Convention The following conventions are used throughout this document. 'Enter' means the

---
Master Index Current Directory Index Go to SkepticTank Go to Human Rights activist Keith Henson Go to Scientology cult

Skeptic Tank!

Convention The following conventions are used throughout this document. 'Enter' means the Enter key or the Return key. Type 'key' means press the specified key without 'Enter'. Press F? means press the specified function key without 'Enter'. Enter means type in the specified text and followed by 'Enter' key. Select/Use an option: use cursor keys to move the highlighted option and press 'Enter', or type the upper-cased character on the left. GAF is not case sensitive, use either upper or lower cased key. Using The Menu The GAF system menu, at the top of the screen, is the primary user interface to the system. System menu provides sub-menu function selection (function menus): Info, Run, Display, Control, Adapt, Method, Segment, Feedback, Eval, Other, and Help. Use the first character of each selection or use cursor keys (right/left arrows) to select the sub-menu. When a function menu is selected a pull down menu appears to provide function selection. To select a function either by the character on the left of that function or move the cursor with up/down arrow keys and then "Enter". When a function is selected, further menu selection or data entry menu may be provided by that function. Press ESC in function menu will return to system menu. Press ESC under system menu will bring run function menu, except when the system menu was ESCaped from any of the running screen (i.e. from adapting or simulation). Please refer GAF.DOC for details. The following steps are typical for function selection under simulation or adapting. press "ESC" to escape back to system menu select desired function menu (see above) select desired function (see above) press "ESC" again under system menu to go back Simulation This simple example demonstrates the application of the fuzzy logic to control a motor based on its angular position feedback. Lets assume that this is a robot arm, and its arm extension limit (in angular position) is from 0 degree to 70 degree. The inputs to the control loop are the position, position error, and the speed, the output is the current to the motor. The example also demonstrates that using fuzzy logic to simulate the feedback loop, e.g. the position feedback and the angular speed. The file "robotarm.gaf" defines the control loop, and the file "robotarm.fbk" defines the feedback loop. Run Simulation After start the demo, you'll find the system menu is at the top of the screen. Select run menu by typing 'R', then select "simulation mode" by either type 'S' or move high lighted cursor to "simulation mode" and press 'Enter'. After simulation mode is selected, GAF displays the control segment on the screen and waits for user input. Please refer to GAF.DOC for details of the display windows. Type 'F' to start the simulation. You should see the trend plots of inputs and outputs on the right and the rules and firing status on the left of the screen. Reset the system Press function key F7 to reset the system. The system reset is defined by the INITIALIZATION declaration in all segments of the system. Type 'F' again to run the simulation. View the feedback loop Now switch to the feedback loop by pressing F3, the screen displays the feedback segment. Press F7 to reset the system, and then type 'F'. Now watch the plot of the inputs and the outputs for the feedback loop. Stepping through Let's switch back to the control loop by pressing F3 twice, and reset it by F7. Now press 'Enter' to see the demo program goes one step. The program allows you to do one step by 'Enter' or multiple steps by typing 2 to 9, or 10 steps by typing '0'. Type '0' and watch the program goes through 10 steps and note different rules are fired. At any time type 'F' will bring you to the end of the plot frame. Continuous mode Press 'C' to run the simulation continuously. You can stop continuous mode by pressing 'C' again or ESCape (press ESC) back to system menu. While in continuous mode press F7 to reset the system. Change variable value Stop continuous mode by pressing 'C'. Now let's change the simulation from raising the arm to lowering the arm. Press F7 and then ESC to escape back to system menu. Press 'S' to select segment menu. The segment menu allows you to switch segment and change characteristics of the segment. Switch to feedback segment by selecting "segment number" under segment menu, and enter '1' for feedback segment. Select segment menu again, and select "set input value". GAF pops all inputs in a menu beneath the segment menu, select "REFERENCE" input and enter its new value to 0. Select segment menu one more time to change the input variable "POSITION" value to 70 (from "set input value"). Now get back to the simulation by pressing ESC, press F3 to switch back to control segment, and start simulation by stepping through it. Note that you can also change the "POSITION" value from the control segment (segment 0). View the plot From system menu type 'D' to select display menu. The display menu allows you to rearrange the display windows. Let's select "select plot only mode" by typing '5', and then escape back to simulation. GAF should redraw the screen with trend plot takes most of the screen. Run simulation by hiting F7 and then 'F'. View the combined result From Display menu select "select whole display mode". The right portion of the screen splits into two parts, the upper half shows the plot, and the lower half displays the combined result for the output. The combined result shows the trapezoid result of all fired rules (rules generate non-zero output). Rerun the simulation by pressing F7, and then 'F'. Now, the combined result tells you about the firing results of all rules and the defuzzified crisp output (the current of the motor) from all fired rules. View details about the fuzzy inference The above exercise only allows you to view the combined result of the output. GAF also allows you to examine the firing details of the rules. This feature along with the stepping mode is the best way to tuneup or debug the control segment. From the segment menu, select "display rule firing", GAF highlights the first