YT YouTube2Text

YouTube2Text → Transcripts → I Tried to Make a Better Fan

I Tried to Make a Better Fan — Transcript

by CaptiveAire · 8,450 words · 1,217 segments · language en · Watch on YouTube

Open in transcriber Summarize with AI Download SRT/VTT

Full transcript

0:00This video is about fan wheel design.
0:02But before we start, I have to say this.
0:04I'm a little bit of a geek. My wife
0:06would say a lot. In school, I was always
0:09that guy that had the teachers getting
0:11back to me after class. My endless
0:14questions always going beyond what was
0:16meant to be taught. And so, it shouldn't
0:18be surprising that when I was helping
0:19produce our first fan videos, I started
0:22asking some questions in particular
0:24about centrifugal fans,
0:27like why are the blades at these angles?
0:29[music] And why do we have eight blades?
0:31Why not six or seven? Or just, "How did
0:34you come up with this fan wheel at all?"
0:36Simple questions with often complex
0:38answers, maybe too complex for what we
0:40were trying to do in those videos.
0:41[music] And here's a part that's kind of
0:43hard to admit, but I should anyways.
0:46[music] And that's that I thought I
0:48could maybe nudge engineering into
0:50designing better [music] fan wheels.
0:52Really, this is where this video was
0:54born, and you'll have to stick around to
0:55see just [music] how that went.
0:58So beyond fan wheel design, this is a
1:01story about theory versus [music]
1:02practice, overconfidence, humility, and
1:06the always always almighty scientific
1:08[music] process. Let's go.
1:15As you might imagine, a lot of theory
1:17goes into designing something that moves
1:19a fluid. Nature presents us with things
1:21like turbulence, cavitation, and we try
1:24to come up with basically math that
1:26predicts these things so we can work
1:27around them or with them the best we
1:29can. But I didn't want this to be a
1:31video all about theory. No, I wanted to
1:33show the real deal. And that meant two
1:35things. Number one is I had to figure
1:38out a way to design and physically make
1:40different fan wheels. And number two, I
1:43had to test these wheels so we know how
1:45they perform. From the start, I knew I'd
1:47be 3D printing these wheels. And I
1:49thought that part would be easy. It
1:52wasn't. Before I even started designing
1:54my own wheels, I tried to make a scaled
1:57version of one of our wheels and set up
1:59my test rig to ensure the whole idea was
2:01going to work. And oh boy, I wasted a
2:04lot of plastic, PLA to be exact. 3D
2:06printing is great when it works, and
2:09when it doesn't work, it's almost always
2:11user error. But it never feels like it's
2:13your fault at the time. It took me a
2:15while to realize the geometry was just
2:17too thin everywhere, causing all kinds
2:19of issues. When I finally got a clean
2:22wheel to print, it looked like this
2:24because I ran out of gray filament
2:25halfway through. It didn't really
2:27matter, though, because when I tested
2:28it, I realized my scale was just too
2:31small and I would have to go bigger. So,
2:33I got lots of practice and throughout
2:34the video, I became a lot better at
2:36properly setting up the prints, but it
2:39was always a bit of a battle. Also,
2:42supports are your friend and also your
2:44worst enemy all at the same time. You
2:47can't live with them and can't live
2:49without them, I guess. Now, let's take a
2:51look at our test rig. First and
2:53foremost, we have to turn the fan. So,
2:56first I tried using an EC motor I had
2:58from a previous project that's made to
3:00power a 12-in fan wheel. It worked, as
3:02in it spun the fan, but this is a big
3:05motor compared to these little fans, and
3:07it draws a relatively large amount of
3:09power just to spin the shaft. and
3:11overcome its own losses even without a
3:13fan wheel attached. When you're testing
3:15a fan, you have to realize that only
3:18some of the energy from the motor goes
3:19into driving the fan. The rest go into
3:22heat, vibration, noise, and this is what
3:25we call motor losses. I didn't want to
3:28be in a situation where motor losses
3:30were greater than the power required to
3:32drive the fan. In fact, I wanted motor
3:34losses to be as small as possible. Also,
3:36I wasn't sure if 1,800 RPM was high
3:39enough to drive some of our fan designs.
3:41So, next I tried this little guy, a 12vt
3:43DC motor from a drill. This one had
3:45fewer losses and could spin more than
3:47twice as fast, but with a 90 W limit, I
3:50thought it might be too small to power
3:52some of the fan designs I wanted to
3:53test, which actually was right. So,
3:55third times a charm, and this 24volt
3:58electric scooter motor became the best
4:00compromise with lots of power available,
4:02a higher RPM limit, and still much lower
4:05losses compared to the big motor. Now,
4:08about those pesky motor losses. Let's
4:10say this motor is using 80 W while
4:12spinning this fan. Okay, then how do I
4:14know how much of that power is used by
4:17the fan versus wasted motor losses? One
4:20way would be to use a dynamometer to
4:22isolate the fan energy. But with this
4:25being such a small setup, I decided this
4:27would just be too complex. So, I didn't
4:29go with that. Another way would be to
4:31use performance data provided by the
4:33manufacturer to calculate the losses.
4:36But good luck getting any kind of
4:37reliable data for a motor that's made
4:39for a scooter. Now, this is usually
4:42reserved for specially calibrated
4:44motors. So, as a sort of solution,
4:46here's what I tried. I measured the
4:48motor power draw spinning by itself at
4:50different speeds. And by subtracting
4:52those values, we can roughly calculate
4:54the power used by the fan. I say roughly
4:57because this is only a halfway solution.
4:59The motor is going to have more losses
5:01when loaded. As you can see on this
5:03sidebyside thermal comparison. So for
5:06most of this video, I'll be using the
5:08total gross power consumed by the motor
5:10with maybe a couple exceptions. Power
5:12consumption is one of four parameters
5:14this test rig will need. The next is
5:16RPM, the speed at which the motor and
5:18fan are turning. At first, I used a
5:20laser tachometer I had laying around and
5:23some reflective tape on the fan wheel.
5:25This worked just fine, but knowing I
5:27would have to record a lot of data, I
5:29decided to go with something more
5:30automated. So, I installed this little
5:32hall sensor tachometer, which works
5:34alongside a magnet I just taped on the
5:36hub with a nice LED display continuously
5:38giving us the fan speed. This will be
5:40much easier. The last two parameters to
5:43measure are air flow and static
5:44pressure. We explained these in detail
5:47in a lot of our recent videos, but in
5:49summary, as the fan moves air in the
5:51duct, friction between the air and the
5:54walls of the duct creates a resistance
5:55to air flow. We can measure this
5:58resistance by reading the static
5:59pressure of the air right before the fan
6:02and comparing it to the atmospheric
6:03pressure around the test rig. In this
6:05setup, the fan is creating a lower
6:07pressure here, which makes the air move.
6:09So, pressure entering the rig will
6:11always be higher than right before the
6:13fan. And the difference between the two
6:15is the pressure loss through the duct.
6:17Just like in our fan curves video, I
6:19made an iris to add some control of the
6:21rig's resistance to air flow. This
6:24simulates different system conditions
6:25the fan might see. From a pretty open
6:28system where resistance is low and air
6:30flow is high to a more restrictive
6:32system flowing less air. Speaking of
6:34air, I chose to use a thermal animometer
6:36to measure air flow. This device works
6:39by measuring the cooling rate of air on
6:41a heated sensor. And I chose to place a
6:43sensor right in the middle of the duct.
6:45Now, I have to tell you that this is not
6:47ideal as air speed in a duct is not
6:50uniform. It's actually a little faster
6:52through the center of the duct. A true
6:54airflow reading requires averaging of
6:56air speeds using a grid or multiple
6:58readings in a duct traverse. But I think
7:00for the purposes of comparing different
7:01wheels for education, this will be
7:03accurate enough. And besides, this
7:06particular unit calculates volume of
7:08flow in cubic feet per minute on the
7:10fly. And that's going to be really
7:11handy. By now, you may have noticed this
7:13little piece here. This is the inlet
7:15plenum. And I actually borrowed the
7:17geometry from one of our fans for this.
7:19This interface is crucial as the air
7:21makes its way into the wheel. And as far
7:23as sizing, I gave myself a little bit of
7:25a margin just in case we had any wobbles
7:27during the test. It's also important to
7:29have good radial alignment, not just to
7:31avoid rubbing, but also to improve
7:33efficiency. We want the air to come from
7:36the duct, not from the surroundings, as
7:38much as possible. To get this all
7:40aligned, I used washers and little shims
7:42made out of tape. And it took a couple
7:44tries, but my eyeball meter says I got
7:46it pretty perfect. So, here's the setup
7:50actually working with the fan wheel. We
7:52have total power consumption in watts,
7:55RPM, air flow in cubic feet per minute
7:58or CFM, and finally static pressure.
8:01This is measured in inches of water
8:03column. With all this set up, I can also
8:06show you that axial alignment from the
8:08wheel to the plenum matters. As I change
8:11this interface, the performance of both
8:13air flow and static pressure changes
8:15quite a bit. Or if you ram the fan into
8:18the plenum, it turns out performance
8:20also changes. Anyways, I try to be
8:22mindful and consistent with this
8:23alignment as I ran these tests. So, now
8:26that we have a working test rig, it's
8:28finally time to start assigning some
8:29fans.
8:34815 23 left takeoff.
8:37>> Turbo machinery is a term used to
8:39describe machines that transfer energy
8:41between a rotating component and a
8:43fluid. This includes things like
8:46turbines, compressors, pumps, and fans.
8:49And it just so happens that some of
8:51these items make my favorite noises.
9:07I could listen to that stuff all day.
9:10Anyways, turbo machinery. Um, yes, it's
9:14complex. There's a lot of information
9:16for out there, but most of it actually
9:18is pretty specific on pumps or turbines,
9:21and everything carries over, but
9:23sometimes the leg work can be a little
9:24bit intense. So, while looking for a bit
9:26of a shortcut, I downloaded this program
9:28called CF Turbo, which really helped me
9:30understand a few things I can pass on to
9:32you. As a disclaimer, this is not a
9:35tutorial. I'm just using this software
9:36as a sort of teaching tool. So, let's
9:38dive right in and see what we can learn.
9:40Thankfully, the software has modules for
9:42all types of different turbo machinery.
9:44And of course, the fan module will be
9:46our go-to today. Right off the bat,
9:48you're asked for machine design points.
9:50And here's where our lesson really
9:52starts. There are three design points
9:54that determine the type of fan you will
9:56or should be making. The flow rate of
9:59the fan, the delta P or the pressure
10:02differential the fan can handle, and
10:04finally fan speed. Hey, these are all
10:07things our rig can measure. What do you
10:08know? It turns out that these three
10:10design points determine the fan specific
10:13speed and the fan specific diameter. And
10:15these are hugely important in the
10:17initial design of the fan. And I'm sure
10:19you're asking, well, what do these mean?
10:21I'll keep it as simple as I can.
10:23Starting with specific speed. Specific
10:26speed is a dimensionless design index
10:28that gives us some idea as to what type
10:31of fan to use. A low specific speed will
10:34correlate to using a purely centrifugal
10:37wheel. A medium specific speed is in the
10:39realm of mixed flow wheels and a high
10:42specific speed is typically reserved for
10:44purely axial fans. So let's look at this
10:47equation. You can see that the whole
10:48equation represents a relationship
10:50between the speed of the fan, the flow
10:53and the head or basically pressure
10:55differential the fan can handle.
