Parallel Pumps 7

[5marc5]

Hello Everyone,

Have done my due deligence and researched books and online resources (including this forum), but it seems to me that there is a bit of information lacking in this subject. I don't have an immediate fellow engineer with expertise in this field to direct my questions to, but have asked other veteran engineers and it seems that there are various takes on the subject; some engineers think that it is ok to simply double the flow while maintaining the same head and others think that there should be an increase in head and the flow would be less than twice as much as only one pump? To me it makes more sense that indeed the head would increase with the increase of flow and that the flow (although increases) will not reach twice as much as for one pump due to the headlosses. Having said that and understanding that concept, I still don't fully understand (and have not found a book, another fellow engineer, or a pump supplier, or anyone) to explain how the system curve is developed considering equal capacity flow pumps in parallel. If someone can finally put this question to rest (first, they should be teaching the subject and second should write a book so we can reference from this point on)?

This is my first post, although have searched and viewed other posts and have been very helpful. If I have done any wrong doing with the format of my post or requires any corrections, please let me know.

Thanks in advance,

MR

 

[JJPellin]

A single pump will operate at the intersection of the pump curve and the system curve. Starting up a second pump in parallel with the first one does not change the system curve. The system curve is a characteristic of the system, and is unchanged by changes in the pump or pumps.

Making a change to a single pump such as a speed change or impeller diameter change will change the pump curve. The operating point will still occur at the intersection of the new pump curve with the unchanged system curve.

It is possible for a system curve to be almost flat. If the piping is large and the velocity is low, the pressure drop at increasing flow may be very little. But, it is never perfectly flat. It has some slope.

When you start up the second pump in parallel, you need to generate a new curve for the combination of the two pumps. But, a few definitions might be in order. The pumps are considered to be in parallel if they share a common suction and a common discharge. There is no imposed pressure drop between the two pumps in the suction and discharge piping. They are fairly close together. Based on you question, I assume you are interested in matched pairs of identical pumps.

The two parallel pumps produce a combined curve that is additive in the direction of flow. There is nothing that would allow either pump to produce more pressure at a given flow than it would otherwise. But, a parallel flow path produces the possibility of adding flow.

The two pumps will operate at the intersection of this new combined pump curve with the system curve. Since the system curve must have some slope, I can only say one thing for certain about the flow. It cannot be twice the flow from a single pump. If the system curve is very steep, the increase in flow could be very small.

I have attached a sample curve showing the curves for one pump and two pumps with the system curve imposed over these. In this example the single pump would produce about 900 gpm. Two pumps running would only increase the flow to about 1025 gpm. With two pumps, the head pressure will be much higher. But, it still must fall on the system curve.


Johnny Pellin

 

[LittleInch]

5marc5, JJpellin above makes some excellent points and the graph attached is exactly the same as I show on set of lectures I do for engineers to demonstrate the fact that centrifugal pumps don't come in "lots", i.e. two pumps, both rated at, say, 250 m3/hr, when coupled in parralel won't add up to 500m3/hr. The only thing to add is that positive displaccement pumps (piston, screw, progressive cavity, gear, axial piston etc) work differently and in those cases, providing that the system pressure is less than their max pressure, then two pumps in parralell will actually be double the flow. However the vast majority of pumps you come across are likly to be centrifugal, but you do need to know which type of pump you are talking about.

Exactly how much an addiitonal identical pump will add to your system will depend on your system as ably explained by JJ above.

The key is developing your system curve to overlay onto the pump curves for one or two pumps. The start point (0 flow) is any static head which exists at zero flow, either a fixed pressure you need to overcome (e.g. when connected ot a fixed pressure other system such as a boiler or separator)or height difference or high point along your pipeline.

The curve part is then the friction losses along your pipe which vary in proportion to the square of the flowrate. The steepness of the curve is dependant on the size of the pipe, viscosity and desnity of the product and the actual velocity in the pipeline ( a big pipe with low flow will be flatter than a small pipe with higher velocity).

Although it refers to slighty different things, Also see this post as it covers a lot of the same issues.

http://www.eng-tips.com/viewthread.cfm?qid=343261

My motto: Learn something new every day

Also: There's usually a good reason why everyone does it that way

 

[Artisi]

To break it down to simple terms, which I always find to be the best solution is to look at in terms of 3 components.
This has already been discussed by JP and show on the curve posted.

