2.75" Fusor Setup and Current Plans

[Liam David]

Hello all,
This'll be a long one.

Main Chamber and Vacuum System
The main fusor chamber is a 6-way 2.75" conflat cross. The viewport faces the floor, is protected internally by a borosilicate glass disc, and is covered by a soldered 1/16" lead cup to reduce x-rays. An older phone looks through a small hole in the cup and a mirror gives it a view of the grid. It streams to another phone using some generic app. I used to have another viewport on the front of the chamber, but during a higher-power run an electron beam cracked it.

A tee on the right side of the chamber connects through a right-angle valve to a 50l/s turbo pump (Pfieffer TPU-040). The other tee port connects through a radius right-angle bend to a Pfeiffer PRK 251 full range gauge. Opposite the tee is a 4-way cross with a viewport, secondary HV feedthrough, and butterfly valve which connects to a RGA (100AMU Spectra Vacscan from the 90s) and a second 56l/s turbo pump (Pfeiffer TPU-056). The base pressure with both turbos at full speed and not throttled is, according to the gauge, about 2e-6 torr. Since the residual pressure is mainly water, however, I'm inclined to trust the RGA's reading of ~5e-8 torr. The latter turbo and butterfly valve allow me to differentially pump the RGA, meaning I can sample the main chamber while it is at fusion pressures. With the valve cracked I can keep the main chamber in the 100mtorr range while the RGA sits comfortably at the low e-5. It's worked well so far, confirming deuterium purity in the chamber (>99%) and giving me an appreciation for the complexities of gas measurement. The tritium that I thought I saw at m/z = 6 (from the lecture bottle enrichment) was instead D3+. More importantly, it has revealed many hydrocarbons in the 20, 30, 40, 50 m/z range that I need to get rid of due to their charge exchange cross-section. A backing pump oil change didn't solve the problem and it's been a while since I've had acetone anywhere near the chamber, so the source remains a mystery. I'm slightly suspicious of the 50l/s turbo, however, as it outgassed badly for the first several weeks.

Since much of that makes little sense without pictures:

There's lead on two of the joints since they use viton seals. X-rays like to shoot out of there at pretty low voltages.

Turbo controllers and RGA power supply in lower left. Deuterium gas and regulator in center. X-ray transformer in the white bucket and Spellman in back right.

Both turbos are backed by a 2-stage rotary vane pump (a cheap refrigeration pump that can pull <20mtorr) through KF16 fittings and hoses and the pressure is monitored by a 275 mini convectron gauge. I recently replaced the manual valve in the backing line with a NC solenoid valve to protect the turbos in case of power loss. I had the breaker trip twice, and since the roughing pump doesn't have an integral safety valve, the line hit atmosphere within seconds. Needless to say, the 50l/s turbo, which was the only one running both times, audibly complained and decelerated from 90kRPM to almost standstill in less than 15 seconds. Thankfully it didn't crash, but never again. My reaction time with the HV on is only so fast, and since I'm setting up remote operation and can't exactly run 20ft in a second, the upgrade was a must.

High Voltage
The primary power supply is a Spellman DXM70N600 modified to regulate voltage and current instead of shutting off upon exceeding the setpoints. Since it's made to run an x-ray tube it has a floating filament supply and additional safety features. A simple jumper swap on the control board and uploading new firmware made to run the SLM series supplies fixed those issues. Like I mentioned in my intro post, it has a rather sensitive arc sensing feature, quenching for a minimum of 100ms and ultimately shutting off if the arc frequency exceeds a programmable value. This presents a problem since small imperfections in the grid and feedthrough insulator protector (discussed later) create localized electron jets at >35kV. The supply sees these as arcs and extinguishes the plasma, requiring I raise and lower the pressure to properly reestablish it. Another challenge with using the supply is its 4-pin Claymount CA11 receptacle and my unwillingness to fork over a couple hundred bucks for a cable. I ended up 3D printing a connector and potting it in silicone with some 150kV wire. Since the wire came unshielded, I added a grounded tinned copper braid and braided sleeving to reduce static buildup and improve safety. Some o-rings and silicone vacuum grease around the perimeter seems to do the trick. So far is has withheld up to 50kV without issue.

Between the supply and feedthrough is a 60k ballast resistor, mounted on top of the feedthrough. While it gets hot, forced-air or oil cooling is unnecessary at this stage. Which brings me to the feedthrough...

Spellman connector pre-slathering in silicone grease

Feedthroughs
The feedthrough is inspired by Doug's and Andrew Seltzman's designs. The current revision consists of a 25mm OD, 21mm ID, 350mm long quartz tube sealed with a conflat compression adapter. The outside HV end is a 1" Swagelok cap with an o-ring to seal the male threaded section to the tube. Only the top ferrule is still in the fitting, acting as a washer to eliminate friction as I tighten it. To connect the resistor and cable, I 3D printed a piece that slides over the cap and screws into place. A threaded rod touches the Swagelok cap and allows me to attach the resistor directly to the feedthrough. I cut down on the corona using large washers on top of the threaded rod It's not an ideal solution, but works well enough for now. Inside the feedthrough column is a threaded rod nested in a 1/4" OD SS tube to reduce the E field. It is held in place by a 25mm OD washer pinched between the quartz and Swagelok cap and kept centered using a ceramic bushing at the grid end. Having learned of deuterium reducing quartz insulators to silicon, I decided to install a ceramic shield between the grid and quartz. It consists of a 1.25" OD ceramic washer and another ceramic bushing, allowing it to just fit inside a 2.75" conflat. It effectively blocks off the sidearm where the insulator passes through and leaves no easy path for ions or fast neutrals to do their thing. The tube is prevented from shooting into the chamber by several layers of kapton tape around the glass.

