Global sensitivity analysis of a feedback-controlled stochastic process model

https://doi.org/10.1016/j.simpat.2013.04.003Get rights and content

Abstract

The purpose of the study was to investigate whether global sensitivity analysis can be utilised in concurrent process and control design to gain insight into the process and its control. This paper addresses the issue of sensitivity of a control law performance to its parameters in a dynamic, hybrid deterministic–stochastic process model. The control law under investigation is a collection of single-input, single-output type tower level controllers in a papermaking process. Global sensitivity analysis is shown to attribute higher importance to certain key parameters, thus providing valuable insight for the designer.

Introduction

This paper examines global sensitivity analysis (GSA) of a mill-wide control system. The purpose of this work was to investigate the potential of GSA in guiding design work in concurrent process and automation design. The current work relates to a simulation and optimisation based design framework under development, see [10], [18], [8], [14]. Although concurrent design has been widely studied (see, e.g., [2], [11] and the literature therein), in practice process design typically begins with flowsheet and equipment design, with the control of the process left as a later consideration. In contrast, the framework under development strives to start the control design as early as possible. This means that dynamic simulation models are needed early on in the design, even though not all relevant information is yet available. Consequently, the models contain a significant amount of uncertainty, especially with regard to their control system parts. This also means, therefore, that any further understanding of the model’s behaviour is useful in the later phases of design, such as in process and control optimisation. The purpose of this paper is thus to investigate what insight can be gained by applying global sensitivity analysis during the early phases of process and automation design.

In this study we restrict the analysis to the control system of the model. This restriction is made since, as mentioned above, this part of the model is likely to contain the largest uncertainty. The aim of this paper is not to analyse one control loop at a time, as many methods and tools are available for that purpose (see, e.g., [15]). Rather, we take a systemic view, analysing several control loops and the process simultaneously. The studied case originates from the papermaking industry. The case control system provides system-wide control of several large storage towers. As such a control structure has an effect on several areas of the system, it follows that it should be designed together with the process structure as early on as possible.

The work presented here relies heavily on simulation models, some of which are stochastic, arising from parts of the studied process for which first principles models are nearly impossible to construct. These models are presented in Section 2. The decision to use partly stochastic models is also justified by the fact that early on in the design of processes and controls, the amount of available information is limited. Thus, the secondary research interest of this paper was to investigate how well global sensitivity could be applied to such models. In addition to simulation models, the present work utilises global sensitivity analysis (GSA), which is outlined in Section 3. Section 4 presents the results of the study, and the final chapter discusses the results.

Section snippets

Modelling

The modelled process is a paper mill. A schematic of such a mill is presented in Fig. 1.

The process uses two raw material streams, thermomechanical pulp (TMP) and chemical pulp (CP), which enter the mill from separate production systems. In the process leading up to the paper machine (PM), the raw materials are diluted, stored and mixed in large towers (3000 m3) and smaller tanks or chests (400 m3). At the PM water is removed from the pulp stock and solids are retained, raising the dry solids

Global sensitivity analysis and simulation setups

Sensitivity analysis (SA) is a method for determining how a selected set of inputs or parameters affects another set of variables called outputs [12]. Typically, one may be interested in which parameters affect the outputs most or which have a negligible effect. A traditional way of producing sensitivity information is by evaluating partial derivatives either analytically or numerically, but this approach has the potential pitfall that it is local [12], [19]. In other words, the derivatives are

Results

For the initial analysis, the model outputs are presented in the figure below, which shows the output variables Yk (k = 1,  , 4) as functions of the SA sample number.

In the top-left corner of Fig. 2 is the time to first over- or underflow variable, Y1. In the top-right corner is the output variable Y2 (replication- and time-averaged squared strength deviation) as a function of the sample number. In the bottom-left corner is Y3 (replication- and time-averaged squared filler content deviation), and

Discussion and conclusions

The aim of this paper was to investigate the potential of global sensitivity analysis in the broader scope of an early-phase concurrent process and automation design framework and to present an example of how to use GSA in a dynamic, stochastic process modelling case to gain insight into the process and automation under design. In this study, the parameters were analysed using the variance-based Sobol’ method. One parameter, c10, rose clearly above the rest in importance. From a practical point

Acknowledgements

This work has been supported by the EffTech and EffNet programmes of FIBIC Ltd., and by VTT. The author would also like to thank Professor Risto Ritala of Tampere University of Technology for his support with the modelling, and Lic.Sc. (Tech.) Jari Lappalainen of VTT for his support with the control law definition. Lic.Sc. (Tech.) Lic.Tech. Petteri Kangas is gratefully acknowledged for his support in preparing the manuscript. Finally, the developers of Simlab are acknowledged.

References (19)

There are more references available in the full text version of this article.

Cited by (4)

  • Greenhouse gas emissions during timber and concrete building construction —A scenario based comparative case study

    2018, Sustainable Cities and Society
    Citation Excerpt :

    It is also frequently used to estimate upstream impacts of a product or process as often these information are difficult to obtain. On the contrary, process is a bottom-up approach that uses process specific information to evaluate emissions and environmental impacts of a product or process (Li et al., 2001; Savolainen, 2013). However, process based approach requires quality information to develop the assessment model in evaluating environmental impacts.

View full text