Peter Nachtwey
Member
You should read the whole thread. I have already demonstrated that.cheeko said:One way to find the parameters is using some search techniques. You can define some lower and higher limits of the gains (acheiveable gains not so large to implement). Search techniques can be like some heuristics. Model is simulated for different set of parameters and best are selected based on error(SP-Y).
In this example the temperature and level set point don't change. It is the flow demand or disturbance that changes. This is unpredictable.One other technique can be: Since you know the model. If you know the SPs then in advance you can find the best control signals which will minimize the error for next five to ten samples.
You are describing feed forwards. Feed forwards have been discussed on this forum and sci.engr.control.again in next sample use this same approach. doing this outputs will reach to desired SP. This is also a search technique but online.
The controller for this problem is using only proportional gains and works well enough.I think try only PI controller.
We took into account saturation but we have ignored dead time. Dead time would make using the iterative Ricatti method of calculating the LQC gains impossible and the optimizing/minimizing methods you mentioned above would need to be used. This is why I like the optimizing/minimizing method of calculating gains.Also take care of delays or saturations in system if any.
Pandiani, I have looked at your linear model.pdf. Think it has a few mistakes. I don't like the way the A array is modified because of the simulation. The bottom of the B arrays don't take into account the pump capacity.
Last edited: