Upload
john-ball
View
213
Download
0
Embed Size (px)
Citation preview
REC 2006 - Georgia Tech University SREC 2006 - Georgia Tech University Savannah Feb 22-24avannah Feb 22-24
Reduction in Space Complexity And Error Detection/Correction of a Fuzzy controller
F. VainsteinGeorgia Institute of Technology
E. MartePontificia universidad Católica Madre y Maestra
V. OsoriaUniversidad Tecnológica de Santiago
R. RomeroInstituto Tecnologico de Santo [email protected]
REC 2006 - Georgia Tech University SREC 2006 - Georgia Tech University Savannah Feb 22-24avannah Feb 22-24
AgendaAgenda
IntroductionIntroductionFuzzification, Rule Base and DefuzzificationFuzzification, Rule Base and DefuzzificationAddress Generation Address Generation Variable Radix Numbers (Multi Radix) Variable Radix Numbers (Multi Radix) Reduction in space complexity Reduction in space complexity Data compression and error correcting codes in Rule Base Data compression and error correcting codes in Rule Base Conclusion Conclusion
REC 2006 - Georgia Tech University SREC 2006 - Georgia Tech University Savannah Feb 22-24avannah Feb 22-24
IntroductionIntroduction
Fuzzy controllers prove to be very useful for practical applications, Fuzzy controllers prove to be very useful for practical applications, especially in the cases when there is no appropriate mathematical especially in the cases when there is no appropriate mathematical model of behavior of the controlled object. Control signal is model of behavior of the controlled object. Control signal is computed by fuzzy controller with the use of rule base table. computed by fuzzy controller with the use of rule base table.
Fuzzy sets and some basic ideas pertaining to their theory were first Fuzzy sets and some basic ideas pertaining to their theory were first introduced in 1965 by Lofti A. Zadeh, a Professor of Electrical introduced in 1965 by Lofti A. Zadeh, a Professor of Electrical Engineering at the University of California a Berkeley. The Engineering at the University of California a Berkeley. The development of fuzzy set theory and fuzzy logic experimented development of fuzzy set theory and fuzzy logic experimented changes since their introduction. Therefore the “Fuzzy Boom” (since changes since their introduction. Therefore the “Fuzzy Boom” (since 1989) has been characterized by a rapid increase in successful 1989) has been characterized by a rapid increase in successful industrial applications that have netted impressive revenues. industrial applications that have netted impressive revenues.
REC 2006 - Georgia Tech University SREC 2006 - Georgia Tech University Savannah Feb 22-24avannah Feb 22-24
Braunstingl (1995) developed a wall-following robot that used a fuzzy logic Braunstingl (1995) developed a wall-following robot that used a fuzzy logic controller and local navigation strategy to determine its movement. The controller and local navigation strategy to determine its movement. The fuzzy logic controller uses the input variables to control the firing of 33 rules.fuzzy logic controller uses the input variables to control the firing of 33 rules.
A fuzzy system developed by Surmann (1995) controls the navigation of an A fuzzy system developed by Surmann (1995) controls the navigation of an autonomous mobile robot. The entire system has about 180 fuzzy rules that autonomous mobile robot. The entire system has about 180 fuzzy rules that associate 30 fuzzy inputs with 11 outputs. Potentially, the number of fuzzy associate 30 fuzzy inputs with 11 outputs. Potentially, the number of fuzzy rules can be very large.rules can be very large.
REC 2006 - Georgia Tech University SREC 2006 - Georgia Tech University Savannah Feb 22-24avannah Feb 22-24
The contribution of this paper is to tackle the space complexity of the The contribution of this paper is to tackle the space complexity of the system by decreasing the number of addresses lines used to store If-Then system by decreasing the number of addresses lines used to store If-Then rules. Also the incorporation of error correcting codes in the memory used to rules. Also the incorporation of error correcting codes in the memory used to store If-Then rules without substantial increasing space complexity as well store If-Then rules without substantial increasing space complexity as well as application of signatures analysis for error detection/location in a fuzzy as application of signatures analysis for error detection/location in a fuzzy controller. controller.
