Want to learn about industrial automation? Go here: www.realpars.com/individual-pricing Want to train your team in industrial automation? Go here: www.realpars.com/pricing-team
I use PID with some fuzzy logic: if the error is close to ideal, I turn off the D part of PID. The result is incredible, the noise is very low. Also, being the I accumulator a measure of the system loses (drift, heat loss, etc.), I preload the I accumulator value with the P value as soon as it is not saturated, and periodically. If I have historical data, I preload the I accumulator at start with the typical value of I fir that setpoint. That way I save time loading the I accumulator, and I reach the setpoint faster. With these two techniques combined my controllers behaves beautifully.
Wow, very nice ideas. I remember back before Kinetix 5xxx brought decent default tuning and adaptive tuning. We altered the PID values during operation when we knew the load was changing. Tell me, what kind of processes are you tuning that made you develop your techniques?
Howdy. Yeah. During my days as a training planner in a paper mill. Operator experience. Shifting to another paper grade only changing the main set value would take almost a whole reel if the PID:s were left to do the adjustments. One whole reel to be repulped. The experienced operators knew the output values of the most important PID:s. They manually set them close to the final values. Then they input the new grade to the master controller and switched back to automatic. The new grade was achieved in about an hour. The shifting reel only needed to be rerolled cutting out the transition part. Enormous time saving and way less to be repulped. Regards.
Recommend looking into PDF (psuedo derivative feedback) where the derivative term is replaced leveraging feedback and the integral. Limiting windup on the integral accumulator is also important. Theoretical results show you can get faster convergence with little or no overshoot which can be very important for some control situations.
The shocks in your car's suspension can be thought of as PID controllers: the spring is the proportional part, the damper is the derivative part, and your manual height setting is the integral part. The shocks in the suspension stabilize the ride height of the car against the constantly changing height of the ground (setpoint) while allowing minimal overshoot and oscillation.
There is also another type of bang-bang control, called "Take Half Back" by Steven Woodward. It offers settling time comparable to PID and useful for on/off systems like AC or home heaters.
I love this refresher. I am not working in controls but I loved the study. Where are my eigenvalues and eigenvector? Does anybody know root locus or Bode diagram anymore?
MPC is indeed a combination between feedback and feedforward. It re-calculates the optimal controller output, so every time-step it takes into account disturbances that may have arisen -> feedback. The fact that it is able to predict the future given the a-priori known reference is why it is also feedforward.
I wonder will you have videos about advanced applications for PID control? For example these applications are commonly used in most DCS: split range, gain scheduling, ratio control, cascade control, and feed forward.
Hi @David_Bruton, Thanks for reaching out. We have not yet developed any videos on advanced applications for PID Control. Stay tuned though as we are planning more videos and courses on such topics.
Long story short, a properly tuned PID controller IS the best solution, when feasible; if not, due to changing dynamics or excessive degrees of freedom, an MPC wil hack together a solution that is "good enough"
It is always a trade-off between complexity and performance. A PID controller is simple to implement for the industry, like he said it basically is just 3 tuning knobs. However, classical (frequency domain) control strategies go way beyond a PID controller called "Loop shaping". This allows you to combine all sort of filters like for example notch filters to suppress resonant process dynamics and low-pass filters to suppress noise. Lastly, MPC is a so called optimal control strategy, it finds the optimal controller output given future predictions, which is something that a feedback controller like a PID-controller is never able to do. Therefore, if you put them side by side, an MPC is superior since it is able to counteract to disturbances outside of your control (feedback) by recalculating the optimization problem every time step, as well as counteract for reference-based disturbances which you know beforehand (feedforward). However, if the MPC fails on site, your average machine operator won't be able to fix it because you need to know what you're doing to be able to tune an MPC-based controller and prove stability is more involved.
MPC consistently outperforms PID, so it's objectively the best solution. However, it has an Achilles heel if the model is more than 10% off reality. The PID is good enough and should be the default control. In my experience, the feature I like best in MPC is that it has programmable limits. Most real systems have limits, and it's hard to optimize PID when there are real-world limits.
Very interesting the FLC part, but to which extent the fuzzy logic gradation is like the proportional part of the PID? Or does it add a quadratic value?
