[Halld-cpp] [EXTERNAL] AIEC CDC info

[Naomi Jarvis]

Hi,

Here is an introduction to the AIEC project, we can talk about it at the
next CPP meeting.

The aim of the AI for Experiment Control project with the CDC is to adjust
the chamber's HV before the start of each run, to maintain uniform gain
from one run to the next, using an AI to calculate the appropriate HV and
predict the calibration constants for the gain correction and time to
distance calibrations.  The target gain would be that for 760mmHg and
2125V.  The inputs to the AI will include environmental data (pressure,
temperature, beam current, flux) and might also include some physics data
(pulse height etc).  By 'each run' I mean each 2h or less period of
data-taking.

We do not envisage using the AI to drive the CDC directly; the plan is for
the AIEC team to use the AI to calculate the desired HV and a human being
would adjust the HV within some constraints, for example, keeping the HV
within +/- 20V of the standard operating HV, which is 2125V.   The HV would
not be altered during the data-taking.

Background information follows below.

This plot from https://logbooks.jlab.org/entry/3759767 shows a nicely
linear relationship between the gain correction factor and gas density,
provided the flux is consistent.  The different coloured bands correspond
to different beam current & different radiators, with some distracting data
from the early runs when the gas mixer was not working properly.

[image: gainfactor_vs_dens_colourcoded.png]

In the logbook entry https://logbooks.jlab.org/entry/3886559  I used the
data from the CDC's HV scans to show that the chamber gain relative to that
at 2125V is independent of the flux, and calculated the new HV that we
would want to use at two fairly extreme pressure values.  I then
ran garfield simulations to see what the effect on the drift times would
be.  These showed that the drift times at the extreme pressures and our new
HV values would be closer to those at the standard pressure and HV values
(2125V and 760mmHg) than the drift times would be using 2125V, which should
bring improvements in time resolution.

The advantage of using an AI to calculate the desired HV over performing
the calculation manually as I did is that the AI has already been trained
for a wide variety of flux and pressure conditions, using the GlueX data
from 2017-2020.  For the manual calculation, I would have to wait until the
atmospheric pressure reaches 760mmHg to know what the chamber gain would be
at that pressure (or extrapolate from the pressure of the day + another
measurement after the pressure has changed sufficiently), and I would have
to assume that the flux would be constant.  The advantage of using the AI
to predict the calibration values is a huge time saving; the calibrations
would be fine-tuned in the conventional manner but having a good starting
point for the time to distance calibration would be very valuable.

We hope to get a little operational experience with the HV optimization
during PrimEx.

Best regards,
Naomi.
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <https://mailman.jlab.org/pipermail/halld-cpp/attachments/20210818/c2fe1f67/attachment-0001.html>
-------------- next part --------------
A non-text attachment was scrubbed...
Name: gainfactor_vs_dens_colourcoded.png
Type: image/png
Size: 16683 bytes
Desc: not available
URL: <https://mailman.jlab.org/pipermail/halld-cpp/attachments/20210818/c2fe1f67/attachment-0001.png>