Those of you who know me know I am quite enamored with a little freeware application called Plot Digitizer. If you are unfamiliar with it, the application allows to to create CSV (Comma Separated Value) files from the lines in an image. Meaning that if you have, for instance, a pump curve as a .pdf or .jpg file, then you can pretty quickly capture the curve shapes and load them into a spreadsheet to create a chart that is an electronic version of the curve.
Once the curve is in the form of a spreadsheet. you can do math on the lines in it. For instance, you can use the affinity laws to project a new impeller size from a know impeller size. or you can plot a system curve from the data you collect in a pump test and the square law.
This link will take you to a page on our commissioning resources web site where I provide more information, and this link will take you to a page where I provide a spreadsheet template that will let you create a formatted pump curve pretty easily from the CSV files you capture with plot digitizer.
My goal in this post is to show you an idea that came to me one day out in the field that saved the day for me and involved using plot digitizer in a way I had not thought of before. Since it happened while I was working at a really interesting chilled water plant that had some unique features, I thought I would give you a peak at those things also.
There were two cool features in the plant I will be discussing that I wanted to highlight in addition to illustrating my new Plot Digitizer application. But if you want to jump straight to that, the links below will let you do that (or jump to any of the other topics for that matter). Each section has a Back to Contents link that brings you back here.
Setting the Scene
Last October, I had the opportunity to support a field class that used the central chilled water plant at the Gaylord Grand Ole’ Opry as a part of the Building Commissioning Association Annual Conference (formerly called the National Conference on Building Commissioning or NCBC). (And still called that by older folks like me who forget they changed the name). It was a 9,000 ton plant; one of the largest I have been around for a while.
The plant was a variable flow primary/secondary plant. Unlike current technology chillers, back when this plant was designed, the chiller technology would not deal well with flow variation in the evaporator. In fact, in the olden days, before we had realized that we needed to pay attention to energy efficiency, the most common chilled water plant design configuration was a constant flow arrangement and a big driver for that was protecting the chiller tube bundles from frosting up and freezing. But you ended up with large pumps moving the design flow rate at the design head for all of the operating hours.
The variable flow primary secondary design evolved as a way to allow the flow rate to the loads to vary with the load profile, saving a significant amount of pump energy, while maintaining a steady flow rate through the chillers, which protected them. My point in bringing this up here is simply to let you know about the configuration of the plant I am about to discuss, not to explain variable flow primary/secondary plant theory. But, if you want to know more about that, you will find a couple of resources at this link.
Return to Contents
The Quick Look
Aside from the size and quality of the plant, there were two technologies they had in place that provided for some added interest. So, I wanted to briefly highlight them here so you recognize them if you run into them.
Return to Contents
Hydraulic Variable Speed Couplings
One unique feature was that the distribution pumps had hydraulic variable speed couplings on them.
A hydraulic coupling is very similar to the torque converter in an automatic transmission and is the blue piece of machinery between the motor (the dark gray thing on the left) and the pump on the right (the black thing with the silver pipes attached to it) in the picture above.
Here is a picture of the drive itself, including the heat exchanger that rejects the heat associated with the efficiency losses.
While fairly efficient at full speed and full load, these drives are much less efficient than a current technology variable frequency drive at part load. That means they represent a good retrofit target and the plant operating team has that on their list of improvements for this year.
From a retrocommissioning standpoint, for a project where you might need to document the inefficiency of the drive to support your case for replacing it, its kind of cool that you could pretty easily document the efficiency of the drive by logging flow and temperature rise across the heat exchanger because that is where the losses show up.
This is the actuator that controls the output speed of the drive by varying how much of the oil that is moved by the impeller in the drive reaches the turbine.
Another interesting thing about this technology when you contrast it with the variable frequency drive most of us are more accustomed to is that the motor is ahead of the drive. That means the motor needs to be sized for the brake horsepower the pump needs at its input shaft, plus the losses in the drive. In contrast, a variable frequency drive serving a motor serving a pump supplies the pump energy plus the motor efficiency losses.
In this case, the pump bhp requirement is probably in the range of 400 – 425 bhp and the motor is a 450 hp motor. Thus, having the drive losses be part of the load that the motor had to serve probably did not affect the motor selection. But I suspect there are instances where the hydraulic drive would have kicked up the motor size by one incrament due to its location in the “food chain”.