10:57Increasing flow rate and RPM both
10:59correspond to an increase in specific
11:01speed. While increasing the head or
11:03pressure requirement decreases specific
11:05speed. Great. So looking at this number,
11:08we can determine what kind of fan is
11:10best to use. The result of this equation
11:12becoming a sort of index pointing to
11:15where we fit in this scale. A word of
11:18caution though, this scale is of course
11:20very dependent on the units used and
11:22I've seen all kinds referenced. So make
11:25sure if actually using this scale for
11:27fan design to have all your units in
11:29alignment. All right. So what about
11:31specific diameter? Specific diameter is
11:34also a dimensionless value which helps
11:36us determine the optimal diameter of a
11:39fan given a certain pressure head and
11:41flow rate. And yes, it's confusing. We
11:44have the specific diameter here and the
11:46actual diameter here. So, how does this
11:49equation work? What what do we solve
11:51for? This is where it gets interesting.
11:53Thanks to our friend Otto Cordier, who I
11:56couldn't find a picture of, so I use
11:57this really awkward stock footage.
11:59Cordier was a German engineer who in
12:011953 published a version of this, the
12:04Cordier diagram. Really the first step
12:06in efficiently designing any turbo
12:08machine. This diagram links together
12:11specific speed and specific diameter
12:13with an ideal best fit curve based on
12:15lots and lots of testing done by Cordier
12:18himself. Let's say you're wanting to
12:20make a fan with a calculated specific
12:22speed of 10. Again, this is based on air
12:24flow, RPM, and pressure. Thanks to all
12:27this empirical data cordier gathered, we
12:29know that when we intercept this cordier
12:31line, we can use that as a reference to
12:33tell us the most optimal specific
12:35diameter for this particular fan. In
12:38this case, it's about 1. Great. So now
12:41we can take the one, plug it into the
12:43specific diameter equation and solve for
12:45the ideal actual diameter of the fan.
12:48This also works in reverse. say that due
12:51to limited space we need to build a fan
12:53of a certain diameter with a certain
12:55flow rate and pressure requirement. We
12:57can just plug those values into this
12:59equation determine a specific diameter
13:02and from there using the cordier diagram
13:04establish a corresponding specific speed
13:07and finally using the specific speed
13:09equation you can figure out ideally at
13:11what RPM the fan will have to turn to
13:14achieve these parameters. If by some
13:16limitations such as RPM, you end up with
13:19a fan that's outside the cordier line,
13:21the fan will still work. But the further
13:23your design is from this line, the
13:25further you will be from the most
13:26efficient design, at least in theory.
13:30Some diagrams even include a
13:31characterization of the type of fan
13:34based on the specific speed, making them
13:36the most useful all-in-one resource to
13:38start your fan design. But once again,
13:40be very careful with these as most of
13:42the time these characterizations are for
13:44pumps rather than fans and the ranges
13:46are actually pretty different. In truth,
13:49it probably doesn't really matter as
13:51you'll likely be using software that's
13:53already referencing the cordier diagram.
13:55But I wanted you to know how the
13:57software is calculating these things.
13:58Otherwise, it becomes this magic black
14:01box, and I don't like those.
14:03>> Put your right hand in the box.
14:05>> What's in the box?
14:06>> Pain,
14:07>> by the way. Great book series. And
14:09surprisingly so far, great movies. So
14:12for our first design, we're going to
14:13make a fan that can move 200 CFM. I
14:16chose that airflow number to keep the
14:17air speed in the 6-in duct at about
14:191,000 ft per minute. I thought 1 in of
14:22pressure was a reasonable place to
14:24start, and 2,000 RPMs puts us at about
14:2770% of the maximum speed of the motor.
14:30You can see the software is already
14:31suggesting based on the specific speed
14:34that this corresponds to a relatively
14:36low pressure centrifugal fan. but a
14:38centrifugal fan nonetheless. Let's go
14:40ahead and make the wheel. Now, as this
14:43is our first design, I want to make the
14:45wheel as simple as possible with
14:46straight blades. Right away, I want to
14:49show you that the software has picked a
14:50wheel diameter based on the cordier
14:52diagram. It's also picked a suction
14:55diameter, and we'll go into this in a
14:57bit, but let's finish up the wheel
14:58first. Here you can see the meridian
15:00profile of the wheel and the 3D model.
15:04As you can see, the software defaulted
15:06to backward incline blades, and it also
15:08picked seven blades. Now, it's actually
15:10giving me a warning saying blade overlap
15:12is too low, meaning I should add more
15:15blades. But you know what? I know you
15:16just want to see this wheel in action.
15:18So, let's go ahead and get it printed.
15:19Once I added the material for the bottom
15:21blade and the top blade or hub, I
15:24exported the wheel and sent it to my 3D
15:26modeling software. This software is made
15:28for animation, not making actual things,
15:30but it works fairly quickly to add these
15:32holes for the hub. And only 13 and 1/2
15:35hours. Wait, I ran out of filament.
15:37Okay, 13 and 1/2 hours later, we have a
15:40finished wheel.
15:42Right. So, with nothing to compare it
15:44to, I'd say this wheel did great. I did
15:47have to spin it a little faster than our
15:49design of 2,000 RPM to achieve the air
15:51flow and pressure goals, but it achieved
15:53them at 77 watts of measured total
15:56power.
15:58So, we already know because the software
16:00told us that this doesn't have enough
16:02blades. So, let's add some more. But how
16:04does this work? How many blades should
16:06this fan have? This was one of those
16:08questions where the answer has more to
16:10do with trial and error than with actual
16:12science and equations. More blades are
16:15always beneficial in controlling the
16:17air, but add too many and you end up
16:19reducing the total wheel area the air
16:21has to travel through, as well as adding
16:23friction and making a heavier wheel. I
16:25ended up going with 11 blades here just
16:27to see what happens. And performance was
16:29actually a little bit worse, so I think
16:31maybe I went too far. But I won't spend
16:34any more time on this because really I
16:35want to start getting away from straight
16:37blades. Which gets me to my next topic,
16:40blade angle. Now these blades are
16:41straight, right? So, you would think,
16:43okay, they just have one angle, right?