The components are,
1. 1 pump operating
2. 2 pumps operating (assuming that you are talking about identical pumps)
3. the system curve.

1 and 2 is simply the sum of the two flows at the same head and a fixed know, with 3 being the calculated total head imposed by the system at varying flow rates with the head increasing as flow increases ideally shown by JP'c curves.
On the JP curve you can see as flow increases so does total head until it intersects the HQ curve of the 2 pumps in parallel, this becomes the operating point for the 2 pumps, ie half the flow from each pump contributing to the total.

Now this is an ideal situation and rarely seem in practice as no 2 pumps will ever perform the same, but in such a system the 2 pumps (or 3 or 4 or 5 depending on the installation) will settle at a point on their own performance curve and operate at the head imposed in conjunction with the system curve - head will the same for both pumps however, flow may vary between pumps.



It is a capital mistake to theorise before one has data. Insensibly one begins to twist facts to suit theories, instead of theories to suit facts. (Sherlock Holmes - A Scandal in Bohemia.)

 

[5marc5]

First of all thanks for all responses, I really do appreciate it.

JJPellin, you are absolutely right about definitions being in order, while parallel pumps are connected from both the suction and discharge, what would be the correct terminology for pumps that are only connected at the discharge and not the suction? In my experience it has always been referred to as parallel pump systems even though the pumps are not connected at the suction.

Maybe I did not ask the right question, but my concern is when developing the system curve. Say for example: You have a pump station with three "identical" pumps rated at 2,500 gpm, two pumps will be providing the capacity requirements for the system while the third pump is standby. My thought when calculating the headlosses is that I know that the suction line of each pump will ONLY see a max flow of 2,500 gpm and then the discharge of each individual pump will also only see a max flow of 2,500; but when two pumps are on, the header pipe will see the "combined" flow of 5,000 gpm. So when I'm doing my headloss calc's it seems wrong to assume that each pumps individual piping will ever see more flow than the 2,500 gpms each pump is rated for? Say you are developing the system curve for that system, and you run multiple flows (ex. 750, 1500, 2500, 3500, 5000, and 6000 gpm's), calculating headlosses through the suction and discharge line of each pump for the flows higher than 2500 gpm would create additional head whithin the suction and discharge that the system would really never see until the header portion of the pipe line. If that is incorrect, what is the process thinking behind that?

Thanks.

 

[LittleInch]

Pumps connected to differnet suction lines are not parralell in the way noted above. Even small differences in suctio presusre can make "identical" pumps run at different flows and in extreme conditions run at 70/30 or 80/20 in ratio terms. Design like that is a bad move. Usually you try and make sure even as near identical flow paths are used.

Normally the head loss in the suction and discharge piping is negligible in the overall system curve calcualtion so your issues can be ignored. however as noted above, what you need to do is make the overall flow path through each pump as equal to each other as posisble to prevent uneven running. usually you would bring the suction lines in from one side of the pumps and then take the discharge off the otther side so that from the common flow point in the scution lines to the common flow point in the discharge lines, the length, no of bends, valves etc is a simialr as possible regardless of which pump it goes through.

Only if your system has actually been designed to have two pumps working will the pumps "rated" flow be 2,500 gpm in your example. If only one pump was operating then flow would be much more than 2,500 and may even go off end of curve, similarly adding the third pump would not increase it to 7,500, but something much less than this.


My motto: Learn something new every day

Also: There's usually a good reason why everyone does it that way

 

[JJPellin]

You have given us a new set of information that is not what you originally described. Now, you are describing a complex system with three pumps designed to run two out of three. The pumps have different suction sources but tie together in a common discharge. I am not going to focus on whether this is a good idea. Instead, I will focus on explaining how this system could be analyzed.

If this system exists, it was probably designed using a modeling program that could model the flows and pressures in the entire system. The software would have taken into account the complex interaction between the two pumps and the common downstream system. If this design was based on two pumps running at 2500 gpm each, then this is probably very close to what you would achieve. But, only because that is what the system was designed for. As LittleInch noted, if you shut one pump down, and do nothing else to restrict the flow, the flow from a single pump will increase well above 2500 gpm.

