Hi everyone! I hope you enjoyed this video. If you have any questions or comments you'd like me to see, please leave them on this comment so that I get notified and can respond. Cheers!
Hi Brian, thanks for this Brilliant explanation of Feed forward control .. i only wished you make an extension to this video where you show also an example with the feedforward through the setpoint path .. and also another example focused on the noise filtering and decoupling the noise error from setpoint and disturbance errors and how this becomes easier with feed forward (what you explained in the beginning) .. really thanks man for all your effort!!
Thanks for the comment, Ayman. I would love to have covered those topics in this video as well but I have to balance getting in as much as possible versus making a video so long people won't sit through the whole thing! Maybe I can make a setpoint feedforward video that shows how noise is also decoupled on my channel as a suplement to this one. Cheers!
Sorry Brian i just watched the video again (really liked it) .. i am very interested on how to find the best causal transfer fn to approximate the inverse of the plants transfer fn .. would you happen to know some techniques for this that i can read about? Or to have already a video from your channel about this topic? .. thanks again and keep up the great work :D
This is an informative read on the topic: buchholz.hs-bremen.de/inversion/inversion_impossible.pdf. If the system has significant dead time (that is, it's long enough that you care about it), then you can approximate the dead time first with an LTI system using a Pade approximation.
Very well done. I wish there was a feed forward controller on the youTube algorithm to increase the value to technical advertiser's for excellent technical content like yours.
This channel is gold. I am trying to make my Betaflight helicopter to dive with terminal speed exactly the same as CC3D controlled same craft. I look at both source code and couldn't understand if their PID schemes can be configured to be equivalent. And this episode explains the feedforward part just the way I need.
Two things that might help people understand this better. (i) Why is it 0=u*Gp+D* Gd. The confusion is because the u in this equation is not the same as input. In actuality the u reffered here is actually u'. The actual equation is: Y(s)=Gp*u+Gd*D+Gd*u' Assuming your plant dynamics are tracking Y(s)=Gp(u) Gd*u'+Gp*D=0 In other words add a perturbation to your input u called u' that perfectly cancels the disturbance response. (ii) In the end example, Brian refers to the Simulink controller as feedforward or non-feedforward. It is important to note that the feedforward only has disturbance rejection feedforward and not the setpoint feedforward he originally discussed in the video.
disclaimer(i might be wrong here in which case pls correct me ). think of a scenario where there is an output with some undesired disturbance value added in it. So to remove this particular value the change in output has to be zero. so y(s)=0 means the change in Y(s)=0 and NOT the absolute value of output Y(s)=0. so u need ur system to produce this ==original control input u(s) + an extra value (which u got after putting y(s)=0).This extra value is u(s)= d(s)*(-Gd(s)/G(p))). u multiply this with Gp(s) u get d(s)*(-Gd(s))(look at the diagram) which is exactly negative of disturbance. so adding with the disturbance(D(s)*Gd(s)) cancel each other out leaving behind the original desired y(s) corresponding to the input.
15:43 why we set "0=u*Gp+D* Gd" ?? if we do, we can't follow R. does't it? I think "R=y=u*Gp+D* Gd" is correct. doesn't it? if i am thinking wrong, enlight me
It's been a long time since you posted, but I thought I'd post in case others have the same question. This equation 0=U*GP+D*GD is really just a bit of shorthand for what he is really doing, specifically, choosing the feedforward controller by setting the transfer function Y/D to zero, which guarantees that the disturbance does not affect the output. Note that he models the disturbance as adding to the plant output (not to the plant input, which is used in some textbooks). The transfer function will be Y/D = (GFF*GP+GD)/(1+GC*GP*GS), where GFF, GD, GP, GC, and GS are transfer functions for the feedforward controller, the disturbance, the plant, the feedback controller, and the sensor, respectively. Setting this equal to zero and solving for GFF gives GFF=-GD/GP, which is the result in the video. Note that the transfer function GFF calculated this way may be improper and thus not realizable. In that case, an approximation of the transfer function would be used.
3 years later, here is the answer - you want to have the output = 0 in response to disturbance. Basically it means you want the disturbances to have no effect on the output.
You are such a good teacher. I hope you made these videos while i was in my undergraduate. Any ways, Brian I am designing a controller for a wind turbine that will predict the wind speed ahead and adjust the out put accordingly in order to control the frequency in the power system. I need some help regarding that. where can I reach you?
