Summary of "Responsi Sistem Pakar Pertemuan 11 - Fuzzy Sugeno + Fuzzy Tsukamoto"

Summary of “Responsi Sistem Pakar Pertemuan 11 - Fuzzy Sugeno + Fuzzy Tsukamoto”

This video covers the concepts and practical application of two fuzzy inference methods used in expert systems: Fuzzy Sugeno and Fuzzy Tsukamoto. It compares these methods with the Mamdani fuzzy inference method, highlighting their differences, advantages, and computational procedures through a case study involving washing machine spin speed control based on inputs like the number of clothes and dirtiness level.

Main Ideas and Concepts

1. Fuzzy Sugeno Method (Takagi-Sugeno-Kang Model)

Developed by Takagi and Sugeno in 1985 as an improvement of the Mamdani method.

Key characteristics:

The output (consequent) of each fuzzy rule is a linear function or a constant, not a fuzzy set as in Mamdani.
Avoids complex defuzzification processes by producing a crisp output directly.
Enhances computational efficiency.
Easier to integrate with mathematical models and machine learning algorithms.

Comparison with Mamdani:

Mamdani creates new membership functions for each rule and uses methods like centroid defuzzification to obtain crisp output.
Sugeno directly calculates crisp outputs using linear functions or constants, thus simpler and faster.

Membership Functions:

Uses increasing and decreasing membership functions for input variables.
Example input variables: number of clothes (few or many), dirtiness level (low, medium, high).

Calculation Steps:

Fuzzification of inputs using membership functions.
Application of fuzzy rules with outputs as constants or linear functions.
Calculation of alpha predicates (degree of truth) for each rule.
Final output is computed as a weighted average of rule outputs:

[ Z = \frac{\sum (\alpha_i \times z_i)}{\sum \alpha_i} ]

Example result: For 50 clothes and dirtiness 58, the washing machine spin speed is approximately 734 RPM.

2. Fuzzy Tsukamoto Method

Developed by Yesun Sukamoto in 1979.
Differs from Sugeno and Mamdani mainly in the inference and defuzzification steps.

Key characteristics:

Uses monotonic membership functions for the output fuzzy sets.
The output for each rule is a crisp value derived from the inverse of the membership function corresponding to the rule’s firing strength (alpha predicate).
The final output is a weighted average of these crisp values, similar to Sugeno.

Calculation Steps:

Fuzzification of inputs.
Calculation of alpha predicates (degree of truth) for each rule.
For each rule, calculate the crisp output ( z_i ) by applying the inverse membership function corresponding to the alpha predicate.
Final output is calculated by weighted average:

[ Z = \frac{\sum (\alpha_i \times z_i)}{\sum \alpha_i} ]

Example results showed Tsukamoto produces a higher output value (e.g., 1061 RPM) compared to Sugeno and Mamdani methods.

Case Study: Washing Machine Spin Speed Control

Inputs:
- Number of clothes (0-100 scale, categories: few, many)
- Dirtiness level (0-100 scale, categories: low, medium, high)
Output:
- Spin speed in RPM (slow ~500 RPM, fast ~1200 RPM)
Membership functions and rules were applied to calculate the spin speed using Mamdani, Sugeno, and Tsukamoto methods.
The Sugeno and Tsukamoto methods were shown to be computationally simpler and faster compared to Mamdani.
The choice of method depends on the specific application requirements.

Methodology / Step-by-step Instructions

For Fuzzy Sugeno:

Define input membership functions (increasing/decreasing).
Define fuzzy rules with consequents as linear functions or constants.
Fuzzify input values.
Calculate the degree of truth (alpha predicate) for each rule using the min operator.
Compute each rule’s output ( z_i ) using the linear function or constant.
Calculate final output ( Z ) by weighted average of ( z_i ) weighted by alpha predicates.

For Fuzzy Tsukamoto:

Define monotonic membership functions for output variables.
Define fuzzy rules.
Fuzzify input values.
Calculate alpha predicates for each rule.
For each rule, find crisp output ( z_i ) by applying the inverse membership function corresponding to alpha predicate.
Calculate final output ( Z ) by weighted average of all ( z_i ).

Speakers / Sources Featured

Primary Speaker: Lecturer explaining the concepts and calculations of Fuzzy Sugeno and Fuzzy Tsukamoto methods.
Bro Angga: Mentioned as the next presenter who will explain Fuzzy Tsukamoto in detail.
No other speakers explicitly identified.

Conclusion

Fuzzy Sugeno and Tsukamoto methods offer computational advantages over Mamdani by simplifying defuzzification.
Sugeno uses linear functions or constants as rule consequents, Tsukamoto uses monotonic membership functions with inverse mapping.
The choice between Mamdani, Sugeno, and Tsukamoto depends on the case and application needs.
Understanding these methods is essential for efficient design of fuzzy expert systems, especially in engineering and control applications like washing machines.

If you want, I can also provide a comparison table or detailed formulas used in the video.