Tuning A Control Loop - The Knowledge Board

hi my name is Kevin Starr I've been asked many times how do I tune the loop so what we're going to do is take about 15 minutes and we're going to give you the the bare minimum requirements in order to tune the loop the thing with automation if you if you've ever heard this idea of the tip of the iceberg what we're talking about is automation continues well under what you see what's interesting is at the very top is there's these three little numbers called PID and how do we set those up it seems very easy but it's a little bit like trying to balance bowling balls you know it's happy how do you do that that's what we're going to talk about here in just a few minutes recognize that there is a lot of information under the surface and if you want more detailed information we have books we have presentations I liked it so much I went back to school and got a master's in automation so what we're going to do is spend a little bit of time talk about how to tune the loop now what we have to do is we have to start off with this thing called a controller that controller it actually has an output and it will talk to I'm just going to draw a let's call that a valve that's a valve and this valve let's say it's there's some process here that we're controlling in this process it has some sort of a you know an i/o module or a transducer or something it's a sensor and that sensor produces some sort of a feedback signal okay so what we have is this controller that takes the measurement and we call that the measured value and we have an output which we call the output now what we need also is a reference so this is our reference or we call that our set point so our set point is what you want that's where I'm trying it like the room this temperature I want it to be it you know 68 degrees and if it gets too hot or too cold the furnace kicks on to adjust it so what we're looking at here is with our process we have a control sending an output to an actuation device that action device can be a valve it could be a drive it can be an actuator and you need to go and do a visual inspection so when you're walking up to a process do it let's just put a visual inspection that's what I like to do I was at one site they said I can you tune this Lewis well I need to go look at it and like this guy's going to fix our problem they're like why is that so because you're the first one to ask to go out and see it so I like if it's safe go out and have the operations or whoever the operators are taking you and show you the valve take a look at the condition of the valve what's it look like is it got an eye to P converter there's got to a p2i converter does it have a valve positioner does it have a cam is it rusty is it a single seated globe or a top hat or a rotary valve look at it is it rusty then you know you're gonna have a hard time then go over and look at the sensor or look at the device the transducer that's converting it from a measurement to a signal that can be transmitted back what's vintage is it is it old is it a mag flow meter is it a pitot tube is it an orifice plate is a differential pressure device try to figure out just take a look at it and kind of get an idea of what you're looking at because you're after the numbers in this are those P I and D you know we're going to talk about what those things are now the class of process that we're talking about there's there's several different things we could talk about flow pressure temperature consistency these types of processes are what we refer to as self-regulating that's different is then what's called a tank in this particular short talk we're just going to talk about self-regulating this represents these this list is probably 80 to 90% of all the industrial loops that you're going to run into so if we get this you've got yourself pretty well covered so the first thing you do is do a visual inspection look at the valve look at the sensor look the thing over second thing you want to do is do a bump test so what's a bump test but what we're going to do here is we're going to come down here and we want to inject energy we want to do a step change this is what's called a step and what we're going to do is we're going to we're going to have a change in this thing called output this little triangle is a change it's a delta it's the change in the output and what we're going to do is typically we do is we put the control in manual so in manual or you know we want to adjust the output so that we can inject a step there's a lot of restrictions depending on the industry on how big of the step you can make I always recommend start small and grow and get larger we also recommend a step cycle where if this is up if this is 1x this is 2x and then you come back up so what happens if the operator says you can do a 5% change you go up five down ten back five and then this is where you could do your identification on so what we're looking at is we do this bump test at the same time we want to record the measure value so we want to record this we want to find the relationship and that's what we call modeling we want to model model or it's called identification identification so we want to come up with the model parameters you have to capture the dynamics you have to be able to if I move this what happens and you have to repeat it and make sure that you understand what those parameters are once you have these parameters the next thing you do is you you come up with tuning alright and that's what you need to know is when you come up with the numbers for these P I and D so you do a visual inspection you do a bump test or more than one you do a model identification to calculate the parameters you use those parameters to come up with tuning and then I always like to stop at a validation step make sure validation is is do a closed loop set point change go and change the thing change the set point and see what happens you're the one in control you're the automation engineer you're the one that did all this thing you're the one that tuned it when you change the set point it out of work now let's quickly go through here the bump test we talked about now let's talk about modeling what we're talking about when we do a model is let's do a little block here we'll call this our process okay so here's where I made my