Feedthrough fully assembled

Ceramic quartz shield

I've pushed it to 50kV without plasma and there are no signs of arcing or other issues. With plasma the max was about 40kV, which is when the electron jets off the ceramic imperfections caused issues. I considered using boron nitride but figured I'd hold off until the quartz alone proved inadequate. The secondary feedthrough I mentioned before used to be the main one. It's a 0.75" OD alumina tube with nested 0.5" and 0.25" tubes, Swagelok cap, and conflat compression fitting. It proved incapable of exceeding 30-35kV, suffering lengthwise arcs between the tubes (Paschen strikes again) and what appeared to be intense static buildup and discharge between the layers. It will now serve as a lower voltage feedthrough for a "control grid" once I get some more things sorted out.

Secondary feedthrough extending over 8" into the chamber so the outside portion is rather short.

I have a 50kV x-ray supply built around a transformer salvaged from an x-ray head for that purpose. One of the secondary windings ended up arcing to itself, requiring a complete rebuild of the transformer. I separated the core and removed what seemed like miles of magnet wire (oh boy, was that painful to do), reinstalling the working secondary after insulating it and the primary with some brown paper. It, along with a half-wave rectifier, voltage divider, and current-sense resistor, are immersed in a couple gallons of mineral oil in a plastic bucket. I haven't pushed it particularly hard yet--just to 25kV--since it won't have to run at full output for the control grid. No point in pushing my luck just yet.

Grid
The grid is cylindrical and made of tungsten and graphite. It's handmade--as in without using a lathe or mill--and thus not particularly accurate, to say the least. Four 40mil pure tungsten welding electrodes connect via friction-fit 38mil holes in the graphite endcaps which are 0.5" OD. I made another 0.4" OD grid with 20mil tungsten, though I've had greater stability challenges with it and have mainly been using the former. I've gotten them red-hot without issue, so I think they'll continue to work well.


And I hit the attachment limit...

Neutron Detection
Neutron detection is done with a SNM-18 tube coupled to a Ludlum model 12. It's set to ~1750V with the discriminator at ~-2mV per the manual's recommendations for neutron counters. It's surrounded by ~3cm of paraffin. Since I don't have a calibrated detector for comparison, I resorted to the gamma rejection and simulation method. I used OpenMC to model the detector tube and used the 3He reaction rate it spit out to estimate my neutron TIER. My methodology is described in-depth on the Fusor Forums here: https://fusor.net/board/viewtopic.php?f=31&t=12906. I have also done the obligatory moderator removal test to prove noise isn't the source, though the tube and related stuff is shielded pretty well from EMI. Ultimately it's no substitute for a bubble detector or other instrument, but it's far better than trying to estimate it based on the manufacturer's claimed sensitivity and hand-waving assumptions about moderator geometry/effects. There's still a manufacturer for these tubes, it turns out: https://consensus-group.ru/radiation-co ... -32300-40l. As built, the moderator and tube configuration are about 13.9% efficient and have a calibration factor of 354,000 counts/mrem for the relevent neutron energy.

Shielding
It's not necessary for the neutrons yet, but the front of the fusor is covered in 1/16" lead to stop 'them x-rays. The strips around the aforementioned viton joints aren't tight enough and the lead viewport cup leaks through the solder joints. I have some 1/32" lead on order to solve some of these issues. I once measured the x-ray dose at the unshielded viewport to be 2 rem/hr at some 35kV. Yikes. And the ion chamber has a flat response curve, so it wasn't some GM counter overreacting. Behind the large sheet I can get the pancake counter to hit 500 kCPM. In front of the lead is nothing but background (~60 CPM)

Gas Control
The 20sccm MFC I recently installed is a huge improvement over the needle valve I started with. It's calibrated for hydrogen, not that that's too important, and affords really nice, precise control of the pressure from the same box I control the Spellman with. I use a +-15 V supply powered from the main 24V fusor bus and a 5V regualator/potentiometer for control. I of course throttle the pumps to 66% speed and just barely crack the valves to save deuterium. The stuff in the regulator stem has been more than enough for hours of fusion, and it seems like it'll last several more.



As can be seen in the pictures, everything is mounted on a 3030 aluminum extrusion frame and some shelving brackets.

Fusion Runs
I first achieved fusion early this past summer, though statistically speaking I made a few neutrons in December 2016 when I first attempted it. The Spellman supply was posing challenges, as was the neutron detector, and my senior year of high school, a cross-country move, and starting college all but minimized the time I could spend working on the fusor.

