Tank Level Tuning - The Knowledge Board

hi my name is Kevin Starr today we're going to talk about tank level tuning in our travels around the globe working in the process automation division we have found that tank level tunings oscillate matter of fact some statistics have shown that 95 percent of all industrial tank levels oscillate in some shape or form and that's why we want to spend the next few minutes showing you some very fundamental techniques that you can use immediately and stabilize your process what we're seeing here is there's two types of processes that we've talked about before is there as what we refer to as self-regulating those are the kinds where you make a change and you get a process nice smooth response and from the other videos that we've done we talk about the process gain and the time constant that define how much it goes and how long it goes those are self regulating things like flows pressures temperatures consistencies those types of things that's not all your processes that's about 80 to 90 percent which is why a lot of times guys are good at that but then we move into the non self-regulating okay so we'll just call them non self-regulating or what we're going to refer to as a tank in this case it's just that whenever I inject it in balance off it goes and that's why they're a little bit scary if you're doing a manual bump test you have to pay attention that's what we really recommend that you do a visual inspection you walk around you get an understanding of your process before you start doing a bump test so what we're going to do is talk about how do we identify the process how do we come up with a model and how do we come up with tuning rules to identify and to make sure that this loop is properly tuned now our process I'll just draw a little picture here of a tank as it has an inlet flow and here we've got let's say an outlet flow but in this case what we have is a controller that is usually a you know it's a fixed so this is our flow out so we have our flow out and our flow in so this is a fixed outlet flow maybe it's a drive maybe it's a control but it's a constant demand so in other words what we've done is we've broken the link between the head pressure and the flow so in this case what we have here is if we have a bump test let's say we have our flow in equals our flow out in this space we would say this is balanced and this would be our level so that's our level so when the flow in and flow out are the same the level doesn't move now what happens if I increase the inlet flow so the flow in is greater than the flow out well what's going to happen is your tank is going to start moving and now let's say at this point we drop the inlet flow back so once again the flow rate is equal to the flow out so we're back into a balance mode what's going to happen to the tank level OOP that's going to stop this is one of the ways I like to determine if I even have a self-regulating process on a self-regulating process if I put the actuator back to where it was the process will come back down so if I draw the rest of this you know we usually come back to where it started if we don't have stiction and hysteresis but in a tank if you put the balance condition your measurement will move based upon how long it was separated this is why we call this an integral type of a process so you can see it's integrating this difference or this imbalance that's why these are tricky so what's the process model how are you going to come up with model parameters you can't use the process gain at time kha'zix that never stops it never settles so our definition of process gain as it applies to self-regulating would virtually be infinite so we have to have a different model so what we do here in this case is we recognize that there are different slopes let's call that m0 m1 and m2 and let's call this you know out change in output 1 and change in output 2 so in this case what we're finding is that the process gain for a tank is equal to the change in slope over the change in the output so the slope how this changed versus that so that would be this process gain let's call that KP 1 and then maybe we do again for this one and say that KP 2 is equal to m2 minus m1 over our output - ok I like to do this these two process gains because I can also identify if KP 1 equals K P 2 then I have what I would call a linear tank you know tanks can come in all shapes and sizes you've seen them that look like a cylinder sometimes they're on the side so that they have a cross section that changes with the level sometimes they look like a comb so we're going to get into that in our non-linearity section in a later segment but right now we're assuming that we have linearity in our process gain but what I wanted to identify here is this slope change so if you remember slope slope is equal to rise over run which would it basically be the change in level over the change in time now this is where that's the slope this is where when we do this in classes and we do this in settings this is kind of hard to do especially if you have noise on your signal it's kind of hard to identify so I recommend that you do this a couple of times but there's a few things I want to point out here is it's the change in slope not the slope I've seen this messed up if if for example let's say that when I walk up to my tank it's doing this and then I do my bump test ok so this is let's just say this is M zero and this is M 1 and this is my change in output 1 I can't just say the slope is equal to that divided by that it won't give you the right answer you have to get the change in slope divided by the output change in slope divided by the outlet produced it now a couple things to remember here is your level most of the tank levels are normalized 0 to 100 but there are few there and feet or meters so you then have to normalize that and I would recommend that you look at your industrial books your technical notes to determine if the game needs to be normalized or if they have a factor to do that already but the time it also has units it can be seconds or minutes or hours what I would recommend is well no it's not recommended you need to set the time units on your slope equal to the integral time of your controller so if your controllers working in minutes you need to do your slope in minutes if it's in seconds it needs to work in seconds I was teaching a class once and a guy came flying in so you told us wrong and he calculated all this but he had messed up on his minutes and seconds so he was off by a factor of 60 on his tuning and the thing went crazy so this theory doesn't work that's something I want you the theory always works now your implementation of it might be an issue so what we have here is we're identifying our process we recognize that a large percentage of our industrial sites are self-regulating now we're talking about tank level