16:45You could look at it that way, but the
16:47software doesn't. In fact, it says you
16:49have a blade inlet angle, angle one, and
16:52a blade outlet angle, angle two, and
16:54they're way different. At first, I
16:56thought this was incorrect. But like
16:58with everything else, while making this
17:00video, it turns out I'm the one that's
17:01wrong. So, let's draw it out. To get
17:04blade angle one, you draw a line from
17:06the center line of the fan to the
17:07leading edge of the blade. Then, you add
17:09a perpendicular line to this. The first
17:12angle measurement is from the blade to
17:14this line. To get blade angle two, you
17:17do the same to the blade trailing edge.
17:19And as you can see, this is a much
17:22bigger angle than angle one.
17:25Now, the software is telling us here
17:26that we have quite a bit of delta
17:28between the ideal blade angle at the
17:30trailing edge and the actual angle we're
17:32using. And that's because the first
17:34angle here is optimized. But since we're
17:36using straight blades, we don't have any
17:38control of the second angle. It's fixed
17:40to the first angle. So, let's change
17:43these straight blades to curved blades
17:45and see what happens. We now have a
17:47backward curved fan.
17:49Thanks to this new curve, we're pretty
17:51much hitting these ideal numbers which
17:53are calculated for this specific fan.
17:56And before we get into how this math
17:58works, I want to show you just how
17:59important this is. So, here's the fan
18:01wheel hitting the design of 200 CFM at 1
18:04in of static while only consuming 63
18:07watts total power. That's a huge
18:09efficiency improvement from the 77 watts
18:11or backward incline wheel required. And
18:14I showed you this first because it's
18:15about to get a little more complicated
18:17when we again ask the question, how did
18:19the software calculate both of these
18:21angles? The first angle is actually
18:23pretty intuitive. Let's follow a
18:25particle of air in a spinning wheel. The
18:27air moves radially outward from the
18:30center of the opening and it meets the
18:31leading edge of the blade. We basically
18:34want the relative angle of the air
18:35entering here to match the angle of the
18:38blade. The software can calculate this
18:40as a function of the speed of the air at
18:42our given design point, the rotational
18:44speed of the fan, and the blade leading
18:46edge angle. This matching of angle and
18:49trajectory results in a shockless entry
18:52where the air is least disturbed and
18:54this is the most efficient.
18:56Okay, now that the air has entered the
18:58blade area, the blades are ready to do
19:00work on it. The air is already moving in
19:03a radial direction this way. But now the
19:06blades will accelerate the air in the
19:08tangential direction increasing the
19:10tangential velocity. This is also called
19:13the whirl velocity. In essence, this
19:16whirl velocity increase is where the fan
19:18adds energy to the air. And this works
19:21exactly the same with any fluid and any
19:23other type of turbo machine that's
19:25adding energy to the fluid. Some turbo
19:27machines like turbines are actually
19:29extracting energy. So everything's kind
19:31of reversed. If we can measure this
19:33increase in tangential velocity and we
19:36know how much fluid we're targeting to
19:38move, we can extrapolate all kinds of
19:41useful data. And this is exactly what
19:43Leonard Oiler, a Swiss polymath, found
19:46out and published throughout the 1750s.
19:50Oilers's turbine or pump equation,
19:52they're one and the same, tells us how
19:54much work is performed by a fan as a
19:56function of the blad's tangential
19:57velocity u at points 1 and two, as well
20:02as a fluid's tangential velocity vt at
20:05the same two points. We can also derive
20:08these equations which can give us
20:09torque, power, and pump head or pressure
20:13rise. So, what does this have to do with
20:16blade angles? Well, it turns out that
20:18the term VT or the fluid's tangential
20:20velocity is heavily heavily dependent on
20:23the blade angle at that point. And it
20:26takes some trigonometry and the use of
20:27these so-called velocity triangles to
20:29calculate them. Something we won't do
20:31today because I actually like you guys.
20:34But we have all the data we need. So,
20:37the software is doing all these
20:38calculations for us. Remember we already
20:41have a blade angle picked out for the
20:43leading edge blade angle 1 based on
20:45shockless entry. So with that data it
20:48can calculate VT1. Then using Oiler's
20:51pump head equation it will solve for VT2
20:54and ultimately give us blade angle 2. In
20:57a previous video I made some animations
20:59to show the difference in the air's
21:00tangential speed between backward radial
21:03and forward bladed fans. And I even had
21:05this comparison to show how different
21:07the path of the air is between these two
21:09kinds. But for this video, I thought it
21:11would be cool to do a practical demo. So
21:13here's a backward curve fan which
21:15imparts very little tangential velocity
21:17on these beads. In fact, I think most of
21:20the scattering is plain centrifugal
21:21force. And if I reverse the speed of
21:24this impeller, it essentially becomes a
21:26forward curved fan. Now look at how much
21:30more energy this impeller is imparting
21:31on the beads. So for the same size fan
21:34and at equal speeds, a forward curve fan
21:37is much more capable but also a lot less
21:40efficient because the energy transfer is
21:42very violent compared to the much more
21:44mellow smooth input of a backward curved
21:47impeller. And I wanted to verify this.
21:49So for fun, I printed a radial and
21:51forward curved fan. Their performance
21:54was abysmal. The radial fan required 137
21:57watts at the design point. Remember,
21:59this is against 63 watts for the
22:01backward curved fan. And the forward
22:03curved fan couldn't even reach the
22:05design point. But a lot of that has to
22:08do with the fact that these two types of
22:10fans perform much better with a housing
22:12around them. Take the forward curve fan
22:14for example. Since this fan literally
22:17scoops the air in the tangential
22:18direction, without a housing controlling
22:21the flow, the air exiting will just
22:23swirl around and kill all the momentum
22:25the fan just added to this air.
22:28To solve this, a scroll housing or in
22:30turbo machinery terms, a volu is that
22:33right?
22:34>> Volute.
