarios. Our methodology finds a diverse set of policies, which allows decisionmakers to better handle and mitigate the consequences of disruptions.

Keywords: Resilient, Safe, and Cyber-Secure Systems, Model Predictive Control, Optimal Control
Abstract: Control-theoretic cyberattack detection strategies are control strategies where control theory can be used in the design of the detection policies and analysis of stability properties with and without cyberattacks. This work provides a step toward understanding how to diagnose cyberattacks using control-theoretic cyberattack detection mechanisms. Specifically, we analyze the conditions under which a control-theoretic cyberattack detection strategy developed in our prior work to handle detection of simultaneous actuator and sensor attacks can be extended to distinguish between whether attacks are occurring on sensors or actuators. We present and evaluate heuristic concepts for attempting to diagnose sensor attacks; these again demonstrate the utility of control-theoretic diagnosis policies and lead to further suggestions for such control-theoretic policies.

Keywords: Model Predictive Control, Manufacturing Plant Control
Abstract: Many large-scale multi-input multi-output systems are treated as a combination of single-input single-output systems in reality. At such times, interference from input signals not focused is regarded as observable disturbances. For observable disturbances, feed-forward controllers are effective to reject the influence. A simple feed-forward controller construction is a combination of transfer functions of controlled and disturbance systems. This paper proposes an extension of the simple feed-forward controller and its parameter tuning method. The controller is designed based on the a Predictive Functional Controller (PFC), one of the Model Predictive Control. Effectiveness of the proposing scheme is verified by some simulation examples.

Keywords: Estimation and Robust Estimation
Abstract: Designing state estimation strategies for discrete time nonlinear dynamic systems often requires incorporation of inequality constraints on the state estimates. These constraints may arise from physical considerations and/or operational perspective which can significantly increase the complexity of the estimator design problem. The estimation problem becomes even more challenging for distributed parameter systems (DPSs) due to spatial dependency of the system states. This paper provides two novel approaches for estimating the state profiles of DPSs while adhering to the imposed bounds. The first approach utilizes the idea of constraining the maximum and minimum values that spatial state profiles can take along the spatial domain. The second approach employs a characteristic property of Bernstein polynomials to achieve state profile estimates within the specified bounds. Proposed approaches are compared with an existing approach that incorporates bound constraints at a large number of discretization points. However, it does not guarantee constraint satisfaction by the entire profile. The performance of the proposed approaches is evaluated by simulation studies conducted on an auto-thermal tubular reactor system. The analysis of the estimation results demonstrates that the proposed approaches are capable of producing accurate profile estimates within the specified bounds.

Keywords: Experiment Design, Estimation and Robust Estimation
Abstract: In the context of parameter estimation, under input uncertainty, the probability distribution function (pdf) of the measurement data mismatches the true pdf of measurement with accurate input. In this scenario, the Cramer-Rao bound (CRB), which is widely used in optimal experimental design, may become an overoptimistic lower bound on parameter estimation error covariance. To tackle this issue of mismatched measurement distribution subject to input uncertainty, in this work, a novel optimal sampling time design is proposed that employs the misspecified Cramer-Rao bound (MCRB), with the aim to collect informative data for high-quality parameter estimation. The MCRB is formed following the Cauchy-Schwarz inequality using the true pdf of the measurement, approximated by the statistics of measurement samples. In the numerical study, large samples from the input uncertainty space are generated and applied to the underlying system model; the outputs are calculated and used to approximate the true measurement pdf. The proposed MCRB-based sampling time design is formulated as a non-convex integer programming optimisation problem solved by a conjugate direction method. Three sampling time designs, the uniform sampling, the CRB-based design and the MCRB-based design, are tested on a benchmark enzyme reaction system model. The results show the necessity and superiority of using MCRB for experimental design under input uncertainty.

Keywords: Fault Detection, Machine Learning Assisted Modeling, Identification Methods
Abstract: Abnormal plant operations are caused by disturbances, process measurement faults, or malfunctioning equipment. Steady-state or dynamic models of the process units are widely available. Since continuous process plants operate under closed-loop control and available plant data often covers a narrow operating window, the process model can generate normal operating data over a wider window to train an autoencoder to represent that data. For deployment in real-time, the plant model accepts process inputs from the plant and calculates outputs; one instance of the autoencoder accepts data from the plant, and the other accepts data from the model. The occurrence of a process fault leads to differences in the latent space variables of the two instances of the autoencoder, which enables fault detection. Compared to a traditional PCA-based fault detection framework, an autoencoder-based framework can model nonlinear processes, which is not possible by using PCA or dynamic PCA.