You may wonder why anyone would use even use a hydraulic drive. My guess is that at the time they were installed, a variable frequency drive for a 450 hp 4,160 volt motor would have been pretty spendy.
For example, in 1980, when I specified my first variable speed drive for a 40 hp air handling unit motor, my choices were a variable frequency drive that cost about $50,000 and was the size of two motor control center sections. Or, I could use an eddy current clutch which cost about $20,000. The eddy current clutch was significantly less efficient at part load, but given the price difference, it was the better choice at the time.
So my guess is that a similar economic assessment prevailed when they built this plant and these drives probably made sense back then. Plus, they are basically mechanical devices so a mechanically inclined person can probably fix one in a pinch.
Given that the point of this post is something other than hydraulic couplings, I’m not going to go any further into them for now. But TMEIC’s brochure titled Selecting Variable Speed Drives for Flow Control provides a lot of good information regarding how they work along with comparing them to variable frequency drives.
Return to Contents
Chiller Based Free Cooling Cycle
Most people in the industry are familiar with water side free cooling cycles that leverage the capacity of cooling towers at low wet bulb temperatures to create chilled water directly with out the need to operate a chiller. Typically, this involves operating the cooling towers to produce water colder than is required by the chilled water system and using a plate and frame heat exchanger in between the condenser water system and the chilled water system to transfer the energy.
A picture of a typical plate and frame heat exchanger is shown below along with a little model I have that shows all of the parts.
Here are a few pictures of some actual plates.
My point here is that there is another way to accomplish the cycle with out the cost of the heat exchanger and the pumps it requires and several of the chillers in the Gaylord plant were equipped with this feature.
More specifically, some chillers can be configured in a way that allows an operating mode to occur where control valves bypass the compressors and expansion device. The compressor bypass allows refrigerant vapor to migrate from the evaporator to the condenser due to the vapor pressure difference created if the temperature in the condenser is lower than the temperature in the evaporator. The expansion device bypass allows liquid refrigerant to circulate by gravity back from the condenser back to the evaporator.
That means that if you run the condenser water temperature down below the desired chilled water supply temperature (just like you would if you were going to use a plate and frame heat exchanger for a free cooling cycle), then there will be a natural circulation pattern set up inside the chiller that transfers heat from the (relatively) warm evaporator to the (relatively) cool condenser.
Here are a few slides I use when I teach about this feature, including one that shows the parts and pieces on the chillers in the Grand Ole’ Opry plant. This first slide shows a schematic of a centrifugal chiller with the two control valves added but with the chiller in the normal operating cycle (warmer colors = warmer temperatures).
This next one shows the free cooling cycle triggered with the valves open and the compressors shut down.
This slide highlights the compressor bypass valve on an actual chiller.
Note that it is in a similar location to where the hot gas bypass connection might be for a chiller of this type. But, since this cycle needs to work at very small pressure differences, the pipe is much bigger than it would be for a chiller where hot gas bypass was installed.
Here is a schematic showing the hot gas bypass connection along with a picture of a similar chiller (same manufacturer and product line but about half the tonnage) with a hot gas bypass connection to give you a sense of what that would look like.
The other difference between what a chiller with a free cooling cycle would look like compared to one with hot gas bypass is that the free cooling cycle requires a second valve that bypasses the expansion device. Hot gas bypass does not require this second valve. Here is the Gaylord chiller with the second control valve highlighted.
I will probably do a more details blog post about this at some point, but for now, that should give you a sense of what the free cooling option looks like on a chiller.
All of my exposure to the free cooling feature on a chiller have been on Trane chillers. But I suspect other vendors can offer it assuming their condenser is higher than their evaporator so you can get the gravity flow back along with some other technical details. There is a section in this Trane manual that provides a description of the cycle working on one of their machines. And this page on their website will give you a few more images to look at.
