pin1at pull request - acisops/psmc_check GitHub Wiki

This page exists to reproduce Dick's notes from his PR to a previously existing version of this repository regarding a fix to the pin1at node in psmc_check. Each comment is prepended by the name of the person who wrote it.

I've added code to psmc_check.py to make it roll aware.

Tom and I have come up with a decent-fitting model, which is rather different in most details from the previous one. This is essentially his fit5_psmc_spec.json file, with minor changes of mine.

I have bumped the VERSION number from 0.3 to 0.4.

Comparison between the old and new PSMC models can be done on recent weeks by looking at these web pages:

old: http://cxc.cfa.harvard.edu/acis/PSMC_thermPredic/APR1816/oflsa/ new: http://cxc.cfa.harvard.edu/acis/tmp/psmc_APR1816/index.html

old: http://cxc.cfa.harvard.edu/acis/PSMC_thermPredic/APR0416/oflsc/ new: http://cxc.cfa.harvard.edu/acis/tmp/psmc_APR0416/index.html

The Detector Housing Heater was turned on on 2015:223. To assess how much of an overprediction we should expect from the present model (0.4) if we turn the heater off again, I've plotted residuals in guifit (data minus model vs. data) for two time periods, using this model: [a] the 400 days prior to turning the heater on in 2015, and [b] the 245 days after turning the heater on. In the hot case, it appears that the model overpredicts the heater-off case by 1 to 2 degrees C, which we can live with until such time as there's enough heater-off data to recalibrate if need be.

Before the heater was turned off:

After the heater was turned on:

As further "it does what we expect" validation, here are two plots for times with a large negative value of off-nominal rolls. On day 2015:351-352, and again on 2015:353-354, the off-nominal roll angle were -18 and -16 degrees. We expect, since the PSMC is physically on the +Y side of the ISIM, that this will be a hot attitude for the PSMC. Indeed, the currently baselined model underpredicts at these times:

while the new model proposed here does a much better job of matching the data:

Other effects (mostly due to refitting the parameters) are responsible for the model improvements after day 2015:361 or so.