A Novel Design of Iterative Learning Control with Pure Feedforward
Structure
CHI Ronghu
1
, LIN Na
1
, HOU Zhongsheng
2
, ZHANG Ruikun
3
1.
School of Automation & Electronic Engineering, Qingdao University of Science & Technology
Qingdao 266042 (
E-mail: ronghu_chi@hotmail.com)
2. Advanced Control Systems Lab, School of Electronics & Information Engineering, Beijing Jiaotong University
Beijing 100044 (E-mail: zhshhou@bjtu.edu.cn)
3.
School of Mathematics and Physics, Qingdao University of Science & Technology
Qingdao 266042 (e-mail: rkzhang@qust.edu.cn)
Abstract: A new design framework of feedforward ILC is developed for nonlinear systems in this work. By supposing that there
exists a desired nonlinear controller, two dynamical linearization methods are proposed for the nonlinear controller. And then, a
CFDL-ILC and a PFDL-ILC are presented. Comparatively, the PFDL-ILC is similar to the higher order ILC schemes that more
errors of previous iterations are used. The proposed approaches are data-driven and no process model is required for the
controller design and analysis. The availability of the proposed approaches is further confirmed by simulation results.
Key Words: Iterative learning control, Nonlinear systems, Data-driven control
1 Introduction
Iterative learning control (ILC) [1-3] is most suitable for
repetitive dynamics operating over a fixed time interval. In
genral, there are three major structures of ILC system: Pure
feedforward (PFF) ILC, Pure feedback (PFB) ILC, and
feedback – feedforward (FBFF) ILC.
The original proposed ILC is a pure feedforward
algorithm, where the controlled plant is supposed stable
over the finite time interval and no system dynamics of the
current iteration is considered. The idea of this structure is
very straightforward, i.e., learning from experience. It is
guaranteed that all the tracking errors over the entire finite
time interval are converging to zero as the learning number
approaches to infinity.
The pure feedforward PID-type ILC has some advantages
such as simple structure, easy to implement, no on-line
computing burden, and so on. However, there are also some
limitations hindering its further applications in practice.
First, although there is a convergence theorem pointing
out a scope about the proper learning gains of the PFF
PID-ILC, the learning gains are difficult to be selected if
there is no enough knowledge known exactly about the
controlled system. Second, the learning gains are generally
fixed and invariant in the controller operation when they are
selected properly. If there is a large disturbance or
uncertainty exposed on the controlled plant, the PFF
PID-ILC with the previous fixed learning gain may become
unsatisfied. Third, the convergence of PFF PID-ILC is
guaranteed under the framework of
norm and requires
*
This work was supported by National Science Foundation of China
(61374102, 61503209, and 61433002), and the High School Science &
Technology Fund Planning Project of Shandong Province of China
(J14LN30).
is large enough, which leads to poor tracking
performance/even divergence may occur in some time.
In fact, there always exists a generic ideal controller
producing the desired control input such that the system
output tracks the reference trajectory exactly. The controller
is unknown and may be nonlinear or linear. The key issue is
how to find such an ideal controller. Recently, model-free
adaptive control (MFAC) [4-6] is proposed for nonlinear
systems. And the key innovation of MFAC is its dynamical
linearization. In [7], the dynamical linearization is also
applied to linearize an ideal nonlinear controller. And thus a
controller dynamic linearization based MFAC scheme is
proposed.
This paper aims to present a general design framework of
pure feedforward ILC, as well as its higher order algorithm,
for a class of nonlinear system directly. Two dynamical
linearization methods for the nonlinear systems and the
general controller are proposed. The unknown learning
gains in the controller are varying with both time and
iterations, and can be updated by the designed estimation
algorithms. A higher order algorithm is also proposed by a
different dynamical linearization of the desired nonlinear
controller. Simulation study shows the efficiency and
application of the proposed method.
The rest of this paper is organized as follows. Section 2
presents problem formulation. In Section 3, a new CFDL
based ILC controller is designed. Section 4 extends the
result to a PFDL based ILC controller. Section 5 provides a
simulation study. Finally, some conclusions are given in
Section 6.
2 Problem Formulation
Consider a nonlinear discrete-time system,
)
)(,),1(),(
),(,),1(),()1(
ukkk
ykkkk
ntututu
ntytytyfty
−−
−−=+
"
"
(1)
5761
978-1-4673-9714-8/16/$31.00
c
2016 IEEE