John at work had been talking for some time about the Tesla
turbine and the Tesla valve, or more properly the Tesla valvular
channel since Tesla never called it a valve and it is not a valve,
it is a channel with a valve-like property.
Christmas 2019 I looked up the Tesla turbine and found that a
nice gentleman had
created a 3D printed version, which was very convenient, though it
turned out to be only the start of my journey. The Tesla
valvular channel was more problematic; browsing the internet
suggested that efficiency wasn't very good. So I decided
that I would use the pre-designed 3D printed Tesla turbine as my
measurement mechanism for the evaluation of the Tesla valvular
Note that there is somewhat of an evolution in the designs below
as I find out what's good and what's bad so, if you intend to
reproduce any of this yourself, do make sure to read through to
the end before doing so. The sections on this page have
ended up being:
Tesla turbine 1: Integza's
completely 3D printed design as mentioned above, useful as an
Tesla turbine 2: GravInert
(AKA Paul Townley)'s Tesla cube design, demon fast in both
acceleration and speed,
Tacho: because of the speed of Paul's
Tesla turbine I needed to design my own tacho which was also
able to plot speed over time,
I downloaded Integza's Tesla turbine files from Thingiverse
and printed them on my Prusa
3D printer in grey PLA, at 0.1 mm resolution as I wasn't in a
hurry. The only problematic part was the axis which was tall
and top heavy and needed supports to stop it toppling at the stage
where the bridges were being printed.
I obtained two bearings with outside diameter 22 mm, inside
diameter 10 mm, a brushed DC electric motor with outside
diameter 22 mm (which will act as a generator) and a collar
with inside diameter 3 mm to attach to the motor shaft.
I used some 8 mm long M3 hex bolts to hold the turbine
together. I managed to fit nine disks inside the turbine
without any fowling, glueing the last one in place with
cyanoacrylate adhesive. I glued one half of the joint for
the electric motor to the end of the axis and the other to the
collar using cyanoacrylate adhesive; everything else was push-fit
or bolted. I glued a pneumatic fitting to the air inlet with
Araldite (standard, not rapid) to allow me to push-fit a 5 mm
inside diameter/8 mm outside diameter air hose.
I mounted the turbine on a piece of wood, leaving room to fit a
lamp holder and a DC power measurement unit that I ordered from
China; the wires running around the back go to a 9 Volt
battery powering the DC power measurement unit.
To make things simple to manipulate, I use quick-connect pneumatic
fittings on the rest of the tubing. There are as many types
of these as there are people on the planet and they all look very
similar so it is important to make a choice and remember it.
I used the "Series 21" type throughout as they work nicely
with 5 mm inside-diameter plastic tubing and are relatively
small and neat.
Here it is on its first test run, driven by compressed air from my
little Jun-Air compressor:
...and here's the same but with the Watt meter in circuit.
As you can see, lots of vibration and a peak of just over
2 Watts output as the stored capacity of my little compressor
is used up.
Further testing showed that the peak speed at that 2 Watts output
was around 6800 RPM. Remove the motor and that rose to
10500, much less than half the 24000 RPM peak achieved by Integza and
nowhere near achieving sync with the incoming air flow. Now
of course I'd just thrown this together with no attempt at surface
finish, vibration control, air flow, etc. but there were other
issues. My little Jun-Air compressor has a 25 litre
tank versus Integza's 100 litre tank and it is just not able
to keep up with the flow requirements of the model. And
you'll see below that putting my first-cut Tesla valvular channel
in the air flow, with its tiny orifice, stopped the whole thing
However, after posting the first video above on YouTube I was
contacted by Paul Townley, AKA GravInert.
It turns out that he and iEnergySupply,
who I guess is in the USA, are trying to build Tesla turbines
which they believe can extract energy from atmospheric pressure in
damp air. In the comments section beneath their videos
someone had posted a link to the Tesla Engine Builders
Association, which, under the heading "The Open Secret"
includes, such articles as "Conquest of Space - Before It Went
Black" and "Flying Saucers 'Explained'". I was either down
the rabbit-hole or through the looking-glass.
Back to the problem at hand: in order to have a Tesla turbine that
I can use to measure the performance of a Tesla valvular channel I
needed one that (a) required a smaller air-flow and probably (b)
employed a metal shaft. I could have tried Integza's first turbine but since
Paul Townley had been in contact I thought I'd try his first.
Tesla Turbine 2
I downloaded the STL files from the Drop Box link under
video and made two modifications:
the rotors wouldn't print well as a single unit, it was
very difficult to get the support material out from between
the blades afterwards, so I split them up into a rotor centre
and rotor blades that can be pushed together after printing,
file seemed to be corrupted so I created my own handle; this
was designed and printed in three parts which clip together to
avoid the need for support structures when printing.
Here are the Blender
and STL files for
the extras. All the parts were printed at 0.1 mm
resolution in PLA with supports everywhere aside from the
gaskets from my extras (see later) which
were printed in flexible PLA at the same resolution, and (added
later) the base plate from my extras which only needed 0.2 mm
resolution. The parts on the right in the picture below
are Paul Townley's originals, the ones on the left are my
additions; Paul's file tesla_cube_single_stage_baseplate.stl is not
required here. Two each of his "nozzle" and "port" casings
were required; note that I later modified these to improve
bearing fit, hence they moved into my extras file and this also
removed the need for the gasket once more. From my extras
five "open" rotor blades were required, one "closed" rotor
blade, two of the star-shaped rotor centres and two sets of
gaskets. After printing, aside from the usual removal of
support materials, the hole in diverter_block.stl through which rotary_shaft.stl
fitted needed some attention with a half-round file so that the
shaft fitted and turned easily.
I purchased four
bearings of 9 mm outside diameter/4 mm inside
diameter/2.5 mm wide, a length of 4 mm diameter metal
rod, some 4 mm inside diameter collars, four 100 mm long
M4 bolts and a pack of M4 brass inserts, though I realised
afterwards that these latter are only required for a single-stage
design; for the two stage design plain-old M4 nuts are used
I began by assembling the rotors and the handle of the diverter
block. Note the single "closed" rotor blade
positioned in the middle of one of the rotors (the right-hand
rotor in the top middle picture below) to make that the "male"
rotor. Cyanoacrylate adhesive was used to keep the rotor
blades in place on the star-shaped rotor centres, to glue the
handle parts securely to each other and to attach the handle to rotary_shaft.stl,
which allowed the direction of the air flow to be changed, being
careful not to accidentally glue the rotary shaft into place at
the same time.
The main body parts lined up as follows:
The body is held together by the long M4 bolts but, to ensure a
good air seal, I chose to attach the diverter block to its
adjacent piece permanently by gripping the two together in a vice
after applying cyanoacrylate adhesive around the air channels
between them. I cut two 50 mm long pieces of the
4 mm diameter metal rod and hammered them through the centres
of the assembled rotors, then pushed a bearing on either end of
the shaft and test fitted them in their housings. Paul
Townley had warned me to ensure that the rotor blades did not
touch the side, so I was careful how I positioned the rotor on the
shaft and did a little filing of the rotor blades here and there
to be sure.
I found that when both halves of the casing
were assembled on a rotor too much pressure was applied to the
bearing, I could feel the bearing "step" as I rotated it, hence I
printed myself the 1 mm thick gaskets out of "flexible PLA"
to fit between the two casing halves, also acting as an air seal.
The female rotor (i.e. the one with all "open" rotor blades) was
fitted into the casing nearest the nozzle, the male rotor in the
casing furthest away from the nozzle. Here is the dual
turbine fully assembled with a couple of collets attached to the
shafts to which I intended to glue something for my RPM counter to
Being impatient to see how it ran I connected it to my compressor
immediately. The first rotor span up very quickly indeed,
making a very pleasing, if slightly panic-inducing, whining
noise. The second didn't start so I increased the flow;
there was a "tick" noise and suddenly all motion stopped.
The first rotor had simply shattered; now we're talking! It
turned out that Paul Townley had his printed in carbon-reinforced
nylon, which I could do but only with a nozzle change in the
printer. And Paul balanced his rotors. And splitting
the rotor into parts would have weakened the blades. A
redesign of my turbine extras was in order.
I modified my rotor blades to print whole disks which were each
hammered onto the 4 mm diameter shaft, no glue
required. I printed them in the much stronger ASA (UV-safe
ABS, only because I didn't have any plain-old ABS to hand), with
support material on the print bed so that they could retain their
slightly tapered profile since Paul had mentioned that this was a
necessary feature; I hand finished the blades somewhat to make
sure that shaping survived the 3D printing process. I
learned from Paul Townley that the second rotor isn't supposed to
spin-up in the same air flow (I need to read more about how that
is meant to work) so this time I quickly assembled the whole thing
but didn't insert the second rotor and placed the solid tesla_cube_single_stage_baseplate.stl
between the two rotor-casing segments so that the second
part wasn't in use at all (see later
for why this was the wrong thing to do).
Just to make sure I had got past the smashy stage, I spun this up
immediately on the compressor with both exit air holes open.
You can just about hear, at around 35 seconds in, that it
wants to sync and it eventually gets up to about 38,000 RPM,
which is fine for an untuned first try; the little blade is
rotating 633 times per second. And no smashy, though also
not the rather exciting whine of the first version. Fiddling
was required as Paul Townley had achieved more than twice that on
his first run, though I'm not sure at what air pressure; mine was
with an initial peak of 8.5 bar.
I trimmed away some gasket which appeared to be touching the shaft
when the bolts were tightened and put the rotor, mounted on the
shaft, into my little Unimat lathe where I abraded the sides and
edges in an attempt to improve the uniformity/taper of the
blades. Paul Townley had pointed out to me a rotor balancing method
that Tesla had used but it was a little too complex for me to want
to set up such a thing just yet. I also found that if I spun
the rotor by hand with the bearings sitting loosely in their slots
it ran for far longer than if I pushed the bearings down into
their slots, so I used my Herzo hand abrading tool to open up the
bearing slots somewhat and I put the rotors back into the lathe
where I abraded them to reduce them in size very slightly in case
they were catching on the turbine case. I bought some
60 mm long hex bolts so that I could assemble just the first
stage turbine using those M4 brass inserts hammered into the bolt
holes in tesla_cube_single_stage_baseplate.stl
(again, see later for why this was the
wrong thing to do). A final innovation, which Paul had
mentioned but which I hadn't thought would be necessary yet given
my low speeds, was to drop some 3-in-1 oil into the shaft-hole at
either end before running.
Note also that the motor on my little Jun Air compressor can't
keep up with "open-valve" demand so while experimenting I needed
to do only timed short runs, which was fine since pulsed air
flow is where the Tesla valvular channel comes into play
anyway. I added a manually-operated valve to the tubing
delivering air to the Tesla turbine to give me convenient local
What was necessary, of course, was to know whether the rotor was
anywhere near achieving sync with the velocity of the incoming
air, a calculation which Integza talks about at this point in
his video. My compressor had a 25 litre tank and a
peak pressure of 8.5 bar with the compressor switching on
again when this drops to 6 bar. The idea is to work
out the volume of air this difference in pressure represents
once that air has escaped compression and, given the size of the
hole it has to escape through and the amount of time for which
it is escaping, work out how fast it must have been travelling.