22:35>> Yeah, volu is used to manage how the air
22:38exits the wheel. Starting with this
22:40cutoff, the air has limited space to
22:42exit, which forces it to continue in
22:45this direction. This then creates a sort
22:47of boundary for the air exiting further
22:48down the wheel. And the same thing keeps
22:50happening as the space continues to
22:52fill. There are different methods of
22:54shaping the housing, but they all rely
22:56on this same principle. So, look at the
22:58difference. This forward curved wheel is
23:00now actually performing, moving up to
23:03370 CFM, much better than the 139 we got
23:07without a housing. I also made this wall
23:09adjustable to demonstrate that even
23:11changes in the width of this housing
23:13affect performance. So control of how
23:16the air exits and expands out of these
23:18aggressive wheels is extremely
23:19important. On the other hand, a
23:21well-designed backward curt may perform
23:24about the same with or without a
23:26housing, which makes it extremely useful
23:28for use in HVAC applications. So with
23:31that in mind, I want to start looking at
23:33different design parameters and how fans
23:36change with them.
23:39Let's make another wheel now and just
23:41change one design point. We're going to
23:43lower the pressure to just half an inch.
23:46This in turn gives us a higher specific
23:48speed. And now we're in mixed flow
23:50territory. But I went ahead and
23:52completed this wheel as a backward curve
23:53wheel so we can more easily compare the
23:55two. And here are the two models. The
23:58new wheel has a smaller overall
23:59diameter, a larger suction diameter or
24:02inlet size, and the outlet is wider.
24:06We know that the overall diameter is
24:08prescribed using a cordier diagram. The
24:10trend will always be to decrease this
24:11diameter as specific speed increases.
24:14The suction or inlet diameter ratio is
24:16also often derived from empirical data.
24:19And the trend with higher specific
24:21speeds is to increase this ratio to
24:23reduce the restriction to flow. This
24:25also comes about because we don't need
24:27to raise the pressure as much. So the
24:29blade area can be smaller.
24:32Finally, the width of the outlet will
24:33also increase because we increase the
24:36inlet area. Ultimately, the number is
24:39determined by some of Oilers's math to
24:41keep the desired velocities through this
24:43wheel. And look, Oilers's math can get
24:46pretty complex. It's definitely worth
24:48spending time on if you're wanting to do
24:50this for a living. But for today, I
24:52mostly want to show you the
24:53relationships. For example, we can go
24:56the other way. Here's a wheel with a
24:57higher pressure requirement of 2 in and
25:00therefore a lower specific speed. The
25:02result is the opposite. We have a larger
25:04diameter wheel with a small inlet and
25:07narrow outlet. We can change different
25:10parameters too like air flow and this
25:12400 CFM/2in
25:14static wheel starts to get bigger,
25:17wider, and its inlet to outlet diameter
25:19ratio is pretty high. So, how did these
25:22wheels perform? I tried testing each
25:25wheel at all these design points, but
25:27not all of them were able to achieve
25:28them. Starting with our baseline of 200
25:31CFM at 1 in of static, all the wheels
25:33could handle this, but clearly the wheel
25:36that was designed for these set points
25:37did the best. Next is the 200 CFM at
25:41half an inch of static test. This one
25:44surprised me. The wheel designed for
25:45these parameters did okay, but it was
25:48actually outperformed by this wheel as
25:51far as gross power consumption. This
25:53goes to show the importance of testing.
25:55Maybe the wheel the software suggested
25:57wasn't as optimized as we thought. Then
26:01we have the high pressure test at 2 in
26:03of static. Only the wheel that was
26:05designed for this was able to reach this
26:07target. All the other wheels may have
26:09made it, but with a 2,800 RPM limitation
26:11on the motor, they just couldn't get
26:13there. And finally, there's the high
26:16flow low pressure test. This one was
26:19interesting because when it comes to
26:21efficiency, none of the other wheels
26:23were able to get even close to the wheel
26:25that was designed for these conditions.
26:27Also, this wheel is very efficient
26:29across a pretty large operating window.
26:32On the other hand, something like this
26:34wheel wasn't even able to make it to
26:36this test point. It's just too small and
26:38maybe optimized for its own design point
26:40with the seemingly narrow operating
26:42window. So when talking about operating
26:44windows, there are two scenarios you
26:46might find yourself in when designing a
26:48fan. In the first scenario, you're
26:50designing for a very specific
26:52application, a very narrow operating
26:54window. In the second scenario, the fan
26:56has to be able to handle a wide range of
26:59operating conditions and applications.
27:01You might still want to optimize one
27:03specific point, but the fan should be
27:05able to perform well outside of this
27:07point. This is a scenario we face at
27:09Captive Air since our equipment is
27:10designed for extremely variable
27:12conditions. So when we test fans, we
27:15can't just look at one or a handful of
27:18conditions. No, we need a bigger
27:19picture. This typically means gathering
27:22data for each fan at one set speed and
27:25with different system conditions. So, I
27:28did exactly that with our baseline fan
27:30by maintaining this fan at the design
27:32speed of 2,000 RPM and changing the iris
27:34position in steps from all the way
27:36closed to wide open. And this is the
27:39data I collected.
27:41The most basic way to look at this data
27:43is to simply plot air flow versus
27:45pressure. This is your basic performance
27:48curve showing us that as we increase
27:50static pressure, the air flow decreases
27:52with it.
27:55We can also add a curve for our power
27:57used. Notice that power usage peaks
28:00close to where this fan was designed to
28:02operate. And although that's a little
28:03counterintuitive, that's likely where
28:06the fan is doing the most work. We can
28:08also easily calculate the fan's
28:10efficiency as a function of these
28:12parameters. So this graph can tell you
28:14where your fan is most efficient. And
28:16not surprisingly, it's most efficient
28:19right around it design point of 200 CFM
28:21and 1 in of static. So when designing
28:24fans for varying conditions, we can have
28:26a look at all these parameters to see
28:28the range of pressures and air flows the
28:30fan can handle as well as the power
28:32usage and in particular the efficiency
28:34curve. An axial fan, for example, tends
28:37to have a much more narrow efficiency
28:39curve as the blades are subject to
28:41stalling much outside its design point,
28:44generally targeted at higher air flows
28:45and lower pressures. This is why we use
28:48backward incline or backward curb fans
28:50quite a bit because not only can they
28:52handle a lot more pressure, but their
28:54efficiency curves are typically much
28:56broader. And that matters with the rise
28:58of variable frequency drives and
29:00electronically commutated motors. Now
29:02more than ever, the same fan design can
29:05be set to operate at different speeds.