Hi, I have a simple plant transferfunction of 1/ms+b, a simple cruise control model for a school assignment. Now I have to use feedforward so I need the inverse of the plant transferfunction. This would be ms+b. But I cant make this in Simulink as the transferfunction block gives the error that the numerator has a higher order than denominator. Is there somebody here who can help me with this problem?
can I send u pictures of a block diagram? there are some inputs that I don't understand. The block diagram is a reference from Process Systems Analysis and Control by Donald R. Coughanowr Steven E. LeBlanc Third Edition
That just means you are changing the goal for your system. The set point is what you system is trying to achieve and when you change that, you're changing the set point. For example, if you set a thermostat to 18 degrees then that is the set point. The is the temperature the system is trying to maintain. If you then bump it up to 20 degrees you've changed the set point.
I'm new to control theory, could anyone please tell me why do we want the controller to ignore the noise from sensor? what do you mean by "it's not real state error"?
That eye tracking analogy was one of the most impressive teaching demonstrations I've ever experienced
What an intuitive way to explain different frequency disturbances, I loved it as well!
Hi everyone! I hope you enjoyed this video. If you have any questions or comments you'd like me to see, please leave them on this comment so that I get notified and can respond. Cheers!
Hi Brian, thanks for this Brilliant explanation of Feed forward control .. i only wished you make an extension to this video where you show also an example with the feedforward through the setpoint path .. and also another example focused on the noise filtering and decoupling the noise error from setpoint and disturbance errors and how this becomes easier with feed forward (what you explained in the beginning) .. really thanks man for all your effort!!
Thanks for the comment, Ayman. I would love to have covered those topics in this video as well but I have to balance getting in as much as possible versus making a video so long people won't sit through the whole thing! Maybe I can make a setpoint feedforward video that shows how noise is also decoupled on my channel as a suplement to this one. Cheers!
@@BrianBDouglas Thanks Brian .. i'll be waiting for it ;)
Sorry Brian i just watched the video again (really liked it) .. i am very interested on how to find the best causal transfer fn to approximate the inverse of the plants transfer fn .. would you happen to know some techniques for this that i can read about? Or to have already a video from your channel about this topic? .. thanks again and keep up the great work :D
This is an informative read on the topic: buchholz.hs-bremen.de/inversion/inversion_impossible.pdf. If the system has significant dead time (that is, it's long enough that you care about it), then you can approximate the dead time first with an LTI system using a Pade approximation.
The red dot was an amazing analogy
Agreed, one of the best
🔥🔥🔥
Very well done. I wish there was a feed forward controller on the youTube algorithm to increase the value to technical advertiser's for excellent technical content like yours.
This channel is gold. I am trying to make my Betaflight helicopter to dive with terminal speed exactly the same as CC3D controlled same craft. I look at both source code and couldn't understand if their PID schemes can be configured to be equivalent. And this episode explains the feedforward part just the way I need.
Brian's other channel is amazing!
Great to see that Matlab chooses from the best
Nice video. Ships using dynamic positioning actually uses feedforward from the wind measurement in order to counteract the gusts.
Thank you so much for this example. Helps me a lot.
Great explanation! This guy is a gold mine for control systems education.
Thanks!
This video is incredibly enlightening and helpful for me. Thank you, Brian!
I'm glad you found it helpful!
Thanks for the brief and easy-to-understand explanation of the feed forward concept. This is going to help my next project.
This is pure gold. Thank You!
Incredibly clear and useful. Thanks!
Brian Douglas; as great as always.
Haven't seen a better description of noise filtration than the red-dot analogy.
Two things that might help people understand this better.
(i) Why is it
0=u*Gp+D* Gd.
The confusion is because the u in this equation is not the same as input.
In actuality the u reffered here is actually u'.
The actual equation is:
Y(s)=Gp*u+Gd*D+Gd*u'
Assuming your plant dynamics are tracking Y(s)=Gp(u)
Gd*u'+Gp*D=0
In other words add a perturbation to your input u called u' that perfectly cancels the disturbance response. (ii) In the end example, Brian refers to the Simulink controller as feedforward or non-feedforward.