Calculating the density of wet air is complicated so I chose to
Calculator. At 60% humidity and room temperature
this gave me 10 kg/m3 at 8.5 bar and
7.12 kg/m3 at 6 bar. A 25 litre
compressor tank is a volume of 0.025 m3 which
gave masses of air of 0.25 kg and 0.19 kg respectively
and so, between the two pressure values, 0.06 kg of air
must had been emitted. Back to Omni
Calculator: at "room" atmospheric pressure (1 bar)
the same wet air had a density of 1.18 kg/m3
which meant that my 0.06 kg of expelled air would have a
volume of 0.07 m3. Changing to more
reasonable units, 1 cubic metre contains 1,000,000 cm3
so the volume of uncompressed wet air is 70,000 cm3.
It took 20 seconds for the pressure to drop from
8.5 bar to 6 bar with the valve on the compressor
completely open so the rate of air flow was
70,000 / 20 = 3,500 cm3/s.
The air entry hole in the Tesla turbine casing, as measured at
the rotary shaft in the diverter block, was 4 mm in
diameter giving a cross-sectional area of 3.142 * 0.22
= 0.125 cm2. So the average linear speed
of the wet air was
3,500 / 0.125 = 28,000 cm/s. The
circumference of my 26 mm diameter rotor was
3.142 * 2.6 = 8 cm, so it should on
average rotate at
28,000 / 8 = 3500 times per second or 210,000 RPM. Quite a lot,
and that's the average number.
With all my improvements what did I achieve? You can see
that the rotation is still relatively poor under hand
encouragement at the start of the video below and there is quite
a lot of vibration when the turbine runs.
A peak of around 60,000 RPM (1000 times per second!),
average probably around 53,000 RPM over the period, with
both exit holes open. I'm about a quarter of the way
there, 25% efficiency in something that can theoretically get to
98%. Makes me wonder if all I'm doing here is measuring
the performance of some bearings.
After I published the video above Paul
Townley got in contact to point out (politely) that I'd been
very silly. I knew what the air flow in a Tesla turbine
was meant to be: from the outer edge of the rotor and then
towards the centre of the rotor to escape from the middle.
I had closed off the two large rectangular holes in the back of
the turbine with tesla_cube_single_stage_baseplate.stl,
leaving the two small holes on either side of the turbine
assuming that they were the intended exit holes:
But that's rubbish, the air was not escaping through the middle
of the rotor. What I wanted was this:
Paul had included the small side holes for some other purpose
because he was sending the objects away to be 3D printed
and hence needed flexibility in the result. Those small
side holes were meant to be tapped and closed off with a screw
when not required, the intended exit holes being the large
rectangular holes. So I screwed an M2 hex bolt into each
of the small side holes, pushed M4 inserts into tesla_cube_twin_stage_baseplate.stl
(with the two rectangular holes in it) instead and reassembled
the turbine. I applied a little cyanoacrylate adhesive on
top of the M2 hex bolts for additional security; I didn't want
small metal bolts exiting the building and entering me at high
I tried again with encouraging results:
Rather than starting at 60,000 RPM and wilting down to
46,000 RPM, the turbine stayed at around 55,000 RPM
for the period. This suggested that it was now
bearing-limited, which I needed to sort out anyway as the
lilting sound is likely a result of the bearing being slightly
loose in the housing due to my fettling.
Paul had suggested that GRW bearings were the best so I went in
search; curiously the two UK agents for GRW were unable to
supply small quantities but a Dutch seller (247industries) on
Ebay could. I also purchased some turbine oil, a thin oil
used with model jet engines (from Kings Lynn model
shop) and some paraffin to mix it with, 5 parts
turbine oil to 95 parts paraffin. Paul suggested that
the bearings be immersed in this when not in use. And I
into my Blender extras file and modified them to remove the
small side holes entirely and increase the width and diameter of
the bearing slots by 0.2 mm so that they would come out the
correct size when FDM printed, no need for the gasket. I
believe Paul's casings where resin printed which has a higher
resolution and so he had no need of this tolerance.
Here are all of the parts before assembly once more: the
diverter block, rotary shaft and twin stage baseplate at either
end are from Paul's files, the rest are from my Blender
extras. Glue (cyanoacrylate) was applied solely to affix
the handle, nowhere else, and no gasket was required.
And I think this time it worked; Paul agreed: possibly the
fastest accelerating Tesla turbine in the world.
As you can see, the tachometer gave up and my measurement disk
flew into pieces that I mostly couldn't find afterwards, apart
from the one that sliced my finger. Very cool and a
wonderful sound too. I belatedly realised that my
tachometer wouldn't go above 99,999 RPM. Gonna need a
bigger measuring device. And a stronger reflective
disk. And some shielding.
During a discussion at work one lunchtime Jonathan suggested I
make my own tachometer using an optical sensor; with this I
could measure acceleration as well as speed. I purchased a
QRD1114 reflective object sensor (basically an LED and a
photo-sensitive transistor in a tiny plastic case) from Hobbytronics
for £1.20, a Raspberry Pi Zero W with the header fitted costing
£13, a power supply for the Pi costing £8 and I dug up an old
8 gigabyte SD card to use in the Pi. I designed a 3D
printable base to hold the Tesla turbine, the sensor and the
Raspberry Pi. I painted length-ways 2/3rds of the
sticky-out metal shaft of the turbine with matt black paint
(actually Humbrol 32 dark grey since I had that to hand) so
that it had reflective and non-reflective sides.
I followed the procedure here
to set up the Raspberry Pi Zero W headless on my Wifi network,
i.e. such that I could log into it from a terminal program (e.g.
PuTTY) from anywhere rather than having to connect
screen/keyboard etc. and use a GUI. I followed the advice
on how to connect the sensor to the Pi, using one of the
3.3 Volt power outputs from the Pi to supply the LED and
connecting the sensor output to a digital input pin on the Pi,
rather than an analogue input pin (so that I could use
interrupts), checking with a multimeter to make sure that it
achieved the necessary 0.8 V to 2.4 V voltage swing
for a digital input pin between the dark and light sides of the
shaft. Below you can see the voltage swing on a scope when
the rotor was spun by hand.