The best runs so far haven't exceeded 70,000 n/s, primarily due to the aforementioned instabilities. That was achieved at 32kV, 5mA, and with 59mtorr of deuterium (corrected since the gauge reads ~2x high). If an exponential fit to my n/s/mA data is a reasonable indicator of performance at higher voltages, I should be able to achieve 1e6 n/s at ~45kV and 8mA. It's not fantastic, but perhaps improvements in the grid symmetry as well as eliminating those hydrocarbons will yield a significant performance increase, as will conditioning through wall-loading and such. The best Q so far is 1.14e-9 at 29.9kV, 2mA, 63.9mtorr, and 60,000 n/s. That run didn't last more than a few seconds, but I'm reasonably confident in the results with the greatest uncertainty in the current. The lowest voltage at which I've quantitatively measured neutrons is 18kV, yielding some 500 n/s isotropic. Counts have noticeably increased in the 12-14kV range, however.

Some eye candy:


Grid at ~30kV producing three well-focus beams and one that's not so nice


Lower voltage star captured with a better camera. Note the borosilicate shield fluorescing blue from the electron beam. I now use a magnet to crash the beam into the wall (x-rays, anyone?). And yes, I have filed away those rough edges on the grid.


Fusor covered in lead and with neutron detector in foreground.


RGA spectrum showing D2+ and D3+

Plans
First and foremost is upping the voltage to get some respectable neutron rates. Once I get some free time no thanks to college physics, some conditioning runs and sanding of the ceramic shield should help reduce instabilities. Most of the fusion runs so far have been pre-Spellman modification where I was constantly juggling the voltage and pressure to keep from going over-current. That will now pose no challenge.

I really need to get around to a local machine shop and make better grids...

I am also working on remote operation, meaning I can spend less time worrying about the neutrons and x-rays. Until I get a new multicore cable the distance is limited to some 20ft, but that's better than sitting directly in front of the thing.

I just recently got an oscilloscope--about time. I'm not willing to spend the full $500 on the cheapest 4-channel scope on Amazon, so a hamfest bailed me out there with a $5 analog 15 MHz, 2 channel scope. It wasn't guaranteed working but ended up fine--and hey, for $5 you can't really go wrong. At least I can finally see some waveforms and the noise the fusor throws out.

Quality data collection is long overdue. I have a bunch of arduinos and will probably end up controlling them through MATLAB. I just need to get around to making an EMI shield box and doing all the fun wiring and voltage dividing it entails.

I can now start work on a HF power supply based on a massive H-bridge unit I purchased from eBay several years ago. It's rated at 400V and 52A, containing 16 MTH13N50 MOSFETS in a four-parallel configuration, beefy protection diodes, and integrated gate-drive transformers. The intent is to investigate Doug's massive success at a mere 10kV by driving the grid at HF with a coupled secondary grid (still learning the details and doing the requisite safety things).

Potentially before the HF stuff is done I'll parallel Joe Gayo's work on 1D grids as documented over on the Fusor Forums. He's achieved over 4e7 n/s at 90kV and has the amateur Q record as recognized by that forum. Using a linear cathode is interesting because the fusion products are preferentially emitted on-axis.


Hopefully that block of interrupted text was of some interest. I'd welcome any suggestions

-Liam David

[johnf]

Liam
Well done
first keep up with your studies
second keep doing what you are doing
We are all here ready to help when you need it!!!!

[Jerry]

You can pick up used Tektronix TDS series scopes for about $100. Like the Tek TDS 320 and 340.

[Doug Coulter]

Thanks for the great report, Liam. I haven't had time yet to go over it all the way, but I hopped on here to make it "sticky" so it won't get lost. I'll probably have things to say once I read it rather than just skim. We may want to move this somewhere else in a little while (when you can continue to post on it in the new location, which should be after one more post from you - see if newtopic and reply buttons then appear on other places and let me know if not).

Personally, a big long report isn't a bad thing - it sure beats chasing over years worth of scattered small ones for conveying information.

Two things I can say right away - I *almost* broke a window as you did, and did ruin a sacrificial piece of 3/8" thick glass (it didn't break, but did turn brown from bombardment and X rays) I'd put in the arm the window was in, to avoid wrecking the good one (which was a rather expensive glass window/door from Lesker). This problem was cured by putting in a grounded stainless steel screen, inside of the main window, spaced out a little, to catch any stray beams. Sometimes it gets red hot, and when you introduce other fields, the hot spot moves around. This not only tends to keep the window cool but it seems to stop most all sputtering type deposits on it. The screen sometimes makes it hard to get good pictures, because some cameras will insist on trying to macro-focus on the screen instead of the grid further away....

One of our earlier grids was kind of twisted, sort of like one of yours. It was precise, in that it was uniformly messed up that way. As luck would have it...it was one of the most stable grids we've ever run here - which has both up and downsides. All the beams were sharp looking, though. Cause of the twist was some tilt in the mount on the drilling table where I did the ends...sigh.

