| Motivation and rationale |
In
many domains - e.g. process control to mention a major one - modelling,
simulation and control are becoming more and more integrated with one
another. In particular, as projects become larger and requirements
stricter, models gain importance as decision aids at virtually any
stage of the overall design, which apparently involves jointly both
control and controlled system. In this context, I am primarily interested in methods and techniques to facilitate the reuse of models throughout the phases of a project, ideally from high-level engineering, through component selection/design, control strategy definition and system assessment, down to control code verification. The research is centred on the object-oriented modelling paradigm, as this seems an adequate basis to build the required methods/techniques upon. Moreover, I am interested in the efficient integration of object-oriented models in control environments. This research has also significant relationships with that on the automatic tuning of industrial controls. |
| Methodological achievements |
A
relevant obstacle for a widespread use of modelling and simulation in
the sense above, is the necessity of tailoring the model complexity to
the particular study at hand. Techniques were proposed for model
manipulation and solution algorithm simplification, based on ideas like
that of dynamic decoupling, and in a view to exploiting the so gained
simulation efficiency for on-line model predictive control. Another issue is that both plant-wide optimised design and software synthesis/verification require a representation of control systems. However, plant-wide optimisation needs continuous-time and simulation-efficient representation of controllers, for example to exploit variable-step solvers and ease the application of simulation-based optimisers. On the contrary, assessing control software in simulation calls for a detailed digital representation of the same controllers, sometimes down to code replica, apparently to the detriment of simulation efficiency. The problem of maintaining consistency between the two required representations of the same controller - while at the same time having one as efficient and the other as detailed as possible - was studied, and some solutions proposed. A third subject is the representation of three-dimensional phenomena when a very detailed replica, of the spatial distribution of variables may not be the main concern, while global quantities need capturing. A typical example is a large room, where the properties of air cannot be taken as uniform, and one is not interested in the fine-grained temperature field but needs to catch the heating energy consumption. A "quasi-3D" modelling approach was devised, based on a purpose-specific simplification of the Navier-Stokes equations, that in low fluid velocity cases, proved well suited for the purpose just mentioned. Finally, somehow lateral yet relevant to the process domain if heavy-duty equipment is considered, interesting results were obtained as for object-oriented modelling of mechanical elements with distributed flexibility. |
| Technological achievements |
The
research initially concerned mainly power plants, but over the years
extended to various industrial applications, to energy-efficient
household appliances, and to energy optimisation in buildings and
plants. Some model libraries were created, and numerous case studies
addressed (several in the context of research projects and industrial
collaborations). One of these libraries, built along the ideas above on
the representation of controllers, won a prize at a Modelica conference. More recently, efficient simulation techniques were also applied to the assessment of control solutions for computing systems, where sometimes an accurate physical simulation and a conveniently abstracted representation of software-induced phenomena have to be joined. |
| Open issues and work in progress |
At present, the focus for this research is mainly set on applications.
In particular, work is under way to address systems of large dimensions
but with a high degree of sparsity, like for example national-level
grids, and to integrate the developed model libraries into optimisation
tools, which quite often opens issues concerning the regularity of the
involved models. |
| Outlook and vision |
In the near future, a significant effort will be required to maintain an orderly methodological view on so rapidly evolving a scenario,
while always keeping an eye on applications. To both these ends, plans
are to establish and strengthen relationships with the involved and
potentially interested communities. In the power domain such
relationships exist since many years. Concerning other contexts, for
example, I have recently started a cooperation, within a national
project, with an architecture department. |