Melted PS Performance

Here I talked about an ugly looking melted PS film. Turns out I was a bit harsh to judge it; melting it didn't change the light output that much. However, a film cast on a disc of the exactly same composition had much higher light output (the light blue triangles).  The green is the before melting; the red is after.

Gamma Count Rates

Been spending a bunch of time in Gamma Town.  It's nice, you don't have to interact with anybody*.  Shown below is comparing results to simulated values from the total interactions. 

Source of all errors

Alright, so not perfect. But hey, I'm gonna take it.  It explains a lot of the problems we have been having; doubling the distance about doubles the counts.   I think at the father distances photons bouncing around off the table and the lab might also be a little bit an issue. It is also unsettling how the simulated and calculated cross each other.  Now the problem is going to make sure that we can have a repeatable geometry; so maybe we want to place it around 6" away, where 1/2" in each direction only changes the count rate by 12%.  At 1/4", moving the sample by 1/4" changes the counts by 30%.

*Groan worthy, but I'm proud of it.  Gamma's are photons, and are characterized by a probability of interaction, not like electrons which are always experiencing E&M fields.

Gamma Spectra Simulations

I've tried to simulate the spectra that our detectors see in MCNPX from a Cs137 source (0.6617 MeV), and Co60 (1.17 and 1.33 MeV).  You can observe the Compton Edge (maximum energy of a Compton scattered electron) in both spectra; notice that the Co60 spectra has two compton edges because of the two photons it emits.  You can see the photopeaks, which is caused by when all of the incident energy is deposited in the detector clearly in the Cs137 spectra.

EJ-200 Gamma Response.  Black is Cs137, blue is Co60

GS20 Gamma Response.  Cs137 is black, Cobalt 60 is blue.

The simulated spectra where then compared. The EJ200 is 1/4" thick 2" diameter, while the GS20 is 1" diameter and about 1/8" thick, that is why the EJ200 has a higher tally than the GS20.  You can notice the extra resolution that using a glass scintillator has a much higher resolution (notice the photo peaks that occur at 1.17 and 1.33 MeV in the black, and the absence of them in the the blue).

Co60 Comparison.  The black is the GS20, the blue is EJ200.

The problem is that the count rate's do not match the observed spectra.  However, the shapes are the same.  Gives hope for tomorrow, anyway.

Observed Spectra from GS20.  Green is the Cobalt 60.

Gonna have to trust me that the shapes are the same, since I don't have a spectra readily available on a log ordinate axis.  The channel numbers are analogous to energy, since the channel number corresponds to a voltage, which in turn corresponds to the amount of light collected by the PMT, which in turn corresponds to the energy deposited by the photon.

Really Dude? (Spun PS LiF Film)

We must be reaching the limits of what we can create for new detectors to measure.  We took a rat's nest of 250 micron spun PS fibers with LiF and some fluor and melted them to create the beauty below.  Amazing enough, it gives some counts.

Before Mounting

After Mounting

It reminds me of frosted Wheaties.  The cool thing, I think, is that once it was melted (adding energy) the fibers relaxed to an agglomeration, not a coating (which we had hoped for).

Beards

http://www.biggerbetterbeards.org/

I didn't know they had so many uses!

MCNPX simulations

The problem with MCNPX (or with any simulation) is that while it may pass one check it doesn't mean that the simulation is correct.  My neutrons aren't working with boron (can't figure out the reaction card for the boron 10 capture reaction, and the total reactions gives results 4x higher than observed, but hey, we don't use boron any way), and for some reason MCNPX is producing a bunch of photon that we don't see (probably because I tried to get real clever with the geometry cards).

I have my last presentation tomorrow.  It's a group project.  I hate group projects.  I hate trying to read other's code, and trying to debug it.  I don't care what we put in the slides, but this is going to be the background of the last slide.

Who could ask a question about neutronics looking at that?

Enough complaining.  In a little over a week I go home! And David's back from Wales!

Killer Turkeys

A movie about Killer Turkeys exist, ThanksKilling. Excited.

Simulated

I have been spending a very long time learning how to use MCNPX to simulate detectors, and an even longer time fiddling with my geometry to ensure what I modeled is actually what is there.  It is an iterative procedure; simulate, look at results, learning something about MCNPX, simulate, change a composition / geometry (mostly adding stuff, Martin's input was very back of the envelope), and continue.  Each simulation takes around 40 hours to run on a single processor, so I run it on the NE computing cluster with anywhere from 30 - 50 cores.  (The cluster only haves 150 cores, and I like to leave space for others to run their jobs.)

Using the (n,triton) reaction card (MT 105) I used an F4 (cell flux) tally modified by an FM card to calculate the interaction rate.  For the GS20 (a standard detector with published efficiency values) I also used an F1 (surface particle) tally to find the number of neutrons (of certian energy ranges) that cross the detector.

Results

For GS20 the simulated interaction rate CPS is  431, the simulated reaction rate from efficiency is 423 cps, and the observed count rate is 428 cps.  Less than a 1% difference.  For the PS based LiF loaded film, I calculated a simulated interaction rates of 108 cps, the measured is 96 cps, a 13% error.  However, this is to be expected since the PS film is opaque, and light generated may be quenched or pushed below the LLD setting, so our measurement may not record all of the light generated.

PSD with ANN

Check it out!  Pulse Shape Discrimination with Artificial Neural Networks!  Pretty baller, I know.
The idea is to take a spectra and figure out what radiation caused it.

So we sanitize the network (a high pass filter to remove some of the noise as well as normalization), do some fitting, extract some properties, and then feed those properties to the ANN.  The ANN is essentially a matrix with different function kernels for elements, trained iteratively in a minimization of the error surface (conjugate -gradient like methods).

But here is the real question.  How can the first network train to figure 3 but not train to figure 4?  A five year old could draw that decision surface!

Figure 3

Figure 4

Below are the results, a classifications of 3%.  Not bad; the ORNL paper with the same data got 7%

.