Control Pi-Pid

Journal of Process Control 20 (2010) 452–463

Contents lists available at ScienceDirect

Journal of Process Control
journal homepage: www.elsevier.com/locate/jprocont

PI/PID autotuning with contextual model parametrisation
Alberto Leva *, Sara Negro, Alessandro Vittorio Papadopoulos 1
Dipartimento di Elettronica e Informazione, Politecnico di Milano, Via Ponzio 34/5, 20133 Milano, Italya r t i c l e

i n f o

a b s t r a c t
In model-based PI/PID tuning regulators, the same set of I/O data and the same tuning rule can produce very different results, depending just on the procedure used to parametrise the process model. The problem is seldom addressed, but extremely relevant for the acceptability of model-based autotuners in the applications. This manuscript proposes amethodology to treat the model parametrisation and regulator tuning phases jointly, so as to circumvent said problem with affordable process stimulation and computational effort. The methodology can be generalised to different regulator structures, and even employed to devise new tuning rules. Ó 2010 Elsevier Ltd. All rights reserved.

Article history: Received 31 March 2009 Received in revisedform 14 January 2010 Accepted 25 January 2010

Keywords: Autotuning Model-based tuning PID control

1. Introduction and motivation In Model-Based AutoTuning (MBAT for short) of industrial regulators, some process input/output measurements are first used to obtain a ‘‘model” of the process, which is subsequently employed – together with convenient specifications – to compute the regulatorparameters with some ‘‘tuning rules”. A review of the huge MBAT literature is impossible to give here, the interested reader can refer, e.g., to [10,17,7,4], and for a broad panorama to the excellent survey [30]. In general, MBAT methods are based on very simple models, of structure decided a priori based essentially on that of the regulator, despite the process dynamics encountered may be complex [1],and require a ‘‘clever” order reduction [34]. Two are the main reasons for the fact above. First, models need identifying on-line, based on data produced by stimuli that must obey to potentially severe process upset constraints and therefore typically lack excitation, which hampers, e.g., model order selection. Second, to achieve tuning procedures suitable for an industrial implementation, explicittuning rules are desirable [13,31], which call for a simple model unconditionally. Such a scenario motivates the widespread use in MBAT of ad hoc parametrisation methods such as that of areas, of moments, of the tangent, and so forth. As a consequence of the heuristics unavoidably introduced, the results obtained with MBAT depend on the used parametrisation method significantly. The problem isseldom addressed in the literature, but hampers a wide acceptance of MBAT in the application domain. In fact, if the same set of data,

* Corresponding author. Tel.: +39 02 2399 3410; fax: +39 02 2399 3412. E-mail address: leva@elet.polimi.it (A. Leva). 1 Graduate students at the Dipartimento di Elettronica e Informazione. 0959-1524/$ - see front matter Ó 2010 Elsevier Ltd. All rights reserved.doi:10.1016/j.jprocont.2010.01.005

the same model structure and the same tuning rules produce different tuning results depending just on which method is selected to find the model parameters, it is not surprising that the industrial community’s confidence in MBAT at large is adversely affected, as easily observed in technology reviews such as [28]. Indeed, in MBAT there is hardly any point indiscussing tuning rules without taking into account the effects of the model parametrisation method [23,27]. This is however complex: in the light of the above remarks, on one hand well established results of the identification theory often prove inadequate to assess the quality of a model for the purpose of MBAT. On the other hand, process stimuli limitations leave limited or no room for experiment...