29:08So it'd be very useful to understand fan
29:10performance beyond a fixed speed. And we
29:13do this through data extrapolation. We
29:15can turn this data into full fan
29:18performance curves. We do have a video
29:20dedicated to the process and math behind
29:22this, but in summary, the curves tell us
29:25the speed and power required by a fan at
29:27any given air flow and static pressure
29:29combination. So, at Captive Air, we make
29:32fan wheels [music] from 10 in to 36 in
29:34diameter. And I'm sure this range will
29:36continue to expand. [music] and fan
29:39performance curves help us understand
29:41where these fans overlap, where they're
29:43most efficient, and where we can
29:45improve. With such a broad range of
29:48wheels, I should point out that we can't
29:50just make one design and simply scale it
29:52up or down. One reason is manufacturing
29:55and [music] material limitations. You
29:57may not always be able to just scale up
29:59a wheel. But even if you could, it turns
30:01out that efficient fan design changes as
30:04we scale the wheel. Let me explain.
30:06Let's say we have a range of four fans.
30:08The first is designed around 1,000 CFM
30:11and 2 in of static. The next is 2,000
30:14CFM and 2 in of static. Then 4,000 and
30:176,000 CFM also at 2 in of static. Notice
30:21that as we increase air flow, we're
30:23keeping static pressure targets
30:25constant. Why? Well, in HVAC, as air
30:29flow increases, equipment size and
30:31ducting diameters should increase as
30:33well. What this means for these fans is
30:35as we make them bigger for more air
30:37flow, their specific speeds increase
30:39because their static stays constant. So
30:42here's what the software came up with
30:43for each of these fans. You can see the
30:46proportions change. The wheels are not
30:49simply scaled up. So actually the bigger
30:52the fans are, the more the software is
30:54telling us we should be starting to
30:56consider a mix flow design. And that's
30:59exactly why at Captive Air, we also make
31:02our own mix flow wheels. Mixflow wheels
31:05combine elements of centrifugal and
31:07axial design, giving us a resulting
31:09diagonal flow through the wheel. These
31:11wheels are typically not as capable as
31:14centrifugal wheels when it comes to
31:15pressure, but are more capable and
31:18efficient when it comes to moving lots
31:20of air. And as far as my little test
31:22wheels go, this would be the point where
31:24this 3D prototyping process would really
31:26shine. At least that's what I thought.
31:34Before even trying to make a mix flow
31:36wheel, I made an air foil wheel to dip
31:38my toes into slightly more complex
31:40geometry. An air foil wheel is a lot
31:43like a backward curved wheel, except
31:45that we have this air foil shape. This
31:47is here to make the wheel more efficient
31:50by ensuring that the air flow remains
31:51attached on both sides of the blade. At
31:54least in theory, because my wheel was
31:56consuming one watt more total power at
31:59design than the original backward curved
32:01wheel. Unfortunately, this lackluster
32:04performance would set the tone for a
32:06little while because next I tried making
32:08a mix flow wheel to compete with a
32:09benchmark backward curve wheel and the
32:12performance was actually worse. So, I
32:14thought, well, maybe this wheel has too
32:16low of a specific speed. Let's try a mix
32:19flow wheel to compete with the lower
32:20pressure wheel we made earlier. And this
32:23time performance was better by a couple
32:25watts. And I said, "Okay, that's the
32:28right direction. Let's make one now for
32:30the high airflow, low pressure wheel.
32:32That one should really favor a mixed
32:33flow wheel." Once again, it was a little
32:36bit better. Three total watts better.
32:38But honestly, I was expecting more out
32:40of these mixed flow wheels.
32:43And finally, it hit me. All these mixed
32:45flow wheels are heavier, sometimes by
32:48quite a bit. Honestly, I knew that all
32:50along. When you go to print the wheel,
32:52it tells you how much it'll weigh. But I
32:54thought the effect would be marginal.
32:57So, I ran this little test with a sort
32:59of flywheel weight, and my intuition was
33:01correct. The power consumption was only
33:03marginally higher with the weight. But
33:05then I added a little tape on the
33:07flywheel to throw it out of balance, and
33:09power consumption went up significantly.
33:11This made me realize that while a
33:13heavier wheel on its own may only have a
33:15small penalty, a wheel that is out of
33:18balance or not concentrically mounted
33:20could have a major effect. And the
33:22heavier that wheel is, the worse that
33:24effect will be. When I realized all
33:27this, I thought I need to go back to the
33:29mix flow wheel that did worse because it
33:31was heavier and somehow compare the two
33:33at the same weight. So, I printed off a
33:35little weight to add to the backward
33:37curved wheel to match the weight of the
33:39two wheels competing against each other.
33:42And here's the data. The weight does
33:44make a difference. Again, when the mix
33:46flow wheel was significantly heavier, it
33:48did worse than the backward curved
33:50wheel. But at the same weight, the mix
33:53flow wheel is actually a little bit
33:54better. The lesson is we may have a
33:57great design, but if it's heavier, any
34:00performance gains or efficiency gains
34:02may go down the drain depending on how
34:05much heavier it is and how well balanced
34:07your whole system is. So, with that in
34:09mind, I went on to try to design a
34:11better mix wheel. Basically, a version
34:13of this on a diet. We'll stick with the
34:15higher airflow, low pressure design
34:17point because that's well within the
34:19range of mix flow wheels. And I think
34:20that's where we'll see the biggest
34:22improvement compared to a backward
34:23curved wheel. Here, we can tell the
34:25software to have free reign on blade
34:27design, including 3D shapes. Since I'm
34:30sort of manufacturing these wheels with
34:32a 3D printer, I don't really have to
34:34think too much about how complex these
34:36shapes may get. I'm purely just trying
34:38to make the best wheel possible,
34:40regardless of how or if this would even
34:42be manufacturable.