I wrote a Python script to run on the
Raspberry Pi which counted the number of falling edges on GPIO4
in each 100 ms interval and saved the values to a CSV file
that I could load into Microsoft Excel. For the first run
I also attached a logic analyser to the optical sensor output
pin to capture a live trace so that I could cross-check the
accuracy of the program.
I lubed up...
And away we went.
Not too shabby. A steady state of 185,000 RPM, ~90%
of my very roughly calculated theoretical maximum of
210,000 RPM given the output of my compressor. And
that kick in the curve followed by the flatness makes me wonder
if the rotor had achieved sync with the incoming air flow.
The logic analyser output agreed with the above: below are a
couple of clips (upper trace digital 3.3 V logic, lower
trace analogue), one of the startup and another from somewhere
around the peak showing a nice clean sawtooth with the number of
transitions over time to match the graph. So the data is
And, as independent proof, John at work had pointed out that I
could probably just listen to the turbine and the audible
frequency would likely be that of the rotation speed. To
test this, here is a 30 second burst of a 3.08 kHz sine wave
(AKA 185,000 RPM) as an MP3 file generated using Audacity. Try
playing it at the same time as the video and you'll see that the
tones match. I tried exporting the audio from the video to
an MP3 file and using Audacity to analyse the spectrum to see if
it would match the speed curve but unfortunately there is far
too much general hissing to pick out the wanted rotor tone; ears
are more clever.
During the initial jump to 100,000 RPM in just one second
the surface speed of the rotor (diameter 28 mm) went from 0
to 147 metres/second, so an acceleration of
147 metres/second/second, Now I thought that was
15 g (147 / 9.8) but after another lunchtime
conversation David, who once worked on centrifuges, pointed me
link about converting RPM to g force which says that
relative centrifugal g force (RCF) is (RPM)2 * 1.118 * 10-5 * r,
where r is the radius of the rotating thing in
centimetres. That calculation makes the force
53,000 g at max RPM.
Then I remembered that one of my brother-in-laws, Toby, works in
the design of fans and he confirmed that, though the RCF number
is interesting, the number I was looking for was hoop stress,
the tangential force which is at a maximum at the edge of the
rotor. The equation for hoop stress is (density * r2 * w2) / 3,
where r is the radius of the thing in metres and w
is the speed in radians/second. The density of ASA is
virtually one, the speed in radians/second is
(2 * 3.142 * rotations per second)
so for my 28 mm rotor this becomes (0.0142 * (2 * 3.142 * (185000 / 60))2) / 3 = 24,500
Newtons. For a rotor with disks of thickness 2 mm the
surface area over which that force is applied would be
2 * 3.142 * 28 = 175 mm2.
24,500 Newtons applied across 175 mm2 is
24,500 / 175 = 140 Newtons/mm2.
Checking on the
web, FDM-printed ASA seems to have a tensile strength of
30ish megaPascals, AKA Newtons/mm2.
Hence with these maximum speeds I'm likely running at more than
four times the strength of my material; Toby also mentioned that
they put concrete walls between them and their experimental
Anyway, the rotor survived without flying apart, thankfully, as
I hadn't installed a shield. The surface speed of the
rotor at max RPM was 271 metres/second. This
article suggests that a .38 calibre pistol, apparently a
popular small hand gun, has a muzzle velocity of
286 metres/second and, interestingly, fires a bullet
weighing 6 gm, just twice the weight of the rotor, though
it is only able to inflict as much damage as a hockey puck at
full pelt. But I'd better make that shield.
To reinforce that point, I thought I'd try a second run and, two
seconds in, there was a bang as the disks of the rotor suffered
rapid unplanned disassembly; 192,000 RPM and rising this
time. Maybe the dip and then rise in the acceleration on
the first run was fortuitous? I will open the air inlet valve
more slowly in future.
While I had been sorting the tachometer out, Paul had suggested
a number of additional experiments: increase the number of disks
and try using an air bearing. I thought it was worthwhile
seeing if I could make the rotor edge go supersonic;
340 metres/second (232,000 RPM for my rotor diameter)
looked like it could be feasible. But first...
Balancing Machine For Littul Rotors
Paul pointed out that if I am going to attempt mach one,
and air bearings, I really needed to balance the
rotor. Paul and his colleague in Idaho had made a rotor
balancing machine of the type that Tesla patented so I
did the same but rather smaller. Here are the Blender and STL
files for the design. The principle is that the rotor
on its shaft/bearings is rotated while the shaft is
suspended on springs; an abrasive surface is then brought up
to the perimeter of the rotor. This is left running for a
few hours and the net effect is that material is removed
from the sides of the rotor that push out further because
they are heavier, resulting in balance.
To accommodate the very light rotor I used tension springs
with a very low Youngs modulus, 0.011 Newtons/mm,
27 mm long from www.springmasters.com
(L40350). The abrading clip in the design was intended
to accommodate those 4 mm x 4 mm sanding sticks
one can buy from Amazon
or a section of 4 mm x 4 mm Kemet Gesswein stone
one can buy from Cutwell,
cut to 33 mm in length. One of the
100 mm long M4 hex bolts I originally purchased for the
double-rotor design above
was re-purposed, thread extended with an M4 die to encompass
the entire length of the bolt, to act as the adjustment
The parts were printed at 0.2 mm resolution in PLA with
supports everywhere. I included two different forms of
upright: one (with the large hoop) is for the balancing process,
the other is for the testing process, see later; two copies of each and two
copies of the bearing ring were required.
