Bart Huyck

Towards Online Model Predictive Control on a Programmable Logic Controller: Practical Considerations

Given the growing computational power of embedded controllers, the use of model predictive control (MPC) strategies on this type of devices becomes more and more attractive. This paper investigates the use of online MPC, in which at each step, an optimization problem is solved, on both a programmable automation controller (PAC) and a programmable logic controller (PLC). Three different optimization routines to solve the quadratic program were investigated with respect to their applicability on these devices. To this end, an air heating setup was built and selected as a small-scale multi-input single-output system. It turns out that the code generator (CVXGEN) is not suited for the PLC as the required programming language is not available and the programming concept with preallocated memory consumes too much memory. The Hildreth and qpOASES algorithms successfully controlled the setup running on the PLC hardware. Both algorithms perform similarly, although it takes more time to calculate a solution for qpOASES. However, if the problem size increases, it is expected that the high number of required iterations when the constraints are hit will cause the Hildreth algorithm to exceed the necessary time to present a solution. For this small heating problem under test, the Hildreth algorithm is selected as most useful on a PLC.

Identification and modeling of distillation columns from transient response data

Mathematical models that predict the behavior of the column are highly useful to improve the working conditions of distillation columns, i.e. increase throughput or lower energy consumption. They are a prerequisite for advanced control strategies. White-box models of distillation columns already exist, but they are often too time consuming to calibrate in real industrial cases. This paper models a binary distillation column using a black-box model in the frequency domain. One of the identification challenges of distillation columns is the long transient response that causes leakage errors in the frequency domain. This problem is solved using a recently developed local polynomial method. Furthermore, we propose a simple expedient to model the variations in the column response that are caused by the changes in ambient temperature. The results are illustrated using both simulation data of a white-box model and using real measurements.

Identification of a Pilot Scale Distillation Column: A Kernel Based Approach

Abstract This paper describes the identification of a binary distillation column with Least-Squares Support Vector Machines (LS-SVM). It is our intention to investigate whether a kernel based model, particularly an LS-SVM, can be used for the simulation of the top and bottom temperature of a binary distillation column. Furthermore, we compare the latter model with standard linear models by means of mean-squared error (MSE). It will be demonstrated that this nonlinear model class achieves higher performances in MSE than linear models in the presence of nonlinear distortions. When the system is close to linear, the performance of the LS-SVM is only slightly better than the linear models.

Implementation and experimental validation of classic MPC on Programmable Logic Controllers

Over the last years, a number of publications were written about Model Predictive Control (MPC) on industrial Programmable Logic Controllers (PLC). They focussed on explicit MPC strategies to provide a fast solution. When sufficient time is available to solve a classic MPC problem, an online solution to the corresponding Quadratic Problem (QP) can be provided. This paper investigates the use of an online quadratic programming solver to exploit MPC on a PLC. This will be illustrated with the classic Hildreth QP algorithm and qpOASES, a recently developed online active set strategy. These algorithms will be investigated on a MISO system.

Constrained Model Predictive Control on a Programmable Automation System Exploiting Code Generation: Practical Considerations

Abstract This paper examines the use of code generators for Model Predictive Control (MPC) on a Programmable Automation Controller (PAC). Recently, a strong interest has been observed in code generators able to produce dedicated code for solving optimization problems on embedded devices. This interest has consecutively raised questions about usability, flexibility and user-friendliness of these code generators in more standard industrial hardware or lab equipment. To this end, the current paper investigates as a benchmark example the constrained model predictive control of a real-life Two Input – Single Output hair dryer system that is controlled with a CompactRIO controller (National Instruments). The aim is to provide practical information throughout the description of the entire development process (i.e., from model identification over code generation to MPC implementation and validation). Hence, advantages as well as possible limitations and pitfalls are explored.

Identification of a Noninsulated Distillation Column From Transient Response Data

Distillation is the most used separation technique worldwide. However, it is also an energy-intensive method. A way to reduce the energy consumption is to optimize the working conditions of distillation processes. This paper uses system identification to find an accurate model of a distillation column that can be used in, for example, model predictive control. Identification of a pilot-scale distillation column is done in the frequency domain in two steps. First, a nonparametric local polynomial method (LPM) is used to eliminate the leakage effects and to estimate the frequency response and noise contributions. A fast and a robust LPMs are compared. Second, results from the first step are used to find a low-order rational Laplace model with a time-delay term. Due to lack of insulation, the systems operating point is varying with ambient temperature. By using a reparametrization of the initial model, it is possible to account for these changes. The final result is a metamodel, which depends on the frequency and the effective separation heat of the column. In the future, this model can be applied in model predictive control to optimize the working conditions of a distillation column. Simulation data and real measurements on a pilot-scale distillation column are used to illustrate the results.