Improvements in accuracy seem to be worth it for beam collisions. That paper we were discussing said those were essentially negligible, but then I'm doing better at 50kv than they were in SS at 70, so perhaps those improvements I made actually improved something. Anyone here who machines things will tell you - making tools and jigs is way worth it for results.

Edit:
We should probably talk about data aq, which was my specialty when I got paid for it. I am right now spinning up a new setup for the new system meant to test the AC/ion trap stuff, which will use some dirt cheap arduino clones (probably two) with some real generic code I will post, mastered by a raspberry pi to do the "smart but doesn't have to be inside a microsecond" stuff - like putting stuff in a database, plotting things, telling the arduinos to set settings, asking them for readings and so on. This will be pretty similar to the one I've already got going on the big fusor, with some lessons learned and a little slimming down.

Each piece of the system does what it does well. Arduinos are really good for "right now simple" kinds of things, like taking periodic measurements on the dot, time-wise, and shipping them off timestamped, to something fancier that can run databases, plotting software, user interfaces and so on - but can't do diddly on time because that's the nature of ALL preemptive operating systems - they preempt, and "go off on their own demented errands" that can take some milliseconds. But their own hardware will buffer some things like serial input from the arduinos, then get it and do good things with it. For this kind of thing, while I have a couple, I don't think a pi 4b is needed at all, unless you want one. I'm using a 3 in the big rig and it's not getting too busy. But that ability to run linux, high level languages, and various nifty already existing code is something an arduino can't hope for, and most PC's are going to burp trying because of other cruft on the PC. In my case I wind up running the pi headless and just running the one data acq/control app, talking to it via VNC from a PC that does the real heavy - but even less time-critical, lifting - like plotting form the database once a second or so, looking at video streams from remote cameras and other things like that.

[Liam David]

Thanks for the feedback everyone.

I spent a couple hours today making a new grid. To step up the accuracy, I 3D printed a jig that accepts the 1/2" graphite stock and lays out the mounting hole as well as those for the tungsten rods. The center is tapped 8-32 and the holes for the rods are drilled with a 0.9mm bit to accept 0.04" ~ 1mm tungsten. While the filing and drilling is still done by hand, the results are far superior to what I've made before; the parallel rods speak for themselves. The most significant asymmetry is that the cap is slightly tilted relative to the base, though a little adjustment on the tungsten lengths should fix it. I'm mostly concerned about ruining the friction fit. Since the grid is hanging upside down, it can only handle so many disassembles before falling apart. Tests to come within the week.

12.8mm OD, 8.36mm rod spacing, 21.61mm OAL.

Jig and grid for comparison. Jig has 8 holes should I want more tungsten rods in the future.

Data aq is probably of more importance than all the other thing's I'm working on. Currently I'm limited by how fast I can perceive and write down panel meter readings for voltage, current, pressure, neutrons, secondary grid voltage, and the list goes on. That's not to mention the lack of simultaneity, especially since the voltage and neutrons jump around a lot. I'm concerned about reporting bad data and certainly don't want to claim a breakthrough where none exists. To address this, at a recent hamfest I acquired some half-dozen Arduinos for $20. I know they're not the best and that I'll have to do some voltage dividing to bring all those 10V signals down, but for a first pass, they're better than nothing. I'll probably end up running the raw signals down a 100ft shielded cable and set up all the daq stuff by my operator station to reduce noise and potential crashes. Matlab is my software of choice and it'll allow me to generate all sorts of fancy plots and statistics on the fly. The tiny old laptop I currently use--which has been to Iraq and back, twice--won't suffice, so something with a little bit more CPU power is on the list too.

While I wait for some scope probes to arrive, a couple days ago I did a plasma cleaning run and used a wire coil to pickup whatever the fusor threw out. Whenever plasma was established there was a very strong signal at ~140Hz. It was about 2/3 negative and at higher voltages (~14kv versus 7kV) the sharp peaks on the positive 1/3 noticeably flattened out. The frequency did not change with voltage and the signal disappeared when the plasma extinguished but the Spellman was still on, so it must have something to do with the plasma. Has anyone noticed a similar signal?

Peaks at about 7kV.

Flattened peaks at 14kV along with pressure, voltage, and current readouts from left to right.

I also picked up an old CDV dosimeter at that hamfest. It's 200 rem full scale, so perhaps I can get it to move behind the lead when I've got 40kV on the grid, not that it's any joking matter. I've measured upwards of 3 rem/hr behind the lead with an ion chamber, and that was over a foot from the action. Scary stuff.

More updates to follow...

Edit: I can now post in the other sections, but I don't think I have permission to move this post.

[Doug Coulter]

"A wire coil" isn't clear enough to know what it is you're picking up here - is it a current transformer, an antenna, or what? 140 is suspiciously close to 120 hz - which would indicate some weak filter capacitor in the power supply (and a scope that needs calibration!) - and would only show up under load, just a guess. It is possible to get a fusor to oscillate in that range, like a neon bulb relaxation oscillator, but it's really unstable and not that easy to make it consistent - the frequency wanders all over the place...

