Chandra multi-mass model tools

This suite of tools provides the tools to derive, use and maintain Chandra multi-mass (aka nmass) models. The key tools and files are:

• nmass.py: Actual multi-mass model evaluation code
• fit_nmass.py: Calibrate a full multi-mass model
• nmass_model.py: Module providing infrastucture for fitting multi-mass model
• predict_nmass.py: Predict temperatures for a multi-mass model
• fit_node.py: Fit for one node of a multi-mass model using telemetry for coupled nodes
• node_model.py: Module providing infrastucture for fitting one node of a multi-mass model
• predict_node.py: Predict temperatures for one node of a multi-mass model
• files_def.py: File naming conventions
• <modelname>/pars_<modelname>.json: Model coefficients for a model
• minusz/pars_minusz.json: Model coefficients for Minus-Z model
• minusz/fit.*: Outputs from fitting for Minus-Z model

These tools depend on elements available via the Ska runtime environment, notably the engineering telemetry archive and the commanded states database.

Multi-mass model definition

The Chandra multi-mass thermal model allows for a linear coupling between an arbitrary number of mass nodes. It includes pitch-dependent heating and an ad-hoc constant thermal bath:

In the actual model implementation the thermal mass (i.e. heat capacity) of each node is normalized to unity. That means that the coupling between two nodes is directional, i.e. U01 != U10. So heat flow from node 1 to 0 is governed by U10 while heat flow from 0 to 1 is governed by U01. This is mathematically equivalent to modeling independent heat capacities with U10 == U01.

Updating model calibration

The fit_nmass.py tool is used to update the calibration coefficients for a model.

Usage: fit_nmass.py [options]

Options:

-h, --help	show this help message and exit
--days=DAYS	Number of days in fit interval (default=90
--stop=STOP	Stop time of fit interval (default=NOW - 7 days)
--method=METHOD
	Fit method (default=simple)
--model=MODEL	Model to predict (default=minusz)
--thaw-pars=THAW_PARS
	List of parameters (space-separated) to thaw (default=none)
--nproc=NPROC	Number of processors (default=1)
--ftol=FTOL	ftol convergence parameter (default=1e-4
--outdir=OUTDIR
	Output directory within <msid> root (default=fit)
--pardir=PARDIR
	Directory containing model params file (default=<model>)
--quiet	Suppress screen output
--nofit	Do not fit (default=False)
--stored-data=STORED_DATA
	Use stored states and telem data from this file instead of database and archive

Outdir

The --outdir parameter is a little special and is configured to conveniently allow many trials of fitting. If the specified output directory already exists (e.g. minusz/fit) then that directory is renamed to be <model>/<outdir>.<N> where <N> is the next in sequence from any other similary named directories. As a specific example, one might find the following files within the minusz directory:

fit/             # Most recent fit trial
fit.1/           # First fit trial
fit.2/           # Second fit trial
pars_minusz.json # Parameters that are used for model calculation

Now if another fit trial is done using fit_nmass.py then fit will be renamed to fit.3 and the new fit results will go in fit. In this way the most recent results are always in fit but one has a chronological record of all trials.

Get the source

First untar copy of the latest <version> of the nmass project tar file:

source /proj/sot/ska/bin/ska_envs.csh
cd <workdir>
untar /proj/sot/ska/share/nmass/nmass-<version>.tar.gz

(If you are planning a calibration update then instead check out a copy of the nmass source repository and enter the nmass directory.)

Single processor example

Fit the pf_* model parameters (Epoch 2010 solar inputs) for the TCYLFMZM and TEPHIN nodes over the date range 2010:001 to 2010:090:

./fit_nmass.py --model=minusz --days=90 --stop=2010:090 --outdir fit \
             --thaw-pars="tcylfmzm__pf_.* tephin__pf_.*"

This will generate a new set of model coefficients named pars_minusz.json in the minusz/fit directory. You can look at a web summary of the results with something like:

firefox minusz/fit/index.html

If the new coefficients and the fit results are acceptable then copy those new coefficients to the like-named file in the minusz/ directory. Then commit the updated file to the source control repository and distribute the file to the FOT for update of the MATLAB model.

A normal “full” calibration update would consist of fitting for the all the Epoch 2010 solar inputs as well as the inter-node couplings over the last 360 days:

./fit_nmass.py --model=minusz --days=360 --thaw-pars="t.*__pf_.* t.*__tau_t.*"

Multi-processor example

When fitting more than about 3-6 months of data it can be useful to take advantage of multiple processors. The fit_nmass.py script is written to allow this.

First, you need to start up the MPD servers on the host machines. The hosts can be either on the localhost or remote machines. The only requirement is that you be able to access them via rsh (e.g. rsh remote_host ls). In order to start a single MPD server process on the ccosmos machine which can allow up to 16 child processes do:

ska   # NEED full ska environment
mpdboot -r rsh --totalnum=1 --file=hosts/ccosmos --verbose --maxbranch=16

The following example will start 4 MPD server processes (each on a different node selected from hosts/mine) that can create up to a total of 12 child processes (3 per node):

mpdboot -r rsh --totalnum=5 --file=hosts/mine --verbose --maxbranch=12

To confirm that the servers have started as expected do:

mpdtrace

Now do a fit including the --nproc option which can be any number up to the --maxbranch value set above:

python fit_nmass.py --days=180 --stop=2010:090 --nproc=4 --model=fssac

If the process hangs up then kill with the kill <pid> command. If this doesn’t seem to work then use kill -9 <pid>. In the worst case shut down the servers (see below) and hopefully all the zombie child processes will die. Be sure to check on the remote hosts that this is the case (ps auxwww | grep fit).

Finally the MPD servers should be shut down:

mpdallexit

This usually generates a warning/error message which can be ignored.

Model validation

The fit_nmass.py tool can be used for model validation to check the accuracy of predictions. This is done with the --nofit option:

python fit_nmass.py --stop=2010:330 --days=30 --nofit --outdir validation

This generates all the usual fit output plots using the current model parameters <model>/pars_<model>json.

If the --stop parameter is not supplied then the validation plot covers the time range that stops 7 days before the current time.

Tools

• nmass
  • Functions