The abrader plate was pressed into mating with the hinge
mounts on the base by scrunching them together in a soft-jawed
vice. Three springs each were used to suspend one of the
bearing rings in each of the hoop-shaped balancing uprights,
taking care not to stretch the spring during assembly, then the
loops on the springs were adjusted with a pliers so that the
bearing rings were visually centred with reference to each
other. A length of abrading stick was cut to size and
mounted inside the abrader clip and the clip was clipped onto
the abrader plate. The head of the modified M4 hex bolt
was pushed into the bottom cylinder and the bottom cylinder was
then clipped into its recess in the base. An M4 nut was
screwed onto the bolt and then the top cylinder was pushed on
after it and clipped into the recess in the abrader plate; thus
the height of the abrader plate could be adjusted.
Since my compressor was not very powerful and this kit would
need to run for several hours I designed myself the tiniest
nozzle, maybe 0.5 mm in diameter, providing the fastest air
flow for the smallest volume of air expended, and pushed it into
a length of 5 mm inside-diameter plastic tubing; here are
the Blender and STL files for the
nozzle, the business end of which was hand-worked with a few
pins after printing at 0.2 mm resolution (a higher
resolution would probably have been better). The nozzle
was held in the correct position with one of those "extra hands"
thingies, deliberately mounted separately from the balancing
machine to avoid passing on any vibrations and, for the same
reason, the balancing machine base was screwed down onto a nice
solid piece of wood.
Since I had no idea whether this
was going to work I arranged to measure the vibration of the
unbalanced rotor first. For this, instead of the hooped
balancing uprights, I inserted into the base two copies of the
testing uprights. With the rotor on its shaft/bearings
held in the testing uprights I could run the rotor while my
mobile phone was placed on the base running a mobile
phone app that measured the amplitude of the vibration
using the phone's accelerometer.
Right then, back to the plot to see the results...
Tesla Turbine 2 With Five
I modified Paul's design to accommodate five rotor disks and to
marginally reduce the aperture of the air inlet (more like
3 mm than 4 mm): here are the Blender and STL files, where
everything is basically 11 mm wider.
I printed the components on the right in PLA at 0.1 mm
resolution, the base in PLA at 0.2 mm resolution, the rotor
disks in the middle in ASA (ABS is also fine) at 0.1 mm
resolution (four outer rotor disks are required) and the
components on the left, the two turbine casing halves, in
polycarbonate at 0.1 mm resolution. This latter was a
trial in anticipation of moving to air bearings; polycarbonate
is very hard (the hardest material I can 3D print without a
nozzle change, and a bugger to print): I hope it will have more
chance of withstanding the load of the metal rotor shaft before
the air flow centres it. All parts were printed with
supports everywhere and, in the case of the ASA and
polycarbonate prints, a brim/skirt to keep them stuck
down. After hammering the rotor disks onto a 60 mm
length of 4 mm diameter stainless steel shaft I balanced
the rotor in the Tesla rotor balancing machine.
First, I did some runs with the balance testing uprights to
record the initial vibration: with the bearings lubed-up I
measured the vibration while sweeping the rotor speed up to
60,000 RPM, estimating the speed by running an on-line tone
generator at 1 kHz and slowly adjusting the air speed
until the sound from the rotor matched. The vibration did
not rise linearly with speed, obviously there was some form of
resonance going on with the worst case of +/- 0.5 g
occurring at somewhere around 30,000 RPM.
With that done I began the actual balancing process. After
turning on the air I adjusted the nut beneath the abrader plate
so that the turbine was just able to run and only occasionally
touched the abrasive. I left this running for three hours,
coming back every so often to add more lube and to increase the
speed of the air flow if the rotor seemed too stable.
Material had been taken off the rotor but had this done any
balancing or had it just polished random bits of the rotor
edges? I repeated the vibration measurement using the
balance testing uprights.
The answer was "maybe". If one discounted the resonance at
around 30,000 RPM as purely some artefact of how I happened
to throw things together the first time around then the result
was actually worse than before. If the resonance was
something which the balancing had removed then it was an
improvement. I needed a more reliable vibration
measurement mechanism. And I needed to address some more
obvious problems: simply hammering the disks onto the shaft
doesn't make them square and the balancing machine isn't going
to fix that, click on the segment of video below for a closer
I needed to make myself a tool to help out with that and then I
could maybe balance each disk individually in the balancing
machine for a better result.
It's rabbit-holes all the way down.
if you're not interested in the construction of the turbine.
I paused to make a few improvements in my construction
method. First of all I worked on the individual rotor
disks to make each of them as balanced and symmetrical as
I took a 60 mm long M4 hex bolt and screwed a nut all the
way onto it tightly so that the hex bolt could be held properly
in a lathe, then I screwed another nut a little way onto it
followed by a rotor disk and another nut to hold the rotor disk
firmly in place; I happened to have some large square M4 nuts
and used these as the latter two to help make sure the rotor
disk was perpendicular to the shaft. This assembly allowed
me to hold the rotor disks individually in my little lathe while
I finished them with abrasive sticks such that they each had a
smooth and symmetrical taper.
I 3D printed (Blender
and STL files here)
a rotor alignment tool in PLA at 0.2 mm resolution (no
supports required) that could be used to hold the stack of rotor
disks at 90 degrees to the 4 mm diameter shaft during
I purchased an ADXL362
accelerometer board from Hobbytronics
and printed a mount for it that could be glued (with
cyanoacrylate adhesive) to the top of one of the bearing rings
of the Tesla rotor balancing machine. I redesigned the
Tesla rotor balancing machine base (revising the Blender and STL
files) to include fittings for a Raspberry Pi Zero W and
connected the ADXL362 accelerometer to the Raspberry Pi Zero W
as shown below. While I was at it I reduced the gap
between the uprights by 10 mm (i.e. the abrasive strip was
now 23 mm long) so that the shaft would stick out and added
a mount for the optical sensor. I could now extended my Python script (I also needed to
enable the SPI interface on the Raspberry Pi Zero W and install spidev) to
measure both peak vibration and speed (in 100 millisecond
intervals) during balancing, no need for the mobile phone app or
the testing uprights.
