New Fuzzy Neural Network With Fast Learning Algorithm And Guaranteed Stability For Manufacturing Process Control.pdf

Fuzzy Sets and Systems 132 (2002) 201–216
ate/fss
A new fuzzy workwith fast learning algorithm
and guaranteed stability for manufacturing
process control
Yunfei Zhou, Shuijin Li ∗, Rencheng Jin
National Engineering Center of Numerical Control System, College of Mechanical Science & Engineering,
Huazhong University of Science & Technology, Wuhan,
Hubei 430074, China
Received 30 June 1999; received in revised form 23 May 2001; accepted 15 October 2001
Abstract
In this paper, a new fuzzy work(FNN) is presented for manufacturing process control. It is di3erent from the
conventional FNN in its structure, learning algorithm and stability analysis method. Firstly, it utilizes the input and output
layer to on-line ÿne-tune scaling factors. It can also use the hidden layers to realize the fuzziÿcation, fuzzy inference,
defuzziÿcation and tune parameters such as membership functions, fuzzy control rules dynamically. Secondly, a new
combining learning algorithm (CL) bines the gradient-based error back-propagation algorithm (EBP) with similar
Newton (SN) algorithm is proposed in order to improve the convergence speed and putational burden during the
learning process. Lastly, a convergence condition for determining the stability of FNN is established. Physical experiments
for manufacturing process control are implemented to evaluate the e3ectiveness of the proposed scheme. c 2001 Elsevier
Science . All rights reserved.
Keywords: Neuro-fuzzy systems; Learning; Lyapunov’s direct method; Stability analysis
1. Introduction Many structures for FNN have been proposed. In
[31], a ÿve-layer FNN for learning rules of fuzzy
Recently, fuzzy workcontrol systems logic control systems was proposed and a two-phase
have been extensively studied [2,7,22,24,29] and suc- learning procedure was developed to delete redun-
cessfully used in manufacturing process control, such dant rules for obtaining a concise fuzzy rule base. In
as tool wear monitoring [21,3