Granted, this adds to the cost of the chiller. But assuming you can get the capacity you need from the feature, it means you can provide free cooling with out buying a plate and frame heat exchanger and piping it into the system. In other words, the pumps and connections serving the chiller also serve the free cooling cycle on the chiller. If you had to do it with a plate and frame heat exchanger, you would need to provide all of that for the heat exchanger in addition to the heat exchanger itself, which is a pretty expensive piece of hardware.
Return to Contents
Using Plot Digitizer to Generate Trend Data from a Graphic
O.K.; I will now return to the main reason I put of this post.
One of the reasons I got to spend time in the Gaylord plant is that the operating team had graciously agreed to allow the commissioning conference to use it for a field exercise in the training class I was supporting for the conference. Originally, I had hoped to use trend data from the plant to illustrate a few techniques I use. But unfortunately, they were in the middle of a control upgrade and there was no trend data readily available.
That changed my plans for the class a bit and time was of the essence since the plan was I would arrive on site the Friday morning before the class, spend Friday exploring the plant, and then develop the class over the weekend in my hotel room.
Back in my hotel room Friday evening, I found myself self studying the pictures I had take of the the chiller graphic displays, longingly wishing the control upgrade was to the point where I could pull the data they contained out of the system.
The pictures below are of of the motor data and evaporator data for one of the chillers and will give you a sense of what I was staring at (note that the time scales are slightly different, which is why things don’t line up exactly).
I was actually contemplating doing a manual transcription of the data. In the olden days, before we had trend data at our fingertips and all we had were log sheets with, if we were lucky, three readings a day with one set of readings taken on each shift, manual transcription and plotting was the approach we were stuck with.
The process was tedious but possible and provided meaningful insights in terms of general trends as long as the data was not highly variable (like a hunting control loop for instance). And, it is the underlying concept behind the process I use now to leverage trend data to start to assess a plant.
In any case, as I was about to make a pass at manual transcription, it hit me; I realized it would a lot faster to use plot digitizer to trace the lines out and create Comma Separated Value files (CSV files) that I could then import into Excel and manipulate to my heart’s content.
So, I loaded one of the images into the tool and tried it out. In hindsight, had I thought about doing it at the time I took the pictures, I would have tried to line myself up more directly with the graphic screen to eliminate the impact of parallax. But I concluded that:
- Since the angle I used was about the same on all of the shots, and
- Since I was not as concerned with absolute values as patterns, and
- Since this was a preliminary analysis and not an exact science
the data I pulled from the photos of the graphics would be good enough for my purposes.
Return to Contents
Using the Trend Data
The purpose of this post is not to go into the details of my analysis technique (but hopefully in a future post I will). Rather, it is to illustrate how I got the data into a form where I could use it for analysis. In general terms, the steps behind what I show towards the end of the post are as follows.
For each chiller, I digitized the motor current data as described above using Plot Digitizer. Specifically, in the first image in the preceding section, I digitized the brownish colored line.
As I clicked across the image in Plot Digitizer to pick up the line, I tried to pick points that would capture the general curve shape, meaning I tried to click on the line anyplace the slope of the line changed significantly, but I didn’t worry too much about ripples that were small relative to the major changes. The assumptions I would be making in my analysis would make them kind of meaningless.
This gave me a little table in the form of a CSV file with a date and time in one column and a percent run loaded amps value for that point in time in another column.
Once I loaded the CSV file into my spreadsheet, I plotted it.
The reason I plotted the chart is that (for me at least), it is a quick way to do data validation. Things like the negative values and points being “backwards” in time jump out at me in the graph faster than they do in the table, all though you can see them both places if you look closely. And for me, as I tweak the data to correct for those issues, the chart gives me a quick visual on the validity (or not) of the adjustment I made.
Obviously, the chiller could not pull negative amps and something in the future could not have happen before something in the past; the reason for the discrepancies was I was slightly off with some of my mouse clicks when I did the digitization. In other words, when I am using Plot Digitizer I am trying to click on a pixel with my mouse that the program then correlates with the pixels I told it represented the X and Y axis for my chart. If I am off by one or two pixels, I get a value that doesn’t really exist.
So, before using the data, I did a bit of data validation. Specifically, using the actual image as a reference:
- I filtered the data to replace negative values with zero.
- I eliminated points that were double clicks on the same point in the line on the image.
