Automatic tuning of industrial controls

Motivation and rationale	The importance of autotuning is well known. For some years, I had this as my main topic. At present, my research is continuing some previous paths, like the tuning of control structures, but aims primarily at exploiting the gathered experiences in domains like computing systems, where autotuning can be crucial for controller parametrisation at startup time, and also to ease maintenance on the part of the typical administrator. who generally is not - and need not be - directly involved in control.
Methodological achievements	Along the years, I have proposed numerous tuning methods for PI/PID controllers. I have also addressed other control laws, including for example filters to shape the high-frequency control sensitivity, and control structures, addressing for example the tuning of a cascade control with a single experiment. His research also concentrates on the synthesis of controllers expressed in industrial form~\cite{bib:IJ-2011-JPC}, and on how to tailor and improve tuning rules on the basis of structural process information as gained by the experiments allowed in industrial contexts. Moreover, I studied the problem of "robust" autotuning, mostly by means of nonparametric model error descriptions obtained by identification data. Another subject is the definition of "tuning quality indices", to allow for the on-line choice of the best rule to use in a given situation, and more in general to provide some basis for a taxonomy of the vast corpus of available and new rules. Studies were also conducted on the automatic detection of the necessity for a controller re-tuning. I also proposed techniques to exploit experimental data that carry local-in-frequency information, like a relay test does, to simultaneously tune the controller, and determine a process model that is inherently precise near the actually obtained cutoff frequency; this led to the so-called "contextual" autotuning method. Also, the problem of realising autotuning controllers on extremely low-end devices, with very strict computational and memory limitations, was addressed.
Technological achievements	Many of the proposed methods were implemented and actually used in the industry. A notable example, referring to the adaptation of a filter included in an on-off control for refrigerators, was patented in 2009. Also, autotuning controllers were included in many model libraries, and frequently employed - in the context of industrial collaborations - to parametrise the involved controllers prior to deploying the results to the field. Computing systems applications also start appearing, and are being actively investigated.
Open issues and work in progress	On the methodological side, the problem of "robust" autotuning still stands open, as in fact no experiment of industrially accepted duration can yield enough information to qualify the class of systems in which the controlled one is expected to vary. As suggested above the chosen solution was to reverse the problem and quantify the tolerable model a priori: in this respect, as on the connected one concerning tuning quality indices, research is still ongoing. On the technological front, the main issues to tackle at present are those posed by realisations in the computing system context, basically owing to the nature itself of those systems, that were not conceived to host controllers, and often are not keen to allow "control" software to take the necessary authority. Also, the research on automatic tuning has strong relationship with that on event-based control, since such controller realisations call for ad hoc tuning procedures, and with that on computing system, as there in many situation autotuning capabilities prove useful.
Outlook and vision	In the industry, many control engineering and design environments offer autotuning capabilities, but the use of these is by common opinion far less widespread than it could. At the same time, recent studies set an increasing emphasis on control and its optimisation as an added value driver - also in the information technology domain, as already said. Plans are to tackle this dichotomy, so as to foster an increasing adoption of autotuning and therefore a corresponding return in research; when convenient, this can be done by "piggybacking" autotuning functionalities, and the consequent technology transfer, to simulation and design studies.