step input and then we get this response this is the question mark how do we identify that what is it how do we find it well what if you can if you can trim this it's better if you can store the data in a text file and load it into an analysis tool or Excel or our loop tune but in the worst case you can do a bump test and then watch the process okay and what worked what I'm trying to show you here is a it's a simple first order model with a simple first order model what you're looking at and I'll put in some parameters here this is our change in our output this is our change in our process variable and this is our change in time remember that little triangle means Delta or change so with those three things that's what I like to think about for the seesee estate is how much did it move for given change and how long does it take to get there so we're looking at two things is the process game is defined as case of P which is equal to the change in measured value I'll see that let me change that to PV over the change in the output that's simply what it is this is answering the question is how how much did it move how much did it move that's that's the first thing you can get that pretty easy just take a look at the change and how you change an output over the change changing process over the change in output second thing you need is a time constant that's things that this funny-looking symbol tau sub P tau is a Greek symbol that universally represents the time constant tau so P can be approximated as the change in time over the four so it's the change in time over for approximately now there's a couple ways to do this you can sort of take this and divide it into fourths you know or you can say okay take this total time and sort of divide it into thirds sure enough right there it is you know that's roughly 63 0.2% of where it's going to go so how long did it take is one time constant so 1 2 3 4 you're pretty much there so in this you can do this with a Timex your watch it's hard because in a real world you've actually got a process that may be wiggling and jiggling it may you know kind of doing this type of stuff but no matter what it'll stop so what I like to do is sort of draw a first order model on my response look as far out as I can to get the process gain and then see how long it takes to get there to get my change in time that's my time constant these two parameters this and this are crucial you have to make sure you've captured them you make sure you get those so do another bump test and see if these change every time then you've got a actuator problem or a process problem or a non-linearity issue that you have to investigate if you can't predict what's going to happen when you change this output then this controller has no chance of getting it right so make sure that after you've done the visual inspection after you've done the bump test take a look at the model and make sure it's repeatable now we've captured the dynamics now we're ready to do tuning so now we come over here and let's talk about tuning first thing we have to do is recognize that not all PID algorithms are the same there's actually three classes there standard parallel and classical they have to do with how the algebra behind the scenes works and I'm not going to go into that right now we're going to work with this one right here and it has a it has a thing looks like this one plus one over this is what it looks like so I would recommend you go and take a look at your technical reference for your particular vendor and you'll see a mathematical expression I'll just show you the parallel one is going to be this plus ki over s plus KD s I guess the reason I like this what does these happen to be my initials KD s now and then classical looks like this it takes KC one plus one over T is and then it multiplies by one plus TD s so what I recommend is I know this is a lot of math but this is what the control modes look like and I always have to go and look in the reference guides for our particular vendors and whether you work I won't name all the vendors but not everybody calls everything the same thing some it's an interval time some it's an integral gain some it's a proportional band controller gain derivative time there's a derivative filter all those different different caveat but if you look at the equation look for the parentheses and then that will tell you what type you have for the analysis that I'm going to use I'm going to assume the standard algorithm I always my brain kind of was tied to that it's not that it's that difficult to convert but I always work towards standard and then I convert it to the parallel or classical forms one other little hint is if the derivative time is set to zero notice that these are identical so you may have a classical or a standard but if you're not using derivative these two are the same so what I wanted to just point out is I've seen too many places where they'll change a dcs system and they'll go for parallel to standard or classical and the guys that tuned by feel they these numbers can't work if you just take this number and stick it here it's not going to work actually it could blow up so you have to recognize what kind of controller you have but once you have it then then you get into the the tuning rule and there's all different theories on to do you want to first order response a second order response you want to have Ziegler Nichols you want to have what do you want but I think for what we're talking about here is I want a first order closed loop response you're the guy that's doing the tuning you set it so what we're saying here is if I change the set point I change the set point I want my measured value to come up in for closed loop time constants so here's a question for you notice that the change in measured value equals the change in set point so what's the process gain in the closed loop form so here we're changing the set point go back over here this is work just so you were at is we're now injecting energy into my set point the loop is in Auto and I'm changing the set point over here our loop was in manual and we change the output this is called a manual bump test here we're doing a closed loop bum test and this is the set point so this the whole thing becomes live and that's what we're after here so what we're saying is when I