- I arbitrarily adjusted the data points a second or two either way where their was a sharp drop in the data so that the line was vertical and/or a data point further down the table (which correlates to a point further to the right in the graph image) was slightly later in time than the value ahead of it.
This only took a few minutes and made my data more representative of the actual data in the image I was trying to capture. And it was much faster and more accurate than manually trying to read points from the graph and enter them into a table. The result of validating the raw data above came out looking like this.
Once I had completed the first two steps for each chiller, I needed a way to combine the data. Given the technique I used to capture the data in the first place, there was not the proverbial “snow ball’s chance in hell” that my time stamps were consistent from chart to chart to chart.
So, I started a new table with the first column being a date and time that incremented by one minute per row. The first date and time value was manually set and simply corresponded to the earliest time I had in my data set. All of the other rows are created by using an Excel formula that added 1 minute to the value in the date and time column I was making relative to the same cell in the row above the cell with the formula.1
In the image below, the table to the left (orange outline) is what the cells looked like for the first few rows of the spreadsheet. The table to the right illustrates those same cells with the raw values and formulas made visible.
Since the data set I was working with covered about two days, I ended up creating about 2,880 rows with time stamps (2 days times 24 hours per day times 60 minutes per hour).
Next, I added columns for each chiller and used the VLOOKUP function to go to the table with the digitized data I had created for each chiller in it and fill in the percent run loaded amps value associated with the time stamp in the first column of the row. Here is what those cells and their formulas looked like for the first few rows.
Certainly, this introduced a bit of an error into my analysis. For instance, let’s say I selected a point at 12:51 PM when I was digitizing my data and then the next value I picked was at 1:51 PM on the same day because the machine was off or drew the same amount of current for that entire period. Because of how the VLOOKUP function works, when it scanned my data table, for all of the times between 12:51 PM and 1:50 PM, it would have reported back the percent run loaded amps value that existed at 12:51 PM.
That means that if I was not fairly meticulous in making sure I captured any significant change as I did my digitizing, I would have introduced some potential issues. But, since I was careful to pick up any meaningful change, this approach would provide a reasonable value for the gap in time.
In other words, if I could visually see a change in the value of the line I was digitizing relative to the previous point I had clicked on, I clicked again. When I reviewed my images, you could visually pick up changes in the range of 2% (basically, the “ripple” you can see in the motor data image earlier in the post between about 9:21 PM pm the 10th and 1:21 AM on the 11th ). So, as long as I was rigorous and methodical in my digitizing, I probably had not skewed the results for any given data interval by more that 2 or 3%.
And again, I want to emphasize that I was just doing this as a first pass to get me pointed in the right general direction. In other words, to quote Pat Murphy (the lead estimator at Murphy Company, where I worked for a while)
It’s an estimate, not an exatamate
When I am out in the field, I have gotten into the habit of taking pictures of the nameplates of the machinery I am looking at. In this case, that included the chillers because the chiller nameplates included a lot of very useful metrics, including the nominal chiller tonnage and the nominal kW at full load.
As you can see, this can be a bit cryptic; i.e. it does not say “nominal chiller tons”; rather there is a code with a number beside it; in this case, NTON.
I happen to be somewhat familiar with the Trane codes so I could read the information from the pictures I took of the nameplates. But lacking that, usually, if you search around a bit on the internet, you can find something that explains manufacturer nameplate codes, either as a separate document or in the form of their installation and operation manual for the equipment in question. Here is an example for Trane centrifugal chillers.
The reason I needed this information was that I wanted to turn the run loaded amps into a tonnage, which I describe next.
To convert the percent run loaded amps to tons, I assumed the relationship between percent run loaded amps and % full load on the chiller was approximately linear. In other words, if the chiller is at 50% run loaded amps, then it is at 50% of its nominal full load tonnage.
This is far from a perfect assumption, especially at low load conditions and especially if the chiller has hot gas bypass. But for the equipment I was looking at, there was not hot gas bypass and the chillers, when running, were typically at 50% load or more. And, I will remind you again of Pat Murphy’s quote.
In any case, using this assumption, I added another column for each chiller and then for each minute in the data set, I estimated the load on the chiller in tons. Finally, for each minute in the data set, I added up the tonnages of all of the chillers that were running, which gave me the total tons on the plant for each minute and allowed me to project a load profile and draw some preliminary conclusions.