Keywords: Fault Detection, Estimation and Robust Estimation, Machine Learning Assisted Modeling
Abstract: Fault estimation is crucial for ensuring reliability and safety throughout industrial processes. However, the increased nonlinearity and complexity in modern systems, as well as their feedback control logic, multiply the challenges when estimating faults. Health monitoring in today’s systems may impact the overall cost substantially. To address such challenges, we present a hybrid fault estimation scheme for nonlinear systems, by incorporating an Extended Kalman Filter along with inferential sensors. These fault-sensitive sensors are developed using symbolic regression combined with information theory, to be cost-effective supplements to the existing hard sensors. The proposed method was applied to open-loop and closed-loop architectures of a cross-flow plate-fin heat exchanger dynamic model toward estimating the fault severity at various levels of measurement noise. To showcase the robustness of the inferential sensors, we compared the performance of the proposed framework to an Extended Kalman Filter designed solely with information from hard sensor measurements.

Keywords: Manufacturing Plant Control, Model Predictive Control, Production Planning
Abstract: Dynamic market conditions as a consequence of increased globalization, coupled with fluctuations in electricity prices brought about by the deregulation of energy markets, require process manufacturing plants to operate in a responsive manner in order to remain competitive. In particular, the quasi steady-state assumption that is typically applied in optimal scheduling does not hold in a highly dynamic operating environment, where the dynamics of transitions have an increasingly significant impact. This has led to a research thrust on the integration of scheduling and control. In this paper, we provide an overview of this topic, highlighting assumptions and formulations related to the plant control system. We then focus on a class of ‘controller aware’ scheduling formulations, in which the predicted closed-loop response of the plant under the action of the plant control system is taken into account. A case study illustrating key concepts is presented.

Keywords: Data-Driven Optimization, Supply Chain and Enterprise Integration, Plant-Wide Optimization
Abstract: This work introduces a methodology to design and optimize chemical process flowsheets through a novel hybrid Proximal Policy Optimization (HPPO) agent. The novelty of this work lies in the use of masking, a technique never considered before in the design of process flowsheets capable of finding correlations between the composing unit-operations (UOs). The performance of this novel agent is tested using case studies, one of which employs ASPEN Plus. The results obtained from both cases demonstrate significant learning on the part of the agent, yielding outcomes in line with the problem’s specifications.

Keywords: Machine Learning Assisted Modeling, Estimation and Robust Estimation, Identification Methods
Abstract: This work presents a Bayesian approach to evaluating the uncertainty of physics-informed neural network models. The proposed strategy uses a hybrid methodology for training and assessing the uncertainty of model parameters. In the first part of the training, a gradient-based algorithm is used to train and obtain the weights. In the second stage, a Markov Chain Monte Carlo algorithm is used to evaluate the uncertainty of the network weights. The developed method was used to solve Burger’s equation, and the results show that it was possible to characterize the uncertainty region of the PINNs’ prediction.

Keywords: Experiment Design
Abstract: Applying model-based design of experiments to compute maximally-informative campaigns with multiple parallel runs is challenging. Effort-based methods can overcome some of these challenges through discretizing the experimental space into a finite set of candidate experiments, then applying convex optimization techniques to determine the optimal efforts for each candidate, and finally rounding the efforts to integer numbers of runs for a target experimental campaign size. For small experiment campaigns in particular, the final rounding can result in large suboptimality. This paper presents an approach to solving the exact design problem, where the effort variables being optimized are constrained to taking integer values. We consider model parametric uncertainty and formulate risk-inclined, risk-neutral and risk-averse exact design problems as mixed-integer nonlinear programs (MINLPs) with convex participating functions. We demonstrate the tractability of an outer-approximation algorithm to solve such MINLPs to global optimality on a case study involving the exothermic esterification of propionic anhydride with over 1000 experiment candidates and 100 uncertainty scenarios.

Keywords: Model Predictive Control, Optimal Control, Manufacturing Plant Control
Abstract: Recent increasing interest in quantum computers has spurred research into practical engineering applications for quantum algorithms. One potential application is process control. The unique quantum phenomena involved with quantum computing brings up interesting considerations for control. This work focuses on non-determinism, first through a motivating simulation utilizing a continuous stirred-tank reactor. Following this, two methods of potentially ensuring system stability in the presence of non-determinism are discussed. The first involves including an additional gate to the modified Grover’s algorithm presented in our previous work, which is designed to prevent a qubit state corresponding to an undesired control input from being measured. The second method involves defining an inner region where, if the state leaves the region, a classical stabilizing controller is activated to drive the state back inside the region.