I took a 60 mm long M4 hex bolt and extended the thread
most of the way down it with an M4 die (just under 20 mm of
unthreaded bolt left). I painted the "marker sleeve"
component from the redesigned Tesla rotor balancing machine
length-ways half matt black (or Humbrol 32 again) and half
gloss white (Humbrol 22) to act as the marker for the
optical sensor. A bearing was pushed into each bearing
ring. It was important to ensure that the rotor disk was
square on the bolt and so I used the top-half of the rotor
alignment tool as a jig to hold the rotor disk square while I
initially screwed the bolt into it. Having done this I
unscrewed the bolt again and, with the non-accelerometer upright
held in my hand (i.e. out of the base) I pushed the M4 hex bolt
through the GRW bearing and carefully screwed the rotor disk
onto the bolt once more, allowing it to follow the same
thread. Then I pushed the upright into the base, pushed
the bolt through the other GRW bearing and finally screwed the
"marker sleeve" onto the sticky-out screw end of the bolt until
both it and, on the other end, the head of the M4 bolt were
loosely touching the bearing.
I lubed up and performed a measurement run (i.e. with the
abrader plate retracted down onto the base) on the first rotor
disk before it was balanced:
As the speed was increased, around 30 seconds in, above
about 25,000 RPM, the rotor disk started to wobble; the
spikes on the peak vibration measurement got larger and the
speed dropped. A speed of less than 27,000 RPM seemed
to be about right. Tesla mentions in his Tesla rotor
balancing machine patent that the momentum of the spinning disk
is likely what stops it bouncing off the abrasive surface in a
chaotic manner; given that each of these rotor disks weigh only
1 gm speed has to compensate for the lack of weight.
In order to be certain I repeated the measurement five times,
getting the rotor disk to stability at around 25,000 RPM
each time, and then (with a little help from here
I persuaded Excel to plot
the peak vibration distribution when the rotor disk was within
its high-but-stable RPM range (24,000 to 27,000).
The aim of my balancing process was to reduce the mean of the
peak vibration of this rotor disk from 1.83 g, moving the
bell curve to the left. After several days of trial and
error, during much of which I only made the balance worse, I
determined a good procedure to be as follows:
Clip a length of the roughest sanding stick into the
Without any air flowing, bring the abrasive up to about
2 mm away from the rotor disk. I positioned a
light on the opposite side of the rotor disk to make this
distance easier to see.
Switch on the air flow and bring the rotor disk up to
about 25,000 RPM or as high as you can without
Carefully rotate the nut to increase the height of the
abrasive until it makes contact with the rotor disk: when
the speed drops as a consequence of fiddling with the nut,
rather than through natural variation, contact has been
made. Don't wiggle the uprights or move the base in
relation to the air nozzle while making the
adjustment. Take it to the limit and, if the rotor
disk begins to oscillate wildly, steady it with your
thumbnail, back the abrader off a little and try
again. See the video below for how I performed this
The edge of the rotor disk will become shiny; leave it to
run and, over time, the wobble will reduce. Continue for as
long as you can, at least 30 minutes, readjusting
abrader height and rotor speed as required to ensure that
contact is maintained between the abrasive and the rotor
disk but that the rotor disk doesn't oscillate wildly or
The video below shows the adjustment procedure in detail.
So far so dandy but how consistent was the improvement? I
spent a day performing more half hourish balancing runs on the
same rotor disk as above with various balancer configurations
one after the other, measuring the change in the mean of the
peak vibration in the 24,000 to 27,000 RPM range. For
each balancer configuration I plotted the effectiveness factor,
which I defined as the reduction in the mean of the peak
vibration divided by the duration of the balancing run.
This graph suggested that the method described in the video,
basically the roughest sanding stick adjusted to be as close as
possible to the rotor disk, gave the greatest likelihood of
achieving a reduction in peak vibration; not a surprise but this
In cases where I found I was making things worse rather than
better I returned the disk to the lathe to restore the tapered
edge and then I was able to work the vibration down
further. However, the lowest number I achieved for the
mean of the peak vibration was 1.23 g, which was still quite
high. The rest of the vibration was likely a matter of
dynamic (as opposed to static) imbalance due to yawing arising
from the rotor disk not being perfectly square on the M4
bolt. Given that, in the balancing machine, the rotor disk
is at the centre of a 60 mm long bolt this yawing probably
has a large effect out at the end where the accelerometer is
mounted. I was hoping it would be fixed through the stack
of rotor disks being held square by the rotor alignment tool
during assembly onto the shaft.
I balanced the four female outer rotor disks for the five-disk
turbine, achieving mean peak vibrations (in the RPM range 24,000
to 27,000) of 1.23, 1.36, 1.52 and 1.50 g. I found it
pretty much impossible to balance the central female rotor disk:
the lack of fins on that rotor gave the air flow little to bite
to on and there was no easy way I could make it stable
presenting the rotor edge to my air feed when the abrading had
flattened that edge, making the aerodynamics worse. I put
the central female rotor disk back in the lathe, restored the
tapered edge and left it at that (1.95 g).
After balancing the disks I weighed the female outer rotor disks
using a jewellers' weighing scales and made sure their weights
were identical. Borrowing a technique from my autogyro
project I added Balsaloc (but see below) to any rotor disk that
needed to weigh slightly more to ensure lateral balance, adding
it evenly around the centre of the disk; this only applied to
rotor disk one, from which about 0.08 gm of material had been
removed in all the messing about above. With that done I
slightly rounded the end of the 4 mm shaft with a file in a
lathe to make sure it could be eased into the rotor disks
without carving off any material. Then I put the stack of
rotor disks into the rotor alignment tool and forced the shaft
through them (mostly applying pressure with a vice, a hammer
only required for final adjustment) until the disks sat at the
centre of the shaft (i.e. with equal lengths of the shaft
sticking out of the rotor alignment tool at either end).