tuning or non self-regulating where we have an imbalance we have an integration effect so we have to kind of a new term balance but we had to calculate the process gain now I always recommend that you link the process gain back to the physical characteristics of the tank and what's interesting here is that the process gain let's let's let me just start here with let's let's draw a picture and let's say this is my level zero and let's say this is 100% so this is my measured value and this is time okay so if I could and let's say this is my let's say this is my another one here let's say this is my flow in so here we've got this is time zero to 100 so what we're trying to do here to help you understand what this process gain is calculating is if my flow is zero I have nothing going into the tank all right now what we do is we take our flow take it to 100% so we have a hundred percent of our flow being pumped into our tank it was empty to start with we've shut the outlet valve off so we have an imbalance of a hundred percent it's the biggest we can get so what's gonna happen to this level and let's say we time it so that we drop this back to zero so what we're doing is we are filling the tank so we have filled the tank so let's call this a fill test so this time from here to here is what we call the fill time the tank fill time now I don't recommend that you go and do this don't say okay mister operator could you shut all your tanks off so I can fill them up and then time it that's the idea if you could could great but you can't do that so what we're trying to say is the fill time is really what we're looking at is if I have a imagine a like a garden hose you know that you're filling a bucket with you know it's going to take a while but now let's say you got this gigantic hose you know how long does it take to fill that bucket it's a function of the amount of volume of fluid going into it so the way I like to think of the process gain is it tells me the relationship between my input device and my tank size if I have a tank as big as this room and I've got a little garden hose this is going to take forever to fill up but if I have a tank that's you know the size of a glass it'll fill up really quick so you kind that's why we said one guy used to say the amount of time it takes to walk around the tank is real close to the fill time now again I don't necessarily recommend that you go walk around your tank but this concept that the fill time is what we're after well it turns out that the process gain that we get here this process gain which is the change in slope over the change in the output is equal to one over the fill time so the process gain and the fill time are related they're inverse of each other which that's where we started things like well maybe we don't even need to do a bump test what if if we have our tank here if I know the volume which is you know PI R squared times the height and here's my you know you've got your your your your level indicator and you got your tap so the heights the distance between the taps and then we have our inlet flow and we know that and we can stand we can identify the maximum flow so if I know the maximum flow and the volume of the tank I can calculate how long it takes to fill that so for example if this max flow is equal to 100 gallons per minute and let's say this volume is equal to a thousand gallons how long how many minutes is it going to take to fill it 10 minutes so this would have a process gain of 0.1 so that's how we do this is I always recommend that after you calculate your process gain time out and then look at it whether you use the slope tech technique or you use the volume technique to calculate the the fill time go and take a look at the tank and see if that makes sense I've run into a lot of cases where they messed up the fill time or they rest up the max they looked at the the span on the transmitter and they had had it calibrated wrong a lot of times this volume you can go look at the tank look at the diameter and you can calculate it but on a lot of industrial P&ID drawings they'll have the volume stamped right there on the drawing so I've gotten this this come up with my process gain and then I apply that directly into my tuning and it's it's incredible how fast you can tune a loop so if you could if you don't have this information then you have to do a bump test if you can if you have this information I find this to be way more accurate because the fill time is what you're after okay now that we have the process gain what we're interested in is coming up with the tuning so now that we've identified the process now we need to come up with tuning now what we're trying to do with this technique is if this is my set point right here and here's my tank and there's been a load change there's been a disturbance of some sort let's just say we had a disturbance so this is a load change what's gonna happen so what we're saying is somebody we're running along someone open the outlet flow what's going to happen to the level the level is going to come down and what we're going to want to do is so this is the flow out we're going to want the flow in to quickly come up to this point what happens when the flow is equal to the flow out the tank stops moving so what we do is we define this as the arrest rate all right so the arrest rate is there's a couple neat things about that that's where the flow in at this point right here the flow in is equal to the flow out in other words we've balanced it however if we stop there we resulted in offset so control has to keep working it has to have the inlet flow go above the outlet flow for a period of time to bring this thing back around and what's interesting and again we're over here we're balanced again where the flow in equals the flow out we're balanced it takes six arrest rates this is what we call a second-order critically damped tuning for a tank it's the fastest you can go recover from a disturbance without oscillating this is the goal this is what we're shooting for for you guys that like root locus I'll real quickly just sort of show you what we're doing here because we have a double integrator with our tank we got our PID control our proportional integral and our tank is an integral so we have a double integral we've got two poles so what happens is as you start increasing your game what happens and this is where most people mess up is they this is the region a lot of people will leave their control tuning which means your tendency to oscillate now I don't mean that if you don't understand root locus that's okay this plot sort of shows what happens to the tendency to oscillate as I increase the control gain so what happens is I start increasing the gain I start to oscillate what do we all do when we're in an oscillation you cut back on the game well you cut back on the game you just change the amplitude well