Failure rate of Outer Detector PMTs

[last edited 11/05/97, HGB]

The Anti-Sorter is one of the 3 online programs on sukant which is directly handling all raw digitized PMT data of the outer detector. In early spring 1997, I enhanced it with a feature to generate a table of suspicious PMT channels at the end of each run.

In average it was observed that with each HE+OD trigger (roughly 5 Hz with the current threshold settings) there's a 10%-20% likelihood of a PMT in the outer detector to generate an electrical pulse within the 16 µsec TDC boundary window of the trigger signal. Each pulse within that window is recorded by a TDC channel where it is digitized and then sent to the OD online workstation (sukant), from where is is processsed by the Anti-Sorter - along with all other TDC, timing, and event status data. At the end of each run, the Anti-Sorter saves an ASCII table to a file on sukant's disk at: /home/online/pmtstats/deadpmts.runXXXX (with XXXX = the run number). The table is listing all "suspicious" PMT channels. See the list of the latest run (SuperK password needed).
With "suspicious" defined as:

"DEAD" = no hits at all.
"LOW OCCUPANCY" = too few hits, average occupancy of < 1% per trigger event (relatively quiet PMT),
"HOT" = far above average hit occupancy of > 50% per trigger event (possible flasher candidate).

This feature has been tested for a few test runs (3600, 3611, 3641) in late February/March 1997 and was finally implemented late March 1997, starting with run 3750. Since then, several hundreds of runs have taken place, resulting in several hundreds of such tables. I ran a quick scanning filter through all these files up to the recent runs and it resulted in a summary table, with one line for each run with more than 10000 OD+HE trigger events, listing the numbers of dead, low/high occupancy and "good" PMTs of each run. Running this summary table through a plotting or math software such as MATLAB then results in curves of "alive" or "quiet" OD PMTs:

According to this curve, the average failure (death) rate since April is rougly 1 OD PMT per 10 days. Not quite flat though ...

Last-not-least, by comparing a new deadpmts.runXXXX file with the most previous file (deadpmts.runYYYY, with XXXX=YYYY+1), one can find out which particular PMT channels were starting to fail at which run. Click here for the latest updates.