Thanks for your question. Fuzzy logic doesn’t actually add directly to the proportional part of the PID response. FLC helps in determining the proportional, integral, and derivative gains dynamically by using linguistic variables, fuzzy sets, and a set of rules
I wonder if any of you-all have seen the introduction of pseudorandom inputs into a system that allows a kind of real-time model building, where the pseudorandom inputs are small relative to the setpoint?
its always funny at work when I see the integrator and they some pid loops to ne tuned and you have 5 guys telling the programmer what you put as p i d😅
Thank you for your topic suggestion, I will happily go ahead and forward this to our course developers. Thank you very much for sharing, and happy learning!
Hi there, thanks for reaching out. Sorry, we can’t answer that question definitively as Boston Dynamics incorporates a combination of control methods, some may not be publicly disclosed. For availalbe information, refer to Boston Dynamics' website at bostondynamics.com
@@VK-tv1eq "sounds like a Kalman filter" - there you got off track a bit. The MPC is way more general, because everywhere the video says the word "algorithm" it can compute many more complex functions than multiply or integrate or derive. Higher order derivatives? Sure. (No they aren't D, they are functions that you can't generally get by tuning the gain on D.) Shortest path algorithms? Sure. All kinds of things implied under MPC. Turing-complete computations. Examples: figure the next engine gimbal for Starship flip, balance a double inverted pendulum, return a ping pong ball using three drones carrying a net, ...
Howdy. Great. There are systems that use two PID:s in a Master - Slave configuration. 1. A servo valve has a built in flow sensor and a built in PID controller. The servo valve's Set Value is the desired flow. For regular control valves the Set Value is the throttle position. The master PID is an algorithm in the plant automation computer. The master PID sends the desired flow to the servo valve. 2. The master PID sends varying P, I, and D variable values to the slave PID. The master PID senses the diffrence between Set Value and Process Value. If the difference is large the master sends aggressive P, I, D values to the slave PID. Small Proportional value (=large k=large gain), large D value and small I value. When the Process Value comes close to the Set Value the master sends mild P, I, D values to the slave. Large Proportional value (=small k= low gain), small D value and large I value for smooth approach of Process value to the Set value. The system will provide fast approach yet with no overshooting when tuned to optimum. It much resembles Fuzzy control. Regards.
Good stuff! Thanks for sharing...... I've heard of this as well. Bottom line - The Master controller provides setpoint values, and the Slave controller adjusts the actual process to match these setpoints.
the fact that PID and fuzzy logic is used means that the control world has failed at creating good algorithms because these are the most basic algorithms based on zero theory
Yea furnaces cuntrull, in other videos that touch on that subject it allways struck me as strange the the furnace wos either on or if an not controlled in a more fine graned manner , well I 'm no hvac engineer and certainly no gas fitter, but why can't ng furnace have it 's gas supply controlled ( thus controlling the ammount of heat actually beeing created) is it justba question of cost or is my lack of knowlage sowing wet clearly?
Hiya...... I'm not an HVAC SME or a gas fitter either. Your question is reasonable indeed. Maybe you'll get a response from somebody in the know. In many large industrial systems, the temperature is controlled by duct louvres and vanes that move when commanded. Also not a terribly efficient way to control. Does anybody else want to chime in?
I think that typical home heating systems have, historically, been designed to run at an "optimum" level of "burn", and have quite simple fixed jetting and air supply... It's quite possible to run a system that has enough "mass" with proportional control, for a smoother response, and even put in an offset value to make the output appear to be the commanded one, especially where the expected target temperature is normally in a quite narrow range.
Want to learn about industrial automation? Go here: www.realpars.com/individual-pricing
Want to train your team in industrial automation? Go here: www.realpars.com/pricing-team
I use PID with some fuzzy logic: if the error is close to ideal, I turn off the D part of PID. The result is incredible, the noise is very low. Also, being the I accumulator a measure of the system loses (drift, heat loss, etc.), I preload the I accumulator value with the P value as soon as it is not saturated, and periodically. If I have historical data, I preload the I accumulator at start with the typical value of I fir that setpoint. That way I save time loading the I accumulator, and I reach the setpoint faster. With these two techniques combined my controllers behaves beautifully.
Wow, very nice ideas.
I remember back before Kinetix 5xxx brought decent default tuning and adaptive tuning. We altered the PID values during operation when we knew the load was changing.
Tell me, what kind of processes are you tuning that made you develop your techniques?
Interesting..... Thanks for sharing!
Howdy. Yeah.
During my days as a training planner in a paper mill. Operator experience.
