Duarte Valério IDMEC/IST, Technical University of Lisbon, Avenida Rovisco Pais, 1049-001 Lisboa, Portugal e-mail: dvalerio@dem.ist.utl.pt Manuel Duarte Ortigueira UNINOVA and DEE of Faculdade de Ciências e Tecnologia of UNL, Campus da FCT da UNL, Quinta da Torre, 2825-114 Monte da Caparica, Portugal e-mail: mdortigueira@uninova.pt José Sá da Costa IDMEC/IST, Technical University of Lisbon, Avenida Rovisco Pais, 1049-001 Lisboa, Portugal e-mail: sadacosta@dem.ist.utl.pt Identifying a Transfer Function From a Frequency Response In this paper, the classic Levy identification method is reviewed and reformulated using a complex representation. This new formulation addresses the well known bias of the clas- sic method at low frequencies. The formulation is generic, coping with both integer order and fractional order transfer functions. A new algorithm based on a stacked matrix and its pseudoinverse is proposed to accommodate the data over a wide range of frequencies. Several simulation results are presented, together with a real system identification. This system is theArchimedes Wave Swing, a prototype of a device to convert the energy of sea waves into electricity. DOI: 10.1115/1.2833906 Introduction The identification of linear systems is an interesting subject that deserved a lot of attention in the past 1. Identification in the frequency domain is a particular case with great interest in appli- cations. Algorithms are traditionally based on Levy’s work 2,a least-squares based algorithm formulated in a real framework by separating the real and imaginary parts of the transfer function. This led to a formulation with lengthy expressions and to results not equally good at all frequencies 3. This frequency depen- dence has been faced 3by using an iterative method; another alternative, without iterations, was also proposed 4. Both adap- tations modify the basic algorithm introducing weights. Here, we reformulate the original approach in a completely complex framework, leading to a set of normal equations from which we could remove the frequency dependency. Expressions obtained are shorter. Thus, we obtain two sets of linear equations for each frequency. The formulation is given for the general case of a fractional, commensurate transfer function. The use of measurements corresponding to several frequencies has been addressed in the past by an average of the coefficients computed from each frequency. This procedure is neither robust nor sensible to order changes. To obtain a more reliable algorithm, we propose a new algorithm with two steps: firstly, collect the matrices corresponding to the different frequencies in a stacked matrix; secondly, compute the coefficients of the model by using the pseudoinverse. To illustrate the behavior of this new formula- tion, we present some simulation results and also a real practical example. Adapting Levy’s Identification Method for Fractional Orders Original Formulation. Let us suppose we have a plant de- scribed by a linear system with a transfer function G and a corre- sponding frequency response Gjand that we want to model it using a transfer function G ˆ s= b 0 + b 1 s q + b 2 s 2q + ¯ + b m s mq a 0 + a 1 s q + a 2 s 2q + ¯ + a n s nq = k=0 m b k s kq k=0 n a k s kq 1 where m and n are the preassigned orders of the numerator and denominator, and q is the fractional derivative order. Without loosing generality, we set a 0 = 1. The frequency response of 1is given by G ˆ j= k=0 m b k j kq 1+ k=1 n a k j kq = Nj Dj = + j + j 2 where N and D are complex valued and , , , and the real and imaginary parts thereofare real valued. From 2we see that = k=0 m b k Re j kq = k=0 n a k Re j kq =1+ k=1 n a k Re j kq 3 = k=0 m b k Im j kq = k=0 n a k Im j kq = k=1 n a k Im j kq The error between model and plant, for a given frequency , will be j= Gj- Nj Dj 4 Minimizing the error power would be an obvious but difficult way of adjusting the parameters of 1. Instead of this, Levy’s method minimizes the square of the norm of Ej= def jDj= GjDj- Nj5 This leads to a set of normal equations, easy to solve. Omitting the frequency argument to simplify the notation, we have Contributed by the Design Engineering Division of ASME for publication in the JOURNAL OF COMPUTATIONAL AND NONLINEAR DYNAMICS. Manuscript received June 6, 2007; final manuscript received November 30, 2007; published online February 4, 2008. Review conducted by J. A. Tenreiro Machado. Paper presented at the ASME 2007 Design Engineering Technical Conferences and Computers and Information in Engineering Conference DETC2007, Las Vegas, NV, September 4–7, 2007. Journal of Computational and Nonlinear Dynamics APRIL 2008, Vol. 3 / 021207-1 Copyright © 2008 by ASME Downloaded 05 Feb 2008 to 193.136.128.14. Redistribution subject to ASME license or copyright; see http://www.asme.org/terms/Terms_Use.cfm