34:44For the dimensions, I changed the inlet
34:46side slightly to fit the test rig. And
34:48off we go. Right away, you can see this
34:50meridian view in a 3D model of our
34:52wheel. This is pretty much where I left
34:54the first prototype, and it ended up
34:56being way too heavy. For now, though,
34:58let's go back to the meridian view. You
35:00can see here at just how shallow of an
35:02angle the air can flow through this
35:04wheel as opposed to a traditional
35:06centrifugal wheel where the flow makes a
35:0790° turn. This mix flow design should be
35:11much more efficient as the air doesn't
35:12have to turn nearly as much. Now, as it
35:15sits, the hub takes up this entire
35:17volume. So, I'll cut that down to save
35:19some weight to something like this.
35:22Next are the blades. Because we have
35:25fully 3D blades now, we can much more
35:28closely, almost perfectly match the
35:30ideal angles calculated by Oilers's
35:32equations. Notice that the leading edges
35:35of the blades are angled pretty much
35:36like an axial fan. And as we work
35:39backwards, they bend more towards
35:41resembling backward curved blades.
35:44There's a lot of fine control within the
35:46software to manipulate these blade
35:47shapes, where they start, where they
35:50end, you know, how many sections they
35:52have. And to be honest, I'm not an
35:53expert in optimizing this. I think it
35:55would take quite a bit of practice. But
35:57one thing I did do is I used some
35:59computer fluid dynamics or CFD software
36:01to help me improve this wheel in general
36:04and to try different blade shapes. I'll
36:06say this though, I thought this part
36:07would be at least a little bit
36:09straightforward. But even modern CFD is
36:12like notoriously difficult to use, and
36:14this was true for me, too. It's it's a
36:18lot. In the end, I ended up mostly
36:20chasing efficiency numbers, and we
36:22didn't stray too far from the default
36:24blades the software suggested.
36:26One thing I noticed when testing the
36:28first prototype was that I was achieving
36:30the target air flow and pressure at a
36:32much lower RPM than designed. I don't
36:34know where the software is getting this
36:36wrong, but I thought about making the
36:37wheel smaller. In the end, though, I
36:39just ended up trimming a bit of the
36:41outlet at an angle like this. This
36:43should cut down on weight a bit and
36:45bring the wheel closer to design RPM
36:47since the average diameter is now a
36:49little bit smaller. So, off to the
36:51printer. Well, that didn't work. Small
36:54tweaks and trying again are the name of
36:56the game here. You may have noticed on
36:58the test rig I have a different inlet
37:00plenum. In fact, it's not an inlet
37:02plenum at all. It's just a straight
37:03cylindrical piece to interface with the
37:05impeller. The fact is this duct is a
37:08little bit smaller in diameter than it
37:09should be to test a fan moving this much
37:11air. At 400 CFM, the air speed through
37:14here will be pretty high at over 2,000
37:16ft per minute. But other than that,
37:18there's really not an issue with using a
37:20straight plenum like this. The reason a
37:22curved inlet plenum is normally used is
37:25because to keep velocities lower, the
37:27duct is usually a larger diameter,
37:30requiring a reduction to interface with
37:32the bottom blade. Okay, enough talking.
37:35The moment of truth.
37:37We finally have a mixed flow wheel
37:39that's outperforming the backward curved
37:41wheel, consuming a gross power of 38
37:44watts versus 46. That's a major
37:47difference.
37:49And this wouldn't be one of my videos
37:51without some more smoke. So, let's
37:53visualize just how differently these two
37:55turbo machines handle the air.
38:00Finally, I had one more mix flow design
38:02to test. the scaled down version of one
38:05of Captiva's mixed flow wheels. The
38:07first wheel that gave me so much grief
38:09to print. And as I was setting up this
38:11wheel, I thought, "This is not really a
38:13fair fight." The other wheel I designed
38:15with no regard to how it's manufactured.
38:17It's 3D printed. I could do whatever I
38:19wanted. This wheel is made out of sheet
38:21metal. It doesn't stand a chance, right?
38:28I couldn't believe it. This relatively
38:31simple wheel design is outperforming the
38:33much more complex wheel I designed with
38:35the help of software and CFD and not
38:38just at one design point. No, it's
38:40edging out the other wheel, especially
38:42at higher air flows where efficiency
38:44reaches 87% compared to 78%. Plotting
38:48these curves, I also see the air flow
38:49and pressure curve is pretty flat. I
38:52think most of it has to do with the fact
38:53that this test rig is a bit too
38:56restrictive at higher air flows. I
38:58wouldn't say this data is incorrect.
38:59It's just maybe a bit incomplete.
39:02Nevertheless, this wheel did amazing. It
39:05kind of surprised me. I'm kind of
39:06scratching my head about how we designed
39:09a wheel like this that's so efficient
39:11but also manufacturable. So, tomorrow
39:14I'm going to be flying to our R&D lab in
39:16Pennsylvania where I'm hoping to get
39:18some answers. It's tomorrow and after a
39:21chill flight I make it to R&D. Out of
39:24all the cool things I get to do in my
39:26job, visiting R&D is always a highlight.
39:29This place is busy in all the right
39:31ways, and I always feel like I'm just
39:32hanging out with my friends. This is
39:34true even with Bill, who's one of the
39:36smartest people I've ever met, and also
39:39the guy running this place.
39:40>> We did take a step back with the mix
39:42flow wheel and said, "Hey, how do we
39:44move a little more air at a good
39:46efficiency level? And how do we make
39:48that wheel fit into our current
39:50products?"
39:51The short answer is a lot of blood,
39:53sweat, and tears and trial and error.