rule by using the left column of the text window. Use down arrow key to move to rule 5 and press 'Enter' to select it. Go back to the simulation, you'll find the selected rule's fuzzy sets are displayed to the left of the combined result. Note that each fuzzy set has the same color as the text window. Try previous exercises to view the changes of the fuzzy rule firing status. Disable rules GAF allows user to disable individual rule for testing fuzzy segment. Let's disable rule 10, this will effectively break the control system. Since there is no rule to respond when the POS_ERROR is VERY_LARGE, so after system reset (press F7) there will be no output from the control segment. From segment menu select "disable rule", GAF highlights the rule ID 0 on the left of the screen. Press 'End' to move the highlight cursor to rule 10, and then type 'Enter'. Now restart the simulation by pressing F7 then 'F'. You should see the rule 10 is marked with 'X' to indicate it's been disabled, and also that there is no response to the position of the robot arm. Enable rules There are two options under segment menu allow you to enable rule(s). The "enable all rules" option enables all rules in this segment. The "enable rule" enables the selected rule. So, type 'A' under segment menu to enable rule 10. Change sampling rate Another important feature in GAF is that you can tune your system by trying different sampling rate (or cycle time defined in each segment). From control menu select "sim schedule rate", GAF prompts for new rate. Enter 0.2 (in second) and restart the simulation. Increment/decrement sampling rate An easy way to change the simulated sampling rate is to use the '+' and '-' keys. Press '+' each time will cause the sample rate to increase 0.005 second, and press '-' will decrease the sample rate by 0.005 second. The amount of increment/decrement (default is 0.005 second) can be changed through the control menu's "Increment value" option. Now put the simulation in continuous mode by 'C', reset it by F7 and then use '+' and '-' keys to adjust the sample rate to find out what's the max stable rate of the control loop. Change output gain You can change the gain of the output, from segment menu's "set output gain" option. The default gain is 1.0. Check individual segment This feature allows you to examine one segment at a time. Let's examine the feedback loop's response at a constant motor current. Press F3 to switch to the feedback segment. From run menu select "check segment". Then from segment menu change the value of input variable "CURRENT" to 1.0. Run the simulation by typing 'F' to see the response of constant current input. Getting online help Press F1 to get online help. Display the system limits Press F2 to see the system limits. View segment information Press F4 to see the variable definition, initialization and normal functions, and other information for the selected display segment. View symbol definition Press F5 to see the fuzzy set symbols' definition for this segment. View preset definition Press F6 to view the preset definition for this segment. Set new plot time frame The default plot frame period is five seconds (except in sim clock mode), this can be changed through display menu's "plot duration" option. View local variables The default display mode is to display IN, OUT, and INOUT variables, the LOCAL variables are not displayed. Use "display local variable" option in display menu to enable/disable LOCAL variables. Quit the program Press 'Alt-Q' to quit the program under simulation or select "quit" option from run menu. Adapting the Fuzzy Segment The genetic adapt mode is an offline adaptation mode. While fuzzy logic mimics the human's imprecise reasoning, the genetic algorithm mimics the evolution of the nature. The genetic algorithm uses randomization techniques to reach local optimum without exhausting searching through the solution state space. With randomization, genetic algorithm is also capable of escaping from local optimum and hopping into a new and better state. Apply genetic algorithm in fuzzy control eliminates problems associated with fuzzy system. GAF uses genetic algorithm to derive proper rules and fuzzy sets from the initial rules. By changing, adding, deleting rules and fuzzy membership sets of the fuzzy system, the genetic algorithm automatically adapts and optimizes the fuzzy control system. The adapting example is built on top of the simulation example, by adding a third evaluation segment. The file "robotarm.evl" defines the evaluation segment. Start Adapting From system menu, type 'R' to bring up run menu. Select "adapt mode" by pressing 'A'. GAF starts adapting the control loop and displays the control segment along with adapt status on the top of the screen. Pause every iteration Press Alt-P to halt the adapting after each iteration. Press 'C' to continue next iteration. Press Alt-C to switch to continuous mode. Pause every cycle Press Alt-P to stop the adapting after each cycle. Press 'C' to continue next cycle. Press Alt-C to switch to continuous mode. Change size of gene pool Press 'Esc' while in continuous adapting, GAF brings up the system menu. Select adapt menu by type 'A'. From adapt menu select 'max best items', and enter a new pool size. This will change the number of genes will be saved (and evaluated) by GAF. Change minimum score Again, from adapt menu, select "minimum score", GAF prompts for new minimum value. Enter 0.5, this tells GAF to accept all evaluation results ,which are greater or equal than 0.5. The default of the minimum score is 0.6. Change method The method menu provides user to control the algorithm of genetic adaptation. Select "adding rule" from method menu, enter 0 for its new weight. With zero weight, GAF will not generate (adding) new rule into the control loop. Print the result Select 'Other' menu from the system menu, and type 'A' to generate report for all genes in the gene pool. The result will be saved in the adapt.rpt file of the current directory. Generate statistic report Again type 'S' (generate statistic log) from 'Other' menu, GAF produces a statistic report file with the file name specified by 'statistic to' at the bottom of the menu. The log includes statistics for all methods. Based

---

E-Mail Fredric L. Rice / The Skeptic Tank