change the set point I want the measured value to respond in this particular shape so in this case the closed loop gain is the change in measured value over the change in set point which is equal to one and the well the time constant is equal to the time constant it's this total time divided by four so that's what it is so what is the tuning rule to do that there's a lot of math remember underneath the surface but the control gain is equal to one over the process gain times this thing called tau ratio and the integral time is equal to the process time constant and the derivative time is equal to zero this is the magic numbers that you have to use so notice that the tuning is a function of your model parameters and this thing we're calling tau my blue markers dying is called tau ratio tau ratio defines speed and let me just take a second and tell you what that means tau ratio is defined as the closed loop time constant divided by the open loop time constant so people will talk about fast or slow and I don't like to do fast or slow when you use tau ratio it balances you this is fixed this is your process so when you say fast or slow it has to be as a function of what's real so I like to use tau ratio of let's say if we make a table let's just make a table tau ratio and let's talk about speed of response if we say 1 that's what I would call very fast and then we'll go with 2 3 4 4 is very slow this is safe and this is a I would say marginal Mart it's just fast faster fast so your speed of response so this tile ratio is really just a knob you decide how fast you want this to run so if you had a six second open-loop time constant in a tower ratio of two you would have a 12 second closed-loop time constant so for example let's let's do an example now let's say we we did our inspection and looks good we did our bump test manual looks good we did our model and we got a process gain of let's say 2.0 and we say a time constant of 6 looks good and we could get that every single time let's say that we want a our tuning and we want a let's say we want to let's say it's aggressive let's say we want to go safe I always recommend start with a safe so what we would say is that our controller gain will equal 1 over KP tau ratio so that's going to be 1 over 2 times our towel ratio of 3 so that's going to equal 1 over 6 for our control gain and our integral time is just going to equal 6 boom so those would be your tuning numbers so then the validation value do the validation validation is in closed loop you change the setpoint what should happen well your process if I can get a little bit of life out of my pin it will start to come up roof how long should it take to get there well it should take for closed loop time constants but we said that we had a tower ratio of 3 so what would be my closed-loop time constant it should equal my towel ratio times my process time constant so I prompt this so I've got a time constant of 6 tower AC o 3 so that's equal to what is that 18 should have picked a little easier numbers so my close loop is 18 so third 72 seconds is how long it should have taken to get there because I have a closed loop time constant of 18 and it should go whatever I told the setpoint to do the other neat little thing is the output this this is the total change needed this is the change in the output needed this first step this first kick if you will will equal the total output needed divided by your towel ratio so in this case it would be 1/3 so that would be so the result would be the kick will be 1/3 of final Delta so I almost got that drawn right so this would be 1/3 and then it gets near the rest of the way so with this you can actually come up very and then you can walk away you know this thing's - so we just went through the whole step here on the board of talking about the process modeling tuning how to pick speeds and then how to validate it you do this you'll have a loop that will run very very well will be stable it won't be calling you in the middle of the night and if something does break it's the theory that does not break that means something broke the the assumptions weren't right the models changed the actuator got stuck but you will have a paper trail that you can record do a bump test and you'll see what changed that's how you do tuning and that's how you tune a self-regulating loops to have a first-order closed loop response the video you just watched literally represents the tip of the iceberg in the space we call process automation if you'll give me just one more minute I'm going to show you where you can get more information on this topic we've actually written a book called single loop control methods we have a training course that you can go to ABB University to learn more about through formal instruction we have a virtual environment version of the same course that you can take at your own pace and we not only have the knowledge that we're making available to you on the lessons that we've learned over 50 years of automation we have tools and services that can that you can use to improve your process one of them is called loop tune loop tune is a package that you can run on your computer that will collect data perform you can do the bump test and you can analyze the things that we looked at in the video and have a report and have a record and do simulation before you actually try it in the real process that's called luke tune this box I'm standing beside is called service port it comes in several different form factors but it is designed to plug into your automation system and evaluate the performance of these control loops so that you can quickly identify problems and fix them so that you can have increased improvement of your process for more information go to the link in the description below

Channel: ABB Process Automation

Views: 347,399

Rating: 4.9021406 out of 5

Keywords: Control System, Loop Performance, Tuning a control loop, control loop, tuning, tune, loop tune, serviceport, ABB Ltd (Business Operation), process automation, kevin starr, service, abb service, PID control tuning, PID, PID tuning, big data, cloud computing, machine learning, data scientist, white box, grey box, black box, non deterministic modeling, prescriptive maintenance, proactive

Id: 3viD5ij60EI

Length: 21min 50sec (1310 seconds)

Published: Fri Oct 31 2014