Shifting to another paper grade only changing the main set value would take almost a whole reel if the PID:s were left to do the adjustments. One whole reel to be repulped.
The experienced operators knew the output values of the most important PID:s. They manually set them close to the final values. Then they input the new grade to the master controller and switched back to automatic. The new grade was achieved in about an hour. The shifting reel only needed to be rerolled cutting out the transition part. Enormous time saving and way less to be repulped.
Regards.
Recommend looking into PDF (psuedo derivative feedback) where the derivative term is replaced leveraging feedback and the integral. Limiting windup on the integral accumulator is also important.
Theoretical results show you can get faster convergence with little or no overshoot which can be very important for some control situations.
What about NON-Linear control. PID suck at that. Actually PID sucks at many processes. Because the world is not linear.
The shocks in your car's suspension can be thought of as PID controllers: the spring is the proportional part, the damper is the derivative part, and your manual height setting is the integral part. The shocks in the suspension stabilize the ride height of the car against the constantly changing height of the ground (setpoint) while allowing minimal overshoot and oscillation.
LQ-PID has worked for me quite well over the years
There is also another type of bang-bang control, called "Take Half Back" by Steven Woodward. It offers settling time comparable to PID and useful for on/off systems like AC or home heaters.
Hi there! Thanks for sharing. If I remember correctly, this concept was introduced over 20 years ago.
It's so rare to watch comparative control methods, thanks a lot!
You're very welcome!
Nice introductory to these concepts.
Glad you liked it! Thank you for sharing.
I love this refresher. I am not working in controls but I loved the study. Where are my eigenvalues and eigenvector? Does anybody know root locus or Bode diagram anymore?
A superb video!!! Thank you for the clear and easy to digest explanation of the various processes.
Glad it was helpful!
MPC sounds like feed forward control with extra steps.
MPC is indeed a combination between feedback and feedforward. It re-calculates the optimal controller output, so every time-step it takes into account disturbances that may have arisen -> feedback. The fact that it is able to predict the future given the a-priori known reference is why it is also feedforward.
The most important condition to use the MPC is having an accurate model. If the model has a low precision, the outcome would be disastrous.
I wonder will you have videos about advanced applications for PID control?
For example these applications are commonly used in most DCS: split range, gain scheduling, ratio control, cascade control, and feed forward.
Hi @David_Bruton, Thanks for reaching out. We have not yet developed any videos on advanced applications for PID Control. Stay tuned though as we are planning more videos and courses on such topics.
Long story short, a properly tuned PID controller IS the best solution, when feasible; if not, due to changing dynamics or excessive degrees of freedom, an MPC wil hack together a solution that is "good enough"
It is always a trade-off between complexity and performance. A PID controller is simple to implement for the industry, like he said it basically is just 3 tuning knobs. However, classical (frequency domain) control strategies go way beyond a PID controller called "Loop shaping". This allows you to combine all sort of filters like for example notch filters to suppress resonant process dynamics and low-pass filters to suppress noise. Lastly, MPC is a so called optimal control strategy, it finds the optimal controller output given future predictions, which is something that a feedback controller like a PID-controller is never able to do. Therefore, if you put them side by side, an MPC is superior since it is able to counteract to disturbances outside of your control (feedback) by recalculating the optimization problem every time step, as well as counteract for reference-based disturbances which you know beforehand (feedforward). However, if the MPC fails on site, your average machine operator won't be able to fix it because you need to know what you're doing to be able to tune an MPC-based controller and prove stability is more involved.
MPC consistently outperforms PID, so it's objectively the best solution. However, it has an Achilles heel if the model is more than 10% off reality.
The PID is good enough and should be the default control.
In my experience, the feature I like best in MPC is that it has programmable limits. Most real systems have limits, and it's hard to optimize PID when there are real-world limits.
Excellent and informative presentation. Thank you.
Glad you enjoyed it!
Thanks for the refresher video. It would be nice to make a video about Adaptive control.
You're very welcome! Thank you for the topic suggestion, I will happily go ahead and add this to the list.
Happy learning!
Very interesting the FLC part, but to which extent the fuzzy logic gradation is like the proportional part of the PID? Or does it add a quadratic value?
Thanks for your question. Fuzzy logic doesn’t actually add directly to the proportional part of the PID response. FLC helps in determining the proportional, integral, and derivative gains dynamically by using linguistic variables, fuzzy sets, and a set of rules
Very nice! Excelent Explanation! Thanks!
