Leyla Özkan

On the way to autonomous model predictive control : A distillation column simulation study

Model Predictive Control (MPC) is a powerful tool in the control of large scale chemical processes and has become the standard method for constrained multivariable control problems. Hence, the number of MPC applications is increasing steadily and it is being used in application domains other than petrochemical industries. A common observation by the industrial practitioners is that success of any MPC application requires not only efficient initial deployment but also maintenance of initial effectiveness. To this end, we propose a novel high level automated support strategy for MPC systems. Such a strategy consists of components such as performance monitoring, performance diagnosis, least costly closed loop experiment design, re-identification and autotuning. This work presents the novel technological developments in each component and demonstrates them on a distillation column case study. We show that automated support strategy restores nominal performance after a performance drop is detected and takes the right course of action depending on its cause.

Generalized predictive control tuning by controller matching

The tuning of state-space model predictive control (MPC) based on reverse engineering has been investigated in literature using the inverse optimality problem ( [1] and [2]). The aim of the inverse optimality is to find the tuning parameters of MPC to obtain the same behavior as an arbitrary linear-time-invariant (LTI) controller (favorite controller). This requires equal control horizon and prediction horizon, and loop-shifting is often used to handle non-strictly-proper favorite controllers. This paper presents a reverse-engineering tuning method for MPC based on transfer function formulation, also known as generalized predictive control (GPC). The feasibility conditions of the matching of a GPC with a favorite controller are investigated. This approach uses a control horizon equal to one and does not require any loop-shifting techniques to deal with non-strictly-proper favorite controllers. The method is applied to a binary distillation column example.

A Lagrangian Method for Model Reduction of Controlled Systems

This paper presents a method for closed-loop order reduction of linear systems. An approximation is carried out on the Lagrangian or Hamiltonian system that is obtained from the problem to minimize an optimization criterion subject to plant dynamics and system constraints. The resulting Hamiltonian system is reduced in complexity by means of a standard reduction techniques. The merits of the method are illustrated on an example of a distillation process.

A Frequency Domain Approach for MPC Tuning

Abstract This paper presents a frequency domain based approach to tune the penalty weights in the model predictive control (MPC) formulation. The two-step tuning method involves the design of a favourite controller taking into account the model-plant mismatch followed by the controller matching. We implement this approach on a SISO example.

Model reduction of multi-variable distributed systems through empirical projection spaces

This paper considers the problem of finding optimal projection spaces for the calculation of reduced order models for distributed systems. The method of proper orthogonal decompositions is popular in the reduction of fluid dynamics models, but may become rather cumbersome for the reduction of systems in which the total dimension of physical variables is large. This paper aims to deal with this problem and proposes the construction of projection spaces from tensor representations of observed, measured or simulated data. The method is illustrated for the reduced order modeling of a tubular reactor.

Dynamic modeling of ultrafiltration membranes for whey separation processes

In this paper, we present a control relevant rigorous dynamic model for an ultrafiltration membrane unit in a whey separation process. The model consists of a set of differential algebraic equations and is developed for online model based applications such as model based control and process monitoring. In this model, membrane resistance concept is adjusted to describe the membrane fouling. Based on the observations regarding the permeate flux, we propose a membrane resistance expression consisting of static and dynamic resistances. The empirical expressions for the membrane resistances are identified by solving a parameter estimation problem. The dynamic model is investigated for its predictive capabilities and is further utilised for the study of optimal operation strategies.

Safety analysis for a class of graph constrained scheduling problems

We consider a class of scheduling problems, common in manufacturing industries that consist of several interconnected subprocesses. We model the scheduling constraints of each subprocess and of the overall system using labeled directed graphs, forming the admissible set of schedules. Furthermore, we consider polytopic constraints on the state space. We approach the scheduling problem as a safety analysis problem utilizing reachability mappings, system and constraint structure. Consequently, we construct the safe schedules that guarantee constraint satisfaction at all times. The proposed framework is illustrated in a case study that concerns a simplified separation process.

Monitoring of Batch Industrial Crystallization with Growth, Nucleation, and Agglomeration. Part 2: Structure Design for State Estimation with Secondary Measurements

This work investigates the design of alternative monitoring tools based on state estimators for industrial crystallization systems with nucleation, growth, and agglomeration kinetics. The estimation problem is regarded as a structure design problem where the estimation model and the set of innovated states have to be chosen; the estimator is driven by the available measurements of secondary variables. On the basis of Robust Exponential estimability arguments, it is found that the concentration is distinguishable with temperature and solid fraction measurements while the crystal size distribution (CSD) is not. Accordingly, a state estimator structure is selected such that (i) the concentration (and other distinguishable states) are innovated by means of the secondary measurements processed with the geometric estimator (GE), and (ii) the CSD is estimated by means of a rigorous model in open loop mode. The proposed estimator has been tested through simulations showing good performance in the case of mismatch in the initial conditions, parametric plant-model mismatch, and noisy measurements.

Monitoring of Batch Industrial Crystallization with Growth, Nucleation, and Agglomeration. Part 1: Modeling with Method of Characteristics

This paper develops a new simulation model for crystal size distribution dynamics in industrial batch crystallization. The work is motivated by the necessity of accurate prediction models for online monitoring purposes. The proposed numerical scheme is able to handle growth, nucleation, and agglomeration kinetics by means of the population balance equation and the method of characteristics. The former offers a detailed description of the solid phase evolution, while the latter provides an accurate and efficient numerical solution. In particular, the accuracy of the prediction of the agglomeration kinetics, which cannot be ignored in industrial crystallization, has been assessed by comparing it with solutions in the literature. The efficiency of the solution has been tested on a simulation of a seeded flash cooling batch process. Since the proposed numerical scheme can accurately simulate the system behavior more than hundred times faster than the batch duration, it is suitable for online applications such as process monitoring tools based on state estimators.

Optimal start-up and operation policy for an ultrafiltration membrane unit in whey separation

Membrane filtration systems are preferred unit operations in industrial applications due to their mild operating conditions. However the performance of a membrane stack drops over time because of the membrane fouling. This decrease is overcomed by introducing clean membrane stacks. The associated scheduling problem, when to include new membrane stacks to the operation, is the main topic of this paper. We construct a dynamic optimization problem to find the optimal time instants of introducing new membrane stacks. Furthermore, the optimal operating pressure profile for the optimal scheduling strategy is constructed from the desired output specifications. The result of the simulation study indicates that the optimal scheduling strategy improves the operation by slowing down the accumulation of fouling.