It is important to note that the feedforward only has disturbance rejection feedforward and not the setpoint feedforward he originally discussed in the video.
Thanks bro! I was wondering the same.
Really good video! Thank you Brian!
I noticed that you mentioned setting Y(s) = 0.Did you mean setting input-Y(s)=0 instead since we want the error to go to zero and not the output?
disclaimer(i might be wrong here in which case pls correct me ).
think of a scenario where there is an output with some undesired disturbance value added in it. So to remove this particular value the change in output has to be zero. so y(s)=0 means the change in Y(s)=0 and NOT the absolute value of output Y(s)=0. so u need ur system to produce this ==original control input u(s) + an extra value (which u got after putting y(s)=0).This extra value is u(s)= d(s)*(-Gd(s)/G(p))). u multiply this with Gp(s) u get d(s)*(-Gd(s))(look at the diagram) which is exactly negative of disturbance. so adding with the disturbance(D(s)*Gd(s)) cancel each other out leaving behind the original desired y(s) corresponding to the input.
Thank you Sir.
15:43 why we set "0=u*Gp+D* Gd" ?? if we do, we can't follow R. does't it? I think "R=y=u*Gp+D* Gd" is correct. doesn't it?
if i am thinking wrong, enlight me
It's been a long time since you posted, but I thought I'd post in case others have the same question. This equation 0=U*GP+D*GD is really just a bit of shorthand for what he is really doing, specifically, choosing the feedforward controller by setting the transfer function Y/D to zero, which guarantees that the disturbance does not affect the output. Note that he models the disturbance as adding to the plant output (not to the plant input, which is used in some textbooks). The transfer function will be Y/D = (GFF*GP+GD)/(1+GC*GP*GS), where GFF, GD, GP, GC, and GS are transfer functions for the feedforward controller, the disturbance, the plant, the feedback controller, and the sensor, respectively. Setting this equal to zero and solving for GFF gives GFF=-GD/GP, which is the result in the video. Note that the transfer function GFF calculated this way may be improper and thus not realizable. In that case, an approximation of the transfer function would be used.
Amazing video OMG many thanks
Is this the type of studd one learns in a control theory course?
Also, did you use perlin noise to implement the red dot animation?
Please explain why at about 10:00 why y[s] shall be set to 0? y[s] shall be equal to setpoint???
3 years later, here is the answer - you want to have the output = 0 in response to disturbance. Basically it means you want the disturbances to have no effect on the output.
How can we just set Y[s] = 0 at 10:00? Isn't our goal to set E[s] = 0?
exactly, couldn't understand y[s]=0??????
Great! Thanks a lot!
same if this (matlab link/coding) is given for the digital electronics it'd be moreeee appreciated and helpful. Thanks
You are such a good teacher. I hope you made these videos while i was in my undergraduate. Any ways, Brian I am designing a controller for a wind turbine that will predict the wind speed ahead and adjust the out put accordingly in order to control the frequency in the power system.
I need some help regarding that. where can I reach you?
Hi, I have a simple plant transferfunction of 1/ms+b, a simple cruise control model for a school assignment. Now I have to use feedforward so I need the inverse of the plant transferfunction. This would be ms+b. But I cant make this in Simulink as the transferfunction block gives the error that the numerator has a higher order than denominator. Is there somebody here who can help me with this problem?
can I send u pictures of a block diagram? there are some inputs that I don't understand. The block diagram is a reference from Process Systems
Analysis and Control by Donald R. Coughanowr Steven E. LeBlanc Third Edition
would do video on fuzzy PID controller in mat lab
Your matlap video is awesome. But I can't solve a question about feedforward. Can you help me share it here? @Brian Douglas
Great ..
Isn't feed-forward just model-based control?
what is a set point change?
That just means you are changing the goal for your system. The set point is what you system is trying to achieve and when you change that, you're changing the set point. For example, if you set a thermostat to 18 degrees then that is the set point. The is the temperature the system is trying to maintain. If you then bump it up to 20 degrees you've changed the set point.
3:54
But I followed the red dot with ease. Does that mean I'm the cat of the controllers?
I'm new to control theory, could anyone please tell me why do we want the controller to ignore the noise from sensor? what do you mean by "it's not real state error"?
Because your sensor measures [real error] + [noise]
The noise that we want to ignore affects the sensor, not the process we want to control