This is a really good reason to do as Richard and I do - and put at least one counter output on some kind of audio to listen to. Your brain will instantly pick up on things like that, and have a really good guess at the cause. A safety thing as well...

I kept getting improvements in Q all the way up to the limit of my precision in building grids, though I did get to a point where the improvements weren't huge. Yep, you can only assemble them a time or two before the friction fit is gone, and sometimes the electrostatic forces are surprisingly strong. The 20 mil rods are better here, and using the pure tungsten ones improves Q by having a higher work function so they don't emit so many electrons when hit by ions - this by actual tests. You can get them plenty hot with no problems, your mount or feedthrough will melt first. But here we don't do that so much in the search for Q vs max output - they don't happen at the same set of conditions. I'm using 8 rods in the big setup (3/4 OD), and am testing 6 rods in the smaller grid that's 1/2" OD.

It's going to be interesting to compare the small tee to the 6" ID setup I have. We did do some tests with a 1" grid (early days) vs effective OD by making some 3,4 and 5" copper sleeves to slide in to reduce the size of the anode. Uniformly, the smaller the outside got, the worse it worked (and took more pressure to get it to stay lit - which was more scattering losses). There is a really strong hint here that my ion grid - out in the main tank which is much larger (14" ID x 18" tall) - has better Q than my main grid, it really lights up the neutron detectors, even the hornyak which is touching the tank wall at the main grid, but a foot away from the ion grid. Not sure if it's ratio thing or just absolute sizes and haven't fully quantified it yet but it's a big enough effect to be obvious just listening to the clicks and noting that the ion grid supply is only 2.5 ma and usually running in current limit at not more than 20-30kv at most. So far here, the smaller the grid, the better it has worked, though the other variable is that I've gone to smaller rods in those, and used fewer to keep the transparency about the same as I went down in size. About 1.2x inside length to outside diameter ratio seems best so far.

We never did, BTW, try such a large grid OD to tank ID ratio. Could be it gets better at that extreme too, though I doubt that.

You should make a "neutron oven" and measure silver activation....it's a great way to make sure your other detectors aren't messed up in some way. Since I'm already plotting a geiger counter in real time here, just moving the silver to it after a run tells all - and the time it takes to move it as long as it's reasonably short, lets you trivially extrapolate back to when the fusor was shut off - using a log plot as I do, the silver decay is a straight line pointing at the "moment" you really wanted the data for. The point is that this is something one can compare across labs with reasonable confidence.
I have no faith in gas tubes - and little in the hornyak, you can easily be off by several to one with a minor change in threshold or HV, one you wouldn't think mattered. Even the load of an HV probe on the detector power supply might toss it totally off - and there's always noise getting counted - see audio, above. It's just a good thing to have.

I'd never, ever put raw signals down a long cable - you'll have noise, a not very decent lowpass due to the large capacity between center wire and shield, ground loop issues and so on - if you add a preamp to hold those problems down, now you have preamp accuracy and noise issues.... When decently set up, arduinos are nearly immune from crashing - I don't think it's ever happened here, actually, with them in a metal box, and with inputs conditioned. I use 10k series R on the a/d with a .1uf to ground at the arduino - since the a/d has extremely high DC impedance that 10k doesn't affect things much (the a/d is calibrated with that in place anyway) - but it now takes quite a wallop to put enough power into an arduino input to fry one against the built in protection diodes. In most cases, the 10k 1/8th watt resistor would fry first anyway (eg, cheap and obvious so easy to fix).
In my setup, I program the a/d to take as many samples as will fit in my reporting interval and sum them up - one could divide by the number of samples later, but actually there's no need - just calibrate it with the larger scale, and it seems one nets a couple of extra bits of resolution when sampling say 8 times per report (which would theoretically give you 3 more bits in a perfect world, but this isn't).
This is due to inevitable little bits of noise toggling the LSB of the a/d on the various readings and it averaging out...the technique is called dithering when done on purpose. Turns out the usual noise you can't get completely rid of anyway is fine for dithering...

I use perhaps 100k pullup on the counter input and a 10k series R there - same idea, let a cheap resistor burn up vs the chip. I haven't had any issues or fried any arduinos yet.

I find that the geometry is the crash indicator - I can't crash arduinos, it's possible to crash a pi but it's difficult and usually means some huge EMI got in there - and intel machines crash immediately with reasonable fusor output - even with EMI really well taken care of (all HV wiring inside home made huge coax using pvc pipes, copper screen for the outside). The smaller the transistors and the more dependent on ram, the easier to crash they are. AFAIK, NASA agrees.

Simultaneity is a real issue, even at the low speed I do the a/d and counting at (100ms per report). This stuff is all a bit off in clock frequency, some things worse than others, and if there is more than one they drift relative to one another even if they're all started exactly in time via an interrupt I generate to sync them on what call a clapboard line (same idea as the ones used on hollywood takes).

