After an awfully long time away, here is a fusion report.
I started building a fusor some time in 2006 and didn't end up getting around to trying for fusion before running off to university (and subsequently taking a job in a research lab). I left behind a machine that was more-or-less fusion-ready, but with an overengineered control and data-acquisition system that was only half done (picture stacks of home-etched PCB's partially assembled). In March 2015 I managed to set aside a few days to come back, dust everything off, patch in some rough-and-ready instrumentation and do some deuterium runs with the neutron detector going. I got some suggestive pulses out of the neutron detector, but in my rush I didn't have the presence of mind to do any running with the moderator removed from the neutron detector. I never got around to really going over the data until just recently when I had already started pulling parts off to sell and even sold and delivered my deuterium bottle to its new home.
To make a long story short, I ended up borrowing back the deuterium, bolting the vacuum system back together, and taking another shot at it.
There were some setbacks. Within a day of closing the system back up, I had managed to blow my HV supply. A couple of the zip ties holding things together in the HV tank had failed after around a decade sitting under oil, causing things to shift around, and eventually one of the HV AC leads arced over to the side of the filter capacitor, carbonized my rubber supplementary insulation and formed a persistent HV short. Thinking I had blown a secondary, I was fully prepared to pull the transformer and saw off the bad one, but fortunately the actual problem was fairly obvious once the lid was off. I got rid of the cap, tossed the burnt rubber pieces and cleaned up the sooty stuff with a pipette, and the HV supply was back in operation.
I also discovered a couple vacuum leaks which prompted an early morning run to the grocery store for some emergency nail polish.
Finally, as I started with the trial runs, I found that the system would start becoming increasingly unstable within a couple minutes of operation at only 100W or so. Neutron counts would drop off, crackling and sparking would crop up and keep getting worse, and eventually the plasma would drop out. Grinding some stray epoxy off the grid stalk insulator, sealing up the aforementioned leaks and increasing gas flow seemed to make a substantial improvement.
My main chamber is a 5-way Conflat junction with 8"CFF main flanges, with a blind reducer on top and a DIY viewport assembly on the bottom. The HV feedthrough was made by soldering a military surplus BeO transformer feedthrough onto a 2.75"CFF flange. My center grid is made up of 3 ~1" dia. hoops of 0.020" tungsten wire.
The pumping system consists of a monster Welch 3012 belt-drive (!) turbomolecular pump backed by a Welch 1402 rotary vane pump. Total pump overkill. The TMP motor was run off a variable-frequency drive, generally run at a fraction of its nameplate speed. The Chamber is connected to the pump via a long SS bellows hose with a right-angle valve and a small butterfly valve in between.
Pressure measurement was done with an MKS 325 pirani gauge. Note that all pressure values here are with respect to the N2 curve supplied by MKS, not corrected in any way for deuterium.
Edit: Pirani, not micropirani.
Edit: The Stanford Research Systems app note "Gas Correction Curves for PG105 Readings" gives an N2->D2 correction factor of 0.79 for thermal conductivity gauges in the molecular flow regime, suggesting an operation pressure of about 18 microns for these runs (discounting water vapor, hot epoxy byproducts, etc.).
Deuterium injection was controlled by a pulsed solenoid valve. Since last time I added an actual PID loop to the pressure control program and this made for vastly more stable operation. I used a manual butterfly valve on the pump port of the chamber to control the relative rate of gas flow while the PID routine maintained the pressure. Varying the TMP speed was also found to be effective at controlling the pumping rate, but obviously the valve offers vastly quicker response.
The neutron detector consisted of an ancient N. Wood BF3 tube, 5/8" OD, 3.5" active length, operated at 1600V, inside a 139mm OD HDPE moderator, oriented vertically, placed about 185mm from the central axis of the chamber. The neutron detector was mounted on a fixture to allow it to be repeatably positioned w.r.t. the chamber, so that the moderator could be removed and reinstalled between runs. A cardboard jig and markings on the chamber helped to ensure accurate positioning (see photo below).
