829 Copyright@ 1996IFAC 13th Triennial World Congre:"i ... , San Francisco. USA KNOWLEDGE, CONTROL, AND REALITY: THE NEED FOR A PLURALISTIC VIEW IN CONTROL SYSTEMS DESIGN Manuel Liz • Margarita Vazqucz· Javicr Aracil·· • Facultad de Filosofia, Dept. de Logica 11 Filosofia de la Ciencia , Unive rsidad de La Laguna, 38fOl Canary Islands, Spain •• Es cuela Superior de ingenieros, Universidad de Se -villa, Avda. Mercedes Sill, 4101£ Sevula, Spain. E-mail: aracil@esi.us.es Abstract. Modelling is a key issue in Control Systems Design. The model embodies the knowledge about the process to be controlled and is interiorized in some way by the controller I so the actions it takes are based on the model. However I th e model never is unique. Different conceptual schemes and different languages produce alternate models. Current control engineering illustrates the fact that inside scientifical and technological disciplines we are very often lead to adopting an unavoida.ble pluralistic perspective. Wit.h t.he help of new developments in recent epistemology and philosophy of science, some ("_onceplu al proposals will be made in order to analyze that pluralism. Keywords. epistemological aspects of modelling, in ternal realism, alte rnate descriptions, conceptual schemes, fuzzy modelling 1. INTRODUCTION language, a system is described as dx -= j(x) dt 6b-OI 4 (I) Most of the designs of control systems are based on ex- plicit models of th e controlled process. Even when the design method does Ilot include t.he model explicitely, this is implicitly used by the controller. Th e model SUlll- marizes the knowledge available about the aspect of rcal- ity to be controlled. The control system n ee ds some sort of representation of the process which it il'i acting upon (as every agent needs a representation of th e system she is trying to govern) . This point has been largely empha- sized by the control systems literature, at least since the classical paper by Conant and Asbhy (1970). wh ere a dynamical system (X,I) is the mathematical object formed by a state space X (a manifold) and a vectorial field j, defined 011 X, X = JR n , j E COO(x, x), and T = IR. This is a continoU9 time system description. In control systems th ere is a control signal u and the formalism takes the form The question is that th e system designer has many "lan- guages" to represent the system to be controlled. The most classical one is that based on dynamical systems (differential equation,). With the tool, supplied by this d", -= !(x , u) dt y=g(",1 (2) where f is the state-transition function aDd 9 i. the lec- ture function .