For the next go, I plan to generate a trigger signal external to all the slave machines that doesn't drift. I'm getting a pretty good approximation right now on a rasberry pi using a high priority virtual timer and direct drive of an IO pin that can trigger arduinos to take a reading and report it. This signal is synced to the pi's clock, which is turn made accurate via NTP over the long term. Keeping the videos in sync with this is still a bit of a trick as a dropped frame is hard to deal with (or even detect). I do have timestamps overlaid on the display so you can see that error manually, however.
The pi does have some facility to generate signals in its hardware that aren't affected by opsys latencies - but they're all much too fast for this, and I didn't want to add a big divider chain to get down to reasonable speeds.

I use gnuplot (which I believe is what matlab uses under the hood?) and rolled my own database and plotting code - on process puts things in the database, another polls that and plots the data once a second or so - which gives you a built in error monitor - the data has to be getting in to that database to show up on a plot, so it's super obvious when something is wrong...and then that same plotting software also works on any previous run as well...and plenty of tools exist for managing and backing up databases. This is where something like a raspberry pi shines brightly, as it can be set to ignore the various other things that sometimes get a laptop etc too busy to really keep up.

FWIW, all my code is copyleft, but to set up a system takes some administration and adaptation to your setup details. For example, one has to set up MySQL or mariadb, I added a web server (nginx) and phpMyAdmin to fiddle in the database, things like that...it's all a job of it's own. It's pretty much all for linux - you can make it go on windows, but it's another layer of junk to make that happen.

[Liam David]

You've given me much to think about. I will change the data aq stuff around accordingly and leave the hard-wired connections for things like HV enable. The laptop I am (was) using crashed all the time above 20kV, so a more robust system is certainly needed.

I use the 6-way tee more because of budgetary constraints than anything else. I missed a very nice chamber and misc. vacuum components lot on Ebay not that long ago and am still on the hunt for something with more volume and ports--something more the size of yours. Six ports just isn't enough for the multiple instrumentation and power feedthroughs I'd like to add, not to mention that 1.5" ID doesn't give much space for larger things...and Paschen doesn't help me with the smaller chamber either. The chamber/grid ratio is tied to the chamber size--I can only make them so small. The next grid I make will have 0.02" tungsten and perhaps I'll give a stab at something more around 1/4" OD, but with school it's the time that's lacking.

A silver oven is certainly on the list, but right now I'm not making enough neutrons for silver to register anything above background. It should be a pretty quick build with all the paraffin I have laying around.

The noise pickup on the scope was more for grins than anything at this point. I think it's calibrated, just based on the built-in square wave, but my question was more akin to "hey, what's that?" than "breakthrough!" Now I can see noise that was invisible before--things that the vast majority of fusioneers never see nor care about nowadays. The more I learn, the more I realize instrumentation can be a major limiting factor and not just the ability to make a good vacuum or HV.

Currently I'm most interested in your claim of having made over 1e11 n/s at under 100W input with HF on the main grid (viewtopic.php?f=42&t=938). I realize such a breakthrough could keep you pretty quiet as to avoid idea theft, as you mentioned in that post, but now that you have remote operation going, have you replicated these results? I've been following along pretty closely and don't think I've missed anything major, but please correct me if I'm wrong. It's this kind of progress that brought me here, and I'd like to help out.

[Doug Coulter]

The 2.75 stuff makes possible some things that haven't really been measured - so to make a virtue out of necessity, those things should be measured! There's a lot not yet known about scaling issues and it's always good to find things out! Even if those things are "this way doesn't work as well". Knowing always beats guessing. If you need help making tiny grids, let me know...

Rather than keep fooling with replicating what was a basically uncontrolled and poorly understood situation - nifty as it was, yet still far from gain - at considerable risk to my health and the lab, I decided to attempt to understand how it happened - those numbers were a gross guess, btw, perhaps +/- factor of 10 or more (still, pretty ginormous). I did get a couple more of those, and even one of the youtube videos actually recorded one (Unstable) - I didn't realize it at the time. This may have caused me to misinterpret the dose I got from that big one - I'd been catching that stuff for a couple of runs without realizing it. After all, once you know how something works, your chances of controlling it to make it work better improve quite a lot. Clearly, having this thing oscillate around both sides of the dynamic equilibrium desired by most workers is the key here. In fact, I've found that almost anything that pushes it off that improves things - some. Of course, we want more than "some".

The thing is, there was no reasonable way to allow remote twiddling of things to run that close to the edge of stability, and no way to safely be there to do it by hand. I also hadn't considered the implications of that much improvement on everything - when you're making microwatts of fusion, it can be fine in your spare room. Multiply it by a thousand or more - still far from gain - and oops, you might not want to activate everything else you own. All of a sudden, the stakes (required investment) are raised a good deal. If I wanted to continue trying the "passive but with tuned circuits inline" approach, not only was the radiation a new issue, but also being able to change those parts was a problem. To keep other electronics alive, I'd been shielding both sides - the HV was all inside shielded pipes, and it's a real bear to get into those to try things, which needed custom engineering each time to not arc inside the pipe. It just wasn't a tenable way forward. If I could only try a couple things a week, given the huge search space - I'd still be at that years later and likely not have learned much - this search space is non-monotonic, there are peaks and dips in all directions - it's analogous to a neural net error plot - a simple gradient descent doesn't work to find the endpoint. Taking off the shield required other adaptations for data aq and survival of other electronics...basically what I'd been doing was cutting the noise down at the source (a parasitic very high quality 50pf capacity discharging 50kv in a arc in a couple nanoseconds can be e9 watts!) AND at the electronics and data aq - 40db is easy to get, but it needed more than that, so both ends...

Clearly there is something serious to be learned, but also more than one way to go about it. In many things, I've found that sometimes baby steps - once you know the prize is there - are actually the fastest way to get there, so that's what I'm up to now. And I've learned a lot more than I would have futzing around with that particular lashup this way.

We have basically found that almost anything that pushes this off (first order dynamic) equilibrium improves the output and Q quite a lot. That oscillation, when all the parameters were right - gave outstanding improvements. Clearly my (I guess I originated it?) theory that recirculation must be driven was correct, and that the apparent stability most fusor workers seem to crave is the opposite of what is really wanted. The trouble is, the search space is crazy huge for those right parameters, and the few times I got it to happen - everything in the room crashed and I had to stop it for my own good - so I wasn't reducing the search space, and a methodical approach vs just trying things in the dark seemed to be called for. It seems necessary to creep up on this, gain more understanding along the way, and really nail it down. There's quite a lot more I know (or think I do) that I haven't manage to clearly report yet - this post would be super long if I tried to do it here and now.

So, what do I think I know in the fewest words possible?
Well, I stared to try and write up a summary, but failed in the time I have right now, and this isn't the place on the board for it. I guess, thanks for the reminder that I need to do that!
Super quickly - I did find that something like the basic Mathieu equations apply, but vastly smeared out and pretty obviously, need some added terms because the original math (used for ion traps and mass specs) - doesn't take any account of charged particles affecting one another - space charge and the like - and for this to work at densities good enough to make significant collisions happen, we need quite a lot of interaction - it's the entire point of the exercise.

What that means in terms of replication is that for one thing, volts and frequency and amplitude of all those (now including terms from the locations and velocities of the charges themselves, not just our applied fields) must be related for the thing to work well, or at all (current fusor efficiency is close enough to zero to be called that, IMO). Even the assumption that holds valid in mass spec math that any waveform will do - sine, square, no difference - is almost surely not valid. This makes the search space so enormous (we are adding dimensions) that just diddling is not going to find it without luck such as no human has ever had - I'm just going to have to creep up on it and get real understanding so I can design a working system feed-forward, rather than accepting a flaky lucky accident here and there. At that point, it's likely possible to design a passive version that self-oscillates - but reliably and always with the right parameters.

The significance of the lucky accidents (I'm not the only one...) is that we have an "existence proof" - we know there is something there worth expending resources to find. At least I'm very sure of it.

Meanwhile, if you just want to know all that I know, as fast as possible...you should simply skype me on a video call where the communication bandwidth is far higher than typing will ever be...I'm dcfusor on skype (and the invite is for anyone here). I'm on the east coast so the times should be reasonable for that. If I'm around (most of the time) I'll answer - the time of year for outdoor work is nearly over.
Yes, I'm a little gun shy after my ideas have been patented by others (who haven't made their first neutron) and I don't need the ridicule I sometimes get either, but my intent was and is to share this as widely as I can.

Re data aq -
I did find I had to get my PC stuff well away, but the arduino/pi stuff was fine nearby - and bits go over wires a lot less lossily than analog. Since I had some real crazy EMP type events, I got a fiber optic converter to isolate the fusor stuff from the rest of my network so accidents wouldn't fry my whole world. That works very well here. Other than the front-end stuff, it's all done over ethernet. While it's not a requirement, here I have a separate machine to be the SQL server (an odroid HC2 running linux) and to be the place my video cameras write data to. The fusor raspberry pi used to do that, but being all sealed up in the box, and at some small risk plus hassle to get into it and replace things like disk drives - I offloaded all I could from it, and it just serves as master control and data collector which takes commands via remote desktop software (vnc), and moves data from arduinos to that database. As well as commands to the arduinos, which control gas flow, power supplies and anything else. At that point, anything else on my network can plot that data in real time - and does, meanwhile keeping permanent records. It's kind of a lot of system administration to set up, but it's pretty slick once it's going. All but the PC are really cheap commodity hardware. And yes, safety critical stuff is hardwired. See the shakedown video for an overview. There is also a camera that just looks at the whole lab, sending a stream back via wifi. That's a backstop in case something I'm not explicitly measuring (because I don't and can't do everything, or even think of it) - goes weird or catches on fire...

BTW, you will get silver out of the noise pretty easily at 100k n/s, especially since you can get closer in your fusor than mine (intercept a bigger % of them). FWIW, HDPE is better than wax, but for where things get hot, water is better than either. I made a little water tank out of soldered copper flashing to sit on the fusor sidearm, so if it gets hot, it can boil and vent out a standpipe. I'd burned up plenty of plastic before that, and it's not pleasant. I'm using a median 1.5" thick on the fusor side of the oven, and a few inches HDPE above that as a reflector. Works great.
Especially handy if you think you might get lucky - a piece of indium foil in the oven wouldn't be misplaced. I wish I'd had that...but no, I trusted the fancy instruments and they let me down by not being able to handle the flux, and then things crashing anyway so as to lose data. A backstop is not unwise.

