@Barp wrote:
Jzee,
Well, you are right about the more information is better to control the system. However, now you are stepping into an Advanced Control System to take 6 measurements and act on 2 outputs. If you look into Control Theory, this is Multi-Input and Multi-Output (MIMO) system and, as such, you will have to come with a Control Strategy to better use this information, as for example, Cascade PID, Feedforward control, decoupler, filtering and/or combination of it to control the plant. You could also use Fuzzy for this purpose, however, our current implementation only uses 4 inputs. Without knowing what kind of plant your are trying to control, it is hard for me to advice in any control strategy right now.
However, if you are have the Model of the plant that you are trying to control, you could easily upgrade to the Control Design Toolkit, and you can use more advanced techniques to control your plant. You can use state feedback control algorithms as Pole-Place or LQR where you just need to provide the matrix multiplication to obtain the controller that you need. This is very easy to use, as long as you have the model. If you want to know more about it, look at our manual online, in special Chapter 11 and 12:
http://digital.ni.com/manuals.nsf/websearch/1B2AF86E66AD7F61862571060055D6AB
If you don't have the model, AND your plant is linear, you also could use System Identification Toolkit to obtain the model of it. However, IF your plant is nonlinear (most of them are), you could use Simulation Module to model it and linearize it around the operating point or use System Identification to "fit" the parameters through measurement.
This is Model-Based Control Design Techniques and you could have access with the Developer Suite with Control Design and Simulation Option. 🙂
Anyway, if you can give more information about your plant, maybe we could better advice on what to do next.
Hope this helps!
Barp - Control and Simulation Group - LabVIEW R&D - National Instruments
Barp
first, i've access to all LabVIEW toolkits, including the CD and identification toolkits through my university.
I tried actually earlier to use pole placement technique (and currently trying with LQR). i'm more interested in my system tracking a reference signal rather than just u=-kx strategy! the thing is using (u=-kx+r) as a strategy is not actually sufficient because this will make a steady state error. so i tried the tracking procedure in Franklin, Powel, and Emami, i guess that was p.380 (not sure) but it's in the state space chapter. anyway, i was faced with some problems, that this strategy only works if the number of inputs is >= the number of inputs (I knew that after few days of designing the controller and faced with weird errors!). anyway, i guess i learnt the hard way.
so that was the reason i resorted for PID but after some reading i found another procedure to track a reference signal with no tracking error to a step signal which is augmenting the integral of the error with the states while calculating the gain K matrix as follows
e_dot= r - y where the dot means it's differential of the actual signal, r: input, y:output, e: error
e_dot=r - Cx where x: states,C: C-matrix
then the full states would be (written with matlab syntax), the ";" means a new row and the 0 is a zero matrix of compatable dimensions
[x; e]=[A 0;-C 0]+[B;0]u+[0;i]r
where A,B,C are the original matrices of the system, i: is identity matrix depending on the number of signals required to track.
i was unlucky enough that my linearised system (originally nonlinear but i linearised it by hand about an operating point) was "ill posed" when i used LQR and wanted to track 6 reference signals. i mean when i added the six other rows of e_dot to the actual system. so i finally decided that i want only 2 outputs to track 2 reference signals. so instead of adding 6 rows of e_dot i added only 2 with the requirement that the 1st, and last output track the reference signals. so far, as i'm testing each part of the design, it's going ok and i got a matrix of the gains from the LQR but didn't complete anything further yet. I'll implement an observer/kalman filter as well but this will be at a later stage as i've all my outputs measured. i will implement the observer only to check the correctness/reliability of the sensors signals.
i'll be soon faced with the problem of implementing an integrator - not able to use the simulation integrator because my controller will work on an fpga of the cRIO, so i've to get an integrator that fits in a while loop. also my design is in the continuous system so not able to use the discrete integrator.
i don't know how to use a cascade PID, or what is the advantage of a feedforward to be honest. also, not sure if i'm able to implement any other strategy than the ones i've previously mentioned, as it seems that my control knowledge sucks
😞my system is a control of a uni-cycle, to at least stabilize it as a first stage. i've a motor driving the wheel to control the pitch direction, and another motor driving a flywheel to control the uni in roll/yaw directions. think of it as an inverted pendulum over a cart but this is harder because it can fall off in direction perpendicular to the wheel movement direction. this is why i've the flywheel. The thing is the system is big and heavy so the control strategy has to be well designed. any help/idea is much appreciated?