Glad you enjoyed it!
Very helpful. Thanks
Glad to hear that, happy learning!
Always good and nice video quality, also i hope the teams talk about linear matrix inequality LMI it quite new automation control type
Thank you very much for your kind support! I will happily go ahead and add that topic to the list.
Excellent RealPars from Brazil.
Many thanks!
I wonder if any of you-all have seen the introduction of pseudorandom inputs into a system that allows a kind of real-time model building, where the pseudorandom inputs are small relative to the setpoint?
Hi there. That's a new one for me. Thanks for sharing.
Amazing information of PID vs other control method..very helpful
Glad you liked it! Thank you for sharing.
its always funny at work when I see the integrator and they some pid loops to ne tuned and you have 5 guys telling the programmer what you put as p i d😅
there is sliding mode control too..
Please make a tutorial on GE FANUC PLC
Thank you for your topic suggestion, I will happily go ahead and forward this to our course developers.
Thank you very much for sharing, and happy learning!
That's great and brilliant 👏
Glad you like it! Thank you for sharing
Which control methods used in Boston Dynamics robots? example Boston Dynamics Atlas humanoid robot?
Hi there, thanks for reaching out. Sorry, we can’t answer that question definitively as Boston Dynamics incorporates a combination of control methods, some may not be publicly disclosed. For availalbe information, refer to Boston Dynamics' website at bostondynamics.com
Thanks!
You're very welcome!
Thankyou for the knowledge
Our pleasure, happy learning!
"Wait it’s all PID?"
"Always has been"
@@VK-tv1eq "sounds like a Kalman filter" - there you got off track a bit. The MPC is way more general, because everywhere the video says the word "algorithm" it can compute many more complex functions than multiply or integrate or derive. Higher order derivatives? Sure. (No they aren't D, they are functions that you can't generally get by tuning the gain on D.) Shortest path algorithms? Sure. All kinds of things implied under MPC. Turing-complete computations. Examples: figure the next engine gimbal for Starship flip, balance a double inverted pendulum, return a ping pong ball using three drones carrying a net, ...
Howdy. Great.
There are systems that use two PID:s in a Master - Slave configuration.
1. A servo valve has a built in flow sensor and a built in PID controller. The servo valve's Set Value is the desired flow. For regular control valves the Set Value is the throttle position. The master PID is an algorithm in the plant automation computer. The master PID sends the desired flow to the servo valve.
2. The master PID sends varying P, I, and D variable values to the slave PID. The master PID senses the diffrence between Set Value and Process Value. If the difference is large the master sends aggressive P, I, D values to the slave PID. Small Proportional value (=large k=large gain), large D value and small I value.
When the Process Value comes close to the Set Value the master sends mild P, I, D values to the slave. Large Proportional value (=small k= low gain), small D value and large I value for smooth approach of Process value to the Set value.
The system will provide fast approach yet with no overshooting when tuned to optimum. It much resembles Fuzzy control.
Regards.
Good stuff! Thanks for sharing...... I've heard of this as well. Bottom line - The Master controller provides setpoint values, and the Slave controller adjusts the actual process to match these setpoints.
the fact that PID and fuzzy logic is used means that the control world has failed at creating good algorithms because these are the most basic algorithms based on zero theory
Yea furnaces cuntrull, in other videos that touch on that subject it allways struck me as strange the the furnace wos either on or if an not controlled in a more fine graned manner , well I 'm no hvac engineer and certainly no gas fitter, but why can't ng furnace have it 's gas supply controlled ( thus controlling the ammount of heat actually beeing created) is it justba question of cost or is my lack of knowlage sowing wet clearly?
Hiya...... I'm not an HVAC SME or a gas fitter either. Your question is reasonable indeed. Maybe you'll get a response from somebody in the know. In many large industrial systems, the temperature is controlled by duct louvres and vanes that move when commanded. Also not a terribly efficient way to control. Does anybody else want to chime in?
I think that typical home heating systems have, historically, been designed to run at an "optimum" level of "burn", and have quite simple fixed jetting and air supply... It's quite possible to run a system that has enough "mass" with proportional control, for a smoother response, and even put in an offset value to make the output appear to be the commanded one, especially where the expected target temperature is normally in a quite narrow range.
Thanks god bless
Always welcome!