[Liam David]

School's kept me busy but I've been making steady progress on the DAQ and control system.

I've opted for a fiber optic/ethernet system for data with a multicore, shielded cable for critical functions like HV enable, both of which pass through 30 meters of expandable sleeving to form an umbilical. My laptop connects via crossover cable to an ethernet/fiber converter and the hardwired connections will be controlled with toggle switches. On the fusor side in a metal box is another fiber converter, a network switch, and an arduino with ethernet shield. It communicates with two slave arduinos (Wire I2C) that perform measurements when the laptop program requests data. A relay board allows for switching of less-critical functions. I have yet to build the fancy enclosures but the electronics are fully functional--minus a few bugs.

Umbilical with LC/LC fiber and 15-core shielded cable

Fiber converter

Neutron detector data logging and control system test

I've set the sample rate to 20/s, but it's adjustable in the control GUI, as are a number of other parameters like transient filtering, the DAQ IP, etc. Data is saved to the disk every 5s. The neutron counts are tallied via an interrupt on the slave arduino and requested in 5s intervals. The pulses are generated by slight modification of the Ludlum model 12 that drives the 3He tube. A pin on a pulse extender IC gets pulled high to 10V for ~15ms per count, and I tapped into that via a 2:1 voltage divider to bring it to the required 5V. The connection is made though a BNC bulkhead on the front of the ratemeter so that it doesn't protrude awkwardly from the side. While it doesn't support super high count rates, it's a good first approach and I can always move the tube farther away.

The GUI's made in Matlab's app designer and features most of the things you'd expect. The "big four," or voltage, current, pressure, and neutrons, have large gauges while other values are displayed in boxes below. Major events are displayed in a console and saved as well. I've incorporated things such as neutron dose estimation, gas usage, and pressure correction for deuterium to make data analysis easier. Plotting will be done in a separate program to prevent data collection from lagging.

I'll view the fusor with two old smartphones used as streaming cameras. A free app hosts the live-stream over wifi which I pull up through Matlab and/or any old browser. One views the grid and the other will watch over the whole setup to, as you said, make sure nothing catches fire.

The DAQ GUI, still very much a work in progress

[Doug Coulter]

Good moves there. The fiber optic link stopped the frying of whatever machine was local to the fusor (a pi and a few unos in my case) as well as whatever was on the other end - some PC, and the stuff in between (switches). The hardwire cable is also very good for a safety. I've set mine up so that two wires carry 9v from my remote operator position to the fusor, and those are used to energize a power relay that the main fusor power is connected to. That way, all I have to do is turn off that control/enable power from my, safe, end, and the fusor is safe-ed immediately and for certain.

Of course you will use whatever software etc works for you. I'd strongly urge you to write the data down somehow as it comes in. This is a basic rule of most experimental science. If you only save some interpretation of the data, but not the raw stuff, your'e foreclosing any further interpretation should out understanding improve (which of course, the the point!).

For example, I use a database (mysql) which is a little overkill but it's efficient code and I didn't have to write that database itself, just talk to it from my code.
The way things work is all the raw data goes into that database. Then, various things can plot it or display it - this way gives me an auto error detection mechanism, as I can't see anything happening if the first stage isn't working. In my case I wrote code that lets me get plots on some run number (defaults to the one in progress, but being able to see older ones is also good) and which can update around once per second, so the plots are near real time.

But then I also wrote a thing that can map entries in the database to axes of a plot using some interpreted code you can type into the gui, and make 3-d (really 4d as I have color as an independent variable) scatter plots you can rotate - and if your hardware is up to it, do that in real time too. In my case, the most useful mapping has volts and amps on x and y, and either raw neutrons on the z axis, or better - the neutron count divided by the volts*amps - Q, in other words. That's a most illuminating plot...

You might be able to do that kind of thing with the stuff you're using, I'm not knowledgeable with it. Definitely something to consider. One of the biggest advantages we have over say, Farnsworth, is that we have better data acquisition and analysis than he ever could have. The hope of course, is that if one of those high output "anomalies" should happen again, we get it recorded for analysis, so we can do it on command later, and don't have to wait for luck anymore.
So this is what I use - it's not perfect, but...it ain't bad either. I made various DB tables that let me put in comments so most database tools let me find things again.
Like this:

fishing for data

And then I can do useful later analysis, like this:

4 d rotatable plot (Q axis arbitrary units)

What this plot shows (and so do most of them) is that our Q goes up at lower pressure and lower current. The working theory is that less scattering helps us get the benefit of better focus.
That paper we looked at assumed there was no beam on beam - well, that's not true here, as that's what I've worked on making better and had some success.
We do also show in time-series plots that our neutron output and Q go down as the tank walls heat up. There is certainly some deuterium embedded in those walls, at least till heat drives it out...