If you wanted to analyze this system by hand, without the software, it becomes a complex problem. There is no single system curve. There are two systems that share some common elements. The analysis could be done in an iterative manner. This is how I might approach it:

Assume that one pump is running at 2500 gpm. With a known suction condition and the pump curve, you determine what the discharge pressure must be. Use that discharge pressure on the second pump. For the second pump, you have a known suction condition and a known discharge pressure. Using the pump curve, determine what the flow from that pump would be. Add these two flows together.

Take the combined flow from the two pumps and take that as the inlet conditions into the common header. Build a system curve for the header based on the incoming flow, known outlet conditions and design. If you put in the flow you calculated above, determine the pressure drop in this header system. With a known head loss and a known outlet condition, you can determine the pressure at the point where the pumps enter the header. If this does not match the discharge pressure that you determined for the pumps, then you have to iterate.

Take this new pressure you just determined as the inlet pressure to the header and impose that on both of the pumps. Calculate the flow that each pump would produce with that discharge pressure. Add those flows together and then take that total flow back to the system curve for the header. Each time you do this iteration, it should come closer to convergence on the true answer. It might take two or three iterations to get a good answer.

You could use the same process if you decided you wanted to run all three pumps. If you want to run a single pump, then the system collapses to a simple system with a single system curve. That single pump will run at the intersection of that system curve with that pump curve.


Johnny Pellin

 

[nabilbasha]

of course; as JJPellin explained; when using parallel pumping, the system piping size and head loss calculation must be based on the total flow of the system. e.g. if the total flow is 5000 and you decided to use 2 pumps, each rated for 2500. Then when two pumps operate will almost certainly give you near double of the capacity of each pump i.e. 5000. However, If for any reason you shut down one pump then the flow rate of a single pump would be higher than 2500 because the operating point will move to the right of the curve, because of less head loss. This situation is typical in closed chilled water circulation in HVAC.

 

[5marc5]

JJPellin, I really don't think that I provided a whole new situation; on the contrary, I just wanted to provide more detail to further explain my question. If that got you a bit confused, I do apologize, that was not my intent.

The only thing that I wanted to get further input on is how the system should be treated when developing the system curve for multiple pumps. From talking to other engineers and from this forum and multiple other forums for that matter it seems to me that many engineers are not really sure what is the proper method to address multiple pumps when deloping a system curve.

I am not an expert at all,but my scope of work is expanding, so I am becoming more involved in the water and wastewater branch of civil engineering. Having said that, I have come to realize that many engineers that have been practicing in this area do not provide a definitive approach to analyzing multiple pumps.

As I am trying to master this new area, I have been reading multiple books and reference manuals, including the "Water Distribution System Handbook" by Larry Mays. His approach determines the headlosses within the main piping system (not including suction or discharge lines), adds the static head to the main piping system headlosses and develops a system curve. Then selects a pump curve, but goes on to explain that the manufacturer's pump curve must be "corrected", by subtracting the headlosses that occur in the suction and discharge piping. This provides corrected head points for the pump curve and therefore creates a new pump curve that obviously changes the operating point of the pump versus the system curve.

It treats the minor losses trough suction and discharge piping independent from the main piping system so that the suction and discharge lines only see the flow capacity of the pump and makes corrections to the pump curve based on that. Typically, I have heard people ask if the headlosses in suction and discharge piping for each pump (prior to joing to a manifold that connects to the main piping system, with multiple pumps) should be calculated based on varying flows starting from 0 gpm but exceeding the pump's capcity flow? I understand why people ask that question, it does not make much sense to assume that the piping for the suction and the discharge of each pump will ever see significantly more flow than that of the pump's capacity. That question also comes up because of the lack of understanding on how to properly analyze the system. In other words, as LittleInch kindly pointed out, the headlosses trough the suction and discharge can normally be discarded because they can be negligible. But that might not always be the case. I am certain that for an experienced engineer, it becomes apparant when those headlosses can be negligible or not based on the system. Again, I would say that comes with experience, but for those engineers like myself that have not yet reached that level of experience, I would like to have a full understanding of the proper way to analyze the system to gain the knowledge and eventually some day have that experience. Based on Larry May's method, the answer would be no, you would not include the headlosses of the suction and discharge piping (when dealing with multiple pumps in parallel) until you have selected a pump curve and then you determine those headlosses through the suction and discharge piping based only on flows from 0 gpm to the pump's flow capacity; while the system curve would use flows varying from 0 gpm to some percentage higher than your required system capacity flow.