REC 2006 - Georgia Tech University SREC 2006 - Georgia Tech University Savannah Feb 22-24avannah Feb 22-24
Fuzzification, Rule Base and DefuzzificationFuzzification, Rule Base and Defuzzification
FuzzySensor 0 Attributes
Values of Membership Functions
0.00.60.30.0
Sensor k 0.0 Crisp Output
0.00.00.70.50.00.00.0
Fuzzification
Fuzzification Addressgenerator
0r
kr
Rule Base(1,2,..r)
Computational Block
Fig. 1 Basic Fuzzy Controller Block diagram
Fuzzification Fuzzification
Rule BaseRule Base
Computational blockComputational block
DefuzzificationDefuzzification
REC 2006 - Savannah, Georgia Feb 22-24
Address Generation
Sensor 0(Position)
NM 0.0 To ComputationalNS 0.6 BlockZE 0.3PS 0.0
Sensor 1 PM 0.0 3(Velocity)
3
NM 0.0NS 0.0ZE 0.7PS 0.5
PM 0.0
Fuzzification
Fuzzification Addressgenerator
50 r
51 r
Fig. 2 Example of Address Generation
NM NS ZE PS PMNM 1 4 3 2 1NS 5 2 1 3 2ZE 1 3 2 1 4PS 5 5 3 2 1PM 1 2 5 1 2
Position
to computationalblock
Rule Base
velo
city
Fig. 3 Bidimensional Table for Decision
Take of the fuzzy Controller
REC 2006 - Savannah, Georgia Feb 22-24
Number of Lines and Space Complexity
Rule Base
02log r
kr2log
r2log
Fig. 4 Straight forward Address Generation
k
iirN
020 ][log
Example: Let k=9,
5... 910 rrr
Using (1) we obtain
300 N
.
(1)
number ofaddress lines
space complexit
y
exponential
Fig. 5 Space Complexity Vs Address Lines
REC 2006 - Savannah, Georgia Feb 22-24
Variable Radix Numbers (Multi Radix)
),...,( 0aaa kr (2)
},...,{ 0 krrr
}1,...,0{ 00 ra
}1,...,0{ 11 ra
……………….
}1,...,0{ kk ra
110102010 ...... kkr rrrarraraaa
The Multy Radix number ra has the valueBy Definition Multy Radix number can be written as follow:
Example: Let’s assume that
7,2,5 210 rrr
Then the multi radix number 4123 it then has the decimal value
10303ra
REC 2006 - Savannah, Georgia Feb 22-24
There exists natural one-to-one correspondence between the set of multi radix numbers and the set of fuzzy rules. It is demonstrated on Fig. 6 and Fig. 7.
Sensor 0
01
Sensor 1 2
01234
0 1 2 3 4
0
0 0
1
01 4
02 3
03 5
0 4 1
1
1 0 3
11 4
12 4
13 1
142
2
2 0 2
211
22 1
231
242
Fig. 6 Set of fuzzy Rules
Fig. 7 One-to-One Mapping
REC 2006 - Georgia Tech University SREC 2006 - Georgia Tech University Savannah Feb 22-24avannah Feb 22-24
Multi Radix to Binary ConverterMulti Radix to Binary Converter
3
3
2
3
Convertor
+
0w
1w
2w
3w
0a
1a
2a
3a
9][log 32102 rrrr
Fig. 8 Multi Radix to Binary Converter
Denote by
kk rrrwrrwrw ...,...,, 1010201
.
Then
kkr wawawaaa ...22110
(4)
0 1 2 35, 5, 3, 6r r r r
REC 2006 - Georgia Tech University SREC 2006 - Georgia Tech University Savannah Feb 22-24avannah Feb 22-24
Reduction in space complexityReduction in space complexity Sensor 0
Sensor k
N
Fuzzification
Fuzzification
Addressgenerator in
noncompact binary
0r
kr
Multiradix to Binary
Converter
Rule Base
][log 02 r
][log 2 kr
Fig. 9 Rule Base with Multi radix to Binary Converter
0202102 ][log...][log]...[log NrrrrrN kk (5)
REC 2006 - Georgia Tech University SREC 2006 - Georgia Tech University Savannah Feb 22-24avannah Feb 22-24
Example:
Let 5..,9 90 rrk Then the number of initial addresses lines
300 N
.
The number of addresses lines after the Multi Radix to Binary Converter is equal to 24][log2 krN
Reduction in space complexity (cont.)Reduction in space complexity (cont.)
Data compression and error correcting codes in Rule BaseData compression and error correcting codes in Rule Base
Example:Suppose we have 2 Sensors 50 rr , 5r
Normally, for this case it will take 15 bits for representing a single row as shown in Fig.10.
Fig. 10 Single Row 2 Sensors Representation
1 4 3 0 1
Fig. 11 Center Row
To convert it to binary we need 12]15[log 52 bits. We saved 3 bits.
These bits can be used for error correction
1 4 3 0 1 001 100 011 000 001 15 Bits
Example:Suppose that we have 3 sensors, ;5210 rrr
5rIn this case we have a number with 25 digits. The biggest possible number
255 1To convert it to binary we need 59 bits. Therefore we saved 75-59=16 bits, see Fig. 12.
2 1 1 2 4 3 5 3 1 2
25 Positions25*3=75 Bits
…...…….
Fig. 12 3 Sensors representation
Data compression and error correcting codes in Rule Base Data compression and error correcting codes in Rule Base (cont.)(cont.)
REC 2006 - Georgia Tech University SREC 2006 - Georgia Tech University Savannah Feb 22-24avannah Feb 22-24
Rule Base with data compression and Error Rule Base with data compression and Error Detection/correctionDetection/correction
N
Addressgenerator
(noncompact binary)
Variable Radix to Binary
Converter
Rule Base with error correction
code
][log 02 r
][log 2 kr
0N
Decoder
…Code Word Code Word Code Word
Decode
check bits(16)
Fig. 13 Rule Base block Diagram with data compression and Error Detection/Correction
REC 2006 - Georgia Tech University SREC 2006 - Georgia Tech University Savannah Feb 22-24avannah Feb 22-24
Testing of a Fuzzy Controller by signature Analysis of test Testing of a Fuzzy Controller by signature Analysis of test
ResponseResponse
Test MPatters N Test
Responses
Variable Radix to Binary
Converter
Rule Base with error correction
code
][log 02 r
][log 2 kr
0N
Decoder
Fig. 14 Testing by signature analysis
MN ZZf 220:
(6)
)(xfS xf (7)
)(xfS xf
(8)
If the signatures fS and fS
are the same, the test is passed.
MZ2 Can be considered as a field FGF m )2(
FZx xN ,0
2
REC 2006 - Georgia Tech University SREC 2006 - Georgia Tech University Savannah Feb 22-24avannah Feb 22-24
Conclusion and future workConclusion and future work
In this paper we introduced a new representation of numbers – Variable Radix Number system. In this paper we introduced a new representation of numbers – Variable Radix Number system. Using a Variable Radix Number system we decreased the number of addresses lines in a ROM Using a Variable Radix Number system we decreased the number of addresses lines in a ROM that is used to store If-Then rules, thus reducing the space complexity of a fuzzy controller. Also that is used to store If-Then rules, thus reducing the space complexity of a fuzzy controller. Also we incorporated error correcting codes in the memory used to store If-Then rules without we incorporated error correcting codes in the memory used to store If-Then rules without substantial increasing space complexity.substantial increasing space complexity.