Approximation of the Preisach function using feedforward neural network

Issue title: ISTET '01

Article type: Research Article

Authors: Dotterweich, Martin^a | Szymański, Grzegorz^{b; *} | Waszak, Michał^b

Affiliations: [a] PROFI Engineering Systems Inc., Darmstadt, Germany | [b] Poznań University of Technology, Poznań, Poland

Correspondence: [*] Corresponding author: G. Szymański, Institute of Control and System Engineering, Poznań University of Technology, ul. Piotrowo 3A, 60-965 Poznań, Poland. Tel.: +48 61 6652 504; E-mail: [email protected]

Abstract: The nonlinear computations based on the Newton-Raphson algorithm may suffer poor convergence provided that the measured data are employed directly. The aim of this work is to present the method to improve convergence rate of magnetic field simulations taking hysteresis into account. The Preisach model is considered. It is proposed to use the neural network approximator to obtain smooth hysteresis model output. The mathematical representation of the neural network is exploited to obtain analytical expression for derivative of the hysteresis model output. Moreover, taking advantage of the analytical form of the Preisach function, it is shown that speed up in calculations of hysteresis energy losses can be achieved.

DOI: 10.3233/JAE-2003-246

Journal: International Journal of Applied Electromagnetics and Mechanics, vol. 17, no. 1-3, pp. 243-250, 2003