Finally I rotated the disks as necessary to align the holes in
Out of curiosity I put the entire turbine rotor, on its shaft,
into the balancing machine and gave it a spin to measure the
Pretty steady, though I should have
let the Balsaloc set first. In fact, splitting the data into
the X, Y and Z axes it was plain that the maximum of the mean peak
vibration in any axis was actually due to the Z axis blip when the
Balsaloc flew away, so the meaningful maximum mean peak vibration
value was 0.23 g and in the Y axis, that of the shaft itself,
a sideways motion. I cut out the Balsaloc, which never
seemed to set entirely and could have been dangerous inside the
turbine casing: better to have the slightest of imbalances, small
as compared to the weight of the rotor shaft anyway.
Turbine 2 With Five Rotor Disks And A Shield
I screwed the five rotor disk Tesla turbine base onto a solid
piece of wood. I had some 2 mm acrylic sheets made up
one 140 mm x 280 mm with counter-sunk fixing holes at
each corner and another 220 mm x 100 mm. The first
I folded into an "n" shape with the aid of a blow-torch following
the advice here
and the other I attached removably to the rear of the "n" with the
aid of a couple of 3D printed slots glued
to the sides of the "n" with cyanoacrylate adhesive. The
front of the shield was left open for filming.
I installed the turbine, lubed up and away we went.
Hmph. Not so fast. From the slow roll-off once the
air was turned off it was clear that there was little frictional
drag on the rotor, yet the top speed of just 106,000 RPM was
not much over half that of the three-rotor disk turbine. Maybe the
air was not spreading out across the surface of the five rotor
disks properly, and of course I had constricted the air flow
slightly to increase the air speed. There was not much scope
for making the rotor disks thinner such that the rotor was less
Comparing the detail of Paul's original air inlet (below left)
with my modified version for five rotor disks and a slight
constriction (below right), his was clearly smoother.
So I went back and did it again.
While I was at it I made the top and bottom inlet nozzles
different sizes, one the original 3.8 mm diameter, the other
the newer 3.1 mm diameter, so that I could switch between
them using the inlet control knob. I printed the inlet and
outlet casings again in PLA at 0.2 mm resolution; printing in
polycarbonate is a pain and, when I need the extra toughness for
air bearings, I can print just an air bearing insert in
Running the revised design demonstrated that making the nozzle
inlet smaller to increase the air speed did slightly increase the
rotor speed achieved. However the improved shaping of the
inlet had no beneficial effect.
Another possibility was that air was leaking away down the sides
of the rotor. I had left a little more room to make sure
there was no contact between the rotor and the casing and
this amounted to 2 mm at either end of the rotor.
I modified the casing to reduce this to 1 mm and tried
Nope, about the same. I wondered if there was too great a
gap between the outer edge of the rotor disks and the casing but I
didn't think it was any larger than the original three-disk rotor
Last thing to try: split the air inlet so that there were two
separate ones; in other words rather like a pair of the original
three-disk rotors side by side, making each aperture smaller so
that the air speed within the turbine remained the same, just with
This was better, with a peak of 109,000 RPM, but still didn't
compete with the three-disk rotor turbine. Hmph. Where
to go next? Was there anything else I could do or should I
cut away the outer two disks and see what happens with a balanced
three-disk rotor solution? Paul wondered if the additional
weight of two more rotors was pressing on the bearings.
Time, I thought, to try making an air bearing.
Turbine 2 With Five Rotor Disks And Air Bearings
I redesigned the "split air inlet" version of the casing halves
above and added air bearings as follows:
removed the outlets at the rear: the air isn't leaving that
way any more,
increased the space for the rotor on either side by
1 mm again to give it a little headroom,
removed the shapes that held the bearings, creating instead
a half cylinder with a small step at the inner edge,
created an air bearing to fit in that cylindrical space,
with a funnel on the inner side, a 4.5 mm diameter hole
(for the 4 mm diameter shaft) and a smaller funnel on the
created an air buffer, designed to fit tightly onto the
rotor shaft at either end and sit loosely inside that smaller
funnel, the intention being to encourage the rotor to remain
created a tool to help fit the air buffers onto the rotor
These parts are all in the Blender and STL files. I
printed the cylindrical parts at 0.1 mm resolution, the
casings at 0.2 mm resolution, all in PLA (can switch to
polycarbonate later if required) with brim and supports
I sanded the paint off the end of the rotor shaft, put the air
bearings in place and then scrunched the air buffers into place
using a soft-jawed vice with the buffer fitting tool, holding the
casing in place as I did so in order to judge the distance
carefully: each air buffer needed to just prevent the rotor disks
from touching the opposite side but I also didn't want to push the
air buffers in too far as they must normally turn without touching
I re-applied the paint, assembled the turbine and gave it a spin.
Not so hot. After about 15,000 RPM it went into some
sort of oscillation and dropped right down to 3000 RPM;
question is, what sort of oscillation? First I tried
decreasing the diameter of the hole in the centre of the air
bearing to 4.3 mm. I ran this without the air buffers
fitted, just to see how it behaved: now the oscillation occurred
at 35,000 RPM: a positive step. So I decreased the
diameter to 4.15 mm: too tight, the rotor failed to turn and
I could feel air escaping from the diverter block rather than out
of the bearing holes. I moved up to 4.2 mm diameter:
this felt right, little play, but (without air buffers) it got no
further than 23,000 RPM and I could still feel air escaping
from the diverter block.
With the casing open I could spin the rotor with my finger and
make it emit the same oscillating noise so I reasoned that the air
probably wasn't escaping through the bearing holes as
intended. I made a version of the diverter block without a
rotary shaft to eliminate the leakage and tried again.
