For more information on the Autoprofit project please contact the coordinator: Paul Van den Hof, .

This project is made possible by


the EU

Project description

WP1 - Management and Coordination

The overall objective of the management WP is the smooth delivery of the project, and in particular:

  • To co‐ordinate and supervise activities to be carried out
  • To carry out the overall administrative and financial management of the project
  • To manage the contract with the European Commission and the Consortium Agreement
  • To manage contacts with the European Commission
  • To monitor quality and timing of project deliverables
  • To establish effective internal and external communication procedures

Lead Partner: Technische Universiteit Delft

WP2 - Models in model-based systems

The life-time performance of model predictive controllers (MPC) is rather limited, particularly due to the fact that over time the underlying dynamic models do not cover the process behavior any more. To recover the control performance, the models have to be adapted. In order to do this, there are basically two options: (i) to keep the model structure and re-identify the model parameters or (ii) to change the model structure and identify the model parameters of the new structure.

The task of WP2 can be divided into two sub-problems. In the first instance, one needs criteria to decide if a new model structure is required. If so, the second task is to provide a new, more accurate model structure, which hopefully leads to a better control performance of the MPC. In particular, these tasks include the adaptation of linear model structures and the decision, whether to switch from a linear to a nonlinear MPC. However, in contrast to the other WPs, WP2 is not limited to linear models and linear MPC. The aim is to find criteria that are also applicable to general nonlinear MPC.

Unfortunately, our literature survey did not yield satisfactory results. On the one hand, there are so-called open-loop measures, which extrapolate only from the open-loop model accuracy to the controller adequacy. For example, they fully disregard the linearizing effect of the closed-loop. On the other hand, there are so-called closed-loop measures. However, they do not apply MPC. They are limited to linear controllers, because in this case it is possible to generate analytical results.

In WP2, we are working on a methodology to assess the potential of a particular model structure extension for an improvement in control performance. This way it could be possible to decide if a change of the model structure could improve the control performance and which model structure should be chosen. In a first step, we limit ourselves to polynomial ARX models. Later on, the goal is to extend the method to other types of nonlinear model structures and to be able to include prior process knowledge in the decision making process of the best model structure.


WP2 Poster

Lead Partner: Rheinisch-Westfaelische Technische Hochschule Aachen (RTHA)

WP3 - Least costly closed loop testing

The overall objective of this work-package is the development of a comprehensive methodology, and tools, for closed-loop testing where the focus is on autonomy and minimum use of resources. The methodology should be adapted to the requirements of constrained control, such as MPC. So far progress with respect to applications oriented open loop experiment design for linear multi-input/multi-output systems has been made. In a sequence of contributions, [1]—[3], a method for open loop experiment design, tailored for MPC-applications, has been developed. Also a first draft of a MATLAB toolbox for experiment design has been launched; it can be downloaded from [3]. It was presented at the ERNSI (European Network on System Identification) workshop in Nice in September 2011.  Prof. Xavier Bombois presented a plenary at JIME 2011 on control oriented experiment design, see [4]. Current focus is on extending the (open loop) experiment design framework to processes operating under feedback with a MPC-controller in the loop.

[1] C. A. Larsson and M. J. E. Annergren and H. Hjalmarsson. “On Optimal Input Design for Model Predictive Control”, Proceedings 49th IEEE Conference on Decision and Control, Orlando, FA, USA, December 2011.

[2] C. A. Larsson. “Toward Applications Oriented Optimal Input Design With Focus on Model Predictive Control. Licentiate Thesis in Automatic Control, KTH, September, 2011.

[3] C. A. Larsson, C.R. Rojas and H. Hjalmarsson. “MPC oriented experiment design”, 18th IFAC World Congress, Milano, Italy, 9966-9971, 2011.


[5] X.J.A. Bombois. “Design of optimal identification experiments for control’. Plenary address. JIME’2011, Douai, France, April 6-7, 2011.


WP3 Poster

Lead Partner: Kungliga Tekniska Hoegskolan

WP4 - Performance monitoring and diagnosis

The objective of this WP is twofold

  1. the development of a performance monitoring algorithm that will trigger the detection algorithm when and only when the performance drop is significant enough and
  2. the development of a detection algorithm that is able, via the least costly experiment on the plant, to detect the cause of the observed performance drop.

In a first stage, a literature survey on performance monitoring and performance diagnosis has been performed (see [1]). Subsequently, based on the insights of this literature survey, a simple and practical economic performance measure, the main element of the performance algorithm, has been devised (see [2]). The theoretical basis of the least costly detection algorithm has also been laid in a series of three conference papers [3,4,5]. In [5], in particular, the detection algorithm is linked with a possible subsequent step (i.e. the re-identification of the model in WP3). The objective is to minimize the overall cost of the detection and re-identification steps in the case of a performance drop caused by a change in the plant dynamics while keeping the detection cost below a certain threshold in the other cases.

Until now, the simplifying assumption of a linear time-invariant controller has been used. In the future, this assumption will be relaxed to also consider the case of MPC controllers using the results developed in WP3. Testing  the developed algorithms on the two benchmarks is now also under way.


[1] A. Mesbah and X. Bombois. Performance Monitoring and Diagnosis. Deliverable 4.1 of the AUTOPROFIT Project, October 2011.

[2] P.E. Moden, A. Mesbah and X. Bombois. Fundamentals on least costly detection and performance monitoring, Deliverable 4.2 of the AUTOPROFIT Project, April 2012.

[3] Ali Mesbah, X. Bombois, J. Ludlage and P. Van den Hof, ``Closed-loop performance diagnosis using prediction error identification'', pp 2969-2974, Proceedings of the 50th IEEE Conference on Decision and Control, Orlando, 2011

[4] A. Mesbah, X. Bombois, J. Ludlage, P. Van den Hof, ``Experiment Design for Closed-loop Performance Diagnosis'', pp. 1330-1335, Proceedings of the 16th IFAC Symposium on System Identification, Brussels, 2012


WP4 Poster

Lead Partner: Technische Universiteit Delft

WP5 - Tuning of model-based systems

The performance of any controller degrades over time if they are not supervised. Model based controllers and operation support systems are not an exception [1]. The successful implementation  of model based controllers depends on the quality/accuracy of model and  its calibration during their lifetime but also the tuning parameters . With the tuning of model based controllers, we mean the selection of weighting matrices in the performance index. In this work-package, we aim to find the optimal tuning parameters to achieve a certain closed loop behavior in case of plant-model mismatch. To this end, we study the impact of uncertainty on the closed loop behavior in frequency domain and use this information for selecting weighting matrices in the performance index. The impact of uncertainty  on the closed behavior  in terms of closed loop system behavior is shown below.

[1] M. Bauer , I. K. Craig, Economic assessment of advanced process control-A survey and Framework, Journal of Process Control 18(1), 2-18, 2008


WP5 Poster

Lead Partner: Technische Universiteit Eindhoven

WP6 - Benchmarking and industrial application

The goal of this work package is to validate and evaluate the developed tools under realistic conditions. The technologies will be tested on a few industrially relevant and internationally recognized benchmark model cases. These represent a distillation column and a pulp digester and will be made available publically on the web site for dissemination of the results. As last step, industrial validation will be performed for a depropanizer unit with Sasol Synfuels and a zinc ore flotation unit with Boliden Mineral.

Achievements up to now include an environment to use for the benchmarking, and the distillation column and pulp digester models to be used plus preliminary scenarios to be applied to them. Visits have been made to the two industrial sites, and initial tests have been performed on the depropanizer unit.


WP6 Poster

Lead Partner: ABB AB