That is what I understood from that book and May's method, If anyone has that book and or the engineers (JJPellin, LittleInch, Artisi and others) in this forum that have ample knowledge and experience in this area that can provide their input 1) on May's method of analyzing system curves and 2)Other methods that may be better (more accurate) 3) any further comments that will help anyone trying to gain knowledge.

Thanks.

 

[JJPellin]

5marc5,

I am sorry if you took my comment as some sort of scolding. It was not intended as such. I was simply trying to state that your system seems more complex than a simple pump curve overlaid with a system curve. With no information on how the suction sources to the pumps are different, I assumed that they were completely different (different pressure, different piping, and different source) to generalize the analysis so it would work with any differences. I was not confused. No need to apologize.

I see no problem with the alternate method you describe. There will be others on this forum who probably would take exception. But, allow me to paraphrase how I interpret your description. If we treat the pump plus its associated suction and discharge piping as one item, we could generate a corrected curve for the pump plus this small portion of the piping. That seems fine to me. At least one major API pump manufacturer routinely does something similar. They often add an orifice plate at the pump discharge flange and then treat it as if it is part of the pump. They test the pump with the orifice plate and generate the test curve as “pump plus orifice”. A purist would argue that this is improper. But it works just fine.

You could create corrected curves for each pump including the nearby piping. Then you could add the two corrected pump curves together as I showed in my first reply. Then you could overlay this combined/corrected curve with the system curve for the downstream header to get the likely operating point. But, this only works if both pumps have the same suction conditions. You stated that they don’t share a common suction. But you did not describe how they are different. If they both pulled out of the same tank, for example, you could assume that the suction conditions are the same even though they don’t use common piping.

If you go through this exercise, you may find that the adjustments you, make to the pump curve by incorporating the nearby piping are insignificant and can be ignored. But, without performing this exercise, you won’t know that for certain.

I would caution you to not feel you need to restrict the results of this analysis to the rated flow of the pump. A centrifugal pump can operate from zero flow to end of curve (sometimes well beyond that). Even if you don’t intend to operate beyond the rated point, it might be possible that you end up there if you drop back to a single pump. In fact, if two pumps are running at rated flow and one of them trips off, you can be sure that the flow through the remaining pump will rise above rated flow. It would be better to generate the corrected pump curve over the entire range from zero to end of curve. Once the analysis is complete, it will be evident where the single pump will end up running. As before, it will always be at the intersection of the pump curve and the system curve.

I am not familiar with the book you referred to. So, I can’t comment on his methods beyond what you described.


Johnny Pellin

 

[77JQX]

5marc5 - For the cases I analyze, I tend to think of it not as a single system curve, but as an envelope of system curves ranging from one extreme to the other. For instance, what are the system curves when the supply tank is full vs. empty? Or the destination tank is full vs. empty? In your case, you seem to be hung up on how many pumps are operating and the differences in individual suction and discharge piping - so go ahead and calculate the system curves all different ways. Calculate friction from the source boundary condition(s) to the destination boundary condition(s) under all the operating scenarios and flows you can envision. Picking the flow through each individual suction and discharge in advance will be difficult, so you'll probably have to iterate. And of course, as you point out, there is no point in calculating the friction through an individual suction and discharge at a flow greater than the pump can handle.

I'm not familiar with the book you mention either, but it seems a bit easier to put individual suction and discharge piping losses on a "corrected" pump curve (for parallel pumping) than putting all combinations of flow into the system curves. You still will need to account for all extremes of source and destination boundary conditions, though.

Or, you can do what I do, and model it. Have you heard of EPANet?

 

[5marc5]

JJPellin, thank you so much for your comments. Very helpful. I have already performed the exercise as you suggested and indeed the adjustments are fairly insignificant. This really helps in understanding your method of analysis and obviously the reasoning behind not considering the headloss through the nearby piping. Regarding the restriction of the rated flow that i had mentioned, that restriction would have only been through the nearby piping (suction & discharge) but not through the main piping system and as you mentioned, the losses through the nearby pipe is insignificant and can be ignored therefore not really affecting the system regardless of flow(?). But to make sure I fully understand your comment regarding the pump rising above rated point? If I have two identical pumps in parallel and operating point for each pump is say 2500 gpm; wouldn't one pump in operation be delivering the 2500 gpm and when the second pump goes on, the two pumps combined would be delivering the corresponding flow based on the operating point for the combined pump curve (the combined flow being less than twice the flow for one pump due to losses of increased flow, so each pump would be delivering lesser flow than the 2500 gpm each)? And then when one of the two pumps shuts off, the remaining pump would now move back to the operating point on the one pump curve therefore once again delivering 2500 gpm? Is this incorrect?