The high voltage power supply was built from an old dental X-ray head. I kept the original housing as a tank, ripped out everything but the transformer and used the gained space to build in a rectifier. The transformer is ballasted with a choke made from an old MOT -- secondary removed, welds ground out and core gapped with plastic cut from a credit card. Control is via a run-of-the-mill variac. Input power was monitored with a Kill-A-Watt.
Pulses from the neutron detector amplifier were fed straight to an oscilloscope which was used for counting. My software downloaded the waveform after each trigger, allowing after-the-fact filtering and analysis. A second oscilloscope was used to monitor the HV probe and pressure gauge. The 8-bit vertical resolution on a digital scope isn't ideal for precision voltage measurement, although this is helped by heavily oversampling. The analog frontend on an oscilloscope is incredibly handy for makeshift data acquisition setups even if the sampling resolution isn't ideal.
Edit: The HV measurement was done by taking waveforms off the HV divider, taking an RMS average in software, and multiplying by a conversion/calibration factor. The setup was calibrated against a multimeter with a Fluke HV probe. The data logging software was set up to also occasionally save the raw waveforms themselves for later inspection, but this failed due to an implementation error.
Here is the whole works. Please excuse the background -- this is my bedroom from high school, and hasn't changed a whole lot since I left.
Data acquisition and gas control were handled by a whole bunch of Python code.
After some experimentation I settled on a pressure setpoint of 23 microns (*indicated* pressure -- see above note about pressure readings). I ran with the TMP at about 20% speed and the butterfly valve open to about 45 degrees. I had tried a lower-flow configuration with the the butterfly valve mostly shut, and although it was easier to get rock-solid pressure, performance seemed to suffer due to outgassing, etc. when things got hot.
I did a series of 8 5-minute "burns" (HV on, discharge going) separated by 5-minute "rests" (HV off). The third, fourth, seventh and eighth runs were done with the moderator removed from the neutron detector. The first four burns were done at a nominal power level of 80W, and the last 4 at 160W. The variac was set to produce the specified power draw at the beginning of each set of runs, and the same setting used for subsequent runs. The power draw generally started out around the nominal value upon ignition, and gradually dropped by about 1/4 over the course of a burn.
Pulse events were filtered by hand after-the-fact: every single event was plotted out, and obvious false triggers, EMI events, etc. were marked and discarded. This was done "blindly", i.e. without looking at the event timestamp or any other data channels. This was followed by a fixed 0.75V peak height cutoff in software. Obviously this sort of scheme would make no sense at any reasonably high count rates, but it worked out fine for my non-optimized fusor and tiny detector tube, and was the easiest solution to set up on the data-acquisition side of things.
Here's a nice plot:
- Code: Select all
Region 0 ( blue): 38 counts -> 7.6 counts/min
Region 1 (green): 1 counts -> 0.2 counts/min
Region 2 ( blue): 52 counts -> 10.4 counts/min
Region 3 (green): 0 counts -> 0.0 counts/min
Region 4 ( red): 0 counts -> 0.0 counts/min
Region 5 ( gray): 0 counts -> 0.0 counts/min
Region 6 ( red): 0 counts -> 0.0 counts/min
Region 7 ( gray): 0 counts -> 0.0 counts/min
Region 8 ( blue): 95 counts -> 19.0 counts/min
Region 9 (green): 1 counts -> 0.2 counts/min
Region 10 ( blue): 106 counts -> 21.2 counts/min
Region 11 (green): 0 counts -> 0.0 counts/min
Region 12 ( red): 2 counts -> 0.4 counts/min
Region 13 ( gray): 0 counts -> 0.0 counts/min
Region 14 ( red): 0 counts -> 0.0 counts/min
Region 15 ( gray): 0 counts -> 0.0 counts/min
Finally, here is the requisite selfie: