Add Center of Gravity Defuzzification Method

[Unknown]

Would it be possible to add a new defuzzification method? National Instruments has a Center of Area or Center of Gravity defuzzification method that is extremely useful. We would like to port the fuzzy controller we’ve implemented in National Instruments to FuzzyLite; however, we are unable to do so because we are using the Center of Area/Center of Gravity defuzzification method, which is unsupported by FuzzyLite.

The following is a description of how that is implemented by National Instruments.

http://zone.ni.com/reference/en-XX/help/370401H-01/lvpidmain/center_of_area/

[Juan Rada-Vilela (admin)]

Hi,

thank you for your question.

There seems to be some differences in the names used by the different players in fuzzy logic control, but the differences are only in the names. The centre of area or gravity that you are asking for is the Centroid in fuzzylite (and also in matlab). In jfuzzylogic, they make a difference between centre of area and centre of gravity, where the centre of area refers to the Bisector (in fuzzylite and matlab), and the centre of gravity refers to the Centroid (in fuzzylite and matlab).

Let me know if this answers your question.

Cheers

[Unknown]

Hello,

Thanks for your explanation. There appears to be something different about what National Instruments is doing, and what everyone else does 🙂 At the end of this post I’ve listed a block of FLL “code” that I’m using. I’ve also linked images to what National Instruments GUI shows.

When I set the “TestInputVar” to a value of 160, National Instruments says the output variable “TestOutputVar” is 100. When I load the FLL block into FuzzyLiteQt, and set the input value to 160, it says that the output variable “TestOutputVar” is 50. So I’m trying to determine what the difference is and how to overcome that.

In my example, the output variable ranches from 0 to 100 where 0 would represent 0% of something and 100 would represent 100% of something (perhaps a duty cycle). The input variable ranges from 0-550.

Engine: TestController
InputVariable: TestInputVar
enabled: true
range: 0.000 549.999
lock-range: false
term: AtSetpoint PiShape 298.000 299.000 301.00 302.00
term: VeryLow Trapezoid 0.000 0.000 260.000 300.00
term: VeryHigh Trapezoid 300.000 305.000 550.000 550.00
term: Low PiShape 260.000 300.000 300.000 300.00
term: High PiShape 300.000 300.000 300.000 305.00
OutputVariable: TestOutputVar
enabled: true
range: 0.000 100.000
lock-range: false
aggregation: Maximum
defuzzifier: Centroid 100
default: nan
lock-previous: false
term: Full Constant 100.000
term: Medium PiShape 10.000 15.000 30.000 35.000
term: Off Constant 0.000
term: High PiShape 15.000 20.000 40.000 45.000
term: Low PiShape 0.000 5.000 25.000 30.000
RuleBlock: mamdani
enabled: true
conjunction: Minimum
disjunction: Maximum
implication: Minimum
activation: General
rule: if TestInputVar is AtSetpoint then TestOutputVar is Medium
rule: if TestInputVar is VeryLow then TestOutputVar is Full
rule: if TestInputVar is VeryHigh then TestOutputVar is Off
rule: if TestInputVar is Low then TestOutputVar is High
rule: if TestInputVar is High then TestOutputVar is Low

Screenshot 1 of NI Fuzzy Logic Controller:

Screenshot 2 of NI Fuzzy Logic Controller:

Screenshot 3 of NI Fuzzy Logic Controller:
[url=https://flic.kr/p/XuYQ8z]Screenshot3[/url]

[Unknown]

Hello,

Thanks for your explanation. There appears to be something different about what National Instruments is doing, and what everyone else does 🙂 At the end of this post I’ve listed a block of FLL “code” that I’m using. I’ve also linked images to what National Instruments GUI shows.

When I set the “TestInputVar” to a value of 160, National Instruments says the output variable “TestOutputVar” is 100. When I load the FLL block into FuzzyLiteQt, and set the input value to 160, it says that the output variable “TestOutputVar” is 50. So I’m trying to determine what the difference is and how to overcome that.

In my example, the output variable ranches from 0 to 100 where 0 would represent 0% of something and 100 would represent 100% of something (perhaps a duty cycle). The input variable ranges from 0-550.

Engine: TestController
InputVariable: TestInputVar
enabled: true
range: 0.000 549.999
lock-range: false
term: AtSetpoint PiShape 298.000 299.000 301.00 302.00
term: VeryLow Trapezoid 0.000 0.000 260.000 300.00
term: VeryHigh Trapezoid 300.000 305.000 550.000 550.00
term: Low PiShape 260.000 300.000 300.000 300.00
term: High PiShape 300.000 300.000 300.000 305.00
OutputVariable: TestOutputVar
enabled: true
range: 0.000 100.000
lock-range: false
aggregation: Maximum
defuzzifier: Centroid 100
default: nan
lock-previous: false
term: Full Constant 100.000
term: Medium PiShape 10.000 15.000 30.000 35.000
term: Off Constant 0.000
term: High PiShape 15.000 20.000 40.000 45.000
term: Low PiShape 0.000 5.000 25.000 30.000
RuleBlock: mamdani
enabled: true
conjunction: Minimum
disjunction: Maximum
implication: Minimum
activation: General
rule: if TestInputVar is AtSetpoint then TestOutputVar is Medium
rule: if TestInputVar is VeryLow then TestOutputVar is Full
rule: if TestInputVar is VeryHigh then TestOutputVar is Off
rule: if TestInputVar is Low then TestOutputVar is High
rule: if TestInputVar is High then TestOutputVar is Low

Screenshot 1 of NI Fuzzy Logic Controller:
NI Screenshot1

Screenshot 2 of NI Fuzzy Logic Controller:
NI Screenshot2

Screenshot 3 of NI Fuzzy Logic Controller:
NI Screenshot2

[Juan Rada-Vilela (admin)]

Hi Aaron,

I am sorry that your posts were marked as spam by wordpress, and hence not shown. Thank you for emailing me.

For completeness, I will elaborate on your issue.

The problem in your controller is the usage of Constant terms in the output variable for Off and Full. The centroid defuzzifier integrates over the terms in the output variable, and the integration over the constant terms is just not supported by this defuzzifier. The people from National must be doing some very specific handling of this case of integration over.

The solution I proposed to you (see at the end of post) substitutes the constant terms by integrative Triangle terms at the locations where you want them (0 and 100), which makes me extend the range of the output variable a bit more to integrate over the terms. In addition, I have increased the resolution of the defuzzifier for better accuracy.

Personally, I do not like my solution much. In the short term, I would avoid using the Engine::process() method to process the engine, and do it manually instead. You can look at the logic inside this method, and add the exceptions to turn the engine off and full if the corresponding memberships are activated.

In the long term, I want to improve the fuzzy consequents to override defuzzification. For example, if x is a then z is 0.9.

Lastly, I am sorry I did not see your messages being marked as spam. I would have elaborated on your case then.

Kindly let me know if this is of any help, or if you found a better solution.

Cheers.

Engine: TestController
InputVariable: TestInputVar
  enabled: true
  range: 0.000 549.999
  lock-range: false
  term: AtSetpoint PiShape 298.000 299.000 301.000 302.000
  term: VeryLow Trapezoid 0.000 0.000 260.000 300.000
  term: VeryHigh Trapezoid 300.000 305.000 550.000 550.000
  term: Low PiShape 260.000 300.000 300.000 300.000
  term: High PiShape 300.000 300.000 300.000 305.000
OutputVariable: TestOutputVar
  enabled: true
  range: -5.000 105.000
  lock-range: false
  aggregation: Maximum
  defuzzifier: Centroid 1000
  default: nan
  lock-previous: false
  term: Off Triangle -1.000 0.000 1.000
  term: Low PiShape 0.000 5.000 25.000 30.000
  term: Medium PiShape 10.000 15.000 30.000 35.000
  term: High PiShape 15.000 20.000 40.000 45.000
  term: Full Triangle 99.000 100.000 101.000
RuleBlock: mamdani
  enabled: true
  conjunction: Minimum
  disjunction: Maximum
  implication: Minimum
  activation: General
  rule: if TestInputVar is AtSetpoint then TestOutputVar is Medium
  rule: if TestInputVar is VeryLow then TestOutputVar is Full
  rule: if TestInputVar is VeryHigh then TestOutputVar is Off
  rule: if TestInputVar is Low then TestOutputVar is High
  rule: if TestInputVar is High then TestOutputVar is Low

[Author]

Posts