 

[JJPellin]

I cannot say. If the system was designed for two pumps operating at a total flow of 5000 gpm that is where it will probably operate. If one pump trips off, the remaining pump will increase in flow above its rated flow of 2500 gpm. On the other hand, if the system was designed for one pump running alone at 2500 gpm and you start up a second pump, the total flow will be something less than 5000 gpm total.

The answer always comes out the same in one regard. A single pump, will run where the single pump curve intersects the system curve. Two pumps operating in parallel will run where the combined pump curve crosses the system curve. Without the curves in front of me, I can’t say more than that.


Johnny Pellin

 

[LittleInch]

I agree with all JJPellin says who has nearly written a text book for you. The only thing I would add to your last point is that, depending on the pump and system curves, you could find that a system designed to operate with two pumps will not operate with one as the single pump would run off the end of the curve as there would be insufficient back pressure, e.g to take your example, two pumps would be 5000 gpm but one pump would be 4000gpm before it intersected the system curve and be overloaded if designed at BEP of 2500.

A JJPellin says, without the system curves we cannot say further what would happen. The other thing I would caution is to read my post carefully and whilst suction and discharge piping losses are often negligible, you do need to make each of the pumps piping losses from the common suction point to the common discharge point are as equal as possible otherwise you could easily get unequal flow from identical units, especially in the water industry where heads and pressures are often much less than the oil industry and hence small differences in head losses could become significant.

My motto: Learn something new every day

Also: There's usually a good reason why everyone does it that way

 

[5marc5]

JJPellin and LittleInch thank you both, very helpfull and very interesting conversation we had. I have learned a lot from the both of you regarding this topic and can only hope it serves others as well. As LittleInch noted, Mr. JJPellin basically did write a book for everyone in this forum to enjoy and learn from and for that I am grateful.

Thanks to all.

MR

 

[Artisi]

5marc5, a star or two wouldn't go astray for the help.

It is a capital mistake to theorise before one has data. Insensibly one begins to twist facts to suit theories, instead of theories to suit facts. (Sherlock Holmes - A Scandal in Bohemia.)

 

[ccfowler]

5marc5,

The advice given above can only be described as superb. My only contribution is to suggest my preferred technique of developing reasonably good descriptions of the pump characteristics and the piping system characteristics in a spreadsheet (my personal favorite is Excel) since this allows relatively easy adjustments to more accurately model the actual pump and system characteristics including the effects of the pump affinity laws and "oddities" in the actual piping system and other equipment. Once the existing system is adequately modeled in the spreadsheet, the effects of possible changes or upset conditions can be subjected to reasonably easy "what if" analyses. By setting up all of the spreadsheet equations or formulae yourself, the uncertainty of unknown idealizations and assumptions included in "canned" computer programs will not be an issue of potential cercern.

The key to everything is to recognize that the piping system and the pump(s) always operate on their respective curves. "Identical" pumps operating in parallel almost never share the load equally, and adding an additional pump will not simply add some nominal rated flow amount to the total. Whatever combination of pumps are in operation, they will find their balance in combination with the connected piping system.

I've encountered somewhat troublesome systems where the varying slip of the driving induction motor due to the variations in flow was enough to account for apparently erratic performance. It is wise to confirm the basis of the pump curves. Often, but not always, the pump curves are based on operation at a synchronous shaft speed rather than the slightly varying speeds provided by induction motors.

Valuable advice from a professor many years ago: First, design for graceful failure. Everything we build will eventually fail, so we must strive to avoid injuries or secondary damage when that failure occurs. Only then can practicality and economics be properly considered.

 

[Artisi]

The following link gives a fairly simple over-view of parallel pumping, it should answer most of your questions.

http://www.lawrencepumps.com/Newsletter/news_v07_i4_Apr10.html

It is a capital mistake to theorise before one has data. Insensibly one begins to twist facts to suit theories, instead of theories to suit facts. (Sherlock Holmes - A Scandal in Bohemia.)