39:56So, we just basically started with a
39:59very simple design. Believe it or not, I
40:01produced that version. And then one of
40:03my engineers at the time, Josh Hes said,
40:06"I can make that wheel better." And I
40:08was happy he said that because he took
40:11this and went through many iterations of
40:14this fan wheel, changing the top
40:17diameters, the number of blades, the uh
40:20blade angles. And while we can use
40:23software to approximate these air flows,
40:26it's a good starting point, but you have
40:29to test test and test. And it starts
40:31with prototyping. And fortunately, we
40:33have a plant two hours from our R&D
40:36location where they were able to rapidly
40:37turn around these designs. We would get
40:40these designs in, put them on our test
40:42duct with an inlet iris and develop all
40:45the parameters of the wheels. The wheel
40:48may work very well by itself in a
40:50plenum. But when you put that wheel
40:52assembly inside of a unit, it actually
40:55performs a little differently. [music]
40:57So the spacing around the wheel between
40:59the fan wheel and the housing that can
41:01cause different reactions [music]
41:03on air flow,
41:05you know, we went through about 20
41:06iterations and [music]
41:08after a lot of testing and a lot of
41:11development, we ended up with our
41:13current design. We've added [music] some
41:15stiffening hems and different features
41:18to the wheel over time to make it more
41:21robust [music] and allow us to spin it a
41:23little faster without wheel failure.
41:26When you're designing wheels, not only
41:27do you have to look at performance,
41:29[music] but you have to look at how well
41:31the wheel will perform over a long
41:33period of time.
41:35And this was over approximately a 2-year
41:37time period from start to finish. So, it
41:40takes a lot of time because when you're
41:42moving air, the fan wheel is the most
41:45important part of what you need to get
41:47right. So, it was worth the time
41:49investment to make sure we did this
41:50correctly. [music]
41:53Some engineers may be tempted to use all
41:55theoretical data to produce their fan
41:57curves. And as we've seen in this video,
42:00that perfect wheel in theory doesn't
42:03perform the same in reality.
42:06So, one thing that we take a lot of
42:08pride in is developing a wheel and then
42:11actually testing those exact geometries
42:14in a real life situation. And our
42:16performance data is all generated from
42:18that real life test data, not from
42:20theoretical software. While this simple
42:22Iris test rig yields good results for at
42:24least comparing different fans, it does
42:27suffer from some of the same constraints
42:28as my little test rig, such as limited
42:31air flows. So, at Captive Air, we're
42:34trying to step up our fan testing
42:35capabilities with the use of an AMA
42:38chamber. So, this is an update to our
42:40AMA chamber. You may remember from
42:41previous videos uh condition that it was
42:43in. It's uh it's starting to shape up
42:46now. So, we've got our supply fans to
42:48the chamber connected. They're powered
42:50up, ready to go. And we're working on
42:53the uh internal guts of the chamber now,
42:56where all the sensors are located, the
42:58pressure pickup points, and the controls
43:01associated with the chamber to allow us
43:03to start testing.
43:04>> I asked Bill how exactly this chamber
43:07works, and I'll summarize by using a
43:09schematic provided by AMC for this
43:10specific configuration. We have the
43:12chamber itself, the inlet fan or fans
43:15which are permanently installed to the
43:16chamber and can be speed controlled and
43:19finally the test fan which is installed
43:21on the opposite side. The guys were
43:23working on a mounting system for this
43:25when I was there. Within the chamber is
43:27a venturi or nozzle wall with an array
43:30of different size nozzles and there's
43:32also these two sets of what calls
43:34settling means. These specifically sized
43:36grits are meant to straighten the air
43:38flow and make it more laminer within the
43:40chamber. We also have some static and
43:42total pressure ports with very very
43:44specific locations specified by AMA. So
43:47when this is running, the test fan spins
43:49at a constant speed and the supply fans
43:52are run at different speeds to change
43:53the inlet pressure the test fan
43:55encounters. In other words, the
43:58resistance. For example, if we want the
44:00test fan to experience zero resistance,
44:03the supply fans are set at a speed that
44:05provides that at the test fan inlet. Or
44:08on the opposite end, if we want the test
44:10fan to experience the most resistance,
44:13the supply fans are simply shut off and
44:15the test fan would experience quite a
44:17pressure drop at the inlet. We can also
44:19manipulate how open or closed this
44:22nozzle wall is by blocking different
44:24nozzles. And really, the nozzle wall is
44:26what makes this chamber really, really
44:28accurate because ultimately the way we
44:30measure air flow through the chamber is
44:32by measuring the static pressure
44:34difference across these nozzles. So
44:36there's published data between measuring
44:39the pressure drop across this nozzle
44:41wall which translates directly into a
44:44CFM. So you can very accurately measure
44:47how much CFM is moving through this
44:49chamber.
44:52Once we have this up and running,
44:54testing fans will be much more accurate
44:56[music] and the data will be much more
44:58reliable than our traditional Iris duck
45:00that we've used in the past.
45:02It turns out there is no magic bullet to
45:05designing better fans. But by combining
45:07theory, testing, [music] and better
45:09tools, we can get closer to harmonizing
45:12with nature instead of just fighting it.
45:15As far as my little quest to make a
45:16better fan wheel, well, for now, it ends
45:19here. I admit defeat, but I'm excited to
45:22see what improvements Captive Air or
45:24anyone else for that matter can come up
45:26with in the future, and how fan designs
45:28continue to evolve.
45:31With that, if you watched this far,
45:32[music] then thank you. Thank you for
45:34coming along in this little journey. I
45:36hope it was as fun for you as it was for
45:38me. See you. [music]

About this transcript

This page contains the full transcript of I Tried to Make a Better Fan by CaptiveAire, generated from the public captions YouTube serves with the video. The transcript has 8,450 words across 1,217 segments, with the original timestamps preserved so you can click any line to jump to that moment in the embedded player.

What you can do with it

Use the transcript to take notes, quote the speaker, build a study guide, generate a summary with ChatGPT or Claude via the YouTube Summary tool, or export it as a timed subtitle file with YouTube to SRT. You can also re-open it in the transcriber to translate the transcript into 100+ languages.

Free YouTube transcript tool

YouTube2Text is a free YouTube transcript generator — no signup, no daily limit. Paste any YouTube link and get the full transcript instantly, with timestamps, click-to-jump, translation to 100+ languages, AI prompts for ChatGPT, Claude, and Gemini, and exports to TXT, SRT, VTT, or Markdown.