Hello
I'am working on cooperative robot manipulator and I want to designing a fuzzy controller for trajectory tracking of manipulator joints.
I used genfis1 to generate fis structure , but it's not working in my case.
I used anfis toolbox in matlab too to generate fis structure, but it's not working too.
please help.
Hi
check this out.
https://www.researchgate.net/post/What_are_the_differences_between_genfis1_and_genfis2_when_used_in_Matlabs_ANFIS?_tpcectx=qa_overview_following
Hello Aboldazl,
In a fuzzy inference system the rules have to be derived from expert knowledge and tuned by trial and error.
By using the genfis commands, you are preparing an initial FIS for an ANFIS. If you are trying to find a sort of automatic way to generate the FIS you can look into the ANFIS kind of system.
I suggest you look deeper into what these two concepts are (FIS and ANFIS) otherwise you will keep having trouble.
A Fuzzy inference system refers to an inference system based on mamdani or takagi sugeno inference rules. In these systems, the number of inputs and outputs can vary and the training process should be performed off line though the composition of linguistic rules by the expert. In this case, you!.
Genfis 1, 2 and 3 commands allows you to use the Adaptive Neuro Fuzzy Inference System ANFIS that is a different structure of FIS in which the pure FIS structure is used as antecedent part, and the consequent is hybridized with a neural network. ANFIS training can be performed from the data set and the initial FIS generated by genfis1, 2 allows you to have a MISO system, while genfis 3 allows you to stablish a MIMO system.
The difference between them is that when you use genfis3 in mamdani scheme, ANFIS training cannot be performed.
With all this in your head my main suggestion is the following:
1. Your main problem fuzzy trajectory tracking should be compared with traditional controllers such as PID´s or something…. I suggest you to check some of my papers.
2. But the point is: try to develop the PIDs with enough performance and after that, stablish a FIS with some inputs such as error, error derivative or integral error and the control action or changes in control action as outputs. This FIS should be written with the linguistic rules that the expert can interpret from the behavior. BE CAREFULL selecting the inputs for the FIS.
3. Later, you can generate an ANFIS structure through genfis1, 2 and 3 with a single output. Use the data from PIDS and the open loop data to train.
4. finally, use the ANFIS command to train the genfis generated and evaluale if is superior to the basic structures.
It is very important for me that my students understand completely the differences between the FIS, ANFIS, neuro fuzzy and neural networks structures.
Speaking of your problem, should be easy if you use our controllers for trajectory tracking (Scaglia et al, 2006……) as teachers to your ANFIS or FIS controllers.
Best regards
Lucia.
Dear Olga Lucía Quintero
I designed a PID controller for my system , and this data set used for :
1 - genfis1 and genfis2 command.
2 - I use neuro fuzzy editor toolbox (anfis editor) to train the genfis generated fis structure .
but this 2 methods for generating the fis structure is not working.
Hi Aboldazi
If you have measurements of the inputs and outputs of the system, you can use a clustering algorithm such as C-means, Gustafson-Kessel, ... , to make the learning premises of a neurofuzzy model.
The consequent of the fuzzy rules may have the form of state-space structure of the system that you want to modeling and identify. Using weighted least squares by the membership values for each clustering. That is, for each clustering you will have a submodel of Takagi-Sugeno type with the resulting state space. This will give an approximation of the robot dynamics with a fuzzy system with linear submodels for each fuzzy rule
Now with identified Fuzzy model, you need to perform cross-validation. In this first part you can use the control techniques proposed by Professor Tanaka: Fuzzy Control Systems Design and Analysis.
This allows you to design a control law for path tracking with the methodology of theorem quadratic Lyapunov inequality using LMIS in the solution.
Best Regards
Julio
I am afraid that you will always be disappointed with fuzzy control. Ist well know it does work.
You should use "Flatness-based control" and especially "Exact feed-forward linearization based on differential flatness". You will be able to achieve whatever your need on your robot.
Article Robustness Analysis With Respect to Exogenous Perturbations ...
Article Continuous-time non-linear flatness-based predictive control...
Conference Paper Nonlinear Model Predictive and Flatness-Based Two-Degree-of-...
- Badges
- Science topic
- Similar topics
- Engineering
- Control Theory
- Control Systems
- Control Systems Engineering