19-05-2012, 05:18 PM
The Architecture of Fuzzy PID Gain Conditioner and Its FPGA Prototype Implementation
[attachment=22506]
Introduction
ASIC techques have been widely used in the
fields of signal processing, image processmg, d i c i a l
neural network and commurucahon In a sense, the
control algonthms are also a special lund of signal
processmg and can be implemented by ASIC Such
ASIC mplementahon of control algonthms can offer
hgh performance, good reliability and low cost for the
control systems. In the meantme, buildmg control
system on a chip will lead to a great leap foward for
control methodology and application, In dus paper, the
ASIC techniques for the implementaion of a fuzzy PID
controller is practised by us
Fuzzification
First, e(k) and Ae(k) are normalized into the
range [-128, 1281 for convenieace Then we transform
the absolute value of e(k), Ae(k) to their loganthms So,
the fuzzy rule is also applicable when the closed-loop
error or its change rate is small In ths paper, we fuzzlfy
e(k),Ae(k) into two fuzzy variables E and R, both of
wluch include the following fuzzy sets { NB, NM, NS,
20, PS, PM, PBJ where N represents negative, P
positwe, 20 zero, B big, M me&um,S small
VLSI Architecture
A new kmd of VLSI architecture as sh
3 is proposed to realize the FPGC. The core of this
archit&ture consists of four parts: fuzzification, fuzzy
inference (including the storage of fuzzy rules),
defuzzfication, tuning of PID controllers
Fuzzification
In fuzzfication, we input digihzed e(k),Ae(k) (8
bits ) to produce the membershp funchon of the fuzzy
sets In ths stage, there are 7 resultmg fuzzy sets { NB,
NM, NS, ZO, PS, PM, PB} So we code the fuzzy set
mto 3-bits Fuzzy Set Code In hardware function
diwsion, we must make a compromise between the
precision of the algonthm and the system scale We
modlfy the algonthm the contmuous membershp
function is &gibzed to only four values 0,113,213 1 The
simulation result confirm that thls digitization does not
affect the control performance
Fuzzy Inference
In fuzzy mference, we adopt the Rule-Dnven
form whose core is to obtain the conclusion and the truth
value for each fuzzy rule We can use ROM to store the
fuzzy rule The mput of ROM is the Fuzzy Rule Code
and the output is the conclusion The truth value of the
fuzzy rule is obtained as the production of the
membershp of the antecedents (e(k) or Ae(k))
[attachment=22506]
Introduction
ASIC techques have been widely used in the
fields of signal processing, image processmg, d i c i a l
neural network and commurucahon In a sense, the
control algonthms are also a special lund of signal
processmg and can be implemented by ASIC Such
ASIC mplementahon of control algonthms can offer
hgh performance, good reliability and low cost for the
control systems. In the meantme, buildmg control
system on a chip will lead to a great leap foward for
control methodology and application, In dus paper, the
ASIC techniques for the implementaion of a fuzzy PID
controller is practised by us
Fuzzification
First, e(k) and Ae(k) are normalized into the
range [-128, 1281 for convenieace Then we transform
the absolute value of e(k), Ae(k) to their loganthms So,
the fuzzy rule is also applicable when the closed-loop
error or its change rate is small In ths paper, we fuzzlfy
e(k),Ae(k) into two fuzzy variables E and R, both of
wluch include the following fuzzy sets { NB, NM, NS,
20, PS, PM, PBJ where N represents negative, P
positwe, 20 zero, B big, M me&um,S small
VLSI Architecture
A new kmd of VLSI architecture as sh
3 is proposed to realize the FPGC. The core of this
archit&ture consists of four parts: fuzzification, fuzzy
inference (including the storage of fuzzy rules),
defuzzfication, tuning of PID controllers
Fuzzification
In fuzzfication, we input digihzed e(k),Ae(k) (8
bits ) to produce the membershp funchon of the fuzzy
sets In ths stage, there are 7 resultmg fuzzy sets { NB,
NM, NS, ZO, PS, PM, PB} So we code the fuzzy set
mto 3-bits Fuzzy Set Code In hardware function
diwsion, we must make a compromise between the
precision of the algonthm and the system scale We
modlfy the algonthm the contmuous membershp
function is &gibzed to only four values 0,113,213 1 The
simulation result confirm that thls digitization does not
affect the control performance
Fuzzy Inference
In fuzzy mference, we adopt the Rule-Dnven
form whose core is to obtain the conclusion and the truth
value for each fuzzy rule We can use ROM to store the
fuzzy rule The mput of ROM is the Fuzzy Rule Code
and the output is the conclusion The truth value of the
fuzzy rule is obtained as the production of the
membershp of the antecedents (e(k) or Ae(k))