Aside from looking at the load profile pattern as a time series, I wanted to look at it in terms of some indicator of a driver behind the load. For plants serving air handling systems with integrated economizer cycles and/or 100 percent outdoor air systems, up until the point where the economizer high limit kicks in, the load on the cooling coils is a 100% outdoor air load, meaning it is a direct function of the outdoor air enthalpy and a fairly direct function of the outdoor air temperature.
As a result, I wanted to be able to plot tons as a function of outdoor air temperature. To do that, I needed outdoor air temperature data which I retrieved from a local ASOS site using the Iowa State University website I mention in the blog post titled Hourly Weather Data Website Update.
Once I had the weather data for a period that correlated with my chiller data period, I used the VLOOKUP function to add an outdoor air temperature value for each row (minute) in my chiller data set.
I always try to do something to cross-check myself, especially when I am going quickly due to the pressure of time or making some pretty general assumptions, both of which were true in this case. Given the data on the evaporator graphic for the chiller (the second graphic I show above), I could also have digitized the evaporator entering and leaving temperatures and used the water side load equation to calculate the tons on each chiller.
To do that, I would need an evaporator flow rate. It turns out there is there is a reasonable assumption I could make to provide that piece of information.
Specifically, since the plant is a variable flow primary/secondary plant, by design, the intent is for the flow through the evaporators to be constant no matter what the flow is in the distribution loop. So assuming a constant flow rate for the evaporators simply reflects the plant design intent and is reasonable. The question then becomes
What is the magnitude of the constant flow rate I am assuming?
Here is how I answered that question.
Since the discharge valves on the pumps were throttled, it was reasonable to assume that the flow being delivered by the pump was the nameplate (design flow).
I say that because the reason balancers throttle pumps is that at their testing has demonstrated that the pump is delivering more flow than needed with the discharge valve wide open. Most balancing specs and good practice require that if the balancer finds this condition, then they should throttle the pump to design since most of the time, this will save some energy (all though not as much as you would save by some other optimization strategy).2
All of that means that time permitting, I could have digitized the evaporator temperature data, done the math, and compared the result I got using evaporator temperature drop data with the result I got using the percent run loaded amps data. But, as I mentioned, I was under a pretty tight schedule and that process would have consumed some of my precious time.
In fact, I initially had considered using the evaporator data but decided on the percent run loaded data instead because it would (in theory) get me similar results. But to accomplish it, I would only need to digitize one line for each chiller, not two. And, I would only need to do one VLOOKUP for each chiller, not two. Given that there were 9 chillers in the plant, saving those steps represented a significant time savings.
However, I did spot check my results by randomly selecting some points in time and then comparing the actual logged evaporator temperature drop with the temperature drop I came up with based on:
- Tons from my percent run loaded amps analysis, and
- An assumption of design flow through the evaporator (the valves were in fact throttled as indicated by the white line being approximately in the middle of the window on the valve actuator)
These spot checks tended to validate each other, so I was reasonably comfortable moving forward with the data set.
Return to Contents
The Bottom Line
The bottom line was that my brainstorm about using plot digitizer to turn a photo of data into actual data paid off, allowing me to generate both a time series view of the plant load profile and related parameters …
… and scatter plots and regressions of the load profile patterns relative to outdoor air temperature.
I’m just about done with a post that takes a look at the clues that were revealed by the charts above. But before I do that, I wanted to show you one other trick that came in handy for my effort working with the data I generated, specifically, how I created the third, number of chillers running axis in the time series chart. That will be the topic of my next post.
Return to Contents
Senior Engineer – Facility Dynamics Engineering Visit Our Commissioning Resources Website at http://www.av8rdas.com/
1. For more on how Excel represents date and time, which you need to understand to be able to do this, see the blog post titled Good News about NWS Weather Data, Plus Working with Date and Time in Excel.
2. For more about optimizing pumps, including case studies illustrating the steps and techniques, you may want to download and read two design briefs that you will find on the Energy Design Resources web site; Centrifugal Pump Application and Optimization, and Pumping System Troubleshooting.