Title: Hybrid Control of Structures Utilizing Fuzzy Logic and Genetic Algorithms
1Hybrid Control of Structures Utilizing Fuzzy
Logic and Genetic Algorithms
David Shook Friday July 18, 2003
2Outline
- Background Information
- Proposed Ideas
- Simulations
- Results
- Conclusions
David Shook, REUJAT Symposium 2003
3Background Information
- Fuzzy Logic
- Whats so fuzzy?
- Hybrid Control
- Why are two heads better than one?
- Genetic Algorithms (GA)
- Evolution of data?
LOGIC
David Shook, REUJAT Symposium 2003
4Fuzzy Logic?
- Non-Classical form of control
- Utilizes simple If.Then statements to control
a building - Basic Example If velocity is large and negative
and displacement is large and negative, then
control force is large and positive
David Shook, REUJAT Symposium 2003
5Big Picture - Earthquake
David Shook, REUJAT Symposium 2003
6Near-Field VS Far-Field Earthquakes
- Near-Field Earthquake
- Categorized by
- Large amplitudes
- Long periods
- High velocity pulses
- Far-Field Earthquakes
- Categorized by
- Smaller amplitude and period
- More consistent movements
David Shook, REUJAT Symposium 2003
7Hybrid Control
- Switching Rules
- Optimized by GA
- Intelligent Fuzzy Optimal Active Control (IFOAC)
- For Far-Field Earthquakes
- Utilizes EQ prediction data, Control Interval
- Reflective Fuzzy Active Control (RFAC)
- For Near-Field Earthquakes
- Optimized by GA
David Shook, REUJAT Symposium 2003
8Big Picture - IFOAC
David Shook, REUJAT Symposium 2003
9IFOAC
- Intelligent Fuzzy Optimal Active Control
- Best for Far-Field Earthquakes
- Takes into account subjective arguments
- Economy
- Comfort of inhabitants
- How?
- Based of man-made rules that can be designed with
such parameters in mind
David Shook, REUJAT Symposium 2003
10IFOAC
- Earthquake prediction
- Uses a control interval for prediction
- Optimal control variable for the next interval is
determined by fuzzy membership functions - Three memberships function based on
- Structural Response (Y) - objective
- Control Force (U) constraint
- Stroke of Actuator (S) constraint
David Shook, REUJAT Symposium 2003
11IFOAC
- These three factors enter a Fuzzy Maximizing
Decision - By these factors ideal control variables are
produced and then the control force is calculated - Why best for far-field earthquakes?
- Far-field earthquakes are slower in variation
- Thus using a control interval for EQ prediction
is suitable - The Pros Takes into account subjective
arguments - The Cons Not as effective in near-field
situations
David Shook, REUJAT Symposium 2003
12Big Picture - RFAC
David Shook, REUJAT Symposium 2003
13RFAC
- Reflective Fuzzy Active Control
- Best for Near-Field Earthquakes
- Fuzzy Rules created and optimized by Genetic
Algorithms - The Pros More effective in Near-Field
Situations - The Cons Doesnt take into account
subjective arguments
David Shook, REUJAT Symposium 2003
14Big Picture Switching Rules
David Shook, REUJAT Symposium 2003
15Switching Rules
- Simple algorithm to determine which control
system to use - IFOAC or RFAC
- Rules based on building response in terms of
displacement and velocity - a and b values are determined by G.A.
- n0 and n1 are determined by G.A.
- Determined to be either IFOAC or RFAC
David Shook, REUJAT Symposium 2003
16Big Picture Genetic Algorithms
David Shook, REUJAT Symposium 2003
17Genetic Algorithms
- Based of the idea of Survival of Fittest
- Single member of population is called a
Chromosome of base10 numbers - Evolution occurs as follows
- Population is created
- Some of the population dies off due to poor
evaluations - Crossovers and mutations produce offspring
- A new generations is produced and the cycle
repeats until a predefined accuracy is reached
David Shook, REUJAT Symposium 2003
18G.A. Example
David Shook, REUJAT Symposium 2003
19G.A. Example
David Shook, REUJAT Symposium 2003
20G.A. Example
David Shook, REUJAT Symposium 2003
21G.A. Example
1 0 0 0 1
David Shook, REUJAT Symposium 2003
22G.A. Example
1 0 0 0 1
David Shook, REUJAT Symposium 2003
23G.A. Example
1 0 0 0 0
David Shook, REUJAT Symposium 2003
24Big Picture
David Shook, REUJAT Symposium 2003
25Computer Simulations
- Programs were custom made in Delphi by Mr. Takagi
- Three parts
- Optimization of RFAC by G.A.
- Optimization of Switching Rules by G.A.
- Actual simulation
David Shook, REUJAT Symposium 2003
26Previous Case
- Published in 2002
- Utilized similar techniques as explained above,
but a different optimization method is used for
the RFAC - Results of Previous Case
David Shook, REUJAT Symposium 2003
27Proposed Ideas
- Match/Improve results from previous case
- How?
- Use different data for optimization of G.A.
- Vary IFOAC parameters
- Change target displacements in optimization
David Shook, REUJAT Symposium 2003
28Varying Parameters
- Positive results from optimization on different
data - Positive results from varying IFOAC parameters
- Ineffective results from changing target
displacements - Reducing the target displacements causes fuzzy
systems to be ineffective
David Shook, REUJAT Symposium 2003
29Optimization of RFAC
- Optimization with the 1995 Kobe Earthquake was
found to have the best results - When tested in other earthquakes it had the best
results - Fuzzy limits allowed for best distribution of
resulting output
David Shook, REUJAT Symposium 2003
30Variation of IFOAC Parameters
- Proposed optimal values for IFOAC
- 50 less than previous case
- Maximum Displacement 3 cm
- Maximum Velocity 75 cm/sec
- Maximum Control Force 100 ton force
- Why reduce these values ?
- By reducing the fuzzy range a more efficient
and effective evaluation occurs
31Results El Centro
David Shook, REUJAT Symposium 2003
32Results
- Percent Reduction of Maximum Displacement in four
proposed earthquakes
David Shook, REUJAT Symposium 2003
33Problems
- RFAC is neglected by control algorithms
- Why?
- RFAC maximum values are too large thus
ineffective in evaluation - Current G.A. program is inefficient in producing
smaller fuzzy maximum values for RFAC while
optimizing on one earthquake - Resolution Use smaller fuzzy limits of RFAC as
proposed in previous case
David Shook, REUJAT Symposium 2003
34Future Work
- Continue work in United States at Texas AM
University - Professor Roschke
- Neuro-Fuzzy control with MR Dampers
- Look for possible implementation of methods
learned - EX Genetic Algorithms, Hybrid Control
David Shook, REUJAT Symposium 2003
35Acknowledgements
- 2003 REUJAT Program
- National Science Foundation
- Professor Dyke
- Professor Christenson
- Kobe University
- Professor Kawamura
- Professor Tani
- Students
- Professor Roschke, Texas AM
David Shook, REUJAT Symposium 2003
36 37Hybrid Control of Structures Utilizing Fuzzy
Logic and Genetic Algorithms
David Shook Friday July 18, 2003