if you increase the game it gets worse nobody by tuning by feel says oh if it's oscillating here let's just increase the gain a little bit further that's the reason you need to use the scientific techniques on how to tune a tank because what we're trying to do is to get over to this area right here which is what we call the region of stability which just happens to be one over the arrest rate okay so what are we doing here how do we calculate all this so when you plug all this math in to come up the tuning rule is actually pretty easy we say that the controller gain is equal to two over the process gain times the arrest rate and that the integral time is equal to two times the arrest rate this is the tuning rule for a PI standard algorithm not parallel or classical and we've talked about that and I also want to make sure that if you use if your if your levels normalize 0 to 100 you're good to go if it's in feet or meters then you may have to scale it so please look at your technical documentation but you know you're 2 over k PT RS or two times the arrest rate is your integral time this tuning technique will result in this type of a response that's so important that you go back and validate your tuning once you've done it cause a load change watch it stop and then watch it recover it's pretty neat to see and it almost looks like magic when you can tell a customer hey it's going to take six of these arrest rates to return now for some of you guys that are really interested is how far does this fall well if I know my load change this change in air the max deviation will equal the change in load almost you times see it's got it so it's going to equal KP TI over two times this e value so it's kind of tough to see is whatever your integral time is times your process gain which and then your load change so if you know that you had a 10% load change you can calculate how far this is going to go so I've had I've actually come up with tuning like that if I'm working with a system that say I can only allow this many inches of movement then I say well how big of a load change do I have once I know the load change this I can calculate my process gain and my integral time that I need to give me those results so this is another neat little equation that you can use once you use this technique now this is great but what people have asked is how do I calculate an arrest time and I'm like well that's a good question so how do we calculate an arrest time well when we did our self regulating we came up with a tower ratio to try to bias our tuning so that it was a function of our process what we've come up with is if we say that our let's come up with a new variable M and say that whatever our fill time is divided by our arrest rate in other words we can flip this around and say that our arrest rate is equal to the fill time divided by M so for example let's say if M is 1 well that would be fast or slow so let me make a little table here because everybody wants to know fast and slow for tanks and let's put this in thing here so if I say M is one that is really slow because that's saying that the fill time is equal to the arrest time that means it's going to empty before you even start to try to recover that's one of the things that I do when I'm analyzing tanks as I look at this relationship and I know that they tuned them so slow that any disturbance I have it's going to it's going to empty or overflow before we even come back you know maybe I like you know picking a five to ten so that we do is we say that our arrest time is five to ten times smaller than our fill time when we do that then when we plug this in you can say that our let me put our controller game here and our integral time here it's pretty neat if you take our tuning rules that we had over here for a rest rate your case C is just equal to well let me just do that here this when we this this results into two times M and T I equals it's two over t RS so you can put in its to fill time divided by M so now what I like about this is if you notice that that's one of the things with tank levels you just can't set the integral and move the P for control speed what happens is this circle literally changes size based upon the type of tuning that you want so what we have here is this M the speed is in both components so if we pick let's just say 5 what that would do is it would give me a gain of 10 and then our integral time would just be 2 times the fill time over 5 and then if we go down to 1 you know you would just say this would be 2 and then it would be 2 times the fill time and then what I do is I usually will build a couple of these maybe go with 3 and 2 and then I'll fill in my table and then I come up with the speed of response that is desirable that gives me the best deviation and the fastest recovery time without overshoot what we found is this process can actually go pretty fast is if you can calculate the volume of the tank and get the maximum flow you can calculate the process gain pretty quick and then what I like to do is just to kind of get a feel is if I take my controller gain times my process gain times my integral time when you use these tuning techniques this should equal 4 if it's on either side of 4 it's either going to take forever to get to the set point or it's going to oscillate around the set point so these are some techniques that we have found when we work with industrial oops remem ninety-five percent of all industrial oops Asli about ten percent of all your control loops our tanks tanks oscillating cause an imbalance in your plant whether it's due to mixing whether it's doing the header pressures consistency bonding there's a lot of issues with tanks that can cause problems when they're oscillating these techniques first identify the process whether you do a bump test to calculate the slope you look at the volume to calculate the fill time you use these tuning techniques that I'm showing you here this tuning rule is the rule for tank levels make sure that the integral time is a function of your you know you know your seconds if it's in seconds or minutes make sure that your slope and your time are lined up then these other things are just ways you can come up with an arrest time that is satisfactory for you and your client that's our lesson happy to you

Channel: ABB Process Automation

Views: 47,080

Keywords: Tank level tuning, Knowledge (Quotation Subject), knowledge vault, tuning, single loop control method, single loop, control tuning, kevin starr, abb service, process automation, service, westerville, ohio, tank, ABB, ABB Ltd (Business Operation), knowledge board, tune a loop, loop performance, big data, cloud computing, machine learning, data scientist, white box, grey box, black box, non deterministic modeling, prescriptive maintenance, proactive

Id: wd9_oC8js8U

Length: 22min 48sec (1368 seconds)

Published: Thu Dec 10 2015