A systematic approach for on‐line identification of second‐order process model from relay feedback test

Abstract

in Wiley InterScience (www.interscience.wiley.com).Low-order process modeling provides a basis for control system design and on-line autotuning in process control. A systematic on-line identification method is proposed in this article to obtain a second-order-plus-dead-time (SOPDT) model from a single biased/unbiased relay feedback test. The relay response shapes of the three types of SOPDT model, overdamped, critically damped, and underdamped, are first examined and categorized for model structure selection before proceeding with the corresponding parameter identification. Exact expressions of the corresponding limit cycles are derived for assessing process response under relay feedback test. By using the measurable parameters of these limit cycles, the corresponding identification algorithms are subsequently derived in a transparent manner. Two denoising methods are given to cope with measurement noises. Illustrative examples are performed to demonstrate the effectiveness and merits of the proposed identification algorithms. 2008 American Institute of Chemical Engineers AIChE J, 54: 1560-1578, 2008 Keywords: identification, relay feedback test, biased, second-order-plus-dead-time (SOPDT) model, limit cycle, denoising IntroductionRelay feedback identification has attracted increasing attention in the process community, since the pioneering work of Å ström and Hägglund 1 which used relay feedback to generate sustained oscillations of the controlled variable for closed-loop identification.Compared with an open-loop identification such as using a step or impulse excitation signal, the process dynamic response is more effectively excited and it can be better observed with sustained oscillation resulted from a relay feedback test. Furthermore, relay feedback test will not cause the process to drift too far away from its operation level. This is important for many practical cases, in particular for highly nonlinear processes with rigorous operating conditions. Based on the concepts of ultimate gain and ultimate frequency obtained from such experiments, the early Correspondence concerning this article should be addressed to F. Gao at kefgao@ust.hk. American Institute of Chemical EngineersAIChE Journal June 2008 Vol. 54, No. 6 1560 relay feedback method, autotune variation (ATV), 2 has become a popular identification tool in process industry. Reviews on the rapid development of relay feedback identification for process regulation have recently been addressed by Atherton 3 and Hang et al. 4 Much research effort has been presently devoted to using a single relay feedback test for process identification. There are in general two kinds of relay feedback structure, unbiased (symmetrical) and biased (asymmetrical). On the basis of a single run of unbiased relay feedback test, Vivek and Chidambaram 5 proposed a first-order-plus-dead-time (FOPDT) modeling according to the Fourier analysis of the process response; Panda 6 reported an identification algorithm for obtaining an underdamped second-order-plus-dead-time (SOPDT) mode...

On-Line Controller Tuning for Critically Damped Soptd Systems

“…Lee et al (2007Lee et al ( , 2010 used the integral of relay responses to estimate the model. Liu et al (2008) proposed a method based on the relay tuning method to calculate the model parameters of the SOPTD system. Liu and Gao (2009) presented a method for estimating the model parameters using the relay tuning method based on either biased or unbiased relay tests.…”

Relay feedback‐based time domain modelling of off‐diagonal elements of linear 2‐by‐2 MIMO systems

“…The process input in the relay feedback test consists of a series of step changes with down amplitude, µ − and up amplitude, µ + . The procedure for obtaining the mathematical model of SISO systems using biased relay is explained (Gu et al, 2006; Liu et al, 2008) in detail. At the first interval (after synchronising input with output by time shift) the response can be described as: …”

Robust Step-Like Identification of Low-Order Process Model Under Nonzero Initial Conditions and Disturbance

“…By triply integrating both sides of (4) and rearranging the resulting equation using (13), (15), (16) and 17, we obtain the LS form of parameter estimation (t) = T (t) + "(t) (18) where "(t) denotes the residual error, and (19)-(23), shown at the bottom of the next page.…”