Overview

This page summarizes the design & implementation of the interface between the live FACET-II controls system and the FACET2E Bmad model. The infrastructure consists of two layers:

• BmadLiveModel: a python class responsible for watching the accelerator and updating a live instance of PyTao based on extant machine settings
• The live model PVA service: a process that runs it's own BmadLiveModel and publishes data to table PVs

This document outlines the design of the modeling infrastructure.Source code can be viewed on GitHub, and API documentation is available online.

• GitHub mirror: F2_live_model.git
• API documentation: f2-live-model.readthedocs.io/

BmadLiveModel

The goal of this class is to provide access to an instance of Tao that can view/manipulate the FACET2 Bmad model with settings that match those of the actual production accelerator.

The BmadLiveModel python class loads the FACET2E lattice and runs a local copy of pytao.SubprocessTao This object connects to the production controls system via EPICS Channel Access, and connects callback functions to each live quantity (PV) of interest. When live quantities change, these callbacks will submit update instructions to a shared queue. Additionally, a daemon thread (model-update) will periodically empty the queue and execute all the scheduled update commands and then update the BmadLiveModel.live data structure.

Modes of use

The BmadLiveModel operates in three distinct modes, depending on the input flags at construction:

f2m = BmadLiveModel(design_only=True)

f2m = BmadLiveModel(instanced=True)
f2m.refresh_all() 
# --> your code here

f2m = BmadLiveModel()
f2m.start()
# --> your code here
f2m.stop

with BmadLiveModel() as f2m:
    # --> your code here

Daemon process design

This class makes use of both thread-level and process-level parallelism. Controls system monitoring, processing of update requests and data structure changes are handled asynchronously by callback functions and a model-update daemon thread. The Tao instance itself is contained in a separate subprocess via pytao.SubprocessTao.  Controls monitoring and device update management are handled asynchronously to optimize performance and minimize delays between changes being broadcast over the network and being mirrored in the model. Tao itself is wrapped in a subprocess for handling Fortran errors, and calls to BmadLiveModel.tao will still block the main python executable

PV callback functions are responsible for the following actions:

1. convert the new value into Bmad units (i.e. from kGm → T, GeV → eV etc)
2. submit a tuple of (element name, attribute, value) to the queue

Separately, the model-update thread will run forever in a loop, at each iteration it will:

1. empty the command queue of (approximately*) all submitted updates
2. run all the set ele commands, then re-calculate lattice parameters
3. update the live.p0c, live.e_tot, live.twiss, live.quads, live.bends, live.rf data structures (← or update on request?)

LEM

LEM (LINAC energy management) is a set of algorithms responsible for estimating the beam momentum profile and calculating the corresponding lattice quad settings. This code only calculates the momentum profile, errors and magnet settings. Actually setting magnets in the accelerator will be handled by a LEM server GUI (coming soon...).

• BmadLiveModel will monitors the klystron complement and bend magnet settings to calculate the live momentum profile of the beam
• Periodically, the server will use the live momentum profile to calculate energy errors and magnet settings, which are published to a PV

Calculating the live momentum profile

1. start by estimating the momentum profile based on reported klystron amplitudes & phases

   

2. The estimated momentum profile is unlikely to add up to exactly the correct energy, so we need to scale things by a fudge factor to match reality. Since the actual final energy of the beam in each linac is known form the bend magnet settings, we can calculate this number by comparing the estimated final energy to the actual energy

   
   1. under the hood this fudge factor just scales all the cavity voltages in a given region up/down
   2. the fudge factor will be closer to 1 when ENLDs are current and linac phases are accurate

Calculating magnet settings

1. calculate relative to design values using a dimensionless "LEM error", i.e. the ratio of the live momentum profile to the design profile

   
2.  calculate relative to design values using design multipole coefficients (lcls style)
   1. get design multipole coefficients (k0_des, k1_des for bends/quads), and (effective) magnet lengths l_eff from the model

   2. calculate the live magnetic rigidity & required field strength (where E_live is in GeV) 

      

3. calculate relative to extant settings

   
   1. note: requires a previously implemented BLEM → if LEM is bootstrapping (i.e. run for the first time without reference to a previous trim) just skip the matching quads
   2. matching quads should always scale from extant values

F2 Live Model Server

The live model server runs its own BmadLiveModel, and periodically writes live model data for NTTables accessible on the controls system via EPICS PVAccess. The code for the live model itself is relatively simple, and is derived from the lcls_live_model server.

Access to the live Bmad model server (for twiss/rmat data), should be provided through the python meme service

The format of the publish tables are as folllows:

Implementation details

Dependencies

The live model needs to run on the production network (primarily on facet-srv02)

• numpy, spicy
• pyEPICS
• p4p
• Bmad + Tao
• Pytao
• facet2-lattice

Source files