Up to 39,000 RPM now. I thought the "tick" sound was
likely to be the rotor really wanting to do its stuff but hitting
the sides of the casing; it was time to see if the air buffers
helped stop that. Also, there was some leakage of air
between the diverterless block and the rotor casing: I sealed that
up with some Loctite 574 sealant I happened to have lying
around. With the air buffers (slightly increased in diameter
from the originals pictured above) fitted, I tried again.
And they made no difference whatosever, same "tick" noise.
Also my sealant refused to cure, even after 24 hours.
I wondered if the air flow was too turbulent. I redesigned
the air bearing to add rifling: a pair of 0.5 mm deep,
1 mm wide spiral notches running down the inside of the air
hole. I printed the rifled air bearing in PLA at the maximum
resolution of my printer, 0.05 mm, no supports required.
In order for the rifling to have any effect I increased the length
of the air bearing to 15 mm. Of course the shaft should
be at least as long but since a new shaft would require a new
rotor I first did a test with the current rotor to see if there
was any positive effect. I had to snap the tacho tower off
as it fouled the sticky-out air bearing so I couldn't monitor
speed directly during the test. And this time I used some
Araldite (standard) around the exit air hole of the diverterless
block to seal the turbine during assembly.
Of course the answer was that rifling made no noticeable
difference in this test. At least the Araldite worked.
I decided to persist: I had a feeling that rifled air bearings
could be the thing and they would need a longer shaft to be
affective. Back to three disks, where I had most success,
and experiments with rifled air bearings.
2 With Air Bearings
While I was waiting for a 3D print to complete I spent some time
fiddling with the open rotor in its air bearing, using the nozzle
from the Tesla rotor balancing machine to direct the air flow
freely, and noticed something interesting:
Maybe all I needed to do to keep the rotor centred was to split
the air flow, just as I had done before, but this time to point at
the gaps between the rotor disks? While I was making
modifications I made the tacho tower a separate component so that
it could be slid back and forth as required. And I modified
the design of the now diverterless block to make a better air seal
and printed myself a gasket to resolve the leakage problem without
having to resort to glue. Here are the Blender and STL files for all of
the parts, printed at 0.2 mm resolution in PLA with supports
everywhere (asking your slicer program for higher resolution
around the nozzle openings if you can) apart from the gasket which
was printed in 0.2 mm in flexible PLA (no supports required).
It took several attempts to get the diameter and positions of the
inlet nozzles correct; I tried to make the nozzle holes pretty
small, like the nozzle used with the Tesla rotor balancing
machine, to obtain maximum air speed. I performed an
open-casing test run with my final configuration.
So dual inlets weren't enough, the suction effect of air escaping
out of the side of the casing brought the rotor with it, randomly
to one side or the other. I needed to concentrate on
stability over speed.
I redesigned the rifling on the air bearing to be much shallower
and have more threads to it; more like true rifling, just viable
when printed at the highest resolution of my FDM 3D printer
(0.05 mm) in PLA. In order to avoid having to balance a
new rotor/shaft I reduced the length of the air bearing once more
and tried again fitting air buffers on the ends of the shaft, this
time making them much larger in the hope that they would have a
That was at least interesting: the ker-chunk noise emitted from
the closed turbine didn't seem to be due to the rotor hitting the
sides of the casing after all; no wonder nothing I was doing was
making a difference. Could it simply be due to yawing of the
shaft somehow? It seemed to me that the possibilities were:
Air bearing hole too large, permitting
I reduced the aperture of the hole in the
air bearing again, to the point where it was initially too
stiff, and then rotated the rotor shaft in a drill while
in the air bearings for a minute to loosen it.
No difference, well maybe a slightly
higher speed before each ker-chunk but still predominantly
Air flow is too little to support the
weight of the rotor.
Re-try with larger air holes now that the
seals have been improved.
To be continued...
Variation on the above: the rotor is too
heavy for the air flow.
Try again with an aluminimum shaft.
To be continued...
Air flow is too large through the bearing
holes, after all it seems to work when open.
Re-open the holes in the end-plate.
To be continued...
The shaft is too short, allowing yawing.
Try again with a longer (aluminium) shaft.
To be continued...
Rifling is the key but it needs a resin
printer, FDM won't cut it.
Buy a resin printer.
To be continued...
To be continued...
Tesla Valvular Channel
I took the picture of the Tesla valvular channel from the patent:
...and followed the procedure described in this
YouTube tutorial to import it into Blender as a usable
shape; basically, from a fixed camera viewpoint, trace around the
salient parts of one section of the shape with Bezier curves and
then duplicate/join that three times to make a set of eight.
Finally I extruded the 2D form vertically and made it into a solid
that could be printed, 115 mm long. I did try a version
smaller than this (90 mm long), as I had the impression that
the pressure exerted on the valvular channel is a factor and the
smaller the geometry the less external pressure would be required,
however 115 mm long is about as small as my 3D printer could
print the features of the valvular channel distinctly. Here
are the Blender and STL files, which need to
be printed with supports on the print bed only (see below for
Here's a cross-section of what it looked like in my printer's
slicer program and during printing at my 3D printer's highest
possible resolution, 0.05 mm, in PLA:
I wanted to be able to push on the 5 mm inside-diameter
plastic tubing so I modified the outside of the valvular channel
to be round, like a pencil, and tapered the ends. Here's a
quick tour, going through it the "wrong" way, courtesy of a couple
of days rendering from Blender.
This had to be printed with supports (on the print bed
only). Even then I had some trouble stopping leaks from the
connection to the tubing so cyanoacrylate adhesive was applied to
the tapers as the tubing was pushed on.
I placed my Tesla valvular channel in the air-flow to the turbine
and... nothing happened. Either way around, the tiny
aperture of my valvular channel constricted the air flow so much
that the turbine would not rotate at all. Either I needed to
make the valvular channel bigger or I needed a turbine which could
work with a reduced air flow; possibly both. To be continued once
I've bottomed-out the Tesla turbine part...