Formalism to design a neural network: Application to an induction machine drive coupled to a non linear mechanical load

¹ Laboratoire d'Électrotechnique et d'Électronique de Puissance de Lille (L2EP), École Centrale de Lille, BP 48, 59651 Villeneuve d'Ascq Cedex, France
² Laboratoire d'Électrotechnique et d'Électronique de Puissance de Lille (L2EP), EUDIL, avenue Paul Langevin, 59655 Villeneuve d'Ascq, France
³ Laboratoire d'Électrotechnique et d'Électronique de Puissance de Lille (L2EP), ENSAM, bd. Louis XIV, 59046 Lille Cedex, France

Corresponding author: betty.semail@eudil.fr

Received: 6 March 2000
Revised: 7 September 2000
Accepted: 26 September 2000
Published online: 15 November 2000

Abstract

This search deals with the control of a process in order to take into account non linearities without parameters identification. Neural networks properties are exploited for the modelling of non linear features, and a formalism is proposed to design a neural model which can be used directly as a controller. We apply this formalism to the modelling of a non linear mechanical load torque feature coupled to an induction machine in order to design a speed controller. A partial and a global neural method are presented. In order to overcome modelling errors or any process changes, an adaptive on line method is proposed. At last, simulation and experimental results are presented.